JMIR Publications

ABSTRACT

Background:

Disease modifying therapies ameliorate disease severity of sickle cell disease (SCD), but hematopoietic cell transplantation (HCT) and more recently autologous gene therapy are the only treatments that have curative potential for sickle cell disease (SCD). While registry-based studies provide population-level estimates they do not address the uncertainty regarding individual outcomes of HCT. Computational machine learning (ML) has the potential to identify generalizable predictive patterns and quantify uncertainty in estimates thereby improving clinical decision-making. There is no existing ML Model for SCD and ML models for HCT for other diseases focus on single outcomes rather than all relevant outcomes.

Objective:

Address the existing knowledge gap by developing, and validating an individualized ML-prediction model, sickle cell predicting outcomes of hematopoietic cell transplantation (SPRIGHT), incorporating multiple relevant pre-HCT features to make predictions of key post-HCT clinical outcomes.

Methods:

We applied a supervised random forest ML model to clinical parameters in a de-identified CIBMTR dataset of 1641 patients who underwent HCT 1991-2021 and followed for a median of 47.8 months (0.3-312.9). We applied forward and reverse feature selection methods to optimize a set of predictive variables. To counter the imbalance bias towards predicting positive outcomes due to the small number of negative outcomes we constructed a training dataset taking each outcome variable of interest, and performed a two-times repeated 10-fold Cross-Validation. SPRIGHT a web-based individualized prediction tool accessible by smartphone, tablet, or personal computer. It incorporates predictive variables of age, age group, Karnofsky/Lansky score, co-morbidity index, recipient CMV seropositivity, history of ACS, need for exchange transfusion, occurrence and frequency of vasocclusive crisis (VOC) before HCT, and either a published or custom chemotherapy/radiation conditioning, serotherapy, and GVHD prophylaxis. SPRIGHT makes individualized predictions of overall survival (OS), Event Free Survival (EFS), Graft Failure (GF), acute graft versus host disease (AGVHD), chronic graft versus host disease (CGVHD), occurrence of VOC or stroke post-HCT.

Results:

The discrimination predictive performance of SPRIGHT evaluated using the area under the curve (AUC), accuracy, and balanced accuracy met or exceeded published predictive benchmarks with AUC for OS (0.7925), EFS (0.7900), GF (0.8024), AGVHD (0.6793) CGVHD (0.7320), VOC post-HCT (0.8779). SPRIGHT revealed good calibration with slope 0.87-0.96, with small negative intercepts (-0.01-0.03), for 4 out of the 5 outcomes. However, OS exhibits non-ideal calibration which may be reflective of the overall high OS in all sub-groups.

Conclusions:

A web-based ML- prediction tool incorporating multiple clinically relevant predictors predicts key outcomes of HCT for SCD has high predictive performance and has potential use in shared decision-making Clinical Trial: Not applicable. This not a clinical trial.