Appendix 2.4: NHBS Rationale and Methods for Population Size Estimation

Size estimates of populations at risk for HIV/AIDS infection help policy makers and program administrators understand the scope of the HIV problem, plan appropriate interventions and allocate sufficient resources. Population size estimates can be used as denominators in calculating program coverage and producing national estimates and projections of the HIV epidemic.Population size estimation for Key Populations has been made a priority by WHO, UNAIDS and CDC GAP.

In the absence of a gold standard, such estimates are imprecise and prone to potential biases. Therefore, multiple approaches are used simultaneously to produce the most rigorous estimates of MSM population sizes. Which of the methods are feasible for each location can be determined during formative assessment (i.e. confirming with service providers that they keep unduplicated counts of participants). The use of multiple methods strengthens confidence in estimates, provides upper and lower plausibility bounds and reduces the likelihood that biases of any single method will substantially alter results.

Service Multipliers. The core NHBS survey already has part of some service multipliers built in. “Multiplier methods” entail two basic steps (UNAIDS, 2003);

1. Adding questions to the survey instrument asking about the use of specific services or facilities, membership in a group, or participation in a research study. (e.g. Where did you get your last HIV test? in the core NHBS survey).

2. Obtaining the unduplicated counts of the MSM using the above services or facilities, membership lists, or participating in a research study for a specific period of time.

Using these two data sources, the multiplier method provides a population size estimate by the formula:

N = n / p

Where N is the MSM population size, given by n as the number of MSM using a particular service in a specified time period and p as the proportion of MSM reporting using the particular service in the time period collected in the survey. For example, if a specific HIV test sites (AHP) reached 533 MSM in San Francisco in 2008 and 0.9% of survey respondents reported having an HIV test at AHP in 2008, then there are an estimated 59,222 (= 533 / 0.009) MSM in San Francisco. Variance for this estimate, and hence confidence interval for a population size estimate can be given by:

V(S) ≈ n12 * (1-m/n2) / { n2 * (m/n2)3} + n1(1/r)2

(M. Morgan, E. Kim, personal communication 10 February 2010).

Several multipliers are used simultaneously to minimize the potential influence of biases of any one multiplier and to produce upper and lower plausible bounds. The final choice of multipliers to use will be determined by the availability and quality of the unduplicated MSM count available in each location during the formative phase of the project. Of note, the program data need to have unduplicated individual contacts, be able to provide this number for a specific time period and ascertain MSM status among their clients for the estimate to be unbiased. Possible sources of service counts include:

A) MSM-friendly STI clinics (i.e., where staff received specific training in MSM STI care and treatment and routinely record MSM status of patients).

Example: Between April 2007 and March 2008 were you tested for an STD at City Clinic (located at 7th and Folsom)?

B) MSM-friendly HIV care and treatment clinics.

Example: In 2007 did you receive HIV related health care at Tom Waddell Clinic?

C) Programmatic data counting MSM reached for outreach education, VCT, and condom and lubrication distribution.

Example:In 2007 did you ever complete a survey for the STOP AIDS Project (either on the street in the community or at a STOP AIDS event)?

D) Other services or membership groups (e.g. gay rugby club) that can report an unduplicated membership number.

Example: Were you a member of Triangle Martial Arts between April 2007 and March 2008?

Where questions assessing service use for particular services do not already exist in the core NHBS survey, questions can be added to local use surveys.

Wisdom of the Crowds: This method is based on the assumption that the average response of a population on the number of members of a group approximates or proportional to the actual number in that population. Assumptions are that persons in a large sample have unique information about the population in question, that when asked individually their estimates are not influenced by others and that in aggregate the biases in estimates tend to cancel out. The method entails asking respondents how many MSM they estimate to be present in the particular location. Median, range and quartiles descriptive statistics can be calculated.

Example:How many men who identify as gay or bisexual or men who have sex with men but don’t identify and gay or bisexual do you believe live in San Francisco?

Modified Delphi: The modified Delphi begins with the synthesis of local and international expert opinion. A synthesis of data from an extensive review of the published and grey literature, searching for relevant data from similar cities and regions which can be used as benchmark estimates. During the analysis phase, the Modified Delphi method is concluded with a convening of stakeholders to synthesize the new information and estimates gathered during the study to revise earlier estimates.

References and further reading.

  • FHI. Estimating the size of populations at risk for HIV: issues and methods. Family Health International (FHI); 2003.
  • UNAIDS/WHO Working Group on HIV/AIDS/STI Surveillance. Estimating the Size of Populations at Risk for HIV: Issues and Methods. July 2003.
  • H. Fisher Raymond,Director, Bio-Behavioral Surveillance. HIV Epidemiology Section, AIDS Office, San Francisco Department of Public Health.