"Application of five different strategies to define a cohort of patients with knee osteoarthritis in a large primary care database".

Background: Electronic health records (EHR) are frequently used for epidemiological research including drug utilisation studies in a defined population such as the population with knee osteoarthritis (KOA). We sought to describe the process of defining a cohort of patients with KOA from a large UK primary care database and estimate the annual incidence of diagnosed KOA between 2000 and 2015.

Method: This was a retrospective study using data from the clinical practice research datalink (CPRD). CPRD is a large primary care longitudinal electronic medical records' database that contains anonymous records of patients from general practices across United Kingdom. Five different cohort definition strategies were applied including symptoms-based or diagnosis-based strategies or a combination of both. To validate results, the annual incidence of KOA was estimated and compared to published data.

Results: The study defined 898,690 patients when symptoms-based strategy was applied, 137,541 patients when diagnosis based and 83,294 when a combination of both strategies were applied. The final cohort was defined using a diagnosis-based strategy that avoided overestimation (with symptoms-based definition) or underestimation (with a combination of symptoms and diagnosis). The incidence of KOA ranged from 1.33 per 1000 CPRD registrants in 2000, 1.76 in 2008 and 1.45 patients in 2015.

Conclusion: This study logically/sensibly defined a cohort of patients with diagnosed KOA through the application of several strategies. This was an essential step to avoid subsequent over or underestimation of the prevalence of drug utilisation and the associated adverse clinical outcomes within primary care patients with KAO.