Premium
Profiling Arthritis Pain with a Decision Tree
Author(s) -
Hung Man,
Bounsanga Jerry,
Liu Fangzhou,
Voss Maren W.
Publication year - 2018
Publication title -
pain practice
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.899
H-Index - 58
eISSN - 1533-2500
pISSN - 1530-7085
DOI - 10.1111/papr.12645
Subject(s) - medicine , machine learning , decision tree , discriminative model , artificial intelligence , physical therapy , predictive analytics , data mining , computer science
Background Arthritis is the leading cause of work disability and contributes to lost productivity. Previous studies showed that various factors predict pain, but they were limited in sample size and scope from a data analytics perspective. Objectives The current study applied machine learning algorithms to identify predictors of pain associated with arthritis in a large national sample. Methods Using data from the 2011 to 2012 Medical Expenditure Panel Survey, data mining was performed to develop algorithms to identify factors and patterns that contribute to risk of pain. The model incorporated over 200 variables within the algorithm development, including demographic data, medical claims, laboratory tests, patient‐reported outcomes, and sociobehavioral characteristics. Results The developed algorithms to predict pain utilize variables readily available in patient medical records. Using the machine learning classification algorithm J48 with 50‐fold cross‐validations, we found that the model can significantly distinguish those with and without pain ( c ‐statistics = 0.9108). The F measure was 0.856, accuracy rate was 85.68%, sensitivity was 0.862, specificity was 0.852, and precision was 0.849. Conclusion Physical and mental function scores, the ability to climb stairs, and overall assessment of feeling were the most discriminative predictors from the 12 identified variables, predicting pain with 86% accuracy for individuals with arthritis. In this era of rapid expansion of big data application, the nature of healthcare research is moving from hypothesis‐driven to data‐driven solutions. The algorithms generated in this study offer new insights on individualized pain prediction, allowing the development of cost‐effective care management programs for those experiencing arthritis pain.