Machine learning for screening and predicting the risk of anti-MDA5 antibody in juvenile dermatomyositis children

Abstract

Objective: The anti-MDA5 (anti-melanoma differentiation associated gene 5) antibody is often associated with a poor prognosis in juvenile dermatomyositis (JDM) patients. In many developing countries, there is limited ability to access myositis- specific antibodies due to financial and technological issues, especially in remote regions. This study was performed to develop a prediction model for screening anti-MDA5 antibodies in JDM patients with commonly available clinical findings.

Methods: A cross-sectional study was undertaken with 152 patients enrolled from the inpatient wards of Beijing Children's Hospital between June 2018 and September 2021. Stepwise logistic regression, least absolute shrinkage and selection operator (LASSO) regression, and the random forest (RF) method were used to fit the model. Model discrimination, calibration, and decision curve analysis were performed for validation.

Results: The final prediction model included eight clinical variables (gender, fever, alopecia, periungual telangiectasia, digital ulcer, interstitial lung disease, arthritis/arthralgia, and Gottron sign) and four auxiliary results (WBC, CK, CKMB, and ALB). An anti-MDA5 antibody risk probability-predictive nomogram was established with an AUC of 0.975 predicted by the random forest algorithm. The model was internally validated by Harrell's concordance index (0.904), the Brier score (0.052), and a 500 bootstrapped satisfactory calibration curve. According to the net benefit and predicted probability thresholds of decision curve analysis, the established model showed a significantly higher net benefit than the traditional logistic regression model.

Conclusion: We developed a prediction model using routine clinical assessments to screen for JDM patients likely to be anti-MDA5 positive. This new tool may effectively predict the detection of anti-MDA5 in these patients using a non-invasive and efficient way.

Keywords: antiMDA5; antibody; juvenile dermatomyositis; machine learning; pediatric.

Figure 1

Flow chart showing the study design.

Figure 2

Variable selection using the LASSO logistic Poisson model. LASSO model coefficient profiles of the 19 candidate variables. The logistic regression coefficients are estimated with an upper bound (“L1 norm”) to the sum of the absolute standardized regression coefficients. The L1 norm regularization term typically shrinks many regression coefficients to 0.

Figure 3

Tuning parameter selection by cross-validation in the LASSO model. The solid vertical lines represent the partial likelihood deviance standard error (SE). The red dotted line indicates the cross-validation curve. The broken vertical lines indicate the optimal values on the basis of the minimum criteria and 1−SE criteria. A λ value of 0.04194893, with a log(λ) value of −4.356293, was chosen according to cross-validation.

Figure 4

The area under the curve (AUC) of the random forest algorithm for the established model.

Figure 5

Nomogram and the calibration curve for the anti-MDA5 prediction model. (A) Nomogram predicting the probability that a JDM patient will be detection positive for anti-MDA5 antibodies. Points for 12 screened variables can be obtained using a point caliper and then summed to obtain a total score that can be matched with the risk. (B) Calibration curve of the prediction model by the actual risk with 500 bootstraps. Broken line represents apparent prediction; solid line represents the performance of the corrected prediction model. A smaller distance between the scatter point and the broken line indicates a better calibration.

Figure 6

Decision curve analysis of the two models. “None” is the net benefit when it is assumed that none of the JDM patients will have the outcome (anti-MDA5 positive). “All” is the net benefit when it is assumed that all the JDM patients will have the outcome. ‘established’’logistic’ represents the net benefit when JDM patients are screened the predicted risk of anti-MDA5 estimated by different prediction models, respectively. Thestrategy with the highest net benefit at any given threshold is the preferred strategy.