(C-152) Impact of design choices that deviate from a target trial emulation. An application to the estimation of the effect of anticholinergic properties of antidepressants on the incidence of dementia

Background: Several observational studies have reported an 11-74% higher relative hazard of dementia comparing long-term users (eg, ≥ 10 years) of antidepressants with anticholinergic properties (AC) compared with nonusers of antidepressants. However, a target trial emulation (TTE) comparing antidepressants with AC and without anticholinergic (non-AC) properties estimated no effect of AC antidepressants on dementia

Objectives: To investigate the potential role of study design in explaining the discrepant estimates from the TTE and prior observational studies

Methods: We consider three three ways in which the previous observational studies deviated from the target trial: (1) the use of a comparator group defined as no antidepressant therapy in the previous observational studies (vs. non-AC antidepressants in the TTE), (2) the inclusion of a post-baseline survival requirement for eligibility (vs. application of eligibility criteria at baseline in the TTE), and (3) the classification of individuals into treatment groups based on treatment received during follow-up (vs. assignment of treatment strategies at baseline in the TTE). To explore the impact of these design decisions on our estimates, we emulated a target trial that sequentially (1) considered no antidepressant therapy as the comparator group, (2) added survival for 10 years after baseline as an eligibility criterion, and (3) classified patients into treatment groups based on the treatment they received in the year after baseline. CPRD GOLD (2001-2017) was used for the analyses. Twelve-year risk ratios (RR) were estimated using parametric survival analysis with adjustment for baseline and post-baseline variables

Results: In the previous TTE, The the estimated 12-year RR of dementia in the TTE was 0.90 (95% CI 0.61, 1.52), when comparing AC vs. non-AC antidepressants (23,545 initiators of AC vs. 16,320 initiators of non-AC). By contrast, the RR was 1.06 (0.77, 1.21) when comparing AC antidepressants with no antidepressant therapy (29,408 initiators of AC vs. 1,114,975 non-initiators), 1.25 (1.01, 1.32) when requiring that individuals survived 10 years after baseline to be eligible (19,000 surviving initiators, vs. 734,979 surviving non-initiators), and 1.30 (1.02, 1.45) when patients were additionally classified into treatment strategies according to the treatment they received in the first year of follow-up (9,862 exposed postbaseline vs. 584,392 non-exposed postbaseline)

Conclusions: Our results suggest that antidepressants with AC properties do not increase the risk of dementia at 12 years compared with antidepressants without AC properties or with no antidepressant use. Discrepancies with the results from other studies may be explained by study design decisions that deviate from a target trial