Study settings and populations
The study was conducted in health facilities from three administrative zones of Amhara regional state, northwest Ethiopia. A total of 24 health facilities (9 hospitals and 15 health centers) were selected. Five hospitals and six health centers from East Gojjam Zone, two hospitals and six health centers from West Gojjam Zone, and Bahir Dar city administrations two hospitals and three health centers were selected. The region has an adult HIV prevalence of 1.2% which was higher than the national average (0.9%) in 2016 [33].
All newly diagnosed HIV positive individuals in the HIV treatment centers of the 24 health facilities who consented to participate in the study were recruited into the study between December 1st, 2018 and July 30, 2019. The health facilities were included based on an average patient flow of three to five new cases per month. Participants were eligible if they were 18 years old or over, newly diagnosed with HIV and had no prior exposure to ART. The target population was all newly diagnosed HIV-positive patients in the three administrative zones.
Design and sample size
The study design was cross-sectional. The sample size was calculated using a formula for a single population proportion. The proportion of late presenters to HIV care was taken from a previous study in Ethiopia, which was 52.89% [5]. Using the formula;
(Zα/2)2 (P) (p-q)/w2; where Zα/2=1.96, q=47.11%, p=52.89% and marginal error (w) =0.05. Total=383
And considering a non-response rate of 10% and a design effect of two, a total of 844 participants were required.
Sampling procedures
Three administrative zones were selected purposively. Then a total of 71 health facilities (13 hospitals and 58 health centers) providing ART services were identified in the three zones. Health facilities were clustered based on average patient flow as “high load” >3 new patients per month and “low load” <3 new patients per month. A total sample of 844 was distributed for 24 health facilities that fulfilled the selection criteria (fig 1).
Fig 1: Sampling techniques of delayed HIV diagnosis amongnewly diagnosed people living with HIV in northwest Ethiopia, 2019
Data collection tools and procedures
Data collection tools were developed from a review of similar literature on the subject matter [14, 15, 26-32, 34, 35]. Data were collected using exit interview techniques via interviewer-administered questionnaires. Interviews were conducted in a quiet separate room after obtaining written informed consent. Data collectors were nurses who have received training on HIV treatment guidelines and working in the HIV clinic. All newly diagnosed HIV positive individuals were included consecutively until the required sample size was achieved.
Variables and definitions
Delayed HIV diagnosis was defined following the European Late Presenter Consensus working group as a patient who presents for care when there is an established AIDS-defining clinical condition (stage III/IV) irrespective of CD4 count [36]. Independent variables were categorized into individual and community level factors. Individual-level variables included: age, sex, religion, ethnicity, marital status, educational status, employment status, occupation, wealth status, opportunistic infections, body mass index (BMI), functional status, pregnancy status, any current health complaints. Other individual-level variables were alcohol use, disclosure status, HIV test history, number of sexual partners, condom use, HIV and ART related Knowledge, level of perceived stigma. Health facility-related variables included types of health facility, zones, residence and distance from the health facility.
Data processing and analysis
Data were entered into Epi-Data version 3.5 and exported to STATA version 14 for further analysis. Taking into account the nested structure of the data, multilevel logistic regression analysis has been employed. Four models containing variables of interest were fitted using STATA version 14. Model I (Empty model) was fitted without independent variables to test random variability in the intercept and to estimate the intra-class correlation coefficient (ICC). Model II examined the effects of individual-level variables, Model III examined the effect of health facility level variables and Model IV examined the effects of both individual and health facility level characteristics simultaneously. The random effects are the measures of variation in delayed HIV diagnosis across communities expressed as ICC and proportional change in variance (PCV). Akaike’s Information Criterion (AIC) and Bayesian Information Criterion (BIC) were used to choose a model that best explains the data and the model with low AIC value was taken.
The predicting ability of the model (model accuracy) was evaluated using the Receiver Operating Characteristic (ROC) and area under the curve. These indicated that in model III in which only individual-level variables were fitted it was 81.90% while it was 84.23% in the final model (IV) that included the health facility level variables. Multivariate multilevel logistic regression analysis was performed to estimate the adjusted odds ratios (AOR) at a 95% confidence interval (CI).