Study design and population
In Ukraine, individuals who screen positive for HIV at any community- or facility-based testing sites are referred to a network of government HIV clinics for confirmatory testing, diagnosis and registration. To assess the HIV risk factors and biological markers of transmission, we conducted a cross-sectional survey among adult patients who were officially registered at all clinical facilities in three most recent years before the study, from 2013 to 2015.
Sampling
We used a two-step random sampling approach to recruit participants into the study. At the first step, we selected seven out of 27 administrative regions of Ukraine using a random number generator. To estimate the prevalence of HIV risk factors with a 95% confidence level, 5% margin of error and assuming a survey design effect of 2.0, the target sample size was 775 per each year. To ensure representation of regions with different sizes of the epidemic (which could potentially be associated with case registration practices), we sorted the list of regions by the number of HIV cases reported in 2013 (13) and chose each fourth unit starting from a randomly generated number. The target sample size was divided across the seven regions proportionally to the number of cases reported in 2013 (Supplement Table S1).
Since there was no significant change in the number of registered patients in the selected regions over time, we chose to sample patients from three equal 3-month periods (October–December) to investigate the change in mode of transmission over time.
At the second step, we used a random sampling approach to recruit survey participants. We extracted data from patient registration forms to create a de-identified registry of patients diagnosed during the specified three periods (Supplement Table S2). The data included personal within-clinic ID code, sex, year of birth, date of registration, date of discharge (if applicable) and reason of discharge, dates of HIV tests used for diagnosis, registered mode of transmission, and stage of HIV disease at the time of diagnosis. In a random order, patients from the registry were contacted and invited to participate in the survey. All patients who attended the study appointment provided written informed consent.
Data collection
After providing informed consent, patients completed an anonymous survey to determine the patients’ HIV risk factors 10 years before finding out about their HIV-positive status. The survey instrument was developed by the investigators specifically for this study, and was pilot tested on 10 patients to adjust wording that could be misinterpreted. The questionnaire (see Supplementary File 1) included the following sections: sociodemographic information, sexual risk factors and protective behaviors, injecting drug use history, and history of sexually transmitted or bloodborne infections. The survey was administered by trained interviewers using the REDCap electronic data capture tools hosted at Ukrainian Institute on Public Health Policy (14). To limit self-reporting bias, interviewers were trained in psychological techniques to enhance rapport with respondents. Although the interviews took place in the HIV clinics, the interviewers were not clinic staff, and the information collected was not disclosed to any clinic staff. No one from the local research teams except the interviewers had access to the REDCap database. The survey data did not include any personal identifiers and were linked to the patient recruitment logs and registries using a participant ID. Before the interview, the interviewers explained that they did not have access to personal information in the clinical records and that the clinical staff did not have access to the interview data. The interview on average took 15 minutes to complete.
After the interview, participants provided a blood sample for hepatitis B virus (HBV) and HCV testing. Samples were centrifuged, and plasma samples were frozen. After recruitment, the frozen samples were shipped to a national viral hepatitis reference lab at the Kyiv City AIDS center. The samples were tested for biomarkers of HBV and HCV infection using the ARCHITECT-i1000SR Immunoassay Analyzer and the following reagent kits: ARCHITECT HBsAg, ARCHITECT Anti-HBs, ARCHITECT Anti-HBc, ARCHITECT Anti-HBc IgM and ARCHITECT anti-HCV. Results were reported both to the study investigators and to regional HIV clinics to inform clinicians and the patients about the results of the tests.
Data were collected between March and October 2016.
Data analysis
The descriptive analysis focused on the distribution of the registered mode of transmission in the official reports and verified registry and on the distribution of HIV risk factors among the survey participants. The data were disaggregated by sex, year of registration, and region. HIV surveillance data represents a census of HIV diagnoses (15) for the selected regions; therefore, no confidence intervals are presented.
In clinical records and official reports, registered modes of transmission were heterosexual exposure, homosexual exposure, injecting drug use, blood product transfusion, transplantation of organs or tissues, other medical exposure, occupational exposure, other non-medical exposure, confirmed mother-to-child, unconfirmed mother-to-child, and unknown. Mother-to-child transmission cases were excluded from this study. We combined all categories with few cases (except heterosexual exposure, homosexual exposure, and injecting drug use) into an “other” category.
For each risk factor identified in the survey, we created a dichotomous variable based on one or more questions. Some participants were not consistent in responding to different questions addressing the same risk factor; therefore, we constructed logical formulas defining absence or presence of the factor (Table 1). Both in descriptive analysis and in hypothesis testing, we treated these variables as not mutually exclusive, recognizing that one person may be exposed to more than one factor at the same time.
We constructed a summary variable representing the most probable mode of transmission based on the survey responses. Given the strong correlation between HCV and injecting drug use, presence of anti-HCV antibodies was considered a marker of injecting drug use-related transmission. If no anti-HCV antibodies were detected, the survey-based mode of transmission was based on self-reported behavior. If only one risk factor was reported by a participant, the survey-based mode of transmission was assigned the corresponding value. If a participant reported multiple exposures, the survey-based modes of transmission took the value of a risk factor associated with greater probability of transmission per act (16) and higher level of prevalence in respective key populations (12) in the following hierarchy: injecting drug use, homosexual exposure between men, heterosexual exposure, and other. We did not create more detailed or mixed categories, such as those developed by CDC and other authors (17), to enable comparison with Ukrainian registered modes of transmission data. Instead, we present prevalence of all possible two risk factor combinations.
Participants who refused to answer one or more questions required for determining the survey-based mode of transmission were excluded from the analysis of modes of transmission but were retained in the dataset for analysis of other variables.
Sensitivity analysis
To assess the randomness of the sampling approach and thus the representativeness of the survey sample, we compared the registered modes of transmission distribution in the survey sample and the rest of the patients in the registry who did not participate in the survey. The significance of difference for each registered mode of transmission was tested using Chi-square tests.
Hypotheses testing
The main research question was whether there is a difference between the proportion of patients in corresponding registered modes of transmission and survey-based modes of transmission categories. Since these two variables were measured using the same participants and could be considered related, we used the McNemar test to determine the significance of difference between proportions for each major mode of transmission (heterosexual exposure, injecting drug use, homosexual exposure, and other). To assess the extent of agreement between the registration data and survey-based determination of mode of transmission on individual level, we computed Cohen’s Kappa and corresponding p-value. We used the Mantel-Haenszel test for trend (for 2 × r tables) to test the significance of change in the proportion of main transmission categories and prevalence of risk factors over time (18).
To determine the accuracy of the official paper-based reporting system, we used the Chi-square test to determine difference in the distribution of the main modes of transmission between the reporting forms and the verified registry.
P-values less than 0.05 were considered statistically significant. Statistical analysis was done using SPSS for Windows version 23 (IBM Corporation, Armonk, NY USA).
Extrapolation
An adjusted distribution of modes of transmission among total national cases registered in 2015 was imputed by extrapolating the magnitude of misclassification observed in our sample. For each mode of transmission category, we calculated an extrapolation coefficient as a ratio of the proportion observed in the survey to the proportion in the registry. The coefficients were then applied to the reported number of cases within each category to calculate percentages representative of the adjusted modes of transmission among all HIV cases registered in 2015.