Application of latent class analysis in diagnosis of graft-versus-host disease by serum markers after allogeneic haematopoietic stem cell transplantation

Graft-versus-host disease (GVHD) is one of the major causes of morbidity and mortality in 25–70% of patients. The gold standard (GS) test to confirm the diagnosis of GVHD has some limitations. The current study was conducted to evaluate the accuracy of three serum markers in diagnosing GVHD without a GS. 94 patients who were hospitalized for allogeneic transplantation were studied. Mean levels from day of haematopoietic stem cell transplantation (HSCT) to discharge of serum uric acid (UA), lactate dehydrogenase (LDH), and creatinine (Cr) were measured for all participants. We adapted a Bayesian latent class analysis to modelling the results of each marker and combination of markers. The Sensitivity, Specificity, and area under receiver operating characteristic curve (AUC) for LDH were as 51%, 81%, and 0.70, respectively. For UA, the Sensitivity, Specificity, and AUC were 54%, 75%, and 0.71, respectively. The estimated Sensitivity, Specificity, and AUC of Cr were 72%, 94%, and 0.86, respectively. Adjusting for covariates, the combined Sensitivity, Specificity, and AUC of the optimal marker combination were 76%, 83%, and 0.94, respectively. To conclude, our findings suggested that Cr had the strongest diagnosis power for GVHD. Moreover, the classification accuracy of the three-marker combination outperforms the other combinations.