MODEL TO PREDICT HEALTH UTILITY SCORES FROM CD4 COUNT IN HIV

Purpose:   To develop a model that predicts health utility scores from CD4 count data for use in economic evaluations of HIV treatments.

Method:   Clinical, demographic and quality of life data was available from the Ontario HIV Treatment Network (OHTN) Cohort Study (OCS).  Data is collected at 11 actives sites across Ontario.  Data from the OCS included: EQ-5D; CD4 count; age at interview; time since HIV diagnosis; sex; prior or current AIDS defining condition (yes or no); and current treatment status (yes or no).  A regression model predicting EQ-5D derived utility scores was developed.  Both linear and squared terms were included in the model for the three continuous independent variables to account for possible non-linear relationships.  The final model was determined by an automated stepwise variable elimination using AIC.  Ten-fold cross-validation was used to assess the model fit.  A 3,000 replicate bootstrap was used to derive 95% confidence intervals (CIs).

Result:   Data was available from 1,051 individuals.  The cohort contained a wide range of individuals.  CD4 counts ranged from 2 to 1,631 per mm³, EQ-5D derived utility scores ranged from 0.20 to 1 and age at interview ranged from 18 to 85 years.  All variables were included in the initial regression model.  The final reduced model included the variables: CD4 count (linear and squared); age (linear and squared); time since HIV diagnosis; and sex.  The mean actual and predicted utility scores for the whole cohort was 0.83.  The 95% CI of the difference was [-0.01, 0.01].  For the subgroup of 357 individuals with a past or current AIDS defining condition, the mean [CI] difference in predicted and actual utility scores was 0 [-0.02, 0.02].

Conclusion:   CD4 counts are the standard measure of efficacy in HIV studies while utility scores are a common measure of effectiveness in economic evaluations.  The development of a model predicting utility scores from CD4 count will facilitate model based economic evaluations in HIV.  The current model demonstrates the feasibility of predicting utility scores from CD4 count at a group level.  However, the model performed poorly at the individual participants level.  In particular, the variability observed in the individual utility scores was not completely captured by CD4 count.