Abstract 5079: Adjudicating complex phenotypes for use in GWAS: A study of transplant-related mortality (TRM) after unrelated donor hematopoietic cell transplantation (URD-HCT)

Abstract

Abstract As a first step in a 2-phase GWAS of TRM in patients with acute lymphoblastic leukemia (ALL), acute myelogenous leukemia (AML) and myelodysplastic syndrome (MDS) after URD-HCT, a consensus panel (CP), composed of BMT physicians and epidemiologists, reviewed all cause-specific deaths (CSDs) reported by 151 US transplant centers (TC) from 2000-11. Agreement measures, statistics and models were used to compared CP and TC CSD and the factors impacting the determination of CSDs. Less affected by prevalence and marginal probabilities than Cohen's K, Gwet's Agreement Coefficient (AC1) measured the agreement between TC and CP CSD and the 95% confidence intervals were generated by bootstrapping the conditional distributions of the CSDs. For each CSD (disease, graft versus host (GvHD), infection, organ failure, other) we determined the proportion of agreement (PA) between the TC and CP and the Bias Index (BI, the proportion of cases the CP moved into or out of a CSD category). Logistic and log-linear models identified factors significantly associated with TC/CP odds of agreement and dependencies between covariates and TC/CP CSD, respectively. The discovery cohort (c1) includes 2609 10/10 HLA matched URD-HCT patients from 2000-8, reported to the Center for International Blood and Marrow Transplant Research (CIBMTR) by TCs. The validation cohort (c2) includes 920 10/10 or 8/8 HLA matched URD-HCT patients from 2000-11. The CP adjudicated CSDs in c1 using in-person meetings and teleconferences over 1 year; c2 cases were reviewed during a 2-day in-person meeting. A total of 1116 (c1) and 388 patients (c2) died ≤1-yr post-URD-HCT. External and internal validity showed >96% agreement. The AC1 statistic was .751 (.726, .801) and .658 (.567, .682), for c1 and c2 respectively. The PA for GvHD was 12% lower in c2 than c1 but the proportion of GvHD cases was higher in c2; these same differences were also seen for organ failure CSD. BI showed the rate of movement between CSDs was almost identical in both cohorts. In c1, TC and CP were >2 fold less likely to agree on CSD for both ALL and MDS (p<.05) when URD-HCT was in the early 2000s. The log-linear models also showed c1 cohort effects, as well as dependencies in both cohorts between age and TC CSD and CP CSD and disease status (early/advanced). The cohort effect and age dependencies with TC CSD in c1 were not present when a systematic review of cases by the CP was performed. In c2 the TC CSD did not show evidence of a cohort effect; however a greater understanding of the lower AC1 in c2 vs. c1 was gained through use of PA. In both cohorts the BI and log-linear models show consistency in CSDs as determined by the CP. Our work illustrates that CPs are a worthwhile part of genomic studies of complex multi-factorial phenotypes, and the use of agreement statistics and dependency models is imperative to monitor CP phenotype reliability and understand data structure for subsequent analyses. Citation Format: Lara Sucheston-Campbell, Theresa Hahn, Leah Preus, Kenan Onel, Xiaochun Zhu, Song Liu, Li Yan, Alyssa Clay, David Tritchler, Marcelo Pasquini, Philip McCarthy. Adjudicating complex phenotypes for use in GWAS: A study of transplant-related mortality (TRM) after unrelated donor hematopoietic cell transplantation (URD-HCT). [abstract]. In: Proceedings of the 105th Annual Meeting of the American Association for Cancer Research; 2014 Apr 5-9; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2014;74(19 Suppl):Abstract nr 5079. doi:10.1158/1538-7445.AM2014-5079