The study site was Makueni County, one of the 47 counties in Kenya located on the South-eastern part of the country. In 2020 the prevalence of HIV in Makueni County was 3.5%, ³⁵ with the county displaying a mixed epidemic pattern where HIV prevalence varied between 3%-10% among the general population and 23-30% among key populations. ¹ Data indicate that challenges faced with HIV epidemic control in Makueni County include moderate ART coverage with 21% unmet need, moderate testing inefficiencies, high mother to child transmission and low VL suppression among children. ³⁶ HIV services in Makueni County include HIV testing and treatment, prevention, and care services. These services are available for high risk populations, PLHIV and their partners, families and caregivers, according to the Kenya national guidelines. ³⁷ An analytical cross-sectional study design was employed to estimate the prevalence and identify the predictors of VF among PLHIV in Makueni County between October 2018 and June 2019, the period for which the data were available.

Data from this study were abstracted from the National Viral Load/early infant diagnosis (EID) monitoring system. This system is a repository of HIV VL and EID data and is managed by the National AIDS and STI Control Programme (NASCOP). Briefly, NASCOP spearheads the Ministry of Health’s interventions in tackling HIV/AIDS through policy formulation, coordination of procurement and supply chain management, training and monitoring and evaluation of the HIV response. ³⁸ The national VL/EID system is an electronic data management system for monitoring patients. It contains patient information on medication and ART history and demographic data. Health facilities feed data into the system which is available publicly via an interactive computer interface tool that graphically represents programme indicators. ^{39 , 40}

Data collected from all PLHIV across all ages, resident in Makueni County and who had VL tests done between October 2018 and June 2019 in point of care facilities under the national VL/EID monitoring system were included in the study. Additionally, those classified as receiving ART by either having detectable blood levels of selected ART or by reporting current ART use were also included. On the contrary, all PLHIV in Makueni County on ART for six months or less and those with missing information on key variables and/or invalid VL outcomes were excluded. An initial screening of the database showed that 23,067 entries were made between 2018 and 2019 from health facilities in Makueni County. A total of 16,340 eligible participants met the inclusion criteria and were selected for this study.

A simple random sampling design was employed with a sampling frame that comprised all PLHIV with VL test results from point of care centers in Makueni County and who were enrolled in the national VL/EID monitoring framework between October 2018 and June 2019.

For this study, cases were defined as those with VF identified as detectable viral load ≥1,000 copies/ml after a minimum of six months on ART, as per the Kenya national treatment guidelines. ³⁷

Ethical approval for this study was granted by the Kenyatta National Hospital and University of Nairobi joint Ethics and Research Committee (KNH-ERC/A/508) on 14 ^th December 2022. To safeguard participant confidentiality, any identifying information contained in the data was removed during the abstraction process.

All analyses were performed using R Statistical Software (v4.3.0; R Core Team 2023). Table 1 shows the predictor variables. Continuous variables were summarized using medians and ranges. For qualitative variables, frequencies and proportions were computed. The prevalence of VF and its associated 95% exact binomial confidence interval were estimated. This was followed by univariable mixed-effects logistic regression analysis to assess the effect of each predictor on VF. Code for the analysis is available as extended data. ⁴¹ The inclusion of age as continuous predictor in the univariable models yielded insignificant results, age was grouped into two categories <15 years and ≥15years. The variables “sub-county” and “facility” were included as random effects to account for clustering. At this stage, a liberal P<0.20, was used to evaluate the significance of each of the predictors. Significant variables from the univariable analysis were included in a multivariable model where a backward stepwise approach was used to eliminate variables at P≥0.05. Notably, to minimise confounding, exclusion of these variables from the model was only considered if their removal resulted in a less than 30% change in the effects of the remaining variables. ⁴² Two-way interactions were fitted between the remaining variables of the final model and assessed for significance.

All PLHIV in Makueni County and enrolled in the national VL/EID database were assessed for eligibility. An initial screening of the database showed that 23,067 entries were made between 2018 and 2019 from health facilities in Makueni County. A total of 16,340 eligible participants met the inclusion criteria and were selected for this study. Among those included, 15,014 were adolescents and adults aged ≥15 years, and 1,326 were children aged 0-14 years ( Figure 1).

Table 2 shows the descriptive statistics of the study. During the study period, the median age of the participants was 43 years (range: 0-93 years). Of the participants, 69.6% (n=11365) were female with 83.3% enrolled on a TDF-based ART regimen (n=13613) The estimated period prevalence of VF between October 2018 and June 2019 was 13.2% (95% CI: 12.7%-13.8%).

From the results of the univariable analysis ( Table 3), sex, age, partner, sample type, ART-regimen, and justification were found to be significantly associated with VF at a 20% significance level and were subsequently offered to the multivariable model.

In the multivariable analysis, only age, justification and ART-regimen were found to be significant predictors of VF at the 5% significance level ( Table 4).

Being 15 years or older reduced the odds of VF by a factor of 0.53 (aOR=0.53; 95% CI: 0.44-0.645) controlling for the ART regimen and the justification for VL testing. Compared to patients taking a TDF-based ART-regimens, those taking ABC-based, AZT-based, or other regimens had about two times higher odds of VF (aOR=1.61; 95% CI: 1.34-1.94), (aOR=1.75; 95% CI: 1.52-2.01), and (aOR=1.55; 95% CI: 0.99-2.44) respectively, controlling for age and the justification for VL testing. Compared to patients whose blood samples were taken for baseline viral load measurement, those who gave samples for other reasons had lower odds of VF: breastfeeding mothers (aOR=0.1; 95% CI: 0.01-0.97); clinical failure (aOR=0.08; 95% CI: 0.01-0.44); confirmation of VF (aOR=0.2; 95% CI: 0.07-0.62); no VL data (aOR=0.06; 95% CI: 0.01-0.31); routine VL (aOR=0.04; 95% CI: 0.01-0.12); drug substitution (aOR=0.03; 95% CI: 0.01-0.08), controlling for their age and ART regimen.

In this study the VF rate was estimated at 13.2% among HIV patients in Makueni County. This is an improvement in the VF rate compared to the national reports in 2015 when more than 60% of adults receiving ART in Makueni County had VF. ⁴³ This prevalence of VF is comparable with the findings from a national cross-sectional survey in Uganda that found a VF rate of 11%. ¹⁰ A similar burden has been observed elsewhere in Sub-Saharan Africa. ^{5 , 11 , 44 – 47} Nonetheless, the VF frequency was higher in Zimbabwe (30.6%) ⁴⁸ and Togo (51.6%), ⁴⁹ likely attributable to poor ART adherence.

This study revealed that younger age was associated with higher odds of VF. This is similar to the findings of a national-based household survey in Kenya where decreasing age was associated with higher risk of VF. ² A possible explanation for this finding could be that the youth face a myriad of challenges encompassing behavioral and psychosocial such as peer-related stigma, anxiety, lack of disclosure, sexual, reproductive and gender health concerns that may undermine adherence. ^{47 , 50 – 53} Similarly, a study in South Africa showed that adolescents aged <15 years had higher risk of VF compared to older patients. ⁵⁴ High pill burden among adolescents could also explain the lack of adherence leading to VF. ⁵¹ These findings have also been replicated elsewhere, with younger patients demonstrating poor adherence or higher levels of drug resistance mutations. ^{11 , 15 , 50 – 52 , 55}

In this study, patients taking Tenofovir (TDF) based ART regimens had lower odds of VF compared to those on zidovudine (AZT) or abacavir (ABC) based or other regimens. This corresponds to the research findings of a study in Uganda which showed that patients initiated on AZT-based regimens as compared with TDF-based ones were more likely to have VF. ⁵⁶ TDF-based regimens have been shown to be better tolerated with fewer side effects and hence better adherence. ⁵⁷ Nonetheless, one study reported that patients experiencing VF on a TDF-based regimen had higher rates of the NRTI-resistance mutation – K65R. ⁵² Some studies have reported that ART-experienced patients have higher odds of VF compared to ART-naïve patients. ^{58 , 59} Data suggests that drug resistance testing should be done for all patients with HIV RNA levels >1000 copies/ml. ¹⁶ The absence of drug resistance in these patients indicates poor adherence. Drug-related reasons such as toxicity, frequency of dosing and pill burden should be investigated and strategies of optimizing ART adherence discussed.

Patients tested for VF because of suspected clinical failure, repeat testers after suspected VF, breastfeeding mothers, those with no VL data, those undergoing routine viral loads and those with drug substitutions had overall reduced odds of VF compared to those tested at baseline. This is corroborated by the finding that patients on routine monitoring registered the lowest levels of VF ¹⁰ due to enhanced adherence counselling and support. A Kenya nationwide analysis showed that the odds of VF were reduced as the frequency of VL monitoring increased. ³⁹ Contradictory findings were reported in Uganda where adolescents who had detectable VL at baseline testing were more likely to have VF upon a repeat viral load test regardless of their adherence level and change in ART regimen. ⁵³ Moreover, repeat testers who had active tuberculosis co-infection had higher odds of VF. ¹⁰

Sex was not associated with VF in this study ( Table 4). This agrees with the findings in other resource-limited settings which found no association between sex and VF. ^{55 , 60 , 61} Other studies have however showed that being male was associated with higher risk of VF due to reluctance to access healthcare and poor adherence compared to women. ^{25 , 46 , 51} In particular, a study in rural Cameroon found that men had higher odds of VF, independent of their adherence behaviours, ascribable to biological differences where men took longer to regenerate their CD4 count while on ART compared to women. ⁶² Contrastingly, in Tanzania and UK, females were reported to be more likely to experience VF than males. ^{15 , 27}

Considering other study variables, the organizations partnering with the HIV care facilities did not emerge as significant predictors for VF. This was contrary to the findings of others studies where different levels of quality of care that the partners offer to the patients, program level differences or factors such as variations in policy and their implementation were attributed to higher odds of VF. ^{11 , 52} Moreover, task shifting - comprising of the redistribution of tasks from highly qualified to less specialized healthcare workers where appropriate, to efficiently utilize the available human resources for health ^{63 , 64} - was shown to be most successful in areas where the community health workers provided care under the supervision of experienced ART providers. ^{63 , 65}

The sample type collected was not independently associated with VF in this study. This is supported by findings from other studies that showed that using either DBS or plasma samples to quantify HIV viral loads resulted in findings that were reliable and comparable. ^{59 , 66 , 67} In Malawi, one study reported that using plasma samples increased the precision of VL monitoring but was plagued by logistical and financial barriers thus unappealing in resource-limited settings. ⁵⁹ On the contrary, DBS samples were shown to be cost effective and easy to transport at ambient temperatures, but associated with reduced sensitivity with viral loads <5000 copies/ml. ^{56 , 68}

This study is not without limitations. The use of routine data that were not collected for research purposes handicapped the research since several exposure variables were not captured. The study did not analyze data on variables like income, education, or distance to the health facility that would have allowed for the control of more concealed sociodemographic confounders, because these are not routinely recorded in public health facilities. Additionally, some important factors leading to VF such as ART adherence, should be considered as potential factors in the study, but such information was not available. Failure to control for these factors may have led to residual confounding that could have biased the estimated odds ratios towards null. Furthermore, the study was based on outcome data collected at a single time point. Therefore, we might have overestimated the virologic failure rate due to individual temporal variability in biological markers. Nevertheless, the strength of this study rests on its sizeable statistical power afforded by the large sample size. Further investigations using cohort studies may be necessary to validate the study’s findings.