The Composite Health Risk Assessment Model (CHARM) to Predict 1-Year Non-Relapse Mortality (NRM) Among Older Recipients of Allogeneic Transplantation: A Prospective BMT-CTN Study 1704

Introduction: Allogeneic hematopoietic cell transplantation (HCT) remains the only therapy for long-term disease control for many high-risk hematologic malignancies (HM). HCT use in older HM patients (pts) continues to increase. However, concerns of excessive transplant-related morbidity and non-relapse mortality (NRM) limit referrals and broader application of curative intent HCT. The HCT-comorbidity index (HCT-CI) was initially developed to improve risk-stratification of NRM. More recently, geriatric assessment (GA) and biomarkers have emerged as promising additional tools that may refine estimates of these risks. We hypothesized that a combination of health assessments by GA and biomarkers would constitute a robust and valid model, CHARM, for accurate personalized estimation of one-year (1-yr) NRM. Here, we report the results of the largest, first of its kind, prospective study of older recipients of HCT through the Blood and Marrow Transplant Clinical Trials Network (BMT CTN) 1704 study (NCT03992352).

Patients and Methods: Adults aged ≥60 years (yrs) with HM institutionally eligible for HCT were enrolled (n=1226) from 49 centers in the US between 2019 - 2021. Within 21 days of conditioning, 13 prospectively identified older pt.-specific health variables, informed from prior studies, were collected: age, HCT-CI, % of weight loss over the past yr, pt reported Karnofsky performance status, PROMIS physical function scale, instrumental activities of daily living (IADL), number of falls, PROMIS depression scale, number of prescribed medications, 4-meter walk speed, cognition by the Montreal Cognitive Assessment (MoCA), C-reactive protein (CRP) and albumin. The CHARM score to predict NRM within 1 yr was built using a multivariable (MVA) Fine-Gray model, with multiple imputation to handle modest missingness in covariates and grouped penalized variable selection using SCAD to identify variables to retain in the model. Final variables were used as continuous. The model further considered adjustment for conditioning intensity, CMV serostatus, and donor type. Area under the ROC curve (AUC) was calculated to validate the CHARM and compare to HCT-CI alone, with bootstrap sampling to correct for optimism in the within-sample AUC. MVA of overall survival (OS) within 1 yr used the CHARM score and adjusted for other significant variables derived from stepwise regression.

Results: The primary analysis included 1105 pts who were confirmed eligible and proceeded to transplant. Median age was 67 (range 60-82) and 32% were ≥ 70 yrs. AML (45%) and MDS (30%) were the most common indications and matched donors the most frequent donor type at 72%. Reduced intensity conditioning with fludarabine (flu)-melphalan (38%) and flu-busulfan (20%) were the most common regimens. Baseline geriatric vulnerabilities among evaluable pts were frequent including slow walk speed (<0.8 meter/second) in 22%, any IADL limitation in 37%, and cognitive impairment (MoCA < 23) in 12%. 1-yr NRM was lower than expected at 14.4% with 1-yr OS of 72%.

In the MVA model, albumin, CRP, HCT-CI, and weight loss were independently associated with NRM (Table) and comprised the CHARM. Of note, increasing age did not significantly affect NRM. The CHARM had within sample and bootstrap corrected AUC of 0.641 and 0.607 whereas the HCT-CI alone achieved corresponding AUC values of 0.601 and 0.598. NRM by tertiles is shown in Figure. Each point increase in the CHARM (median -2.2504, range -3.6327 to 0.1381), was associated with a hazard ratio (HR) of 2.83 (95% CI 2.12 – 3.78, p < 0.0001) for NRM. The CHARM also predicted overall mortality within 1-yr with a HR of 2.04 (1.65– 2.53, p <0.0001) per one point increase, independent of the disease risk index, which had a HR of 1.5 (95% CI 1.12 – 1.99, p =0.0059) for high/very high vs low/intermediate categories.

Conclusions: We report on the first prospectively established tool, CHARM, to risk-stratify older adult HCT comprised of simple and readily available parameters in transplant and oncology clinics. Adopting CHARM in practice may promote HCT referrals and access for older pts, given low risks of NRM among CHARM lower 2/3^rd tertiles. Identifying high-risk CHARM pts pre-HCT will promote developing novel strategies to reduce NRM. Further analysis of the impacts of the 13 health assessment variables on geriatric morbidity such as disability trajectories is in-process.