Risk prediction for ALS using semi-competing risk models with applications to the ALS Natural History Consortium dataset.

Abstract

Background and objectives: Important landmarks in progression of amyotrophic lateral sclerosis (ALS) can occur prior to death. Predictive models for the risk of these events can assist in clinical trial design and personal planning. We propose a predictive model, using a semi-competing risks modeling approach, for five important disease progression landmarks in ALS. Methods: Data on 1508 participants from the ALS Natural History Consortium (ALS NHC) were used, including baseline characteristics and the ALS Functional Rating Scale-Revised (ALSFRS-R) score collected at clinic visits. A semi-competing risks modeling approach was used to study the time to disease progression landmarks, accounting for the possibility of death. Specifically, time to gastrostomy, use of noninvasive ventilation (NIV), continuous use of NIV, loss of speech, and loss of ambulation were chosen and modeled individually. To measure the predictive capabilities of the model, the integrated Brier score was computed for each model using cross-validation for the NHC data. Data from Emory University were used for external validation of the models. Results: We present model results using gastrostomy as the intermediate outcome. Similar trends in disease progression groups were found across all model pathways. Diagnostic delay, age, and site of onset were the most important covariates. Predictive metrics in both internal and external validation are presented across all models and for different pathways. Conclusion: Semi-competing risks modeling is a flexible approach to studying disease progression. The models have good predictive capabilities across different outcomes and pathways. These are replicated in the external validation dataset.