(A-177) An Alternative to Non-Responder Imputation for Study Participants with Missing Outcomes in Single-Arm Drug Effectiveness Studies: A Case Study in the CorEvitas Psoriasis Registry

Background: The goal of many single-arm drug effectiveness studies is to assess patient responses to a drug of interest within a specific follow-up (FU) period. The prospective, non-interventional multi-center CorEvitas Psoriasis Registry aims to collect patient and provider information during routine dermatology clinic sites at 6-month intervals. However, a proportion of study-eligible patients will lack a qualifying FU Registry visit. Excluding these patients from effectiveness analyses raises selection bias concerns and can impact the validity of study findings. Non-responder imputation (NRI) is a common approach for binary outcomes that imputes non-response to all patients with a missing outcome. However, this is an extreme, and not always clinically appropriate, assumption.

Objectives: We demonstrate an alternative to NRI for binary outcomes that allows for more clinically informed assumptions about true response rates for patients with missing outcomes, in the context of a 6-month, single-arm follow-up study.

Methods: We selected patients with diagnosed plaque psoriasis who initiated a biologic therapy for psoriasis (4/15/2015-3/31/2017) and were followed for a 6-month (6M) FU visit (5- to 9-month window). Several outcomes were investigated; here we report results for PASI (Psoriasis Area Severity Index) ≤ 2. Response to therapy was defined as achieving PASI ≤ 2. For patients lacking a FU visit, the number of responders was singly imputed assuming (based on expert knowledge) that the most common reason for lacking 6M FU is that a patient was responding well to therapy and chose not to return to the dermatology clinic within 9 months after initiation. This assumption was represented by odds ratios (OR) of 1.5, and 3.0, representing moderate and large increases in the odds of responding to therapy for patients without a 6M FU visit relative to those with a 6M FU visit. For comparison, results were also calculated under NRI. Proportions of responders, and 95% confidence intervals (CI), were first calculated using complete cases only, then for all patients using imputation under each assumption.

Results: 407 patients were included. Of the 341 (84%) that had an observed outcome at 6M FU,168 achieved response [49.3% (44.0%-54.6%)]. Imputing number of responders based on ORs of 1.5 and 3.0 yielded estimates of 207/407 [50.9% (46.1%-55.7%) and 217/407 [53.4% (48.5%-58.1%)], respectively. By contrast, NRI yielded 168/407 [41.3% (36.6%-46.1%)].

Conclusions: For binary outcomes, this alternative to NRI allows investigators to incorporate subject matter expertise and clinical experience into the specification of assumptions about the responses of patients with missing outcomes in sensitivity analyses.