Ferrer-Rosende et al. BMC Medical Research Methodology (2023) 23:217 BMC Medical Research https://doi.org/10.1186/s12874-023-02042-z Methodology RESEARCH Open Access Implementation of web-based respondent driven sampling in epidemiological studies Pedro Ferrer-Rosende1,2,3, María Feijoo-Cid4,5, María Isabel Fernández-Cano4,5, Sergio Salas-Nicás1, Valeria Stuardo-Ávila6 and Albert Navarro-Giné1,2,7* Abstract Background Respondent-driven sampling (RDS) is a peer chain-recruitment method for populations without a sampling frame or that are hard-to-reach. Although RDS is usually done face-to-face, the online version (WebRDS) has drawn a lot of attention as it has many potential benefits, despite this, to date there is no clear framework for its implementation. This article aims to provide guidance for researchers who want to recruit through a WebRDS. Methods Description of the development phase: guidance is provided addressing aspects related to the formative research, the design of the questionnaire, the implementation of the coupon system using a free software and the diffusion plan, using as an example a web-based cross-sectional study conducted in Spain between April and June 2022 describing the working conditions and health status of homecare workers for dependent people. Results The application of the survey: we discuss about the monitoring strategies throughout the recruitment process and potential problems along with proposed solutions. Conclusions Under certain conditions, it is possible to obtain a sample with recruitment performance similar to that of other RDS without the need for monetary incentives and using a free access software, considerably reducing costs and allowing its use to be extended to other research groups. Keywords Epidemiologic methods, Hard-to-reach populations, Web based sampling, Respondent-driven sampling, WebRDS Background *Correspondence: Albert Navarro-Giné Respondent-driven sampling (RDS) is a sampling method albert.navarro@uab.cat that has gained popularity in epidemiological studies 1 Research group on psychosocial risks, organization of work and health, over the years for hard-to-reach populations or those Universitat Autònoma de Barcelona, Cerdanyola del Vallès, Spain 2 Unitat de Bioestadística, Facultat de Medicina, Departament de without a sampling frame [1, 2]. This method is based on Pediatria, d’Obstetrícia i Ginecologia i de Medicina Preventiva, Universitat a chain-referral process, which involves three main steps: Autònoma de Barcelona, Avda. Can Domènech S/N, formative research, data collection, and data analysis [3]. Cerdanyola del Vallès 08193, Spain 3 Programa de doctorat en Metodologia de la Recerca Biomèdica i Salut Formative research is a crucial phase where researchers Pública, Universitat Autònoma de Barcelona, Cerdanyola del Vallès, Spain delve into social network properties of the target popu- 4 Nursing Department, Faculty of Medicine, Universitat Autònoma de lation, evaluate the acceptability of RDS as a viable sam- Barcelona, Cerdanyola del Vallès, Spain 5 Multidisciplinary Research Group in Health and Society (GREMSAS), pling method, determine the selection of initial members Barcelona, Spain (or ‘seeds’), and address survey logistics, including incen- 6 Instituto de Salud Pública, Universidad Andrés Bello, Santiago, Chile tives and coupon design [4]. 7 Institut d’Estudis del Treball, Universitat Autònoma de Barcelona, Cerdanyola del Vallès, Spain © The Author(s) 2023. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data. Ferrer-Rosende et al. BMC Medical Research Methodology (2023) 23:217 Page 2 of 9 During the data collection process, these ‘seeds’ are software, monitoring strategies throughout the recruit- required to answer the survey, which should include one ment process, and potential problems, along with pro- or more questions about the size or ‘degree’ of their per- posed solutions. sonal network (e.g., ‘How many people with the charac- teristics of the target population of the study do you know Development and/or can you contact right now?‘). This information Example is crucial, as RDS estimates use it to calculate the prob- We demonstrate the use of WebRDS in the formative ability of selection for each respondent. Additionally, it is research, implementation, and follow-up recruitment recommended to include some other questions that can phases through data from the CUIDÉMONOS Project serve as diagnostics for the sampling method at the end [12], a web-based cross-sectional study conducted in of the study (for a comprehensive discussion on diagnos- Spain from April to June 2022. The objective of the proj- tics, see Gile et al. 2015) [5]. ect was to describe the working conditions and health Once the survey is completed, ‘seeds’ are instructed to status of homecare workers for dependent individuals. recruit a limited number of participants from their per- Homecare workers provide routine personal assistance in sonal network, usually through a coupon system. These daily activities to people who require it in private homes. recruits are asked to take the survey and subsequently Research indicates that exposure to harmful working become recruiters themselves. This process continues conditions can adversely affect the health of workers in for as many waves as necessary until the desired sample the homecare sector [13]. However, despite the growing size is reached, the participants’ characteristics have sta- social and economic significance of this sector, scientific bilized, or until the recruitment chains become extinct. evidence on their working conditions and its impact on For the effectiveness of the sampling method, it is ideal occupational health remains scarce. The fact that homec- for the seeds to have a large and diverse social network. are workers develop their job within private homes fur- For data analysis, there is a growing theoretical frame- ther complicates the assessment of their workplace work with several proposed estimators for RDS data exposures, as there is no sampling frame available. which under certain assumptions can generate asymptot- Given the absence of a sampling frame, we proposed ically unbiased population estimates [6]. Consequently, it WebRDS as a suitable sampling method for this popula- becomes vital for researchers to carefully select the RDS tion. This decision was reinforced by the apparent high inference approach that provides more robust estimates, motivation of the target population towards the study, taking into account the assumptions met by study design the opportunity to collect data from a large geographi- [7]. cal area (all of Spain), and the fact that formative research Although RDS has traditionally been conducted face- indicated this population is well-connected and regularly to-face, the online version (WebRDS) has gained sig- uses mobile phones. nificant attention in the last decade due to its potential benefits over face-to-face RDS [8, 9]. WebRDS offers easy Formative research access for participants and ensures anonymity, elimi- As part of our preparatory work, we conducted a com- nating time and location-related barriers. Moreover, it prehensive literature review of previous research and allows data collection within a short time frame and at held interviews and focus groups with the invaluable sup- low cost, providing a more efficient recruitment medium port of a national homecare workers association. These [10]. However, similar to other web survey methods, meetings served two primary purposes: first, to evalu- WebRDS may also encounter challenges, such as bias ate the characteristics of the social network within the from differential internet access, the possibility of mul- target population and assess its suitability for the pro- tiple responses, concerns regarding the credibility of posed sampling method, and second, to identify suit- online research, and the absence of face-to-face interac- able ‘seeds’ for the study. Additionally, these interactions tions [11]. proved instrumental in refining the final questionnaire to Despite the increasing use of WebRDS, to our knowl- encompass the objectives of the study while aligning with edge there is no clear framework for the implementa- the interests and concerns of homecare workers. This tion of this online recruitment method, which is of great alignment played a crucial role in motivating their active importance considering that online approach could participation. be almost as feasible and effective as face-to-face RDS For seed selection, we specifically targeted workers if the population and the application of the method is from diverse profiles based on age, gender, migratory sta- appropiate [9]. This article aims to provide guidance for tus, size of geographic working area, and type of contract, researchers who want to recruit through a WebRDS. It ideally, with a large social network. Considering that long covers various aspects, including formative research, chains of a few seeds are preferable over short chains of implementation of the coupon system using a free acces many seeds [3, 14] a total of eight seeds were recruited Ferrer-Rosende et al. BMC Medical Research Methodology (2023) 23:217 Page 3 of 9 at the beginning and three coupons were allowed per after they completed the survey, with a maximum of recruiter. three uses. Once a participant received and completed After selecting the seeds, we conducted virtual meet- the survey through their link, a new link was automati- ings with them, providing detailed explanations of the cally generated and provided to them, which they could study and the recruitment method. Furthermore, we cre- then share with three other individuals, and so on (Fig. 1). ated promotional videos for the study, which were dis- To facilitate this process, we utilized Limesurvey to cre- seminated on YouTube, Facebook groups of homecare ate a participants database, where each recruiter’s iden- workers, and shared through WhatsApp. tification number (Id) was linked to a predefined access code or “token,“ corresponding to a unique link with a Designing the questionnaire maximum of three uses. Additionally, a “token” variable The questionnaire and coupon system were implemented was automatically generated in the response database, using the free-to-use application Limesurvey (https:// enabling us to identify the recruiter’s ID for each partici- www.limesurvey.org). The questionnaire was based on pant (Fig. 2). previously validated instruments in Spain and tailored to At the conclusion of the survey, participants were pre- align with the specific study objectives. It encompassed sented with two options to share the generated link (Fig- a range of essential aspects, including sociodemographic ure S3). The first option allowed them to simply press a information, labor management practices, psychosocial box, which automatically opened WhatsApp on their exposure at work, and health status. Furthermore, follow- phone or computer, enabling them to send the link to ing the recommendations of Gile (2015) [5], we included their contacts along with a pre-defined message encour- specific questions to capture participants’ self-reported aging participation in the study. The message also pro- network size: vided an explanation of the study methodology and 1) How many homecare workers do you know in Spain? included contact information in case of any doubts (Fig- 2) How many of those “n” people could you invite to ure S4). The second alternative was to manually copy the participate in the survey right now because you have link and share it with three different individuals. Upon their phone or email contact? survey completion, participants were given the option to Response to the second question was the degree used for provide their contact information, facilitating resolution estimations. Also, to account for finite population and of any potential issues or future communication (Figure reciprocity assumptions respectively we asked the follow- S2). ing questions: 3) Besides the person who gave you the link, how many Diffusion plan other homecare workers do you know who have The success of the project relied heavily on effective already participated in this study? promotion before and during recruitment. Several vid- 4) Would the person who sent you the link to eos featuring homecare worker platform members were participate in the survey be one of your three shared on social media platforms. The first video was contacts if you had received it from someone else? released before the study began, aiming to motivate par- Ensuring clear instructions for new recruits was of ticipation among homecare workers. A second video utmost importance, as it would not be possible to deliver followed, explaining the recruitment process. Two addi- them personally. To address this, we included a brief tional videos were published after recruitment started, description of the study along with a five-minute video encouraging participation with the slogan “No rompas on the first screen of the survey, before obtaining consent la cadena” (“Don’t break the chain”), visually illustrating from participants. This video provided detailed explana- the consequences if a branch of the recruitment chain did tions of the research method and the recruitment process not recruit. Homecare workers actively participated in (https://www.youtube.com/watch?v=tN9M4abXczM&t). sharing these videos to promote the study. Additionally, we provided the contact information of the research team, including an email address and phone Application number, in case participants had any questions or doubts Launching day (see Figure S1). On the day scheduled for the start of the study, a vir- tual meeting was conducted with the seeds to discuss Coupon system final details and make sure that everyone was ready and One of the most challenging and crucial aspects of informed about the recruitment process. After the meet- designing the study was establishing the coupon distribu- ing, everyone received their respective link to the survey tion system, which plays a vital role in the success of the through WhatsApp since this would be the preferred method. To accomplish this, a unique and personalized platform for sharing the links generated afterwrds. link was automatically generated for each participant Ferrer-Rosende et al. BMC Medical Research Methodology (2023) 23:217 Page 4 of 9 Fig. 1 Recruitment process using Web based Respondent-Driven Sampling (WebRDS) from a ‘seed’ respondent Fig. 2 Coupon system in Limesurvey, 2A: Survey participant list 2B: Survey responses. Each access code in 2A is represented by the “token” variable in 2B with a limited amount of uses defined by “Uses left” in 2A which is related to a unique personal link. After completion of the survey, the “Id” variable in 2B become the access code for that participant in 2A and therefore the token variable for his three recruited new participants. In this example, participants with Id = 48, 66 and 67 in 2B were invited by participant Id = 30 which invitation had 3 uses (2A), now each of them has 3 invitations that will be reflected in the survey participant list with access codes 48, 66 and 67 respectively Follow-up variables of interest was assessed using convergence and During the recruitment process, constant monitoring bottleneck plots. These ongoing evaluations ensured the and evaluation of the system’s performance were crucial. reliability and accuracy of the data collection process. To achieve this, a semi-automated report was generated To assess the presence of multiple answers, a two-step every two days using “RDS” package of R software [15]. process was implemented. First, if a participant had pro- This report provided a descriptive table of the charac- vided contact information, it was verified to ensure there teristics of the sample and visual representation of the were no repetitions. Second, the combination of specific recruitment chain stratified by variables of interest. Addi- responses was checked across different participants to tionally, on a weekly basis, the convergence for specific identify any potential duplicates. Ferrer-Rosende et al. BMC Medical Research Methodology (2023) 23:217 Page 5 of 9 Contact with the seeds throughout the recruitment Recruitment performance process was crucial. They were regularly updated of The study initially involved 8 seeds; however, during the the progress of their recruitment tree (anonymized) to recruitment process, we became aware that our focus was encourage participation. The objective was to empower primarily on the profiles of the seeds, neglecting their all seeds to have an influence on the completion at least network size and willingness to stay engaged with the of the first two waves of recruitment. research. Consequently, certain initial seeds responded Lastly, for participants whose links had available uses to the survey but failed to maintain contact, leading to after 5 days of completing the survey and had provided unproductive chains. Therefore, we made the decision to their contact information, a message was sent to them. introduce new seeds during recruitment that exhibited The message informed them of the number of respon- improved performance (a total of six additional seeds dents and emphasized the significance of continuing the were included). After incorporating those new seeds and recruitment chain. Their respective link was also attached removing the only one who did not recruit, we ended up to facilitate forwarding if needed. with 337 responses from 13 seeds. Median recruitment These last steps turned out to be very helpful. An chain length (waves) was 4 (Range, 1–10), 162 partici- example of this is displayed in Fig. 3, where a recruitment pants (48.1%) recruited at least one person, 103 (30.6%) tree of a seed who had not advanced for days is shown recruited at least two and 59 (17.5%) recruited three. The and the effect that a single contact had on its growth a largest recruitment chain contained 123 participants, few days later. 36.5% of all recruits. The final recruitment tree and a example of a convergence and a bottleneck diagnosis Challenges encountered during recruitment process plots can be seen in Figs. 4 and 5 respectively. Table 2 Despite our efforts to anticipate and address potential includes RDS estimates for selected sociodemographic challenges during the recruitment process, we encoun- characteristics and recruitment homophily (the ratio of tered a few problems along the way, ranging from tech- the number of recruits who have the same characteristic nical issues to other diverse natures. Table 1 provides a as their recruiter to the number we would expect if there summary of the main challenges we faced and the solu- was no homophily). tions we applied to overcome them. Throughout the study, we diligently recorded all these issues in a study Discussion log, and if necessary, we made corrections in the analysis. Conducting RDS sampling online allowed us to obtain a It is worth noting that the high number of participants nationwide sample with a total of 338 responses (324 par- who left their contact information proved to be invalu- ticipant after seeds removal). This number of participants able in resolving most of these problems, as we were able is comparable to those reported in face-to-face RDS to communicate directly with them and find suitable both in sample size and in number of waves [16]. It was resolutions. also consistent with that reported in other WebRDS [8]. When comparing sample and RDS-adjusted proportions, Fig. 3 The effect that a single contact had on a recruitment tree of a seed that had not advanced for days Ferrer-Rosende et al. BMC Medical Research Methodology (2023) 23:217 Page 6 of 9 Table 1 Issues encountered during recruitment process and the solutions applied Issue Description Solution Invite link has Some participants shared the link that belonged When this happened, it was addressed by getting the ID of the participant who already been to them, instead of using the one generated sent the repeated link. After finding it, the misused token is given one more use. used. specifically after completing the survey. In addition, the participant who sent the wrong link is given his correct link to be forwarded to his contacts. To prevent this from happening, the survey open- ing message was edited to add clarity to this instruction. Multiple One participant answered the survey three The participant was contacted since she left her contact information, the situ- response times in a row using her unique link and then ation was corrected considering the first response as the official response. Two using the link she generated because of confu- more uses to the link were given. sion about the coupon system. One participant answered the survey in two The participant was contacted since she left her contact information, her first re- different time periods through the invitation of sponse was considered as the official one and she was asked to tell her recruiter two different people. to send another invitation to replace her. One extra use was given to the link. Distrust for the A few participants called us or told recruiters If the participant contacted us, we tried to assure them of the seriousness of the study that they have some distrust for the interest of study, if it was a recruiter who contacted us, we gave them the information to the study. transfer. If the person recruited still did not want to participate, they were asked to invite another person to participate. Seeds whose Some seed’s recruitment trees did not advance Since we had contact with the seeds, we communicated with them by showing recruitment did beyond the first waves. them their anonymized recruitment tree so that they could try to secure the not advance at first two waves with complete answers. If after a while they did not advance, we the beginning added a new seed of same characteristics. Trouble using One participant called saying that she did not We tried to help her through the process but as she showed no interest, she the phone want to participate because she did not know was asked to tell her recruiter to replace her with someone else. how to answer the survey on the phone. we estimate that, without adjusting for RDS, we would and for each person recruited (secondary reward). In be underestimating the prevalence of young individuals our case, we did not provide any rewards, which initially and foreigners (Table 2). The recruitment homophily for raised concerns about potential participation rates. For- sociodemographic variables was close to one, indicat- tunately, this was not the case due to the extensive for- ing that participants tended to recruit others with simi- mative research, a population very interested in the study lar characteristics. However, the homophily was slightly and Permanent contact with the seeds. An advantage higher in terms of working area size, suggesting that par- of not offering rewards is that it reduces the likelihood ticipants were more likely to recruit individuals working of fraud or multiple participation, a concern that arises in the same area size as themselves. when conducting WebRDS or any type of online survey We attribute the success of this sampling method in [3]. our study to four key factors. Firstly, extensive formative Throughout the recruitment process, we also encoun- research provided us with a profound understanding of tered some challenges. For instance, when selecting the target population and their concerns. This enabled seeds, despite acknowledging the significance of net- us to design an instrument that not only aligned with work size and diversity, we faced difficulties in reaching the study’s objectives but also intrigued participants to suitable seeds that could ensure effective recruitment, engage without the need for monetary incentives. Sec- particularly for certain profiles. As a result, while not ondly, an efficient social media diffusion plan, aided by disregarding the desired profiles, we placed greater the active involvement of homecare workers themselves, emphasis on choosing new seeds with large and diverse played a significant role in attracting participants before networks rather than solely focusing on the characteris- and during the recruitment process. Thirdly, maintaining tics of the seed itself [4]. consistent communication with the seeds proved crucial Before implementing the instrument, it is crucial to in encouraging participation, especially during periods anticipate any potential issues that may arise during the of reduced recruitment and in promptly addressing any process and take proactive measures to minimize prob- issues that arose during the process. Lastly, offering par- lems. For instance, we encountered a situation where ticipants the option to provide their contact information some participants shared their invitation link instead allowed us to send reminders and address problems that of the one specifically generated for them to share after may have emerged deeper in the recruitment chain. This completing the survey (Table 1). Fortunately, the soft- aspect was particularly vital, given that 54% of the par- ware utilized enabled us to make adjustments during ticipants chose to share their contact details. the recruitment process, allowing us to provide clearer When conducting an RDS participants are usually instructions and consequently reducing the number of rewarded for answering the survey (primary reward) participants facing this problem. Ferrer-Rosende et al. BMC Medical Research Methodology (2023) 23:217 Page 7 of 9 Fig. 4 Final recruitment network of the study. (n = 337) One issue that still needs to be addressed in WebRDS or regression method would perform best, taking into is data analysis. There is extensive debate concerning consideration the specific challenges that may arise with the various mean or variance estimators for RDS [7], as this type of RDS sampling. well as regression models used to assess risk factors [17]. These discussions primarily revolve around the poten- Conclusion tial biases that different methods may exhibit depend- In response to the issues of high cost or effort that face- ing on the violation of various RDS assumptions. When to-face RDS requires, online RDS have emerged as a applying WebRDS, certain assumptions may be more cost-effective alternative. We propose that under cer- susceptible to violation compared to face-to-face meth- tain conditions, i.e., extensive formative research, good ods, such as the misspecification of network size. Conse- diffusion plan, a population interested in the study and quently, it becomes crucial to evaluate which estimator contact with the seeds, it is possible to obtain a sample Ferrer-Rosende et al. BMC Medical Research Methodology (2023) 23:217 Page 8 of 9 Fig. 5 Example of a Convergence and a Bottleneck diagnosis plot Authors’ contributions Table 2 RDS Estimates and recruitment homophily for some P.F.R., M.F.C., M.I.F.C., S.S.N., V.S.A. and A.N.G. conceived the study. P.F.R. and selected sociodemographic characteristics (seeds were removed A.N.G. carried out all analyses. P.F.R., M.F.C., M.I.F.C., S.S.N., V.S.A. and A.N.G. from the analysis) contributed to interpretation of the results. P.F.R. drafted the original Characteristic N (%) RDS-II Recruit- manuscript. P.F.R., M.F.C., M.I.F.C., S.S.N., V.S.A. and A.N.G. contributed to editing (N = 324) Estimatea ment the manuscript and approved the version to be published. A.N.G. will act as guarantor for the paper. homophily Gender 1.03 Funding Men 18 (5.6%) 4.8 [1.0-13.3] This work was partially supported by Fundación Prevent, XV Becas I + D in Women 304 (94.4%) 95.2 [86.7–99.0] PRL 2021. The funding bodies played no role in the design of the study and collection, analysis, and interpretation of data and in writing the manuscript. Age 1.12 39y or less 49 (15.2%) 21.5 [7.0-36.1] Data Availability 40-54y 174 (54.0%) 55.2 [46.0-64.4] The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request. 55y or more 99 (30.8%) 23.3 [10.4–36.2] Place of birth 1.22 Declarations Spain 250 (77.9%) 67.1 [55.3–78.8] Other 71 (22.1%) 32.9 [21.2–44.7] Ethics approval and consent to participate Live with partner 1.05 All aspects related to this project have been validated and approved by the No 104 (35.9%) 41.3 [26.9–55.7] Committee on Ethics in Animal and Human Experimentation of Universitat Autònoma de Barcelona (CEEAH-5920). All methods were carried out in Yes 186 (64.1%) 58.7 [44.3–73.1] accordance with relevant guidelines and regulations. All participants provided Working area size 1.42 informed consent to participate in the study. Rural 80 (24.7%) 26.7 [11.3–42.1] Consent for publication Intermediate 99 (30.6%) 34.0 [21.7–46.3] Not applicable. Urban 145 (44.8%) 39.2 [25.3–53.2] a Bootstrap 95%CI Competing interests The authors declare no competing interests. with recruitment performance similar to that of other Received: 25 January 2023 / Accepted: 26 September 2023 RDS without the need for monetary incentives and using a free access software, considerably reducing costs and allowing its use to be extended to other research groups without the need for a large budget. References Supplementary Information 1. Malekinejad M, Johnston LG, Kendall C, Kerr LRFS, Rifkin MR, Rutherford GW. The online version contains supplementary material available at https://doi. Using respondent-driven sampling methodology for HIV biological and org/10.1186/s12874-023-02042-z. behavioral surveillance in international settings: a systematic review. AIDS Behav. 2008;12(1):105–30. Supplementary Material 1 2. Baraff AJ, McCormick TH, Raftery AE. Estimating uncertainty in respondent- driven sampling using a tree bootstrap method. Proc Natl Acad Sci USA. 2016;113:14668–73. Acknowledgements 3. Sosenko FL, Bramley G. Smartphone-based Respondent Driven Sam- The research team would like to acknowledge Plataforma Unitaria SAD for pling (RDS): a methodological advance in surveying small or ‘hard-to- their help in planning and organizing the study. P.F.R. acknowledges the reach’populations. PLoS ONE. 2022;17:e0270673. support of ANID in the means of scholarship for PhD studies. Ferrer-Rosende et al. BMC Medical Research Methodology (2023) 23:217 Page 9 of 9 4. Johnston LG, Whitehead S, Simic-Lawson M, Kendall C. Formative research to y estado de salud. Barcelona: POWAH, GREMSAS, UAB; 2023. Available at: optimize respondent-driven sampling surveys among hard-to-reach popula- https://ddd.uab.cat/record/272314 tions in HIV behavioral and biological surveillance: lessons learned from four 13. Kouvonen A, Mänty M, Lallukka T, Pietiläinen O, Lahelma E, Rahkonen O. case studies. AIDS Care. 2010;22(7):784–92. Changes in psychosocial and physical working conditions and psychotropic 5. Gile KJ, Johnston LG, Salganik MJ. Diagnostics for respondent-driven sam- medication in ageing public sector employees: a record-linkage follow-up pling. J R Stat Soc. 2015;1(1):241–69. study. BMJ open. 2017;7(7):e015573. 6. Heckathorn DD. Respondent-driven sampling: a new approach to the study 14. Lachowsky NJ, Sorge JT, Raymond HF, et al. Does size really matter? A sensitiv- of hidden populations. Soc Probl. 1997;44(2):174–99. ity analysis of number of seeds in a respondent-driven sampling study of gay, 7. Abdesselam K, Verdery A, Pelude L, Dhami P, Momoli F, Jolly AM. The develop- bisexual and other men who have sex with men in Vancouver, Canada. BMC ment of respondent-driven sampling (RDS) inference: a systematic review Med Res Methodol. 2016;16(1):1–10. of the population mean and variance estimates. Drug Alcohol Depend. 15. Handcock MS, Gile KJ, Fellows IE, Neely WW, Package. ‘RDS’. 2016. https:// 2020;206:107702. cran.r-project.org/web/packages/RDS/index.html (21 December 2022, date 8. Wejnert C, Heckathorn DD. Web-based network sampling: efficiency and last accessed). efficacy of respondent-driven sampling for online research. Sociol Methods 16. Johnston LG, Hakim AJ, Dittrich S, Burnett J, Kim E, White RG. A systematic Res. 2008;37(1):105–34. review of published respondent-driven sampling surveys collecting behav- 9. Helms YB, Hamdiui N, Kretzschmar ME, et al. Applications and recruitment ioral and biologic data. AIDS Behav. 2016;20(8):1754–76. performance of web-based respondent-driven sampling: scoping review. J 17. Yauck M, Moodie EE, Apelian H, et al. General regression methods for respon- Med Internet Res. 2021;23(1):e17564. dent-driven sampling data. Stat Methods Med Res. 2021;30(9):2105–18. 10. Latkin CA, Knowlton AR. Social network assessments and interventions for health behavior change: a critical review. Behav Med. 2015;41(3):90–7. 11. Wright K. Researching internet-based populations: advantages and disad- vantages of online survey research, online questionnaire authoring software Publisher’s Note packages, and web survey services. J Comput Commun. 2006;10(3):A. Springer Nature remains neutral with regard to jurisdictional claims in 12. Fernández-Cano MI, Navarro A, Feijoo-Cid M, Salas-Nicás S. Estudio Cuidé- published maps and institutional affiliations. moNos. Auxiliares de ayuda a domicilio en España, 2022. Riesgos laborales