The Misclassification of Depression and Anxiety Disorders in the Multiple Sclerosis Prodrome: A Probabilistic Bias Analysis.

Abstract

Background: Studies suggest that depression/anxiety form part of the multiple sclerosis (MS) prodrome. However, several biases have not been addressed. We re-examined this association after correcting for: (i) misclassification of individuals not seeking healthcare, (ii) differential surveillance of depression/anxiety in the health system, and (iii) misclassified person-time from using the date of the first MS-related diagnostic claim (i.e., a demyelinating event) as a proxy for MS onset.

Methods: In this cohort study, we applied a validated algorithm to health administrative ('claims') data in British Columbia, Canada (1991-2020) to identify MS cases, and matched to general population controls. The neurologist-recorded date of MS symptom onset was available for a subset of MS cases. We identified depression/anxiety in the 5-years preceding the first demyelinating claim using a validated algorithm. We compared the prevalence of depression/anxiety using modified Poisson regression. To account for misclassification and differential surveillance, we applied probabilistic bias analyses; for misclassified person-time, we applied time-distribution matching to the MS symptom onset date.

Results: Our cohort included 9,929 MS cases and 49,574 controls. The prevalence ratio for depression/anxiety was 1.74 (95%CI: 1.66-1.81). Following correction for misclassification, differential surveillance using a detection ratio of 1.11, and misclassified person-time, the prevalence ratio increased to 3.25 (95%CI: 1.98-40.54). When the same correction was conducted, but a detection ratio of 1.16 was applied, the prevalence ratio increased to 3.13 (95%CI: 1.97-33.52).

Conclusions: Previous conventional analyses were biased towards the null, leading to an under-estimation of the association between depression/anxiety and MS in the prodromal period. This first application of probabilistic quantitative bias analysis within MS research demonstrates both its feasibility and utility.