Readmissions Due to Hospital-Acquired Conditions (HACs

September 2012

Readmissions Due to Hospital-Acquired

Conditions (HACs):

Multivariate Modeling and Under-coding Analyses

Final Report

Prepared for

Susannah G. Cafardi, MSW, LCSW, MPH

Centers for Medicare & Medicaid Services

Rapid-cycle Evaluation Group

Innovation Center Mail Stop C3-19-26

7500 Security Boulevard

Baltimore, MD 21244-1850

Prepared by

Richard D. Miller, Jr., PhD

Terry Eng, RN, PhD (c)

Amy M.G. Kandilov, PhD

Jerry Cromwell, PhD

Nancy McCall, ScD

RTI International

3040 Cornwallis Road

Research Triangle Park, NC 27709

RTI Project Number 0209853.231.002.128

_________________________________

Readmissions Due to Hospital-Acquired Conditions (HACs):

Multivariate Modeling and Timing of Clinical Presentation Analyses

Draft Final Report

by Richard D. Miller, Jr., PhD

Terry Eng, RN, PhD (c)

Amy M.G. Kandilov, PhD

Jerry Cromwell, PhD

Nancy T. McCall, ScD, Project Director

CMS Project Officer: Susannah G. Cafardi, MSW, LCSW, MPH

RTI International

CMS Contract No. HHSM-500-2005-00029I

August 2012

This project was funded by the Centers for Medicare & Medicaid Services under Contract No.

HHSM-500-2005-00029I. The statements contained in this report are solely those of the authors

and do not necessarily reflect the views or policies of the Centers for Medicare & Medicaid

Services. RTI assumes responsibility for the accuracy and completeness of the information

contained in this report.

RTI International is a trade name of Research Triangle Institute.

ACKNOWLEDGMENTS

We would like to acknowledge assistance we received in conducting analyses related to

hospital readmissions and preparing this report for submission. First, we would like to thank

Merry Rabb, Matt Urato, and Arnold Bragg who provided valuable assistance in the construction

of the episode-of-care linked file that made this analysis possible and in programming assistance

through the analysis phase. Lastly, we would like to thank Loretta Bohn and Norma DiVito for

assistance with preparation of this report.

This page intentionally left blank.

CONTENTS

SECTION 1 INTRODUCTION AND OVERVIEW OF RESEARCH APPROACH ....................1

1.1 Introduction and Findings from Previous Research.......................................................1

1.2 Summary of Phase II Findings on Readmissions and Potential Estimation Bias ..........2

1.3 Overview of Phase III Research Questions and Analytic Approach .............................4

1.4 Organization of Report ..................................................................................................5

SECTION 2 TECHNICAL APPROACH ........................................................................................7

2.1 Study Sample and Data ..................................................................................................7

2.2 Defining Readmissions ..................................................................................................7

2.3 Defining Discharges to PAC Settings ............................................................................8

2.4 Selection of the Three Study HACs ...............................................................................8

2.5 Comparison Group Matching Criteria ...........................................................................8

2.6 Multivariate Analyses ....................................................................................................9

2.7 Co-morbid Condition Measures ...................................................................................11

SECTION 3 MULTIVARIATE RESULTS FOR LIKELIHOOD OF READMISSION

WITHIN 30 DAYS OF DISCHARGE .................................................................................15

3.1 Choosing among Potential Model Specifications ........................................................15

3.2 Logistic Model Results ................................................................................................15

3.3 The Excess Likelihood of Readmission Attributable to Three Hospital-

Acquired Conditions ....................................................................................................18

SECTION 4 DISCHARGE TO POST-ACUTE CARE SETTINGS ............................................21

4.1 Descriptive Statistics on Discharge Destination ..........................................................21

4.2 Relationship between Discharge Destination and the Likelihood of

Readmission .................................................................................................................22

4.3 Probability of Discharge to a Post-Acute Care Setting ................................................22

SECTION 5 SPECIAL STUDIES .................................................................................................27

5.1 Potential Under-Coding of Hospital-Acquired Conditions..........................................27

5.1.1 Introduction .........................................................................................................27

5.1.2 Data and Methods ...............................................................................................28

5.1.3 Correlation Between Hospital-Acquired Conditions Coded on Hospital

Claims and Hospital-Acquired-Related Conditions Coded on Physician

Claims .................................................................................................................30

5.1.4 Under-Coding of Hospital-Acquired Conditions and the Effect on

Readmission Rates ..............................................................................................35

5.2 Timing to Clinical Presentation of Selected Conditions ..............................................38

5.2.1 Time to Physician Diagnosis of Conditions During Hospitalization ..................38

5.2.2 Post-Discharge Diagnosis of Conditions in Inpatient Settings ...........................42

5.3 Post-Discharge Use of Outpatient Department Drugs for Infection Treatment ..........44

SECTION 6 SUMMARY AND CONCLUSIONS .......................................................................49

BIBLIOGRAPHY .........................................................................................................................51

List of Tables

Table 2-1 The Yale co-morbid condition measures ................................................................. 11

Table 3-1 Multivariate regression estimates of the likelihood of a 30-day readmission for

selected hospital-acquired conditions (HAC) .......................................................... 17

Table 3-2 Excess likelihood of readmission for selected hospital-acquired conditions ........... 19

Table 4-1 Discharge destination for selected hospital-acquired conditions ............................. 21

Table 4-2 Relationship between discharge to a post-acute care (PAC) setting and the

likelihood of readmission for beneficiaries in the hospital-acquired condition and

comparison groups ................................................................................................... 22

Table 4-3 Multivariate regression models for the likelihood of discharge to a post-acute care

setting ....................................................................................................................... 23

Table 5-1 Frequency of hospital and physician coding of selected hospital-acquired

conditions ................................................................................................................. 32

Table 5-2a Readmission rates per 100 discharges for 7-day readmission window for selected

hospital-acquired conditions (HACs) identified from hospital and or physician

claims and for comparisons with discharges with no reported HAC ....................... 36

Table 5-2b Readmission rates per 100 discharges for 15-day readmission window for selected

hospital-acquired conditions (HACs) identified from hospital and or physician

claims and for comparisons with discharges with no reported HAC ....................... 36

Table 5-2c Readmission rates per 100 discharges for 30-day readmission window for selected

hospital-acquired conditions (HACs) identified from hospital and or physician

claims and for comparisons with discharges with no reported HAC ....................... 37

Table 5-3 Time to presentation of selected conditions from physician claims during initial

hospitalization, all claims with a physician diagnosis of selected hospital-acquired

conditions ................................................................................................................. 39

Table 5-4 Physician specialty from the first diagnosis of HAC-related condition, among all

hospital claims linked to a physician claim with a HAC-related diagnosis code .... 41

Table 5-5 Time to presentation of selected conditions on inpatient claims following index

hospital discharge with no hospital or physician diagnosis of selected hospital-

acquired conditions .................................................................................................. 43

Table 5-6 Outpatient department (OPD) drug claims within 30 days of hospital discharge, for

patients with and without a HAC ............................................................................. 46

SECTION 1

INTRODUCTION AND OVERVIEW OF RESEARCH APPROACH

1.1 Introduction and Findings from Previous Research

This report continues RTI’s analysis of the effects of the hospital-acquired conditions-

present on admission (HAC-POA) program on utilization, with a specific focus on readmissions.

The HAC-POA program was designed to improve the quality of inpatient care to Medicare

beneficiaries by providing a negative financial incentive, in which inpatient prospective payment

system (IPPS) cases can no longer be assigned to higher-paying MS-DRGs on the basis of

reasonably preventable complications or co-morbid (CC) conditions or major complications or

co-morbid (MCC) conditions that are acquired during the hospital stay. The reimbursement

effects are limited to the initial or index admission only. Thus, even though the hospital may not

receive a higher rate of payment for index admissions in which a HAC occurred under the HAC-

POA program, hospitals could receive additional payments from the Medicare program for care

provided during readmissions related to the hospital-acquired condition. A recent study has

found that an additional $103 million in payment would be withheld if Medicare expanded the

non-payment policy to HAC-related readmissions (see McNair and Luft, 2012).

Previous research has shown to varying degrees that the likelihood of readmission is

greater for patients who experience adverse events than for similar patients who have no such

adverse events (Ashton et al., 1997; Herwaldt et al., 2006; Encinosa and Hellinger, 2008;

Friedman et al., 2009; Friedman and Basu, 2004). Ashton and colleagues (1997) conducted a

meta-analysis of the relationship between early readmission rates (31 days) and inpatient

processes of care and concluded that substandard care was correlated with higher readmission

rates; patients who were readmitted for unplanned reasons were 55% more likely to have had

poor quality of care.

Encinosa and Hellinger (2008) studied the occurrence of seven categories of Agency for

Healthcare Research and Quality patient safety indicators (PSIs) among 161,004 privately

insured patients undergoing surgery. These seven groups of PSIs span the 10 HAC categories

included in the HAC-POA program. Excessive 90-day readmission rates, calculated as the

difference between the readmission rate estimated if all patients had the PSI and the readmission

rate estimated if none of the patients had the PSI, were found for four of the seven PSI groups:

infections (7.7%), pulmonary and vascular problems (3.4%), acute respiratory failure (4.3%),

and metabolic problems (6.3%).

Infections after surgical procedures are an important reason for early readmissions and

have been the focus of a number of recent studies. Herwaldt and colleagues (2006) studied

postoperative nosocomial infections associated with general, cardiothoracic, and neurosurgical

procedures in a large tertiary care medical center and associated VA hospital. They found that

the risk adjusted odds ratio of being readmitted within 30 days of surgery ranged across the three

surgical services from 2.15 to 5.62 for patients with a SSI compared with patients with no SSI.

Friedman and colleagues (2009) used an all payer data set of hospitalizations for surgical

procedures from seven states and found that the relative risk of readmission was higher for

patients experiencing at least one of nine PSIs. The unadjusted rate of 3-month readmission was

25% among patients with a positive PSI compared with 17% among those without a positive PSI.

Risk adjustment reduced the 3-month readmission rate differences yet the rates remained

statistically higher for patients with each of the nine PSIs.

The most recent literature points to a similar relationship between hospital-acquired

conditions and readmissions. Morris et al. (2011) considered unplanned 30-day same-hospital

readmissions among 1,808 surgical patients in an urban, tertiary hospital in FY 2009 and found

that deep vein thrombosis significantly increased the probability of a readmission, with an odds

ratio of 4.7. The reasons for readmission among these patients, however, did not seem to be

related to the deep vein thrombosis.

1.2 Summary of Phase II Findings on Readmissions and Potential Estimation Bias

RTI completed a descriptive analysis of the relationship between the Medicare hospital-

acquired conditions and readmissions earlier this year (see Kandilov et al., 2012). In that study,

we examined the rates and reasons for all-cause readmissions among all discharges in FY 2009

and the first 10 months of FY 2010 in which a HAC was coded by the hospital and the patient

was discharged alive. The rates of readmission varied considerably across the different HACs,

with the lowest readmission rate for deep vein thrombosis or pulmonary embolism (DVT/PE)

following certain orthopedic procedures and the highest readmission rate for blood

incompatibility and surgical site infection (SSI) of mediastinitis following a coronary artery

bypass graft (CABG) procedure. Readmission rates increased as the readmission window

expanded from 7 days to 60 days.

Between FY 2009 and FY 2010, we did not discover any large changes in the

readmission rates for any of the HACs, except for among the low-volume surgical site infections,

where fluctuations in the readmission rate from year to year likely have more to do with small

sample sizes than with actual changes in readmissions for this patient population. Septicemia

and pneumonia were among the most common primary diagnoses for readmission across many

of the HACs, and for the surgical site infections, post-operative infections were a common

reason for readmission. Comparing FY 2009 and FY 2010 data, we did not detect any

substantive changes in the reasons for readmission following the development of a HAC during

the initial hospitalization.

To address the incremental effect of a HAC on readmissions for falls and trauma,

vascular catheter-associated infections, and DVT/PE following certain orthopedic procedures,

we developed comparison groups for each of the three HACs using a random sample of

discharges matched to the HAC cases by key clinical and demographic characteristics. For all

three HACs, we found large and statistically significant differences in the readmission rates

between the HAC cases and the matched comparison groups. FY 2009 and FY 2010,

readmission rates were 3 to 6 percentage points higher for discharges with falls and trauma, 6 to

7 percentage points higher for discharges with a vascular catheter-associated infection, and 2 to 3

percentage points higher for discharges with a DVT/PE following certain orthopedic procedures.

Although we found that readmission rates vary by key patient criteria, such as age,

Medicaid status, disability status, and HCC scores, differences in readmission rates between

discharges with the HAC and its respective comparison group persisted across most of these

stratifications. For patients with a fall or trauma, readmissions remained significantly higher

than among the comparison group within all age groups, those with and without Medicaid, all

eligibility groups (aged, disabled, ESRD), both genders, within two racial groups (white and

other), within all levels of HCC score (low, medium, or high), and within those who were not

institutionalized. Significant differences in readmissions also remained when we stratified the

vascular catheter-associated infection patients and comparisons by these same categories, and

additionally there was a significant difference in readmissions among the black patients. For

HACs and comparisons in the DVT/PE group, the only patient characteristics where a significant

difference in readmissions did not persist were within patients over 85, those enrolled in

Medicaid, those with ESRD, those whose race was other, those who had medium or high HCC

scores, and those who were institutionalized.

The significant differences in readmission rates also persisted when we stratify by

important hospital characteristics such as Census division, urban location, teaching status, and

bed size. For falls and trauma, the rate of readmissions was significantly higher for the HAC

cases than for the control cases across all of these stratifications of hospital characteristics. For

vascular catheter-associated infections, readmission rates for HAC cases were also significantly

higher within all Census divisions, urban and rural hospitals, teaching and non-teaching

hospitals, and for hospitals with 100-299 beds and hospitals with 300 or more beds.

While the rates of readmission for the beneficiaries who acquired one of the three

conditions during their hospitalization were much higher than for comparison beneficiaries, we

found many of the same reasons for readmission for these two groups across our two years of

data. The primary exception was the “infection of a central venous catheter,” which was one of

the top five reasons for readmission among those with a hospital-acquired vascular catheter-

associated infection, while it was not among the top reasons for readmission among the

comparison group beneficiaries.

Finally, we created a separate study sample to conduct further investigations of

mediastinitis following coronary artery bypass graft (CABG) surgery. The number of

mediastinitis cases that were HACs is very small so we undertook this study to examine the

possible degree of under-reporting of mediastinitis during the hospital period or clinical

presentation of mediastinitis after discharge. The primary motivation for this study was to

examine the degree to which readmission estimation bias may exist due to identification errors in

the dependent variable because of either under-reporting of the HAC by the hospital or a delay in

clinical presentation until after discharge.

The mediastinitis study sample included all discharges with a CABG procedure in either

FY 2009 or FY 2010. We linked the MedPAR records for these discharges with all physician

claims billed during the admission and all physician and hospital outpatient department claims

for a 60-day follow-up period and explored the reporting of mediastinitis by physicians during

the hospitalization and follow-up periods.

Of the 195 cases of mediastinitis identified during the index hospitalization using a

hospital or physician diagnosis, 65% were coded only on the physician claims, and 21% were

coded only on the hospital claims. The rate of agreement between hospital and physician coding

of mediastinitis was poor, with only 14% of all mediastinitis cases identified in the hospital

coded on both hospital and physician claims. Most physician-reported diagnoses of mediastinitis

occurred between day 9 and day 23 following the CABG surgery. The first physician diagnosis

of mediastinitis was most likely to be made by an infectious disease specialist.

Overall, we found low rates of reporting of mediastinitis by physicians after discharge

from the hospital. However, the rate of observed interactions between the patient and their

primary surgeon post-discharge was extremely low; only 14 out of 149,395 Medicare

beneficiaries had a follow-up appointment within 30 days. This low number is likely a reflection

of the global billing payment policy. Thus, it would appear that the use of Medicare claims with

the global billing convention may not be an adequate source of information to conduct post-

discharge analyses for beneficiaries having major surgical procedures subject to the global

surgical payment policy.

1.3 Overview of Phase III Research Questions and Analytic Approach

The readmission multivariate analyses presented in this report represent an extension of

the descriptive analysis of readmissions that we completed in Phase II of the study. The Phase

III research questions that we address in this study are:

• Does the likelihood of readmission differ between cases that report HACs and similar

cases that do not report HACs?

• Does the likelihood of use of post-acute care (PAC) services differ between cases that

report HACs and similar cases that do not report HACs?

We use a two stage estimation strategy for readmissions. The first stage is the estimation

of a single period, mixed-effects level model where logistic regression is used to estimate the

likelihood of having a readmission within 30 days of discharge. A dichotomous variable is

included for the presence or absence of a HAC (1 = HAC; 0 = no HAC). We report the odds

ratio (OR) on the presence of a HAC from the logistic regressions along with an indication of the

level of statistical significance of the effect. Odds ratios greater than 1.0 indicate an increased

likelihood of a readmission; odds ratios less than 1.0 indicate a decreased likelihood of a

readmission.

In the second stage, we replicate the method of Encinosa and Hellinger (2008) and report

the excess likelihood of readmission associated with HAC status and its associated standard error

using the multivariate logistic regression models developed in the first stage. In general, the

strategy is to use the results of the logistic regression models to estimate the likelihood of

readmission assuming that all hospitalizations had an adverse event (HAC) and then estimate the

likelihood of readmission assuming that all hospitalizations had no adverse event (no HAC).

The difference between the two sets of predictions is the “excess” likelihood of readmission that

can be attributed to the HAC after controlling for hospital- and patient-level characteristics,

including co-morbidities.

To better understand the role of post-acute care services in observed readmission rates,

we also estimate the single-period, multilevel model on the likelihood of any post-acute care

services following the general estimation strategy for readmissions. We do not generate

estimates of the “excess” likelihood for PAC admissions, but do report the odds ratios from the

logistic regressions. We also provide descriptive statistics regarding the relationship between

two discharge destinations (PAC and home) and the likelihood of readmission. These analyses

consider this relationship for patients both with and without HACs present.

We also conduct a special study which is an extension of last year’s work to examine the

degree to which readmission estimation bias may exist due to identification errors in the

dependent variable because of either under-reporting of the HAC by the hospital or a delay in

clinical presentation until after discharge. The Phase III research questions that we address in

this special study are:

• What proportion of HAC cases are identified during the index hospitalization period

from physician claims and what is the degree of concordance with the MedPAR claim

diagnosis for these HACs?

• What is the typical timing to a physician diagnosis of a HAC-related condition during

a hospitalization? And what proportions of patients with a HAC and without a HAC-

related condition are diagnosed on a subsequent inpatient claim within 7, 15, or 30

days following hospital discharge?

• What evidence do we find of treatment for an infection among the outpatient

department drug claims for patients at-risk of developing a HAC-related infection

within 30 days following discharge?

To conduct further investigation of potential under-coding or post-discharge presentation

of the more frequent HACs, we conducted a series of descriptive analyses. To analyze the

potential degree of under-coding during the hospitalization, we compare the degree of

concordance in coding a HAC between hospital claims and linked physician claims for the

hospitalization period. We examined the timing between procedures that put patients at risk of

developing a HAC and physician diagnosis of the HAC-related conditions in the hospital, and we

also analyzed post-discharge presentation of the HAC-related conditions by linking hospital,

SNF, IRF, and LTCH claims in the 30-days following the index hospitalization to the index

hospital claim and reporting the 7-, 15-, and 30-day readmission rates for patients at risk of

developing the HAC during the hospitalization and for those that have a HAC reported. We do

not include physician or hospital outpatient department (OPD) claims in this analysis as the rate

of follow-up for patients with a major surgical procedure that puts them at risk of developing a

HAC is extremely low during the post-discharge period due to the Medicare global surgical

payment policy that does not allow physicians to routinely bill for services within 90 days of the

procedure that are related to follow-up care for the procedure. For the infection-related HACs,

we also linked outpatient department drug claims within a 30-day period of discharge and report

the percentage of beneficiaries who received antibiotics that would be appropriate for treatment

of a HAC-related infection.

1.4 Organization of Report

Section 2 of this report describes the study sample, data, and methods to answer the first

two research questions related to likelihood of readmission and likelihood of PAC transfer for

patients with a HAC and those at risk for development of a HAC but for whom no HAC is

recorded during the hospital stay. Section 3 provides results of the multivariate modeling of

readmission. Section 4 provides descriptive and multivariate analyses of likelihood of using

PAC services. Section 5 describes the study sample, data, methods, and presents the results of

the special study of potential under-coding and time to clinical presentation for selected HACs.

Section 6 presents a summary of the findings.

SECTION 2

TECHNICAL APPROACH

2.1 Study Sample and Data

For the analysis of readmissions among patients with hospital-acquired conditions

(HACs), we created our study sample by linking Medicare claims data to “index” HAC inpatient

prospective payment system (IPPS) hospital claims. These “index” claims were defined as

claims with the HAC-associated diagnoses coded as not present on admission (POA indicator =

“N” or “U”). The index HAC claims were taken from MedPAR files for FY 2009 and the first

ten months of FY 2010, to allow for a 60-day look-forward period. From these index HAC

claims, we used a cross-referenced beneficiary identifier (HIC number) to look back 180 days

prior to the index admission date in order to identify any Medicare claims (inpatient, outpatient,

home health, and physician claims) for that patient within that period. The claims data for the

look-backs came from FY 2008, FY 2009, and FY 2010, as needed. These look-back claims

were used to calculate concurrent Hierarchical Condition Category (HCC) indicators for these

patients, which were then used to generate indicators of pre-existing medical conditions as

described in Section 2.7. We then used the HIC number to look forward 30 days from the index

discharge date for additional hospital admissions. If a patient was discharged from their index

HAC hospitalization and admitted to another IPPS hospital within a day (with a discharge

designation of an acute care transfer), then the 60-day follow up period began with the discharge

date from that second transfer hospitalization.

The study sample was limited to beneficiaries who were residents of the U.S., who were

enrolled in Medicare Parts A & B, who did not have Medicare as a secondary payer, and who

were not enrolled in managed care during their HAC index claim, during the 180 days prior to

the index admission, and during the 30-day period following the index discharge. The sample

was also limited to patients who were discharged alive from their index hospitalization. These

exclusions allowed us to focus on Medicare patients with HACs who could possibly have a

readmission and whose readmission claims we would likely find using MedPAR claims data.

For example, if a Medicare beneficiary with an index HAC admission switched to Medicare

managed care during the 60-day follow-up period, any hospital readmissions they might have

had would not be present in the MedPAR claims data. Including these beneficiaries in the

sample could lead to an under-estimation of the readmission rates.

2.2 Defining Readmissions

For the statistics presented in this report, we use a measure of hospital all-cause

readmissions and include all admissions to acute care hospitals that occur within 30 days of the

index claim discharge date, regardless of the clinical reason for the admission. In addition to

IPPS hospitals, an admission to a critical access hospital (CAH) or to another non-IPPS hospital

that is paid under Medicare Part A (such as a Cancer hospital or a Children’s hospital) following

an index IPPS hospital discharge is considered a readmission. This measure of readmissions

does not include admissions to an inpatient rehabilitation facility (IRF) or to a long-term care

hospital (LTCH), which are included among our measure of post-acute care. Discharges from

the index hospitalization to another acute care IPPS hospital, where the index discharge date is

within one day of the next admission date and the discharge destination is a transfer, are treated

as transfer cases and so are not included as readmissions. The 30-day look-forward period

begins with the discharge date of the transfer hospitalization, if there is one.

2.3 Defining Discharges to PAC Settings

For the analyses of post-acute care utilization, we created a measure based on the

discharge destination variable in the MedPAR data. The following settings were included in our

PAC definition: skilled nursing facilities, organized home health service organizations,

intermediate care facilities, inpatient rehabilitation facilities, and long-term care hospitals.

2.4 Selection of the Three Study HACs

Based on our initial descriptive statistics produced for the Strategic Memo: Strategy to

Estimate Readmissions Due to Hospital-Acquired Conditions (HACs), we selected three HACs

from the current set of HACs for further analysis in this report. The primary criterion for our

selection was that the chosen HACs have a sufficient volume to estimate statistically reliable

descriptive statistics, allowing us to examine variation in readmission rates across beneficiary

characteristics. Using this criterion, we selected the following three HACs for the Phase II

report:

• Falls and trauma, with 7,954 HAC-associated live discharges in FY 2009 and the

first 10 months of FY 2010.

• Deep vein thrombosis or pulmonary embolism (DVT/PE) following certain

orthopedic procedures, with 4,195 HAC-associated live discharges in FY 2009 and

the first 10 months of FY 2010.

• Vascular catheter-associated infection with 5,167 HAC-associated live discharges in

FY 2009 and the first 10 months of FY 2010.

We continue to analyze these three HACs in this Phase III report.

2.5 Comparison Group Matching Criteria

To develop a valid comparison group we selected discharges based on a small set of

clinical or demographic characteristics held in common with the specific HAC cases, and then

used a larger set of covariates in the outcome regressions. Matching is a common technique

found among empirical studies on this topic. For the descriptive analysis in this report, we took

a multivariable matching approach. Multivariable matching uses a limited number of specific

characteristics and identifies controls that match on all of the variables.

To construct appropriate comparison groups for the three selected study HACs, we

matched each index claim identified with a HAC to 10 IPPS claims without a HAC but with the

same MS-DRG and demographic characteristics (sex, race, and age) as the HAC claim. In the

cases where a 10:1 match was not obtainable, we reweighted the matches that were made to

simulate a 10:1 match. Any claims with the HAC-associated diagnosis codes identified as

present on admission (POA indicator equal to “Y” or “W”) were excluded from the comparison

group, since conditions coded as present on admission could potentially be true HACs that were

miscoded. Including true HACs in the comparison group could introduce bias in our results.

Thus, the comparison group for each of the three HACs contained no index claims with the

specified HAC-associated diagnoses.

No additional restrictions were placed on the comparison group for the falls and trauma

HACs. For the DVT/PE following certain orthopedic procedures, the set of claims from which

the comparison group was drawn was further limited to those claims containing the orthopedic

procedure associated with this HAC. To better target the population who would be at risk for a

vascular catheter-associated infection, we limited this comparison group to index claims that had

one of two vascular catheter procedure codes (38.93 or 38.95). Note that among patients with

the vascular catheter-associated infection HAC, 38% did not have a vascular catheter procedure

code on their claims. The vascular catheter codes may have been coded after the fifth surgical

procedure code, and thus not picked up by the MedPAR data, or may have been left off of the

claim completely. Readmission rates were similar between the HAC claims that included the

vascular catheter procedure codes and those that did not include the codes.

From these index comparison claims, we linked additional claims data both before and

after the index comparison claim, as described in Section 2.1, in order to calculate readmission

rates and co-morbid conditions. The same sample exclusions – residents of the U.S., enrolled in

Medicare Parts A & B, Medicare not the secondary payer, and not enrolled in managed care –

were applied to the identified comparison groups to ensure analogous samples.

2.6 Multivariate Analyses

To estimate the impacts of each of the three study HACs on the likelihood of readmission

within 30 days, we estimated mixed effects (or multi-level) logistic models. The mixed effects

models are necessary due to the multi-level nature of the data being analyzed. The idea is to

control for both patient- or discharge-level covariates such as co-morbid conditions and age as

well as hospital-level covariates such as size (number of beds). Also, the discharges are

clustered within hospitals, so it is necessary to model this clustering.

The mixed effects logistic model is derived through using the logistic function to model

the probability of readmission based on the value of a latent variable , where i indexes

discharges within hospitals and j indexes hospitals.

The variable can be thought of as a function of HAC status as well as other patient- or

discharge-level characteristics (X) and an individual-level error term (r) as follows:

The mixed effects model is implemented by allowing the βs to vary across hospitals,

which constitute a second level of data. We considered three different specifications for

modeling the likelihood of readmission for each of the three HACs and the likelihood of

discharge to a PAC setting. The first was a random intercept model. The second was a random

intercept model where we allowed the intercept to be a function of hospital-level covariates. The

third was a model, which built upon the second model by estimating a random effect for the

HAC indicator variable.

Model I

In this case the intercept term is equal to an average intercept across all hospitals and a hospital-

specific random error term ( ). The model may be re-written as one equation as follows:

This model is composed of fixed effects (the and ) and random effects (the random error

terms).

Model II

In this case the intercept term is a function of hospital-level covariates ( ) and a hospital-

specific random error term ( ). The model may be re-written as one equation as follows:

This model is composed of fixed effects (the and ) and random effects (the random error

terms).

Model III

In this case the intercept term is a function of hospital-level covariates ( ) and a hospital-

specific random error term ( ). The coefficient on the HAC indicator is equal to the average

value of the coefficient across all hospitals ( ) and a hospital-specific random error term ( ).

The model may be re-written as one equation as follows:

This model is composed of fixed effects (the and ) and random effects (the random error

terms).

In each of the specifications, we control for the following discharge- or patient-level

characteristics (the Xs): age, Medicaid enrollment, original eligibility status, gender, race,

institutional status, and several co-morbid conditions (we discuss the measures for co-morbid

conditions in Section 2.7). In the Models II and III, we control for the following hospital-level

characteristics (the Ws): whether the hospital is located in an urban area, number of beds, and

whether the hospital is an academic medical center (teaching hospital).

2.7 Co-morbid Condition Measures

To control for co-morbid conditions in our models, we included a series of indicator

variables suggested in a report to CMS by the Yale New Haven Health Services

Corporation/Center for Outcomes Research and Evaluation (referred to as Yale for the remainder

of this report). Based on several factors, the Yale team grouped CMS condition categories

(CMS-CCs) into a series of 31 co-morbid risk variables.

The grouping of the CMS-CCs into the

risk variables is presented in Table 2-1.

Table 2-1

The Yale co-morbid condition measures

Co-morbid condition measure CMS Co-morbid conditions included

Severe infection

1 HIV/AIDS

3 Central nervous system infection

4 Tuberculosis

5 Opportunistic infections

Other infectious disease

6 Other infectious disease

111 Aspiration and specified bacterial pneumonias

112 Pneumococcal pneumonia, emphysema, lung abscess

113 Viral and unspecified pneumonia, pleurisy

Metastatic cancer/acute leukemia

7 Metastatic cancer/acute leukemia

Severe cancer

8 Lung, upper digestive tract, and other severe cancers

9 Other major cancers

Other major cancers

10 Breast, prostate, colorectal and other cancers and tumors

11 Other respiratory and heart neoplasms

12 Other digestive and urinary neoplasms

(continued)

See pages 29-30 of Horwitz et al. (2011) for a fuller description of the rationale for the creation of the comorbid

risk variables.

Table 2-1 (continued)

The Yale co-morbid condition measures

Co-morbid condition measure CMS Co-morbid conditions included

Diabetes mellitus

15 Diabetes with renal manifestation

16 Diabetes with neurologic or peripheral circulatory

manifestation

17 Diabetes with acute complications x

18 Diabetes with ophthalmologic manifestation

19 Diabetes with no or unspecified complications

20 Type I diabetes mellitus

119 Proliferative diabetic retinopathy and vitreous hemorrhage

120 Diabetic and other vascular retinopathies

Protein-calorie malnutrition

21 Protein-calorie malnutrition

End-stage liver disease

25 End-Stage Liver Disease

26 Cirrhosis of Liver

Other hematological disorders

44 Other hematological disorders

Drug and alcohol disorders

51 Drug/alcohol psychosis

52 Drug/alcohol dependence

Psychiatric comorbidity

54 Schizophrenia

55 Major depressive, bipolar, and paranoid disorders

56 Reactive and unspecified psychosis

58 Depression

60 Other psychiatric disorders

Hemiplegia, paraplegia, paralysis

and functional disability

67 Quadriplegia, other extensive paralysis

68 Paraplegia

69 Spinal Cord Disorders/Injuries

100 Hemiplegia/hemiparesis

101 Diplegia (upper), monoplegia, and other paralytic syndromes

102 Speech, language, cognitive, perceptual

177 Amputation status, lower limb/amputation

178 Amputation status, upper limb

Seizure disorders and convulsions 74 Seizure disorders and convulsions

Congestive heart failure

80 Congestive heart failure

(continued)

Table 2-1 (continued)

The Yale co-morbid condition measures

Co-morbid condition measure CMS Co-morbid conditions included

Coronary atherosclerosis or angina,

cerebrovascular disease

81 Acute myocardial infarction

82 Unstable angina and other acute ischemic heart disease

83 Angina pectoris/old myocardial infarction

84 Coronary atherosclerosis/other chronic ischemic heart disease

89 Hypertensive heart and renal disease or encephalopathy

98 Cerebral atherosclerosis and aneurysm

99 Cerebrovascular disease, unspecified

103 Cerebrovascular disease late effects, unspecified

104 Vascular disease with complications

105 Vascular disease

106 Other circulatory disease

Specified arrhythmias

92 Specified heart arrhythmias

93 Other heart rhythm and conduction disorders

COPD

108 Chronic Obstructive Pulmonary Disease

Dialysis status

130 Dialysis status

Ulcers

148 Decubitus ulcer

149 Decubitus ulcer or chronic skin ulcer

Septicemia/shock

2 Septicemia/shock

Cardio-respiratory failure and shock 79 Cardio-respiratory failure and cardio-respiratory shock

Acute renal failure

131 Acute renal failure

Pancreatic disease

32 Pancreatic disease

Rheumatoid arthritis and

inflammatory connective tissue

disease

38 Rheumatoid arthritis and inflammatory connective tissue

disease

Respirator dependence

77 Respirator dependence/tracheostomy status

Transplants

128 Kidney transplant status

174 Major organ transplant status

Hip fracture/dislocation

158 Hip fracture/dislocation

SOURCE: Horwitz et al. (2011)

This page intentionally left blank.

SECTION 3

MULTIVARIATE RESULTS FOR LIKELIHOOD OF READMISSION WITHIN 30

DAYS OF DISCHARGE

3.1 Choosing among Potential Model Specifications

We estimated three separate specifications or models to estimate the impact of HAC

status on the likelihood of readmission, while controlling for patient and hospital characteristics.

In this section, we provide a rationale for the choice of final specification that we made for each

of the three HACs considered in this study. The results for the three specifications for each HAC

can be found in Appendix Tables 1 through 3.

Falls and Trauma. The result regarding the effect of having a HAC on the likelihood of

readmission within 30 days is quite consistent across all three specifications. This is also true for

the effect of the control variables. Including the hospital-level covariates has very little impact

on the coefficients for the patient- or discharge-level covariates, although at least some of the

hospital-level covariates are significant.

In all three models, the random effect on the intercept is significant. At the same time,

only roughly 7% of the total variance is determined to be due to differences in readmission rates

across hospitals. In Model III, the random effect on the HAC indicator is not significant and

neither is the covariance between the intercept and the HAC indicator. Allowing for the random

effect on the HAC indicator has no significant impact on the results. Based on these results, we

report the results from Model II for falls and trauma.

Vascular Catheter-Associated Infection. Again, the result regarding the effect of having

a HAC on the likelihood of readmission within 30 days is quite consistent across all three

specifications. This is also true for the effect on the control variables. Including the hospital-

level covariates has very little impact on the coefficients for the patient- or discharge-level

covariates, although at least some of the hospital-level covariates are significant.

In all three models, the random effect on the intercept is significant. At the same time,

only roughly 5% of the total variance is determined to be due to differences in readmission rates

across hospitals. In Model III, the random effect on the HAC indicator is not significant and

neither is the covariance between the intercept and the HAC indicator. Allowing for the random

effect on the HAC indicator has no significant impact on the results. Based on these results, we

report the results from Model II for vascular catheter-associated infection.

DVT/PE Following Certain Orthopedic Procedures. In this case, the result on the effect

of having a HAC is quite different when allowing for a random effect on the HAC indicator in

Model III. In addition, the random effect on the HAC indicator is quite significant. Based on the

significance of the random effect on the HAC indicator, we report the results from Model III for

DVT/PE.

3.2 Logistic Model Results

The results of the chosen mixed effects logistic model for each HAC are presented in

Table 3-1. The main finding is that the presence of each HAC has a significant positive impact

on the likelihood of readmission within 30 days. For the falls and trauma HAC and the DVT/PE

HAC, the presence of the HAC is associated with a 21 to 23% increase in the odds of being

readmitted within 30 days, respectively. The presence of the vascular catheter-associated

infection HAC has an even greater impact on the likelihood of readmission. It is associated with

a 33% increase in the odds of being readmitted within 30 days.

As far as the hospital-level covariates are concerned, larger hospitals tend to have higher

readmission rates. For instance, in the falls and trauma sample, the odds ratio for the largest

hospitals (those with 300 or more beds) is equal to 1.139, which indicates that the odds of a

readmission are 14% higher for patients from these larger hospitals than for patients from

hospital with fewer than 100 beds. The effect of hospital size is largest for patients in the

DVT/PE sample, or almost 40% higher. There is no association between bed size and likelihood

of readmission for patients at risk to develop a DVT or PE. Discharges from academic medical

centers are also associated with a higher likelihood of readmission for patients at risk of a fall or

trauma, 18%, and vascular catheter-associated infection, 22%. The results of the influence of

level of urbanicity are mixed. Urbanicity has no correlation with the likelihood of readmission in

the falls and trauma sample. At the same time, discharges from urban hospitals are associated

with a higher likelihood of readmission in the vascular catheter-associated infection sample,

10%, and with a lower likelihood of readmission in the DVT/PE sample, 12%.

The results on patient age are also mixed. In the falls and trauma and DVT/PE samples,

there is a positive relationship between age and the likelihood of readmission after controlling for

other factors, including co-morbidities. In the vascular catheter-associated infection sample,

there is to be a negative relationship. The effect of Medicaid enrollment is more consistent

across the three samples. In each case, Medicaid enrollment is associated with a greater

likelihood of readmission. In the DVT/PE sample, the odds of readmission is 35% higher for

Medicaid enrollees than for non-enrollees, while for the falls and trauma and vascular catheter-

associated infection samples, the odds of readmission among Medicaid enrollees are 20 and 15%

higher, respectively.

Among the discharges in the falls and trauma and DVT/PE samples, original Medicare

eligibility status and gender are important determinants of the likelihood of readmission.

Patients who initially became eligible for Medicare due either to disability or ESRD status have a

greater likelihood of readmission than patients who initially became eligible due to age. Women

generally have a lower likelihood of readmission than men. In the falls and trauma sample, the

odds of readmission were 9% lower for women than for men and in the DVT/PE sample, the

odds of readmission were more than 20% lower. Generally, race and institutional status have no

effect on the likelihood of readmission. Co-morbidities are very important determinants of the

likelihood of readmission. In each sample, at least one-half of the Yale co-morbidity measures

have significant odds ratios. In all cases where the odds ratios are significant, the odds ratio is

greater than one, indicating that the co-morbidities are associated with a greater likelihood of

readmission.

Table 3-1

Multivariate regression estimates of the likelihood of a 30-day readmission for selected

hospital-acquired conditions (HAC)

Variable

Falls and

trauma

(n=78,827)

Vascular

catheter-

associated

infection

(n=44,981)

DVT/PE

following

certain

orthopedic

procedures

(n=41,432)

HAC Indicator

1.214**

1.330**

1.229**

Hospital-Level Covariates

Urban

0.998

1.096*

0.875*

Number of beds (reference is “fewer than 100”

100-299

1.112**

1.028

1.376**

300 or more

1.139**

1.097

1.393**

Academic medical center

1.175**

1.218**

1.109

Discharge-Level Covariates

Age (reference is “less than 65”)

65-74

0.989

0.950

1.078

75-84

1.106*

0.893*

1.682**

85 and older

1.225**

0.779**

2.371**

Enrolled in Medicaid

1.197**

1.145**

1.346**

Original eligibility (reference is “aged”)

Disabled

1.092**

1.020

1.348**

ESRD

1.558**

1.080

2.091**

Gender: Female

0.910**

1.033

0.771**

Race (reference is “white”)

Black

0.960

0.984

Asian

0.891

1.018

0.876

Other

0.915

0.952

0.915

Institutionalized

0.778

0.823

0.400

Yale Comorbidity Measures

Severe infection

1.089

1.222*

0.510

Other infectious disease

1.142*

1.059

1.122

Metastatic cancer/acute leukemia

1.286**

1.260**

1.662*

Severe cancer

1.235**

0.993

1.226

Other major cancers

1.039

0.967

0.876

Diabetes mellitus

1.182**

1.134**

1.132*

Protein-calorie malnutrition

1.152**

0.987

1.503**

End-stage liver disease

1.368**

0.953

2.346**

Other hematological disorders

1.404**

1.160**

1.099

Drug and alcohol disorders

1.410**

1.116

1.789**

Psychiatric comorbidity

1.121*

1.185**

1.148

Hemiplegia, paraplegia, paralysis and functional disability

1.175**

1.027

1.124

Seizure disorders and convulsions

1.100

1.121*

1.674**

Congestive heart failure

1.331**

1.113**

1.353**

Coronary atherosclerosis or angina, cerebrovascular Disease

1.147**

1.076*

1.222**

Specified arrhythmias

1.120**

1.053

1.101

COPD

1.328**

1.072*

1.381**

(continued)

Table 3-1 (continued)

Multivariate regression estimates of the likelihood of a 30-day readmission for selected

hospital-acquired conditions (HAC)

Variable

Falls and

trauma

(n=78,827)

Vascular

catheter-

associated

infection

(n=44,981)

DVT/PE

following

certain

orthopedic

procedures

(n=41,432)

Dialysis status

1.242**

1.181**

0.997

Ulcers

1.138**

0.973

1.389*

Septicemia/shock

1.127**

1.147**

1.282

Cardio-respiratory failure and shock

1.057

1.068

1.017

Acute renal failure

1.342**

1.230**

1.657**

Pancreatic disease

1.254**

1.217**

1.427

Rheumatoid arthritis and inflammatory connective tissue disease

1.178**

1.127**

1.314**

Respirator dependence

1.497**

1.074

5.516**

Transplants

1.560**

1.152

0.794

Hip fracture/dislocation

0.938

0.915

1.330**

NOTES:

* indicates statistically significant difference using negative binomial regression with p<0.05.

**indicates statistically significant difference using negative binomial regression with p<0.01.

SOURCE: falls_re_readmt_models.log, vcath_centered_xtmelogitJun20_2012.log,

dvt_centered_xtmelogitJun20_2012.log

3.3 The Excess Likelihood of Readmission Attributable to Three Hospital-Acquired

Conditions

In Table 3-2, we present our multivariate regression results on the excess likelihood of

readmission attributable to three selected HACs. We generated the excess likelihood by using

the results of the logistic models to estimate the likelihood of readmission assuming that all

hospitalizations had an adverse event (HAC) and then to estimate the likelihood of readmission

assuming that all hospitalizations had no adverse event (no HAC). The difference between the

two sets of predictions is the “excess” likelihood of readmission that can be attributed to the

HAC after controlling for patient and hospital characteristics. We find that the falls and trauma

HAC leads to an excess likelihood of readmission of 2.9 percentage points while the vascular

catheter-associated infection HAC leads to an excess likelihood of readmission of 5.6 percentage

points and the DVT/PE HAC leads to an excess likelihood of readmission of 1.8 percentage

points. All of these results are statistically significant.

Table 3-2

Excess likelihood of readmission for selected hospital-acquired conditions

Hospital-acquired condition Excess likelihood Standard error

Falls and trauma

2.9%

0.5%

Vascular catheter-associated infection

5.6%

0.7%

DVT/PE following certain orthopedic procedures

1.8%

0.7%

NOTES: DVT/PE = Deep vein thrombosis or pulmonary embolism

SOURCE: DVT_predicted.xlsx, Vcath_predicted.xlsx, fall_readmt_predict.log

This page intentionally left blank.

SECTION 4

DISCHARGE TO POST-ACUTE CARE SETTINGS

4.1 Descriptive Statistics on Discharge Destination

Table 4-1 presents descriptive statistics on the discharge destinations of patients in the

HAC group and the comparison group. In the falls and trauma sample, 76.9% of the patients

with the HAC were discharged to one of the PAC settings, while 57.4% of the patients in the

control group were discharged to a PAC setting (statistically significant difference). On the

other hand, patients with the HAC were significantly less likely to be discharged to home—

13.7% compared to 33.7% among the controls. There is no significant difference in the

likelihood of discharge to other settings between those patients with the falls and trauma HAC

and the comparison group.

Table 4-1

Discharge destination for selected hospital-acquired conditions

Hospital-acquired condition HAC group

Comparison

group

Difference

Falls and trauma

Discharged to PAC setting

76.9%

57.4% 19.5%**

Discharged home 13.7%

33.7%

-20.0%**

Other 9.4%

8.9%

0.5%

Vascular catheter-associated infection

Discharged to PAC setting 67.9% 53.2% 14.7%**

Discharged home 25.0%

24.9%

0.1%

Other 7.1%

21.9%

-14.8%**

DVT/PE following certain orthopedic procedures

Discharged to PAC setting

88.0%

81.8% 6.2%**

Discharge home 8.5%

13.9%

-5.4%**

Other 3.5%

4.3%

-0.8%**

NOTES: HAC = Hospital-acquired condition; PAC = Post-acute care; DVT/PE = Deep vein thrombosis

or pulmonary embolism

**indicates statistically significant difference using negative binomial regression with p<0.01.

SOURCE: falls_pac_descriptive.log, vcath_pac_descriptive.log, dvt_pac_descriptive.log

In the vascular catheter-associated infection sample, 67.9% of the patients with the HAC

were discharged to a PAC setting, compared with 53.2% of the patients in the control group

(statistically significant difference) Patients with the HAC were discharged home at the same

rate as patients in the control group, leaving a significant difference in the rate of discharge to

other settings.

In the DVT/PE sample, there were significant differences between the HAC group and

the comparison group for all discharge destinations, with the HAC group more likely to be

discharged to a PAC setting (88.0% vs. 81.8%) and less likely to be discharged home (8.5% vs.

13.9%).

4.2 Relationship between Discharge Destination and the Likelihood of Readmission

Table 4-2 illustrates the relationship between the likelihood of readmission and whether

the patient was discharged to a PAC setting for patients in the HAC and comparison groups. The

results indicate that patients discharged to PAC settings are more likely to be readmitted within

30 days than those patients who were not discharged to PAC settings.

Table 4-2

Relationship between discharge to a post-acute care (PAC) setting and the likelihood of

readmission for beneficiaries in the hospital-acquired condition and comparison groups

Hospital-acquired condition

Likelihood of

readmission

HAC group

Likelihood of

readmission

Comparison group

Falls and trauma

Discharged to PAC setting 23.7%

19.8%

Not discharged to PAC 19.7%

14.5%

Vascular catheter-associated infection

Discharged to PAC setting 32.0% 30.0%

Not discharged to PAC 28.0%

16.6%

DVT/PE following certain orthopedic procedures

Discharged to PAC setting 12.8%

9.8%

Not discharged to PAC 6.8%

5.2%

NOTES: HAC = Hospital-acquired condition; PAC = Post-acute care; DVT/PE = Deep vein thrombosis

or pulmonary embolism

**indicates statistically significant difference using negative binomial regression with p<0.01.

SOURCE: falls_pac_descriptive.log, vcath_pac_descriptive.log, dvt_pac_descriptive.log

4.3 Probability of Discharge to a Post-Acute Care Setting

We estimated a series of mixed effect logistic models to predict the probability of

discharge to a PAC setting. As we did for readmissions, we estimated each of the three models

described in Section 2.6 for each of the HACs. Based on the results of these models, we present

the results of Model II for falls and trauma and vascular catheter-associated infection. We do

this, because the random effect on the HAC indicator is insignificant for each of these HACs.

We present the results of Model III for the DVT/PE sample, due to the fact that the random

effect on the HAC indicator is significant for this HAC. The model results are reported in

Table 4-3.

Table 4-3

Multivariate regression models for the likelihood of discharge to a post-acute care setting

Variable

Falls and

trauma

(n=78,827)

Vascular

catheter-

associated

infection

(n=44,981)

DVT/PE

following

certain

orthopedic

procedures

(n=41,432)

HAC Indicator

2.668**

1.943**

1.393**

Hospital-Level Covariates

Urban

1.311**

1.390**

1.891**

Number of beds (reference is “fewer than 100”

100-299

1.320**

1.093

2.226**

300 or more

1.297**

1.042

2.585**

Academic medical center

0.858**

0.894*

1.110

Discharge-Level Covariates

Age (reference is “less than 65”)

65-74

1.570**

1.453**

1.363**

75-84

2.074**

1.925**

2.550**

85 and older

2.630**

1.968**

3.009**

Enrolled in Medicaid

1.175**

1.180**

1.199**

Original eligibility (reference is “aged”)

Disabled

1.107**

1.144**

1.351**

ESRD

0.711**

0.740**

1.625

Gender: Female

1.368**

1.145**

1.533**

Race (reference is “white”)

Black

1.229**

0.948

1.321**

Asian

0.584**

0.925

1.249

Other

1.011

0.809**

1.032

Institutionalized

1.333

0.881

0.293

Yale Comorbidity Measures

Severe infection

0.907

0.962

1.141

Other infectious disease

1.220**

1.081

0.810

Metastatic cancer/acute leukemia

0.701**

0.618**

0.833

Severe cancer

0.702**

0.835**

0.840

Other major cancers

0.912**

1.102*

1.074

Diabetes mellitus

1.105**

1.098**

1.149**

Protein-calorie malnutrition

1.276**

1.173**

1.512

End-stage liver disease

0.932

0.844*

1.068

Other hematological disorders

0.825**

0.537**

2.084*

Drug and alcohol disorders

0.852*

0.870*

1.616

Psychiatric comorbidity

0.938

1.061

1.405*

Hemiplegia, paraplegia, paralysis and functional disability

1.310**

1.564**

1.093

Seizure disorders and convulsions

1.055

1.070

1.303

Congestive heart failure

0.913**

0.944*

0.907

Coronary atherosclerosis or angina, cerebrovascular disease

0.956*

1.022

0.904

Specified arrhythmias

1.023

1.029

1.044

COPD

0.924**

1.026

1.237**

(continued)

Table 4-3 (continued)

Multivariate regression models for the likelihood of discharge to a post-acute care setting

Variable

Falls and

trauma

(n=78,827)

Vascular

catheter-

associated

infection

(n=44,981)

DVT/PE

following

certain

orthopedic

procedures

(n=41,432)

Dialysis status

0.796**

0.740**

0.756

Ulcers

1.682**

1.360**

1.039

Septicemia/shock

1.328**

1.227**

1.182

Cardio-respiratory failure and shock

1.064

1.173**

0.856

Acute renal failure

1.018

0.976

1.142

Pancreatic disease

0.751**

0.796**

0.947

Rheumatoid arthritis and inflammatory connective tissue disease

1.137**

0.961

1.230*

Respirator dependence

1.585**

1.333**

1.369

Transplants

0.916

0.740**

1.047

Hip fracture/dislocation

2.395**

1.527**

1.299*

NOTES:

* indicates statistically significant difference using negative binomial regression with p<0.05.

**indicates statistically significant difference using negative binomial regression with p<0.01.

SOURCE: falls_pac_re_models.log, vcath_pac_re_models.log, DVT melogit Jun2012 Req2.log

The main finding is that the presence of each HAC is associated with a greater likelihood

of discharge to a PAC setting. The odds of being discharge to a PAC setting is 2.7 times greater

for patients with the falls and trauma HAC, two times greater for patients with the vascular

catheter-associated infection HAC, and 40% greater for patients with the DVT/PE HAC than for

similar patients without the HACs.

Among the hospital-level covariates, we find that patients discharged from teaching

hospitals are generally less likely to be discharged to a PAC setting, while patients from medium

and large hospitals (those with at least 100 beds) are more likely to be discharged to a PAC

setting than patients from smaller hospitals (those with fewer than 100 beds). Patients from

urban hospitals are more likely to be discharged to a PAC setting than patients from rural

hospitals.

Among the discharge-level covariates, we find that women are more likely to be

discharged to a PAC setting than men and that there is a positive relationship between age and

the likelihood of being discharged to a PAC setting. Medicaid enrollees are more likely to be

discharged to PAC settings than non-enrollees, while patients who were initially eligible for

Medicare due to disability are more likely to be discharged to a PAC setting than patients who

became eligible due to age. Interestingly, patients with a history of ESRD are less likely to be

discharged to a PAC setting, at least in two of the samples (falls and trauma and vascular

catheter-associated infection), although these sample sizes are quite small, 126 and 316 in the

HAC groups, respectively.

The results on the co-morbidity measures are mixed. Several of the measures that are

significant are associated with a greater likelihood of PAC admission, but about half are

associated with a smaller likelihood of PAC admission. It is likely that these co-morbidities are

related to discharge to other inpatient settings that are not included in our PAC measure.

This page intentionally left blank.

SECTION 5

SPECIAL STUDIES

5.1 Potential Under-Coding of Hospital-Acquired Conditions

5.1.1 Introduction

In the Phase II report for this task, “Readmissions Due to Hospital-acquired Conditions

(HACs),” we developed a mathematical model of readmission estimation bias that can occur

when there is error in the measurement of hospital-acquired conditions, particularly when HACs

are not reported. This can occur when clinical manifestation of the HAC occurs after the initial

hospital discharge, such as for a SSI, or under-reporting by hospital staff. In this section, we

focus on under-reporting of HACs on hospital claims, and in Section 5.2, we look more closely

at post-discharge presentation of HAC-related conditions and post-discharge treatment that could

be an indicator of a HAC-related condition.

The model demonstrates how one hospital can have a higher reported HAC rate if (a) it

has more infections in general than average, and/or (b) if it has a higher likelihood of reporting

its HACs. Conversely, a hospital with a lower-than-average HAC rate may truly have fewer

HACs than other hospitals, or it may be under-reporting the incidence of HACs. Thus, two

hospitals may have the same reported HAC rates but different readmission rates per admission

leading to little correlation between the presence of a hospital-acquired condition and the

likelihood of a readmission. It is also possible that one hospital has a lower reported HAC rate

yet has a higher true infection readmission rate. The paradox is explained by the fact that the

HAC rate calculated from claims data reflects two factors: the hospital’s true, overall, HAC rate

(once unreported, post-discharge, infections are accounted for) as well as the hospital’s rate at

which it reports HACs. The latter term may be both positive and negative; thus, an ambiguous

net effect on the overall readmission rate. The reported or coded HAC rate can also vary

positively or negatively with hospitals’ overall infection rate. Thus, it is possible that a hospital

has a high reported infection rate of all infections but a low readmission rate, thereby producing

a zero correlation of reported HAC rates with readmission rates.

Model Implications. Conceptually, we would expect that the relationship between HAC

rates and readmission rates to be positive; a HAC worsens a patient’s health and could require

multiple hospitalizations to treat. However, the “observed HAC” measure is imperfectly

sensitive by failing to capture all true HACs. As a result, the observed relationship between HAC

rates and readmission rates will not match the true relationship.

If the sensitivity is unrelated to the readmission rate and does not vary across providers,

then this situation is analogous to the classic errors-in-variables regression problem, and the

correlation between observed HAC rates and readmission rates will be lower than the true

correlation. This biases the reported HAC coefficient in any readmission model towards zero,

producing an under-estimate of the effect of true HACs on readmissions.

However, it is quite likely that the sensitivity of the observed HAC measure does vary

systematically across providers (and type of HAC). To see this, consider two hospitals which

differ only in their length of stay. One hospital tends to discharge patients as quickly as possible,

whereas the second hospital tends to permit patients to stay in the hospital longer. In this

hypothetical situation, we assume that the procedure infection rates and other aspects of

underlying quality are identical but only the lengths of stay differ. In the early-discharge

hospital, the infection may not be identified until after the patient is discharged. The inpatient

HAC rate for this hospital will be low, but the readmission rate will be high. In contrast, in the

second hospital, since the underlying length of stay is longer, the HAC may be identified and

treated in the hospital prior to discharge (even further lengthening that patient’s stay length).

Assuming the patient is discharged with the HAC fully treated, no readmission would be

necessary. Thus, the second hospital’s reported inpatient HAC rate will be high, but its

readmission rate will be low.

This confounding relationship between observed inpatient HAC rates and readmission

rates is due to the fact that hospitals vary on two dimensions. First, hospitals vary in their true

HAC rates because of differences in their quality of care. Second, hospitals will vary in their

lengths of stay (or any other factor that would impair the sensitivity of the HAC measure). To

counteract the confounding length of stay effect, one option must be to extend the time window

for measuring (recording) HACs into the post-discharge period. Using readmissions to enhance

the measure of true HAC rates can significantly improve the sensitivity of the initial HAC

measure and produce a higher, more accurate estimate of the HAC-readmission link. Care must

be taken, however, in inferring a HAC when using readmission data. Infections not acquired

during the earlier admission will likely be picked up in using readmission data and make the

measure somewhat less specific. Readmission data will also be imperfect to the extent that

infections and other late-appearing HACs are treated in an ambulatory setting without a

subsequent readmission. The modeling suggests taking a careful look at the complex

relationship between a very imperfectly measured estimate of hospital-acquired conditions and

any subsequent readmission rates. The shorter the window, the greater the likelihood that a HAC

had gone unreported during the earlier hospitalization. It also calls for using non-readmission

claims to track ambulatory follow-up of HACs (e.g., physician and outpatient department bills).

5.1.2 Data and Methods

From the twelve hospital-acquired conditions included in the HAC-POA payment policy,

we selected seven for this analysis. Foreign object retained after surgery, air embolism, and blood

incompatibility were excluded due to their relative infrequency in the hospital claims data. For the

remaining HACs – pressure ulcer stage III and IV, falls and trauma, catheter-associated urinary

tract infection, vascular catheter-associated infection, manifestations of poor glycemic control,

surgical site infections (included mediastinitis following CABG, SSI following certain orthopedic

procedures, and SSI following bariatric surgery for obesity), and DVT/PE following certain

orthopedic procedures – we constructed an episode of care file containing all IPPS hospitalizations

in FY 2009 and FY 2010 with at least one of these conditions coded as hospital-acquired (POA

indicator equal to “N” or “U”). In order to allow for a 30-day follow-up period after the initial (or

“index”) hospitalization, we excluded IPPS discharges that occurred on or after September 1, 2010

(approximately 4% of the initial sample).

For each of these HACs, we selected a comparison sample of IPPS hospitalizations that 1)

did not have any of these HACs coded on the hospital claim, 2) did not have any of these HAC-

related conditions coded as present on admission (POA indicators equal to “Y” or “W”), and 3) did

not have any of these HAC-related diagnosis codes as the primary diagnosis on the claim. Since

all hospital patients are potentially at risk for hospital-acquired pressure ulcer or falls and trauma,

we selected a 5% sample of all of the claims that met the above criteria. To look for evidence of

the manifestations of poor glycemic control HAC, we used a comparison group of 5% of IPPS

claims that had a principal diagnosis of diabetes (ICD-9_CM diagnosis codes 250.00 – 250.99),

since diabetic patients would be those at most at risk for poor glycemic control.

The remaining HACs, being specific to particular procedures or surgeries, had comparison

groups selected based on the presence of ICD-9_CM procedure codes on the hospital claim. To

check for under-coding of catheter-associated UTI, we selected all IPPS claims with a urinary

catheter procedure code (57.94 or 57.95); note that this procedure code greatly under-estimates the

actual rate of urinary catheters among hospital patients, as evidenced by the fact that only 5% of

IPPS claims with a catheter-associated UTI have one of the urinary catheter procedure codes. We

looked for evidence of vascular catheter-associated infections among a 25% sample of IPPS index

claims that had a vascular catheter ICD-9_CM procedure code (38.93 or 38.95). For the three SSIs

considered, we used the entire set of claims with the specified surgical procedures as our

comparison sample, and for DVT/PE following certain orthopedic procedures, we took a 50%

sample of all total hip replacement and total knee replacement surgeries to use for the population at

risk for a DVT/PE. The sampling that we chose allowed us to significantly reduce computational

time of these analyses while maintaining large enough comparison samples to find evidence of

HAC under-coding. When presented in the following tables and descriptions, all samples have

been adjusted to reflect 100% of the population; for example, any frequencies for the comparison

sample of diabetic patients have been multiplied by 20 to reflect their size in the entire at-risk

population in the FY 2009 and FY 2010 MedPAR files.

After selecting the IPPS index claims with HACs and the at-risk comparison samples for

each of the HACs, we then used the unique beneficiary identifiers and admission and discharge

dates on the index hospital claims to link physician claims that occurred during the index

hospitalization to the hospital claim. Approximately 2% of index hospital claims had no linked

physician claims. For index hospital claims that were linked to physician claims, we examined the

reported diagnosis codes on both types of claims to look for evidence of HAC-related conditions.

We also looked forward 30 days from the hospital discharge date to identify any hospital

readmissions that occurred for the beneficiaries in our sample, in order to examine readmission

rates for those with and without hospital- and physician-identified HAC-related conditions. This

analysis was limited to the hospital-acquired infections (catheter-associated UTI, vascular

associated-catheter infections, and SSIs) and DVT/PE following certain orthopedic procedures.

For our analysis of post-discharge clinical presentation of infections and DVT/PE, we

linked inpatient and outpatient Medicare claims – hospital, SNF, IRF, LTCH, outpatient, and

physician claims – that occurred within 30 days of the index hospital discharge—and report

frequency of post-discharge reporting of the HAC-related conditions and rates of readmission

within 30 days. We restrict the reporting to only institutional claims. Using the outpatient claims,

we further examined claims with HCPCS codes for antibiotic administration, to examine post-

discharge treatment consistent with HAC-related infections.

5.1.3 Correlation Between Hospital-Acquired Conditions Coded on Hospital

Claims and Hospital-Acquired-Related Conditions Coded on Physician

Claims

One way in which to determine the potential extent of under-coding of HACs on hospital

claims is to compare the diagnosis codes on the hospital claims with the diagnosis codes on

physician claims that were billed for the same patient during the hospitalization. In Table 5-1,

we show the frequencies of the HAC diagnosis codes on the hospital and the physician claims,

for both the HAC populations and the at-risk comparison samples. In the first row, we present

the number of index claims that had neither a hospital nor a physician diagnosis code of a HAC-

related condition. The vast majority of claims in our analysis fall into this category. The second

row shows, for each of the HACs in the analysis, the number of claims where the HAC is not

coded on the index MedPAR hospital claim, but at least one physician claim linked to the index

hospital claim has one of the HAC-related diagnosis codes. This is our set of potentially under-

reported HACs.

Some of these HAC-related conditions identified on the physician claims but not on the

hospital claim may have actually been coded on the hospital claim, but not picked up by the

MedPAR file and used for determining payment. Prior to January 2012, hospitals were not

required to submit claims using the 5010 electronic format, which captures up to 24 secondary

diagnosis codes. The older system captured only 8 secondary diagnosis codes. So while the

HAC-related diagnosis code may have actually been reported on the hospital claim, for the

purposes of analysis using pre-2012 hospital claims data, these HACs go unreported.

Appendix Table 4 provides some evidence that HAC diagnosis codes were reported after

the eighth secondary diagnosis code on the hospital claims. In this table, we show the rates of

HAC coding in FY 2010, when only the first 8 secondary diagnosis codes on the claim were

recorded, compared to the rates of HAC coding in FY 2011 for those hospitals that began using

the 5010 electronic format earlier than January 2012 and could report up to 24 secondary

diagnosis codes. We found that hospitals using the 5010 electronic format represented about

94% of claims in FY 2011. As we would expect, for almost all of the HACs,

pressure ulcer

stages III and IV, catheter associated urinary tract infection(CAUTI), vascular catheter

associated infection(CLABSI), surgical site infections(SSIs), mediastinitis, following coronary

artery bypass graft surgery, following certain orthopedic procedures, following bariatric surgery

for obesity, and manifestations of poor glycemic control, the HAC rates increase when more

diagnosis codes are used in the FY 2011 data. For example, see the Number of Discharges

Identified as a HAC per Thousand for pressure ulcers stage III and IV. In FY 2010, there were

0.14 discharges per thousand with a pressure ulcer HAC, while in FY 2011, among the claims

where up to 24 secondary diagnosis codes could be reported, the rate increased to 0.20.

Interestingly, HAC rates fall for both falls and trauma and DVT/PE following certain orthopedic procedures

between FY 2010 and FY 2011, despite the fact that more diagnosis codes are used in FY 2011. This could

reflect a significant, system-wide reduction in these HACs that is larger than the effect of the increased number

of diagnosis codes. It could also reflect that these HACs have a greater likelihood of facing a payment penalty,

and thus there is a greater incentive for under-reporting of these HACs. From Appendix 4, we see that more than

one quarter of claims with a falls and trauma HAC face a payment penalty, and more than 40% of DVT/PE HAC

claims face a payment penalty.

Therefore, some of the under-reporting we see in Table 5-1 is likely due to the limitation of 8

secondary diagnosis codes.

Table 5-1

Frequency of hospital and physician coding of selected hospital-acquired conditions

Hospital

Diagnosis

of HAC

Physician

Diagnosis

of HAC

Pressure

Ulcer –

Stages III &

Falls and

Trauma

Catheter-

Associated

Urinary

Tract

Infection

(CAUTI)

Vascular

Catheter -

Associated

Infection

(CLABSI)

Manifestation

of Poor

Glycemic

Control

Surgical Site

Infection

(SSI),

Mediastinitis

following

Coronary

Artery

Bypass Graft

(CABG)

Surgical

Site

Infection

(SSI)

following

Certain

Orthopedic

Procedures

Surgical

Site

Infection

Following

Bariatric

Surgery

for

Obesity

Deep Vein

Thrombosis

Pulmonary

Embolism

(DVT/PE)

Following

Certain

Orthopedic

Procedures

No No 17,742

,320 17,392,660 151,795 1,365,324 4,683,940 156,114 203,780 29,679 673,928

No Yes 1,720 3

5,380 117 4,148 14,340 126 849 163 3,622

Yes No 2,508 3,

103 6,175 6,534 655 41 227 24 1,511

Yes Yes 60 7,064 81 4

88 226 28 151 8 3,740

— Tota

l 17,746,608 17,754,207 158,168 1,376,494 4,699,161 156,309 205,007 29,874 682,801

NOTES:

2.6% of MedPAR index admissions with no hospital diagnosis of pressure ulcer, and 2.2% of admissions with a pressure ulcer HAC, were excluded because

they had no linked physician claims during the hospitalization period. 5% sample of hospital claims without the HAC multiplied by 20 to estimate the full

population.

Includes fracture, dislocation, intracranial injury, crushing injury, burn, and other injuries. 2.6% of MedPAR index admissions with no hospital diagnosis of

falls and trauma, and 2.0% of admissions with a falls and trauma HAC, were excluded because they had no linked physician claims during the hospitalization

period. 5% sample of hospital claims without the HAC multiplied by 20 to estimate the full population.

2.4% of MedPAR index admissions with no hospital diagnosis of CAUTI, and 2.0% of admissions with a CAUTI HAC, were excluded because they had no

linked physician claims during the hospitalization period.

2.0% of MedPAR index admissions with no hospital diagnosis of CLABSI, and 2.4% of admissions with a CLABSI HAC, were excluded because they had no

linked physician claims during the hospitalization period. 25% sample of hospital claims with vascular catheter procedure without the HAC multiplied by 4 to

estimate the full population.

(continued)

Includes diabetic ketoacidosis, nonketotic hyperosmolar coma, hypoglycemic coma, secondary diabetes with ketoacidosis, and secondary diabetes with

hyperosmolarity. 2.6% of MedPAR index admissions with no hospital diagnosis of poor glycemic control, and 2.8% of admissions with a poor glycemic

control HAC, were excluded because they had no linked physician claims during the hospitalization period. 5% sample of hospital claims with diabetes without

the HAC multiplied by 20 to estimate the full population.

0.2% of MedPAR index admissions with no hospital diagnosis of mediastinitis, and 0.2% of admissions with a mediastinitis HAC, were excluded because they

had no linked physician claims during the hospitalization period.

Includes spine, neck, shoulder, and elbow surgeries. 2.4% of MedPAR index admissions with no hospital diagnosis of SSI, and 4.5% of admissions with an

SSI following certain orthopedic procedures HAC, were excluded because they had no linked physician claims during the hospitalization period.

Includes laparoscopic gastric bypass, gastroenterostomy, and laparoscopic gastric restrictive surgery. 2.4% of MedPAR index admissions with no hospital

diagnosis of SSI, and 5.9% of admissions with an SSI following bariatric surgery HAC, were excluded because they had no linked physician claims during the

hospitalization period.

Includes total hip replacement and total knee replacement. 2.2% of MedPAR index admissions with no hospital diagnosis of DVT/PE, and 2.3% of admissions

with a DVT/PE HAC, were excluded because they had no linked physician claims during the hospitalization period. 50% sample of orthopedic procedures

without the DVT HAC multiplied by 2 to estimate the full population.

SOURCE: MedPAR hospital claims from FY 2009 and the first eleven months of FY 2010 linked to Medicare Part B physician claims during the

hospitalization.

Additionally, some of these HAC-related conditions from the physician claims could

have been present on admission; the large numbers of physician-identified falls and trauma and

manifestations of poor glycemic control would seem to be more likely to reflect conditions that

were present on admission rather than hospital-acquired. However, it seems less likely that the

infections related to specific procedures or surgeries performed during the index hospitalization

would be present on the admission before the procedure or surgery occurred, meaning that some

of these physician-identified diagnoses could be truly un-reported HACs. For conditions such as

mediastinitis following CABG, the 126 cases of mediastinitis identified only on the physician

claims are likely to be true HACs. Regardless of the source of the under-reporting, the presence

of true HACs among the hospital claims identified as not having HACs has the potential to bias

our analysis of readmissions.

While the numbers of physician-identified HAC-related conditions are generally small

relative to the entire sample examined, they are often large relative to the number of HACs

identified on the hospital claims. For example, from the set of orthopedic surgery claims that did

not have a DVT/PE coded on the hospital claim (the first two rows in the table), only about 0.5%

had a DVT/PE coded on a linked physician claim (3,622 out of 677,550). But when we compare

that to the total number of DVT/PE HAC claims (the last two rows of the table) – 5,251 – it

seems to be a much more significant number. If all of the physician-identified DVT/PE claims

are true HACs, then the total count of DVT/PE HACs across the two years of data would be

about 70% higher, 8,873 instead of 5,251.

The physician claims for mediastinitis also point to potential under-reporting of HACs.

From the MedPAR data, a total of 156,309 Medicare claims for CABG surgery were identified

and linked to at least on physician claim during the period of hospitalization. Of those claims, 69

(.04%) had a diagnosis of mediastinitis on the hospital claim. In addition to these 69 hospital-

reported mediastinitis HACs, there were 126 physician claims linked to hospital claims for

CABG surgeries with a physician diagnosis code for mediastinitis. If all of these 126 physician-

identified cases of mediastinitis were true HACs, then this data suggests that approximately 65%

of mediastinitis HACs were unreported.

The third row of the table counts the claims where the HAC diagnosis code was on the

hospital claim but not on the physician claim, and the fourth row shows the claims where both

hospital and physician claims report the HAC. The rate of agreement between hospital and

physician coding of these HACs is poor. Except for falls and trauma and DVT/PE following

certain orthopedic procedures, there are more hospital-identified HACs without an

accompanying physician diagnosis of the HAC than there are claims with both hospital and

physician claims report the HAC. It would appear that neither hospital claims nor physician

claims are fully coding all of the HACs that are occurring in the hospital.

126 physician-identified mediastinitis cases + 69 mediastinitis HACs = 195 cases of mediastinitis. 126

physician-identified mediastinitis cases ÷ 195 mediastinitis cases = .646.

5.1.4 Under-Coding of Hospital-Acquired Conditions and the Effect on

Readmission Rates

Having identified a set of potentially under-coded HACs, we now examine how these

under-coded HACs could affect measures of readmission rates. In Tables 5-2a through 5-2c,

we present the 7-day, 15-day, and 30-day all-cause readmission rates for hospital claims with and

without HACs, linked to physician claims with and without the HAC-related diagnosis codes.

The rates presented calculate the number of discharges per 100 discharges that had at least one

hospital readmission within the specified window. We limit this analysis of readmission rates to

catheter-associated urinary tract infections, vascular catheter-associated infections, the three

subsets of SSIs, and DVT/PE following certain orthopedic procedures.

For four of the HACs in the 7-day readmission window, and for all of the HACs in the

15-day and 30-day readmission windows, the lowest readmission rates are seen for the claims

with neither a hospital- nor a physician-identified a HAC-related condition. When considering

discharges with a physician-identified (but no hospital-identified) HAC-related condition, the

second row in each series of 4 rows, the readmission rates are typically more similar to those of

the hospital-identified HACs than to those with no HAC-related diagnoses. For example,

consider the final column where we report the readmission rates for claims with certain

orthopedic procedures (total hip replacement and total knee replacement), with and without a

diagnosis of a DVT/PE. The readmission rate within 7 days is 3.3 per 100 discharges for

surgical patients with no DVT/PE-related diagnosis code, 9.2 per 100 discharges for those with

DVT/PE diagnosis code reported on the physician claim only, 7.2 per 100 discharges for those

with DVT/PE diagnosis code reported on the hospital claim only, and 7.1 per 100 discharges for

those with both a hospital and physician reported DVT/PE. The readmission rate for orthopedic

surgery patients with no DVT/PE diagnosis is half or less than half of the readmission rate for

patients with a DVT/PE diagnosis, regardless of whether the DVT/PE is coded on the physician

claim or the hospital claim or both.

Table 5-2a

Readmission rates per 100 discharges for 7-day readmission window for selected hospital-acquired conditions (HACs)

identified from hospital and or physician claims and for comparisons with discharges with no reported HAC

Hospital

Diagnosis of

HAC

Physician

Diagnosis of

HAC

Catheter-

Associated

Urinary Tract

Infection

(CAUTI)

Vascular Catheter

-Associated

Infection

(CLABSI)

Surgical Site

Infection (SSI),

Mediastinitis

following

Coronary Artery

Bypass Graft

(CABG)

Surgical Site

Infection (SSI)

following Certain

Orthopedic

Procedures

Surgical Site

Infection

Following

Bariatric Surgery

for Obesity

Deep Vein

Thrombosis and

Pulmonary

Embolism

(DVT/PE)

Following Certain

Orthopedic

Procedures

No No 9.0 1

0.3 8.5 4.0 5.3 3.3

No Yes 7.7 1

2.4 15.9 9.9 14.7 9.2

Yes No 9.1 1

2.4 14.6 9.3 4.2 7.2

Yes Yes 11.1

13.5 11.1 7.3 12.5 7.1

Table 5-2b

Readmission rates per 100 discharges for 15-day readmission window for selected hospital-acquired conditions (HACs)

identified from hospital and or physician claims and for comparisons with discharges with no reported HAC

Hospital

Diagnosis of

HAC

Physician

Diagnosis of

HAC

Catheter-

Associated

Urinary Tract

Infection

(CAUTI)

Vascular Catheter

-Associated

Infection

(CLABSI)

Surgical Site

Infection (SSI),

Mediastinitis

following

Coronary Artery

Bypass Graft

(CABG)

Surgical Site

Infection (SSI)

following Certain

Orthopedic

Procedures

Surgical Site

Infection

Following

Bariatric Surgery

for Obesity

Deep Vein

Thrombosis and

Pulmonary

Embolism

(DVT/PE)

Following Certain

Orthopedic

Procedures

No No 14.7

16.8 13.6 6.7 8.2 4.9

No Yes 17.1

22.5 22.2 17.7 21.5 12.3

Yes No 15.8

20.8 22.0 13.7 12.5 9.9

Yes Yes 16.0

22.5 18.5 11.9 25.0 10.6

Table 5-2c

Readmission rates per 100 discharges for 30-day readmission window for selected hospital-acquired conditions (HACs)

identified from hospital and or physician claims and for comparisons with discharges with no reported HAC

Hospital

Diagnosis of

HAC

Physician

Diagnosis of

HAC

Catheter-

Associated

Urinary Tract

Infection

(CAUTI)

Vascular Catheter

-Associated

Infection

(CLABSI)

Surgical Site

Infection (SSI),

Mediastinitis

following

Coronary Artery

Bypass Graft

(CABG)

Surgical Site

Infection (SSI)

following Certain

Orthopedic

Procedures

Surgical Site

Infection

Following

Bariatric Surgery

for Obesity

Deep Vein

Thrombosis and

Pulmonary

Embolism

(DVT/PE)

Following Certain

Orthopedic

Procedures

No No 21.7

24.8 18.9 9.6 11.6 6.9

No Yes 28.2

33.2 28.6 24.7 30.1 15.6

Yes No 24.2

30.0 34.1 22.0 20.8 13.2

Yes Yes 18.5

34.4 29.6 22.5 37.5 14.5

NOTES:

See Ta

ble 5-1 for frequencies for each cell, and see notes on Table 5-1 for further details on the sample

SOURCE: MedPAR hospital claims from FY 2009 and the first eleven months of FY 2010 linked to Medicare Part B physician claims during the hospitalization

and linked to MedPAR inpatient claims up to 30 days after the hospital discharge.

5.2 Timing to Clinical Presentation of Selected Conditions

5.2.1 Time to Physician Diagnosis of Conditions During Hospitalization

For catheter-associated urinary tract infections, vascular catheter-associated infections,

SSIs, and DVT/PE following certain orthopedic procedures that were identified on physician

claims linked to the hospital claim, we looked at the dates of the relevant procedures on the

hospital claim and compared them to the dates on the first physician claim with a HAC-related

diagnosis, to show the amount of time that passes between the procedure and the HAC-related

condition. We combined data from the physician-identified HAC-related conditions that were

identified as HACs on the hospital claim with those who were not identified as HACs on the

hospital claim, such that the sample sums the second and fourth rows of Table 5-1. In Table 5-3,

we report the time to a physician diagnosis of a HAC-related condition.

In the second row, we report the number and percent of hospitalizations (for HACs and

those at risk for HACs who had a physician diagnosis of the HAC-related condition) where the

physician diagnosis code of the HAC-related conditions occurred prior to the catheter insertion

or surgery data. These diagnoses occurring prior to procedure could be error in coding, either of

the diagnosis or procedure dates, or they could also represent conditions that were POA. CAUTI

and CLABSI are more likely to occur before the procedure date, with 22.5% of physician-

identified CAUTI diagnoses, and 45.9% of physician-identified CLABSI diagnoses, occurring

prior to the hospital procedure date for the catheter insertion. This could be caused by a catheter

being replaced (and coded) after the discovery of the infection.

Time to physician diagnosis of CAUTI and DVT/PE following certain orthopedic

procedures is typically much faster than the timing to the other HAC-related infections. For

CAUTI, 60% of physician-identified cases occur within 3 days of the catheter insertion

procedure, and 77.5% of DVT/PE cases diagnosed by a physician occur within 3 days of the

orthopedic surgery. In contrast, more than half of two of the SSIs are first diagnosed more than a

week after the relevant surgical procedures. For example, for all cases of mediastinitis following

CABG surgery identified on a physician claim, 35.7 % were first diagnosed between 8 and 15

days after the surgery and 44.8 % were first diagnosed 16 or more days after the surgery.

We also examined the physician specialty type for patients who had a physician-

identified HAC-related condition, whether or not the HAC was coded on the hospital claims, to

see what physician specialties might be potential indicators for the HAC-related conditions. In

Table 5-4, we present the top five physician specialties for the first physician claim linked to the

hospital claim with a HAC-related diagnosis code. Infectious disease specialists were the most

likely type of physician to make the initial diagnosis of the HAC-related condition for four of the

five infection conditions (CAUTI, CLABSI, mediastinitis following CABG, and SSI following

certain orthopedic procedures), accounting for more than 20% of initial diagnoses for these

conditions. For SSI following bariatric procedures, infectious disease specialists first diagnosed

18% of HAC-related conditions.

Table 5-3

Time to presentation of selected conditions from physician claims during initial hospitalization, all claims with a physician

diagnosis of selected hospital-acquired conditions

Time to Presentation

Catheter-

Associated

Urinary Tract

Infection

(CAUTI)

Vascular

Catheter-

Associated

Infection

(CLABSI)

Surgical Site Infection

(SSI), Mediastinitis

following Coronary

Artery Bypass Graft

(CABG)

Surgical Site

Infection (SSI)

following Certain

Orthopedic

Procedures

Surgical Site

Infection

Following

Bariatric Surgery

for Obesity

Deep Vein Thrombosis

or Pulmonary Embolism

(DVT/PE) Following

Certain Orthopedic

Procedures

Hospital claim with procedure

lin

ked to physician claim with

HAC diagnosis codes

Freq.

Percent

120

100.0%

4,442

100.0%

154

100.0%

1,000

100.0%

171

100.0%

7,362

100.0%

Physician diagnosis of HAC-

rela

ted condition prior to date

of procedure

Freq.

Percent

22.5%

2,039

45.9%

7.8%

7.2%

12.3%

292

4.0%

Physician diagnosis of HAC-

rel

ated condition 0 to 3 days

after procedure

Freq.

Percent

60.0%

1,469

33.1%

5.2%

400

40.0%

19.3%

5,708

77.5%

Physician diagnosis of HAC-

condition between 4 and

7 days after procedure

Freq.

Percent

7.5%

334

7.5%

6.5%

219

21.9%

13.8%

1,160

15.8%

Physician diagnosis of HAC-

condition between 8 and

15 days after procedure

Freq.

Percent

8.3%

332

7.5%

35.7%

207

20.7%

36.8%

168

2.3%

(continued)

Table 5-3 (continued)

Time to presentation of selected conditions from physician claims during initial hospitalization, all claims with a physician

diagnosis of selected hospital-acquired conditions

Time to Presentation

Catheter-

Associated

Urinary Tract

Infection

(CAUTI)

Vascular

Catheter-

Associated

Infection

(CLABSI)

Surgical Site Infection

(SSI), Mediastinitis

following Coronary

Artery Bypass Graft

(CABG)

Surgical Site

Infection (SSI)

following Certain

Orthopedic

Procedures

Surgical Site

Infection

Following

Bariatric Surgery

for Obesity

Deep Vein Thrombosis

or Pulmonary Embolism

(DVT/PE) Following

Certain Orthopedic

Procedures

Physician diagnosis of HAC-

condition between 16

days or more after procedure

Freq.

Percent

1.7%

268

6.0%

44.8%

102

10.2%

17.5%

0.5%

NOTES:

1. Ex

cludes 78 CAUTI HAC claims that did not have a urinary catheter procedure coded.

2. Excludes 194 CLABSI HAC claims that did not have a vascular catheter procedure coded.

See notes on Table 5-1 for further details on the sample

SOURCE: MedPAR hospital claims from FY 2009 and the first eleven months of FY 2010 linked to Medicare Part B physician claims during the

hospitalization.

Table 5-4

Physician specialty from the first diagnosis of HAC-related condition, among all hospital claims linked to a physician claim

with a HAC-related diagnosis code

Physician Specialty

Catheter-

Associated

Urinary Tract

Infection (CAUTI)

Vascular

Catheter-

Associated

Infection

(CLABSI)

Surgical Site

Infection (SSI),

Mediastinitis

following Coronary

Artery Bypass Graft

(CABG)

Surgical Site

Infection (SSI)

following Certain

Orthopedic

Procedures

Surgical Site

Infection

Following

Bariatric

Surgery for

Obesity

Deep Vein

Thrombosis or

Pulmonary

Embolism

(DVT/PE)

Following Certain

Orthopedic

Procedures

Number of phys

ician-identified

HAC-related diagnosis linked to

hospital claim

198 4,636 154 1,000 171 7,362

Most common physician

special

ty (Percent)

Infectious disease

specialist (31%)

Infectious disease

specialist (21%)

Infection disease

specialist (24%)

Infectious disease

specialist (24%)

General surgery

(21%)

Diagnostic

radiology (46%)

Second most common physician

special

ty (Percent)

Ambulance

service supplier

(22%)

Diagnostic

radiology (15%)

Plastic surgeons

(14%)

Anesthesiology

(22%)

Diagnostic

radiology (19%)

Internal medicine

(19%)

Third most common physician

special

ty (Percent)

Internal medicine

(17%)

Internal medicine

(14%)

Internal medicine

specialists (12%)

Orthopedic

surgery (15%)

Infectious

disease

specialist (18%)

Vascular surgery

(6%)

Fourth most common physician

special

ty (Percent)

Urology (7%)

General surgery

(13%)

Cardiac surgeons

(9%)

Neurosurgery

(9%)

Anesthesiology

(11%)

Pulmonary disease

(5%)

Fifth most common physician

special

ty (Percent)

Diagnostic

radiology (5%)

Anesthesiology

(8%)

Thoracic surgeons

(8%)

Internal medicine

(6%)

CRNA (7%) Cardiology (5%)

NOTES:

See no

tes on Table 5-1 for further details on the sample

SOURCE: MedPAR hospital claims from FY 2009 and the first eleven months of FY 2010 linked to Medicare Part B physician claims during the

hospitalization.

Internal medicine physician are often the first to diagnose many of the HAC-related

conditions. Urologists are among the top 5 physician types for diagnosing CAUTI, and

orthopedic surgeons are the third common physician specialty to diagnose an SSI following

certain orthopedic procedures. Interestingly, 22% of CAUTI conditions are diagnosed by

ambulance service suppliers. Recall from Table 5-3 that just over 22% of CAUTI diagnoses on

the physician claims occur before the date of the insertion of a urinary catheter on the hospital

claim. It is likely, then, that these infections identified by ambulance service suppliers occurred

in another healthcare setting (or at home) and are coded during ambulance transfer to the

hospital.

For those beneficiaries who had a DVT/PE coded on the physician claims linked to a

hospital claim for certain orthopedic surgeries, the most common first physician specialty type

coded was diagnostic radiology (46%), followed by internal medicine (19%), vascular surgery

(6%), pulmonary disease (5%), and cardiology (5%). It is not surprising that diagnostic

radiology was the most frequently coded first physician specialty and internal medicine was the

next most common physician specialty as patients are referred to diagnostic radiology for testing

to confirm the presence or absence of a DVT or PE, often by an internal medicine physician’s

assessment of a possible DVT or PE.

5.2.2 Post-Discharge Diagnosis of Conditions in Inpatient Settings

In Table 5-5, we consider the sample of index claims that had neither a hospital- nor a

physician-identified hospital-acquired condition, and we look for subsequent inpatient claims

with diagnosis codes that are used to identify hospital-acquired conditions. We do not include

any of the follow-up hospital claims where the relevant diagnosis codes are identified as being

hospital-acquired (POA indicator equal to “N” or “U”), since we are looking specifically for

evidence of the conditions that stem from the procedure or surgery in the index hospital claim.

We also do not include any of the conditions that can occur due to poor care in any inpatient

setting (pressure ulcer stages III or IV, falls and trauma, and manifestations of poor glycemic

control).

The number and percent of inpatient claims (hospital, SNF, LTCH, IRF, and other

inpatient claims) within the 7-day, 15-day, and 30-day window following the index hospital

discharge that have the relevant diagnosis codes for each group are presented. Note that we do

not refer to these as hospital-acquired conditions, because the first evidence from the claims data

that these conditions occurred comes after the index hospital discharge, and thus we cannot

determine if the conditions were actually acquired in the index hospitalization. Instead, we use

the term “HAC-related” to refer to diagnosis codes/conditions that are considered HACs when

they are coded as not present on admission (POA indicator equal to “N” or “U”) in an IPPS

hospital, but can present themselves in any health care setting. Some of these may have been

uncoded HACs in the index hospital setting, but many may be HAC-related conditions whose

clinical presentation did not occur until after the index discharge. As we saw in Table 5-3, the

time to presentation for many of these HAC-related conditions, particularly the surgical site

infections, can be more than 1 or 2 weeks. Short hospital stays increase the likelihood that an

infection will not clinically present and be reported until after the initial hospital discharge.

Table 5-5

Time to presentation of selected conditions on inpatient claims following index hospital discharge with no hospital or physician

diagnosis of selected hospital-acquired conditions

Time to Presentation

Catheter-

Associated

Urinary Tract

Infection

(CAUTI)

Vascular

Catheter-

Associated

Infection

(CLABSI)

Surgical Site Infection

(SSI), Mediastinitis

following Coronary

Artery Bypass Graft

(CABG)

Surgical Site

Infection (SSI)

following Certain

Orthopedic

Procedures

Surgical Site

Infection

Following

Bariatric Surgery

for Obesity

Deep Vein Thrombosis

or Pulmonary Embolism

(DVT/PE) Following

Certain Orthopedic

Procedures

Index hospital claim with

procedu

re code, without HAC

diagnosis codes, and linked to

physician claims without HAC

diagnosis codes

Freq.

Percent

151,795

100%

1,365,324

100%

156,114

100%

203,780

100%

29,679

100%

673,928

100%

7-day window:

any inpatient

claim with HAC-related

diagnosis codes

Freq.

Percent

230

0.2%

5,728

0.4%

132

0.1%

1,620

0.8%

243

0.8%

4,314

0.6%

15-day window: a

ny inpatient

claim with HAC-related

diagnosis codes

Freq.

Percent

436

0.3%

8,088

0.6%

238

0.2%

2,942

1.4%

364

1.2%

5,404

0.8%

30-day window: a

ny inpatient

claim with HAC-related

diagnosis codes

Freq.

Percent

715

0.5%

11,504

0.8%

329

0.2%

4,062

2.0%

465

1.6%

6,718

1.0%

NOTES:

See no

tes on Table 5-1 for further details on the sample

SOURCE: MedPAR hospital claims from FY 2009 and the first eleven months of FY 2010 linked to Medicare Part B physician claims during the hospitalization

and linked to MedPAR inpatient claims up to 30 days after the hospital discharge.

Within the 7-day follow-up window, about half of the claims identified with a HAC-

related condition are readmissions to a hospital, which illustrates the strong relationship between

post-discharge presentation of these conditions and hospital readmissions. Although a

readmission to a hospital is the primary source of post-discharge diagnosis of these HAC-related

conditions overall, DVT/PE is more likely to be found on a SNF claim and vascular catheter-

associated infection is more likely to be found on an LTCH claim.

Even though the rates of these conditions are fairly low within our samples, the

frequencies are often sizable relative to the number of hospital- and physician-identified

conditions. For example, among SSI following certain orthopedic procedures, there were 1,620

cases of surgical site infections identified on inpatient claims that occurred within 7 days of the

discharge from the index hospital claim where the surgery was performed, compared to 378

infections identified on the hospital claims and 849 infections identified on the physician claims

only (see Table 5-1).

As these are cumulative frequencies and rates, the numbers of these conditions identified

on inpatient claims increases as the follow-up window increases. Hospital readmissions are

responsible for an even larger percentage of these HAC-related diagnoses in the 15-day window,

around two thirds of the total, with other post-acute settings playing a less prominent role. After

hospitals, vascular catheter-associated infections, mediastinitis following CABG, and SSI

following bariatric surgery are most likely to be diagnosed in an LTCH, while SSI following

orthopedic surgery is most likely to be diagnosed in an IRF and DVT/PE following certain

orthopedic procedures is most likely to be diagnosed in a SNF.

At the 30-day window, there are more of HAC-related conditions diagnosed on the

follow-up inpatient claims than on the index hospital claims alone, except for catheter-associated

UTI; with the further exception of DVT/PE following certain orthopedic procedures, there are

more of these conditions diagnosed on the follow-up claims than on index hospital and physician

claims combined. While these may or may not have been true HACs, the presence of these

conditions post-discharge is leading to more readmissions and resource use.

5.3 Post-Discharge Use of Outpatient Department Drugs for Infection Treatment

In this portion of the analysis, we used outpatient department (OPD) drug claims in the

30 days after a hospital discharge to examine post-discharge treatment consistent with HAC-

related infections. We looked at four HAC-related conditions – CAUTI, SSI-mediastinitis after

CABG surgery, SSI following bariatric surgery, and SSI following certain orthopedic procedures

– to identify potential under-reporting of these conditions during the hospital period or clinical

presentation of infections after discharge. As with the previous analyses, our initial sample

contained all IPPS claims with one of these four HACs, and also those beneficiaries who did not

have a reported HAC on their hospital claims, but who were identified as being at risk for a HAC

due to the surgery (CABG, bariatric surgery, or orthopedic surgery) or procedure (placement of

urinary catheter) that was performed during the index hospitalization. Our population of interest

was the sample of beneficiaries who had at least one OPD drug claim that contained one of the

Healthcare Common Procedure Coding System (HCPCS) procedure codes for antibiotics known

to treat these infections.

We identified specific medications used to treat the HAC-related infections a priori

through peer-reviewed literature, drug-related websites, drug reference books, and limited

validation with physicians. We also identified the drugs that were used in the OPD data to treat

beneficiaries with a reported HAC and compared them to the a priori list of antibiotics to ensure

that these medications were included in our analysis. It should be noted that the HCPCS

antibiotic codes included in this analysis are specific to antibiotics that are administered

intramuscularly or intravenously. The OPD file provides a count of the number of occurrences

of the antibiotic HCPCS codes at the revenue center on a claim. This means that an individual

beneficiary may have multiple OPD drug claims for a specific antibiotic treatment plan that

includes one or more administrations within the 30 day period. The outcome of interest in our

analysis is the presence of at least one OPD drug claim for one of the specified antibiotics. In

this section of the report, we present a summary of our findings for all of the four studied HACs.

In Table 5-6, we present a summary of OPD drug claims 30 days after hospital

discharge, for beneficiaries with and without one of the studied HACs. The first row gives the

total number of hospital discharges with the HAC or at risk for the HAC. The second and third

rows report the number and percent of hospital discharges with a 30-day post-discharge OPD

drug claim for antibiotic treatment. The fourth row reports the most frequently administered

OPD antibiotics and the percent of the total among those with OPD antibiotics.

Using mediastinitis following CABG as an example, there are a total of 156,612

discharges with CABG procedures (ICD-9_CM procedure codes 36.10-36.19) without a

diagnosis code for mediastinitis on the hospital claim, and 71 discharges with CABG procedures

and a hospital-acquired diagnosis of mediastinitis. Among the 156,612 discharges without a

reported HAC that were at risk for mediastinitis, there are 2,401 discharges (1.5%) with OPD

antibiotic drug claims appropriate for the treatment of mediastinitis. The ratio of at-risk

discharges with an OPD antibiotic drug claim to reported hospital-based mediastinitis discharges

is 34-to-1.

There are at least five potential reasons 2,401 CABG patients are reporting anti-infective

drug use within 30 days after discharge. First, some patients may have already had an infection

of another origin when admitted for the procedure and so are continuing antibiotic treatment after

discharge. Second, some patients may have been infected after discharge from poor adherence to

post-discharge instructions. Third, some patients could have been treated for a post-discharge

infection unrelated to the surgery. Fourth, patients may have contracted a sternal wound

infection that advanced to mediastinitis post-discharge. And fifth, the hospital might not have

reported a mediastinitis infection during the CABG hospitalization.

Table 5-6

Outpatient department (OPD) drug claims within 30 days of hospital discharge, for patients with and without a HAC

OPD Drug Claims

Catheter-

Associated

Urinary Tract

Infection

(CAUTI):

HAC group

Catheter-

Associated

Urinary Tract

Infection

(CAUTI):

At-risk group

Surgical Site

Infection

(SSI),

Mediastinitis

following

Coronary

Artery

Bypass Graft

(CABG):

HAC group

Surgical Site

Infection

(SSI),

Mediastinitis

following

Coronary

Artery

Bypass Graft

(CABG):

At-risk group

Surgical Site

Infection

(SSI)

following

Certain

Orthopedic

Procedures:

HAC group

Surgical Site

Infection

(SSI)

following

Certain

Orthopedic

Procedures:

At-risk group

Surgical Site

Infection

Following

Bariatric

Surgery for

Obesity:

HAC group

Surgical Site

Infection

Following

Bariatric

Surgery for

Obesity:

At-risk group

Total of hospital claims 6,382 155,

726 71 156,612 396 209,569 34 30,573

Number with 30-day pos

discharge OPD drug claim

for antibiotic treatment

124 3,500 2 2,401 16 2,143 0 513

Percent with 30-day pos

discharge OPD drug claim

for antibiotic treatment 1.9% 2.2% 2.8% 1.5% 4.0% 1.0% 0% 1.7%

Most frequently observed

OPD dr

ugs/ Percent of

most frequently observed

OPD drugs

Cefazolin/

28.6%

Vancomycin/

25%

Levofloxacin/

21.4%

Unclassified

drugs/10.7%

Ceftriaxone/

3.6%

Vancomycin/

26.1%

Ceftriaxone/

16.2%

Cefazolin/

12.5%

Gentamicin/

8.0%

Levofloxacin/

7.9%

Unclassified

Drugs/ 7.2%

Daptomycin/

93.3%

Cefazolin/

6.7%

Vancomycin/

37.4%

Daptomycin/

10.7%

Ceftriaxone/

10.1%

Cefazolin/

8.9%

Daptomycin/

41.4%

Vancomycin/

26.7%

Ertapenem/

15.2%

Vancomycin/

31.5%

Daptomycin/

21.3%

Ceftriaxone/

16.5%

Cefazolin/

8.2%

Ertapenem/

4.7%

N/A

Cefazolin/

19.63%

Vancomycin/

15.5%

Unclassified

Drugs/ 13.9%

Ertapenem/

10.7%

Ceftriaxone/

10.3%

SOURCE: Me

dPAR hospital claims from FY 2009 and the first eleven months of FY 2010 linked to Medicare OPD drug claims up to 30 days after the hospital

discharge.

The most frequently observed OPD antibiotic drugs across these four HACs are similar

between both the reported and at-risk groups for each of the conditions as well as across all four

HACs. This is most likely attributed to the bacterial origin of the infection and the specific

antimicrobial therapy as all of these HACs have either a surgical or invasive procedure

component that places patients at higher risk for an infection. For example, vancomycin, a

glycopeptides antibiotic used for prophylaxis as well as treatment of infections caused by Gram-

positive bacteria, is observed in all four at-risk groups and two of the three HAC groups with

OPD antibiotic drug claims. (There are no observed OPD antibiotic drug claims for the 34

reported SSIs following bariatric surgery.) Gram-positive bacteria include such organisms as

Staphylococcus aureus and Staphylococcus epidermis. Vancomycin is also used in the treatment

of methicillin-resistant S. aureus (MRSA) infections that are resistant to other antibiotics. The

use of vancomycin across all HACs is highest among the CABG patients who are at risk for

mediastinitis (37.4% of those with OPD antibiotics received vancomycin) and lowest in the at-

risk group for SSIs following bariatric surgery (15.5%). Daptomycin, a newer Gram-positive

antibiotic, is used to treat skin and soft tissue infections as well as MRSA. Daptomycin claims

are present in both the reported and at-risk groups for mediastinitis and SSIs following

orthopedic surgery. Prescriptions for Daptomycin were not observed in OPD drug claims for

SSI following bariatric surgery nor CAUTI. Clinical trials are ongoing to test the efficacy of

Daptomycin in treating urinary tract infections.

When combined there are a total of 553,319 discharges used in this analysis, with only

6,883 reported HACs. Among the 552,480 discharges that are at risk but without a reported

HAC, there are 8,557 discharges with one or more OPD drug claims for antibiotics appropriate

for treatment of one or more of the hospital-acquired conditions. Some of these 8,557 discharges

could potentially have been true HACs that were not reported by the hospital.

We observed diverse rates of potentially unreported HACs based on outpatient drug

claims for CAUTI and SSIs for orthopedic, bariatric surgery, and CABG. The ratio of at-risk

patient with outpatient antibiotic treatment to reported HACs ranged from .55-to-1 for CAUTI to

34-to-1 for mediastinitis following CABG. Among the studied conditions, CAUTI had the

highest number of discharges with the HAC (6,382) as compared to the other three HACs. The

at-risk group for CAUTI had 3,500 discharges with an OPD antibiotic claim, which was higher

both in frequency terms and in percentage terms than the other three HACs. This difference may

be attributed to the fact that patients with urinary catheters are at higher risk for infection due to

the hospital environment and their own co-morbidities. Another explanation may be that CAUTI

infections are more readily identified when they do occur in the hospital due to the recent

national attention and success in implementing evidence –based procedures to prevent urinary

catheter infections. Or, they manifest clinically earlier than other infections, as we saw in

Table 5-3.

There are a number of limitations to using drug data to identify potentially unreported

healthcare-acquired infections. One is that many of the drugs used to treat a specific HAC, such

as mediastinitis, are also used to treat other infections. For instance Cefazolin may be used to

treat infections related to CABG surgery, but it is also used to treat the other three HACs, as

observed in our analysis. Antibiotics are prescribed based on the presenting bacterial pathogen

and the results of bacterial blood cultures and drug sensitivity testing rather than the specific

condition. A beneficiary with mediastinitis may present with the pathogen staphylococcus

aureus and be treated with a specific antibiotic for that pathogen and a different drug, like

daptomycin for a more serious infection like MRSA. Third, beneficiaries may have more than

one site at risk for infection. For example, a beneficiary undergoing a CABG procedure is at risk

for both a sternal infection, mediastinitis, as well as an infection at the surgical site where the

vein was harvested for the bypass graft (saphenous or internal mammary artery); these infections

may be treated with different antibiotics depending on the presenting pathogen.

OPD drug claims for administered antibiotics for patients without a reported HAC

provide some additional evidence that HACs related to infections are potentially unreported.

Limiting the observation period for a hospital-acquired condition to just the hospitalization

period may be too narrow, given the presence of these conditions in later inpatient claims and

evidence of antibiotic treatment post-discharge.

SECTION 6

SUMMARY AND CONCLUSIONS

In this report, we have investigated the impact of three different HACs on the likelihood

of readmission within 30 days and on the likelihood of discharge to a PAC setting. We used

mixed effect logistic modeling to control for various other characteristics that may explain the

likelihood of readmission and to account for the clustering of discharges within hospitals. The

results suggest a very strong relationship between the presence of a HAC and the likelihood of

both readmission and discharge to a PAC setting.

The relationship between the presence of a HAC and the likelihood of readmission within

30 days varied across the three HACs included in our analyses. For the falls and trauma HAC

and the DVT/PE HAC, the presence of the HAC is associated with a 21 to 23% increase in the

odds of being readmitted within 30 days. The presence of the vascular catheter-associated

infection HAC has an even greater impact on the likelihood of readmission. It is associated with

a 33% increase in the odds of being readmitted within 30 days.

The relationship between the presence of a HAC and the likelihood of discharge to a

PAC setting also varied across the three HACs. The odds of being discharged to a PAC setting

are 2.7 times greater for patients with the falls and trauma HAC, two times greater for patients

with the catheter-associated infection HAC, and 40% greater for patients with the DVT/PE HAC

compared to similar patients without the HACs. Discharge to a PAC setting does not appear to

be mutually exclusive with a hospital readmission within 30 days. In fact, patients who were

discharged to PAC settings were more likely to be readmitted within 30 days. This was true for

patients with each of the HACs as well as for patients in our control groups.

And lastly, we examined the degree to which readmission estimation bias may be present

in the Medicare claims data due to the presence of unreported HACs in the claims, using both

physician claims linked to the hospital claim, and also 30 days of follow-up claims. We found

that significant numbers of HAC-related conditions were reported on physician claims linked to

hospital discharges where no HAC (or POA) was coded. These potentially unreported HACs

could create bias in readmission estimations. There was fairly poor correspondence between

HACs coded on the hospital claims and HAC-related conditions coded on the linked physician

claims.

We examined physician claims linked to the hospital claims for the infection HACs and

DVT/PE following certain orthopedic procedures, and found that, particularly for the SSIs, the

time to clinical presentation of the infection can be more than a week, longer than the typical

hospital stay for some of these surgical procedures. However, physician claims from infectious

disease specialists identifying HAC-related conditions are common across many of the

conditions studied during the hospitalization and could signal the presence of the infection where

it is not otherwise identified on the hospital claim.

Significant numbers of HAC-related conditions are identified on inpatient claims within

the 30 days following a hospitalization with the relevant HAC-related procedure or surgery.

Outpatient claims data for administered antibiotics also points to potentially unreported HACs,

or HAC-related conditions that manifest after the initial hospitalization.

This page intentionally left blank.

BIBLIOGRAPHY

Ashton, C.M., DelJunco, D.J., Souchek, J., et al.: The association between the quality of

inpatient care and early readmission: a meta-analysis of the evidence. Med. Care 35(10):1044-

1059, October 1997.

Averill, R.F., McCullough, E.C., Hughes, J.S., et al.: Redesigning the Medicare Inpatient PPS to

Reduce Payments to Hospitals with High Readmission Rates. Health Care Financ. Rev.

30(4):1-15, Summer 2009.

Baser, O., Supina, D., Sengupta, N., & Kwong, L. (2010). Impact of postoperative venous

thromboembolism on Medicare recipients undergoing total hip replacement or total knee

replacement surgery. American Journal of Health-Systems Pharmacy, 67, 1438-1445.

Beckman, M., Hooper, C., Critichley, S., & Ortel, T. (2010). Venous thromboembolism a

public health concern. American Journal of Preventive Medicine, 38, S495-S501.

Carter, K., (2010). Identifying and managing deep vein thrombosis. Primary Health Care, 20,

30-39.

Encinosa, W.E., and Hellinger, F.J.: The impact of medical errors on ninety-day costs and

outcomes: an examination of surgical patients. HSR: Health Serv. Res. 43(6):2067-2085,

December 2008.

Friedman, B., and Basu, J.: The rate and cost of hospital readmissions for preventable

conditions. Med. Care Res. Rev. 61(2):225-240, June 2004.

Friedman, B., Encinosa, W., Jiang, J.H., et al.: Do patient safety events increase readmissions?

Med. Care 47(5):583-590, May 2009.

Fuji, T., Takahiro, O., Shigeo, N., et al. (2008). Prevention of postoperative venous

thromboembolism in Japanese patients undergoing total hip or knee arthroplasty: two

randomized, double-blind-placebo-controlled studies with three dosage regimens of enoxaparin.

Journal of Orthopaedic Science, 13, 442-451.

Geerts, W., Bergqvist, D., Pineo, G. et al. (2008). Prevention of venous thromboembolism:

American College of Chest Physicians evidence-based clinical practice guidelines (8

edition).Chest, 133 (6), (suppl): 381S-453S.

Goldfield, N.I., McCullough, E.C., Hughes, J.S., et al.: Identifying potentially preventable

readmissions. Health Care Financ. Rev. 30(1):75-91, Fall 2008.

Haines, S. (2010). Improving the quality of care for patients at risk for venous

thromboembolism. American Journal of Health-System Pharmacy, 67, S3-S8.

Herwaldt, L.A., Cullen, J.J., Scholz, D., et al.: A prospective study of outcomes, healthcare

resource utilization, and costs associated with postoperative nosocomial infections. Infect.

Control Hosp. Epidemiol. 27(12):1291-1298, December 2006.

Horwitz, L., Partovian, C., Lin, Z., et al.: Hospital-Wide (All Condition) 30-Day Risk-

Standardized Readmission Measure: Draft Measure Methodology Report. Contract no. HHSM-

500-2008-0025I/HHSM-500-T0001, Modification No. 000005. New Haven, Conn.Yale New

Haven Health Service Corporation/Center for Outcomes Research & Evaluation

(YNHHSC/CORE), August 2011.

Huang S.S., Placzek H., Livingston J., et al. Use of Medicare claims to rank hospitals by

surgical site infection risk following coronary artery bypass graft surgery. Infect. Control Hosp.

Epidemiol. 2011 Aug;32(8):775-83.

Jencks, S.F., Williams, M.V., and Coleman, E.A.: Rehospitalizations among patients in the

Medicare fee-for-service program. N. Engl. J. Med. 360(14):1418-1428, April 2009.

Kandilov, A., McCall, N., Dalton, K., and Miller, R.D. February 2012. Readmissions Due to

Hospital-Acquired Conditions (HACs). Report prepared for Centers for Medicare and Medicaid

Services Office of Research, Development, and Information under Contract no. HHSM-500-

2005-000291. Research Triangle Park, NC: RTI International.

Kim, E., & Bartholomew, J. Venous Thromboembolism. Retrieved November 5, 2010 from

http://www.clevelandclinicmeded.com/medicalpubs/diseasemanagement/cardiology/

Kohut, K. Guide for the Prevention of Mediastinitis Surgical Site Infections Following Cardiac

Surgery. Association for Professionals in Infection Control and Epidemiology. 2008.

Lip, G., & Tse H. (2007). Management of atrial fibrillation. Lancet, 370, 604-618.

McNair, P. D., & Luft, H. S.: Enhancing Medicare’s hospital-acquired conditions policy to

encompass readmissions. Medicare Care & Medicaid Res. Rev. 2012; 2(2), E1-E15.

Morris, D.S., Rohrbach, J., Rogers, M., Sundaram, L.M.T., Sonnad, S., Pascual, J., Sarani, B.,

Reilly, P., and Sims, C.: The Surgical Revolving Door: Risk Factors for Hospital Readmission.

Journal of Surgical Research, January 2011.

National Nosocomial Infections Surveillance (NNIS) System Report, data summary from

January 1992 through June 2004, issued October 2004. Am. J. Infect. Control 2004; 32:470-85.

Rosenbaum, P.R., and Rubin, D.B.: The central role of the propensity score in observational

studies for causal effects. Biometrika 70:41-55, 1983.

Peterson, E., Coombs, L., Ferguson, T., Shroyer, A., DeLong, E., Grover, F.,& Edwards, F.,

(2002). Hospital variability in length of stay after coronary artery bypass surgery: results from

the Society of Thoracic Surgeons’ National Cardiac Database. Annals of Thoracic Surgery, 74,

464-473.

Swenne, C.L., Linholm, C., Borowiec, J., Carlsson, M.: Surgical-site infections within 60 days

of CABG, J. Hosp. Infect. (2004); 57(1):14-24 - Reported .5 - 5% within 60 days post discharge.

Wells, P., Borah, B., Sengupta, N., Supina, S., McDonald, H., & Kwong, L. (2010). Analysis of

venous thromboprophylaxis duration and outcomes in orthopedic patients. The American

Journal of Managed Care, 16, 857-866.

Zierler, B., Wittkowsky, A., Peterson, G., Lee, J., Jacobsen, C., Glenny, R., Wolf, F., Robins, L.,

Mitchell, P., Wolpin, S., Payne, T., Hendrie, P., Han, G., & Oh, H. (2008). Venous

thromboembolism safety toolkit: a systems approach to patient safety. Retrieved June 26, 2012

from http://www.ahrq.gov/downloads/pub/advances2/vol3/Advances-Zierler_81.pdf.