Evaluating Methods to Estimate United States Multiple Sclerosis Prevalence from Administrative Health Claims Data
MetadataShow full item record
Introduction The prevalence of Multiple Sclerosis (MS) in the United States (US) historically has been poorly described, in part because disease activity is intermittent, making detection using administrative health claims (AHC) data challenging.1 In a recent study Wallin and colleagues inflated a 3-year cumulative prevalence estimate by 37% to 47% to estimate a 10-year cumulative prevalence to account for case under-ascertainment in AHC data.2 However, their new estimate is approximately double that of previous analyses.3-5 The objective of my study was to assess the validity of the 3-year vs. 10-year inflation factor using a US AHC dataset between 2008 and 2017. Methods Patients with MS were identified in AHC data from IBMÂ® MarketScanÂ® Research Databases – one of the largest samples of employer-sponsored health plan data comprised of over 250 million unique patients since 1995.6 Multiple Sclerosis patients were identified based on inpatient, outpatient, and disease-modifying therapy (DMT) outpatient pharmaceutical claims using a validated algorithm developed by Culpepper and colleagues.7 Cumulative prevalence was estimated annually over ascertainment periods of 3 years (2015 – 2017) and 10 years (2008 – 2017), and the 2017 cumulative prevalence was compared between the two ascertainment periods. To ensure the I was implementing the algorithm correctly, I compared my 2008 through 2010 results to those obtained in the same analysis conducted by the algorithm developers. Lastly, because the algorithm was validated using DMTs approved through 2010, the effect of including DMTs approved through 2017 was assessed. Results The 2017 10-year cumulative prevalence (251/100,000) was 7.6% higher than the 3-year cumulative prevalence (233/100,000). Using the algorithm with only DMTs approved in the US through 2010, the 2017 10-year cumulative prevalence (240/100,000) was 10.1% higher than the 3-year cumulative prevalence (218/100,000). Compared with Culpepper’s MarketScanÂ® analysis (2008 – 2010), I identified 5,115 fewer cumulative cases (67,728 total) in the same time period, although the proportion of cases in each age group was not significantly different (Chi-square heterogeneity test, p = 0.35). Conclusion I found the 3-versus-10-year ascertainment period cumulative prevalence difference for 2017 to be 30 to 40 percentage points smaller than that identified by Culpepper et al. for 2010. The different datasets and time periods used may contribute in part to this discrepancy, but the reason for this difference remains unclear. Until future work elucidates a more robust 3-vs-10-year difference, this method should be used cautiously to extrapolate prevalence estimates.