Setting and objective: The Antiretroviral Therapy (ART) Cohort Collaboration published models predicting progression to AIDS or death (the complement of AIDS-free survival) and death (the complement of absolute survival). The objective is to validate the model on independent data from CASCADE.
Study design: Discrimination was assessed using concordance statistics, and calibration was examined by comparing predicted survival curves with the corresponding Kaplan-Meier estimates. Accuracy was assessed by comparing predicted percentage probability of survival with the Kaplan-Meier estimate at yearly intervals after start of therapy.
Results: There was little loss of model discrimination when applying the model to CASCADE. Overall predicted calibration curves agreed with Kaplan-Meier survival curves. Predicted probabilities of AIDS or death at 3 years after starting HAART ranged from 4.3% in the low-risk group to 20.5% in high-risk patients, with corresponding Kaplan-Meier estimates ranging from 4.0% to 18.3%; for death predictions, the probabilities ranged from 1.2% to 7.3% and estimates from 1.1% to 8.6%.
Conclusion: The predictions from the model agree with observed outcomes in CASCADE to within the 95% upper and lower Kaplan-Meier estimates. The prognostic model appears to be accurate in terms of discrimination and calibration, giving reliable and transportable predictions up to 3 years after the start of HAART.