Date: Monday, November 9, 2020
Session Title: Measures & Measurement of Healthcare Quality Poster
Session Type: Poster Session D
Session Time: 9:00AM-11:00AM
Background/Purpose: Hydroxychloroquine (HCQ) is associated with varied cutaneous side effects but only few studies in literature characterizing the risk factors for this. Recently machine learning tools have emerged as a promising option for predicting various outcomes. The aim of our study was to use “traditional” multivariate analysis along with novel machine learning (ML) methodology to analyze and predict for the occurrence of cutaneous side effects with HCQ usage.
Methods: Demographic, clinical details, laboratory parameters, treatment details of HCQ (dose/mg/kg, cumulative dose, duration of therapy) and coexistent DMARD prescription were collected for patients who were treated with HCQ for autoimmune rheumatic diseases from 2015 to 2019. Development of cutaneous side effects was the primary outcome of the study and the subtypes of cutaneous side effect were also noted (hyperpigmentation, photosensitivity and pruritus) retrospectively. Univariate and multivariate logistic regression for the primary outcome was done using SPSS 20.0. To develop the ML classification models for prediction of the primary outcome, we used Random Forest, Logistic Regression and K-nearest neighbour (KNN) algorithms from Python-3.0’s scikit library. Hyperparameter tuning and five-fold cross validation were done for all three ML classifiers. After the three classifer models were trained, they were compared using accuracy scores on the test dataset. Precision and area under ‘receiver operating curve’ (AUC) of the highest accuracy model were also analyzed.
Results: Baseline characteristics of the 430 participants recruited are shown in Figure 1(A). 198 patients (46.04%) developed cutaneous side effects of which the most common was hyperpigmentation (n =182, 42.3%).15 patients stopped HCQ due to the cutaneous side effects. Figure 1(A) also shows the results of the univariate and multivariate regression analysis for development of cutaneous side effects. The risk factors identified were the cumulative dose of HCQ > 216,000 mg (OR 1.82, 1.22-2.73), combination treatment with MMF(OR 2.13, 1.10-4.09), or AZA(OR 2.05, 1.04-4.01) ANA positive (OR 1.72, 1.06-2.8), and RF positive(OR 2.38, 1.51-3.76) . Accuracy scores of three ML models are shown in figure 2A, with the KNN classifier model having the best performance. It achieved a 62.79% accuracy with precision of 0.6 and AUC of 0.64 (figure 2B).
Conclusion: Cutaneous side effects are common with HCQ treatment. While traditional statistical methods can help identify the risk factors, usage of supervised ML methods could help predict development of cutaneous side effects in an individual patient. Larger dataset analysis could allow for development of tailored therapeutic regimens avoiding adverse events.
To cite this abstract in AMA style:Surendran S, CB M, Tiwari A, Marwaha V, Easwar S. Cutaneous Side Effects of Hydroxychloroquine in Rheumatic Diseases –Combination of “Traditional” Multivariate Analysis for Risk Factors AndClassification Model Development Using Supervised Machine Learning –Single Centre Retrospective Cohort Study in India [abstract]. Arthritis Rheumatol. 2020; 72 (suppl 10). https://acrabstracts.org/abstract/cutaneous-side-effects-of-hydroxychloroquine-in-rheumatic-diseases-combination-of-traditional-multivariate-analysis-for-risk-factors-andclassification-model-development-usin/. Accessed November 27, 2020.
« Back to ACR Convergence 2020
ACR Meeting Abstracts - https://acrabstracts.org/abstract/cutaneous-side-effects-of-hydroxychloroquine-in-rheumatic-diseases-combination-of-traditional-multivariate-analysis-for-risk-factors-andclassification-model-development-usin/