THE COST-EFFECTIVENESS OF SYMPTOM-BASED TESTING AND ROUTINE
SCREENING FOR ACUTE HIV INFECTION IN MEN WHO HAVE SEX WITH MEN
IN THE UNITED STATES
J.L. JUUSOLA, M.L. BRANDEAU, E.F. LONG, D.K. OWENS, E. BENDAVID
We developed a dynamic compartmental model to assess the effectiveness and cost-
effectiveness of various strategies of testing for and treating acute HIV infection in men who
have sex with men (MSM) in the United States. Figure A1 shows a schematic representation of
the model. Table A1 summarizes model notation. The model calculates Xi(t), the number of
individuals in each compartment i = 1, …, 13 at time t. Individuals transition between
compartments and age into or transition out of the system at rates defined by population
demographics, disease progression parameters, and screening and treatment interventions. We
incorporated epidemiologic, clinical, and economic data in order to estimate HIV prevalence,
incidence, quality-adjusted life years (QALYs), and healthcare costs for various testing and
treatment strategies. We implemented the model in Microsoft Excel 2007 using a weekly time
step and considered a time horizon of 20 years.
We estimated that there are approximately 6.4 million MSM aged 13-64 in the US [1-6].
We subdivided this population into 13 compartments, based on the following factors:
• HIV infection status (uninfected, infected)
• HIV disease stage if infected (acute infection, asymptomatic HIV: CD4 >350 cells/mm3,
symptomatic HIV: CD4 200-350 cells/mm3, AIDS: CD4 <200 cells/mm3)
• Screening status (unidentified, identified)
• Treatment status (receiving antiretroviral therapy (ART), not receiving ART)
We assumed that individuals enter the model at age 13 into compartment 1 (uninfected,
unidentified) at a rate ?, calculated based on US population data for males by age group [5, 6].
Individuals exit the population due to death or maturation. Individuals mature out of the
population upon turning 65 at rate µi. All individuals die from non-AIDS related causes at rate ?i,
and individuals with AIDS (those in compartments 9, 10, and 13) die from AIDS at rate ?i. We
allowed µi and ?i to vary by compartment, but assumed that they were the same for all MSM,
based on US population data for males [5-7]. In the absence of new interventions, the size of the
population of MSM at the end of 20 years matches expected US population growth trends .
Testing for acute HIV infection can be implemented with either p24 antigen tests or with
HIV-1 viral load (VL) testing. Tests for p24 antigen are more specific and less expensive than
VL tests, but they are less sensitive [8, 9]. In our analysis, we considered VL testing.
Studies on adding VL testing to routine antibody testing have shown that this additional
testing increases the diagnostic yield for HIV infection by 4-10% [10-15]. Most of these
screening programs have used pooling schemes, where serum or plasma samples that are
negative for HIV antibodies are pooled and then tested using VL testing. Pooling is used to
reduce cost and the potential for false positives. The additional cost of pooled VL testing in these
programs, as compared to standard antibody screening alone, ranged from $3-17 per processed
specimen, in populations with HIV prevalence ranging from 0.6-4.1% [10-12, 15]. Pooled VL
testing in populations with higher HIV prevalence, such as the MSM population we modeled,
would require more re-testing of positive samples, leading to a higher cost per specimen. Also,
pooling increases the turnaround time for notifying the patient of test results, especially in
facilities with low volume of testing. During the acute infection phase, time is of the essence due
to high transmission risk. Therefore, although pooling algorithms can be used with VL testing to
reduce costs, we assumed in our analysis that all specimens are tested individually. In sensitivity
analysis we considered pooled testing for routine VL testing, but not for symptom-based testing.
For symptom-based testing, we assumed that uninfected MSM present with influenza-
like symptoms at the baseline rate of influenza in the population and are tested for HIV. This
results in extra costs incurred for symptom-based testing strategies.
We assumed that HIV transmission occurs via homosexual contact. The total rate of
contacts sufficient to transmit infection is the sum of the sufficient contact rates between an
uninfected individual in compartment i (i = 1, 2) and an infected individual in compartment j (j =
3, …, 13), which we denote by ?i,j(t).
We modeled the homosexual contact rate as a binomial process, similar to Long et al.
[16, 17]. A “success” is defined as transmission of HIV, and uninfected individuals select n(1-
ui?) risky partners per year (i.e., partnerships involving unprotected sexual contact) with the
probability of transmission per partnership being the probability of “success”. The rate ?i,j(t) is
affected by condom usage, where the percent of partnerships involving condom usage is ui, and
by condom effectiveness, ?. The annual probability of transmission in a partnership, ?i,j,
assuming no condom usage or ineffective condom usage, depends on the disease stage of the
infected individual and whether he is receiving ART. Uninfected individuals choose risky
partners from infected compartment j based on the number of risky partnerships individuals in
compartment j have (homogeneous mixing). Hence the probability of selecting a risky partner
from compartment j is:
Combining all these factors, ?i,j(t) can be calculated as follows:
Uninfected individuals in compartments i = 1, 2 thus acquire HIV infection at time t with
We modeled the effects of HIV testing and counseling in reducing risky behavior as reductions
in the number of sexual partnerships [16-20].
After acquiring HIV infection, individuals progress through the disease stages at rate ?i.
These progression rates are inversely proportional to the average length of each stage and are
based on previous models of the natural history of HIV infection [16-18]. ART lowers the
progression rate ?12 and the AIDS death rate ?13 by increasing the time spent in those states.
System of Equations
We modeled HIV transmission and progression with a system of 13 nonlinear differential
equations. The equations describing the change in the number of individuals in each
compartment at each time step are listed below. We let Xi denote Xi(t) for ease of notation.
Change in number of unidentified uninfected individuals (compartment 1):
Change in number of identified uninfected individuals (compartment 2):
Change in number of unidentified acutely infected individuals (compartment 3):
Change in number of identified, untreated acutely infected individuals (compartment 4):
Change in number of unidentified asymptomatic infected individuals (compartment 5):
Change in number of identified asymptomatic infected individuals (compartment 6):
Change in number of unidentified symptomatic infected individuals (compartment 7):
Change in number of identified, untreated symptomatic infected individuals (compartment 8):
Change in number of unidentified individuals with AIDS (compartment 9):
Change in number of identified, untreated individuals with AIDS (compartment 10):
Change in number of identified, treated acutely infected individuals (compartment 11):
Change in number of identified, treated symptomatic infected individuals (compartment 12):