Prediction of outcomes after second-line treatment for acute graft-versus-host disease

Abstract

Acute graft-versus-host disease (GVHD) requiring second-line treatment represents a highly morbid complication of allogenic hematopoietic cell transplantation (HCT). Recent studies have defined short-term outcomes after second-line treatment for acute GVHD, but longer-term outcomes have not been well defined. We examined overall survival (OS) and failure-free-survival (FFS) of 216 patient who had HCT who received second-line treatment for acute GVHD. Failure time for FFS was defined as the earliest of death, relapse, or implementation of third-line treatment. Multivariable Cox regression was used to identify risk factors for mortality and failure, and predictive models were derived for 6- and 12-month mortality. Point estimates of OS at 6 and 12 months were 59% (95% confidence interval [CI], 52-65) and 52% (95% CI, 45-68), respectively. Point estimates of FFS at 6 and 12 months were 42% (95% CI, 35-48) and 37% (95% CI, 31-43), respectively. Predictive models for both end points included serum albumin and total bilirubin concentrations at the onset of second-line treatment, patient age at onset of second-line therapy, and a combination of abdominal pain/stage 4 gut involvement. Optimism-corrected areas under the receiver-operator characteristic curve and Brier scores were 77.4 and 0.169 for 6-month mortality, respectively, and 80.0 and 0.169 for 12-month mortality. We identify risk factors associated with mortality and failure after second-line treatment of acute GVHD, provide historical benchmarks for assessment of FFS and OS in other studies, and propose predictive models for 6- and 12-month mortality that could be used to generate population-specific benchmarks.

Graphical abstract

Figure 1.

Third-line treatment and non-relapse mortality without third-line treatment account for most failures after second-line treatment. (A) Survival, (B) FFS, and (C) causes of failure after second-line treatment of GVHD. Dotted lines above and below the Kaplan-Meier survival plot (solid line) show the pointwise 95% CI.

Figure 2.

Statistical models predict the probability of mortality at 6 and 12 months after second-line treatment. Calibration curves for mortality at (A) 6 and (B) 12 months after starting second-line treatment with (C) an example for a hypothetical individual patient. The solid blue line shows the fit between the observed probabilities of mortality (y axis) with the estimated probabilities of mortality as predicted by the model (x axis) (see Methods for details). The fit is obtained by using locally estimated scatterplot smoothing. This approach is like standard linear least-squares regression, but this simpler model is fit to localized subsets of the data, leading to a more flexible representation of the fit between the predicted and observed outcomes than could be achieved assuming a linear association across the entire span of the data. The dashed red line shows results that would be expected for a perfect correlation. The shaded area shows the pointwise 95% CIs of the observed probabilities of mortality across the range of predicted probabilities. In the example, values for each risk factor are multiplied by the respective coefficients. The sum of the products plus the intercept is used to predict the probability of mortality from the formula in the bottom row. In this example, the predicted probability of mortality is 0.39 at 6 months and 0.58 at 12 months.