Trends and predictors of HIV positivity and time since last test at voluntary counselling and testing encounters among adults in Kilifi, Kenya, 2006-2017

Background: Little is known about HIV retesting uptake among key populations (KP) and general populations (GP) in Kenya. We assessed trends and predictors of first-time testing (FTT), late retesting (previous test more than one year ago for GP or three months for KP), and test positivity at three voluntary counselling and testing (VCT) centres in coastal Kenya. Methods: Routine VCT data covering 2006-2017 was collected from three VCT centres in Kilifi County. We analysed HIV testing history and test results from encounters among adults 18-39 years, categorized as GP men, GP women, men who have sex with men (MSM), and female sex workers (FSW). Results: Based on 24,728 test encounters (32% FTT), we observed declines in HIV positivity (proportion of encounters where the result was positive) among GP men, GP women, first-time testers and MSM but not among FSW. The proportion of encounters for FTT and late retesting decreased for both GP and KP but remained much higher in KP than GP. HIV positivity was higher at FTT and late retesting encounters; at FSW and MSM encounters; and at encounters with clients reporting lower educational attainment and sexually transmitted infection (STI) symptoms. HIV positivity was lower in GP men, never married clients and those less than 35 years of age. FTT was associated with town, risk group, age 18-24 years, never-married status, low educational attainment, and STI symptoms. Late retesting was less common among encounters with GP individuals who were never married, had Muslim or no religious affiliation, had lower educational attainment, or reported STI symptoms. Conclusions: HIV positive test results were most common at encounters with first-time testers and late re-testers. While the proportion of encounters at which late retesting was reported decreased steadily over the period reviewed, efforts are needed to increase retesting among the most at-risk populations.


Introduction
Kenya has the fifth-largest human immunodeficiency virus (HIV) epidemic in the world 1 , with 1.3 million adults living with HIV in 2018 2 .Data from sentinel surveillance and national population-based surveys indicate that national HIV prevalence peaked at 10%-11% in the mid-1990s and declined to about 6% in 2006 1,3,4 .Prevalence has remained relatively stable at that level for several years with a modest decline observed from 2010 to 2017 5 .In 2018, national prevalence was estimated at 4.9%, higher in women (6.6%) than men (4.5%) 2 .The epidemic is geographically diverse, with prevalence ranging from 19.6% in Homa Bay county in the west to <0.1% in Garissa county in the north-east 2 .There were approximately 36,000 new infections in 2018 2 , with more than a third occurring among young women 15-24 years 5,6 .Key populations, including men who have sex with men (MSM) and female sex workers (FSW) remain disproportionately affected by HIV.In 2017, prevalence was estimated at 18% among MSM and 29% among FSW 6 .County-level prevalence estimates for key populations are not available.
The proportion of Kenyan adults 15-64 years who have ever tested for HIV increased from 37% in 2007 to 70% in 2012 4,7 , and to 80% in 2014 8 .This tremendous increase in testing coverage is the result of an expanded testing program, including voluntary counselling and testing (VCT), routine (opt-out) provider-initiated testing in health facilities, routine testing in prevention of mother-to-child transmission programs, home-based (door-to-door) testing, mobile testing, and annual testing campaigns.However, knowledge of HIV status remains low.In 2018, it was estimated that 79.5% of people living with HIV knew their status 2 .This falls short of the UNAIDS target of 90% and plays a major role in ongoing transmission 9 .It is estimated that 54-90% of new transmission events arise from persons with undiagnosed infection [10][11][12][13] .
Low knowledge of HIV status may be attributable in large part to infrequent testing.Current national HIV testing guidelines recommend retesting quarterly for key populations (KP) and annually for the general population (GP) 14 .In 2012, a population survey estimated national retesting uptake at 55% among all adults 15-64 years 7 .A more up-to-date estimate is not available, but a repeat survey was underway in 2018.Little is known about retesting uptake at the sub-national level or the factors that predict adherence to recommended retesting frequency.To address such information gaps, data collected at VCT centres can supplement population-based surveys 15,16 , if regularly and rigorously analysed.Currently, test data collected at various testing facilities are reported to county and national headquarters only in summary form, combining VCT and other testing points, and not disaggregated by risk groups.
Kilifi county, one of the six counties in the coastal region of Kenya, is among the poorest counties in Kenya 17 , with low literacy levels and high rates of school dropout affecting both girls and boys 18,19 .In 2017, 30,597 adults were living with HIV in the county, for an estimated HIV prevalence of 3.8% 5 .In the same year, the county experienced 1,380 new infections, with a third occurring among adolescents and young people in the age-group 15-24 years 5 .
In the present study, we used routine data collected over a period of 12 years at three VCT centres located in three neighbouring towns in Kilifi county, to assess trends in HIV positivity (proportion of test encounters where the result was positive), and the proportion of encounters at which clients reported first-time testing (never tested before) or late retesting (previous test more than one year ago for GP or three months for KP).Information on these outcomes in different sub-groups who utilize VCT services can support the targeting of HIV prevention efforts.

Study setting and population
Data were collected at three VCT centres operated by the Kenya Medical Research Institute (KEMRI)-Wellcome Trust Research Programme (KWTRP) in Kilifi county (Figure 1, population 1.4 million 20 ).The three centres followed the serial testing strategy recommended in national guidelines 14,21 .The centres served clients seeking testing out of their own initiative (walk-in clients) and clients mobilized during periodic campaigns by KWTRP outreach workers (mobilized clients).
The oldest of these centres started operating in 2006 and is situated within the KWTRP main campus in Kilifi town, 60 kilometres (km) north of Mombasa (the second largest town in Kenya), and approximately 500 meters from the Kilifi County Hospital.The estimated catchment population for the county hospital is 125,500 22,23 .HIV testing for the general population at the hospital started in 2004, and a large comprehensive HIV care centre was set up in 2005.
The second centre started operating in 2010 and is situated on the premises of the KWTRP clinic in Mtwapa town, 20 km north of Mombasa.Its estimated catchment population is 116,000 22,23 .The town has a busy nightlife, with a large number of bars, nightclubs and hotels among other businesses, including many private health facilities and pharmacies 24 .Since 2005, the KWTRP clinic has conducted cohort studies among KP, including MSM and FSW 25 .The centre was set up following a request by community leaders who wanted clinic services to be accessible to the general population in the area.

Amendments from Version 2
Figure 2 and Figure � � were updated to include 24 participants that had been excluded erroneously.In the limitations section of discussion, two references were added: Soni et al., 2020 to support the statement on potential bias caused by misreporting of HIV status; and Ellison et al., 2014 to support the new statement on the potential bias caused by the use of p-value cut-offs in model building.Also, a number of typographical errors were corrected.

Any further responses from the reviewers can be found at the end of the article
The third centre also started operating in 2010 and is situated at a KWTRP-supported drop-in centre within the sub-county hospital in Malindi town, 120 km north of Mombasa.Its estimated catchment population is 128,000 22,23 .This centre initially targeted MSM and FSW, but increasingly served the general population.During the period 2014-2015, KWTRP collaborated with community-based organizations to mobilize KP for testing.

Data collection procedures
For each test encounter during the study period, a data collection form was completed by VCT staff capturing type of client (walk-in or mobilised in an outreach campaign), test location (Kilifi, Mtwapa, or Malindi), test date, gender, date of birth, highest level of education, religious affiliation, marital status, reason for testing, HIV test results, whether the client had ever tested before, and date of previous test (whether at our VCT or any other testing site).Starting in 2010, data were collected on self-reported HIV risk behaviour in the past six months, including: gender of sex partners, receipt of payment for sex, and current symptoms of sexually transmitted infections (STI).STI symptoms included: for men, urethral discharge and dysuria; for women, excessive or foul-smelling vaginal discharge; and for both men and women, genital sores and history of rectal discharge for those who reported anal sex.VCT records were extracted in early 2018, cleaned, and prepared for analysis.

Sample selection
Our sample selection was guided by our goal to assess trends in adult walk-in VCT clients (i.e., clients seeking testing out of their own initiative).We therefore excluded data from mobilized clients who were tested during outreach campaigns and may have felt social pressure to test, even if previously diagnosed.In addition, the frequency and intensity of outreach campaigns varied over time, making it difficult to evaluate time trends.We also excluded VCT clients seeking confirmatory testing after a positive test done elsewhere, partners of HIV-positive index clients, Malindi clients from 2010-2011 (a period when testing exclusively targeted MSM), and clients outside the age group 18-39 years, where HIV incidence is highest in Kenya.We included 24,728 (52%) of all 47,893 test encounters in the original dataset (Extended data: Supplementary

Data analysis
Data cleaning, recoding and analysis was conducted using Stata ® version 15 (StataCorp, USA).Based on sex, sex of sex partners, and report of transactional sex (collected since 2010), we categorized clients into four risk groups: GP men, MSM, GP women, and FSW.As sexual behaviour data was not collected before 2010, test encounters from that period (all Kilifi-based) were categorized as GP.
The three main outcomes were HIV positivity (proportion of test encounters where the result was positive), proportion of encounters at which clients reported first-time testing (FTT), and proportion of encounters at which clients reported late retesting (previous test more than one year ago for GP or three months for KP).One year was defined as 365 calendar days, and three months as 90 days.We assessed change in outcomes over calendar year using locally weighted regression ( 26 , Stata package "lowess").
Using multivariable log binominal regression ("binreg") and data from the period when information on sexual behaviour was complete (2012-2017, n=19,298), we assessed factors associated with the three outcomes.Given the difference in definitions of late retesting for GP and KP, we fit separate GP and KP models for this outcome.Age and sex were included a priori in all models; all other variables for which p<0.10 in bivariable analyses were carried forward in multivariable models.Factors with p<0.05 in the multivariable model were considered to have statistically significant associations with the outcome in question.For the FTT model, we assessed interactions between study area and risk group.

Characteristics of testing encounters
Of 24,728 tests conducted in the period 2006-2017, 50% were conducted in Mtwapa, 33% in Kilifi, and 16% in Malindi (Table 1).Overall, 56% of encounters were among men, 68% among never-married individuals, 73% among Christians, and 41% among those with secondary education; 92% were among GP and 9% among KP; 32% were FTT encounters and 22% involved clients who were late retesters, that is, had a previous test more than a year ago for GP or three months for KP.

Time trends in the proportion of encounters with firsttime testers
For GP, we observed a decline in the proportion of encounters where the client was testing for the first time among men overall, women overall, and women aged 18-24 years (Figure 2).Slopes were similar for all three sub-groups.For KP, the proportion of encounters that involved FTT declined less steadily, with the lowest percentage-point decline per year observed in MSM.
For the final year assessed (2017), the proportion of encounters involving FTT was 15% for GP clients: 16% for men, 13% for women, and 20% for women aged 18-24 years.The proportion of encounters involving FTT was 29% for KP: 42% for MSM and 9% for FSW.

Time trends in the proportion of encounters with late retesters
We observed declines in the proportion of encounters involving late retesting for both GP (previous test more than one year ago) and KP (previous test more than three months ago) (Figure 3).Throughout the period assessed, the proportion of encounters involving late retesting among the KP remained much higher than that in GP.The percentage-point changes per year were similar for all sub-groups assessed.
For the final year assessed (2017), the proportion of encounters involving late-retesting was 28% for GP: 29% for encounters with men, 28% for encounters with women, and 25% for encounters with women aged 18-24 years.The proportion of encounters involving late retesting was 83% for KP: 81% for MSM encounters and 85% for FSW encounters.

Time trends in HIV positivity
For GP, there was a decline in overall HIV positivity at encounters with both men and women, as well as with the sub-group of those testing for the first time, but not among female late re-testers (Figure 4).
For encounters among MSM, HIV positivity was steady among encounters involving late re-testing (Figure 5).For encounters among FSW, there was an increase in HIV positivity over time.This was also true for encounters with FSW involving first-time testing (large increase at +4.9 percentage points per year) and late re-testing, but not for encounters with FSW involving on-time re-testing.
For the final year assessed (2017), overall HIV positivity for GP encounters was 2.3%: 1.1% for encounters with men, 3.9% for encounters with women, and 2.8% for encounters with women aged 18-24 years.Overall HIV positivity in KP encounters was 7.8%: 6.0% for MSM encounters and 10.7% for FSW encounters.     4 and 5).MSM: men who have sex with men; FSW: female sex workers.

Factors associated with first-time testing encounters
As presented in Table 2, factors associated with increased probability of FTT at VCT encounters included: test location (Kilifi and Malindi), age 18-24 years, never-married status, lower educational attainment.Compared to GP women encounters, MSM and GP men encounters were more likely to involve FTT, while FSW encounters were less likely to involve FTT.First-time testing encounters were less likely during earlier testing periods and among clients with current STI symptoms.No interactions between study area and risk group were identified (data not shown).

Factors associated with late retesting encounters
The GP model is presented in Table 3, and the KP model in Extended data: Supplementary Table 10.The KP model did not identify any predictors of late retesting (previous test more than three months ago).
In the GP model, encounters involving late retesting (previous test more than one year ago) were less likely among never-married clients, clients professing Muslim or no religious affiliation, those with secondary education, and those with current STI symptoms.Encounters involving late retesting were more likely during 2012-2014 and among clients served in Kilifi compared to Mtwapa.6 and 7).8 and 9).Encounters where a positive HIV result was less likely were among clients under 35 years, those who were never married, and GP men.

Discussion
Analysis of 12-year data from three VCT centres in Kilifi county, Kenya, revealed a decline in the proportion of encounters involving first-time testing (those who had never tested before) among GP men, GP women, GP women aged 18-24 years, and FSW; suggesting increasing coverage of HIV testing in the county, in line with national trends 6 .However, the proportion of encounters involving FTT among MSM was relatively constant, and the prevalence of FTT encounters among MSM in the final year assessed (2017) was relatively high at 42% (compared to 15% in GP and 9% in FSW).We also found an overall decline -albeit more modest -in the proportion of encounters involving late retesting, but this remained, in absolute terms, much higher among KP, for whom more frequent testing is recommended, compared to GP. services [30][31][32] HIV self-testing (HIVST) services were introduced in Kenya in 2017 in order to improve test uptake among hardto-reach populations including KP, men and young people 33 .Scaling up of HIVST and partner notification services among KP, including through innovative strategies such as peer test distribution, has been shown to increase test uptake in this population [34][35][36][37] .
In our study, low educational attainment was associated not only with encounters for FTT but also with HIV positivity.Testing encounters among clients with primary or no education were 1.6 times more likely to result in a positive HIV test, compared to those among clients with higher education, while those among clients with secondary education had 1.5 times the likelihood.Kilifi County is amongst the poorest counties in the country 17 , and has low literacy levels 18,19 .Specifically, in Kilifi county, educational outreach and targeted HIV testing programs tailored to the needs of low-literacy, rural populations might improve HIV testing services.For instance, HIV knowledge and literacy could be assessed among patients seeking healthcare, and patients with no or low level of education could be offered brief education sessions with visual aids and confidential HIV testing with clear and simple messages.
Community outreach could also help to dispel myths about HIV and increase awareness and uptake of services.
In Kenya and other similar settings, adolescents and young women 15-24 years are disproportionately affected by HIV 38 .
In 2017, this sub-population accounted for more than a third of all new adult HIV infections in Kenya 5,6 ; HIV prevalence in this group was estimated at 2.6% 5,6 .In the present study, HIV positivity at VCT encounters by young women 18-24 years was 2.8% in 2017.Initiatives that tackle social determinants of HIV risk in this vulnerable group, such as poverty, gender inequality, and sexual violence are needed [39][40][41] .However, resources to implement such initiatives may be limited, since Kilifi is categorized as a medium priority county for HIV prevention and care.Less donor-dependent interventions, such as sex education at primary and secondary schools, will be crucial and could be rolled out in tandem with HIV education aimed at improving health literacy.
Our findings suggest unequal delivery of HIV prevention services across the county.Testing history and HIV positivity at VCT encounters varied by town, with Malindi having the lowest testing prevalence and highest HIV positivity.
Malindi is more geographically isolated, being furthest from Mombasa -Kenya's second largest city, main seaport and former administrative headquarters for the coast province.On the other hand, the town has a vibrant tourism sector which attracts large numbers of KP.Clearly, greater coverage of HIV testing and prevention services is needed in this area, with a strong focus on KP.
This study demonstrates the utility of rigorous analysis of routinely collected data to evaluate trends in first-time testing, late retesting, and HIV positivity at VCT encounters at a county level 42 .Currently, test data collected at various testing facilities are reported to county headquarters only in summary form, combining data from VCT centres and other testing points such as provider-initiated testing in outpatient and antenatal clinics; the data is also not disaggregated by risk groups.Our findings also show that additional socio-demographic, sexual behaviour, and testing history data can be useful in identifying subpopulations in need of additional education and outreach, as well as targeted HIV prevention and care services.
Our study had a number of limitations.First, we cannot be sure that encounters with a positive test result documented were new diagnoses of HIV infection, as stigma and social desirability bias may lead some clients to report their previous test result as negative even if it was positive 43,44 .Second, social desirability bias may also have resulted in over-reporting of previous HIV test uptake.Third, stigma and discrimination towards MSM may have resulted in under-reporting of same-sex behavior practices among men and sex work stigma may have resulted in under-reporting of transactional sex among women.Additionally, our dataset lacked information on sexual behaviour prior to 2010, limiting our ability to describe trends by risk group in that period.Fourth, the data capture system we used did not track individual testers longitudinally, precluding our ability to analyse individual testing practices over time.As one individual's multiple retesting episodes were counted as individual encounters, this may have biased our modelling.Fifth, cross-site comparisons of time trends may have been biased by changes in covered populations in the different VCT centers over time.Sixth, although we excluded clients mobilized through outreach activities, some of the clients registered as walk-in may have been influenced indirectly by outreach activities, hence the sample used may not be wholly representative of the walk-in VCT clientele.Seventh, our data do not enable us to hypothesize about mechanisms underlying some findings, such as associations with religion, and some findings may be due to chance or residual confounding.In particular, the use of p values to select variables for model building can be misleading 45 .Finally, the three VCT centres included in the study are close to KWTRP research clinics, hence clients may not be representative of the whole VCT clientele in the county.

Conclusions
Our study showed that in Kilifi county, HIV positivity at encounters in the three VCT centres studied was most common when encounters involved first-time testing, testing less than annually, key populations, and persons with lower educational attainment.While encounters involving first-time testing and late retesting decreased over time, potentially reflecting increased testing coverage, there is an urgent need to evaluate actual HIV test coverage in different sub-populations and to implement non-stigmatizing HIV testing programs accessible to all in order to achieve the 90% diagnosis target set for the county.In their answer to our previous comment saying that inclusion of variables based on their statistical significance could be misleading, we don't understand why the authors said: "The issue of variables being excluded because of p>0.1 in bivariable analysis that could have had P<0.1 in adjusted analysis does not arise for our study since, as explained here above, all variables were included in the multivariable model.[Pages 8, (data analysis)]".Indeed, in the Data analysis section of the manuscript, it is rather clearly said that "Variables with p<0.10 from bivariable analyses were included in the multivariable model", which suggest that variables with p>0.1 in bivariable analyses were excluded from multivariable models and thus that not all variables were included in the multivariable model.

2.
Figures 4 and 5: the Y axis title should be "HIV positivity" and not "Proportion newly diagnosed".

3.
To support the following statement in the Discussion section: "First, we cannot be sure that encounters with a positive test result documented were new diagnoses of HIV infection, as stigma and social desirability bias may lead some clients to report their previous test result as negative even if it was positive."we suggest to reference a meta-analysis of misreporting of known HIV status (Soni et al. 2020 1 .Under-reporting of known HIV-positive status among people living with HIV: a systematic review and meta-analysis) that can be found here as a 4.
misleading conclusions, and have added this as a limitation of our study.
Figures 4 and 5: the Y axis title should be "HIV positivity" and not "Proportion newly diagnosed".
We have corrected this error.
To support the following statement in the Discussion section: "First, we cannot be sure that encounters with a positive test result documented were new diagnoses of HIV infection, as stigma and social desirability bias may lead some clients to report their previous test result as negative even if it was positive."we suggest to reference a meta-analysis of misreporting of known HIV status (Soni et al. 20201.Under-reporting of known HIV-positive status among people living with HIV: a systematic review and meta-analysis) that can be found here as a preprint.
We thank the for bringing this to our attention.We have added the suggested reference.
We have corrected the typo.Thanks.

Parinita Bhattacharjee
Centre for Global Public Health, University of Manitoba, Winnipeg, Canada The subject of the paper is very important.As the country is debating the need for retesting or frequent testing, this paper would be an important resource.
The paper is being reviewed and possibly published in 2020.The national data on the epidemic and testing used in the introduction is 3 years old.In the meantime the country did a population based survey in 2018 (KENPHIA) and the preliminary report is available now.I think the paper should use the updated data and cite current literature.
The definition of resting is different for GF and KP.Though it is explained clearly in the method section, in other sections of the paper only one definition is used.This needs to be consistent.
One of the limitations could be access to HIV testing services for populations in multiple sites.It could be possible that clients retested in other sites.How was that captured?
The other limitation could be that KPs did not identify themselves as KPs in the testing sites due to stigma and discrimination and hence some of the KPs may have been categorised as GF in a public facility.

Is the work clearly and accurately presented and does it cite the current literature? Partly
Is the study design appropriate and is the work technically sound?Yes

Are sufficient details of methods and analysis provided to allow replication by others? Yes
If applicable, is the statistical analysis and its interpretation appropriate?I cannot comment.A qualified statistician is required.

Are the conclusions drawn adequately supported by the results? Yes
Competing Interests: No competing interests were disclosed.

Reviewer Expertise: HIV prevention among key populations
I confirm that I have read this submission and believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard.
Overall, we found that the paper was innovative and well written.We congratulate the authors for leveraging the available wealth of testing data to gain a better understanding of longitudinal HIV testing trends.However, we have some concerns regarding limitations of these data and feel that some conclusions are not supported by the paper's results.
Please find our comments, suggestions, and questions below.

Methods
The most important limitation of the study is the use of self-reported information, especially with regards to new diagnoses.Several studies have found considerable under-reporting of knowledge of HIV status (see list at the end of review).We would recommend that the authors use extreme caution when interpreting these self-reported "new diagnoses".Indeed, if one examines HIV national testing program data from several countries, it is difficult to reconcile the national total annual number of HIV positive tests performed with the estimated size of the population living with HIV without allowing for considerable retesting of already diagnosed individuals.For example, in many countries the annual numbers of positive tests reported can represent up to 25 -30% of the whole estimated PLHIV population, which is inconsistent with survey data on the proportion of PLHIV ever tested.For more details on these issues, the authors can consult this preprint (Maheu-Giroux et al. 2019 1 ) (also in press at AIDS).
A total of 47,893 tests encounters were recorded.Yet, the authors exclude almost half of these to obtain an "homogenous sample".It is unclear to us why this is warranted and how this achieves a sample that is more representative of the "at-risk adult population in the area ".By definition, the authors' data constitute a convenience sample (i.e., health-care seekers) that is probably not representative of the underlying "at-risk population".Could the authors confirm that excluding 46% of observations does not change their main conclusion (qualitatively)?
○ "HIV positivity" is usually defined as the number of positive tests over the total number of HIV tests performed.Can the authors consider replacing what they call "HIV positivity" with " % of self-reported new diagnosis"?We understand this is a verbose change but it is also more accurate.

○
It is said that risk behaviors were collected from 2010 onward.However, figures showing trends among key populations show estimates from 2009.Were risk behaviors rather collected from 2009?
○ Why are the authors smoothing the time trends using LOWESS?Why is that warranted?Since their data do not constitute a sample -but rather contains all testing encounters in Kilifi county (akin to a census) -we are also wondering why the authors are using statistical test of significance to measure trends.If the proportion of self-reported first-time testers was 60% in the general population in 2006 and that it decreases to 20%ish in 2017 in Kilifi County, why perform statistical tests?The decrease was 40% point over that period: there are no random errors.Instead, it could be more appropriate for the authors to look at meaningful changes in testing metrics and disregard p-values to assess if changes are important.

○
In the multivariable regression models, the inclusion of variables based on their statistical significance could be misleading.For example, a variable with a p-value>0.1 could become important after adjustments for other potential confounders.Could the authors include variables (and confounders, if any) that were determined a priori to plausibly be associated with the outcomes?The authors should, at the very least, recognized the limitations of their model-building process.(That being said, we appreciate the use of log-binomial models over logistic regressions.)○ Ethical approval was not sought because the data was de-identified.De-identification of individual-level observations that contains potentially sensitive information is usually not a sufficient reason for not obtaining ethics approval.Can the authors clarify?Results Figure 5: the authors conclude that there was no change over time for most of the outcomes presented in this figure.Actually, we see that some of the presented curves tend to be quadratic.It could be there was a (quadratic) change over time for some of these outcomes, but that the authors could not detect it because they tested a linear trend.

○
It is written that, in 2017: "Overall FFT prevalence in KP was 28%" and "Overall late-retesting prevalence in KP was 83%".How could the sum of these two estimates be over 100%?Especially considering that we would expect that at least some of the tests in KP were ontime testing.

Discussion
In the discussion section, it is written: "Annual retesting coverage in GP was 72% in 2017, which signals a population-level increase from the 2012 estimate of 55%.On the other hand, less than 20% of KP in the current study were retesting quarterly."Both estimates (72% and 20%) were not provided in the results section.Please make sure that discussed results are also presented in the results section.
○ Some parts of the discussion are disconnected from the authors' analyses.For example, the authors' second paragraph of page 8 that discussed low educational attainment seems out of place.Why are the authors making statement such as "this suggests that HIV prevention programmes are insufficiently reaching those with low educational attainment"?The authors only looked at HIV testing program data and HIV prevention programs are way broader than this.In fact, the whole paragraph is not supported by data presented in their article.The same applies for the discussion of the likely impact of "underage sex", "early childhood marriage", and "sex tourism".Could the authors avoid such wide extrapolations of their findings?

○
The same applies for other sections of the discussion.For example, the authors conclude that PrEP users will likely need intense engagement and tailored support services in order to adhere to the recommended retesting frequency.This conclusion is surprising since the authors did not assess HIV testing among PrEP users.Please consider re-wording that section.

"HIV positivity" is usually defined as the number of positive tests over the total number of HIV tests performed. Can the authors consider replacing what they call "HIV positivity" with "% of self-reported new diagnosis"?
We understand this is a verbose change but it is also more accurate.
Our calculation of positivity was as per the reviewer's definition, i.e., "number of positive tests divided by total number of HIV tests performed".We have changed the definition to reflect this, viz: "proportion of test encounters where the result was positive".In the revised manuscript, we have avoided referring to "new diagnosis," in response to this reviewer's earlier comment.[Pages 1 (title), 5 (abstract), 6 (intro), 8 (data analysis), 18 (factors associated with HIV positivity), 21-23 (discussion).}

Why are the authors smoothing the time trends using LOWESS? Why is that warranted?
We smoothed the plots to improve visibility of the trends, especially for graphs where we have plots that cross each other.We have added in the supplementary materials the actual data points on which the plots are based.[Pages 11 (figure 2),12 (figure 3 and 4) and 13 (figure 5).Supplementary tables 2-9.]

Since their data do not constitute a sample -but rather contains all testing encounters in Kilifi county (akin to a census) -we are also wondering why the authors are using statistical test of significance to measure trends. If the proportion of self-reported firsttime testers was 60% in the general population in 2006 and that it decreases to 20%ish in 2017 in Kilifi County, why perform statistical tests? The decrease was 40% point over that period: there are no random errors. Instead, it could be more appropriate for the authors to look at meaningful changes in testing metrics and disregard p-values to assess if changes are important.
In response to this comment, we have omitted the test for trend.As our goal was to describe the change over the years and not simply the start and end points, we now focus, as suggested by the reviewer, on the slope (percentage-point change per year).[Pages 5 (abstract), 8 (data analysis), 11, 12, 13 (time trends in FTT, late retesting and positivity), 21, 22 (discussion)] In the multivariable regression models, the inclusion of variables based on their statistical significance could be misleading.For example, a variable with a p-value>0.1

could become important after adjustments for other potential confounders. Could the authors include variables (and confounders, if any) that were determined a priori to plausibly be associated with the outcomes? The authors should, at the very least, recognized the limitations of their model-building process. (That being said, we appreciate the use of log-binomial models over logistic regressions.)
We thank the reviewer for commending our use of log-binomial regression which gives estimates that are more intuitive.
We did not initially highlight our a priori variables since these variables met criteria of p<0.1 in bivariable analysis and so were included in the full multivariate model.However, sex and age were considered confounders a priori; this is now clarified in the manuscript.
The issue of variables being excluded because of p>0.1 in bivariable analysis that could have had P<0.1 in adjusted analysis does not arise for our study since, as explained here above, all variables were included in the multivariable model.[Pages 8, (data analysis)]

Ethical approval was not sought because the data was de-identified. De-identification of individual-level observations that contains potentially sensitive information is usually not a sufficient reason for not obtaining ethics approval. Can the authors clarify?
Ethical approval for the analysis has since been granted by the KEMRI Scientific and Ethical Review Unit (SERU).We have amended the ethical statement to reflect the same.[Page 8 (ethical statement)] Figure 5: the authors conclude that there was no change over time for most of the outcomes presented in this figure.Actually, we see that some of the presented curves tend to be quadratic.It could be there was a (quadratic) change over time for some of these outcomes, but that the authors could not detect it because they tested a linear trend.
We have omitted the statistical test for trend and now focus on the estimated slope (percentage-point change per year), also in response to another comment above.[Pages 11 (figure 2),12 (figure 3 and 4) and 13 (figure 5).] It is written that, in 2017: "Overall FFT prevalence in KP was 28%" and "Overall lateretesting prevalence in KP was 83%".How could the sum of these two estimates be over 100%?Especially considering that we would expect that at least some of the tests in KP were on-time testing.
FTT is based on data from clients who reported never testing before, while late retesting is from clients who ever tested before, so the numerators and denominators for these two metrics are different.We have edited the text for clarity, as these statistics refer to the proportion of all encounters that involved first-time testing or late-retesting within subgroups.[Page 8 (data analysis)] In the discussion section, it is written: "Annual retesting coverage in GP was 72% in 2017, which signals a population-level increase from the 2012 estimate of 55%.On the other hand, less than 20% of KP in the current study were retesting quarterly."Both estimates (72% and 20%) were not provided in the results section.Please make sure that discussed results are also presented in the results section.
We thank the reviewer for pointing out this problem.We edited the discussion to ensure that it only includes results that were presented in the results section.[Page 21 (discussion)] Some parts of the discussion are disconnected from the authors' analyses.For example, the authors' second paragraph of page 8 that discussed low educational attainment seems out of place.Why are the authors making statement such as "this suggests that HIV prevention programmes are insufficiently reaching those with low educational attainment"?The authors only looked at HIV testing program data and HIV prevention programs are way broader than this.In fact, the whole paragraph is not supported by data presented in their article.The same applies for the discussion of the likely impact of "underage sex", "early childhood marriage", and "sex tourism".Could the authors avoid such wide extrapolations of their findings?
We thank the reviewer for the comment and critical review of our discussion points.The discussion section has been amended to reflect potential strategies based on literature to tackle the socioeconomic inequalities such as poverty and education and their impact on HIV.[Pages 21-23 (discussion)]

It is written: "Testing coverage and HIV positivity varied by town […]" Testing coverage was not assessed in this study. The only metric the authors are able to calculate are the % of first-time tester -but this is not testing coverage per se. You could well have a region where only 10% of the population has ever been tested for HIV but where repeat testing is very high among this 10% of "testers". Even though the program data would show a very low proportion of first-time testers, testing coverage would still be <10%. Please refrain from making conclusions about testing coverage when examining program data only.
We thank the reviewer for pointing this out.In the revised manuscript, we have avoided the use of "coverage" and "prevalence" when referring to the testing encounters were studied.

Figure 1 .
Figure 1.Map of study area.

Figure 2 .
Figure 2. Time trends in the proportion of encounters involving first-time testers among clients attending voluntary counselling and testing centres in Kilifi County, Kenya.Plots drawn using locally weighted scatterplot smoothing (LOWESS).Slope is percentage-point change per year.Data points on which the plots are based are included in the supplemental materials (Extended data: Supplementary Table2and 3).MSM: Men who have Sex with Men; FSW: Female Sex Workers.

2
Figure 2. Time trends in the proportion of encounters involving first-time testers among clients attending voluntary counselling and testing centres in Kilifi County, Kenya.Plots drawn using locally weighted scatterplot smoothing (LOWESS).Slope is percentage-point change per year.Data points on which the plots are based are included in the supplemental materials (Extended data: Supplementary Table2and 3).MSM: Men who have Sex with Men; FSW: Female Sex Workers.
Figure 2. Time trends in the proportion of encounters involving first-time testers among clients attending voluntary counselling and testing centres in Kilifi County, Kenya.Plots drawn using locally weighted scatterplot smoothing (LOWESS).Slope is percentage-point change per year.Data points on which the plots are based are included in the supplemental materials (Extended data: Supplementary Table2and 3).MSM: Men who have Sex with Men; FSW: Female Sex Workers.

Figure 3 .
Figure 3.Time trends in the proportion of encounters involving late retesting among clients attending voluntary counselling and testing centers in Kilifi County, Kenya.Plots drawn using locally weighted scatterplot smoothing (LOWESS).Slope is percentage-point change per year.(Extended data: Supplementary Table4 and 5).MSM: men who have sex with men; FSW: female sex workers.

Figure 4 .
Figure 4. Time trends in HIV positivity at testing encounters among general population clients attending voluntary counselling and testing centres in Kilifi County, Kenya.Plots drawn using locally weighted scatterplot smoothing (LOWESS).(Extended data: Supplementary Table6 and 7).

Figure 5 .
Figure 5.Time trends in HIV positivity at testing encounters among key population clients attending voluntary counselling and testing centres in Kilifi County, Kenya.Plots drawn using locally weighted scatterplot smoothing (LOWESS).Slope is percentage-point change per year.(Extended data: Supplementary Table8 and 9).

Table 4 . Factors associated with HIV positivity at testing encounters among adult clients attending voluntary counselling and testing centres in Kilifi, Kenya, 2012-2017.
higher education, first-time testers and late re-testers, MSM and FSW, and clients with current STI symptoms.Of note, encounters involving FTT and late retesting were about twice as likely to result in a positive test compared to on-time retesting encounters.Encounters among clients with primary or no education were 1.6 times more likely to result in a positive HIV test, compared to those among clients with higher education, while those among clients with secondary education were 1.5 times more likely.
Late retesting was defined as previous test more than one year ago for GP or three months for key population.
While the proportion of encounters involving late retesting (i.e.previous test more than a year ago) was 28% for GP in 2017, 83% of encounters for KP in the current study involved late retesting (previous test more than 3 months ago).FTT encounters were more common among men (both GP men and MSM), among younger (18-24 years) or single persons, and among persons with lower educational attainment.Such persons may perceive themselves to be at higher risk.Among GP, late retesting encounters were less common among single persons, those with secondary education, those professing Muslim or no religious affiliation and those who had current STI symptoms.While the association with religious affiliation is less clear, the other associations may indicate increased awareness of risk for HIV.These findings are of interest given that encounters involving first-time and late re-testing were more likely to yield a positive test result, compared to on-time re-testers.Increased education and a mobilization strategy targeting sub-groups with these attributes could potentially contribute significantly to achieving the 90% UNAIDS HIV diagnosis target in Kilifi county and other similar settings.Because current STI symptoms were associated with a near doubling of HIV positivity at encounters, such approaches should incorporate integrated sexual reproductive health services that include screening, diagnosis and treatment for STI27.The low uptake of quarterly retesting (implied by the low proportion of on-time retesting encounters among KP in our study) and consequent continuing transmission among MSM and FSW may be due to stigmatizing attitudes of healthcare workers, discrimination, and concerns about confidentiality; factors that have been shown to decrease access to health 2 Gender and transactional sex were excluded from the model due to collinearity with the risk group variable.GP: General population; MSM: Men who have sex with men; FSW: Female sex workers; STI: Sexually transmitted infection.

sexual transmission of HIV from persons aware and unaware that they are infected with the virus in the USA
NASCOP: Kenya AIDS Indicator Survey 2012: Final Report.National AIDS and STI Control Programme (NASCOP), Kenya, 2014. .AIDS.2006; 20(10): 1447-50.PubMed Abstract | Publisher Full Text 12. Bezemer D, de Wolf F, Boerlijst MC, et al.: A resurgent HIV-1 epidemic among men who have sex with men in the era of potent antiretroviral therapy.AIDS.2008; 22(9): 1071-7.PubMed Abstract | Publisher Full Text 13.Baggaley R, Dalal S, Johnson C, et al.: Beyond the 90-90-90: refocusing HIV prevention as part of the global HIV response.J Int AIDS Soc.2016; 19(1): 21348.PubMed Abstract | Publisher Full Text | Free Full Text

Testing among Key Populations, Adolescent Girls and Men in Eastern and Southern Africa: A Review of Research, Policy and Programming
. Methods.2016; 2: 1. Reference Source 29.Chirro O, Wahome E, Kimani M, et al.: Test

and Treat Care Cascade in General and Key Populations. HIV
Research for Prevention.Madrid, Spain.2018.30.Beyrer C, Sullivan PS, Sanchez J, et al.: A

call to action for comprehensive HIV services for men who have sex with men. Lancet. 2012; 380(9839): 424-38. PubMed Abstract | Publisher Full Text | Free Full Text 31
. Campbell CK, Lippman SA, Moss N, et al.: Strategies

Self-Testing: An operational manual for delivery of HIV Self-Testing services in Kenya
. 2017; [cited 2020 26 October].

Reporting of Known HIV-Positive Status Among People Living with HIV: A Systematic Review and Meta- Analysis
. SSRN Electronic Journal.2020.Publisher Full Text 45.Ellison AM, Gotelli NJ, Inouye BD, et al.: P values, hypothesis testing, and model selection: it's déjà vu all over again.Ecology.2014; 95(3): 609-10.PubMed Abstract | Publisher Full Text 46.Mugo P: Trends

tables for: Trends and predictors of new HIV diagnosis and time since last test at testing encounters among adults in Kilifi, Kenya, 2006-2017. Harvard
Dataverse, V1. 2019.http:/

Is the work clearly and accurately presented and does it cite the current literature? Partly Is the study design appropriate and is the work technically sound? A total of 47,893 tests encounters were recorded. Yet, the authors exclude almost half of these to obtain an "homogenous sample". It is unclear to us why this is warranted and how this achieves a sample that is more representative of the "at-risk adult population in the area". By definition, the authors' data constitute a convenience sample (i.e., health-care seekers) that is probably not representative of the underlying "at-risk population". Could the authors confirm that excluding 46% of observations does not change their main conclusion (qualitatively)?
It is written: "Testing coverage and HIV positivity varied by town […]" Testing coverage was not assessed in this study.The only metric the authors are able to calculate are the % of firsttime tester -but this is not testing coverage per se.You could well have a region where only 10% of the population has ever been tested for HIV but where repeat testing is very high among this 10% of "testers".Even though the program data would show a very low proportion of first-time testers, testing coverage would still be <10%.Please refrain from making conclusions about testing coverage when examining program data only.Among limitations of the study, please discuss also potential selection bias in assessment of ○ trends due to the changes in covered populations in the different VCT centres over time.We also suggest discussing potential information bias from self-report "new diagnosis" and, importantly, key population status.It is highly likely that a potential high proportion of MSM do not disclose same-sex activity to their health-care providers.The same applies for FSW.Minor commentsIntroduction: The 2019 AIDS report published by UNAIDS has the updated figure for knowledge of HIV status in Kenya.Consider updating it.The conclusions regarding on-time testing would benefit from mentioning potential limitations associated with these self-reports.In particular, "telescoping bias" is likely to occur if respondents inadvertently recall testing that occurred beyond the last 3 (KP) or 12 (GP) months.This would not necessarily affect the conclusion regarding trends but would affect the absolute proportion testing on time.Our sample selection was guided by our goal to assess trends in "adult walk-in VCT clients", i.e., clients seeking testing out of their own initiative and not those testing as a result of outreach activities.Clients mobilized during outreach activities may be more open to social pressure to test or retest, and differences in the frequency and intensity of outreach activities over the years would make time trends difficult to assess.In response to this comment, we have added supplementary table 1 with data on the numbers of encounters and HIV positivity of the categories excluded.[Pages 7, 8 (Data collection procedures and sample selection)] ○ ○ ○ ○ Figure 4.There is a "T" missing at the beginning of the title.○ References 1. Maheu-Giroux M, Marsh K, Doyle C, Godin A, et al.: National HIV testing and diagnosis coverage in sub-Saharan Africa: a new modeling tool for estimating the "first 90" from program and survey data.bioRxiv.2019.Publisher Full Text 2. Rohr J, Xavier Gómez-Olivé F, Rosenberg M, Manne-Goehler J, et al.: Performance of selfreported HIV status in determining true HIV status among older adults in rural South Africa: a validation study.Journal of the International AIDS Society.2017; 20 (1).Publisher Full Text 3. Fuente-Soro L, Lopez-Varela E, Augusto O, Sacoor C, et al.: Monitoring progress towards the first UNAIDS target: understanding the impact of people living with HIV who re-test during HIV-testing campaigns in rural Mozambique.J Int AIDS Soc.