Based on the Lung Health Study (LHS), a customized FEV1 prediction model was recently established. However, a continuous examination of the efficacy of prediction models in realistic contexts outside of clinical trial design is required. For a study, researchers sought to assess the external validity of the LHS FEV1 prediction model in a sample of nonsmokers with COPD.
They used data from the Canadian Cohort of Obstructive Lung Disease (CanCOLD), an ongoing prospective, longitudinal, multicenter population-based cohort research, to identify ever-smokers with spirometrically verified COPD. At follow-up visits, they compared the mean anticipated FEV1 to the mean observed FEV1. The root means square error (RMSE) and actual coverage probability of the 95% personalized prediction intervals were determined.
The final CanCOLD sample included 360 nonsmokers with COPD who contributed 970 FEV1 observations over an average of 3 years (standard deviation (SD): 0.6 years). The mean observed vs. expected FEV1 for the first visit (1.67 years from baseline) was 2.28L vs. 2.28L, while the mean observed vs. anticipated FEV1 for the second visit (3.13 years from baseline) was 2.19L vs. 2.18L. The forecasts’ RMSE was 0.205L, while the 95% prediction range coverage probability was 93%. The model performed similarly across many subgroups.
The model performed well in predicting FEV1 in both the overall population and specific CanCOLD subsets. In addition, the findings showed that LHS forecasts are accurate for at least three years in the general COPD population.