Antibiotic risk score for acute graft-versus-host disease Free

Introduction: Certain antibiotic exposures have been associated with increased rates of acute GVHD (aGVHD) after alloHCT, likely due to their effects on the gut microbiota. Using data from 2,023 alloHCTs performed at our center (2010-2021), we recently developed an antibiotic risk score based on antibacterial antibiotic exposure between days -7 and +30 (JAMA Netw Open 2023;6:e2317188). The objective was to estimate the added risk for aGVHD due to early exposures to antibiotics. Here, we validated the score in an independent, more recent cohort.

Methods: Data from 297 consecutive alloHCT recipients (2022-2023) including 8,382 antibacterial antibiotic use records between days -7 and day +30 were studied. Eligibility criteria mirrored those in the previous cohort (age≥18, first T-replete alloHCT, and 100 days follow-up) and there were no changes in our institutional antibiotic practice. The outcome measures were grade III-IV (severe) and grade II-IV aGVHD. The marginal structural model developed in our previous analysis was used, without alteration, for validation on the new database. The model estimates the association of time-dependent antibiotic exposures with GVHD hazard in the presence of potentially confounding, previous time-dependent antibiotic exposures. The model included 9 most frequently used classes of antibacterial antibiotics, binarily coded interval-specific exposures (five ~weekly intervals), and 6 non-antibiotic covariates known to influence the risk of GVHD (graft source, conditioning intensity, ATG use, donor type, GVHD prophylaxis regimen, neutrophil engraftment). We applied the original model coefficients to the current database to calculate, for each patient, a total non-antibiotic risk score and an antibiotic risk score using their non-antibiotic covariates and interval-specific antibiotic exposures, respectively. Next, we classified patients into the antibiotic high-risk group if the antibiotic score was in the upper 25% quantile and antibiotic low-risk group otherwise. We fitted a multivariable Fine and Gray model on the occurrence of aGVHD, treating non-GVHD death as a competing risk. The main predictor was the antibiotic risk group, and the non-antibiotic score was included as a covariate. To ensure the models were robust to the choice of score quantiles, we performed sensitivity analysis using different quantiles.

Results: Median (range) age was 61 (19-78) years. 175 (59%) patients were male. Thirty-five (12%) and 162 (55%) patients developed grade III-IV and II-IV aGVHD, at a median of 27 and 35 days after HCT, respectively. In our previous cohort, these rates were 15% and 72%, respectively. This decline in aGVHD rates over time is consistent with more frequent use of post-transplant cyclophosphamide in the more recent cohort (38% vs. 16%, respectively). An antibiotic score in the upper 25% quantile was associated with a HR of 2.37 (95%CI:1.21-4.63) for severe aGVHD compared to the lower 75% quantile (P=0.01). The cumulative incidence of severe aGVHD in the antibiotic high- and low-risk groups was 20.3% (95%CI:11.0-29.5) and 9% (95%CI:5.2-12.7), respectively. In sensitivity analysis, a wide range of quantile thresholds (12% to 30%) were strongly predictive of risk (HR>2). The same patterns, albeit weaker and not statistically significant, were observed for grade II-IV aGVHD (HR:1.30, 95%CI:0.92-1.84; P=0.13). Next, we performed a deeper analysis of non-antibiotic variables to see whether the observed association between antibiotic risk and aGVHD might be confounded by non-antibiotic variables known to be independently associated with aGVHD. The non-antibiotic score, capturing most such variables, was similar between high- and low-risk groups (P=0.39). We examined two additional variables not included in the non-antibiotic score: disease risk index and donor age. Again, we found no significant differences between the groups (P=0.48 and 0.39, respectively). The match between antibiotic high- and low-risk groups for these non-antibiotic variables supports an association between antibiotics and aGVHD independently of other measured variables.Conclusions: We validated a recently developed antibiotic score for added aGVHD risk. A free online antibiotic risk calculator was developed, identifying antibiotic exposures associated with higher added risk for severe aGVHD. When possible, such exposures should be avoided. When not possible, exposed patients should be monitored for aGVHD more closely.