NEW ORLEANS, La.—Researchers have developed a clinical risk score to identify patients at the greatest risk for select acute graft versus host disease (aGVHD) grades in the 100 days following transplant, a study found.
The findings showed that the odds of developing aGVHD grades 2 to 4 by day 100 after transplant were 3.1 (P <.0001) for the greater than 75th percentile cohort, 2.0 (P <.0001) for the 50th to 75th percentile cohort, and 1.50 (P <0.0001) for the 25th to 50th percentile cohort in comparison with the 25th or less percentile cohort in the validation group, reported Caden Ulschmid and colleagues of the Medical College of Wisconsin in Milwaukee.
The adjusted day-100 probability of aGVHD grades 2 to 4 was 53% in the greater than 75th percentile cohort and 26% in the 25th and less percentile cohort, they also found.
The findings were presented at the 64th ASH Annual Meeting and Exposition from December 10 to 13, 2022.
Read more to learn about treatments for paroxysmal nocturnal hemoglobinuria
“Acute graft versus host disease (aGVHD) is a complication seen following allogeneic hematopoietic cell transplantation (allo-HCT) that contributes to significant morbidity and mortality. Over the years, a number of studies have been conducted to determine the risk factors, severity, treatment strategies, and clinical outcomes of aGVHD,” said Ulschmid. “However, to our knowledge, there is no widely inclusive, validated, clinical based model available for predicting the risk of developing acute GVHD following allogeneic HCT,” he continued.
“Here, we have created the first widely inclusive validated clinical scores for the development of aGVHD using both traditional logistic regression and machine learning methods. These scores can guide personalized clinical decision-making, patient counseling, and design of clinical trials for testing novel prophylactic interventions,” Ulschmid said.
The researchers set out to generate and authenticate a clinical risk score to determine which patients might have a significantly different risk of developing aGVHD grade 2 to 4 and 3 to 4 by day 100 after transplant. They then assessed the traditional logistic regression model and the machine learning approach of Bayesian additive regression trees (BART) and compared the performance of each.
The primary outcome was aGVHD grade 2 to 4. The secondary outcome was aGVHD grade 3 to 4. Both of these measures were taken as event rates 100 days following transplant.
There were 15,258 patients in the training cohort and 6538 patients in the validation cohort. All participants underwent allo-HCT from 2008 to 2019. The median age was 55 years, and males accounted for 59% of the study group.
A number of disease groups were represented among the study’s cohorts, including acute myeloid leukemia with intermediate cytogenetics, non-Hodgkin lymphoma, acute lymphocytic leukemia , acute myeloid leukemia with poor cytogenetics, and myelodysplastic syndrome.
The following conditioning regimen was used: myeloablative conditioning with cyclophosphamide and busulfan, reduced-intensity conditioning with fludarabine with melphalan, myeloablative conditioning with busulfan and fludarabine, reduced-intensity conditioning with busulfan, and nonmyeloablative with cyclophosphamide, fludarabine, and total body irradiation.
Patient-associated variables assessed included race, ethnicity, hematopoietic cell transplantation-specific comorbidity index, age, and Karnofsky performance status. Disease-associated factors included disease status at allo-HCT, disease type, and cytogenetics.
aGVHD prophylaxis, donor-recipient sex matching, in vivo T-cell depletion status, donor-recipient CMV serostatus, donor-recipient ABO matching, conditioning regimen, and graft and donor type were among the transplant-associated variables included.
Training sets were used to generate models. Logistic regression with a stepwise selection technicque was usd to pick prognostic factors for each outcome. Based on the size of their odds ratios, weighted scores were given to each variable related to each outcome.
The BART model was employed to accommodate the training information and make projections on the validation information.
Risk scores fell into 4 categories, including 25th, 50th, and 75th percentiles from the training set as cut off points, and the relationship of the risk scores was tested using the independent validation group.
When compared with the 25th or less percentile cohort in the validation group, the odds of developing aGVHD grade 3 to 4 by day 100 after transplant were 3.2 (P <.0001) in the greater than 75th percentile cohort, 2.0 (P <.0001) in the 50th to 75th percentile cohort, and 1.4 (P =.0043) in the 25th to 50th percentile cohort.
Other results showed that the adjusted day-100 probability of aGVHD grade 3 to 4 was 24% in the greater than 75th percentile cohort and 9% in the 25th or less percentile cohort.
These data also demonstrated that there was no statistically significant difference between aGVHD grade 3 to 4 (P =.99) and aGVHD grade 2 to 4 (P =.078) when comparing the performance of logistic regression-based models and BART-based models using concordance index.
“The BART-based machine learning models showed similar performance to the traditional statistical-based models, but provided additional information on variable importance. These models will be used to build a risk score calculator for the CIBMTR website,” Ulschmid concluded.
Ulschmid C, Li X, Wang T, et al. Development of a validated clinical risk score to predict the incidence of acute graft versus host disease after allogeneic hematopoietic cell transplantation: a CIBMTR analysis. 64th ASH Annual Meeting and Exposition. New Orleans, LA; December 10-13, 2022.