Comparing Different As-Treated Approaches: A Methodological Study Using Real-World Data on Psoriasis Treatments

BACKGROUND: Accurate pharmaco-epidemiological assessment of the safety of biological treatments is essential but methodologically challenging.

AIMS: This study evaluated how different assumptions regarding treatment switching and treatment episode definitions affect outcome estimates in psoriasis patients, focusing on malignancies and serious infections.

METHODS: Data from Danish national registers (2009-2022) were analyzed for hospitalized psoriasis patients treated with ustekinumab, non-biologics, tumor necrosis factor-α inhibitors, and other interleukin inhibitors excluding ustekinumab. Various as-treated approaches were assessed and compared to the Intention-to-Treat (ITT) approach. The as-treated approaches included: (1) RC-switch, considering switching only between ustekinumab and the comparator in question, (2) Bio-switch, considering switches among all biologics, (3) Multi-switch, analyzing all groups simultaneously, and (4) Hierarchy-switch, analyzing switching by a predefined hierarchy. Hazard ratios (HRs) and 95% confidence intervals (CIs) were analyzed using Cox models. Sensitivity analyses incorporated treatment episodes based on the defined daily dose (DDD) and grace periods.

RESULTS: Broader treatment switching allowances affected outcome estimates. For malignancies, the Bio-switch approach yielded the largest HR changes relative to the ITT. For serious infections, also, Bio-switch produced the largest HR shifts, for example, HR 1.32 (CI 1.09, 1.58) versus ITT HR 1.08 (CI 0.91, 1.29). Sensitivity analyses showed that accounting for episode gaps influenced HRs and widened CIs.

CONCLUSION: The number of events within the exposure group varied by approaches, influencing the HR estimates. The multi-switch approach effectively captured treatment switching and reduced statistical uncertainty, supporting clearer safety conclusions. Tracking each treatment period with gaps reflects real-world practice but may lower statistical power and should be carefully considered in analyses.