Prediction of Health Preference Values from CD4 Counts in Individuals with HIV. (May 2013)

• Record Type: Journal Article
• Authors: Isogai, Pierre K.; Rueda, Sergio; Rachlis, Anita R.; Rourke, Sean B.; Mittmann, Nicole
• Is Part Of: Medical decision making. Volume 33:Number 4(2013)
• Page Start: 558
• Page End: 566
• Publication Date: 2013-05
• Subjects: CD4 count -- health preference -- HIV -- health utility score -- model; Medical policy -- Periodicals; Clinical medicine -- Decision making -- Periodicals; Medicine -- Periodicals; Médecine clinique -- Prise de décision -- Périodiques; 362.1
• DOI: 10.1177/0272989X12453499
• Languages: English
• ISSNs: 0272-989X

Abstract

Background . A common measure of health benefit in technology assessments is the quality-adjusted life year, which incorporates health preference or utility scores.Objective . To build and test a predictive model using CD4 counts to derive health preference scores.Design . Predictive modeling.Setting . Ontario HIV Treatment Network Cohort Study.Measurement . The relationship between HUI3-derived health preference score and HIV health status measured by CD4 count was examined by a regression model. Additional independent variables considered included age, time since HIV diagnosis, AIDS-defining condition, sex, and education level. A polynomial regression model was fit to predict health preference scores. The final model was established using automated backwards stepwise variable elimination using the Akaike information criterion. Tenfold cross-validation was used to assess the model.Results . Data from 841 participants were available. Mean age and time since diagnosis were 46.78 and 11.03 years, respectively. CD4 counts ranged from 2 to 995 cells per mm 3 with 267 (31.75%) individuals having less than 350 cells per mm 3 . Mean HUI3 utility score was 0.72 and ranged from −0.25 to 1. The final model retained squared terms for CD4 counts, age, and time since HIV diagnosis and eliminated history of AIDS-defining condition and the nonsquared time since HIV diagnosis. Prediction error was assessed in 14 subgroups using the validation set. Two subgroups had mean prediction errors greater than 0.02.Limitations . All statistical models are limited by the data used to develop and test the model. The model estimates health utility scores primarily through CD4 counts. Therefore, the model may be inappropriate if noninfectious diseases are a significant factor.Conclusions . Results provide a model for predicting health preference values from CD4 counts.