Ethical and regulatory considerations in HIV prevention studies employing respondent-driven sampling☆☆☆

Overview of RDS and its application 

Scientific advantages of RDS are illustrated by its contribution to sampling IDUs. Because IDUs are a hidden population, and because recruiting representative samples are important for drawing reliable and valid conclusions, RDS is useful in HIV studies. First, it is difficult to construct sampling frames that list names of IDUs for use in recruiting probability samples (Kish, 1991). Second, IDUs do not seek healthcare regularly, and many of them do not live in traditional dwellings, so they may not be represented in public data sets (Kalton, 1993). Door-to-door surveys, telephone surveys, and facility-based surveys tend to underestimate the number of IDUs, exclude them from their samples, or include biased samples (Catania, Gibson, Chitwood, & Coates, 1990; Gagnon et al., 1989). Although IDU samples recruited from venues or institutions may be more representative than household or general-population samples, they exclude IDUs who do not access these facilities. For example, drug treatment facility samples tend to include older IDUs with longer drug use histories. Third, obtaining traditional probability samples of IDUs is expensive, time-consuming, and resource intensive (Laumann, Michael, Gagnon, & Michaels, 1994; Sudman, Sirken, & Cowan, 1988). Fourth, because HIV research necessitates collection of data about illegal drug use and intimate sexual behaviours, developing rapport with IDUs through community-based relationships and participant-driven sampling is necessary to obtain valid responses and high response rates (Magnani, Sabin, Saidel, & Heckathorn, 2005). Finally, RDS allows investigators to overcome many of the sampling biases observed in other methods of chain-referral sampling (Heckathorn, 2002, Salganik, 2006). Thus, for scientific and practical reasons, RDS is often considered superior to other IDU recruitment methods. The resulting samples include IDUs who might not have been included using venue-based, street outreach, or advertising approaches.

Operationally, RDS uses a highly structured referral process referred to as participant-driven recruitment (PDR). Studies that use RDS depend on participants to nominate and refer or recruit their peers for study participation. Referral quotas limit the number of recruits a recruiter can invite to the study, and limit remuneration for recruitment. Recruiters receive a pre-determined number of coupons, each with a unique serial number, to refer people they know and who are likely to meet the study's eligibility criteria. Recruiters give the coupons to potential recruits who agree to consider participation and to bring their coupon to study staff. Recruiters can invite friends, acquaintances, sex partners, or drug-injection partners to participate in the study. Staff members explain the study to the recruits, screen them, and (if they are eligible) invite them to participate in the study. The consent process is designed to ensure that recruits do not feel pressured by the recruiter. If recruits choose to enrol, the recruiters are reimbursed for their time and effort. Both recruiters and enroled recruits receive remuneration for study participation.

In terms of the recruitment process, RDS and snowball sampling operate somewhat similarly. Both methods begin with a set of recruiters who are chosen by study staff in a non-random fashion. The initial recruiters, referred to as seeds, recruit the first “wave,” who in turn recruits the second wave, who then recruits the third wave, and so forth, until the sampling goal is reached. Thus, the sample grows wave by wave, similar to the way a snowball increases in size as it rolls down a hill (Heckathorn, 2002). In RDS, the number of waves needed to obtain a representative sample of the target population is usually calculated in advance of recruitment. Four to six waves are recommended to produce a sample that is independent of the initial seeds (Heckathorn, 2002). The relationship between recruiters and recruits is documented in the coupon-management system to assess the participant-driven recruitment process and selection biases. In contrast to snowball sampling, RDS offers participants a higher level of privacy and confidentiality because recruited peers choose whether they would like to come to the study site or not. In snowball sampling, enroled participants often give investigators the name of their peers who wait on investigators to find them.

RDS differs from snowball sampling statistically in that RDS is based on a mathematical model of the network-recruitment process that functions like a corrective lens, controlling the distorting effects that network structure exerts on the sampling process (Heckathorn, 2007). In this way, RDS provides information on relative inclusion probabilities, from which unbiased estimates of population characteristics are calculated. If mathematical model assumptions are satisfied, RDS yields population estimates that are asymptotically unbiased (Heckathorn, 2007, Salganik, 2006).

The assumptions of the mathematical model influence the recruitment process. First, long recruitment chains eliminate bias related to the selection of the seeds. Thus, a distinctive feature of RDS is two interrelated means of cultivating long chains achieved by enhancing participants’ motivation to recruit their peers, and by using recruitment quotas (generally set at three recruits per recruiter). Thus, achieving any given sample size will require a substantial number of waves. Recruitment quotas are consequential from an ethical and regulatory perspective because they limit the amount of money any recruiter can receive; hence, precluding problems arising from excessive payments. Remuneration for participant-driven recruitment is a key aspect of RDS and is not contingent on the target population.

A second constraint on the sampling process is the need to document who recruited whom. As Galton recognized when he formulated the homophily principle, friends tend to resemble one another in race, ethnicity, education, income, religion, and other attributes (Galton, 1874). In studies of IDUs, homophily also includes drug preference and HIV status (Heckathorn, 2002, Heckathorn et al., 2002). Mathematical modelling controls for the non-random recruitment bias by quantitatively assessing the degree to which recruiters’ and recruits’ characteristics are correlated. Thus, network data are collected on network relationships and characteristics to assess homophily and to permit the calculation of selection probabilities and population estimates (Heckathorn, 1997, Heckathorn, 2002). Fortunately, information on potential network size (number of acquaintances, friends, or partners who qualify for participation) does not pose privacy-related constraints. Investigators use the data on network size to control for differential network size and to assess the reach of the recruitment process. Investigators use the information on the relationship between recruiters and recruits (e.g. sex partner, spouse, drug-injection partner, relative, friend, acquaintance, stranger, or other) to control for differences between participants, weight the sample, and produce probability estimates. The implication of requiring documentation of the recruiter–recruit relationships and characteristics is that when HIV information is obtained study investigators will also gather information on discordant partnerships.

In many other network-based studies, investigators can destroy the linking code and de-identify the data after the social network has been mapped and sampling weights have been calculated. Data analysis can then be carried out using de-identified data (Klovdahl, 2005, Margolis, 2000). However, this option may not be ethically justifiable or permissible by institutional review boards (IRBs) or ethics review committees (ERCs) for studies that use RDS if investigators had access to information on discordant partnerships during the study period. Thus, deciding on how to disclose information on discordant partnerships to participants is important because this information is clinically relevant to their health.

Ethical and regulatory considerations 

Ethical considerations and regulatory requirements regarding remuneration for participant-driven recruitment and for informing participants of their discordant partnerships are relevant to all studies employing RDS. These topics gain importance as use of RDS gains momentum. Since information on RDS first appeared (Heckathorn, 1997, Heckathorn, 2002), and since its initial application to recruit IDUs (Broadhead et al., 1998, Heckathorn et al., 2002), researchers have used RDS for its practical and statistical advantages to recruit other populations at risk for HIV infection. RDS has been employed for research and surveillance purposes in the U.S. and more than 30 other countries (Frost et al., 2006, Heckathorn et al., 2007, Platt et al., 2006, Simic et al., 2006, Stormer et al., 2006; Yeka, Maibani-Michie, Prybylski, & Colby, 2006). For example, RDS was used in the U.S. to recruit heroin and cocaine users (Broadhead et al., 1998, Heckathorn et al., 2002), MDMA (ecstasy) users (Wang et al., 2005), transgender persons (Clements-Nolle, Marx, Guzman, & Katz, 2001), and MSM (Ramirez-Valles, Heckathorn, Vazquez, Diaz, & Campbell, 2005). Our review and example of studies that used RDS with IDUs in the U.S. show that the confluence of the ethical and regulatory considerations and the scientific literature can help contribute to the development of evidence-based ethics (Halpern, 2005).

With expanding worldwide use of RDS, knowledge of relevant guidelines and regulations is increasingly important. For example, the U.S. federal regulations require all U.S. federally funded research to be reviewed by an IRB. When U.S. federally funded research is conducted outside the U.S., the regulations require that the research protocols be reviewed in both the U.S. and the host country. Studies conducted outside the U.S. should follow the regulations and codes of the host country and relevant literature (Akabayashi & Slingsby, 2006; Hearnshaw, 2004; Rivera, Borasky, Rice, Carayon, & Wong, 2007). Additionally, societal factors can influence the ethics of studies that use RDS. In the case of drug users, these factors include mistrust of authorities, density of networks, mobility of drug users, interaction between IDUs and other populations, female-to-male ratio of drug users, stigma, partner violence, and cultural expectations about privacy, confidentiality, and disclosure of HIV discordant results (Des Jarlais & Semaan, 2005; Klein, 2001; Paxton, 2002, Semaan and Leinhos, 2007; Warszawski & Meyer, 2002).

During the past decade, controversies have emerged over the applicability of regulatory and ethical considerations of resource-rich countries, particularly those of the U.S., to resource-limited countries (Weijer & Anderson, 2001). Subsequently, different ethical frameworks and benchmarks were published on how to resolve such controversies in clinical research, which can be useful for behavioural research, including for remuneration for participant-driven recruitment and for investigators’ responsibilities for informing participants of their discordant partnerships. The benchmarks include those that relate to social value, scientific validity, fair selection of study population, favourable risk-benefit ratio, independent review, informed consent, and respect for recruited participants and study communities (Emanuel, Wendler, Killen, & Grady, 2004; Fitzgerald, Wasunna, & Pape, 2003).

Epidemiologic and intervention studies are classified as research for regulatory purposes, and are subject to specific regulations and policies. Conversely, surveillance activities abide by different policies that pertain to public health activities (Centers for Disease Control and Prevention, 1999, Snider and Stroup, 1997). More recently, several calls have been made to reconsider the policies of surveillance activities (Fairchild, 2005, World Health Organization, 2004). This emphasises the need to inform participants of the purpose of surveillance activities and to share relevant test results with them. More specifically, the calls question the ethics of the practical and economic reasons that have traditionally been used to argue against notification of HIV test results in surveillance activities (Fairchild & Bayer, 2004; World Bank Group, 2002). Project staff training and sufficient resources can increase the feasibility of notification, but are not always available, in which case alternative strategies for notification might be warranted.

Safeguards used in RDS 

Because of the safeguards in studies involving RDS, remuneration for time and cost for participant-driven recruitment is generally accepted as non-coercive, does not cause undue inducement, is not excessive, and does not compromise the voluntariness necessary for informed consent. Coercion typically occurs when the threat of a specific harm compels a person to act in a manner that they would not otherwise choose (Hawkins & Emanuel, 2005). It can be argued that there is some risk for coercion in studies involving RDS if recruiters intimidate recruits to join the study. Undue influence is defined as an offer that leads people to use poor judgment and assume substantial risks of harm that compromise their welfare (Emanuel, 2004, Wilkinson and Moore, 1997). Undue influence could occur when recruiters overzealously encourage their peers to consider study participation while thinking more of their own remuneration than the welfare of their peers. Another important question is whether remuneration payments for participant-driven recruitment pose any risk of harm that is beyond what the participants experience in the course of their daily lives.

Because of these concerns, studies that use RDS adopt several safeguards to protect participants and ensure that coercion or undue influence does not impair the voluntary decision of potential enrolees to participate in a study or their assessment of the risks and benefits of participation. The safeguards include giving recruiters a maximum of three coupons to minimize the recruiters’ influence on the study sample, and to limit the emergence of quasi-professional recruiters (Heckathorn et al., 2002). Recruits’ informed consent is obtained by study staff. Remuneration for participant-driven recruitment is modest, even for those who recruit the maximum of three recruits. Furthermore, investigators do not mention payments in recruitment advertisements or coupons. As additional safeguards, recruiters have to meet with study staff to learn if their potential recruits were enroled in the study. Once study staff validates recruitment information in the coupon management system, recruiters are remunerated for their efforts. Only recruiters are allowed to redeem their non-transferable coupons for participant-driven recruitment. Investigators can choose to conduct interviews with potential participants about the recruitment process to assess relevant concerns. If the risk of harm from recruitment or participation is identified during the interview process, study staff can choose to withhold recruitment or remuneration.

Remuneration for participant-driven recruitment is different from finders’ fees. Finders’ fees, which were used for remunerating physicians for recruiting their patients into clinical studies, are disallowed by IRBs because they might have influenced the judgment of health care providers about the appropriateness of research participation for their patients (Lemmens & Miller, 2003). Recruiters and recruits in studies that use RDS do not share a clinical or consulting relationship; thus, they do not have a financial or professional conflict of interest. Studies that have used RDS with IDUs in the U.S. have been behavioural studies that evaluated prevention interventions or collected behavioural surveillance data, and have not offered medical treatment (Abdul-Quader, Heckathorn, Sabin, & Saidel, 2006; Gallagher, Denning, Allen, Nakashima, & Sullivan, 2007; Lansky et al., 2007). Behavioural surveillance activities and prevention research studies are usually classified as minimal-risk, in which the probability and magnitude of harm or discomfort associated with participation is not greater than those ordinarily encountered in daily life; or during the performance of routine physical or psychological examinations (Office of Protection from Research Risk, 1991).

Characteristics of studies and participants 

We located and reviewed all the articles employing RDS with IDUs in the U.S. Our strategy provided us with eight articles describing four studies (two studies were described in more than one article) (Abdul-Quader, Heckathorn, McKnight, et al., 2006; Broadhead, Heckathorn, Grund, Stern, & Anthony, 1995; Broadhead et al., 1998, Des Jarlais et al., 2007, Heckathorn et al., 2002, Hughes, 1999, McKnight et al., 2006, Robinson et al., 2006) (Table 1). Two earlier studies, published during 1995–2002, used an evaluation design (Broadhead et al., 1998, Heckathorn et al., 2002). Two more recent studies, published in 2006 and 2007, used a cross-sectional design (Abdul-Quader, Heckathorn, McKnight, et al., 2006; Des Jarlais et al., 2007, McKnight et al., 2006, Robinson et al., 2006), and were part of a behavioural surveillance activity (Lansky et al., 2007). The development of RDS (Heckathorn, 1997, Heckathorn, 2002), and its initial application (Heckathorn et al., 2002), and more recent application (Des Jarlais et al., 2007) with IDUs show a core group of researchers who have expertise in research with IDUs and an interest in all aspects of RDS—scientific, ethical, and regulatory.

Table 1. U.S.-based studies that used RDS with IDUs, 1995–2006

	Author et al. (journal, year)	Purpose of study	Study design	Period of data collection and study location (city, state)	Eligibility criteria	Sample size, Age, %male	Race and ethnicity	HIV test and serostatus	Remuneration for participant-driven recruitment	Remuneration for study participation	Efforts to assess or prevent coercion
	Broadhead et al., 1995, Broadhead et al., 1998, Hughes (1999)	To compare outreach-driven intervention with peer-driven intervention	Pre-post intervention	1994–1998; New London, Middletown, and Windham, CT	NR	552 IDUs, Mean age=37, %male: 69	Peer-driven—B: 11%, W: 64%, H: 22%, O: 3%; Out-reach-driven: B: 12%, W: 45%, H: 40%, O: 3%	HIV test offered; %positive—Recruiters: 7%; Non-recruiters: 14%	$10 for each recruit	$20 for initial interview; $30 for follow-up interview; bus fares	A maximum of three coupons; asked if coerced to come to study site
	Heckathorn et al. (AIDS and Behavior, 2002)	To evaluate role of steering incentives in recruiting 18–25-year-old IDUs	Interrupted time-series design	1995–1998; Middletown and Meridan, CT	Injected drugs last 30 days, ≥18 years old	386 IDUs, 31% 18–25 years, %male: 72	B: 11%, W: 50%, H: 38%, O: 1%	NR	$10 for each recruit ≥26 years old; $20 for each younger recruit 18–25 years old	$20 for older interviewee; $30 for younger interviewee	A maximum of three coupons; asked if coerced to come to study site; recruits decide whether to participate
	Abdul-Quader, Heckathorn, McKnight, et al. (JUH, 2006), McKnight et al. (JUH, 2006) and Des Jarlais et al. (AIDS, 2007)	To assess prevalence of risk behaviours and seroprevelance and effectiveness of RDS in recruiting a diversified group of drug users	Cross-sectional	2004; New York, NY	Injected, smoked, or snorted an illicit drug, other than Cannabis last 6 months, ≥18 years, spoke adequate English for consent and interview	618 DU*, Mean age=44, %male: 76	B: 46%, W: 14%, H; 35%, O: 5%	HIV test offered; %positive—current injectors: 15%; never injectors: 17%	$10 for each recruit	$20 for interview	Consent by study staff; a maximum of three coupons; training on recruiting peers; hard to duplicate coupons; Information and characteristics on refusals; only recruiter gets remuneration
	Robinson et al. (JUH, 2006)	To compare and contrast RDS and targeted sampling methods on sample characteristics, recruitment efficiency, and logistics concerns	Cross-sectional	2004; Detroit, MI; Houston, TX; New Orleans, LA	Injected illicit drug past 12 months, ≥18 years, resident of area	Detroit: 102 IDUs, 2%: 18–34 years, %male: 71; Houston: 48 IDUs, 20%: 18–34 years, %male: 77; New Orleans: 118 IDUs, 29%: 18–34 years, %male: 81	Detroit—B: 96%, W: 1%, H: 1%, O: 2%; Houston—B: 45%, W: 23%, H: 32%; New Orleans—B: 58%, W: 29%, H: 2%, O: 11%	NR	$10 for each recruit	$20 for interview; New Orleans: transportation tokens	A maximum of three coupons; non-replicable coupons; only recruiter gets remuneration

RDS: Respondent-driven sampling; JDI: Journal of Drug Issues; PHR: Public Health Reports; JUH: Journal of Urban Health. IDU: injection-drug user; *DU: drug users. 43% current injectors (≤6 months); 19% former injectors (>6 months); 38% never injectors. B: Black; W: White; H: Hispanic; O: Other. M=mean. NA: not applicable; NR: not reported.

Regarding eligibility for study participation, three of the four studies required participants to be at least 18 years old (Table 1), and three studies required eligible participants to have used drugs for a specified number of months. Three studies were conducted in the Northeast, and one was conducted in the Midwest and South. The number of study participants ranged from 48 to 618. Although the method for reporting age was not consistent across studies, the mean participant age was mid-to-late 30s. In all four studies, over two-thirds of participants were male. The samples were racially and ethnically diverse. HIV testing was offered in two studies. Prevalence of HIV ranged from 7 to 17%.

Remuneration and efforts to assess and prevent coercion 

In terms of remuneration strategies (Table 1), peer recruiters received $10 for each recruit, for a maximum of $30. Recruiters and recruits received a range of $20 to $30 for participating. Two studies offered bus fares or tokens to their participants. The most commonly reported methods for assessing or preventing coercion were limiting the number of coupons to three, asking the recruits whether they had been coerced into coming to the study site, and having study staff emphasize that study participation was voluntary (Table 1).

Remuneration for participant-driven recruitment 

Several arguments and topics are relevant to the justification for remuneration for participant-driven recruitment (Table 2). The topics are: (1) respect for the participant, (2) sociological and psychological obligations of reciprocity, (3) participants’ motivations, (4) categorisation of research as a business, and (5) reimbursement and compensation payments.

Table 2. Ethical arguments for and against remuneration for participant-driven recruitment and relevant safeguards

	Ethical arguments for remuneration
	Respect	Responding to a recruiter request to visit a study site does not constitute coercion
		Financial or consulting relationship does not exist between recruiter and recruits
		Motivation for study participation is multidimensional and does not seem to be different for drug users than for those who do not use drugs
		Remuneration for time and effort shows respect and does not distort one's decision about visiting a study site or participating in a study, and is not associated with coercion
		Monetary payments show respect for participant's judgment to meet personal needs
	Beneficence	RDS provides reliable population estimates to support HIV prevention and control efforts
		Altruism and advancement of public health are not sufficient reasons to encourage recruitment
		Remuneration eliminates psychological obstacles and time requirements
		Remuneration for participant-driven recruitment is cost effective
		Non-monetary payments are not as effective as monetary payments in motivating study participation
	Justice	Environmental, economic, and societal conditions that lead to addiction constitute a stronger form of coercion
		Remuneration for participant-driven recruitment forms a business element in study budgets
		Remuneration for participant-driven recruitment is fair given recognition achieved by investigators and benefits to public health
	Ethical arguments against remuneration
	Respect	Concern about bartering of coupons or reimbursement payments, or coercion or undue influence to do so
	Beneficence	Concern about use of remuneration payments to buy drugs
		Concern about potential of remuneration payments to subvert altruistic motivations for study participation
	Safeguards	Informed consent with sober drug users emphasizes voluntary participation
		Remuneration is spelled out in informed consent and is modest even for referring a maximum of three recruited peers
		Study staff interview recruiters and recruits about participant-driven recruitment process
		Only recruiters can redeem their non-transferable remuneration

Disclosure of information on HIV discordant partnerships 

Investigators of studies that use RDS and who plan to offer HIV testing or to collect information on HIV status through surveys need to consider the arguments and the options they could use to inform participants of their HIV discordant partnerships (Table 3). Some studies of HIV transmission and prevention have been criticized for failing to notify uninfected individuals of their sex partner's infection. Our discussion focuses on ethics-related considerations, ethical responsibilities, timing of consent, options for consent, and data on disclosure and notification preferences.

Table 3. Ethical arguments for and against disclosure to participants of their HIV discordant partnerships and relevant safeguards

	Ethical arguments for disclosure
	Respect	Highlight investigators’ obligation for the public health of networks and communities
		Inform participants during informed consent:
		of investigators’ potential knowledge of discordant partnerships
		of the options they have for learning about their discordant partnerships
		of/if sharing positive HIV test results with local health departments
		of/if sharing positive HIV test results with their health care providers
		Offer participants a graduated or sequential consent to learn about discordant HIV partnerships at the time participants are informed of their own test results
		Emphasize protection of participants’ confidentiality and adherence to their choices
	Beneficence	Learning that investigators will know about their HIV discordant partnerships can influence decision on benefits and risks of study participation
		Offer participants training and support in discussing serostatus with primary, secondary, and casual partners
		Offer participants counselling on risk reduction with primary, secondary, and casual partners
	Ethical arguments against disclosure
	Respect	Use de-identified data
		Withholding information about discordant partnerships shows respect for participants’ privacy
	Beneficence	Protection of existing relationships
		Prevention of abuse or breakdown in relationships
	Justice	Role of investigators is different from role of health care providers
		Reporting to state and local health departments is ethically or legally sufficient
		Reliance on HIV partner counselling and referral services is ethically or legally sufficient
	Safeguards	Conduct the analysis using de-identified data after destroying the linking code, subsequent to mapping the social network structure and calculation of sampling weights

Reporting checklist for RDS studies 

Although information about procedures affecting the rights and welfare of recruiters and recruits in studies that use RDS is common in protocols submitted to IRBs or ERCs, limited data have been published in peer-reviewed publications. The limited reporting is influenced by space limitations in journals and by the modest attention to the importance of such information, and has led, in general, to calls for development of evidence-based ethics (Halpern, 2005, Miller and Rosenstein, 2002). Investigators have accepted the importance of reporting statistical methods for the interpretation of scientific results and for advancing the field of meta-analysis (Cooper & Hedges, 1994; Johnson et al., 2002). Similarly, investigators and editors can accept the importance of reporting information on procedures that protect the rights and welfare of study participants.

We suggest that investigators consider the various formats for reporting ethical and regulatory considerations that affect studies that use RDS. These formats can include (1) a subheading labelled “ethical and regulatory considerations,” under the methods section, similar to the commonly used subheading labelled “statistical methods,” (2) topical articles in peer-reviewed journals, or (3) web-based reports. Based on our results, there is a need to collect and report data, as feasible, on variables related to the process of participant-driven recruitment and to the options for informing participants of their discordant partnerships. We propose a list of variables that can be included, as feasible, in studies employing RDS. Variables related to remuneration for participant-driven recruitment include (1) motivation for recruitment, (2) motivation for study participation, (3) assessment of coercion or undue influence, (4) amount of payment for each recruit, (5) amount of payment for study participation, (6) methods used to determine remuneration payments, (7) return rates to collect remuneration payments, (8) mode of spending remuneration payments, and (9) differential effects of monetary versus non-monetary payments. Variables related to disclosure of discordant partnerships include (1) procedures used to protect privacy and personal information, (2) disclosure options provided to participants, (3) procedures used to protect relationships, and (4) beliefs and behaviours of participants regarding their responsibility for disclosure of HIV status and adoption of safer behaviours.