Bayesian modelling of disease evolution in musculoskeletal epidemiology

Psoriatic arthritis (PsA) is a progressive and often destructive joint disease affecting approximately 20% of patients with psoriasis. The inflammation, joint damage and deformity associated with PsA adds substantially to the disease burden for psoriasis patients, leading to an impaired quality of life. Understanding the role of risk factors, such as obesity, is essential in determining which patients are at the greatest risk of developing PsA. The ability to utilise information from routinely collected data contained in databases such as the Clinical Practice Research Datalink (CPRD), which contains medical records for ca. 11.7 million individuals, has the potential to greatly increase this understanding. However, there may be issues when using databases for purposes beyond that for which they were originally designed. In a traditional epidemiology study, data would be collected in accordance with a specified study design, for example recording measurements of risk factors, such as body mass index (BMI), at specified points in time. This is not the case in the CPRD where, from a statistical point of view, the non-routine collection of data results in missing values. Bayesian hierarchical models (BHM) provide a coherent framework within which the inherent uncertainty present in clinical measurements, and the estimation of missing data, can be propagated throughout the modelling process and incorporated in resulting estimates of risk and associated measures of uncertainty, such as confidence intervals. A case study using data extracted from the CPRD is presented in which BHMs were used to estimate the risk of developing PsA in patient with psoriasis. Using data on 90,182 psoriasis patients, the possibility of lagged and cumulative risks associated with changing BMI were investigated. Results showed that increased BMI was significantly associated with a higher risk of PsA, with evidence that reductions in BMI over time reduce risk.