Application of the network scale‐up method to estimate the sizes of key populations for HIV in Singapore using online surveys

S Quaye, Yuwei Cheng, Rayner Kay Jin Tan, Joel Ruihan Koo, Kiesha Prem, Alvin Kuo Jing Teo, Alex R. CookJournal of the International AIDS Society 2023 · DOI

TL;DR

This paper adapts the network scale-up method (NSUM) to estimate the sizes of key populations at risk for HIV in Singapore using online surveys. The study successfully demonstrates that this approach is scalable and cost-effective, even in socially conservative environments, providing crucial data for public health interventions.

What Problem Does This Paper Address?

Estimating the size of "key populations" – groups disproportionately affected by HIV, such as men who have sex with men (MSM), female sex workers (FSW), their clients (MCFSW), people who inject drugs (PWID), and transgender individuals – is critical for effective HIV prevention and treatment strategies. However, these populations are often hidden, stigmatized, or criminalized, making traditional survey methods difficult and prone to bias. Singapore, despite having a low overall HIV prevalence, faces this challenge, with a significant proportion of new infections occurring within these key populations. Previous estimates for these groups in Singapore were imprecise due to small sample sizes. This study aims to provide more accurate and reliable estimates using an adapted NSUM approach.

Methods

The researchers employed a cross-sectional online survey utilizing a network scale-up instrument. The core of the NSUM method involves asking participants about the number of people they know within specific "reference" groups (e.g., people who attended a national parade) and "hidden" groups (the key populations of interest). By comparing the number of contacts reported in each group and considering the known sizes of the reference groups, the method can infer the sizes of the hidden populations.

Key methodological adaptations and features include:

  • Online Survey Platform: Leveraging online surveys and advertising for recruitment, which is crucial for reaching individuals who might be hesitant to participate in face-to-face interviews due to stigma.
  • Bayesian Hierarchical Model: A sophisticated statistical model was used to estimate population sizes, accounting for sampling weights and potential biases.
  • Network Scale-Up Instrument: An existing NSUM instrument, previously piloted in Singapore, was adapted. Participants were asked to estimate the number of people they knew within 10 reference groups and the 5 key populations.
  • Contact Definition: A "contact" was defined as someone known by name and sight, residing in Singapore, and with whom the participant had spoken at least once in the past year via phone or in person.
  • Post-hoc Sample Weighting: The sample was reweighted to better match the demographic profile of the general Singaporean population (age, sex, ethnicity, education).
  • Addressing Bias: The study incorporated adjustments for "barrier effects" (where certain groups are less likely to be included in the sampling frame) and "transmission error" (where individuals may not know the membership of their contacts in hidden groups).

Datasets

The study utilized data collected from a cross-sectional online survey conducted between March and May 2019.

  • Participant Data: 2802 participants completed the survey. Recruitment was done through an existing online panel (Singapore Population Health Studies Online Panel) and targeted Facebook advertisements.
  • Population Data: Singapore's Census of Population 2020 was used for reweighting the sample to match the general adult population's demographics.

Key Insights and Ideas

  • Scalability of Online Surveys for Hidden Populations: The study strongly suggests that online surveys are a highly scalable and efficient method for conducting size estimation studies of hidden populations, even in socially conservative societies like Singapore.
  • Cost-Effectiveness of NSUM: The NSUM approach, when combined with online recruitment, offers a cost-efficient way to gather data on multiple key populations simultaneously, reducing the need for separate, expensive studies for each group.
  • Stigma Mitigation: The anonymity and online nature of the survey helped mitigate stigma-related barriers to participation and data disclosure, which is crucial for accurate estimations.
  • Demographic Insights: The study provided detailed insights into the age and sex demographics of individuals within the key populations, revealing interesting patterns (e.g., older individuals knowing more MCFSW and transgender people, younger individuals knowing more MSM).
  • Social Acceptability: The study also explored participants' views on the social acceptability of behaviors associated with key populations, finding that transgender people were generally more accepted than homosexuality.

Key Results

The study estimated the sizes of five key populations in Singapore:

  • Male clients of female sex workers (MCFSW): 76,800 (95% CI: 64,200–91,800)
  • Men who have sex with men (MSM): 139,000 (95% CI: 120,000–160,000)
  • Female sex workers (FSW): 8,030 (95% CI: 3,980–16,200)
  • People who inject drugs (PWID): 3,470 (95% CI: 1,540–7,830)
  • Transgender people: 18,000 (95% CI: 14,000–23,200)

The study also found that:

  • Men generally reported knowing more people in all high-risk groups.
  • Older individuals reported knowing more MCFSW, FSW, and transgender people.
  • Younger individuals reported knowing more MSM.
  • A bimodal age effect was observed for knowing PWID, with individuals in their early 20s and 50s-60s reporting more contacts.
  • Ethnic differences were observed in contact patterns, with Malays reporting more MSM contacts and Indians and Malays reporting more transgender and PWID contacts.

Limitations and Future Work

The authors acknowledge several limitations:

  • Non-Resident Foreigners: The study could not distinguish between local residents, resident foreigners, and non-residents among contacts, limiting insights into the residency status of key populations.
  • Technological Literacy Bias: The online recruitment method might have inadvertently biased the sample towards individuals with higher technological literacy, although reweighting helped control for age imbalances.
  • Transmission Error Modeling: While adjustments were made for transmission error, the specific form of this inflation might not perfectly represent reality, and older individuals might still be less likely to disclose certain affiliations (e.g., being MSM).
  • Potential for Mis-specification: The estimates for MCFSW, transgender people, and MSM, where the adjustment for transmission error had the greatest effect, might be more prone to mis-specification.

Future work suggested includes:

  • Exploring other study design-based approaches, such as sampling directly from hidden populations if feasible.
  • Utilizing generalized scale-up estimators to overcome limitations of the basic NSUM model.
  • Further research into the nuances of transmission error and how it impacts different groups.

Summary updated

Abstract

This study demonstrates that a size estimation study of hidden populations is quickly and efficiently scalable through using online surveys in a socially conservative society, like Singapore, where key populations are stigmatized or criminalized. The approach may be suitable in other countries where stigma is prevalent and where barriers to surveillance and data collection are numerous.
BibTeX
@article{quaye2023application,
  author = {S Quaye and Yuwei Cheng and Rayner Kay Jin Tan and Joel Ruihan Koo and Kiesha Prem and Alvin Kuo Jing Teo and Alex R. Cook},
  title = {{Application of the network scale‐up method to estimate the sizes of key populations for HIV in Singapore using online surveys}},
  journal = {Journal of the International AIDS Society},
  year = {2023},
  doi = {10.1002/jia2.25973},
}
Figure