All statistical analyses were conducted in Stata 12 (StataCorp, Texas, USA). All statistical tests used were two-tailed and interpreted at the 5% significance level.
For the analyses which looked at the uptake of HTC services (either CO-HTC use during sero-surveys, or clients testing at the WI-HTC or via ANC-HTC), explanatory variables were prepared using questions from the sero-surveys and linked information from the DSS. HIV status was based on sero-survey research test results. Socio-demographic, behavioural and clinical risk factors for service use were investigated separately by sex using a priori hypotheses, drawing on factors
80 identified in the literature or associated with access to other HIV services in Kisesa (42, 43). Data were first analysed using cross tabulations and chi square tests to explore associations between exposure variables and HTC use. Logistic regression models were used to compare the characteristics of those who tested and didn’t test in crude and adjusted analyses. Multivariable models were fitted using a forward- stepwise approach and including all variables significant in univariable analyses at the p≤0.10 level. Likelihood ratio tests were used to assess the inclusion of variables in multivariable models (variable retained if it significantly improved model fit at the p≤0.10 level). Interactions were explored between HIV status and other characteristics found to be strongly associated with HTC use in this setting (area of residence, level of education and previous HTC use), and retained where statistically significant.
The denominator for analyses which explored CO-HTC use during sero-survey rounds included all sero-survey attendees. For the analysis of risk factors for WI- HTC and ANC-HTC uptake using the linked clinic-cohort dataset, some cohort participants were initially matched to a WI-HTC or ANC-HTC clinic record but then dropped during the validation procedures which created the final linked dataset. The analyses assessing risk factors for WI-HTC and ANC-HTC use among sero-survey attendees therefore used case-control methods, because the dataset did not contain the full denominator of all sero-survey attendees. For WI-HTC use, cases were defined as Sero6 attendees who were linked to a WI-HTC client with a clinic visit occurring within two years after participation in Sero6 (Sero6 was in 2010 and WI- HTC clinic data were available up to 2012. Given the potential for reverse causality, as sexual behaviours are hypothesized to change as a result of attending HTC, explanatory variables from the sero-survey and DSS were taken as measured prior to the time of HTC service use. Only clinic visits occurring up to three years after participation in a sero-survey were considered, in order to ensure that data had been recently collected). Controls were defined as Sero6 participants who were not linked to any WI-HTC clinic record.
For the analysis of risk factors associated with ANC-HTC use, women attending either Sero5 or Sero6 were included in order to increase sample size. Cases were defined as either i) women who participated in Sero5, reported a pregnancy between 2007-2010 and were linked to an ANC-HTC client with a testing visit within
81 three years of Sero5, or ii) women who participated in Sero6, reported a pregnancy between 2010-2012 and were linked to an ANC-HTC client with a testing visit within two years of Sero6. Controls were women who participated in either Sero5 or Sero6, reported a pregnancy between 2007-2010 (Sero5 attendees) or 2010-2012 (Sero6 attendees) and were not linked to any ANC-HTC clinic record.
Trends in coverage, i.e. the proportion of persons repeat testing or testing at different service types, were assessed using data on known HTC use (i.e. linked CO-HTC or clinic based HTC use) or on reported HTC use as indicated in responses to sero-survey questionnaires. The analyses which investigated risk factors for repeat testing relied on known CO-HTC use during sero-surveys, due to variability in completeness of reporting of previous HTC use during sero-surveys, and in identification of repeat testers at the clinic HTC services.
For the analysis which explored the impact of CO-HTC use on sexual behaviour change, nine indices of behaviour change were created by comparing sexual behaviours reported at one sero-survey round to those reported at the next. In order to explore any potential differences over time, analyses were carried out separately for those a) attending both Sero4 and Sero5 and using the CO-HTC service at Sero4, b) attending both Sero5 and Sero6 and using the CO-HTC service at Sero5. Multinomial logistic regression was used to assess associations between HTC use and three outcomes: an increase, decrease or no change in sexual risk behaviour, stratified by HIV-status. Crude associations were adjusted for potentially confounding socio-demographic variables (age, sex, marital status, area of residence, level of education and reported previous HTC use) in multivariable analyses. Poisson regression models were used to calculate crude and adjusted HIV incidence rates comparing those who used and didn’t use CO-HTC.