OBJECTIVES: Livingstone is a trusted research environment that allows the user to define cohorts from real-world data and output reproducible epidemiology reports. Incidence and prevalence are key epidemiological metrics that may vary by case definition, the acute or chronic nature of the disease, and denominator selection. In this study, we compared incidence and prevalence estimates derived automatically from Livingstone with published data.

METHODS: We used Livingstone to replicate findings from a Clinical Practice Research Datalink (CPRD) study which reported the prevalence and incidence of six conditions: inflammatory myopathy, muscular dystrophy (MD), Charcot-Marie Tooth disease, Guillain-Barre syndrome, myasthenia gravis, and motor neurone disease (MND). Estimates for the annual incidence and prevalence 2004–19 of the six conditions were compared. In addition, annual rates were compared for each condition. Lin’s concordance correlation coefficient (CCC) was used to measure the concordance between Livingstone and the published data. This study received CPRD RDG approval (22_001781)

RESULTS: For the overall annual estimates, the CCC for incidence was 0.97 (95% CI 0.97–0.98) and 0.99 (0.99–0.99) for prevalence. For each condition by year, the CCCs for incidence were: Charcot-Marie tooth disease 0.61 (0.27–0.82); Guillain-Barre syndrome 0.80 (0.58–0.91); inflammatory myopathy 0.86 (0.70–0.94); MND 0.52 (0.25–0.72); MD 0.78 (0.56–0.89), and myasthenia gravis 0.89 (0.75–0.95). For prevalence, the corresponding CCCs were 0.96 (95% CI 0.92–0.98); 1.00 (0.99–1.00); 0.95 (0.89–0.98); 0.20 (0.02–0.37); 0.24 (0.08–0.38); and 0.97 (0.94–0.99), respectively.

CONCLUSIONS: For the six conditions, there was excellent concordance between Livingstone and the published results for the overall estimates of both incidence and prevalence. There were also high levels of concordance for the majority of estimates for the individual conditions by year. Due to the relative consistency of prevalence estimates over the study period for MDN and MD, the underlying bivariate distribution was heavy-tailed and thus Lin’s CCC was less robust in these two conditions.