To inform efforts addressing individual-level factors relating to NPI effectiveness, research has focused on understanding knowledge, attitudes and practices (KAPs) regarding COVID-19. These efforts were particularly important in many African countries where delayed access to vaccines resulted in increased reliance on NPIs.^4,5,6 Knowledge gained from this sort of research has become critical to both addressing the COVID-19 pandemic and preparing for future similar pandemics.

A review of 84 studies across 45 countries found wide variations in KAPs, particularly for adherence to accepted COVID-19 NPIs, which ranged from 27% to 97% with the lowest levels of adherence in low- and middle-income European and African countries.⁷ Furthermore, although knowledge regarding COVID-19 might be expected to translate to NPI practice, this has not been consistently true. A review of studies in healthcare workers around the world found adherence to accepted practices to vary between 30% and 90% despite knowledge of NPIs.⁸ A scoping review of COVID-19 KAPs in sub-Saharan Africa similarly found many studies demonstrating discrepancies between disease knowledge and use of NPIs.⁹ Finally, patterns of COVID-19 KAPs have also varied, with a study in Ethiopia noting a positive correlation between knowledge and practice,¹⁰ whereas a study in Kenya found a negative correlation between knowledge and practice.¹¹ The Health Belief Model, which suggests that individuals weigh perceived benefits of protective behaviours against perceived barriers to implementation, provides a framework for understanding how perceived susceptibility, severity, benefits and barriers influence health behaviours such as NPI adoption. Taken together, these studies highlight the complexity among KAPs for COVID-19 in countries vulnerable to future pandemics.

For individuals with high risk of COVID-19 and its sequelae, understanding KAP patterns is crucial. Populations with behavioural vulnerability to human immunodeficiency virus (HIV), such as individuals who engage in sex for income or inject drugs, may also have particular risk for COVID-19, given overlapping social determinants for COVID-19 and HIV.^12,13 A study within the United States (US) examining the incidence of COVID-19 and HIV found a spatial correlation between the two.¹³ Factors that contributed to the correlation through associations with both COVID-19 and HIV included binge drinking, chlamydia and income inequality.¹³ Finally, social stigma towards individuals with behavioural vulnerability to HIV has been found to increase perceived healthcare stigma, potentially creating barriers to healthcare.^{14,15,16,17,18} In Kenya, the first cases of SARS-CoV-2 infection were diagnosed in March 2020, although the virus was likely circulating even earlier, with lockdowns and other nationwide mitigation efforts implemented almost immediately thereafter.¹⁹ Two studies identified large barriers to providing HIV testing because of social stigma that were exacerbated by COVID-19 policies such as restrictions on clinic capacity and operating hours.^20,21 Transient reductions in HIV viral load testing and food security were observed early in the pandemic in several African countries.²² Therefore, understanding COVID-19 KAPs in populations with high HIV incidence is critical to preventing further COVID-19-related morbidity and mortality.

Kenya has a particularly large burden of HIV with an estimated national adult HIV prevalence in 2023 of 3.2%.²³ There is geographic heterogeneity in Kenya’s HIV epidemic with HIV prevalence of over 16% in certain regions.²⁴ The Kenyan HIV epidemic also disproportionately affects specific subgroups of the general population such as men who have sex with men (MSM); people who engage in transactional sex; and adolescent girls and young women.^25,26 Understanding COVID-19 KAPs among these populations is crucial given their shared barriers to healthcare access. We used cross-sectional questionnaire data from enrolment into an observational cohort of people with behavioural vulnerability to HIV in Western Kenya to examine COVID-19 KAPs with the goal of informing strategies to decrease transmission risk of SARS-CoV-2 and other future pathogens.

This study is a cross-sectional analysis of questionnaire data collected at enrolment into a longitudinal cohort study.

Participants were recruited from two sites in Kenya: the Kericho Clinical Research Center in Kericho and its satellite location in Homa Bay. Sites were selected because of a long-standing partnership between the US Military HIV Research Program and the Kericho Clinical Research Center.

From December 2021 to March 2023, the Multinational Observational Cohort of HIV and other Infections (MOCHI) enrolled individuals aged 14–55 years without HIV. Participants were recruited through community outreach from hotspot areas in Kericho and Homa Bay. Participants were invited to the clinic for a briefing session, and if interested in participating in the study, were consented and screened for eligibility. Participants aged as young as 14 years were included because of known early sexual debut in Western Kenya region and previous data showing that sexual behaviours associated with vulnerability to HIV and other sexually transmitted infections (STIs) tend to be common in the years following sexual debut.^27,28,29 The primary objective of the longitudinal cohort study was to estimate HIV incidence within populations with behavioural vulnerability to HIV. Behavioural vulnerability was defined as satisfying one or more of the following criteria in the 24 weeks prior to screening for study eligibility: (1) newly diagnosed syphilis, gonorrhoea, chlamydia, Mycoplasma genitalium, herpes simplex virus and/or acute hepatitis C virus; (2) engagement in vaginal, oral or anal intercourse in exchange for money as a regular source of income; (3) engagement in condomless vaginal or anal intercourse with at least three different partners living with HIV or of unknown status; (4) engagement in injection drug use or (5) engagement in insertive or receptive anal intercourse with one or more different male partners. These characteristics do not represent the general population, but reflect the target population of participants with high risk of STIs. Exclusion criteria included individuals who were already living with HIV, pregnancy at the time of screening, diagnosis of a substance dependence that would impair participation in the study or having previously received an investigational HIV vaccine. These exclusion criteria were included to mimic a study population that may someday be recruited into a clinical trial of an HIV prevention product.

Sample size for the cohort was calculated based on a goal of confidently detecting an HIV incidence rate of at least 3 cases per 100 person-years. With target accrual of 400 participants and an anticipated dropout rate of 30%, 19 incident HIV cases would need to be observed to determine with 95% confidence that the true HIV incidence was no less than 3 cases per 100 person-years (estimated incidence 4.90 [95% CI: 3.01–7.45] per 100 person-years). These cross-sectional analyses included all participants who completed a COVID-19 questionnaire at enrolment.

At enrolment, participants completed questionnaires using computer-assisted self-interview. The socio-behavioural questionnaire ascertained, among other questions, if participants were male and reported male sex partners in the past 12 weeks or engaged in transactional sex (responding ‘Yes’ to ‘Are you a person who is a sex worker [sex in exchange for things such as money, drugs, food, shelter or transportation]?’). The questionnaire also ascertained participant weekly income, which was dichotomised to ≤ KES 1500.00 (Kenyan shillings) and > KES 1500.00, which is approximately the 2023 Kenyan poverty level.³⁰

Questionnaire items were designed to assess key constructs from behavioural theory including perceived benefits (knowledge statements), perceived barriers (attitude questions) and behavioural outcomes (practice measures). The COVID-19 questionnaire focused on COVID-19 KAPs in the preceding month, including any use of NPIs to prevent SARS-CoV-2 acquisition or transmission. Coronavirus disease 2019 knowledge was assessed with two questions. The first question asked participants to ‘Please classify the following statements as either “true” or “false”’ with statements regarding SARS-CoV-2 transmission (e.g., ‘COVID-19 can be transmitted through coughing or sneezing’). Responses were dichotomised to true versus false or don’t know. The second question asked participants ‘For each behaviour, please indicate the extent to which you agree or disagree that it prevents the spread of COVID-19’ with behaviours including personal NPIs (e.g., handwashing) and social NPIs (e.g., reducing shopping trips). Responses were dichotomised to agree (strongly agree or somewhat agree) or disagree (strongly disagree, somewhat disagree, neutral or don’t know).

COVID-19 attitudes were assessed by asking participants two questions. The first question ascertained participants’ opinions regarding COVID-19. Participants were asked, ‘Please select the answer depending on how much you agree with each statement below’ with statements such as ‘My becoming infected with COVID-19 poses a risk to others’. Responses were dichotomised to agree (strongly agree or somewhat agree) versus disagree (strongly disagree, somewhat disagree, neutral or don’t know). The second question ascertained participants’ beliefs regarding how their communities behaved. Participants were asked ‘In your community, to what extent do you think that people do the following when they go out in public?’ with statements of social distancing and wearing a face mask. Responses were dichotomised to always or often (always or often) versus occasionally or less (occasionally, rarely, never, or do not know).

COVID-19 prevention practices were assessed by asking participants ‘In the past month, what measures have you taken to prevent infection from COVID-19?’ with a list of personal NPIs (e.g., face masks) and social NPIs (e.g., meeting in the open). Responses were dichotomised into yes or no (no, don’t know, refuse to answer or no response). To assess personal experience with COVID-19, participants were also asked ‘Have you been infected with COVID-19’ and ‘Do you personally know anyone other than yourself who has been told by a doctor that they have COVID-19?’, and responses were dichotomised into yes (Yes confirmed or Yes suspected but not confirmed) or no (No, do not suspect or don’t know).

The date of questionnaire completion was recorded and later assigned to a SARS-CoV-2-variant wave, if applicable. Waves were defined to begin on the first day of two consecutive 7-day periods with > 20% increase in COVID-19 cases.³¹ Waves were defined to end where two consecutive 7-day periods had fewer cases than at the beginning of the wave or at the beginning of two consecutive 7-day periods with < 10% absolute weekly change in the number of cases, whichever occurred first. The waves were defined as Omicron (29 November 2021 – 14 February 2022), Omicron BA.4/BA.5 (02 May 2022 – 22 August 2022) and Omicron BQ1/BQ.1.1 (10 October 2022 – 02 January 2023) (Appendix 1 Figure 1-A1).³²

In total, there were 28 potential independent variables including demographic items, COVID-19 knowledge, COVID-19 attitudes, personal experience with COVID-19 and COVID-19 wave. Given that COVID-19 NPIs directly decrease SARS-CoV-2 transmission, NPIs were the outcome variables of interest. NPIs that had near universal agreement were described but excluded from further analyses. Descriptive statistics included median and interquartile range (IQR) for continuous variables, and frequency and percent for categorical variables, which were estimated overall and stratified by site. Missing data for descriptive statistics were addressed by using an available-case approach, which utilised all data available for each descriptive analysis. Robust Poisson regression was used to estimate prevalence ratios (PRs) and 95% confidence intervals (CIs) for all inferential analyses. Univariable analyses were performed to estimate unadjusted PRs and 95% CIs for the relationship between each independent variable and each NPI. In multivariable analyses for each COVID-19 NPI, purposeful variable selection was used to determine which of the 28 candidate independent variables to include in each model.³³

Purposeful variable selection included the following four steps: (1) Univariable regression analyses were run for each candidate predictor variable. Those variables with p < 0.25 were then entered into a multivariable model. (2) One variable at a time was removed among those with p > 0.05 in the multivariable model from step 1. If a variable had p > 0.05 but its removal resulted in a > 20% change in one or more other regression coefficient from step 1, then that variable was retained; otherwise, that variable was removed. This process continued until all variables with p > 0.05 were examined. The remaining variables were retained for all further analyses. (3) Next, each of those variables that had p > 0.25 in univariable analyses (step 1) were added, one at a time, to the multivariable model (from step 2). Any variables with p < 0.05 were included in the final model. (4) Lastly, diagnostic checks were performed on the final model from step 3. Missing data for regression analyses were addressed by using a complete-case approach, which included only participants with non-missing variables for all candidate variables; this restricted analytic population was used for each univariable regression and the final multivariable regression.

Statistical analyses were performed using SAS version 9.4 software (SAS Institute, Cary North Carolina, US). Figures were developed using R version 4.2.2 (R Foundation for Statistical Computing, Vienna, Austria). P-values < 0.05 were considered statistically significant. P-values were not corrected for multiple comparisons.

Ethical clearance to conduct this study was obtained from the Walter Reed Army Institute of Research (No. WRAIR #2877B). The investigators have adhered to the policies for protection of human research participants as prescribed in Army Regulation 70-25. All participants provided written informed consent prior to any study procedures. Participants aged 14–17 years with behavioural vulnerability to HIV and other STIs were considered outside of parental influence or control and provided their own informed consent as emancipated minors (mature minors), consistent with the Guidelines for Conducting Adolescent HIV Sexual and Reproductive Health Research in Kenya. Procedures for participant protection and confidentiality included the use of study identification numbers and locked, secured data storage. The study was approved by institutional review boards at the Kenya Medical Research Institute, the Walter Reed Army Institute of Research and all collaborating institutions.

Of the 486 participants screened, 413 were deemed eligible, of whom 407 enrolled in the study. Of the 407 participants enrolled in Homa Bay and Kericho, Kenya, 399 (98.0%) completed any of the COVID-19 questionnaire (Table 1). Their median age was 22 years (Q1–Q3 19–24), with 318 (79.7%) participants between 15 years and 24 years old. Key demographic patterns showed predominance of young females (79.4%) engaged in transactional sex (79.2%), with most participants having secondary education or less (65.9%).

Knowledge of COVID-19 NPIs was generally high, with over 80% of respondents agreeing that each of the solicited NPIs may reduce the risk of SARS-CoV-2 transmission (Figure 1a). When asked ‘In the past month, what measures have you taken to prevent infection from COVID-19’, the NPIs with the most agreement were face masks (n = 353 participants, 88.5%), social distancing (n = 347, 87.0%) and handwashing (n = 343, 86.0%). The NPIs with the lowest agreement were reducing shopping trips (n = 325, 81.5%) and meeting in the open (n = 326, 81.7%) (Appendix 1 Table 1-A1).

However, COVID-19 knowledge varied (Figure 1b). Nearly all participants correctly identified that transmission can occur through coughing or sneezing (n = 384, 96.2%). Over 80% of participants correctly identified that it is possible to have no symptoms from SARS-CoV-2 infection (n = 333, 83.5%) and also to transmit the virus without having symptoms (n = 350, 87.7%). However, there was lower knowledge of SARS-CoV-2 genetic variants, with smaller majorities of participants correctly identifying that SARS-CoV-2 variants can have differing virulence (n = 309, 77.4%) and that vaccines have lower effectiveness in preventing some SARS-CoV-2 variants (n = 268, 67.2%) (Appendix 1 Table 2-A1).

Attitudes about COVID-19 were assessed for one’s personal risk and belief in their community’s actions. When asked about the risk of SARS-CoV-2 infection (Figure 2a), most participants agreed that becoming infected with COVID-19 poses a risk to others (n = 287, 71.9%). Most participants were also concerned about the spread of COVID-19 in their communities (n = 270, 67.7%) and getting infected with COVID-19 (n = 263, 65.9%) (Appendix 1 Table 3-A1). However, fewer than half of participants believed that people in their community used NPIs such as social distancing (n = 171, 42.9%) and wearing face mask (n = 188, 47.1%) (Figure 2b, Appendix 1 Table 4-A1).

Among the COVID-19 NPIs, personal NPIs were the most practiced in the month prior to enrolment including handwashing (n = 390, 97.7%) and mask-wearing (n = 380, 95.2%) (Figure 3). Furthermore, social distancing and avoiding contact with sick individuals were also endorsed by 372 (93.2%) and 365 (91.5%) respondents, respectively. The least endorsed measures were the social NPIs of meeting in the open (n = 312, 78.2%), reducing shopping (n = 325, 81.5%) and avoiding crowds (n = 331, 83.0%). There were 250 (62.7%) participants who endorsed adhering to all seven NPIs (Appendix 1 Table 5-A1).

Among the COVID-19 NPIs, the social NPIs of meeting in the open, reducing shopping and avoiding crowds were modelled as outcome variables given their lower levels of uptake. Among the 399 participants who completed any of the COVID-19 questionnaire, 349 participants completed the entire questionnaire and were included in inferential analyses. In univariable analyses (Appendix 1 Table 6-A1 to Table 8-A1), participants responding during the Omicron BA.4/BA.5 wave were less likely to endorse reducing shopping and avoiding crowds (PR = 0.85 [95% CI: 0.76–0.96] and 0.87 [95% CI: 0.77–0.97], respectively). Knowing someone previously diagnosed with COVID-19 was associated with increased probability of avoiding crowds (PR = 1.10 [95% CI: 1.00–1.20]). Broadly, participants with correct knowledge of COVID-19 (asymptomatic transmission and of the impact of genetic variants on transmission) and of NPIs relating to reducing close contact (reducing shopping trips and meeting in the open) were more likely to endorse practicing those same NPIs. Finally, attitudes regarding individual concern for COVID-19 and community use of NPIs were not related to practice of any of the NPIs evaluated.

In adjusted analyses, purposeful variable selection improved the model performance compared to models that included all variables. For meeting in the open, purposeful variable selection resulted in a model with Akaike information criterion (AIC) of 696.8 compared to AIC of 725.3 for the full model. For reducing shopping, the AIC after purposeful variable selection was 694.1 compared to 726.5 for the full model. For avoiding crowds, the AIC after purposeful variable selection was 714.0 compared to 733.9 for the full model.

Male sex and engagement in transactional sex were positively associated with meeting in the open (adjusted PR [aPR] = 1.27 [95% CI: 1.11–1.45] and 1.24 [95% CI:1.02–1.50], respectively) (Table 2). Older participants (25–55 years vs. 14–24 years) and those who were married were more likely to reduce shopping (aPR = 1.13 [95% CI: 1.00–1.28] and 1.19 [95% CI: 1.01–1.41], respectively) (Table 3). Finally, participants with more years of education (> 12 years vs. < 12 years) were more likely to avoid crowds (aPR = 1.19 [95% CI:1.08–1.31]) (Table 4).

For all outcomes, NPI usage varied by COVID-19 wave. Compared to non-wave periods, the initial Omicron-variant wave was associated with less meeting in the open (aPR = 0.85 [95% CI: 0.73–0.98]), reduction in shopping (aPR = 0.83 [95% CI: 0.73–0.96]) and avoidance of crowds (aPR = 0.81 [95% CI: 0.71–0.93]). On the other hand, the second wave was associated with increased probability of reducing shopping (aPR = 1.14 [95% CI: 1.03–1.25]).

Knowledge of the efficacy of personal COVID-19 NPIs was associated with a lower probability of practicing social NPIs. In particular, participants who agreed that wearing face masks or covering correctly prevented the spread of SARS-CoV-2 were less likely to meet in the open or reduce shopping (aPR: 0.44 [95% CI: 0.27–0.73] and 0.48 [95% CI: 0.31–0.76], respectively), and those who agreed that washing hands regularly using soap and water or sanitiser prevents the spread of SARS-CoV-2 were less likely to endorse avoiding crowds (aPR: 0.73 [95% CI: 0.60–0.89]). On the other hand, belief in the efficacy of social NPIs including meeting in the open, reducing shopping and avoiding crowds were each associated with endorsing practicing the same NPI (aPR: 1.57 [95% CI: 1.11–2.21], 1.94 [95% CI: 1.31–2.87], and 1.36 [95% CI: 1.05–1.77], respectively). Finally, participants who believed that people in their communities often wear a face mask or covering were more likely to reduce shopping (aPR: 1.12 [95% CI:1.02–1.23]).