Prediction Models for Risk of Cardiorespiratory Morbidity/Mortality and Fracture Among Young Adults With Cerebral Palsy

Background: There is a dearth of screening tools for cardiorespiratory disease and fracture risk, such as risk prediction models, for adults with cerebral palsy (CP). There is heterogeneity of pathophysiology related to the severity of CP and aging, such that a suite of risk prediction models may be needed. Differentiating by sarcopenia versus frailty syndromes may be a useful, physiologic-based framework to develop a suite of cardiorespiratory disease and fracture risk prediction models for adults with CP. The study objective was to determine if risk prediction models including widely available variables that are CP non-specific and that may capture the physiologic components of frailty provide clinically useful prediction of salient health issues for young adults with CP. Methods: This retrospective cohort study used medical records from 01/01/2012 to 10/2/2022 from 18-40-year-olds with CP at a single Medical Centre. Logistic regression models were developed for three separate outcomes: 3-year risk of respiratory morbidity/mortality, cardiovascular morbidity/mortality and fracture. The following predictors were included: age, sex, intellectual disabilities, epilepsy and four serum biomarkers (creatinine, glucose, calcium, carbon dioxide) from the clinical Basic/Comprehensive Metabolic Panel assay. Model performance measures were evaluated, including discrimination (c-statistic) and calibration. Internal validation was assessed, and optimism-corrected c-statistics were computed using the bootstrap resampling method to assess model overfitting. Results: There were 805 young adults with CP with an average (SD) age of 26.5 (6.6) years and 47.8% were female. Over the 3-year follow-up, 24.6% had incident respiratory morbidity/mortality, 8.9% had incident cardiovascular morbidity/mortality and 7.0% had an incident fracture. The c-statistic (95% CI) was 0.74 (0.70-0.78) for respiratory morbidity/mortality, 0.63 (0.57-0.70) for cardiovascular morbidity/mortality and 0.65 (0.58-0.73) for fracture. The optimism-corrected c-statistic was similar for respiratory morbidity/mortality (0.73) but lower for cardiovascular morbidity/mortality (0.58) and fracture (0.59), suggesting evidence of model overfitting. All models showed good calibration based on the slope of the predicted risk versus observed risk around 1.0 and the Hosmer-Lemeshow goodness-of-fit test, P = 0.305-0.903. However, the range of predicted risks was limited to similar to 20% for cardiovascular morbidity/mortality and similar to 55% for fracture, suggesting that these models have limited value for those with greater risk. Conclusions: Using widely available, CP non-specific clinical variables, a well-calibrated model was developed to predict 3-year risk of respiratory morbidity/mortality among young adults with CP (discrimination, similar to 73%). The predictor set appeared less useful for predicting 3-year risk of cardiovascular morbidity/mortality and fracture in this cohort.