Methods to estimate when to stop a treatment and to compare treatment durations using observational data: a methodological scoping review

For many diseases, the question of the optimal treatment duration is important, and rarely addressed in randomized controlled trials. We searched Medline, Embase, CINAHL and Google scholar for observational studies published between 2005 and 2023 that explicitly determine stopping versus continuing treatment at a given timepoint or treatment duration. Two independent reviewers extracted in duplicate data on study characteristics, target trial emulation (eligibility, treatment assignment, follow-up), and other information related to the topic of interest. Out of 13980 studies, we identified 71 eligible studies, 7 describing epidemiological or statistical methods, and 64 applied observational studies that compared treatment durations ( n = 56) or stopping vs. continuing a treatment at a specific timepoint ( n = 8). In the 56 studies comparing treatment durations, 17 (30%) studies reported possible risk of post-baseline eligibility. Post-baseline treatment assignment with an immortal time period was present for 14 (25%) of studies. Misalignment of timepoints was observed in 12 (21%) studies. 14 (25%) studies indicated the presence of time-fixed confounders. 48 (86%) studies failed to consider time-varying confounders. Overall, only six of 56 studies (11%) had no bias or considered unclear on all design bias considered, and 50 studies (89%) still had at least one other design bias. Many observational studies comparing treatment durations or stopping versus continuing a treatment had possible design biases, the most frequent one being the lack of control for time-varying confounders. Using adequate methods could effectively control the potential risks of bias in observational studies.