Date: Sunday, November 5, 2017
Session Type: ACR Poster Session A
Session Time: 9:00AM-11:00AM
Background/Purpose: Rheumatoid arthritis (RA) is associated with increased mortality. Traditional survival techniques used to identify mortality risk factors, such as the Cox proportional hazards model (CPH), have several limitations, such as reliance on restrictive assumptions. To overcome these limitations, machine learning methods have been developed. Random survival forest (RSF), a non-parametric ensemble tree method, was proposed as an alternative approach to CPH. Our aim is to develop and validate (internally and externally) a predictive model for RA mortality using RSF.
Methods: Retrospective longitudinal study involving 2 independent RA cohorts from Madrid (Spain): the Hospital Clínico San Carlos RA Cohort (HCSC-RAC: 1,461 patients diagnosed between 2001 and 2011, followed-up until death or September 2013; used for model development), and the Hospital Universitario de La Princesa Early Arthritis Register Longitudinal study (PEARL: 280 patients diagnosed between 2001 and 2014, followed-up until death or January 2017; used for external validation). Demographic and clinical-related variables collected during the first two years after disease diagnosis were used. RSF models were developed with the randomForestSRC R package, based on 1,000 trees. 100 iterations of each model were performed to measure the mean and standard deviation of the prediction error. Based on the predicted mortality estimated by the RSF model, mortality risk groups were established through a survival tree created with the R package rpart.
Results: 148 and 21 patients from the HCSC-RAC and the PEARL died during a median follow-up time of 4.3 and 5.0 years, respectively. Age at diagnosis, median erythrocyte sedimentation rate, and number of hospital admissions in the first 2 years after RA diagnosis showed the higher predictive capacity. The prediction errors of our model in the training and in the validation cohorts were 0.187, and 0.233, respectively. The survival tree analysis identified 5 risk groups. After combining those three with intermediate risk, we observed that the intermediate and the high risk groups were significantly associated with higher mortality risk compared with the low risk, both in the HCSC-RAC and PEARL cohorts (Figures 1 and 2).
Conclusion: We developed and externally validated a clinical prediction model for RA mortality using RSF.
Figure 1: Kaplan Meier curves for the observed mortality of patients from the HCSC-RAC, grouped in mortality risk categories
Figure 2: Kaplan Meier curves for the observed mortality of patients from the PEARL, grouped in mortality risk categories
To cite this abstract in AMA style:Rodriguez-Rodriguez L, Lezcano-Valverde JM, Salazar F, León L, Toledano E, Jover Jover JA, Soudah E, Fernández-Gutiérrez B, Gonzalez-Alvaro I, Alcazar LA. Machine Learning in Rheumatology: Development and Validation of a Predictive Model for Rheumatoid Arthritis Mortality Using Random Survival Forests [abstract]. Arthritis Rheumatol. 2017; 69 (suppl 10). https://acrabstracts.org/abstract/machine-learning-in-rheumatology-development-and-validation-of-a-predictive-model-for-rheumatoid-arthritis-mortality-using-random-survival-forests/. Accessed April 2, 2020.
« Back to 2017 ACR/ARHP Annual Meeting
ACR Meeting Abstracts - https://acrabstracts.org/abstract/machine-learning-in-rheumatology-development-and-validation-of-a-predictive-model-for-rheumatoid-arthritis-mortality-using-random-survival-forests/