The relevance of attributable risk measures to HIV prevention planning

In this issue, Tyndall et al. describe their investigation into the role of cocaine injection in the HIV epidemic among Vancouver injecting drug users (IDU) [¹]. Several features of the analysis support a strong causal relationship between cocaine injection and HIV, including the magnitude of the association, the possible dose-response relationship, and the temporal sequence (with cocaine injection at both study enrollment and during follow-up being related to HIV infection). Moreover, there is a plausible explanation for the association - namely, that sessions of cocaine injection may be characterized by multiple opportunities for HIV exposure. A consistency between this and earlier epidemiologic studies showing that cocaine injection is a risk factor for HIV infection adds to the evidence supporting a causal relationship [^2,3]. Further, the findings of the Vancouver study do not appear to be biased by serious losses to follow-up, confounding, or measurement error. In all, it is a well-conducted study and a well-reasoned interpretation of the results. The only remaining items of significance missing from the paper are the estimation of the relative importance of cocaine injection versus other factors to HIV control in Vancouver using measures of attributable risk, and further examination of cut-off points used to classify subjects according to cocaine injection.

At this stage of our understanding of HIV occurrence, estimates of attributable risk are particularly relevant because they estimate the reduction in disease that could ensue (in exposed individuals or in the population at risk) if the risk factor in question were eliminated [⁴]. For this reason, attributable risk estimates are meaningful only when the association is believed to be causal. The proportion of infections in the underlying population of IDU that may be attributed to the risk factor [the population-attributable risk percent (PAR%)] depends upon both the magnitude of the relative risk and the prevalence of the risk factor in the population. Thus, elimination of a common risk factor with a large relative risk would lead to a very substantial reduction in disease occurrence, whereas control or elimination of a rare risk factor with a large relative risk would have a much smaller impact on the health of a community (unless, of course, it is a necessary cause). To illustrate this point, data from this and one other published study of risk factors for HIV infection in a large cohort of IDU [⁵] were used to estimate measures of attributable risk using standard methods [⁶]. The large sample size in both studies would increase the likelihood that exposure rates in the sample represent exposure in the underlying population; this is necessary for calculation of the PAR% [⁷]. In both cases, the relative risk estimates were taken from multivariate analyses of risk behavior during follow-up, and the proportion of subjects with the risk factor in question was estimated using baseline data. Thus, the calculations assume that exposure levels remained approximately the same throughout the studies.

As shown, in the Vancouver study, 73% of HIV infections among weekly cocaine injectors may be attributable to cocaine injection [the attributable risk percent (AR%)]. As a large proportion of study subjects injected cocaine more than once a week (42%), 31% of HIV infections in the underlying population of IDU (73% of 42%) may be attributed to that risk factor. The AR% for incarceration was 64%, and the PAR% appears substantially smaller (21%) because fewer IDU had been incarcerated (33%). Because unstable housing was relatively common in the Vancouver sample (59%), the PAR% was 34%, even though the relative risk was smaller than for other factors. In the San Francisco study, trading sex for money was associated with a large RR of HIV infection among women, and thus the estimate of the AR% was 80%, but since only 24% of women controls had traded sex for money, the PAR% for this factor is estimatd to be 19%. The PAR% for MSM (with a relative risk of 8.8) may be very small (8%) because few subjects were MSM (Table 1).

Table 1

This shows that interpretation of risk factor analyses in terms of prioritizing prevention activities may benefit from calculation of measures of attributable risk. Ranking of HIV risk factors using the relative risk will identify particularly high-risk activities or characteristics, but at this stage of our understanding, many of these have already been quantified. In the case of Vancouver, cocaine injection is associated with both a high relative risk of HIV seroconversion and a large PAR%. Thus, taken together with strong causal evidence, reduction of cocaine injection would appear essential to HIV control in that city.

After 15 years of risk factor research, we have a fairly complete understanding of how HIV is transmitted. The current challenge is to make risk factor studies more directly relevant to public health prevention activities. One way to do this is to examine attributable risk measures, to understand the potential impact of interventions that address specific risk factors. Another way is to classify risk factors for analysis according to categories that correspond to these potential activities. The cocaine injection categories in the Vancouver study are an example of risk factor categorization that cannot easily be linked to a specific intervention. The cutpoint for 'intensive' injection of cocaine used in the multivariate analysis (once a week or more often) combined a broad spectrum of users with a wide assortment of prevention needs. The univariate analysis of HIV risk in relation to 'dose' of cocaine injection also utilized cut-off points that do not necessarily correspond to natural groupings of cocaine users with similar barriers to safe injection. Specifically, the mid-level risk category of those who inject cocaine 'twice per week to three times per day' included those with substantial cocaine dependency for whom some form of treatment to bring their use under better control may be a necessary prerequisite to safe injection. At the lower bound of the category are individuals who are injecting cocaine only twice per week and are probably not addicted or dependent, and who might benefit immediately from better access to sterile syringes. Estimation of relative and attributable risk for subgroups for whom specific strategies may be appropriate should serve as a better guide to decision making and will aid in evaluation of their success in HIV control. Nowadays, measures of attributable risk are rarely calculated, perhaps because they are simple and seem unsophisticated alongside more elaborate statistical methods for calculating relative effects. However, their potential contribution to prevention planning and to our understanding of the underlying dynamics of occurrence of HIV in a community may be more relevant now than ever before.