Clinical trials with longitudinal outcomes typically include missing data due to missed assessments or structural missingness of outcomes after intercurrent events handled with a hypothetical strategy. Approaches based on Bayesian random multiple imputation and Rubin's rules for pooling results across multiple imputed data sets are increasingly used in order to align the analysis of these trials with the targeted estimand. We propose and justify deterministic conditional mean imputation combined with the jackknife for inference as an alternative approach. The method is applicable to imputations under a missing-at-random assumption as well as for reference-based imputation approaches. In an application and a simulation study, we demonstrate that it provides consistent treatment effect estimates with the Bayesian approach and reliable frequentist inference with accurate standard error estimation and type I error control. A further advantage of the method is that it does not rely on random sampling and is therefore replicable and unaffected by Monte Carlo error.

PURPOSE We compared overall survival (OS) in patients with human epidermal growth factor receptor 2 ( HER2)–amplified, treatment-refractory metastatic colorectal cancer (mCRC) receiving pertuzumab plus trastuzumab (PER-HER) in the phase IIa MyPathway multibasket study (ClinicalTrials.gov identifier: NCT02091141 ) with OS in those receiving routine clinical care in an electronic health record–derived external control arm. METHODS A noninterventional study was conducted using patient-level data from MyPathway participants receiving PER-HER and real-world patients with HER2-amplified treatment-refractory mCRC receiving routine clinical care. This study used a deidentified US-based clinico-genomic database (CGDB). For patients in the CGDB who met study eligibility criteria at multiple index dates (treatment initiation dates in the treatment-refractory setting), all eligible index dates were used for the analysis. Standardized mortality ratio weighting on the basis of propensity score derived a pseudopopulation (postweighting population) balancing key prognostic variables between arms. Multivariate Cox proportional hazards models were used for estimation of the hazard ratio (HR) in the primary OS analysis. A series of sensitivity analyses were conducted to investigate the robustness and consistency of the primary analysis. RESULTS The PER-HER arm comprised 57 patients enrolled in the MyPathway study by August 1, 2017 (data cutoff); the external control arm comprised 18 patients (27 index dates) with HER2-amplified mCRC who met the major MyPathway eligibility criteria in CGDB collected between 2011 and 2019. The estimated HR for OS from the multivariate Cox proportional hazards model in the postweighting population was 0.729 (95% CI, 0.184 to 3.900). The results of sensitivity analyses were consistent with the primary analysis in terms of the point estimate of HR. CONCLUSION Despite a small sample size, these findings suggest that PER-HER could have a potential OS benefit for this population.

No abstract

Clinical trials with longitudinal outcomes typically include missing data due to missed assessments or structural missingness of outcomes after intercurrent events handled with a hypothetical strategy. Approaches based on Bayesian random multiple imputation and Rubin's rule for pooling results across multiple imputed datasets are increasingly used in order to align the analysis of these trials with the targeted estimand. We propose and justify deterministic conditional mean imputation combined with the jackknife for inference as an alternative approach. The method is applicable to imputations under a missing-at-random assumption as well as for reference-based imputation approaches. In an application and a simulation study, we demonstrate that it provides consistent treatment effect estimates with the Bayesian approach and reliable frequentist inference with accurate standard error estimation and type I error control. A further advantage of the method is that it does not rely on random sampling and is therefore replicable and unaffected by Monte Carlo error.

36 Background: ML28897 (MyPathway) is a multi-basket trial evaluating the efficacy and safety of targeted therapies in non-indicated tumor types harboring relevant genetic alterations. In MyPathway, patients with treatment-refractory HER2-amplified metastatic colorectal cancer (mCRC) were enrolled and received pertuzumab plus trastuzumab. In order to facilitate contextualization of the outcome of pertuzumab plus trastuzumab in MyPathway, we conducted a retrospective study to compare with the outcome in real-world HER2-amplified mCRC patients. Methods: Overall survival (OS) was used as the endpoint to compare outcomes from MyPathway (PER/HER arm) and the external control of HER2-amplified mCRC patients treated with any therapy except anti-HER2 therapy in the refractory setting (EC arm) from the US-based de-identified Flatiron Health-Foundation Medicine Clinico-Genomic Database (CGDB). The de-identified data originated from approximately 280 US cancer clinics (̃800 sites of care). OS was defined as time from first treatment to death in the PER/HER arm and from index date (initiation of treatment in the refractory setting) to death in the EC arm. For patients in the CGDB who had met study eligibility criteria at multiple index dates, all eligible index dates were used for the analysis. Standardized mortality ratio weighting based on propensity score was used for deriving the pseudo-population (post-weighting population). In the post-weighting population, multivariate Cox regression models and Kaplan-Meier analyses were used to compare between the arms. A series of sensitivity analyses were conducted to investigate the robustness and consistency of the primary analysis results. Results: The PER/HER arm consisted of 57 patients who had treatment-refractory mCRC with HER2 amplification and enrolled in the MyPathway by August 1, 2017 data cutoff. For the EC arm, 64 HER2-amplified mCRC patients were selected from CGDB collected between January 1, 2011 and December 31, 2019. After applying the predefined inclusion/exclusion criteria set to be similar to those in the MyPathway, 27 eligible index dates were selected from 18 eligible patients and used for primary analysis. In the post-weighting population, the hazard ratio (HR) for OS estimated by multivariate Cox regression model was 0.729 (95% CI: 0.184-3.900) and median survival in the PER/HER arm and the EC arm were 11.47 months (95% CI: 7.72-22.11) and 9.72 months (95% CI: 7.43-22.21), respectively. The results of the all sensitivity analyses were consistent with those in the primary analysis in terms of the point estimate of HR. Conclusions: Despite a small sample size, the totality of findings suggests that the combination of pertuzumab and trastuzumab could have a potential benefit in OS for this population.