Prediction of Skin Trouble in Patients Undergoing Allogeneic Hematopoietic Stem Cell Transplantation Using Generalized Additive Model

Topic Significance & Study Purpose/Background/Rationale: Although skin trouble (ST) is a major problem in patients undergoing stem cell transplantation (SCT), there are few reports on the prediction of ST development in SCT units. One of the difficulties in conducting a study of the SCT setting is that variables are highly dependent on the post-transplant day (time-dependent variables), and that the post-transplant day does not always behave as a linear function. In this study, we aimed to predict ST development in SCT patients using a generalized additive model (GAM), an extended model originally used in social science. Methods, Intervention, & Analysis: This retrospective study involved 81 consecutive patients undergoing SCT at Hyogo College of Medicine between April 2012 and March 2013. Among them, 28 patients developed ST (ST group), and 53 patients did not (Control group). On applying GAM, a multistate model was used for data collection in order to avoid overlearning. We determined the following events as states in our multistate model, and collected the time-dependent variables on the day of each event, consisting of diarrhea, need for oxygen supply, hemorrhagic cystitis, skin GVHD, encephalitis, and disease relapse. The time-dependent variables consisted of the serum level of albumin, ALT, CRP, blood glucose, creatinine, hemoglobin, WBC, body temperature, blood pressure, body weight, daily activity score (DAS), and the Functional Independence Measure score. Findings & Interpretation: The age, gender, underlying disease, and donor type were almost identically distributed between the ST group and control groups. Among the events described above, diarrhea, need for oxygen supply, and hemorrhagic cystitis were observed more frequently in the ST group. GAM revealed that albumin, blood glucose, DAS, and the post-transplant day were significant factors affecting ST development. Albumin, blood glucose, and DAS were linear, but the post-transplant day described a reversed U-shaped curve showing that the most susceptible time to ST development was around day 30 after SCT. Discussion & Implications: GAM enabled calculation of the probability of ST development with the values of albumin, blood glucose, DAS, and the post-transplant day. GAM could be a powerful tool for prediction analysis involving non-linear, time-dependent variables in an SCT setting.