In our study population HIV-1 incidence (1.35 per 100 PY) was four times as high as recent estimates from Tanzanian household-based prevalence data for the whole country (0.34/100 PY) 4
. Cohort studies from northern Tanzania report an incidence between 0.39 and 1.20 per 100 PY depending on year of data collection and study site 5,6
. The difference between our and the national estimates can partially be explained by the differences in HIV-1 prevalence; the HIV-1 prevalence in our study (14.8% for males and 21.2% for females in Mbeya; 11.5% for males and 13.4% for females in Itende) 7
was almost twice the national average 8
. Even though estimating incidence from cross-sectional prevalence data is well established 9
, incidence measured in cohort studies remains the gold standard for accuracy, especially for intervention trials. In the Step study 10
which assessed the efficacy of a cell-mediated immunity vaccine no increased HIV incidence was observed during the trial, even though cross-sectional data prior to the study indicated that women were at high risk.
The main burden of new infections in our study is concentrated in women, both in urban and semi-urban settings. This is consistent with the above mentioned prevalence survey from Tanzania and other African countries 4
. An elevated risk due to casual partnerships with older men 11,12
during early sexual experience which could have explained this pattern was not observed in our study (data not shown). A possible explanation for the higher incidence in women could be the increased HIV-1 susceptibility of women which has been reported in many previous studies 13
The age distribution differs considerably between our urban and rural populations: while in Mbeya town most of the infections occurred prior to the age of 25 years, Itende showed a peak incidence rate in the group of 25-29 year old females and 30-34 year old males. The age distribution of HIV-1 infected participants from Itende is similar to that of a rural cohort study close to Mwanza 6
. Differences in behaviour and socioeconomic status between urban and rural settings itself could explain this delayed acquisition of HIV-1 in Itende, as well as the steady spread of HIV from urban settlements into the rural population 14
. Important for the planning of recruitment strategies in future phase III vaccine trials is that a similar HIV-1 incidence was found in the two groups from Mbeya who were recruited by advertisement and by door-to-door campaign. Our data thus do not provide evidence that advertisement leads to an increase in response bias.
We did not observe a significant decline in seroconversions over time attributed to a closed cohort effect like other studies 15,16
, even though we performed regular health education sessions, offered counselling and promoted free condom use.
Most studies on risk factors for HIV infection are based on prevalent rather than incident cases. Hence, their results also take account of the time after HIV-1 acquisition and are biased by determinants of disease duration. Our approach of analyzing incident cases for association excludes this bias.
We identified risk factors that are potentially preventable by public health interventions. We observed a strong association between alcohol consumption and newly acquired HIV infection which has already been described in other studies from Africa 3,17
, and we were able to demonstrate that the incidence rate was positively associated with frequency of alcohol consumption. This is usually explained by the effect of alcohol to reduce inhibition and to diminish perception of exposure to risky sexual behaviour, violence, forced sex and rape 18,19
Tanzanians consume alcohol frequently in restaurants, bars and local brew establishments where they also encounter new sex partners 20
. In a previous cohort study of female bar workers in Mbeya region, 67% of the participants were HIV-1 sero-positive at enrolment 21
, demonstrating the high risk potential for HIV acquisition in their clients.
Another notable result is the association between the risk of HIV-1 acquisition and the number of lifetime sexual partners which – in the absence of data regarding concurrent sexual partnerships and frequency of sexual intercourse - we use as a proxy for current sexual behaviour 22
. A similar association was found in a cross-sectional study of prevalent HIV infection in Northern Tanzania 23
. Indeed, there is no evidence that Africans typically have more sexual partners than elsewhere in the world 24
, but still sexual behaviour might be an important risk factor since having more than one concurrent partnership seems to increase the risk of HIV acquisition 25
and is in addition more frequently found in Africa 24
While in the early phase of the epidemic HIV risk was linked to higher educational attainment 26,27
some studies in Africa observed a similar shift towards reduced risk among higher educated people as found in our study 28-30
. Staying in school for a longer time plays an important role for being more frequently and intensively exposed to health and basic education. Subsequently, better educated people seem to understand and adapt health education messages faster, leading to behavioural changes like delay in sexual debut, reduction of sexual partners or increased condom use 28,31
. Furthermore educational status is likely to reflect socio-economic-status which is also negatively related to HIV incidence 32
Lastly, we found that study participants without religious denomination were at higher risk of HIV-1 acquisition. This finding could be explained by the role of religious communities as influential social networks that can influence health promotion messages 33
. Previous studies addressing correlations between religious beliefs, behaviour and risk for HIV-1 infections often focused on Muslim populations, with a particular interest in male circumcision which is a known protective intervention 34,35
. Our results indicate such a correlation as well (23% risk reduction in the adjusted univariable analysis, HR=0.77, 95% CI 0.34-1.77), even though male circumcision did not reach statistical significance in the final multivariable model.
Our study and this analysis have certain limitations
The main objective of the CODE study was to explore the best recruitment strategy for future HIV vaccine trials, thus our different recruitment strategies might have caused some sampling bias. However, the results of the two urban strata did not differ significantly, thus we consider them as representative for the urban population of this age group in Mbeya. However, choosing a semi-urban setting in walking distance of the urban centre of Mbeya might be another source of selection bias because the population in this setting might closely resemble the urban population. Secondly, losses to follow-up might have influenced our results, although we attained a very high retention rate and identified characteristics for discontinue the study. The exact bias is difficult to assess because we were unable to collect information about reasons for loss to follow-up. It should however be noted that our losses to follow-up (31% over the study period including participants who died) are relatively low for a study in sub-Saharan Africa with a duration of four years. Third, our risk analysis only relies on data collected during enrolment because this was the most complete dataset. It is therefore possible that some of our co-variates changed over time which is not reflected in this analysis. Fourth, the validity of our data obtained during face-to-face interviews is subject to underreporting bias, especially for socially sensitive behaviours. Our study participants might have felt pressure to please our study staff in addition to the normal pressure to underreport sensitive behaviours 36
. A method to overcome this bias – namely audio-computer-assisted self-interviewing – was not considered to be appropriate in this setting. Lastly, as in other closed cohorts we might have encountered a cohort attrition effect, leading to a reduced incidence rate towards the end of the study. However, the relatively stable HIV-1 incidence over time that we found does not indicate that this applies.
In conclusion, this study enhances knowledge for understanding of the ongoing HIV epidemic in East Africa and for implementing further public health actions in this setting. We observed a relatively high incidence of HIV, indicating an ongoing dynamic - especially among young women who are at higher risk - despite huge preventive efforts in the past decades. Alcohol consumption, low school education level and number of sex partner represent key risk factors; not belonging to a religious group seems to be a risk factor as well. To reduce HIV infection rates behavioural interventions towards the reduction of alcohol intake and safer sex practices should be intensified in these communities. The association with a low school education level that we found indicates that health education programs should be strengthened and go beyond formal schooling. Existing structures of social networks (e.g. religious groups) should be used and new networks should be identified to increase the catchment population for preventive measures. Finally, prevention programmes should empower young women since they are the group with the highest risk of HIV-1 infection in this region.