A scoping literature review [13] was conducted to gain insights into the application and performance of web-based RDS. We chose to conduct a scoping review because our aims were primarily exploratory, in the sense that we intended to provide an overview of the work done with web-based RDS so far and to identify factors related to the recruitment performance of web-based RDS. Table 1 provides an overview of the web-based RDS terminology used in this review (partially adapted from [14]).

We searched PubMed, Web of Science, and Scopus for articles. First, a preliminary search was conducted in PubMed to gauge the quality and quantity of web-based RDS related articles and to identify keywords to formulate the search syntaxes.

The term web-based RDS was introduced in 2008. In order to ensure that potentially relevant articles from before the term was introduced were included, we set our search range as 2000 to 2019. The following search terms were included in the final search syntaxes (see Multimedia Appendix 1 for the full syntaxes used):

Two researchers (YH and MS) independently screened the titles and abstracts of all unique records identified. The full text of selected records were then screened by one researcher (YH), to apply the below eligibility criteria. After this, the remaining articles were critically reviewed by YH, MS, and NH, before being included.

We included peer-reviewed articles that described the use of web-based RDS for the recruitment of participants for research purposes (ie, for making population estimates or for studying social networks) or for health intervention delivery. Articles that at least reported the numbers of seeds, subsequent recruitees, and discussed barriers and facilitators to the application of web-based RDS were included. We excluded studies that combined online and offline RDS without reporting on both approaches separately; if a study reported online and offline RDS separately, the online component was included. As we meant to provide an overview of the applications and performance of web-based RDS, our study was not limited to any particular target population or geographical area. We excluded studies that were not available in English or Dutch.

A data extraction table was developed to collect and organize data. The table’s topics were iteratively identified and selected based on (1) STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) RDS guidelines [15], (2) topics discussed in a literature review with a similar purpose and context (offline RDS) to this study [1], and (3) discussion between YH, MS, and NH. Additional topics included study design, main findings, recommendations for further research, and limitations.

We used the process evaluation of complex interventions framework [16] to analyze the application and recruitment performance of web-based RDS. This framework explains the outcomes of an intervention as a function of implementation characteristics, mechanisms of impact, and contextual factors. We adapted this framework to fit web-based RDS specifications. In this study, we viewed web-based RDS as an intervention with the purpose of recruiting individuals. We defined outcomes as web-based RDS recruitment performance, implementation characteristics as the seed selection and recruitment process, mechanisms of impact as mechanisms to stimulate peer recruitment, and context as the setting in which web-based RDS was conducted.

Figure 1 (adapted from [16]) shows the analytical framework integrated with a schematic representation of the web-based RDS recruitment process. Table 2 shows the extraction table’s topics, and operationalized measures thereof, grouped by components of the process evaluation framework.

Analyses focused on uncovering factors that influenced recruitment performance, based on comparing implementation characteristics, peer recruitment and recruitment performance measures between included studies. Data were presented in a narrative fashion.

No ethical issues were foreseen in this study.

We identified 393 unique records. The final number of articles included in this review was 18. See Figure 2 for a detailed account of the study inclusion procedure.

Six studies took place in the United States [8,17-21], 6 studies took place in Western Europe [22-27], 4 studies took place in southeast Asia [23,28-31], and 2 studies took place in Oceania [32,33]. Studies were published between 2008 and 2019, with the majority [17,18,22-24,26,27,30,32,33] from 2015 onward (see Table 3).

In 12 studies, participants were recruited with the aim of generating population estimates [8,19-23,26,28,29,31-33]; 3 studies aimed to study social networks and contact patterns relevant to the spread of infectious diseases [24,25,28], and 3 studies recruited participants for delivering interventions [17,18,27].

Studies focused on a diverse set of target populations: the general population [19,24,25,28,30,32], university students [8,20], men who have sex with men [23,29,31], individuals who smoke [17], individuals using wheelchairs [33], parents of children aged 10 to 14 [18], people with precarious employment [22], young adults at risk of chlamydia infection [27], individuals who have migrated from Syria [26], and individuals who use marijuana [21].

This is the first review investigating the application and recruitment performance of web-based RDS; a novel online sampling method. We identified 18 articles that described the use of web-based RDS. Out of all studies, 12 recruited participants for making population estimates, 3 recruited participants to study social network characteristics (contact patterns relevant to the spread of infectious diseases), and 3 recruited participants to deliver interventions. Studies were conducted in 8 countries, including both high- and low-middle income countries, over 4 continents. Between 19 and 3448 participants were recruited from various populations, including some without a sampling frame, such as men who have sex with men. The heterogenous nature of the included studies (with respect to their aims and setup) made it difficult to compare their recruitment processes and to draw generalizable inferences regarding recruitment performance.

We found that studies that recruited relatively more seeds, through online platforms, and with less rigorous selection procedures reported lower percentages of successfully recruiting seeds. The exact reasons for this observation remain unclear. However, we suggest that recruiting more seeds relatively limits the time and resources available to thoroughly prepare (ie, motivate and inform) and select seeds. In turn, this may limit seeds’ motivation to initiate peer recruitment, or lead to less suited (eg, less socially connected) seeds being selected. Both of these factors are known to be important for inducing and sustaining seed and peer recruitment [34,35].

Studies that did not offer at least one guaranteed material incentive reached relatively lower percentages of successfully recruiting participant, and fewer waves. We thus suggest that such incentives are particularly important to sustain recruitment, as sampling waves increase (ie, monetary incentives appear to carry further than nonmonetary incentives). This is in agreement with wider offline RDS literature [34] and indicates that benefits of online recruitment for participants (eg, easy access for participants) and nonmonetary incentives do not render material incentives redundant if the primary aim is to generate recruitment waves.

However, some studies that recruited participants for interventions reported relatively low percentages of successfully recruiting participants, despite offering substantial monetary incentives. This indicates that online peer recruitment for interventions benefits (or suffers) from factors other than peer recruitment for research purposes. Potentially, peer recruitment for interventions depends more on participants’ affinity toward an intervention (eg, related to intervention framing, packaging), or its anticipated or experienced outcomes. Note, however, that these findings are based only on few studies and require further research to substantiate.

The majority of studies took between 2 and 3 months to collect data. However, online recruitment was relatively faster in some studies. For example, one study [8] achieved their targeted sample size (N=150) in 72 hours, and another [19] recruited 3448 participants in 6 weeks. Both these study populations were composed of university students who may be particularly digitally literate and have extensive well-connected online networks. This finding indicates the importance of these factors when considering applying this online method.

One strength of this literature review was the wide search strategy. It provides an extensive overview of peer reviewed literature relevant for investigating web-based RDS peer-recruitment processes. Another strength was the application of the process evaluation framework, which offered a practical structure for investigating different factors influencing web-based RDS recruitment performance.

One limitation is that we excluded all articles not reporting on the recruitment process in sufficient detail and all studies not exclusively using online peer recruitment or reporting on online and offline peer recruitment separately. Some valuable contextual and comparative information between online and offline recruitment might therefore have been missed. For some crucial recruitment performance measures (eg, the percentage of successfully recruiting participants), we had to rely on a manual count of recruitment trees, since the original data sets were unavailable.

Based on this review, it remains difficult to assess how successful web-based RDS is at achieving the aims for which it is employed (ie, generating population estimates, studying social networks, delivering interventions). For example, only 5 out of 12 studies aiming to generate population estimates reported that the sample composition reached equilibrium. Several studies likely achieved equilibrium (estimated from reported sample size and observed number of waves), but did not report this as such. Studies that used web-based RDS for studying social networks or delivering interventions were mostly feasibility or implementation studies, making it difficult to assess how successful the online method is at reaching the endpoints. At this point in time, we believe that there are not enough studies to draw meaningful conclusions regarding the overall success of web-based RDS for generating population estimates, studying social networks, or delivering interventions. Nevertheless, web-based RDS may be a particularly suited recruitment method when random sampling techniques are not feasible, the target population is geographically dispersed or hidden (which is a challenge for offline sampling), and the target population is extensively connected online [8]. Therefore, despite the heterogenous nature of the studies included in this review, which limits the generalizability of the studies’ recruitment processes and performance, we outline several recommendations for future research into, or using web-based RDS.

First, consistent with offline RDS literature, the results suggest that recruiting a relatively small and thoroughly selected group of seeds (to whom a significant amount of resources can be dedicated for motivational and informing purposes) and providing at least one guaranteed material incentive is the most successful strategy for generating a substantial number of waves [1,34]. As such, this is the preferred setup for studies aiming to reach equilibrium for population estimates. If this is not the primary objective, for example when recruiting individuals for studying network characteristics or delivering interventions, recruiting a larger number of seeds through less rigorous means and providing lower or nonmaterial incentives may be preferred.

Second, despite the limited number of studies that recruited participants for interventions in this review, some implications in this regard stand out. For example, one study [27] found that through web-based RDS, individuals could be reached for sexually transmitted disease testing who were not reached before through traditional sexual health services. Another study [17] similarly noted that with each successive wave, the proportion of not-ready-to-quit smokers in the sample increased. These findings indicate that web-based RDS recruitment is particularly interesting for interventions if the aim is to reach more reluctant, or previously unreached individuals. The challenge here is to adequately incentivize peer recruitment (as discussed under principal findings).

Third, the results indicate that providing multiple recruitment options and facilitating the use of mobile platforms for participation and recruitment may enhance web-based RDS recruitment performance. However, it remains largely unclear how these factors influence peer recruitment across different settings, or even within certain target populations. For example, as indicated also by several studies included in this review, differential access to mobile communications, or the internet in general, may impose barriers to peer recruitment to readily excluded members of a given population. In addition, online communication behavior and the types of digital communication platforms used may differ between different networks, which could affect even relatively well-connected individuals. This is exemplified by one study [32] included in this review that compared web-based RDS to traditional offline RDS and found that individuals with lower socioeconomic status were less likely to be recruited through web-based RDS. Besides socioeconomic status, other factors known from literature that influence access to or use of the internet (and may therefore also influence online peer recruitment) include sociodemographic (eg, age, gender), socioeconomic (eg, household income, educational attainment), social (eg, degree of isolation, political context), and personal (eg, self-efficacy, mental health) factors [36]. To account for these potential sources of bias, we suggest thorough exploration of the target population’s online networks and communication behaviors, in a formative research stage.

Finally, we recommend that researchers using web-based RDS follow STROBE-RDS guidelines when reporting their studies [15]. A number of studies did not consistently report the numbers of total distributed and returned coupons, the numbers of recruitment waves, the numbers of individuals collecting their incentives, and the numbers of recruitees by seeds. Similar gaps in reporting on offline RDS data have been found in a previous review [37]. In addition, we encourage researchers to report relevant recruitment performance measures, such as the percentages of successful recruiters or the average numbers of recruitees per participant. This information is crucial for studying how to optimize peer recruitment in the future.

We have given a comprehensive overview of web-based RDS, what it is used for, how it is applied, and what factors influence its recruitment performance. Web-based RDS can be successfully applied to recruit individuals for making population estimates, studying social networks, and delivering health interventions. Peer recruitment may be enhanced by rigorously selecting and motivating seeds, offering at least one guaranteed material incentive, and facilitating adequate recruitment options regarding target populations’ online connectedness and communication behavior. Potential trade-offs should be taken into account when implementing web-based RDS. Examples are recruiting many seeds and less opportunities for rigorous seed selection procedures, as well as issues around online rather than physical participation, such as risks of cheaters through repeated participation. Furthermore, we have demonstrated the feasibility of—and described outcome measures for—research into web-based RDS recruitment using a process evaluation approach. The main points discussed in this literature review provide researchers with guidelines on key aspects and technicalities to consider when planning their web-based RDS research.