Developing a Risk Prediction Model for Long-Term Physical and Psychological Functioning after Hematopoietic Cell Transplantation

• We propose risk prediction models for long-term functioning in hematopoietic stem cell transplantation survivors
• Readily accessible, pretransplantation predictors can predict long-term functioning
• Accuracy of the prediction models should be improved for use in clinical practice

Hematopoietic cell transplantation (HCT) is associated with impaired physical and psychological functioning for some long-term survivors. A risk prediction model would help clinicians estimate their patients' physical and psychological functioning after HCT and determine when to refer to added supportive care when appropriate. The purpose of the present study was to develop risk prediction models for physical and psychological functioning in HCT survivors. This was a secondary analysis of data from a randomized controlled trial (NCT00799461) that included 3- to 10-year HCT survivors. Risk predictions for physical and psychological functioning were developed by using backward logistic regression. The models were internally validated using bootstrapping techniques. Regression coefficients were converted into easy-to-use risk scores. Finally, the sensitivity, specificity, and positive and negative predictive values of the total risk score were calculated. The analyses included 489 survivors, with a mean age of 45.6 (SD, 12.4) years; 47% were female and with a mean of 6.1 years (SD, 2.0) after transplantation. Younger age, higher body mass index (BMI), no or part-time work, more comorbid diseases, autologous transplantation, and chronic graft-versus-host disease predicted impaired physical functioning. Female gender, younger age, higher BMI, not living with a partner, autologous transplantation, and chronic graft-versus-host disease predicted impaired psychological functioning. Although both models had predictive value for long-term functioning, diagnostic accuracy was moderate. For the physical functioning receiver operating characteristic, area under the curve (AUC) after internal validation was .74 with sensitivity 51.9 and specificity 82.8 at the optimal cut-off. For psychological functioning, AUC after internal validation was .69 with sensitivity 83.3 and specificity 42.9 at the optimal cut-off. We conclude that it is possible to predict long-term physical and psychological functioning with readily accessible, mostly pretransplantation predictors. The accuracy of the risk prediction models can be improved for use in clinical practice, potentially by adding pretransplantation patient-reported functioning and comorbidities.