Advertisement

Peak HIV prevalence: a useful outcome variable for ecological studies

Open AccessPublished:January 30, 2013DOI:https://doi.org/10.1016/j.ijid.2012.12.020

      Summary

      A key question for ecological studies with HIV as the outcome variable is what measure of HIV prevalence to use. In this study we compared the strengths and weaknesses of a variety of measures of HIV prevalence, focusing on peak HIV prevalence and HIV prevalence measured at the same time as the exposure variable. We explored the theoretical problems with each of the two measures of HIV prevalence. We then investigated the difference that substituting one variable for the other made to two published ecological studies. One published study found a strong relationship between migration intensity and HIV prevalence measured at the time the migration was measured. When we repeated the analysis using peak HIV prevalence as the outcome variable, there was no evidence of an association. The second study found evidence of a strong relationship between concurrency and peak HIV prevalence. On repetition of the analysis (but utilizing HIV prevalence at the time the concurrency was measured as the outcome variable) there was no longer a significant association. The choice of HIV measure as outcome variable in ecological studies makes a large difference to the study results. The choice of peak HIV prevalence as outcome variable offers the advantage of avoiding the HIV introduction time bias.

      Keywords

      1. Introduction

      Population-level risk factors, such as increased network connectivity, may be key to the generation of generalized HIV epidemics (GHEs – defined as adult (15–49 years old) HIV prevalence ≥5%). Being population-level risk factors, they need to be investigated at the population or ecological level. A key question for ecological studies with HIV as the outcome variable is what measure of HIV to use; for instance, incidence, prevalence, peak prevalence, antenatal prevalence, or prevalence as measured closest to the time of the exposure variable are all possible measures of HIV that could be used.
      A number of ecological studies have used HIV incidence,

      Oster E. Routes of infection: exports and HIV incidence in Sub-Saharan Africa. National Bureau of Economic Research; 2007.

      or prevalence at one point in time,
      • Drain P.K.
      • Halperin D.T.
      • Hughes J.P.
      • Klausner J.D.
      • Bailey R.C.
      Male circumcision, religion, and infectious diseases: an ecologic analysis of 118 developing countries.
      • Drain P.K.
      • Smith J.S.
      • Hughes J.P.
      • Halperin D.T.
      • Holmes K.K.
      Correlates of national HIV seroprevalence: an ecologic analysis of 122 developing countries.
      • Over M.
      The effects of societal variables on urban rates of HIV infection in developing countries: an exploratory analysis. Confronting AIDS: Evidence from the Developing World.
      as the outcome variable. A refinement of this approach has been to use HIV prevalence at the date that the exposure variable was measured
      • Voeten H.A.
      • Vissers D.C.
      • Gregson S.
      • Zaba B.
      • White R.G.
      • de Vlas S.J.
      • et al.
      Strong association between in-migration and HIV prevalence in urban Sub-Saharan Africa.
      • Reniers G.
      • Tfaily R.
      Polygyny, partnership concurrency, and HIV transmission in Sub-Saharan Africa.
      – here termed date-of-exposure-variable (DEV) HIV prevalence. More recently, studies have used national peak HIV prevalence as the outcome variable.
      • Kenyon C.
      • Colebunders R.
      Strong association between point-concurrency and national peak HIV prevalence.

      McIntosh C. Has better health care contributed to higher HIV prevalence in Sub-Saharan Africa? Mimeographed document. University of California at San Diego; 2007. Available at: http://irps.ucsd.edu/assets/015/6810.pdf, Accessed 10/10/2012(accessed).

      In this article, we use two empirical examples to discuss the strengths and weaknesses of peak HIV and DEV HIV prevalence as the outcome variables in ecological studies.
      The use of DEV HIV prevalence follows in the tradition of ecological studies of other infectious diseases.
      • Koopman J.S.
      • Longini Jr., I.M.
      The ecological effects of individual exposures and nonlinear disease dynamics in populations.
      In diseases with a short incubation and duration of symptoms followed by recovery or death it is appropriate to measure the outcome variable at a short time period after the exposure.
      • Koopman J.S.
      • Longini Jr., I.M.
      The ecological effects of individual exposures and nonlinear disease dynamics in populations.
      In the case of HIV this can be misleading for two reasons, as illustrated in Figure 1. This figure represents the adult HIV prevalence for Uganda and South Africa and the prevalence of two hypothetical causes of GHEs: E1 and E2.
      Figure thumbnail gr1
      Figure 1The prevalence of HIV in Uganda and South Africa (RSA) (derived from UNAIDS

      UNAIDS report on the global AIDS epidemic. Geneva: UNAIDS; 2010.

      excluding Uganda's prevalence pre-1990, which is from Kirby
      • Kirby D.
      Changes in sexual behaviour leading to the decline in the prevalence of HIV in Uganda: confirmation from multiple sources of evidence.
      ) and two hypothetical causal factors (E1 and E2) from 1980 to 2009.
      The first problem relates to the fact that even without antiretroviral therapy (ART), infected persons remain as part of the prevalent population of infected persons for around 10 years.
      • Bongaarts J.
      • Buettner T.
      • Heilig G.
      • Pelletier F.
      Has the HIV epidemic peaked?.
      The point prevalence of HIV in a population is therefore a product of the interactions between its component causes over the preceding decade or longer. Measuring Uganda's HIV prevalence in 1991, a year after the hypothetical exposure variable (E1), as is commonly done in ecological studies, will result in ignoring the effect of that variable on HIV prevalence in the years 1980–1989 (when most HIV infections occurred but E1 was 0). This could result in a spurious association between E1 and HIV prevalence. The use of peak HIV prevalence is also prone to this misclassification bias. Dealing with this problem requires either calculating a cumulative measure of E over the relevant time period or establishing that E is relatively stable over this time (such as E2).
      A second problem and one that only affects DEV HIV prevalence, is the HIV introduction time (HIT) bias. Countries with GHEs had fairly similar rates of increase in HIV prevalence, but these epidemics began at widely differing times, depending largely on when the virus was introduced into the population.
      • Bongaarts J.
      • Buettner T.
      • Heilig G.
      • Pelletier F.
      Has the HIV epidemic peaked?.
      This is illustrated in Figure 1, where South Africa's epidemic only starts around 1991, the same year that Uganda's epidemic peaked. The dashed line (E2) in the figure hypothetically represents the prevalence of an established component cause of South Africa's HIV epidemic. If an ecological study of the relationship between E2 and HIV prevalence was made in 1989, it would likely miss the association, as HIV prevalence is close to zero at this point. The same study repeated a decade later, at the time of peak HIV prevalence, would be less likely to miss the association. In an ecological study seeking to uncover the determinants of GHEs it would seem misleading to represent a country with a high peak HIV prevalence such as South Africa as having an HIV prevalence of 0% merely because the exposure variable was measured in 1989. The various component causes that made South Africa's sexual network so conducive to HIV spread were likely just as prevalent in 1989 (as attested to by the high rates of other sexually transmitted infections) as in 2000.
      • Johnson L.F.
      • Coetzee D.J.
      • Dorrington R.E.
      Sentinel surveillance of sexually transmitted infections in South Africa: a review.
      Using the peak HIV prevalence as the outcome variable removes the bias introduced by the time of introduction of HIV into a network. An alternative strategy that has been used to deal with the HIT bias is to include, in multivariate analysis, a variable that measures the time between the first case of HIV infection or AIDS in a country and the year the HIV prevalence was measured.
      • Drain P.K.
      • Smith J.S.
      • Hughes J.P.
      • Halperin D.T.
      • Holmes K.K.
      Correlates of national HIV seroprevalence: an ecologic analysis of 122 developing countries.
      • Over M.
      The effects of societal variables on urban rates of HIV infection in developing countries: an exploratory analysis. Confronting AIDS: Evidence from the Developing World.
      The problem with this approach relates to the lack of evidence that this measure accurately measures the time of introduction of HIV.
      • Bello G.
      • Passaes C.P.
      • Guimaraes M.L.
      • Lorete R.S.
      • Matos Almeida S.E.
      • Medeiros R.M.
      • et al.
      Origin and evolutionary history of HIV-1 subtype C in Brazil.
      We will now illustrate these problems with two examples from the literature.

      2. Examples

      2.1 DEV HIV but not peak HIV prevalence, strongly associated with migration

      In this study, Voeten et al.,
      • Voeten H.A.
      • Vissers D.C.
      • Gregson S.
      • Zaba B.
      • White R.G.
      • de Vlas S.J.
      • et al.
      Strong association between in-migration and HIV prevalence in urban Sub-Saharan Africa.
      found evidence of a strong association between female in-migration (percentage of 15–49-year-old women who migrated into major cities in 28 African countries in the preceding 12 months) and urban antenatal HIV prevalence in the same year (or if this was not available, then adjacent years) (R2 = 0.57; p < 0.001). We repeated the same linear regression analysis using the same datasets but using peak HIV prevalence as the outcome variable and using female in-migration prevalence closest to the year of peak HIV as the exposure variable. In the repeat analysis there was no association between the two variables (R2 = 0.085; p = 0.132).

      2.2 Peak HIV but not DEV HIV prevalence, strongly associated with concurrency

      In this study, we found a strong association between point prevalence of male concurrency and peak HIV prevalence at a national level (R2 = 0.78; p < 0.001)
      • Kenyon C.
      • Colebunders R.
      Strong association between point-concurrency and national peak HIV prevalence.
      in 11 countries where concurrency data were collected in 1989. When we repeated the analysis, with HIV prevalence in 1990 as the outcome variable, the association between concurrency and HIV prevalence disappeared (R2 = 0.04; p = 0.534).

      3. Discussion

      How should one go about deciding which of these results to believe? We offer two arguments that suggest that in both cases the results using peak HIV as outcome variable offer more plausible results. The first argument is methodological – unlike DEV, peak HIV prevalence is not affected by the HIT bias.
      In both of the studies above, the most noticeable determinant of the dramatic changes in the relationship between HIV and the exposure variable was the HIT bias. An example of this in the concurrency study is that Lesotho in 1990 had an HIV prevalence of 0.8% and this was used in the DEV analysis. Only in 2000 did it reach its peak HIV prevalence of 24.5%. Peak HIV was higher than DEV HIV by a factor of three or more in the case of three other countries – all of which had high concurrency rates. As with the example of South Africa, it seems misleading to classify these four countries as having an HIV prevalence so much lower than their peak, simply because this was their HIV prevalence when the surveys measuring concurrency rates were performed. Peak HIV prevalence is by its nature a composite measure of all the factors promoting and preventing the spread of HIV in a population that is unaffected by the HIT bias. As such it represents a useful way to compare risk factors underpinning the genesis of high HIV prevalence rates.
      Similarly in the migration study, Namibia was represented by an HIV prevalence of 4.2%, which is what it was in 1992 when the migration measurements were taken, rather than its peak of 16.5%. There were also significant discrepancies between the HIV prevalence rates derived from antenatal statistics used in the paper and the UNAIDS estimates of adult (15–49 years old) prevalence data we used for the peak HIV estimates.

      UNAIDS report on the global AIDS epidemic. Geneva: UNAIDS; 2010.

      The second argument is empirical. We have been unable to find an ecological association between migration and HIV prevalence in analyses of 13 different measures of internal and international migration (authors’ unpublished results). As far as the ecological relationship between concurrency and HIV prevalence is concerned, there has been considerable debate in the literature,
      • Sawers L.
      • Stillwaggon E.
      Concurrent sexual partnerships do not explain the HIV epidemics in Africa: a systematic review of the evidence.
      • Lurie M.N.
      • Rosenthal S.
      Concurrent partnerships as a driver of the HIV epidemic in Sub-Saharan Africa? The evidence is limited.
      but our study was the first to evaluate this in a comparable cross-country dataset. Further studies are however needed to better understand these dynamics.
      A potential newly arising problem in using peak HIV prevalence is the effect of ART in increasing prevalence: by decreasing mortality, ART will increase HIV prevalence. However, HIV prevalence was declining in all countries that have experienced GHEs before ART was in widespread use.

      UNAIDS report on the global AIDS epidemic. Geneva: UNAIDS; 2010.

      In some countries HIV prevalence is once again increasing due to ART/increasing HIV incidence.

      UNAIDS report on the global AIDS epidemic. Geneva: UNAIDS; 2010.

      This may make it more appropriate to use initial peak HIV prevalence as the outcome variable.
      While the HIT bias provides a strong rationale to use peak HIV prevalence as outcome variable, much remains to be worked out on how best to conduct ecological studies of the population-level determinants of GHEs. In particular, thought needs to go into how best to represent the strength of the exposure variable through the period of increasing HIV prevalence. So too, it would be useful to know the optimal way to include existing knowledge of transmission matrices into models of HIV transmission. The combination of the failure of individual-level studies to illuminate the determinants of GHEs
      • Aral S.O.
      • Leichliter J.S.
      • Blanchard J.F.
      Overview: the role of emergent properties of complex systems in the epidemiology and prevention of sexually transmitted infections including HIV infection.
      and the increasing evidence that population-level factors may be responsible, make this an important task.
      • Blanchard J.F.
      • Aral S.O.
      Emergent properties and structural patterns in sexually transmitted infection and HIV research.
      • Kirby D.
      Changes in sexual behaviour leading to the decline in the prevalence of HIV in Uganda: confirmation from multiple sources of evidence.
      Conflict of interest: The authors declare that they have no conflict of interest.

      References

      1. Oster E. Routes of infection: exports and HIV incidence in Sub-Saharan Africa. National Bureau of Economic Research; 2007.

        • Drain P.K.
        • Halperin D.T.
        • Hughes J.P.
        • Klausner J.D.
        • Bailey R.C.
        Male circumcision, religion, and infectious diseases: an ecologic analysis of 118 developing countries.
        BMC Infect Dis. 2006; 6: 172
        • Drain P.K.
        • Smith J.S.
        • Hughes J.P.
        • Halperin D.T.
        • Holmes K.K.
        Correlates of national HIV seroprevalence: an ecologic analysis of 122 developing countries.
        J Acquir Immune Defic Syndr. 2004; 35: 407-420
        • Over M.
        The effects of societal variables on urban rates of HIV infection in developing countries: an exploratory analysis. Confronting AIDS: Evidence from the Developing World.
        European Commission and World Bank, Brussels and Washington DC1998
        • Voeten H.A.
        • Vissers D.C.
        • Gregson S.
        • Zaba B.
        • White R.G.
        • de Vlas S.J.
        • et al.
        Strong association between in-migration and HIV prevalence in urban Sub-Saharan Africa.
        Sex Transm Dis. 2010; 37: 240-243
        • Reniers G.
        • Tfaily R.
        Polygyny, partnership concurrency, and HIV transmission in Sub-Saharan Africa.
        Demography. 2012; 49: 1075-1101
        • Kenyon C.
        • Colebunders R.
        Strong association between point-concurrency and national peak HIV prevalence.
        Int J Infect Dis. 2012; 16: e826-e827
      2. McIntosh C. Has better health care contributed to higher HIV prevalence in Sub-Saharan Africa? Mimeographed document. University of California at San Diego; 2007. Available at: http://irps.ucsd.edu/assets/015/6810.pdf, Accessed 10/10/2012(accessed).

        • Koopman J.S.
        • Longini Jr., I.M.
        The ecological effects of individual exposures and nonlinear disease dynamics in populations.
        Am J Public Health. 1994; 84: 836-842
        • Bongaarts J.
        • Buettner T.
        • Heilig G.
        • Pelletier F.
        Has the HIV epidemic peaked?.
        Population and Development Review. 2008; 34: 199-224
        • Johnson L.F.
        • Coetzee D.J.
        • Dorrington R.E.
        Sentinel surveillance of sexually transmitted infections in South Africa: a review.
        Sex Transm Infect. 2005; 81: 287-293
        • Bello G.
        • Passaes C.P.
        • Guimaraes M.L.
        • Lorete R.S.
        • Matos Almeida S.E.
        • Medeiros R.M.
        • et al.
        Origin and evolutionary history of HIV-1 subtype C in Brazil.
        AIDS. 2008; 22: 1993-2000
      3. UNAIDS report on the global AIDS epidemic. Geneva: UNAIDS; 2010.

        • Sawers L.
        • Stillwaggon E.
        Concurrent sexual partnerships do not explain the HIV epidemics in Africa: a systematic review of the evidence.
        J Int AIDS Soc. 2010; 13: 34
        • Lurie M.N.
        • Rosenthal S.
        Concurrent partnerships as a driver of the HIV epidemic in Sub-Saharan Africa? The evidence is limited.
        AIDS Behav. 2010; 14 (discussion 5–8): 17-24
        • Aral S.O.
        • Leichliter J.S.
        • Blanchard J.F.
        Overview: the role of emergent properties of complex systems in the epidemiology and prevention of sexually transmitted infections including HIV infection.
        Sex Transm Infect. 2010; 86: iii1-iii3
        • Blanchard J.F.
        • Aral S.O.
        Emergent properties and structural patterns in sexually transmitted infection and HIV research.
        Sex Transm Infect. 2010; 86: iii4-iii9
        • Kirby D.
        Changes in sexual behaviour leading to the decline in the prevalence of HIV in Uganda: confirmation from multiple sources of evidence.
        Sex Transm Dis. 2008; 84: ii35-ii41