Quantitative modelling to predict candidates for outpatient blood and/or marrow transplantation (BMT)

There are no quantitative models identifying candidates for outpatient BMT. This study's objective was to determine if pre-transplant characteristics could define good and poor outpatient BMT candidates. We retrospectively reviewed 234 consecutive inpatient BMT admissions to RPCI between 1/1/96 and 12/31/98 and categorized patients as either inpatient or outpatient candidates based on infectious and treatment-related acute toxicities (regimenrelated toxicity and GvHD). 72% (94 of 131) of autologous BMT (autoBMT) admissions qualified as outpatient candidates compared to 45% (34 of 75) of related allogeneic BMT (RalloBMT) admissions and 7% (2 of 28) of unrelated allogeneic BMT (UalloBMT) admissions. 23 treatment and patient characteristics were considered as predictors for outpatient BMT candidates by recursive partitioning analysis (RPA) and logistic regression (LR) modelling, As determined by RPA, good outpatient BMT candidates had the following characteristics: 1) AutoBMT and Karnofsky Performance Score (KPS) ≥ 90 (p=0.007); 2) AutoBMT, KPS ≤90 and body mass index (BMI) ≥30 kg/m2 (p=0.0002); 3)RalloBMT and KPS ≥80 (P=0.04) 4) RalloBMT, KPS ≥80 and no prior therapy (p=0.07); 5) RalloBMT, KPS ≥ 80, any prior therapy, AST≤ 33 U/L and female gender (p=0.06). Poor outpatient BMT candidates had the following characteristics: 1 ) UalloBMT (p=0.0003); 2) RalloBMT and KPS < 80 (p=0.04); 3) RalloBMT, KPS > 80, any prior therapy and AST > 33 U/L (p=0.06). RPA identified very high and low risk patients for outpatient therapy at a higher sensitivity, specificity and accuracy when compared to standard LR. Sensitivity Specificity Accuracy RPA - auto and RalloBMT 83% 76% 80% LR - RalloBMT 65% 68 67% LR-autoBMT 98% 8% 73% Subgroup analysis by RPA of larger populations and the addition of other pre-transplant variables should allow for a more detailed prediction of good and poor risk populations. Model development such as the one presented can assist clinicians with predicting patient outcomes as well as facilitate resource untilization and hospital planning for the future of transplant programs.

Duke Scholars

Author Kathleen Donohue Biology