Using Bayesian Networks to identify musculoskeletal symptoms influencing the risk of developing Psoriatic Arthritis in people with psoriasis.

Objectives To explore Bayesian networks (BNs) in understanding the relationships between musculoskeletal symptoms and the development of PsA in people with psoriasis. Methods Incident cases of psoriasis were identified between 1998 and 2015 from the UK Clinical Research Practice Datalink. Musculoskeletal symptoms identified by medcodes were concatenated into primary groups, each made up of several sub-groups. Baseline demographics for gender, age, body mass index (BMI), psoriasis severity, alcohol use and smoking status were also extracted. Several BN structures were composed using a combination of expert knowledge and data-oriented modelling using: 1) primary musculoskeletal symptom groups, 2) musculoskeletal symptom sub-groups and 3) demographic variables. Predictive ability of the networks using the receiver operating characteristic curve (AUC) was calculated. Results Over one million musculoskeletal symptoms were extracted for the 90,189 incident cases of psoriasis identified, of which 1409 developed PsA. The BN analysis yielded direct relationships between gender, BMI, arthralgia, finger pain, fatigue, hand pain, hip pain, knee pain, swelling, back pain, myalgia and PsA. The best BN, achieved by using the more site-specific musculoskeletal symptom sub-groups, was 76% accurate in predicting the development of PsA in a test set and had an AUC of 0.73 (95% confidence interval (CI): 0.70-0.75). Conclusion The presented BN model may be a useful method to identify clusters of symptoms that predict the development of PsA with reasonable accuracy. Using a BN approach, we have shown that there are several symptoms which are predecessors of PsA, including fatigue, specific types of pain, and swelling.