Context: Available diabetes risk calculators were developed for able-bodied individuals, but their metabolic profile is different from individuals with spinal cord injury.
Objectives: We aimed to develop a diabetes risk assessment tool specific to individuals with spinal cord injury.
Methods: We used national Veterans Affairs data to identify patients with at least a 2-year history of spinal cord injury and no prior history of diabetes with a Veterans Heath Affairs visit from 2005-2007, and followed the 11,054 individuals that met inclusion criteria for up to 17 years to assess diabetes development. We used least absolute shrinkage and selection operator (LASSO) Cox regression to develop prognostic diabetes prediction models and evaluated these models on discrimination and calibration.
Results: 2937 subjects developed diabetes during follow-up; median follow-up time was 8.7 years (IQR 3.3, 15.4). The first model selected 17 predictors and demonstrated median discrimination of 0.70 (IQR 0.69, 0.72) at 15 years. The second, more parsimonious model with 4 selected predictors demonstrated median discrimination of 0.69 (IQR 0.68, 0.71) at 15 years. Both models demonstrated good calibration across predicted risk, with better calibration in the 17-predictor model.
Conclusion: These spinal cord injury-specific risk calculators can be used by both patients and providers in assessing risk of diabetes development, and in shared decision making regarding surveillance and prevention.
Keywords: Prevention; Risk prediction; Spinal cord injury; Surveillance; Type 2 diabetes.