Determinants of sexually transmitted infections among female sex workers in Ethiopia: a count regression model approach

Background Sexually transmitted infections (STIs) remain a major public health problem worldwide, with the burden of these infections being high among female sex workers (FSWs), who are often not aware of their infection status. This study aimed to determine the factors that are associated with the number of STIs among FSWs in Ethiopia. Methods A cross-sectional bio-behavioral study involving respondent-driven sampling (RDS) was conducted among 6,085 FSWs in 16 towns in Ethiopia. The hurdle Poisson regression model was fitted using STATA Version 16.2. The incident rate ratio and adjusted odds ratio with a 95% confidence interval were employed to show the strength and direction of the association. A p-value of ≤0.05 was used as a threshold for statistical significance. Results At least one STI was identified in 1,444 (23.64%) of the FSWs. Age group 35–49 years [IRR = 2.32; 95% CI (1.43, 3.74)], forced first sex [IRR = 1.32; 95% CI (1.01, 1.74)], condom breakage [IRR = 1.32; 95% CI (1.01, 1.74)], and a history of depression [IRR = 1.55; 95% CI (1.12, 2.18)] increase the number of STIs. FSWs aged 25–34 years [AOR = 2.99; % CI (2.54, 3.52)] and 35 = 59 years [AOR = 8.05; % CI (6.54, 9.91)], who were selling sex for 5–10 years [AOR = 1.30; 95% CI (1.1, 1.55)], and above 11 years [AOR = 1.21; 95% CI (1.03, 1.43)] were more likely to get STIs. Conclusion STIs are common in Ethiopia. The covariates age, educational status, monthly income, condom failure, age at the first sexual encounter, and long duration of sexual practice are significant predictors of STIs. Health interventions among FSWs need to include awareness generation about the prevention and control of STIs and address the determinants identified in this analysis.

Background: Sexually transmitted infections (STIs) remain a major public health problem worldwide, with the burden of these infections being high among female sex workers (FSWs), who are often not aware of their infection status. This study aimed to determine the factors that are associated with the number of STIs among FSWs in Ethiopia.

Methods:
A cross-sectional bio-behavioral study involving respondent-driven sampling (RDS) was conducted among , FSWs in towns in Ethiopia. The hurdle Poisson regression model was fitted using STATA Version . . The incident rate ratio and adjusted odds ratio with a % confidence interval were employed to show the strength and direction of the association. A p-value of ≤ . was used as a threshold for statistical significance. Conclusion: STIs are common in Ethiopia. The covariates age, educational status, monthly income, condom failure, age at the first sexual encounter, and long duration of sexual practice are significant predictors of STIs. Health interventions among FSWs need to include awareness generation about the prevention and control of STIs and address the determinants identified in this analysis. KEYWORDS serology, epidemiology, sexually transmitted infections, female sex workers, hurdle poison regression model Introduction Sexually transmitted infections (STIs) are common and constitute public health concerns globally (1,2). They include viral infections such as human immunodeficiency virus (HIV), hepatitis B virus (HBV), and hepatitis C virus (HCV), as well as other bacterial STIs, specifically syphilis. According to the World Health Organization's (WHO) estimate, more than 1 million people are newly infected worldwide with STIs each day (2). This is equivalent to 374 million people infected per year, of which 96 million are in Africa. Globally, more than 4.5 million people contract HIV and viral hepatitis each year, and among adults aged 15 to 49, 7.1 million new cases of syphilis are identified (2-4).
STIs remain a major public health problem in Africa (5, 6). The Global Burden of Disease (GBD) (1) estimates that the agestandardized incidence rate (ASIR) for STIs is 9,535 per 100,000 person-years, with the highest rate estimated at 19,973 per 100,000 person-years in 2019 being in sub-Saharan Africa (SSA). STIs constitute the second-leading cause of mortality and disabilityadjusted life years (DALYs) in low-and middle-income countries, particularly among those aged 20-24 years (3). Globally, over 2.3 million people died as a result of STIs (4) in 2021, which also accounted for the increased number of years of life lost (1). According to a study conducted in Botswana, DALYs increased with time due to HIV and other STIs (7).
STIs also increase the risk of cancer and account for 13% of global cancer incidence, with the highest rate in SSA in 2018 (4). HBV and HCV are the most common primary causes of cancer (8). STIs increase medical costs (9), are drug resistant, and are associated with an array of maternal and neonatal morbidities (3,10), as well as a stigma among FSWs (11).
STIs are prevalent, particularly among key populations, including adolescents, young adults, and FSWs (4), owing to the high probability of co-infections, overlapping routes of transmission, and common health determinants. FSWs bear the greatest burden of STIs (4) because of their sexual behavior, which puts them at higher risk of acquiring the infection (1,12); their vulnerability to violence and having the most limited access to health and social services exacerbate the problem (4).
The prevalence of STIs and co-infections is high among FSWs (9,18,20). Several factors determining the occurrence of STIs among FSWs have been reported, including age, lower education level (21), and unemployment (21,22). Factors such as frequent unsafe intercourse with various sex workers (21,23), current drug use, inconsistent condom use (24), HIV stigma (22), previous exposure to violence (25), lack of access to treatment, and the ability to pay for services (25) are among the most frequently identified determinants.
Concurrent multiple STIs occurring in individuals are often associated with having multiple sexual partners and can be a source of an STI epidemic among FSWs (26). In this population group, there is variation in STI co-infection prevalence and the associated determinants in different settings (14,18,19). Indeed, there is a need for further studies on the drivers of the number of STIs. The suggested STI categories (zero STI and at least one STI) in the data count used in previous studies might exhibit excess zero counts (no STI), and excluding the zero counts increases the possibility of biased estimates (27). In addition, counting data as non-negative might be over-dispersed and contain excess zeros, making data analysis complex (28). Therefore, in this analysis, we aimed to determine the factors that are associated with the number of STIs among FSWs in Ethiopia using robust statistical methods.

Study design, setting, and population
This study was a cross-sectional bio-behavioral study among FSWs conducted in 16 towns in Ethiopia between December 2019 and May 2020. The target towns include Adama, Addis Ababa, Arba Minch, Bahir Dar, Kombolcha/Dessie, Dilla, Dire Dawa, Gambella, Gonder, Harar, Hawassa, Jimma, Logia/Semera, Mizan, Nekemite, and Shashemane. We conducted respondentdriven sampling (RDS) on 6,085 FSWs aged 15 years and older who had received money or other benefits from selling sex to four or more people in the previous 30 days and had lived or worked in the surveyed town for at least the last month.

Sampling and data collection
To recruit the participants, we employed respondent-driven sampling (RDS), a technique with benefits well-documented in previous reports (29, 30). As an initial step, 5 to 12 initial study participants referred to as "seeds" were selected from each study town. The seed participants were informed about the study, consent was obtained from each, and each participant was provided with three coupons for recruiting three eligible participants from her social network. All newly recruited participants were given three coupons as was done for the initial seeds to invite additional study participants. The data were collected through an anonymous .
interview administered by the study team in a private room using an Open Data Kit (ODK) electronic data management system with built-in skip patterns and logical validations.

Study variables
The outcome variable in this study was the total number of STIs per FSW, categorized as zero STI, one STI, two STIs, three STIs, and four STIs. The independent sociodemographic variables included participants' age, the age at first sex sale, educational and married status, average monthly income, and the duration of sex work. Behavioral factors included alcohol and drug use, condom breakage, desired or forced first sex, HIV knowledge, and depression status.
Compressive HIV knowledge was computed from the four prevention and treatment and three misunderstanding knowledge questions. Respondents with seven true answers were labeled "had comprehensive knowledge" and otherwise "did not have comprehensive knowledge".

The testing procedure and quality control
The study used whole blood to test for HIV, HBV, HCV, and syphilis using a rapid diagnostic kit. HIV testing was done using the national algorithm, which included three rapid tests: assay 1 (STAT-PAK (HIV1/2, USA), assay 2 (ABONE, HIV1/2/O Tri-Line Device, Hangzhou, China), and assay 3 (SD Bioline, HIV1/2, USA). According to the algorithm, those who tested positive for all three were considered HIV-positive.
Hepatitis B surface antigen (HbsAg) detection was performed by the Virucheck HbsAg test kit manufactured in India. A onestep test for HbsAg detects the presence of HbsAg in serum or plasma specimens. Hepatitis C was screened using the Flaviscreen PLUSTM Test Kit produced in India. Flaviscreen is a rapid, third-generation, two-site sandwich immunoassay for the detection of total antibodies specific to the hepatitis C virus. It utilizes the principle of agglutination of antibodies or antisera with the respective antigen in the immunochromatography format.
Syphilis was screened using the Syphicheck-WB Screen and Confirm Assay produced in Kerala, India. Syphicheck is a rapid, qualitative immunoassay for the detection of antibodies to Treponema pallidum. It utilizes the principle of agglutination of antibodies in immunochromatography format. The interpretation of HBV, HCV, and Syphilis test results was according to the manufacturer's guide using the test kit insert.
The study staff received training to ensure the safety and effectiveness of the testing techniques, and standard operating procedures were followed throughout the process, including specimen collection, transportation, testing, and storage. The temperature was monitored while transporting the specimens, and invalid test results were repeated.

Method of data analysis
Data were collected on tablet computers using the ODK software, exported to MS Excel, cleaned, and imported into STATA Version 16 for analysis. The RDS recruitment process (Tree of recruitment), RDS assumption assessment, and RDS weight generation were all carried out using the RDS package inbuilt into R statistical software (30,32). Homophily and convergence, two common assumptions in RDS, were checked in HIV status, consistent condom usage, and FSW type and met the RDS criteria. The RDS weights were exported to STATA using the RDS-II function and merged with the entire dataset for further analysis. Descriptive statistics such as the crude and RDS-adjusted frequency, mean, and standard deviation were computed using RDS-II as a weighting variable. Univariable analysis was conducted to select potential risk factors to be considered in the final multivariable analysis using a modest level of significance (α = 0.25).
The Poisson, negative binomial, zero-inflated Poisson, zeroinflated negative binomial, hurdle Poisson, and hurdle negative binomial models were employed. The Poison regression model is a baseline count model for count data in which the variance of the dependent variable is equivalent to its mean (33). The deviance and Pearson's chi-square statistic values corresponding to their degree of freedom were used to test the presence of overdispersion after fitting the Poisson regression model. In this case, the mean and variance were 0.27 and 0.53, respectively; thus, the assumption is violated, indicating that the data were dispersed. Then, a negative binomial regression model, the extension of the Poisson regression model was fitted to handle the problem of overdispersion in the dataset (33). However, count data often exhibit an excess number of zeros (one cause of over-dispersion) which cannot be accommodated by the Poisson and negative binomial regression models (33). In the presence of zero inflation and overdispersion, zero-inflated (zero-inflated Poisson and zero-inflated negative binomial) and hurdle models (hurdle Poisson and hurdle negative binomial) were frequently used to fit epidemiological data (27,28,33), and they provide a flexible and effective framework for modeling (33). The models have two parts: the first predicts nonzero STI counts (i.e., at least one STI), and the second predicts the zero-hurdle model (zero infections vs. not zero infections) among FSWs.

Hurdle Poisson regression model
The hurdle Poisson regression model has two components: a truncated Poisson component with a rate parameter u, u > 0 that models non-zero positive counts, and a logit component with success probability, π 0 , π 0 ≥ 0 that models the probability of zero counts. If the discrete random variable Y i follows hurdle Poisson distribution, then the hurdle Poisson probability mass function is given as follows: where 0 ≤ π 0 ≤ 1, and defined by π 0 = p (y = 0) For the logit part, the conditional mean is given by E {p . Taking natural logarithms in both sides of the equation, we have as follows: ′ is a vector of regression coefficients, and log( π 0 1−π 0 ) is the log transformation of the odds of at least one STI.
Similarly, the conditional mean for the truncated Poisson is given by Therefore, the truncated Poisson regression model is given by: are a vector of independent variables and regression coefficients, respectively. Each model's goodness-of-fit was evaluated using the Akaike Information Criteria (AIC), and rootogram visual assessment (34,35). Both the count and the zero-inflated parts were analyzed. Finally, a 95% confidence interval (CI) was reported for the incident rate ratio and adjusted odds ratio. A p-value of ≤0.05 was used to define statistical significance.

Ethical considerations
Ethiopian Public Health Institutes' Scientific and Ethical Research Office provided ethical clearance for the survey protocol (Ethical approval number: EPHI-IRB-108-2018). Each survey participant gave her consent to be interviewed, have blood specimens taken, and have the biospecimens stored for testing. Individuals who tested positive for STIs were transferred to the nearest or preferred health facility for appropriate clinical care. All collected information including the test results and seed contact information were kept entirely confidential.

Patient and public involvement
Locally available organizations working on HIV prevention interventions, such as the HIV/AIDS Prevention and Control Office (HAPCO), District Health offices, and Drop-in Centers (DICs) were used to identify the initial participants (seeds) of the survey.
The seeds were selected based on the type of sexual worker, age category, and geographic location of the site. An FSW with a known social network was given three coupons so that she could invite her friends or other FSW contacts that were in her network. This approach allowed the study to reach as many eligible FSWs as possible. Finally, the findings of the study were shared with the FSW associations, HAPCO, and District Health offices through officially written letters and documents and using different platforms such as technical working group meetings and workshops.

Results
The magnitude of STIs Among the 6,085 FSWs involved in the study, 18.2% had HIV, 6.2% had syphilis, 2.5% had HBV, and 0.5% had HCV (Figure 1

Model selection
The distribution of the number of STIs is skewed to the right, signifying the likelihood of over-dispersion. The zero STIs on the bar charts in Figure 2 are highly selected, suggesting that count data models that account for excess zeros, such as zero-inflated models and hurdle models, would better fit the data of the number of STIs. The hurdle Poisson model has the smallest AIC value and  is considered the final model (see Table 3). Furthermore, a visual assessment of the fit was made in terms of the rootogram. Hurdle negative binomial has an inferior fit, with some low numbers over-predicted. The hurdle Poisson model was found to be the best fit for the data based on their respective log-likelihood, AIC, and rootogram ( Figure 3).

Factors associated with the number of STIs
The model is divided into two sections (Table 4): the first predicts non-zero counts of STIs (truncated negative binomial with log link), and the second predicts the zero-hurdle model (binomial with logit link) with zero STIs vs. no zero STIs.

Zero-hurdle model (binomial with logit link)
The second predicts the zero-hurdle model (binomial with logit link) with zero STIs vs. no STIs. The zero-hurdle model's estimated adjusted odds ratio (AOR) and 95% CI for the factor change in the odds of experiencing at least one STI are shown in Table 4. After controlling for all other factors in the model, the probability of having an STI was found to be significantly associated with age, education level, average monthly income, depression, age at selling sex, years spent as FSWs, and condom breakage.
From the results of the hurdle model (binomial with logit link), the odds of having STI (at least one STI) were lower by 31% [AOR = 0.69; 95% CI (0.59, 0.81)] and 48% [AOR = 0.52; 95% CI (0.42, 0.64)] for FSWs who had attended primary and secondary school or above, respectively, compared to those who had no formal education. FSWs with a history of condom breakage were 1.4 [AOR = 1.4; 95% CI (1.21, 1.61)] times more likely to have at least one STI than those without a history of condom breakage. The odds of experiencing at least one STI were 1.21 [AOR = 1.21; 95% CI (1.01, 1.45)] times higher in FSWs who were moderately or seriously depressed compared to those who did not have depression.

Discussion
This study found that 18.2% of the 6,085 FSWs studied had HIV, 6.2% had syphilis, 2.5% had HBV, and 0.5% had HCV. This study found that at least one STI was found in 23.64% of the FSWs, which is consistent with the findings of previous reports from Ethiopia (15), Mexico (16), and Ecuador (13) where at least 17.6% of FSWs were infected with STIs. Our finding could be a higher estimate because the other studies used polymerase chain reaction (PCR) to test for STIs.
The finding among the FSWs in our study is higher than those of the reports from Brazil of 13.3% (14), but lower than the Russian report (11), which showed that 43.2% of participants had at least one STI. This disparity could be due to differences in sociodemographic characteristics, STI diagnostic methods, types, and the number of STIs included in the specific studies.
Age, educational status, and average monthly income, as well as a history of depression, condom breakage, early initiation of sex selling, and living as a sex worker for a long time were associated with the frequency of STIs in our study. This suggests that the country has to make a greater effort to work toward increasing awareness among FSWs as well as the general population and improving prevention, care, and treatment services for STIs among FSWs.
Our study found that FSWs over the age of 30 were more likely to have at least one STI. This finding is comparable to those reported by studies conducted in Namibia (21), the Republic of Congo (29), South Africa (36), Ecuador (13), and Rwanda  Hanging rootograms for count regression models of STIs among female sex workers in Ethiopia, bio-behavioral survey . (19). Studies show an increase in STIs and co-infections with age, probably due to older FSWs not considering themselves high risk and failing to persuade customers to use condoms (13,20). This could also be due to the cohort effect, where older women have more chances of acquiring STIs. Consequently, FSWs in this age group could be the major drivers of STIs among FSWs and their clients unless effective and comprehensive programs are implemented. In contrast, a report from Iran indicates that being under the age of 25 is independently associated with increased STIs (37). This disparity could be associated with sociocultural differences between population groups, but it requires further exploration. Similar to a finding in Rwanda (19), FSWs who were in the business for a long time in our study had higher odds of getting at least one STI. Similarly, those who began selling sex at a young age had the highest odds of having at least one STI, observations also reported from Iran (37). This may be explained by the fact that young sex workers are more likely to report inconsistent condom use and condomless sex with their last clients (38,39). FSWs who started the business at earlier ages are also more likely to drink alcohol heavily (40) and experience multiple clients each day (41). However, these findings contrast the Rwandan study (38), which found no statistically significant association between sex work starting at earlier ages and STIs.
Our finding shows that, compared to those who had no formal education, the odds of experiencing at least one STI was lower among educated FSWs is consistent with findings from the studies in Rwanda (19), Russia (11), and Namibia (21). As reported by others (21), this could be because those with low educational levels are more likely to be unemployed and engage in risky sexual  behaviors associated with STIs (42). Similarly, FSWs with low average monthly income were more likely to experience at least one STI in our series, which could be because the sex workers earned less money as they got older and had to do sex work more often with many clients, and the clients decided to use condoms (41). Condom breakage increases the frequency of STIs and increases the probability of having STIs among FSWs as shown by our study, which identified that those with a history of condom breakage had an increase in the rate of developing at least one STI by 46%. This may be because those experiencing condom breakage were exposed to pornography and used sexual enhancement products (43). The finding is consistent with previous reports for Ethiopia (15) and China (44). Available evidence has also shown that consistent and high levels of condom use among FSWs decrease the incidence of STIs among sex workers as well as the general population.

Strengths and limitations
Overall, the advantage of our study, which was a nationwide survey involving a large sample of FSWs recruited by using the RDS technique from 16 towns across the country, outweighs a venuebased selection approach in terms of obtaining a representative sample. However, as this was a cross-sectional study, temporal relationships between determinants and the outcome cannot be established. In addition, key measures rely on self-report; biases .
/fpubh. . such as social desirability response bias could have some effect. The rapid test/serological markers we used in this survey do not detect the duration of infections as recent or long-term. In this study, we only considered HIV, HBV, HCV, and syphilis, our results could underestimate the overall prevalence of STIs.

Conclusion
The prevalence of STIs among FSWs is high in Ethiopia. Age, educational status, average monthly income, history of depression, history of condom breakage, early initiations of selling sex, and living as FSWs for a long period of time are identified as independent predictors for developing at least one STI. Targeted STI prevention and control programs need to be improved, with a focus on promoting higher education among women, condom distribution, and the creation of awareness on proper and consistent use, enhancing STI testing, prevention, care and treatment, interventions, and supporting income-generating activities. Further epidemiological research is needed on STIs among FSWs in Ethiopia to determine the magnitude of the problem, which should include a broader list of STIs, confirmatory diagnostic tests, and recency testing.

Data availability statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

Ethics statement
The studies involving human participants were reviewed and approved by PHI-IRB-108-2018 and the Ethiopian Public Health Institute. The participants provided their written informed consent to participate in this study.