Human immunodeficiency virus recency testing coverage and partner-notification-services among people-living with human immunodeficiency virus in low- and middle-income countries

Abstract

BACKGROUND

Human immunodeficiency virus (HIV) recency testing provides data that can be used to monitor the trend of new HIV infections. The effectiveness of using people identified with recent infection to identify partners with new HIV infection through partner notification services (PNS) is not well documented.

AIM

To determine the pooled prevalence of recency testing coverage, recent infection, reclassification (recent to long-term infection) and PNS cascade among newly diagnosed people living with HIV.

METHODS

PubMed, Cochrane Library and Embase were searched for articles published between January 2018 and November 2024. Studies were included if they reported recency coverage and/or PNS among people newly diagnosed with HIV and used recent infection testing algorithm (RITA). Recency coverage was defined as proportion of people tested using rapid testing for recent infection (RTRI) among those newly diagnosed with HIV. RITA further classifies RTRI results using viral load results (≥ 1000 copies/mL vs < 1000 copies/mL) to confirm recency status. For studies with PNS, we evaluated the cascade: Number of partners elicited, successfully contacted, eligible for HIV testing, tested and HIV diagnosis. PNS effectiveness was measured by proportion of new HIV diagnoses from tested partners. Using random effects models, we computed the pooled estimate of recency outcomes and 95% confidence intervals (CIs).

RESULTS

Twenty-five studies from 17-low- and middle-income countries were included. Of 276315 newly diagnosed people living with HIV, 79864 underwent RTRI with an overall pooled recency coverage of 87% (95%CI: 67-96). The pooled prevalence of RTRI and RITA recency were 12% (95%CI: 9-16) and 7% (95%CI: 4-10), respectively. Pooled prevalence of RTRI reclassification was 34% (95%CI: 22-49). Of the recent cases who agreed to PNS, 253 partners were elicited with an estimated elicitation ratio of 1:1.6. Among partners elicited, 99% were successfully contacted, 75% were eligible for testing, 68% tested for HIV, and 15% were diagnosed with HIV.

CONCLUSION

High recency testing coverage among newly diagnosed individuals demonstrates the feasibility of monitoring new HIV infections in LMIC. While PNS yielded moderate HIV diagnoses, its targeted approach remains a critical strategy for identifying undiagnosed cases.

Key Words: Human immunodeficiency virus recency testing coverage; Rapid test for recent infection; Recent infection testing algorithm; Partner notification services; Misclassification; Recent infection surveillance; Low- and mid-income countries

Core Tip: This systematic review and meta-analysis synthesizes evidence from low- and middle-income countries on the implementation of human immunodeficiency virus (HIV)-recency testing and its integration with partner notification services (PNS). The findings highlight high but uneven rapid testing for recent infection (RTRI) coverage, moderate RTRI-recent prevalence, and substantial diagnostic misclassification. One in every three recent infections was reclassified as long-term following a viral load confirmatory testing. While PNS demonstrated strong partners contact rates and high HIV yield, critical attrition occurred throughout the cascade. These results highlight the value of RITA-confirmed recency testing and suggest that recency-informed PNS could enhance HIV surveillance and epidemic control when guided by standardized protocols and supported by confirmatory testing.

INTRODUCTION

The global human immunodeficiency virus (HIV) epidemic remains a major public health challenge, with an estimated 39.9 million people living with HIV (PLWH) in 2022, disproportionately affecting low- and middle-income countries (LMICs)[1]. Although progress has been made towards the UNAIDS 95-95-95 targets - improving diagnoses, expanding antiretroviral therapy (ART) coverage, and achieving viral suppressions - significant gaps persist, particularly in the timely diagnosis of new infections and interrupt ongoing transmission[1-3]. Strengthening HIV surveillance and implementing targeted interventions are essential to identifying new infections and halting transmission, especially in high-incidence settings where health care access and prevention programs face systematic challenges.

HIV recency testing, designed to differentiate recent infection (acquired within the past 6-12 months) from long-standing infections, offers a promising tool to address these challenges[4-7]. By identifying individuals with recent infection at diagnosis, recency testing enabled public health program to target sub-populations and geographical areas where transmission is most active, thereby informing more effective public health intervention[3,4,8,9]. When integrated with partner notification services (PNS), recency testing can amplify the detection of undiagnosed infections, interrupt transmission chain and accelerate progress towards epidemic control[10-12].

Several LMIC’s have introduced recent infection surveillance within routine HIV testing services, often as part of national case-based surveillance systems[13-15]. However, the success and utility of recency-informed strategies depend heavily on achieving adequate recency testing coverage, accurately estimating the proportion of recent infections, and applying confirmatory tests to reduce misclassification. These operational metrics vary widely across programs and settings[13,14]. The World Health Organization (WHO) recommends using a recent infection testing algorithm (RITA), which combines a rapid test for recent infection (RTRI) with supplementary viral load testing to minimize false recent results and ensure reliable estimates, particularly in LMIC’s where ART coverage is expanding[4,6,10].

A critical application of recent testing lies in optimizing PNS outcomes[10,11]. Individuals with recent infections often have high viral loads and are unaware of their status, increasing their risk of transmitting HIV[8,13]. Identifying such index cases allows programs to prioritize partners elicitation and testing, leading to earlier diagnosis, linkage to care, and prevention intervention for partners[11,16,17]. This integration has been shown to improve efficacy along the entire PNS Cascade, from partner elicitation through HIV positivity yield[18,19], which is critical for maximizing limited resource in high-burden resource-constrained low- and middle-income countries (LMICs).

Despite increasing implementation of recency testing and PNS in LMIC’s, no comprehensive synthesis has pooled quantitative outcomes across diverse settings. This systematic review and meta-analysis address this gap by estimating pooled prevalence of recency testing coverage, recent infection rates, reclassification from recent to long-term infection, and PNS cascade outcomes among newly diagnosed PLWH in LMIC. Our findings will inform national HIV strategies, guide targeted prevention programs, and optimize resource allocation in LMICS to achieve epidemic control.

MATERIALS AND METHODS

Registration

The protocol for this systematic review was registered in the International Prospective Register of systematic Reviews (PROSPERO) under registration number CRD420251081733. The review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and meta-Analysis (PRISMA) guidelines.

Ethical approval

As this study involved secondary analysis of published studies and programmatic reports ethical approval was not required.

Search strategy

We searched PubMed, EMBASE, and the Cochrane Library for studies published between January 1, 2018, and November 30, 2024, using combinations of keywords and MeSH terms related to HIV recency testing, the RITA RTRI and PNS including index testing, contact tracing, and elicitation. The search strategy focused on LMIC’s and included only English-language publications. The databases search yielded 268 records. We also identified 13 additional records through manual searches of conference abstracts and grey literature sources, bringing the total to 281 records. The full search was conducted on May 13, 2025.

We uploaded the records from electronic data bases into Rayyan software for deduplication and screening. Of the 281 records, we identified two duplicates manually removed one and ultimately included 280 records before initiating the screening process detail. Search strategies are provided in Supplementary Table 1.

Eligibility criteria

We included studies that reported on any component of the HIV recency testing cascade or the PNS cascade among newly diagnosed PLWH in LMICs. Eligible recency cascade elements included initial testing for recent infection, classification, or reclassification to long-term infections, and confirmatory viral load results. Recency testing coverage was defined as the proportion of newly diagnosed individuals who received RTRI at the point of HIV diagnosis. Studies were required to have implemented RITA which combines an initial recency assay (such as RTRI, Lag avidity EIA, or another validated serological method) with confirmatory viral load testing (≥ 1000 copies/mL) in accordance with WHO guidance to be considered for reclassification from recent to long-term infection[4].

Studies were also eligible if they reported on PNS outcomes including the number of sexual/needle sharing partners elicited, successfully contacted, found eligible for testing, tested for, and diagnosed with HIV. Only studies conducted in LMICS among newly diagnosed PLWH were included. We excluded studies that did not report any relevant recency testing or PNS data, studies that were non-primary (e.g., reviews, editorials, commentaries), those conducted in ineligible population and those lacking sufficient quantitative data for extraction.

Study selection

Two reviewers independently screened all 280 records for eligibility using Rayyan. After title and abstract screening, we excluded 238 records. We sought full texts for the remaining 42 records, all of which were successfully retrieved. We assessed these 42 full-text articles for eligibility and excluded 17 articles, 11 due to inclusion of the wrong population and six due to irrelevant outcomes.

At each stage of review, discrepancies were resolved through discussion and consensus. Ultimately, we included 25 studies in the final analysis. These included both peer-reviewed publication and high-quality conference abstracts that met all inclusion criteria despite the absence of full manuscripts. The PRISMA 2020 flow diagram (Figure 1) summarizes the selection process.

Figure 1 PRISMA flow diagram of study selection. Flowchart illustrating the systematic review process. A total of 280 records were screened after removing 1 duplicate. Of these, 42 full text reports were assessed for eligibility, and 25 studies met the inclusion criteria. 17 reports were excluded due to wrong population (11) or outcome (6).

Data extraction

Data were independently extracted by two reviewers using a pre-defined Excel template. Extracted data included study author and year of publication, country and setting, study design, and sample characteristics. Recency testing indicators included the number of newly diagnosed individual, number tested with RTRI, coverage of RTRI testing, number of cases reclassified from recent to long-term infection, and details of the reclassification process. For PNS, extracted outcomes include the number of partners elicited, successfully contacted, tested for HIV, found eligible for HIV testing, and diagnosed with HIV. PNS effectiveness was measured as the proportion of newly diagnosed HIV positive partners among those tested. Any discrepancies in data extraction were discoursed and resolved by consensus.

Quality assessment

The methodological quality of included studies was assessed using the Joanna Briggs Institute (JBI) tools for observational studies. The tool consists of nine questions with four responses: (Yes, No, Not clear, Not applicable). We assigned a score of 1 to a “Yes” response and 0 to a “No” response. Each scored question was totaled and classified into three categories. Studies with (0-3), (4-6) and (7-9) scores were regarded as being of low, medium and high quality respectively. Two pairs of reviewers (Beatrice Kelvin Mpimo and Haji Mbwana Ally) and (Hassan Fredrick Fussi and Upendo Kayeke Chenya) independently performed and rated the quality of the studies using the JBI tools. Discrepancies of the scores between the two pairs were sorted by a third pair of reviewers (Habib Ramadhani Omari and Hafidha Mhando Bakari).

Definition of variables

The primary outcomes of interests were recency testing coverage at point-of-care, prevalence of RTRI-recent infection, prevalence of RITA-recent infection, and the proportion of RTRI-recent cases reclassified as long-term infections based on confirmatory viral load results. Furthermore, we also reported the proportion of individuals successfully reached across the PNS Cascade including elicitation ratio, prevalence of partners of recent HIV cases who were successfully contacted, prevalence of partners of recent HIV cases who were eligible for HIV testing, prevalence of partners of recent HIV cases who tested for HIV, and prevalence of partners of recent HIV cases who were diagnosed with HIV. The secondary outcome was the effectiveness of partner notification, measured as the proportion of tested partners who were newly diagnosed with HIV.

Statistical analysis

Using random and fixed effects models, we computed pooled prevalence of HIV recency testing uptake, RTRI-recent infection, RITA-recent and reclassification from recent to long-term HIV infections. Additionally, we also evaluated PNS cascade by quantifying several components of the cascade including elicitation ratio, prevalence of partners of recent HIV cases who were successfully contacted, prevalence of partners of recent HIV cases who were eligible for HIV testing, prevalence of partners of recent HIV cases who tested for HIV, and prevalence of partners of recent HIV cases who were diagnosed with HIV. Subgroup analysis on the pooled estimates of RTRI and RITA prevalence were performed to compare studies conducted from surveys/Laboratory samples vs those conducted from HIV programs using χ² tests. Studies heterogeneity was assessed by computing the I² statistic and Cochran’s Q. The score values of I² statistics were categorized at 75%, 50% and 25% to signify the presence of high, moderate and low heterogeneity respectively as previously described[20]. To assess publication bias, the Egger regression asymmetry test was used. To declare the presence of either heterogeneity or publication bias, a P value threshold of < 0.05 was used. For the prevalence outcome that showed either moderate or high degree of heterogeneity, we assessed its possible sources by conducting an influential analysis using the leave-one-out method[21]. In addition, we conducted a meta regression analysis to discern the variation of the HIV recency testing uptake. Using trim-and-fill method, we conducted a sensitivity analysis to assess possible small-study. All statistical tests were performed using R version 4.2.2 (R Foundation for Statistical Computing, Vienna, Australia).

RESULTS

Our search identified a total of 281 records comprising 268 from databases and 13 from manual sources. After deduplication, 280 unique records were screen by title and abstract using Rayyan. Of these, 238 records were excluded for not meeting the eligibility criteria. The remaining 42 records underwent full text review from which we excluded 17 additional studies (11 due to wrong population, 6 due to irrelevant outcomes). A total of 25 studies were included in the final analysis (Figure 1). These studies represent a diverse range of geographic regions within LMICs, and varied in study design, populations and covering various study periods. Collectively, thirteen peered-reviewed articles and twelve high-quality conference abstracts contributed data on HIV recency testing and PNS outcomes.

Study quality assessment

Assessment of study quality showed that all studies were of high quality, with scores ranging from 7-9. Those with scores less than 9, the common reasons were smaller sample size and a response rate of less than 80%. Neither study was of low nor medium quality (Table 1).

Table 1 Summary of characteristics of the included studies.

Ref.	Country	Data source	Number of newly diagnosed	Number of people received RTRI test	RTRI coverage	Prevalence of RTRI recent infection	Prevalence of RITA recent infection	Prevalence of Reclassification from recent to long-term infection	Quality scores
Truong et al[14], 2022	Eswatini, Nigeria, Rwanda, Thailand, Uganda, Vietnam, Zambia	HIV program	231410	41688	18.0				8
Aungkulanon et al[22], 2022	Bangkok	HIV program	1291	694	53.8	8.5	7.8	8.5	8
Simanovong et al[23], 2022	Thailand	HIV program	196	153	78.1	4.6	3.9	14.3	7
Okiror et al[37], 20241	Uganda	HIV program	-	-	-	-	-	-	7
Herce et al[28], 2024	Zambia	HIV program	344	322	93.6	-	3.2	-	9
Mochama et al[17], 20241	Kenya	HIV program	-	-	-	-	-	-	8
Ouk et al[18], 20221	Cambodia	HIV program	-	-	-	-	-	-	7
Lerdtriphop et al[19], 20221	Thailand	HIV program	-	-	-	-	-	-	8
Rice et al[12], 2020	Kenya and Zimbabwe	HIV program	1582	1269	80.2	16.9	7.2	5.9	9
Negedu-Momoh et al[29], 2021	Nigeria	Survey	-	-	-	5.1	3.0	4.2	9
Welty et al[16], 2020	Kenya	HIV program	628	532	84.7	11.7	9.4	19.4	9
Voetsch et al[5], 2021	≥ 10 countries2	Survey	-	-	-	10.3	1.5	-	9
Rwibasira et al[24], 2021	Rwanda	HIV program	10701	7919	74.0	9.5	6.0	36.4	8
Mohloanyane et al[35], 2023	Lesotho	Survey	-	-	-	9.9	7.4	25.0	9
Alemu et al[25], 2022	Ethiopia	HIV program	5689	3129	55.0	-	14.2	-	8
Parmley et al[30], 2022	Zimbabwe	Survey	-	-	-	8.6	1.1	87.5	7
Young et al[31], 2023	Kenya	Survey	-	-	-	11.8	5.6	86.6	8
Telford et al[8], 2022	Malawi	HIV program	9295	9168	98.6	6.1	3.3	45.3	9
Ang et al[32], 2021	Singapore	Survey	-	-	-	25.7	19.0	26.1	8
Msukwa et al[26], 2020	Malawi	HIV program	14022	13838	98.7	6.3	3.2	43.1	9
Agyeman et al[33], 2022	Malawi	HIV program	-	-	-	19.7	13.9	29.3	8
Singh et al[34], 2023	South Africa	Laboratory	-	-	-	43.0	38.6	10.3	8
Zhou et al[36], 2023	China	HIV program	-	-	-	-	19.5	-	7
Saito et al[11], 20231	Rwanda	HIV program	-	-	-	-	-	-	7
Zhu et al[27], 2020	China	HIV program	1157	1152	99.6	17.9	15.1	15.1	9

¹Indicates studies that reported partner notification services only among recently infected people living with human immunodeficiency virus.

²Indicates ≥ 10 countries included Cameroon, Côte d’Ivoire, Eswatini, Ethiopia, Kenya, Lesotho, Malawi, Namibia, Rwanda, Tanzania, Uganda, Zambia, and Zimbabwe.

HIV: Human immunodeficiency virus; RITA: Recent infection testing algorithm; RTRI: Rapid testing for recent infection.

HIV recency testing coverage

We included 11 studies reporting on HIV recency testing coverage using RTRI at point-of -care[8,12,14,16,22-28]. The studies spanned diverse geographic and programmatic settings within low- and middle-income countries and represented a combined total of 276315 individuals newly diagnosed with HIV. Individual study estimates of RTRI coverage ranged from 18% to 100%, with substantial variation across studies. The pooled RTRI coverage using a random-effect model was 87% (95%CI: 67-96), indicating significant heterogeneity across studies (I² = 100%, τ² = 3.95, P < 0.001). By contrast, the fixed effect model yielded a lower pooled estimate of 29% (95%CI: 29-29), largely influenced by disproportionate large sample size in Truong et al[14], 2022, which alone contributed 83.7% of the total sample size (Figure 2). Despite the wide variability, a leave-one-out Sensitivity analysis for RTRI coverage revealed that the overall pooled estimates remain stable across all iterations, with minimal change in heterogeneity metrics (I² = 100%, τ² range = 3.00-4.35) suggesting that no single study disproportionately influenced the summary effect size (Supplementary Figure 1).

Figure 2 Pooled recency testing coverage among newly diagnosed people living with human immunodeficiency virus in low- and middle-income countries. Forest plots showing the proportion of newly diagnosed individuals who received a rapid test for recent infection across 11 studies. Estimates are pooled using a random effects model. Substantial heterogeneity observed (I² = 100%).

Prevalence of RTRI-Recent HIV infection

We included 15 studies that reported the proportion of individuals testing recent for HIV infection using either RTRI, or other validated recency assays, such as limiting antigen avidity immunoassay (Lag-Avidity)[5,8,12,16,22-24,27,29-35]. These studies together represented a combined sample size of 49196 newly diagnosed PLWH across diverse, LMICs settings.

The pooled estimate from the random-effect model was 12% (95%CI: 9-16), while the fixed effect model produced a slightly lower estimate of 10% (95%CI: 10-11). The point estimate from individual studies varied widely, ranging from 5% to 4.3% (Figure 3). There was substantial heterogeneity between studies (I² = 98%. τ² = 0.4710, P < 0.01) suggesting considerable variability in testing approaches and populations.

Figure 3 Pooled prevalence of preliminary-recent human immunodeficiency virus infection across included studies. Forest plots summarizing the proportion of individuals identified as recently infected using recent infection testing algorithm assays in 15 studies. The pooled estimates is generated using a random effect model. Heterogeneity was high (I² = 98%).

To assess the robustness of this finding, we conducted a leave-one-out sensitivity analysis, which showed no significant deviation in the pooled estimates when each study was omitted in turn (range: 0.1 to 0.11; Supplementary Figure 2). This indicates that the summary estimate was not driven by any single influential study. We further stratified the analysis by data source, distinguishing between programmatic- (n = 8 studies) and laboratory/survey-based (n = 7 studies) data and the pooled estimates were 11% (95%CI: 8%-15%) and 14% (95%CI: 8%-23%) respectively. Nonetheless, the test for subgroup differences using the random effect model was not statistically significant (χ² = 0.43, df = 1, P = 0.51) (Table 2), indicating no meaningful differences between the two data sources. Thus, indicating a moderate but consistent burden of RTRI-recent infection among newly diagnosed individual in LMIC’s with variation by study context other than data source.

Table 2 Subgroup analysis on the prevalence of recency coverage, rapid testing for recent infection, recent infection testing algorithm and reclassification among newly diagnosed people living with human immunodeficiency virus in low- and mid-income countries.

	Sources of data		P value
	HIV program, % (95%CI)	Laboratory/surveys, % (95%CI)
RTRI-recent HIV infection	11 (8-10)	14 (8-23)	0.510
RITA-recent HIV infection	6 (5-6)	5 (2-13)	0.440

HIV: Human immunodeficiency virus; RITA: Recent infection testing algorithm; RTRI: Rapid testing for recent infection.

Prevalence of RITA-recent HIV infection

Eighteen studies reported on the prevalence of recent infection using a WHO-defined RITA, which incorporated an initial recency assay with a confirmatory viral load testing (≥ 1000 copies/mL)[5,8,12,16,22-36]. These studies collectively evaluated 66675 individuals. RITA-recent prevalence from individual study ranged from 1% to 39%, with an overall pooled prevalence of 7% (95%CI: 4-10) from the random-effect model and 4 (95%CI: 0.04-0.04) from the fixed effects model. High heterogeneity among the included studies was observed (I² = 99%. τ² = 1.0079, P < 0.001) reflecting that the variation observed across studies could not be explained by chance alone (Figure 4).

Figure 4 Pooled Prevalence of recent infection testing algorithm - recent human immunodeficiency virus Infection based on World Health Organization - algorithm. Forest plots showing the prevalence of recent infection using recent infection testing algorithm (rapid testing for recent infection + viral load ≥ 1000 copies/mL from 18 studies. Estimates are derived using random-effects and fixed-effect models with I² equal 99%.

To assess the robustness of this pooled estimate, we conducted an influential analysis using a leave-one-out approach (Supplementary Figure 3). The pooled prevalence remained consistently between 4% and 6%, regardless of which study was omitted, suggesting that no single study exerted undue influence on the overall estimate. Heterogeneity measures remain unchanged across iterations (I² = 99%, τ² = 1.0079) reinforcing the stability of the model. A subgroup analysis was further performed to explore whether the source of data contributed to the observed heterogeneity. Among the 7 programmatic studies the pooled prevalence was 6% (95%CI: 5-6), while the 7 Laboratory/survey-based studies yielded a slightly lower estimate of 5% (95%CI: 2-13) (Table 2). However, the test for subgroup differences using the random effect model failed to reach statistical significance (χ² = 0.60, df = 1, P = 0.44), implying that the data source alone does not explain the heterogeneity. Together, this analysis confirm that while the pooled prevalence of RITA-recent infection is approximately 7%, substantial variability exists across studies, likely due to differences in population characteristics, implementation fidelity and recency assay protocol.

Prevalence of reclassification

Fourteen studies reported on the proportion of individuals initially classified as recent who were reclassified as long-term after confirmatory viral load testing. Together, these studies encompass a total of 2612 individuals identified as recent prior to viral load confirmation, with a reclassification rate ranging from 8% to 88% (Figure 5). The random-effect pooled estimate was 34% (95%CI: 21-50), while the fixed-effects model yielded a slightly higher and more precise estimate of 0.40 (95%CI: 38-41). Heterogeneity across studies was high (I² = 96%, τ² = 1.4863, P < 0.01) reflecting differences in ART exposure, testing timing, and viral load suppression.

Figure 5 Reclassification rate of individuals initially classified as recent by rapid testing for recent infection. Forest plots of 14 studies reporting the proportion of rapid testing for recent infection - recent cases reclassified as long-term after viral load confirmation. Random-effects estimates with substantial heterogeneity (I² = 96%).

To assess whether any single study disproportionately influenced the pooled estimates, we performed a leave-one-out sensitivity analysis (Supplementary Figure 4). The pooled reclassification estimates remain relatively stable, ranging between 38% and 42% when individual studies were excluded one at a time. All recalculated heterogeneity statistics remain high, reinforcing the robustness of the pooled estimates while also underscoring persistent heterogeneity in the data. These findings confirm that approximately one-third of the individuals initially identified as recent are ultimately reclassified as long-term infections. This high reclassification rate underscores the critical role of viral load confirmation in RITA and may signal gaps in identifying true recent infection in high-ART coverage settings.

Prevalence of partner notification cascade

Proportion of partners successfully contacted: Five studies reported this indicator[17-19,28,37], with a pooled random-effect estimate of 99% (95%CI: 67-100) and fixed-effects estimates of 96% (95%CI: 93-98). Heterogeneity was low (I² = 11%, τ² = 11.2974, P = 0.34) suggesting high and consistent success in partners contacts (Figure 6).

Figure 6 Partner notification services among patients of newly diagnosed individuals. Bar charts showing the proportion of partners successfully contacted (99%), deemed eligible for human immunodeficiency virus (HIV) testing (75%), tested (68%) and diagnosed HIV-positive (15%) across 5 studies. The cascade highlights strong partner reach but notable drop-offs at eligibility and testing stages with a high positivity yield among those tested. HIV: Human immunodeficiency virus; PNS: Partner notification services.

Proportion of partners eligible for HIV testing: The pooled random-effects estimates across five studies was 75% (95%CI: 59-86) (Figure 6), while the fixed effects estimates was 76% (95%CI: 70-0.81). Heterogeneity was moderate-to-high (I² = 84%, τ² = 0.5361, P < 0.01), likely reflecting differing definitions and population characteristics.

Proportion of eligible partners tested for HIV: Among partners eligible for testing, the pooled random-effect estimate was 68% (95%CI: 56-79) (Figure 6). The fixed effects estimates was 69% (95%CI: 63-75), with moderate heterogeneity (I² = 75%, τ² = 0.2507, P < 0.01).

Prevalence of HIV among tested partners: Across five studies, the pooled HIV prevalence among tested partners was 15% (95%CI: 10-22) (Figure 6) using a random-effects model and 15% (95%CI: 11-20) using a fixed effect-model. Heterogeneity was low (I² = 39%, τ² = 0.0906, P = 0.16).

Collectively, these results demonstrate variable performance across the HIV recency testing and PNS Cascades, with strong partner tracing outcome, but more modest testing uptake and case detection.

DISCUSSION

We conducted a systematic review and meta-analysis to synthesize current evidence on the implementation and performance of HIV recency testing and its integration with PNS among newly diagnosed individuals in LMICs. Our findings provide critical insights into testing coverage, diagnostic accuracy, and programmatic impacts, while identifying key gaps that must be addressed to strengthen epidemic control efforts.

The high RTRI coverage of 87% across 11 studies demonstrates substantial integration of recency testing into routine HIV services in many LMICs. High RTRI uptake is crucial for implementing real-time surveillance and response to ongoing HIV transmission. However, the wide range in coverage estimates (range: 18%-100%) raise concerns about programmatic consistency and scalability. This finding is in line with earlier reports from programs in Kenya, Zimbabwe, Zambia, and Rwanda[10,12,16,24,28], where coverage was influenced by policy adoption, training quality and test kits availability. The results suggest that while national programs may report high RTRI uptake, subnational heterogeneity remains a key challenge. Programs must therefore strengthen coverage equity across districts, improving training and addressing logistical constraints to maximize the utility of recent infection surveillance.

Our meta-analysis reveals critical insights about HIV recency testing’s utility and limitations in LMICs. The pooled prevalence of 12% based on RTRI results from 15 studies encompassing over 49000 newly diagnosed individuals across LMICs highlights a moderate but epidemiologically meaningful burden of likely incidents infections, consistent with reports from Kenya, Malawi and Rwanda (> 10%)[5,12,16,24]. However, when restricted to studies using the WHO-recommended RITA, the pooled prevalence declined to 7% across 18 studies with 66675 individuals. This 42% reduction demonstrates the critical role of viral load confirmation in excluding false-recent cases, and exposes RTRI vulnerability to misclassification, particularly in high-ART coverage settings where early treatments and rapid viral suppression are common[8,24,29-31,33].

The discrepancy between RTRI and RITA estimates is further clarified by our pooled reclassification rate of 34% across 14 studies, including 2612 individuals, indicating that nearly one in three individuals initially classified as recent were ultimately deemed long-term after VL testing. These findings reinforce previous observations from Kenya, Malawi, Rwanda and Nigeria, where unconfirmed recency testing led to overestimation of recent infections and misdirected prevention efforts[8,24,29-31]. High heterogeneity observed in all three analyses likely reflects contextual and operational variability, including diverse recency assays, ART coverage levels, differences in VL suppression rates, implementation fidelity of VL testing, and variation in data type (Facility vs community-based). Notably, our sensitivity analysis revealed that no single study unduly influenced the pooled estimates, and that there were no significant prevalence differences between programmatic and research settings. These findings further affirm that real-world data, when rigorously collected, can yield reliable surveillance estimates[10-12], which supports the WHO endorsement of recency testing for dynamic surveillance, but underscores that its accuracy hinges on confirmatory testing infrastructure[7,10].

For public health programs, these findings carry important implications - while RTRI provide an accessible tool for transmission hotspots identification, its stand-alone use possess substantial misclassification risks in high-ART-coverage setting potentially distorting resource allocation[28,38]. In contrast, despite RITA's superior accuracy, implementation barriers persist - particularly VL processing delays and decentralized testing infrastructures that hinder timely recency classification[30,39]. National programs must therefore prioritize fidelity to RITA protocols, ensuring timely VL-testing and results return and integrate recent infection surveillance with partner services and outbreak response mechanisms[40]. Emerging solutions like multi-assay algorithm (Sedia HIV recency assay) and machine learning approaches integrating clinical meta-data show promise but require further validation in programmatic context[6,41,42]. While limitations like residual misclassification and assay variability persist[43] a tiered RTRI-RITA approach with strengthen laboratory systems, offers LMIC's optimal balance of feasibility and accuracy for recency-based surveillance and epidemic control.

Our meta-analysis highlights both the strengths and persistent gaps in PNS implementation across LMIC, with implications that align closely with insights from the HIV recency cascade. The near-universal success in Partner contacts, 99% demonstrates the consistent feasibility of index-case-based approaches across diverse settings, mirroring documented successes in Rwanda and Kenya where provider-assisted notification achieved > 85% contact rates[10,11,28,37]. However, significant cascade attrition emerges at subsequent stages, where only 75% of contacted partners met eligibility criteria and merely two-third of eligible partners completed HIV testing. The moderate-high heterogeneity in testing eligibility and uptake likely reflects operational variations in partner definitions, consent procedures, and persistent structural barriers including stigma and patient mobility - challenges consistently identified in comparable LMIC implementation[44-46].

The substantial 15% HIV prevalence among tested partners - nearly triple typical general population testing yields[47-49], confirms PNS as a high value case-finding strategy. This prevalence closely corresponds with our finding of 12% RTRI-recent infections among index cases, suggesting PNS effectively captures active transmission networks. However, the 34% RTRI-RITA reclassification rates introduce a critical programmatic consideration: Partners of virally suppressed index cases (likely long-term infection) may represent established rather than acute infections, potentially diluting PNS’s outbreak interception value[5,17,30]. To optimize impact, programs should prioritize integrating RTRI-screening with confirmatory RITA testing to strategically direct PNS resources towards truly viremic index cases most likely to yield recent partner infection[10,11]. Concurrently, standardizing eligibility criteria and testing protocols could substantially reduce the observed cascade attrition while improving cross-program comparability. Finally, given the elevated transmission risk associated with recent infections, strengthening post-test linkage systems for partners identified through recency-triggered PNS remains essential for maximizing prevention benefits. While these findings underscore PNS epidemiological value, limitations include potential underreporting of sensitive partnerships warrant consideration in implementation. Future research should evaluate integrated recency-PNS models to determine their optimal configuration for maximizing both individual and population level prevention outcomes.

This study provides the most comprehensive evaluations to date of integrated HIV recency testing and partner notification services across LMIC’s, synthesizing data from 25 studies encompassing tens of thousands of newly diagnosed individuals. Its principal strength lies in the novel integration of recency and PNS cascade analysis, revealing critical intersections between diagnostic accuracy and intervention effectiveness that previous studies have addressed separately. The applications of both random- and fixed-effect models with rigorous sensitivity analysis enhance the robustness of findings, while the consistent HIV prevalence among tested partners across diverse settings validates the epidemiological utility of well implemented PNS programs. Furthermore, our inclusion of both programmatic and research data provides unique insights into real-world implementation challenges and best practices.

We acknowledged several limitations in the study including substantial heterogeneity (I² = up to 100%) across many analyses despite subgroup explorations, reflecting unavoidable variations in recency assay performance, PNS eligibility criteria, and program maturity levels. While our modeling approaches accounted for this variability, residual confounding from a measured contextual factor (e.g., local stigma levels, health system resilience) may remain. Additionally, the predominance of routine program data introduces potential reporting biases, particularly for sensitive indicators like partner refusal rates, or exact testing timelines. Furthermore, while RITA confirmation reduced misclassification, residual challenges persist in high ART coverage settings where atypical viral suppression patterns may still lead to under-detection of acute infections. Finally, the geographic concentration of studies (predominantly East and Southern Africa) may limit generalizability to regions like West/ Central Africa where epidemic dynamics and health systems differ.

CONCLUSION

Based on these findings, we propose 3 priority actions for programs: (1) Implementation of tiered recency testing protocols that prioritize RITA-confirmed cases for PNS resource allocation; (2) Standardizations of PNS eligibility criteria and quality metrics across programs to reduce cascade attrition; and (3) Investments in point of care viral load platforms to minimize confirmation delays. Future research should focus on the following key areas, cost effectiveness analysis of integrated recency PNS models, validation of next generation multi-assay algorithms in routine care settings, and implementation science studies to optimize approaches for key populations currently underrepresented in recent infection surveillance (particularly Men who have Sex with Men & People Who Inject Drugs). This meta-analysis demonstrates that recency-informed HIV programs can effectively identify active transmission networks when implemented with confirmatory testing and strong partner services. The 12% RTRI-recent prevalence and corresponding 15% partner HIV-prevalence confirm substantial ongoing transmission, while the 34% reclassification rate underscores the necessity of viral load confirmation. Moving forward, the strategic integration of recency testing with PNS-guided by RITA confirmation and supported by robust linkage systems, offers a transformative opportunity to focus limited resources on the highest risk transmission networks. As LMIC’s advance toward epidemic control, these findings provide an evidence-based road map for optimizing surveillance and prevention investments to maximize population-level impacts.