Application of logistic regression and machine learning methods for

Clinical and Experimental Rheumatology 2022

Clinical and Experimental Rheumatology 2023; 41: 330-339.

Application of logistic regression and machine learning

methods for idiopathic inammatory myopathies

malignancy prediction

W. Zhang

, G. Huang

, K. Zheng

, J. Lin

, S. Hu

, S. Zheng

, G. Du

G. Zhang

, C. Bruni

, M. Matucci-Cerinic

5,6

, D.E. Furst

5,7,8

, Y. Wang

Department of Rheumatology and Immunology, Shantou Central Hospital, Shantou, Guangdong, China;

Department of Blood Purication, Shantou Central Hospital, Shantou, Guangdong, China;

Department of Radiology, Shantou Central Hospital, Shantou, Guangdong, China;

Department

of Pathology, Shantou University Medical College, Shantou, Guangdong, China;

Department of

Experimental and Clinical Medicine, Division of Rheumatology, University of Florence, Italy;

Unit of Immunology, Rheumatology, Allergy and Rare diseases (UnIRAR), IRCCS San Raffaele

Hospital, Milan, Italy;

Division of Rheumatology, Department of Medicine, University of California

at Los Angeles, CA, USA;

University of Washington, Seattle, WA, USA.

Abstract

Objective

Malignancy is related to idiopathic inammatory myopathies (IIM) and leads to a poor prognosis. Early prediction

of malignancy is thought to improve the prognosis. However, predictive models have rarely been reported in IIM.

Herein, we aimed to establish and use a machine learning (ML) algorithm to predict the possible risk factors for

malignancy in IIM patients.

Methods

We retrospectively reviewed the medical records of 168 patients diagnosed with IIM in Shantou Central hospital,

from 2013 to 2021. We randomly divided patients into two groups, the training sets (70%) for construction of the

prediction model, and the validation sets (30%) for evaluation of model performance. We constructed six types of ML

algorithms models and the AUC of ROC curves were used to describe the efcacy of the model. Finally, we set up a

web version using the best prediction model to make it more generally available.

Results

According to the multi-variable regression analysis, three predictors were found to be the risk factors to establish

the prediction model, including age, ALT<80U/L, and anti-TIF1-γ, and ILD was found to be a protective factor.

Compared with ve other ML algorithms models, the traditional algorithm logistic regression (LR) model was as

good or better than the other models to predict malignancy in IIM. The AUC of the ROC using LR was 0.900 in the

training set and 0.784 in the validation set. We selected the LR model as the nal prediction model. Accordingly,

a nomogram was constructed using the above four factors. A web version was built and can be visited on the website

or acquired by scanning the QR code.

Conclusion

The LR algorithm appears to be a good predictor of malignancy and may help clinicians screen, evaluate and

follow up high-risk patients with IIM.

Key words

machine learning, malignancy, idiopathic inammatory myopathies

331

Clinical and Experimental Rheumatology 2023

Prediction model for malignancy in IIM / W. Zhang et al.

Weijin Zhang, MD*

Guohai Huang, MD*

Kedi Zheng, MD

Jianqun Lin, MD

Shijian Hu, MD

Shaoyu Zheng, MD

Guangzhou Du, MD

Guohong Zhang, MD

Cosimo Bruni, MD, PhD

Marco Matucci-Cerinic, MD, PhD

Daniel E. Furst, MD

Yukai Wang, MD

*These authors contributed equally

and are to be considered co-rst authors.

Please address correspondence to:

Yukai Wang

Department of Rheumatology

and Immunology,

Shantou Central Hospital,

no. 114 Waima Road,

515041 Shantou,

Guangdong, China.

E-mail: [email protected]

Received on January 26, 2023; accepted

in revised form on February 13, 2023.

ExpErimEntal rhEumatology 2023.

Competing interests: C. Bruni reports

consultancy fees from Boehringer-

Ingelheim and Eli Lilly; grants from

Gruppo Italiano Lotta alla Sclerodermia

(GILS), Fondazione Italiana Ricerca

sull'Artrite (FIRA), European Scleroderma

Trial and Research (EUSTAR), Foundation

for Research in Rheumatology (FOREUM),

Italian Society for Rheumatology (SIR),

Scleroderma Clinical Trials Consortium

(SCTC), Scleroderma Research Foundation

(SRF) outside the submitted work.

M. Matucci-Cerinic has received honoraria

from Eli Lilly, BI and Galapagos.

D.E. Furst has received research support

from Actelion, Amgen, BMS, Prometheus,

Galapagos, GSK, NIH, Novartis, Pzer,

Sano, Roche/Genentech, Horizon and

Emerald; he has received consultancies

from Actelion, Amgen, BMS, Corbus,

Galapagos, Novartis, Pzer and Horizon,

and has been member of speaker's bureau

for CME only.

The other authors have declared

no competing interests.

Introduction

The idiopathic inammatory myopa-

thies (IIM) are a heterogeneous group

of autoimmune rheumatic diseases in-

volving skeletal muscle, the respiratory

system, skin and joints (1-3). The sub-

types of these IIM patients have differ-

ent clinical characteristics, including

proximal muscle weakness, rapid pro-

gressive interstitial lung disease and

severe skin lesions. Of note, one dis-

tinguishing feature of IIM, particularly

dermatomyositis (DM) or polymyositis

(PM), is a signicant association with

a risk of cancer (4). The relationship

between DM and cancer in a patient

with skin rash, muscle weakness and

gastric carcinoma was rst reported by

Stertz in 1916 (5). Since then, a variety

of malignancies have been reported to

be closely related with IIM, involving

multiple systems, including the naso-

pharyngeal, lung, breast and gastroin-

testinal systems (6). In a meta-analysis

of case control and cohort studies in-

cluding 4538 IIM patients in 5 studies,

the overall standardised incidence ratio

(SIR) as a risk for cancer was 4.66 and

1.75 for DM and PM correspondingly

(7). Moreover, the incidence of can-

cer is highest in the rst year after IIM

diagnosis (8), and its prognosis is ex-

tremely poor owing to the complexity

of these two diseases and the discrep-

ancy between tumour and IIM treat-

ment. Hence, it is of great importance

to predict the risk of malignance in IIM

patients as early as possible.

Recently, various studies have demon-

strated the close relationship between

cancer and risk factors with regard to

multiple demographics, clinical and

laboratory features. Patients with IIM

onset after 50 years old and male gen-

der may be at higher risk for developing

cancer (9, 10). In addition, increased

risk of malignancy is associated with

skin involvement, with skin necrosis

as the strongest association (9). Higher

levels of inammatory markers such as

C-reactive protein, erythrocyte sedi-

mentation rate, and creatine kinases

were also often observed in IIM pa-

tients with malignancy (11, 12). Nu-

merous myositis-associated antibodies

have been discovered and veried to

indicate different phenotypes of IIM,

with some indicators strongly suggest-

ing a high risk of cancer. Anti-p155/140

(anti-TIF1-γ) is associated with the

highest positive rate in patients with

cancer-associated myositis and this

is the predominant diagnostic sero-

logical indicator for malignancy (13).

This relationship between TIF1-γ and

cancer-associated myositis is so tight

with odds ratios reaching as high as 23

(95% CI 5.23-101.2) (14). Recent stud-

ies also showed a higher prevalence of

malignancy in IIM patients with anti-

nuclear matrix proteins (NXP)-2 and

anti-3-hydroxy-3-methyglutaryl-coen-

zyme A reductase (HMGCR) antibod-

ies. To date, a quantitative predictive

model has rarely been developed to

predict the risk for malignancy in IIM

patients. A nomogram risk prediction

model by Zhong et al. (15) showed that

patients older than 50-year-old, dys-

phagia, refractory itching and elevat-

ed creatine kinase were risk factors,

while interstitial lung disease was a

protective factor for dermatomyositis-

related-malignancy, with an area under

curve (AUC) of 0.756. However, this

model did not incorporate TIF1-γ and

only applied to patients with DM.

Notably, machine learning (ML) algo-

rithms have been widely utilised in re-

cent years to develop predictive models

which appear to have better predictive

ability than the traditional regression

approaches (16). In this retrospective,

case-control study, we aimed to use

machine learning to predict and com-

pare algorithms to establish the best

risk factors algorithm for malignancy

in IIM patients.

Materials and methods

Patients

We retrospectively reviewed the medi-

cal records of patients diagnosed with

IIM in Shantou Central Hospital, Chi-

na, from 2013 to 2021. We included

168 patients after excluding 1 patient

with too much missing information.

This study was approved by the Shan-

tou Central Hospital Ethics Commit-

tee (no. 2022-037). Patients included

into this study met the classication

criteria for IIM (1), including dermato-

myositis (DM), polymyositis (PM),

immune-mediated necrotising myopa-

332

Clinical and Experimental Rheumatology 2023

Prediction model for malignancy in IIM / W. Zhang et al.

thy (IMNM), anti-synthetase syndrome

(ASS), and inclusion body myositis

(IBD). Exclusion criteria were: 1) ab-

sence of complete clinical data; 2) un-

conrmed diagnosis of IIM; 3) diagno-

sis of hepatitis.

Data collection

The following data were collected:

(i) baseline information including age

and gender; (ii) clinical symptoms in-

volving muscle weakness, myalgia,

arthralgia, rash (typical skin involve-

ment of Gottron’s rash, Gottron’s sign,

mechanical hand, heliotrope rash, V-

neck sign, shawl sign and holster sign),

pruritus, dry mouth and dry eye, dys-

phagia, respiratory syndrome, fever,

oedema, cutaneous ulcer, and Raynaud

phenomenon; (iii) clinical signs includ-

ing rash. For high-risk IIM patients, es-

pecially those with several risk factors

including the elderly, DM, dysphagia,

tumour markers positivity and TIF-1γ

positivity, patients were required to be

screened for malignancy through PET/

CT, as well as gastrointestinal endo-

scope if digestive symptoms occurred

or with their consent. Otherwise, espe-

cially those with protective factors in-

cluding interstitial lung disease (ILD)

and negative TIF-1γ antibody, age-

appropriate screening, including naso-

pharyngeal MR, chest CT, abdominal

CT, gastrointestinal endoscope, breast

ultrasound and thyroid ultrasound were

performed as clinically indicated; (iv)

laboratory data including white blood

cell (WBC), lymphocyte (LY), alanine

transaminase (ALT), aspartate ami-

notransferase (AST), creatinine (Cr),

blood urea nitrogen (BUN), lactic de-

hydrogenase (LDH), creatine kinase

(CK), D-dimer, C-reaction protein

(CRP), erythrocyte sedimentation rate

(ESR), ferritin, carcino-embryonic an-

tigen (CEA), alpha fetoprotein (AFP),

carbohydrate antigen 199 (CA199),

carbohydrate antigen 125 (CA125),

complement 3 (C3), complement 4

(C4), antinuclear antibody (ANA) and

myositis antibody prole.

Statistical analysis

Statistical analysis was performed us-

ing R (v. 4.05) and SPSS 22.0 (IBM,

USA) software. Continuous variables

were expressed as mean±SD and were

analysed by Student’s t-test when

they were normally distributed. Vari-

ables were analysed by nonparametric

methods and described using medians

(Q1, Q3) when their distribution was

skewed or kurtotic. Categorical vari-

ables were analysed using χ

. When

univariate analysis revealed variables

with a p<0.05, they were included in

the predictive models and p>0.1 as the

criterion for removing variables.

Odds ratios (ORs) and 95% condence

intervals (95% CIs) were calculated for

all potential predictors of malignancy

when using multivariate regression.

When the regression showed a variable

to have a p<0.10, it was included in the

prediction model.

The R packages of “glmnet”, “rms”,

“caret”, “rpart”, “partykit”, “e1071”,

“MASS”, “randomForest”, “xgboost”,

and “neuralnet” were used to establish

the prediction model of ML algorithms.

The R packages of “pROC” and “rmda”

were used to validate the prediction

ability of the model. The R packages of

“corrplot” and “ggcorplot” were used

to establish the heat map. The R pack-

age of “shiny” and “shinyPredict” was

used to establish the web application.

The R packages of “ingredients” and

“DALEX” were used to show the rela-

tive importance of variables of predic-

tion model.

In this study, we randomly split pa-

tients into two groups, namely the

training sets (70%) for construction of

prediction model, and the validation

sets (30%) for evaluation of model per-

formance. We constructed six types of

ML algorithms models, i.e. Logistic re-

gression (LR), Support vector machine

(SVM), random forest (RF), Classi-

cation and regression tree (CART),

Extreme gradient boosting (XGBoost),

and Neural network (NNET). Then we

used the area under curve (AUC) of

the receiver operating characteristic

(ROC) curve to evaluate and compare

the predictive ability of the models in

the training and validation sets. The

value of the AUC of the ROC curve

was used to describe the efcacy of the

model. Finally, we set up the web ver-

sion using the best prediction model.

The prediction probability of malig-

nancy in IIM can be easily calculated

and displayed on the website after in-

putting clinical features.

Results

Population characteristics

We identied 168 patients with IIM di-

agnosed between 2013 and 2021. Twen-

Table I. Demographics, subtypes and ma-

lignancy distribution in IIM patients.

n =168

Age (years, IQR) 56.0 (44.0, 64.8)

Male n (%) 50 (29.8)

Diagnosis

DM n (%) 86 (51.2)

PM n (%) 40 (23.8)

ASS n (%) 29 (17.2)

IMNM n (%) 13 (7.7)

Malignancy (n=37)

Nasopharyngeal cancer n (%) 11 (29.7)

Breast cancer n (%) 10 (27.0)

Lung cancer n (%) 7 (18.9)

Oesophagus cancer n (%) 5 (13.5)

Cervical adenocarcinoma n (%) 1 (2.7)

Ovarian cancer n (%) 1 (2.7)

Mediastinum cancer n (%) 1 (2.7)

Multiple cancers n (%) 1 (2.7)

Time relationship between tumourigenesis

and disease diagnosis

simultaneous n (%) 20 (54.1)

before n (%) 5 (13.5)

after n (%) 12 (32.4)

DM: dermatomyositis, PM: polymyositis, ASS:

anti-synthetase syndrome, IMNM: immune-me-

diated necrotic myopathy.

Table II. Prevalence rate of malignancy in different subtypes of IIM patients.

Category DM PM ASS IMNM p

n 86 40 29 13

male n (%) 24 (27.9) 17 (42.5) 4 (13.8) 5 (38.5) 0.055

age (year) 56.0 (44.0, 63.0) 57.0 (41.3, 63.8) 58.0 (49.5, 66.5) 41.0 (31.5, 61.5) 0.137

malignancy n (%) 27 (31.4) 4 (10.0) 6(20.7) 0 0.002*

DM: dermatomyositis, PM: polymyositis, ASS: anti-synthetase syndrome, IMNM: immune-mediated

necrotic myopathy. *p<0.05.

333

Clinical and Experimental Rheumatology 2023

Prediction model for malignancy in IIM / W. Zhang et al.

ty-nine and eight tenths of them are

male. Median age at IIM diagnosis was

56.0 (44.0, 64.8) (Table I). Eighty-six

patients (51.2%) were classied as DM

(with 31.4% malignancy), 40 patients

(23.8%) as PM (with 10% malignancy)

and 29 patients (17.2%) as ASS (with

20.7% malignancy) the rest (7.7%) as

other IIM patients (Table II). Among

these 168 patients, 37 patients had a

malignancy (the Malignancy group).

The top three malignant tumours were

nasopharyngeal cancer (29.7%), breast

cancer (27%) and lung cancer (18.9%).

The remaining 131 patients were des-

ignated as the Non-Malignancy group.

When contrasting these two groups, we

found that patients with malignancy

were statistically signicantly older

and more frequently diagnosed as DM

(p<0.05 for both). Clinically, patients

with the following characteristics were

more likely to develop malignancies:

dysphagia (p=0.021), Gottron’s sign

(p=0.011), V-neck sign (p<0.001), and

shawl sign (p=0.016). In contrast, the

following characteristics were associ-

ated with a lower likelihood of malig-

nancy: arthralgia (p=0.011), respira-

tory involvement (p=0.04), and ILD

(p=0.003). There were no differences in

gender, muscle weakness, myasthenia,

myalgia, pruritus, dry mouth and dry

eye, fever, oedema, cutaneous ulcer,

Raynaud’s phenomenon, mechanical

hand, heliotrope rash, and holster sign

(Table III).

Among the laboratory data, the Ma-

lignancy group had statistically high-

er likelihood of a positive TIF1-γ

(p<0.001) and a lower ALT (p=0.011).

Categorical variables of ALT level were

adopted because it did not meet with

linear correlation, and ALT<80U/L

was selected as the threshold. The two

groups did not differ in term of WBC,

LY, AST, Cr, BUN, LDH, CK, D-dimer,

CRP, ESR, ferritin, CEA, AFP, CA199,

CA125, C3, C4 and the rest of the my-

ositis antibody prole (Table IV and V).

Risk factors for malignancy

among IIM patients

In the univariable analysis, age, DM,

arthralgia, dysphagia, respiratory in-

volvement, Gottron’s sign, V-neck

sign, shawl sign, ILD, ALT<80U/L,

CEA>2.0 ng/ml and anti-TIF1-γ were

statistically signicantly different and

were included in the multi-variable

regression analysis (Table VI). The

relationships among variables are il-

lustrated by a heat map analysis (Fig.

1). Based on the multivariable analysis,

the only independent risk factors that

Table III. Patients’ characteristics and symptoms in the malignancy group and non-malig-

nancy group. Data are expressed with interquartile range (Q1, Q3) if the distribution was

abnormal, and otherwise with mean ± SD for continuous data. For categorical variables,

data are expressed with number (%).

Malignancy Non-malignancy Statistics p

n=37 n=131

Age (years, IQR) 59.0 (51.5, 65.0) 55.0 (39.0, 63.0) 2.100 0.036

Male n (%) 15 (40.5) 35 (26.7) 2.637 0.104

DM n (%) 27 (73.0) 59 (45.0) 9.011 0.003

Myasthenia n (%) 28 (75.7) 87 (66.4) 1.146 0.284

Myalgia n (%) 15 (40.5) 64 (48.9) 0.801 0.371

Arthralgia n (%) 2 (5.4) 32 (24.4) 6.467 0.011

Pruritus n (%) 10 (27.0) 26 (19.8) 0.883 0.347

Dry mouth and dry eye n (%) 3 (8.1) 12 (9.2) 0.039 0.843

Dysphagia n (%) 13 (35.1) 23 (17.6) 5.295 0.021

Respiratory involvement n (%) 7 (24.3) 54 (41.2) 3.515 0.040

Fever n (%) 0 11 (8.4) 2.094 0.148

Oedema n (%) 3 (8.1) 4 (3.1) 0.797 0.372

Cutaneous ulcer n (%) 2 (5.4) 5 (3.8) 0.000 1.000

Raynaud phenomenon n (%) 0 6 (4.6) 0.679 0.410

Gottron’s rash n (%) 13 (35.1) 33 (25.2) 1.435 0.231

Gottron’s sign n (%) 18 (48.6) 35 (26.7) 6.426 0.011

Mechanical hand n (%) 4 (10.8) 16 (12.2) 0.000 1.000

Heliotrope rash n (%) 18 (48.6) 42 (32.1) 3.458 0.063

V-neck sign n (%) 18 (48.6) 23 (17.6) 15.117 <0.001

Shawl sign n (%) 11 (29.7) 17 (13.0) 5.830 0.016

Holster sign n (%) 3 (8.1) 6 (4.6) 0.708 0.400

ILD n (%) 11 (29.7) 75 (57.3) 8.747 0.003

DM: dermatomyositis; ILD: interstitial lung disease. *p<0.05.

Table IV. Comparison of laboratory data between malignancy group and non-malignancy

group.

Malignancy Non-malignancy Statistics p

n=37 n=131

WBC 10

/L 7.68 ± 3.14 8.71 ± 4.27 -1.369 0.173

LY 1 0

/L 1.2 (0.6, 1.7) 1.3 (0.9, 2.0) -1.188 0.235

ALB g/L n=166 35.03 ± 5.75 35.2 ± 6.4 -0.102 0.919

ALT U/L 35.0 (19.0, 69.0) 68.0 (29.0, 160.0) -2.540 0.011*

AST U/L 62.0 (27.0, 128.5) 88.0 (37.0, 190.0) -1.678 0.093

Cr umol/L 50.7 (45.6, 70.4) 55.0 (42.9, 66.7) 0.239 0.811

BUN mmol/L 4.3 (3.6, 5.5) 4.5 (3.5, 6.0) -0.136 0.892

LDH U/L n=163 405.0 (318.0, 705.5) 459.0 (331.0, 777.0) -0.750 0.453

CK U/L n=165 724.0 (156.3, 2489.0) 696.0 (171.0, 4830.0) -0.572 0.567

D-dimer ug/L n=138 880.0 (430.0, 2020.0) 696.0 (432.5, 1662.5) 0.207 0.836

CRP mg/L n=159 5.5 (3.0, 12.3) 6.4 (2.3, 14.6) 0.465 0.642

ESR mm/h n=147 14.0 (9.0, 29.0) 22.0 (10.0, 45.0) -1.928 0.054

Ferritin ng/mL n=110 704.4 (382.0, 1033.3) 583.3 (279.5, 1217.0) 0.960 0.337

CEA ng/mL n=161 2.4 (1.6, 4.5) 1.8 (1.2, 3.1) 1.889 0.059

AFP IU/ml n=160 2.1 (1.5, 3.0) 1.9 (1.4, 3.2) 0.713 0.476

CA199 U/mL n=131 7.5 (5.5, 16.2) 9.3 (5.4, 18.4) -0.628 0.530

CA125 U/mL n=130 10.2 (7.2, 15.7) 11.2 (7.3, 18.6) -0.691 0.489

C3 g/L n=146 0.91 ± 0.17 0.88 ± 0.22 0.667 0.661

C4 g/L n=146 0.23 ± 0.07 0.22 ± 0.08 0.439 0.914

WBC: white blood cell; LY: lymphocyte; ALB: albumin; ALT: alanine transaminase; AST: aspartate

aminotransferase; Cr: creatinine; BUN: blood urea nitrogen; LDH: lactic dehydrogenase; CK: creatine

kinase; CRP: C-reactive protein; ESR: erythrocyte sedimentation rate; CEA: carcino-embryonic anti-

gen; AFP: alpha fetoprotein; CA199: carbohydrate antigen 199; CA125: carbohydrate antigen 125; C3:

complement 3; C4: complement 4. *p<0.05.

334

Clinical and Experimental Rheumatology 2023

Prediction model for malignancy in IIM / W. Zhang et al.

predicted malignancy were age (per

ten years, OR=1.612; 95% CI [0.997,

2.607]; p=0.052- included despite be-

ing slightly greater than 0.05 based on

the literature (9, 15, 17) and clinical

judgement), ALT<80 U/L (OR=11.175;

95% CI [1.367, 91.355]; p=0.024),

and anti-TIF1-γ (OR=4.963; 95% CI

[1.193, 20.642]; p=0.028). Intersti-

tial lung disease (OR=0.193; 95% CI

[0.058, 0.643]; p=0.007) was a negative

predictive factor.

The use of machine learning

algorithms

The performance of six different ML

algorithm models as predictors of ma-

lignancy in training sets and validation

sets are shown and compared in Fig-

ures 2 and 3 and Table VII and VIII,

respectively. The results showed that

the NNET and RF model possessed ex-

cellent predictive ability in the training

set, but did not do well in the validation

set. The traditional Logistic Regression

algorithm model did as well or better

than other machine learning algorithms

in predicting malignancy of IIM (AUC

of ROC was 0.900 in the training set

and 0.784 in the validation set) (Table

VII and VIII). Therefore, we selected

the LR model as the nal prediction

model.

The relative importance of

variables in prediction models

The relative importance of variables

in each prediction model is shown in

Figure 4. Although the importance of

different variables in different models

was variable, and differences among

the variables were small, anti-TIF1-γ

and low ALT were numerically the two

most importance positive predictive

variables among the 6 models and ILD

was most useful as negative predictor.

Prediction and validation of

model for malignancy in IIM patients

On the basis of the four factors selected

by multivariate analysis (Age, TIF1-γ,

ALT and ILD), for the convenience of

clinical application, we constructed a

nomogram to predict the probability

of malignancy in IIM patients (Fig. 5).

One determines the numerical value of

each factor based on the vertical line

intersection between the variable and

the point axis, and then adds all vari-

able points to calculate the total risk

score, with each risk score correspond-

ing to the probability of malignancy.

The usefulness of this nomogram will

need to be tested in several other data-

sets but it seemed useful as applied in

our patients.

The web version of model

For extending the application of the

model established in this study, a web

version was built and can be visited

on the website https://hgh-163.shin-

yapps.io/DynNomapp/ or acquired by

scanning the QR code (Fig. 6) with

a smartphone. The algorithm deter-

Table VI. Univariate and multivariable logistic regression analysis of risk factors for idio-

pathic inammatory myopathies. (Step backward, Wald test, entry condition 0.05, deletion

condition 0.10).

Factors Univariate Multivariate

p OR(95%CI) p OR (95%CI)

Age (per 10 year) 0.009 1.456 (1.099-1.931) 0.052* 1.612 (0.997-2.607)

DM 0.004 3.295 (1.476-7.355)

Arthralgia 0.022 0.177 (0.040-0.776)

Dysphagia 0.024 2.543 (1.130-5.725)

Respiratory involvement 0.065 0.458 (0.200-1.049)

Gottron’s sign 0.013 2.598 (1.225-5.512)

V-neck sign <0.001 4.449 (2.027-9.765)

Shawl sign 0.019 2.837 (1.189-6.771)

ILD 0.004 0.316 (0.144-0.693) 0.007* 0.193 (0.058-0.643)

ALT <80U/L 0.006 3.739 (1.460-9.577) 0.024* 11.175 (1.367-91.355)

CEA >2.0 ng/ml 0.031 2.316 (1.081-4.964)

TIF1γ <0.001 15.781 (4.415-56.410) 0.028* 4.963 (1.193-20.642)

DM: dermatomyositis; ILD: interstitial lung disease; ALT: alanine transaminase; CEA: carcino-embry-

onic antigen; TIF1γ: transcription intermediary factor 1γ. *p<0.10.

Table V. Comparison of myositis antibody prole between malignancy group and non-

malignancy group.

Malignancy Non-malignancy Statistics p

n=37 n=131

ANA (n=162) 21 (60.0) 69 (54.3) 0.357 0.550

MDA5 1 (3.8) 19 (17.9) 2.217 0.137

TIF1γ 10 (38.5) 4 (3.8) 22.965 <0.001

NXP2 0 9 (8.5) 1.221 0.269

SAE1 1 (3.8) 1 (0.9) 0.356

Mi-2 0 4 (3.8) 0.585

ARS

EJ 0 4 (3.8) 0.585

OJ 1 (3.8) 1 (0.9) 0.356

PL-7 2 (7.7) 4 (3.8) 0.112 0.738

PL-12 0 1 (0.9) 1.000

Jo-1 (n=161) 4 (11.4) 23 (18.3) 0.914 0.339

HA 0 1 (0.9) 1.000

SRP 0 11 (10.4) 1.767 0.121

HMGCR 0 1 (0.9) 1.000

Cn1a 0 2 (1.9) 1.000

PMSCL75 0 2 (1.9) 1.000

KU 1 (3.8) 1 (0.9) 0.356

RNA-PIII 0 1 (0.9) 1.000

Th/To 0 2 (1.9) 1.000

Ro-52 (n=161) 16 (45.7) 56 (44.4) 0.018 0.894

ANA: antinuclear antibody; MDA5: melanoma differentiation-associated gene 5; TIF1γ: transcription

intermediary factor 1γ; NXP2: nuclear matrix protein 2; SAE1: small ubiquitin-like modier 1; ARS:

aminoacyl tRNA synthetases; EJ: glycyl; OJ: isoleucyl; PL-7: threonyl; PL-12: alanyl; Jo-1: histidyl;

HA: tyrosyl; SRP: signal recognition particle; HMGCR: 3-hydroxy 3-methylutaryl coenzyme A re-

ductase; Cn1a: cytoplasmic 5’ nucleotidase 1A; PMSCL75: polymyositis-scleroderma 75; RNA-PIII:

RNA polymerase III. *p<0.05.

335

Clinical and Experimental Rheumatology 2023

Prediction model for malignancy in IIM / W. Zhang et al.

mines the predicted probability of

malignancy by inputting the clinical

characteristics.

Discussion

Malignancy may be disguised as in-

ammatory myositis, and myositis may

be a manifestation of the paraneoplas-

tic syndrome of malignancy. Thus, the

rashes of IIM may be a skin window

for recognising malignancy. About one-

third of patients with myositis develop

a malignancy within 3 years of “IIM”

diagnosis, while patients with IIM have

the highest risk of developing malig-

nancy within 1 year of diagnosis (18).

Further, malignancy remains one of the

leading causes of death in IIM patients

(19, 20). Thus, it is of great importance

to be able to predict malignancy in IIM

patients, and as early as possible, so

that treatment of the malignancy can

begin quickly, thus increasing the prob-

ability of a good outcome.

The aim of our work was to establish,

in a retrospective, case-control study,

a potentially clinically useful, multi-

variate risk prediction model for malig-

nancy in IIM patients. After examining

and comparing 6 machine learning al-

gorithms, the logistic regression model

remained at least as good as and per-

haps slightly better than the machine

learning algorithms. Based on LR, four

factors were selected as the nal com-

ponents in the model to predict malig-

Fig. 1. The relationship between different variables and malignancy.

Mali: malignant; Age: age per ten years; DM: dermatomyositis; Arth: arthralgia; Rash: rash; dysp: dysphagia; Resp: respiratory involvement; Gott: Gottron’s

sign; V nec: V-neck sign; Shaw: shawl sign; ILD: interstitial lung disease; ALT: alanine transaminase <80U/L; CEA: carcino-embryonic antigen >2.0 ng/

ml; TIF1γ: transcription intermediary factor 1γ.

The Pearson coefcient value shown on the left of the heat map. On the right of the heat map, the red ball indicates negative correlation, and the blue ball

indicates positive correlation.

The size and colour depth is positively proportional to the correlation coefcient. *p<0.05, **p<0.01.

336

Clinical and Experimental Rheumatology 2023

Prediction model for malignancy in IIM / W. Zhang et al.

nancy in IIM. The presence of ALT <80

U/L, increasing age and anti-TIF1-γ

predicted increased probability of ma-

lignancy, while the presence of ILD

was a protective factor.

Although age was a borderline statis-

tically signicant factor in LR, it con-

tributed to the other algorithms and age

has been associated with malignancy

in the literature. A meta-analysis in-

corporating 380 IIM patients and 1575

controls in 20 studies showed that older

age inuences susceptibility to cancer

(9). Another study reported that the

median age of DM patients with can-

cer was older than those without cancer

(17). In another study, age >50 was se-

lected as one of the factors to construct

a nomogram for predicting malignancy

in dermatomyositis patients (15). In our

study, too, the malignancy group was

older than the non-malignancy group,

with the median age 59.00 vs. 55.00,

p=0.036. And, of course, increasing

age is associated with malignancies in

general. Thus, systemic cancer screen-

ing is strongly recommended for IIM

patients older than 50 years old.

Most previous studies focused on DM

associated malignancy, while we ex-

amined the subtypes of IIM- DM, PM,

IMNM, ASS and IBM. We found that

the prevalence of malignancy in pa-

tients with DM was associated with

the highest risk of cancer-31.4% in

our cohort, which is within the 13% to

42% range found in the literature (7,

21, 22). The V-neck sign, shawl sign

and Gottron’s sign, which are typical

signs of DM, also predicted malig-

nancy (p<0.05). Non-statistical differ-

ence in Gottron’s rash, holster sign and

mechanical hand may be due to small

sample size in this study and weaker

correlation with malignancy. Thus, a

larger and prospective study should be

carried out for verication of different

rashes in malignancy prediction.

We found a relatively high prevalence

of cancer in ASS, with 6 malignan-

cies in 29 patients. This is the second

most common prevalence after DM.

The literature reports this relationship

rarely, perhaps because ASS is a rela-

tively rare disease. The literature re-

ports a prevalence of up to 16.6% and

our prevalence was 20.7%, in the same

general range as the literature (23).

Malignancy was found in 10.0% of PM

patients, ranking the third highest risk,

which was also consistent with the re-

ported risk range of 3% to 18% (21, 24).

Among the 13 patients with IMNM in our

cohort, no cancer was found and IMNM

was not a risk factor for malignancy. Of

course, the very few patients with IMNM

make any predictive algorithm suspect.

The lack of malignancy in these patients

is supported by the literature, as extra-

muscular involvement is rare in IMNM.

The two serological markers of IMNM,

Fig. 2. ROC curve analysis of different machine algorithms prediction model for malignancy of

idiopathic inammatory myopathies in the training set.

SVM: support vector machine; LR: logistic regression; XGBoost: Extreme Gradient Boosting; CART:

classication and regression tree; NNET: neural network; RF: random forest.

Fig. 3. ROC curve analysis of different machine algorithms prediction model for malignancy of

idiopathic inammatory myopathies in the validation set.

NNET: neural network; XGBoost: Extreme Gradient Boosting; RF: random forest; SVM: support

vector machine; CART: classication and regression tree; LR: logistic regression.

337

Clinical and Experimental Rheumatology 2023

Prediction model for malignancy in IIM / W. Zhang et al.

anti-SRP antibody and anti-HMGCR an-

tibody, were not associated with malig-

nancy in a meta-analysis (25) although

one author did nd such a relationship

(26). More research clearly needs to be

done in this area.

It was interesting to examine the use-

fulness of antibodies when predicting

malignancies in IIM. The presence of

TIF1-γ (155-kDa) increased the risk

of a malignancy vefold, and it was an

important variable as a predictor in all

models. This nding is supported by

the literature. A meta-analysis regard-

ing the usefulness of TIF1-γ in DM

showed that 80% of myositis patients

with malignancy tested positive for this

antibody, and 90% of patients without

malignancy tested negative, indicating

high sensitivity and specicity (27). Al-

though the sensitivity and specicity in

our study were not as high as the above

(31.25% and 71.42% respectively), it

might be a simple screening tool which

would alert the clinician that a given

patient needs to be closely followed.

Another well-recognised cancer-as-

sociated antibody is NXP-2, which is

linked with muscle weakness and ele-

vated CK (28). Ichimura et al. (29) col-

lected 445 cases of DM and 62 cases of

PM, of whom 7 (1.6%) and 1 (1.6%)

tested positive in NXP-2, respectively,

although another study found no differ-

ence in NXP2 positivity among those

with or without cancer (30). Of the 8

patients described by Ichimura et al., 3

cases (37.5%) developed visceral ma-

lignancies within 3 years of IIM diag-

nosis, indicating that NXP-2 might be

a marker of increased cancer risk in

IIM. This hypothesis was supported by

Fiorentino et al. (31). Thus far, no can-

cer has been observed in our 9 NXP2

positive patients, with follow-up of up

to 8 years. Larger studies are needed to

determine the usefulness of NXP2.

In our study, ILD was identied as a

protective factor for malignancy in pa-

tients with IIM, which was consistent

with the medical literature (32, 33).

According to a retrospective study by

Zhong et al., there is a negative corre-

lation between ILD and tumours in DM

patients, with an OR value as low as

Table VII. Predictive performance of different machine algorithms model for malignancy

of idiopathic inammatory myopathies in the training set.

Models Accuracy Sensitivity Specicity AUROC

LR 0.867 0.875 0.866 0.900

CART 0.856 0.727 0.873 0.910

SVM 0.856 0.727 0.873 0.870

RF 0.867 0.800 0.875 0.963

NNET 0.879 0.722 0.918 0.923

XGB 0.878 0.818 0.886 0.904

LR: logistic regression; CART: classication and regression tree; SVM: support vector machine; RF:

random forest; NNET: neural network; XGB: Extreme Gradient Boosting.

Fig. 4. Relative importance of each variable in different machine algorithms prediction model.

LR: logistic regression (A); SVM: support vector machine (B); CART: classication and regression tree (C); RF: random forest (D); XGBoost: Extreme

Gradient Boosting (E); NN: neural network (F).

Table VIII. Predictive performance of different machine algorithms model for malignancy

of idiopathic inammatory myopathies in the validation set.

Models Accuracy Sensitivity Specicity AUROC

LR 0.784 0.500 0.818 0.784

CART 0.811 0.667 0.824 0.780

SVM 0.811 0.667 0.824 0.776

RF 0.811 0.667 0.824 0.748

NNET 0.730 0.375 0.828 0.728

XGB 0.784 0.500 0.818 0.744

LR: logistic regression; CART: classication and regression tree; SVM: support vector machine; RF:

random forest; NNET: neural network; XGB: Extreme Gradient Boosting.

338

Clinical and Experimental Rheumatology 2023

Prediction model for malignancy in IIM / W. Zhang et al.

0.367 (95% CI [0.147, 0.913]) in mul-

tivariable analysis (15). Furthermore, a

meta-analysis conrmed the protective

role of ILD in cancer of IIM (9). On

the other hand, MDA5 positive patients

(some with lung disease) have a poor

prognosis in DM, and the medications

used to treat DM predispose to cancers

and ILD per se can increase the risk of

lung cancer (34, 35). These conicting

factors are puzzling and certainly will

require further research.

Similar to the 151 patient study of So

et al. (21), our results pointed out that

the following were risk factors for ma-

lignancy in IIM: (i) older age; (ii) ALT

<80 U/L; (iii) the presence of TIF1-γ;

and (iv) the absence of ILD. However,

we did not nd that dysphagia predicted

cancer in multivariable analysis. Coin-

cident with our result, their laboratory

data also showed that lower serum AST

was related with malignancy in IIM pa-

tients and this phenomenon was rarely

reported. Possibly, IIM patients without

malignancy may have more probability

to behave with another system involve-

ment, for instance and dominantly in

muscle injury wherein the ALT or AST

are usually elevated. This variable in-

deed surprised us due to its high weight

among the ML models (21).

Several studies (4, 6, 36) have shown

that inammatory indicators, CRP and

ESR, can be used as predictive markers

for malignancy. However, it was not the

case in our research, perhaps because of

the other factors were more important

in the multi-variable models, making

these not statistically signicant. Also,

inammation is not merely associated

with the tumour, but also may be related

to patient’s multiple other concurrent,

co-morbidities (e.g. infection, necrosis,

drugs, other inammatory conditions)

which may confound their usefulness.

Our data have some notable strengths.

It includes a relatively robust number

of patients which have been sub-set into

IIM subtypes and carefully followed

and analysed. Also, updated serologi-

cal testing has been done (e.g. TIF1-γ).

Further a relatively sophisticated anal-

ysis was undertaken using machine

learning technology, thus increasing the

robustness of the results, increasing the

probability of credible results and al-

lowing internal consistency. Further we

tried to make the results easily available

for clinical use thru the development of

the web-based nomogram.

However, our data have some limita-

tions. First, our data comes from a

single-centre, which may enrol sicker

patients and increase the prevalence of

malignancy. Furthermore, because the

machine learning algorithm itself is

closely related with sample size, better

performance of LR was probably due

to the relatively small sample. Thus, it

needs to be further veried by multi-

center studies. Second, this is a retro-

spective study and data were missing,

especially the anti-myositis antibody

proles. Third, the length of follow-up

would ideally be longer. Fourth, tests

for cryoglobulinaemia and elevated

aldolase were not available in our unit

and confounding with other diseases

may have occurred, although it was un-

likely based on clinical results.

This study summarised the possible

risk factors for malignancy in a retro-

spective and case-control study of IIM

patients, and established a multivariate

risk prediction model of LR, which has

good usefulness for clinical application

and may help clinicians screen, evalu-

ate and follow up those high-risk pa-

tients with IIM. However, it still needs

more cases to optimise this model.

References

1. MARIAMPILLAI K, GRANGER B, AMELIN D

et al.: Development of a new classication

system for idiopathic inammatory myopa-

thies based on clinical manifestations and

myositis-specic autoantibodies. Jama Neu-

rol 2018; 75: 1528-37. https://

doi.org/10.1001/jamaneurol.2018.2598

2. CARDELLI C, ZANFRAMUNDO G, COMETI L

et al.: Idiopathic inammatory myopathies:

One year in review 2021. Clin Exp Rheuma-

tol 2022; 40: 199-209. https://

doi.org/10.55563/clinexprheumatol/vskjxi

3. LI Y, LI Y, WANG Y et al.: Nomogram to

predict dermatomyositis prognosis: A pop-

ulation-based study of 457 cases. Clin Exp

Rheumatol 2022; 40: 247-53. https://

doi.org/10.55563/clinexprheumatol/0ddt88

4. MOGHADAM-KIA S, ODDIS CV, ASCHERMAN

DP, AGGARWAL R: Risk factors and cancer

screening in myositis. Rheum Dis Clin North

Am 2020; 46: 565-76.

Fig. 5. The nomogram prediction model of the malignancy probability in IIM patients. Determine the

value of each factor based on the vertical line intersection between the variable and the point axis, and

then add all variable points to calculate the total risk score, with each risk score corresponding to the

probability of malignancy.

Fig. 6. QR code to be scanned with a smart-

phone for nomogram prediction.

339

Clinical and Experimental Rheumatology 2023

Prediction model for malignancy in IIM / W. Zhang et al.

https://doi.org/10.1016/j.rdc.2020.05.006

5. PRZYBYLSKI G, JARZEMSKA A, CZERNI-

AK J, SIEMIATKOWSKA K, GADZINSKA A,

CIESLINSKI K: A case report of a patient with

dermatomyositis as a prodromal sign of lung

cancer. Pol Arch Med Wewn 2008; 118: 143-7.

6. TINIAKOU E, MAMMEN AL: Idiopathic in-

ammatory myopathies and malignancy: a

comprehensive review. Clin Rev Allergy Im-

munol 2017; 52: 20-33.

https://doi.org/10.1007/s12016-015-8511-x

7. QIANG JK, KIM WB, BAIBERGENOVA A,

ALHUSAYEN R: Risk of malignancy in der-

matomyositis and polymyositis. J Cutan Med

Surg 2017; 21: 131-6.

https://doi.org/10.1177/1203475416665601

8. CHEN D, YUAN S, WU X et al.: Incidence

and predictive factors for malignancies with

dermatomyositis: A cohort from southern

China. Clin Exp Rheumatol 2014; 32: 615-

21.

9. WANG J, GUO G, CHEN G, WU B, LU L, BAO L:

Meta-analysis of the association of dermato-

myositis and polymyositis with cancer. Br J

Dermatol 2013; 169: 838-47.

https://doi.org/10.1111/bjd.12564

10. ANDRAS C, PONYI A, CONSTANTIN T et al.:

Dermatomyositis and polymyositis associat-

ed with malignancy: A 21-year retrospective

study. J Rheumatol 2008; 35: 438-44.

11. SPARSA A, LIOZON E, HERRMANN F et al.:

Routine vs extensive malignancy search for

adult dermatomyositis and polymyositis: a

study of 40 patients. Arch Dermatol 2002;

138: 885-90.

https://doi.org/10.1001/archderm.138.7.885

12. GALLAIS V, CRICKX B, BELAICH S: [Prog-

nostic factors and predictive signs of malig-

nancy in adult dermatomyositis]. Ann Der-

matol Venereol 1996;123:722-6.

13. KAJI K, FUJIMOTO M, HASEGAWA M et al.:

Identication of a novel autoantibody reac-

tive with 155 and 140 kDa nuclear proteins

in patients with dermatomyositis: An asso-

ciation with malignancy. Rheumatology (Ox-

ford) 2007; 46: 25-8.

https://doi.org/10.1093/rheumatology/kel161

14. TRALLERO-ARAGUAS E, LABRADOR-HOR-

RILLO M, SELVA-O’CALLAGHAN A et al.:

Cancer-associated myositis and anti-p155

autoantibody in a series of 85 patients with

idiopathic inammatory myopathy. Medicine

(Baltimore) 2010; 89: 47-52. https://

doi.org/10.1097/MD.0b013e3181ca14ff

15. ZHONG J, HE Y, MA J, LU S, WU Y, ZHANG J:

Development and validation of a nomogram

risk prediction model for malignancy in der-

matomyositis patients: A retrospective study.

Peerj 2021; 9: e12626.

https://doi.org/10.7717/peerj.12626

16. WU Y, RAO K, LIU J et al.: Machine learn-

ing algorithms for the prediction of central

lymph node metastasis in patients with papil-

lary thyroid cancer. Front Endocrinol (Laus-

anne) 2020; 11: 577537.

https://doi.org/10.3389/fendo.2020.577537

17. ANTIOCHOS BB, BROWN LA, LI Z, TOSTE-

SON TD, WORTMANN RL, RIGBY WF: Malig-

nancy is associated with dermatomyositis but

not polymyositis in Northern New England,

USA. J Rheumatol 2009; 36: 2704-10.

https://doi.org/10.3899/jrheum.090549

18. YANG Z, LIN F, QIN B, LIANG Y, ZHONG R:

Polymyositis/dermatomyositis and malig-

nancy risk: A metaanalysis study. J Rheuma-

tol 2015; 42: 282-91.

https://doi.org/10.3899/jrheum.140566

19. DOBLOUG GC, GAREN T, BRUNBORG C,

GRAN JT, MOLBERG O: Survival and can-

cer risk in an unselected and complete Nor-

wegian idiopathic inammatory myopathy

cohort. Semin Arthritis Rheum 2015; 45:3

01-8. https://

doi.org/10.1016/j.semarthrit.2015.06.005

20. NUNO-NUNO L, JOVEN BE, CARREIRA PE et

al.: Mortality and prognostic factors in idio-

pathic inammatory myositis: A retrospec-

tive analysis of a large multicenter cohort of

Spain. Rheumatol Int 2017; 37: 1853-61.

https://doi.org/10.1007/s00296-017-3799-x

21. SO MW, KOO BS, KIM YG, LEE CK, YOO B:

Idiopathic inammatory myopathy associ-

ated with malignancy: A retrospective cohort

of 151 Korean patients with dermatomyosi-

tis and polymyositis. J Rheumatol 2011; 38:

2432-5.

https://doi.org/10.3899/jrheum.110320

22. SIGURGEIRSSON B, LINDELOF B, EDHAG O,

ALLANDER E: Risk of cancer in patients with

dermatomyositis or polymyositis. A popula-

tion-based study. N Engl J Med 1992; 326:

363-7. https://

doi.org/10.1056/NEJM199202063260602

23. DUGAR M, COX S, LIMAYE V, BLUMBERGS

P, ROBERTS-THOMSON PJ: Clinical hetero-

geneity and prognostic features of South Aus-

tralian patients with anti-synthetase autoanti-

bodies. Intern Med J 2011; 41: 674-9. https://

doi.org/10.1111/j.1445-5994.2010.02164.x

24. STOCKTON D, DOHERTY VR, BREWSTER

DH: Risk of cancer in patients with dermato-

myositis or polymyositis, and follow-up

implications: A Scottish population-based

cohort study. Br J Cancer 2001; 85: 41-5.

https://doi.org/10.1054/bjoc.2001.1699

25. OLDROYD A, ALLARD AB, CALLEN JP et al.:

A systematic review and meta-analysis to

inform cancer screening guidelines in idio-

pathic inammatory myopathies. Rheuma-

tology (Oxford) 2021; 60: 2615-28. https://

doi.org/10.1093/rheumatology/keab166

26. ALLENBACH Y, KERAEN J, BOUVIER AM et

al.: High risk of cancer in autoimmune ne-

crotizing myopathies: Usefulness of myositis

specic antibody. Brain 2016; 139: 2131-5.

https://doi.org/10.1093/brain/aww054

27. TRALLERO-ARAGUAS E, RODRIGO-PENDAS

JA, SELVA-O’CALLAGHAN A et al.: Useful-

ness of anti-p155 autoantibody for diagnos-

ing cancer-associated dermatomyositis: a

systematic review and meta-analysis. Arthri-

tis Rheum 2012; 64: 523-32.

https://doi.org/10.1002/art.33379

28. CASCIOLA-ROSEN L, MAMMEN AL: Myo-

sitis autoantibodies. Curr Opin Rheumatol

2012; 24: 602-8. https://

doi.org/10.1097/BOR.0b013e328358bd85

29. ICHIMURA Y, MATSUSHITA T, HAMAGUCHI

Y et al.: Anti-NXP2 autoantibodies in adult

patients with idiopathic inammatory myo-

pathies: Possible association with malignan-

cy. Ann Rheum Dis 2012; 71: 710-3. https://

doi.org/10.1136/annrheumdis-2011-200697

30. FREDI M, CAVAZZANA I, CERIBELLI A et al.:

An Italian multicenter study on Anti-NXP2

antibodies: Clinical and serological asso-

ciations. Clin Rev Allergy Immunol 2022; 63:

240-50.

https://doi.org/10.1007/s12016-021-08920-y

31. FIORENTINO DF, CHUNG LS, CHRISTOPHER-

STINE L et al.: Most patients with cancer-

associated dermatomyositis have antibodies

to nuclear matrix protein NXP-2 or transcrip-

tion intermediary factor 1gamma. Arthritis

Rheum 2013; 65: 2954-62.

https://doi.org/10.1002/art.38093

32. IKEDA S, ARITA M, MISAKI K et al.: Inci-

dence and impact of interstitial lung disease

and malignancy in patients with polymyosi-

tis, dermatomyositis, and clinically amyo-

pathic dermatomyositis: A retrospective co-

hort study. Springerplus 2015; 4: 240.

https://doi.org/10.1186/s40064-015-1013-8

33. LU X, YANG H, SHU X et al.: Factors pre-

dicting malignancy in patients with poly-

myositis and dermatomyostis: A systematic

review and meta-analysis. Plos One 2014; 9:

e94128.

https://doi.org/10.1371/journal.pone.0094128

34. BALLESTER B, MILARA J, CORTIJO J: Idio-

pathic pulmonary brosis and lung cancer:

Mechanisms and molecular targets. Int J Mol

Sci 2019; 20: 593.

https://doi.org/10.3390/ijms20030593

35. LI J, YANG M, LI P, SU Z, GAO P, ZHANG J:

Idiopathic pulmonary brosis will increase

the risk of lung cancer. Chin Med J (Engl)

2014; 127: 3142-9.

36. LEE SW, JUNG SY, PARK MC, PARK YB, LEE

SK: Malignancies in Korean patients with

inammatory myopathy. Yonsei Med J 2006;

47: 519-23.

https://doi.org/10.3349/ymj.2006.47.4.519