A MICROSIMULTION MODEL FOR ESTIMATING THE LIFETIME DIRECT MEDICAL COSTS OF HIV IN BRITISH COLUMBIA

Purpose

The objective of this study was to estimate the lifetime direct medical cost associated with treating individuals infected with HIV and receiving highly active antiretroviral therapy (HAART) in British Columbia (BC), Canada.

Methods

A microsimulation was developed using several statistical models which described the clinical course of HIV and the treatment costs associated with various categories of health services utilization, including inpatient care, outpatient visits, laboratory tests, emergency room visits, and medications (HAART and non-HAART). The clinical course of HIV was characterized by CD4 cell count and plasma viral load.

All statistical models were fit using individual-level longitudinal data, population-based for BC. Costs were estimated using random effects models which adjusted for CD4 cell count, sex, and injection drug use. Trajectories of CD4 cell count and plasma viral load were estimated using additive models which allowed for non-linear trajectories in time. Additional variables included in modelling the clinical course of HIV were adherence to antiretroviral therapy and cumulative number of medication resistance mutations over time. Resistance mutations were estimated via a Weibull survival model with gamma-distributed frailty terms. Mortality was estimated using a Cox proportional hazards model, with CD4 cell count included as a time-dependent covariate.

Probabilistic sensitivity analysis was performed to gauge the impact of second-order uncertainty. In addition, one-way sensitivity analyses were performed for the discount rate and the choice of CD4 cell count threshold at which to initiate treatment with HAART. The maximum time horizon considered was 30 years post-infection. All costs are reported in 2005 $Canadian.

Results

In the base case analysis, the estimated lifetime cost was $274,090 (standard error (SE)=4,818). This was associated with an estimated 22.7 (SE=0.32) life years and 13.8 (SE=0.16) quality-adjusted life years. Across 1-way sensitivity analyses, the estimated lifetime cost ranged from $217,189-$407,950. Factors associated with higher lifetime cost were female sex, injection drug use, and lower CD4 cell count at HAART initiation. Regular use of laboratory tests were associated with lower lifetime costs.

Conclusion

The microsimulation described here includes several novel strategies in describing the clinical course and direct medical costs of HIV, including the use of non-linear models to describe CD4 cell count and plasma viral load trajectories, and the generation of correlated random effects to account for the relationship between various categories of health resource utilization.