This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR Public Health and Surveillance, is properly cited. The complete bibliographic information, a link to the original publication on http://publichealth.jmir.org, as well as this copyright and license information must be included.
Normative guidelines from the World Health Organization recommend tracking strategic information indicators among key populations. Monitoring progress in the global response to the HIV epidemic uses indicators put forward by the Joint United Nations Programme on HIV/AIDS. These include the 90-90-90 targets that require a realignment of surveillance data, routinely collected program data, and medical record data, which historically have developed separately.
The aim of this study was to describe current challenges for monitoring HIV-related strategic information indicators among key populations ((men who have sex with men [MSM], people in prisons and other closed settings, people who inject drugs, sex workers, and transgender people) and identify future opportunities to enhance the use of surveillance data, programmatic data, and medical record data to describe the HIV epidemic among key populations and measure the coverage of HIV prevention, care, and treatment programs.
To provide a historical perspective, we completed a scoping review of the expansion of HIV surveillance among key populations over the past three decades. To describe current efforts, we conducted a review of the literature to identify published examples of SI indicator estimates among key populations. To describe anticipated challenges and future opportunities to improve measurement of strategic information indicators, particularly from routine program and health data, we consulted participants of the Third Global HIV Surveillance Meeting in Bangkok, where the 2015 World Health Organization strategic information guidelines were launched.
There remains suboptimal alignment of surveillance and programmatic data, as well as routinely collected medical records to facilitate the reporting of the 90-90-90 indicators for HIV among key populations. Studies (n=3) with estimates of all three 90-90-90 indicators rely on cross-sectional survey data. Programmatic data and medical record data continue to be insufficiently robust to provide estimates of the 90-90-90 targets for key populations.
Current reliance on more active data collection processes, including key population-specific surveys, remains warranted until the quality and validity of passively collected routine program and medical record data for key populations is optimized.
The 2015 World Health Organization (WHO) guidelines for HIV-related strategic information (SI) [
Disaggregation of HIV surveillance indicators [
The guidelines recommend estimating key population-specific indicators drawn from multiple data sources; however, the operational challenges of this are significant. First, there are often no standardized definitions for key populations and subgroups. Transgender women are not consistently classified separately from MSM; women who exchange sex for goods or services or work part-time may or may not be included as sex workers; and MSM are often considered one group, despite the significant diversity of HIV acquisition and transmission risks among MSM [
Second, as progress is made in reaching the most accessible members of key populations, the HIV epidemic will reach a phase [
Strategic information cycle.
Third, beyond the challenges of aligning population definitions, identifying subgroups, and geographic reach, there is the challenge of reconciling the different sources of bias in and across survey methods, program monitoring systems, and routine medical record data systems. Different survey and population size estimation methods can result in different HIV prevalence and size estimates for the same population in the same location [
Finally, although estimation of the number and percent of new HIV infections is one of the 10 global indicators recommended in the new guidelines, there are methodological and operational challenges of validly measuring incidence among key populations [
We reviewed the published history of HIV surveillance among key populations to put the current challenges into context; identify future opportunities to enhance the use of surveillance data, programmatic data, and medical record data; describe the HIV epidemic among key populations; and measure the coverage of HIV prevention, care, and treatment programs.
We described the historical expansion of HIV surveillance using published guidelines from WHO and UNAIDS, early surveillance reports identified using MEDLINE, and coauthor recollection.
We described the current salient features of four sources of SI: national household surveys, targeted bio-behavioral surveys, medical record data, and program data sources. We included known strengths, weaknesses, and opportunities for improved use of these data sources, with a focus on estimating elements of the treatment and prevention cascades [
We also review published estimates of 90-90-90 indicators based on these data sources to identify the availability of cascade estimates for key populations and the extent to which they drew on program, survey, and treatment databases. We conducted a title or abstract search in MEDLINE using the terms: HIV AND [HIV Testing OR population size estimate OR Viral Suppression OR antiretroviral therapy] AND [key populations OR MSM OR sex workers OR injection drug use OR prison OR transgender OR concentrated epidemic] AND [program data OR surveillance data OR routine data OR medical records]. We included manuscripts based on two criteria: (1) it reported any elements of the 90-90-90 HIV treatment cascade (population size, the proportion of the population who know their HIV status, the proportion of the population with HIV that is receiving antiretroviral therapy [ART], and the proportion of the population that is achieving viral suppression) for key populations from low- or middle-income countries and (2) it described strengths or weaknesses of data sources, or issues of aligning data from different sources. For studies where inclusion criteria could not be determined based solely on the abstract, we searched the full article.
Surveillance of the HIV pandemic has evolved over the past three decades, but from the earliest days it has included information about key populations. Early clinical case reports of patients with AIDS in Haiti in 1983 [
Subsequently, guidance surrounding data collection and monitoring for high-risk subgroups started to appear. In 1985, a WHO Coordinating Center Report [
In 1999, WHO published a comprehensive guide outlining the specific data elements required for STI case reporting in clinical settings [
In 2000, WHO published guidelines for second generation surveillance [
In 2013, these guidelines were updated to incorporate the experiences of countries implementing second generation surveillance over the past 10 years and to incorporate changes in survey methods and laboratory diagnostics [
In response to each set of guidelines, surveys and surveillance systems were dynamically changed, though the implementation varied by country, region, epidemic profile, and study objective. Currently, many countries have yet to achieve the recommendations from 2103 to align program and surveillance data to describe the HIV epidemic and evaluate the response. Alignment with health record data is a further challenge. Surveillance activities are often implemented with little regard for clinical programs, partly because HIV status and ART were considered too confidential to ask about in surveys. Clinical data can be challenging to triangulate with survey data because definitions of geography or reference period and population do not necessarily align. The current recommendation to estimate the HIV treatment cascade represents a paradigm shift in surveillance and program monitoring because valid estimates of cascade indicators for a district require alignment of definitions, geography, and reference period across survey, program, and treatment databases.
Bio-behavioral surveys of key populations have been the backbone of HIV surveillance for key populations over the past 15 years, particularly for HIV prevalence estimates and more recently for size estimates. They provide probability surveys that facilitate representative estimates. Survey data are used for reporting country-specific indicators to UNAIDS; for use in mathematical models, including spectrum estimates; funding requests; and to guide country-level program reviews. Survey instruments have varied by country but commonly include HIV prevalence, knowledge of HIV transmission routes, sexual behavior (including condom use), and information to estimate the size of populations. Knowledge of HIV status, ART status, and indicators of viral suppression are increasingly included in survey instruments.
Bio-behavioral surveys have strengths relative to health or program data to estimate HIV prevalence and the size of the HIV population. Properly designed and executed bio-behavioral surveys aim to obtain a probability sample of the population in contrast to health sector or program data, which have data only from clients using their services. Surveys can provide an independent evidence-based assessment of gaps in coverage for programs and health facilities and identify emerging epidemics [
Bio-behavioral surveys, however, have limitations. They are expensive and time-consuming to implement well, including effective engagement of stakeholders. Well-conducted surveys require formative research to guide protocol development, care in translation and back translation, ethical review by the appropriate organizations, interviewer training, ongoing monitoring of data quality, recording of deviations from the protocol, strategies to ensure data confidentiality and protection of participants, and strategies to provide participants with test results and linkage to care if indicated. Because HIV surveillance often takes place outside the health care system and involves contracting an outside implementing institution, there must be careful collaboration between the survey team and those in the health care system to ensure linkage to care for those who test positive as part of the survey.
Although survey design may minimize the effects of selection bias relative to clinical and program data, the effects of self-presentation bias on the validity of self-reported data may be considerable [
With support from the Global Fund and other donors, countries are using programmatic mapping to identify where to reach key populations and to estimate the size of key populations [
There are several opportunities for improving the value of surveys for program improvement: (1) aligning size estimates from surveys with program catchment areas; (2) characterizing those reached by the survey, but missed by programs; (3) using the Internet for recruitment of survey participants; (4) measuring the 90-90-90 cascade, including viral suppression; (5) characterizing subgroups [
For example, providing a cluster of differentiation 4 count at the time of the survey in Malawi to female sex workers living with HIV improved the acceptability of HIV testing and facilitated collecting an additional blood sample to estimate the proportion of female sex workers who had achieved viral suppression [
Routinely collected data from medical records and case-based surveillance systems have recently become a focus of development to facilitate measuring progress along the cascade. Where they exist, these data provide the number of persons on treatment. More sophisticated systems monitor progress along the cascade at the individual level from the first positive HIV test to viral suppression. Advocates argue that after HIV diagnosis, all cascade indicators recommended in the new Consolidated Guidelines for HIV Surveillance [
Challenges in using case-based surveillance and other analyses of medical records for measuring the cascade among key populations in resource-constrained countries are evident already. The system requirements for tracking individual medical records are not available in many settings. In addition to the usual issues associated with solving the problem of deduplicating reports from various sources (eg, interoperability of computer systems, lack of standardization across providers for reporting, lack of timeliness in reporting, and lack of a unique identifier, UID, protocol), a successful case-based surveillance system to monitor the cascade among key populations would require an indicator of key population status in the surveillance record. Inclusion of this indicator is problematic. People may not self-identify as a member of a key population or want the indicator on their medical record [
In addition, other factors pose challenges for the unique identification of key population members. It is not unusual for sex workers or gay men to adopt a second identity to hide their affiliation and present at different clinics based on identity. Defining membership in a key population may be differently interpreted across facilities. Membership in a key population may be quite dynamic, causing problems for interpretation of the cascade over time. Biometric UIDs such as fingerprint scans facilitate monitoring at the individual level across data sources but require careful introduction into the community, technical support, data protection schemes, ethical review, and ongoing monitoring.
Due to the stigma associated with being a member of a key population and the lack of a key population identifier in the record, it is likely that case-based surveillance systems will underestimate the size of key populations. Health sector data will overestimate prevalence if those who are infected are more likely to seek clinical care (Berkson’s bias) or if clinics with a higher prevalence of infection among patients are selected for inclusion in surveillance [
Some promising methods, however, are emerging to improve linkages between data sources when UIDs are not available, or an identifier fails to uniquely identify persons. One approach being piloted in the Dominican Republic is a follow-up survey of a sample of persons living with HIV in the treatment database to determine key population membership so that the cascade can be estimated for this subset of persons in the database. Improved probabilistic matching strategies based on available data such as name and birthday may be able to link an individual’s records across multiple programs when UIDs are not available. A MEDLINE search in January 2017 of articles related to probabilistic matching of medical records revealed 68 articles on the topic of probabilistic matching of medical records for HIV, of which 24 were published in the past 5 years. Free computer programs to improve deduplication of records are available, and efforts to evaluate the validity of probabilistic matching have been conducted [
Finally, new analytic tools are being developed to provide longitudinal measures of the cascade from treatment databases. These cascades indicate the time spent on the pathway from first positive HIV test to reaching viral suppression, the last 90 in the treatment cascade. The longitudinal HIV care and treatment cascade provides an estimate of the person time spent in each of the compartments of the HIV care continuum [
Since the early 1990s, there has been an acknowledgment that key populations are at greater risk of acquiring and transmitting HIV, are less likely to obtain services, and require specific services. Nongovernmental organizations or special outreach programs operating from government clinics may provide more acceptable and tailored services for key populations and may be more willing to engage key populations in target setting, advocacy, and addressing barriers to uptake of services. Some mature programs such as the Avahan sex worker interventions in India [
Some of the challenges of using program data revolve around the variable quality of program data, arising partially from the broad array of data collection strategies, training, and available support. The main challenge is that program data is not readily generalizable to the entire key population (selection bias), as those who do not visit programs are likely to be different than those who do. Latecomers to programs differ from the early volunteers; thus, the maturity of a program will affect the risk profile of its participants. Even an umbrella program with multiple service delivery sites that employs a UID may comprise an unstable cohort if there is significant mobility across programs and loss to follow-up. Recruitment, attrition, and reach are often not measured systematically within programs even if the capacity to do so theoretically exists [
Probability samples of national household surveys such as the Demographic and Health Surveys (DHS), the AIDS Indicator Surveys, and the newer Population-Based HIV Impact Assessment provide insight into the geographic distribution of HIV across a country but often fail to provide much insight into the HIV epidemic among key populations [
Going forward, however, there may be opportunities for greater use of national household survey data. A new incidence model drawing on DHS data estimates the distribution of new infections in a population for groups, including key populations [
We identified 14 publications where at least one of the 90-90-90 indicators was estimated. Three provided estimation of each indicator. The first was an RDS survey among MSM in Moscow, Russia using RDS [
Current reliance on active data collection processes, including key population–specific surveillance surveys, is warranted both to collect specific critical information that cannot be obtained from service or other program data and to provide a representative depiction of the HIV epidemic and response. Elements of the latter may be replaced in the future by passively collected routine program and medical record data for key populations. Even in mature programs with years of investment in reaching and treating key populations, national programs must leverage data from program data, medical records, and surveys, as any single source will be insufficient to understand the HIV epidemic, monitor care, and track progress in prevention and along the treatment cascade. Bio-behavioral surveys, although expensive, have proven successful in measuring gaps in program coverage that are not yet revealed by program or medical record data. Improving methods to estimate the treatment cascade from medical records and maintain data security and patient confidentiality will remain a high priority. Strategies to gain insight from multiple sources will require efforts to align geographic catchment areas, definitions, subgroups, and indicators.
There are several limitations of this analysis. Important events in the history of surveillance were omitted for brevity. An exhaustive account of the global history of HIV surveillance in the context of key populations is out of scope; however, an overview of the progress and challenges were important to provide some context for the WHO SI guidelines. Important issues regarding measurement of community engagement, data quality, data use, mobility, and the effects of interventions on HIV transmission could not be addressed sufficiently. We did not describe the many size estimation methods available or address issues regarding the validity of these size estimation methods. For transparency, we recommend reporting the methods used to estimate population size and construct denominators for 90-90-90 estimates [
Although most peer-reviewed studies identified restricted the analytic sample to highlight a specific key population group, these analyses often included details about additional high-risk behaviors, documenting overlap in population membership. Due to the different risk behaviors of key population subgroups, the overlap between and among groups, and the variety of legal restrictions across countries, the opportunities and challenges related to disaggregating and aligning surveillance data, programmatic data, and medical record data are considerable.
Currently, complete estimates of indicators for the size of key population groups and estimates for indicators of the 90-90-90 targets are not generally available. There has been a movement to increase the use of programmatic data to inform the HIV epidemic among key populations, although evidence of the quality and validity of estimates from these data are lacking [
Using programmatic data or medical record data for reporting disaggregated estimates of the 90-90-90 targets is particularly challenging when high-risk behaviors are overlapping in key populations, and programs address just one risk behavior. All individuals at risk of HIV acquisition and transmission may not identify as a member of the benefactor population [
UIDs may offer a solution to linking individuals across surveys, programs, and health care settings if the ethical, logistic, and technological challenges of implementing UIDs for marginalized and criminalized populations can be resolved. In many cases, especially when injection drug use, sex work, or homosexuality is illegal, providing details about high-risk behaviors and linking such information to a permanent medical record is a risk that leaves these already marginalized populations more vulnerable. WHO’s new case reporting and patient monitoring guidelines specifically do not include risk behaviors in the patient monitoring data forms because of the potential for harm to patients from stigmatized populations [
In conclusion, we recommend ongoing engagement with key population communities in the improvement and alignment of SI indicators across current data sources and exploration of new sources of data. The goal of SI is to improve the adequacy, acceptability, safety, and effectiveness of the public health response to the HIV epidemic among these populations.
antiretroviral therapy
Demographic and Health Surveys
men who have sex with men
respondent-driven sampling
strategic information
sexually transmitted infection
unlinked anonymous testing
unique identifier
Joint United Nations Programme on HIV/AIDS
World Health Organization
None declared.