Characterization of human immunodeficiency virus drug-resistance mutations among individuals with low-level-viremia in low-income and middle-income countries: A meta-analysis

Abstract

BACKGROUND

Low-level viremia (LLV) defined as [human immunodeficiency virus (HIV)-RNA 51-999 copies/mL] has been associated with an increased risk of drug resistance and treatment failure. Advances in next-generation sequencing enabled the detection of drug resistance mutations (DRM) among people with LLV. However, evidences remain limited in low-income and middle-income countries (LMIC) where surveillance is most needed to inform global epidemic control strategies.

AIM

To determine the prevalence of HIV DRM among people living with HIV who have LLV in low- and middle-income countries.

METHODS

PubMed, Cochrane Library, and EMBASE were systematically searched for articles published between January 2015 and May 2025. Studies were included if they reported DRM among adolescents and/or adults with LLV in LMIC. Mutations were interpreted using Stanford University HIV Drug Resistance Database. Prevalence of DRM was computed as the proportion of resistance mutations among successfully sequenced samples. Pooled estimates of resistance mutation and 95%CI were calculated using random-effects models with stratified analyses comparing mutations by geographic location (Africa vs Asia).

RESULTS

Twenty studies including 7613 people with LLV were included. Of these, 5252 (73.9%) had their samples successfully sequenced. Eleven studies were from Africa and nine from Asia. Overall, the pooled prevalence of DRM was 50.4% (95%CI: 38.3-62.5) with a significantly higher prevalence observed among Africa studies compared to Asia’s (58.0% vs 40.7%; P < 0.0001). The prevalence of mutations associated with nucleoside reverse transcriptase inhibitor and non-nucleoside reverse transcriptase inhibitor were 44.6% (95%CI: 34.8-54.4), and 50.9% (95%CI: 41.4-60.4) respectively and were significantly higher in Africa than in Asia. Protease inhibitor associated mutations were also common in Africa than in Asia (7.3% vs 4.1%; P < 0.001), though the overall prevalence remains low.

CONCLUSION

Most individuals with LLV have resistance mutations and remain on a failed regimen over an extended period. Because resistance testing is not routinely performed in LMIC, lowering the viral failure threshold may improve timely switching to effective regimens, preserve treatment options, and reduce resistance accumulation in high HIV burden regions.

Key Words: Low level viremia; Antiretroviral therapy; Drug resistance mutations; Human immunodeficiency virus; Low-income and middle-income countries

Core Tip: Next-generation sequencing enabled successful sequencing and determined drug resistance mutations (DRM) among people with low-level viremia (LLV). Data on the DRM among people with LLV in low-income and middle-income countries (LMIC) is limited and may help achieve human immunodeficiency virus epidemic control. This systematic review and meta-analysis analyzed 7613 people living with human immunodeficiency virus who had LLV from 20 studies conducted in LMIC between 2015 and 2025. Overall, the pooled prevalence of DRM was 50.4%. Pooled prevalence of nucleoside reverse transcriptase inhibitor, non-nucleoside reverse transcriptase inhibitor and protease inhibitor-associated mutations were 44.6%, 50.9% and 5.1% respectively. In LMIC, most patients with LLV have resistance mutations and remain on a failed regimen over an extended period. Because resistance testing is not routinely performed in LMIC, lowering the viral failure threshold may hasten patients switch to effective drugs.

INTRODUCTION

Over the past three decades, the global response to human immunodeficiency virus (HIV) has yielded remarkable progress, with expanded access to antiretroviral therapy (ART), widespread adoption of routine viral load (VL) monitoring, and substantial reduction in acquired immunodeficiency syndrome-related mortality. As of 2024, nearly 77% of the 40.8 million people living with HIV (PLWH) were receiving ART, and 73% had achieved viral suppression, a major global health milestone[1]. Despite this gain of expanded ART uptake and VL suppression, progress remains uneven across regions, particularly in low-income and middle-income countries (LMIC), which account for over two-thirds of all PLWH and face persistent threats from geographical instability, funding disruption, and weak health system[1,2].

Within this context, low-level viremia (LLV) has emerged as a clinically significant challenge. Although the World Health Organization (WHO) does not operationalize LLV as a formal treatment category in its current guidelines, it recognizes LLV as detectable plasma VL between 50 copies/mL and 999 copies/mL in individuals on ART[2]. Several studies and national programs similarly define LLV within this range with varying sub-categories[3,4]. In contrast, guidelines from high-income countries often define virologic failure (VF) at lower VL thresholds ranging from > 50 copies/mL to > 200 copies/mL[5-7]. This lack of consensus complicates cross-study comparisons and global surveillance. LLV, once considered biologically benign, has been linked with higher risk of subsequent VF, immunologic deterioration, onward HIV transmission and increased mortality, particularly when persistent[8,9]. Persistent LLV unlike transient “blips”, reflects ongoing viral replication under drug pressure and may signal suboptimal treatment efficacy[10]. Crucially, persistent LLV creates a selective environment for the emergence of HIV drug resistance mutations (DRM), particularly in individuals with sub-optimal adherence or exposure to partially suppressive ART regimen[11,12]. These DRMs often remain undetected in LMICs, where genotypic resistance testing is typically reserved for individuals with VL < 1000 copies/mL due to cost, infrastructure limitations, and policy constraints[2].

This diagnostic gap has important implications for treatment outcomes and population level transmission. Emerging and multi drug resistant strains not only drive VF, but also increase the risk of disease progression and mortality in resource limited settings where resistance testing and VL monitoring is lacking[13,14]. The predominance of non-B HIV subtypes in these regions adds further complexity, as such subtypes may follow distinct mutational pathway and exhibits varying susceptibility to antiretroviral agent, complicating resistance interpretation[15-17].

Despite emerging observational data, no systematic review has synthesized the burden and mutational patterns of HIV DRM during LLV in LMICs. This gap hinders efforts to refine VL thresholds for resistance testing and guide ART policy. Advanced technologies like next generation sequencing could improve detection at lower viremia levels but remain inaccessible in most LMICs due to costs and infrastructure gaps. Addressing this knowledge gap is essential to improve clinical decision-making and sustain global HIV control in high-burden, resource limited settings.

This meta-analysis aims to characterize the prevalence, mutational profiles, and clinical implications of HIV DRMs among PLWH experiencing LLV in LMIC. By synthesizing available evidence from real-world, resource-constrained contexts. This meta-analysis evaluated the appropriateness of current VL monitoring thresholds and explore the potential utility of resistance testing during LLV in informing timely decision-making and global policy recommendations.

MATERIALS AND METHODS

Registration

This systematic review protocol was registered in the International Prospective Registry of Systematic Review with registration number CRD420251081973.

Ethical approval

Ethical approval was not required as the review utilized only published data.

Search strategy

PubMed, Cochrane CENTRAL, EMBASE and clinicaltrials.gov were systematically searched for articles published between January 2015 and May 2025. Search terms were used to capture information on LLV, DRM, among PLWH in LMIC. The searches were restricted to papers published in English. Search results were uploaded to Covidence Systematic Review Software (Melbourne, Australia), where deduplication and screening was performed.

Eligibility criteria

Eligibility criteria were guided by the guidelines of systematic reviews and meta-analysis with prevalence approach (CoCoPop)[18]. The CoCoPoP acronym stands for condition/problem, context and study population. In this review, under condition, we included studies that reported prevalence of DRM among PLWH who had low level viremia; context, we explored components of different studies that could explain variation in the reported prevalence of DRM among PLWH who had low level viremia such as study designs, number of study sites (single vs multiple sites), country of origin; population, we included PLWH who had low level viremia.

Inclusion criteria

Studies that involved PLWH who had LLV, underwent DRM testing in LMIC and written in English were included in the study.

Exclusion criteria

Studies that reported the percentage of individuals who underwent DRM testing without actual numerators and denominators used to compute those percentages. Studies that were not written in English. Figure 1 describes the literature search process of all included studies as shown below. The reporting of this systematic review and meta-analysis was done according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines[19].

Figure 1  Preferred Reporting Items for Systematic Reviews and Meta-analyses flow diagram of study selection.

Study selection and data extraction

After the duplication in Covidence, two pairs of reviewers (El-Imam IA, Mpimo BK, Ally HM, and Dababneh SW) independently screened titles/abstracts and assessed full texts for eligibility. Disagreements were resolved through discussion or adjudicated by a third reviewer pair (Fussi HF and Ramadhani HO). Using a pre-specified excel spreadsheet template, the same team independently abstracted the following data from the final list of manuscripts; authors, year of publication, the country in which the study was conducted, the year(s) in which data was collected, study design, study sites (single vs multi sites), number of people with low level viremia, number of people who were successfully sequenced, number of people who had DRM to nucleoside reverse transcriptase inhibitor (NRTI), non-NRTI (NNRTI), protease inhibitor (PI) and integrase strand transfer inhibitor (INSTI). Thereafter, extracted data was compared from two pairs of data abstractor, any discrepancies between the two pairs of reviewers were handled by consensus; and the third pair of reviewers (Fussi HF and Ramadhani HO) was consulted when necessary. Strategies in the Cochrane Handbook for Systematic Reviews of Interventions for data management were followed[20].

Quality assessment

Study quality was assessed using the Joanna Briggs Institute checklists for cross-sectional and cohort studies[21]. Each study was rated on nine questions with responses coded as: (1) Yes; (2) No; (3) Not clear; and (4) Not applicable. A score of 1 was assigned to a “yes” response and 0 to a “no”, or “not clear” response. The total score was summed up and categorized into three groups, with 0-3, 4-6 and 7-9 scores indicating low, medium and high quality.

Definition of variables

Outcome variables: The main outcome of interest was the prevalence of DRM defined as percentage of PLWH with LLV who had confirmed DRMs among those who were successfully sequenced for resistance testing. The secondary outcomes were the prevalence of DRM to NRTI, NNRTI, PI and INSTI.

Exposure variables: This was a prevalent study and therefore no main exposure of interest. Although there was no main exposure of interest, the prevalence of DRM was compared between studies conducted from Africa vs Asia and those conducted from single vs multiple sites to understand disparities of DRM by continent and site.

Statistical analysis

Random effects models were used to compute pooled prevalence of DRM and Freeman-Tukey double arcsine transformation was used to stabilize variance of proportions prior to the computation of pooled estimates. Stabilization of variance adjusts for potential biases and safeguard the robustness of our meta-analysis. In addition, stabilization of variance allows for the inclusion of all studies, including those with proportions at the boundaries, facilitating the computation of confidence intervals that remain within permissible bounds. Subgroup analysis on the pooled estimates of DRM was performed to compare studies done in Africa vs Asia and those conducted from single vs multiple sites. Two sample tests for proportions were used for comparison in these subgroup analyses. We evaluated heterogeneity across studies using the I² statistic and Cochran’s Q test. The I² statistics explain the variance attributable to study heterogeneity with scores of 75%, 50% and 25% indicating high, moderate and low heterogeneity, respectively[22]. Publication bias was assessed through funnel plots and the Egger regression asymmetry test (P < 0.05 implies publication bias and heterogeneity). Furthermore, proactive comparison of the study protocol and final results was done to ensure all that was deemed to be reported from the protocol has been reported in the manuscripts. An influential analysis using leave-one-out method was used to identify potential sources of heterogeneity[23]. Studies with missing information, such as those that reported proportions of outcomes without actual numerators and/or denominators, were excluded from the analysis. All statistical tests were performed using STATA version 18 (Stata Corporation, College Station, TX, United States).

RESULTS

Study selection

A total of 307 articles were identified through databases searches (EMBASE: 198, MEDLINE: 96, and CENTRAL: 13). Of these, 81 articles were duplicates and deleted. The remaining 226 articles were eligible for title and abstract screening. Of the 21 articles eligible for full text review, 20 articles met inclusion criteria and were finally included in our analysis (Figure 1).

Study quality assessment

All included studies were assessed as high quality using the Joanna Briggs Institute critical appraisal tool (Table 1)[3,12,21-38]. Studies with scores less than 9 typically had smaller sample sizes but met all other methodological standards.

Table 1 Context descriptions of included studies.

Ref.	Study design	Continent	Number of sites	Low level viremia definition	Quality assessment
Bangalee et al[24], 2021	-	Africa	Single	51-999	8
Bareng et al[3], 2022	Randomized clinical trial	Africa	Single	51-999	9
Brown et al[26], 2021	Randomized clinical trial	Africa	Single	100-999	7
Djiyou et al[20], 2023	Cross sectional	Africa	Single	200-999	7
Kao et al[21], 2021	Cross sectional	Asia	Single	20-999	7
Liu et al[12], 2024	-	Asia	Single	50-999	8
Kantor et al[27], 2018	Cross sectional	Africa	Single	40-999	7
Mundo et al[28], 2024	Cross sectional	Africa	Multiple	< 1000	8
Yuan et al[23], 2022	Cross sectional	Asia	Single	< 1000	9
Lan et al[29], 2023	Cross sectional	Asia	Single	200-999	9
Li et al[30], 2022	Cohort	Asia	Single	50-999	7
Rupérez et al[31], 2015	Cross sectional	Africa	Single	150-999	7
Liu et al[32], 2018	Cross sectional	Asia	Single	50-999	8
Bareng et al[25], 2022	Randomized control trial	Africa	Single	400-999	7
Chenwi et al[33], 2024	Cross sectional	Africa	Single	< 1000	8
Choga et al[34], 2025	Cross sectional	Asia	Multiple	200-999	7
Shu et al[35], 2025	Cross sectional	Asia	Single	50-999	9
Cao et al[36], 2023	Cross sectional	Asia	Single	50-999	7
Liu et al[37], 2022	Cross sectional	Asia	Multiple	50-999	9
Labhardt et al[38], 2015	Cohort	Africa	Single	80-999	7

Pooled prevalence estimates

Sequencing success rate: A total of 20 studies reported on sequencing outcomes among PLWH experiencing LLV[3,12,20,21,23-38]. The random effects model estimated that 73.9% (95%CI: 66.4-81.4) of them were successfully sequenced (Figure 2)[3,12,20,21,23-38]. This indicates that, on average, nearly three-quarters of people with LLV had their VL samples successfully sequenced. Heterogeneity was observed (I² = 99.4%), reflecting substantial variability in sequencing success rates across studies.

Figure 2  Sequencing success rate.

Pooled prevalence of DRM: The pooled prevalence of DRM among people with LLV was estimated at 50.4% (95%CI: 38.3-62.5) using a random-effects model presented in Figure 3A[3,12,20,21,23-35,37]. Subgroup analysis by site showed a pooled prevalence of 41.7% (95%CI: 11.0-72.3) in multi-site studies and 52.8% (95%CI: 39.3-66.3) in single-site studies (Table 2). When stratified by continent, the pooled prevalence was 58.0% (95%CI: 30.8-85.2) in studies conducted in Africa and 40.7% (95%CI: 26.0-55.5) in studies conducted in Asia (Table 2). Heterogeneity was considerable across studies (I² = 98.9%) and remained considerable even after subgroup analyses by site and continent. Leave-one-out sensitivity analysis showed that no single study substantially influenced the overall pooled estimate, as all P values remained statistically significant and pooled effect sizes were consistent across iterations (Supplementary Figure 1)[3,12,20,21,23-35,37].

Figure 3 Pooled prevalence. A: Pooled prevalence of drug resistance mutations (DRM); B: Pooled prevalence of DRM due to nucleoside reverse transcriptase inhibitors; C: Pooled prevalence of DRM due to non-nucleoside reverse transcriptase inhibitors; D: Pooled prevalence of DRM due to protease inhibitor drugs; E: Pooled prevalence of DRM due to integrase strand transfer inhibitor drugs.

Table 2 Prevalence of drug resistance mutations: Subgroup analysis, % (95%CI).

Outcome	Continent			Number of sites
	Africa	Asia	P value	Single	Multiple	P value
Overall	58.0 (30.8-85.2)	40.7 (26.0-55.5)	< 0.0001	52.8 (39.3-66.3)	41.7 (11.0-72.3)	< 0.0001
Nucleoside reverse transcriptase inhibitor mutation	57.9 (53.9-61.9)	34.7 (33.3-36.1)	< 0.0001	47.3 (42.2-52.4)	16.7 (7.4-30.4)	< 0.0001
Non-nucleoside reverse transcriptase inhibitor mutation	61.6 (44.5-78.8)	36.2 (26.4-46.1)	< 0.0001	48.7 (39.3-58.1)	78.9 (63.7-88.9)	< 0.0001
Protease inhibitors mutation	7.3 (3.4-11.2)	4.1 (3.4-4.9)	< 0.0001	4.9 (3.5-6.3)	15.8 (7.4-30.4)	< 0.0001

Pooled prevalence of DRM due to NRTIs: The random effects model estimated the pooled prevalence of DRM due to NRTIs at 44.6% (95%CI: 34.8-54.4) as shown in Figure 3B[3,12,20,21,23-30,33,35]. Subgroup analysis by study setting revealed a significantly higher prevalence in single-site studies (47.3%, 95%CI: 42.2-52.4) compared to those conducted across multiple sites (16.7%, 95%CI: 7.4-30.4; P < 0.0001; Table 2). Similarly, subgrouping by geographical region showed a higher prevalence in studies from Africa (57.9%, 95%CI: 53.9-61.9) than in Asia (34.7%, 95%CI: 33.3-36.1; P < 0.0001) as presented in (Table 2). Heterogeneity was initially substantial (I² = 96.5%) but decreased to 50.5% following subgroup analysis, indicating reduced variability between studies. Leave-one-out sensitivity analysis confirmed the stability of the findings, as no single study disproportionately influenced the overall estimate. The overall prevalence estimate was stable irrespective of sequential exclusion of each single study (Supplementary Figure 2)[3,12,20,21,23-30,33,35].

Pooled prevalence of DRM due to NNRTIs:Figure 3C presents the random effects model which estimated pooled prevalence of DRM due to NNRTIs at 50.9% (95%CI: 41.4-60.4)[3,12,20,21,23-30,33,35]. Subgroup analysis by study setting revealed a significantly lower prevalence in single-site studies (48.7%, 95%CI: 39.3-58.1) compared to multiple-site studies (78.9%, 95%CI: 63.7-88.9; P < 0.0001), as shown in (Table 2). Similarly, regional subgrouping showed a higher pooled prevalence in studies conducted in Africa (61.6%, 95%CI: 44.5-78.8) than in Asia (36.2%, 95%CI: 26.4-46.1; P < 0.0001; Table 2). Heterogeneity was initially substantial (I² = 96.0%) but decreased in the stratified analyses (I² = 0.0% for multiple-site and I² = 95.9% for single-site studies; I² = 94.8% for both regions). Leave-one-out sensitivity analysis confirmed the stability of the findings, as no single study substantially influenced the overall pooled estimate; all P values remained statistically significant and the effect sizes consistent across iterations (Supplementary Figure 3)[3,12,20,21,23-30,33,35].

Pooled prevalence of DRM due to PI drugs: As illustrated in Figure 3D, the pooled prevalence of DRM associated with PIs was estimated at 5.1% (95%CI: 3.6-6.6) using a random-effects model[3,12,20,23,24,26-30,33,35]. When stratified by study setting, prevalence was significantly higher in studies conducted across multiple sites (15.8%, 95%CI: 7.4-30.4) compared to those conducted at a single site (4.9%, 95%CI: 3.5-6.3; P < 0.0001; Table 2). A similar pattern was observed across regions, with studies from Africa reporting a higher pooled prevalence (7.3%, 95%CI: 3.4-11.2) than those from Asia (4.1%, 95%CI: 3.4-4.9; P < 0.0001; Table 2). Between-study heterogeneity was moderate overall (I² = 50.5%) and was further reduced within the subgroups, particularly among multiple-site studies (I² = 0.0%) and studies from Asia (I² = 1.8%). Finally, leave-one-out sensitivity analysis confirmed the robustness of the pooled estimate, as exclusion of individual studies did not substantially alter the effect size or statistical significance (Supplementary Figure 4)[3,12,20,23,24,26-30,33,35].

Pooled prevalence of DRM due to INSTI drugs: The pooled prevalence of DRM associated with INSTIs was estimated to be 3.5% (95%CI: 0.6-6.4) using a random-effects model (Figure 3E)[3,12,20,30]. Although moderate heterogeneity was observed (I² = 51.1%), the leave-one-out sensitivity analysis demonstrated that the overall effect estimate was stable across iterations, with each omitted study yielding consistently significant pooled estimates (Supplementary Figure 5)[3,12,20,30].

Sensitivity analysis for the pooled prevalence of DRM

To assess the influence of studies reporting no DRM, a sensitivity analysis was conducted excluding Rupérez et al[31], which reported zero cases. Upon its exclusion, the pooled prevalence slightly increased to 53.5% (95%CI: 41.3%-65.7%) compared to 50.4% (95%CI: 38.3-65.5) when it was included. The two prevalences are comparable with considerable overlapping confidence intervals, indicating that the exclusion of this study did not substantially affect the overall estimate. Heterogeneity was similar prior and after its exclusion (98.9% vs 98.8%), indicating similar variation in true effects size across different studies prior and after its exclusion.

Assessment of publication bias across meta-analyses

Egger’s regression test for small-study effects was used and it detected no significant publication bias for the pooled prevalence of successfully sequenced individuals (P = 0.1487), or overall DRM (P = 0.2338). However, significant small-study effects were detected in the analyses of DRM due to NRTIs (P = 0.0157), non-nucleoside NNRTIs (P = 0.0014), and PIs (P = 0.0245). These findings suggest the presence of publication bias in these subgroup analyses, which may influence the interpretation of the pooled estimates and should be interpreted with caution.

DISCUSSION

This meta-analysis synthesized global evidence on the prevalence of HIV DRM among PLWH who had LLV in LMIC. Approximately 74% of the VL samples from these patients were successfully sequenced. Overall, half (50.4%) of the sequenced individuals harbored at least one resistant mutation. Stratified prevalence by antiretroviral drug class revealed striking differences in DRM prevalence. DRM due to NNRTI was highest at 50.9%, followed closely by NRTI resistance at 44.6%. In contrast, PI-resistance and NSTI-resistance remained low at 5.1% and 3.5% respectively. This pattern reflects historical ART prescribing patterns in LMICs, where NNRTI-based regimens (e.g., efavirenz, nevirapine) were widely used, often in combination with NRTIs like lamivudine or tenofovir. Regional disparities were evident, with studies from Africa showing higher pooled DRM prevalence than those from Asia. Single-site studies also reported higher DRM rates than multi-sites, possibly reflecting patient complexity or differences in sequencing thresholds. Further investigation into study setting and participant characteristics is warranted.

These findings reinforce growing clinical and public health concerns that the current WHO VL threshold of < 1000 copies/mL to define virologic suppression may be insufficient to detect clinically meaningful resistance. Under the current WHO guidelines, PLWH who had LLV are regarded as virally suppressed and therefore remain in their current ART regimen without resistance testing or regimen adjustment. However, this analysis demonstrates that half of those with LLV had evidence of resistance mutations indicating that these patients remained on failing regimens. Continued low level viremia in the context of ineffective therapy increases the risk of acquiring additional mutations[37], progressing to advanced disease[4], acquiring opportunistic infections and potential onward transmission of drug resistance viruses[29]. The findings of this analysis support the argument of previous researchers and recommendations to lowering the threshold of VL suppression from < 1000 copies/mL to 200 copies/mL particularly in LMIC where genotypic resistance testing is not routinely done[3,27,30]. The recommendation is based on the following observations: (1) VL < 200 copies/mL have lower likelihood of developing DRM[12]; (2) Align suppression definition of VLS across settings; (3) Prevent prolong exposure to ineffective regimens; and (4) Reduce cumulative DRM burden[6,7].

The findings of this analysis are consistent with Swenson et al[11], who reported that 30% of patients had detectable resistance during their first episode of LLV, with NRTI resistance in 28%, NNRTI resistance in 16%, and PI resistance in 7% of patients. In their study, 50% of LLV patients eventually failed treatment, and the risk of failure increased as resistance intensified. Similarly, finding from this review also reflect similar patterns reported in China by Liu et al[12], where DRMs among PLWH with LLV were identified in 42.3% of patients. NNRTI resistance was again dominant (35.6%), followed by NRTIs (23.8%), PIs (5.8%), and INSTIs (3.9%). Their identification of M184V and K65R mutations, both common in our reviewed papers, further validates the widespread circulation of key mutations in LMICs, particularly in regions with long-standing reliance on lamivudine-based and tenofovir-based regimens.

The apparent differences in prevalence DRM between African and Asian studies may reflect underlying epidemiologic and programmatic factors including poor adherence, HIV subtypes, genetic diversity, duration on treatment, and type of regimen)[39,40]. Gupta-Wright et al[14] observed that 88% of hospitalized patients in Malawi with VLS between 400-999 copies/mL had multi-class resistance, and nearly all harbored resistance to NNRTIs and at least one NRTI. Despite low prevalence of PI and INSTI resistance, their cohort experienced high mortality (25% in those with VF vs 18% in those without), reinforcing the life-threatening implication of under-detected resistance. While INSTI resistance is uncommon, it is emerging. Only four studies in this review reported INSTI resistance, likely due to more recent introduction of dolutegravir-based regimens in LMICs. Still, as seen in Swenson et al[11], and Liu et al[12], resistance to INSTIs was documented, particularly in patients with prior treatment failure or incomplete adherence. This highlights the need for prospective surveillance in LLV populations now receiving INSTI-based regimens. The disproportionately high resistance found in single-site studies (particularly for NRTIs and NNRTIs) may be attributable to their location in tertiary or referral centers, where more complex or treatment experienced patients are typically managed. However, caution is warranted in over-interpreting this trend, as few studies explicitly reported facility type or patient ART history. Future research should report study setting, ART duration, and patient-level factors to improve contextual interpretation of DRM estimates.

Due to its effectiveness, improved side effects profile and high genetic barrier to resistance mutations, between 2018 and 2019, the WHO recommended the use of dolutegravir based regimen as first line treatment in PLWH[41]. Subsequently, many countries in Asia and Africa adopted its use. Current data showed that dolutegravir based regimen uptake ranging from 66%-77%[42]. Of the 175 HIV clinics surveyed in LMIC, 90% reported adoption of dolutegravir based regimen as first line treatment[43]. Expanded use of dolutegravir based regimen would prevent development of DRM, however, increased adherence support and frequent viral monitoring should still be emphasized to minimize development of DRM and preserve limited treatment options.

Several limitations were acknowledged in this study. First, although this analysis reported DRM among PLWH who had LLV; it was not possible to distinguish whether these mutations were accumulated prior to LLV periods. The temporal relationship between DRM emergence and LLV remains unclear due to limited longitudinal data in the included studies. Second, the geographical distribution of included studies was uneven. While the review focused on LMICs, most eligible studies were from sub-Saharan Africa and Asia with noticeable gaps from Latin America, Eastern Europe, and francophone countries. Additionally, all included studies were limited to English-language publications, which may have introduced language bias. Thirdly, heterogeneity in study design, patient populations, and sequencing methodologies may have influenced the pooled estimates. For example, some studies were single-site investigations conducted at tertiary referral centers that typically manage more complex or treatment-experienced cases. These sites may report higher DRM prevalence limiting generalizability. Additionally, small-study effects were observed in some subgroup analyses, particularly for some drug-class subgroup analyses (NRTI, NNRTI, PI); therefore, those subgroup pooled estimates should be interpreted with caution. Despite these limitations, the inclusion of 20 studies from different LMICs countries resulted in a large sample size and strengthened results observed in this study. This review is important because it summarizes DRM among PLWH who had LLV and provides recommendations to consider lowering VL suppression threshold to prevent keeping people on failed regimen.

CONCLUSION

This meta-analysis demonstrates that a substantial proportion of PLWH with LLV harbors DRM particularly to NRTIs and NNRTIs regimens, placing them at risk of silent VF. Routine genotyping is largely unavailable in LMICs, and WHO current VL threshold of ≥ 1000 copies/mL may delay identification of failing regimens. These findings support a global reconsideration of suppression thresholds, particularly in resource-limited settings, to prevent accumulation of resistance and preserve remaining treatment options. Enhance adherence support, periodic VL monitoring, and surveillance of emerging dolutegravir resistance should be prioritized to safeguard ART effectiveness in low level viremia population.

ACKNOWLEDGEMENTS

We acknowledged Emilie Ludeman for systematic search of the manuscripts and their compilation into the Covidence software.