Value of a Consensus Panel to Adjudicate Cause-Specific Mortality after Unrelated Donor Allogeneic Hematopoietic Cell Transplantation (URD-HCT) for Use As the Primary Endpoint in a Genome-Wide Association Study (GWAS)

Cause-specific death (CSD) due to transplant-related complications (TRM) after URD-HCT is often multi-factorial. Consensus adjudication panels (CP), commonly used in clinical trials, are rarely used in observational research. Given the risk of endpoint misclassification due to CSD complexity and reporting variability, we hypothesized that a CP would uncover misclassifications that could cause confounding effects in a GWAS. A CP for CSD was developed, including 2 adult BMT physicians (MP, PLM), a pediatric oncologist (KO), and BMT epidemiologist (TH). A genetic epidemiologist (LSC) and statisticians (LP, DT, XZ) observed the CP. The training cohort includes 2,609 10/10 HLA-matched T-replete myeloablative and reduced intensity URD-HCT patients treated 2000 to 2008 for AML, ALL or MDS, reported to the Center for International Blood and Marrow Transplant Research (CIBMTR) by 151 transplant centers (TCs) with banked recipient and donor samples; 1116 died ≤1-yr post-URD-HCT. The CP held 3 sessions at the CIBMTR and weekly teleconferences over 8 months. Consensus was reached by reviewing clinical data summaries, autopsy reports, error corrections and source documentation. Both internal (random re-review of 11 cases) and external validity (separate CP review of 21 cases) were measured. Of 1116 deaths, 99 (9%) needed additional information from the TC; 75/99 (76%) of data queries were resolved after TC response, and 24/99 (24%) were re-reviewed by the CP for final resolution. Several agreement measures were generated to assess concordance on CSD between TC/CP for all cases, and stratified by gender, age, disease, disease status and year of URD-HCT. Logistic models determined which covariates were associated with odds of TC and CP CSD agreement and log-linear models were used to assess dependencies between covariates and TC/CP CSD. Of 465 cases reported by the TC as disease-related death, the CP agreed with 99.6% (Table). Of 651 cases reported by the TC as TRM, the CP agreed with 80% and adjudicated 130 as disease-related death, hence overall agreement was 87.5%, κ_max=0.773. When considering CSD (disease, GVHD, infection, organ failure, other) the overall agreement was 72.8%, κ_max=0.786. Logistic models found that TC and CP were about 2 fold less likely (p<.05) to agree on CSD for ALL and MDS, when URD-HCT was in 2000-2003 (early in cohort) or when the patient was not in CR pre-URD-HCT. Log-linear models showed the TC CSD, age and year of URD-HCT were dependent, while the CP CSD and disease status were dependent. While the TC and CP have good agreement, the CP's systematic review of cases eliminated a cohort effect and dependencies between CSD and patient characteristics. In analyses of transplant-specific failure, CPs are a necessary and worthwhile component of genetic and translational studies correlating outcomes with patient and transplant-related characteristics.