Assessing the geographic coverage and spatial clustering of illicit drug users recruited through respondent-driven sampling in New York City

Abstract

We assess the geographic coverage and spatial clustering of drug users recruited through respondent-driven sampling (RDS) and discuss the potential for biased RDS prevalence estimates. Illicit drug users aged 18-40 were recruited through RDS (N = 401) and targeted street outreach (TSO) (N = 210) in New York City. Using the Google Maps API™, we calculated travel distances and times using public transportation between each participant's recruitment location and the study office and between RDS recruiter-recruit pairs. We used K function analysis to evaluate and compare spatial clustering of (1) RDS vs. TSO respondents and (2) RDS seeds vs. RDS peer recruits. All participant recruitment locations clustered around the study office; however, RDS participants were significantly more likely to be recruited within walking distance of the study office than TSO participants. The TSO sample was also less spatially clustered than the RDS sample, which likely reflects (1) the van's ability to increase the sample's geographic heterogeneity and (2) that more TSO than RDS participants were enrolled on the van. Among RDS participants, individuals recruited spatially proximal peers, geographic coverage did not increase as recruitment waves progressed, and peer recruits were not less spatially clustered than seeds. Using a mobile van to recruit participants had a greater impact on the geographic coverage and spatial dependence of the TSO than the RDS sample. Future studies should consider and evaluate the impact of the recruitment approach on the geographic/spatial representativeness of the sample and how spatial biases, including the preferential recruitment of proximal peers, could impact the precision and accuracy of estimates.

FIG. 1

Map of 611 START study participants (RDS = 401, TSO = 210) by recruitment location and recruitment strategy with a New York City subway map overlaid. RDS participants are in red, TSO participants are in yellow, the stationary study office is represented with a green star, and subway lines are in dark green.

FIG. 2

Geographic coverage of RDS and TSO participants. Regions of New York City where participants were recruited through only RDS (pink), only targeted street outreach (yellow), both (green), and neither strategy (white) are displayed. All blocks are of equal area and are approximately 3 miles by 2.5 miles (length × width). Because the number of RDS and TSO participants enrolled in the study differed, the top number in each box represents the percent of the RDS sample recruited in that area and the bottom number in each box represents the percent of the TSO sample recruited in that area.

FIG. 3

Comparing the spatial intensity of RDS and TSO respondents with van recruits a included and b excluded. The maps display the spatial intensity of RDS and TSO participants (by recruitment location), respectively. Darker shades indicate greater clustering. The graphs below display the difference between K functions for RDS and TSO participants (solid black line). When the difference in K functions is positive, RDS participants are more spatially clustered than TSO participants; when the difference in K functions is negative, TSO participants are more spatially clustered than RDS participants. The 95 % confidence envelopes for a null difference in the K functions (H0: K _RDS(h) = K _TSO(h)) (dotted red lines) are based on 1,000 Monte Carlo simulations. At distances where the difference in K functions exceeds the 95 % confidence envelopes, differential spatial clustering is observed.

FIG. 4

Average public transportation travel distance (miles) and time (minutes) between RDS study participants and the study office (van recruits excluded; N = 388) and between RDS recruits and his/her recruiter by wave (N = 348).

FIG. 5

Average distance (miles) and time (minutes) by public transportation (per blocked area) between RDS recruits and his/her recruiter by recruit’s recruitment location (N = 348). Although there were 357 peer recruits in the respondent-driven sample, two individuals could not be geocoded, which resulted in the loss of two ties. Additionally, four individuals who were initially eligible to participate in the study were removed from the analysis due to inconsistencies in their self-reported drug use, which resulted in the deletion of seven additional ties. Therefore, the final sample size for recruiter–recruit distance calculations was 348.

FIG. 6

Comparing the spatial intensity of RDS seeds and peer recruits. The first two maps display the spatial intensity of a RDS seeds (N = 46) and b peer recruits (N = 355), respectively. Darker shades indicate greater clustering. c The difference between K functions for RDS seeds and peer recruits (solid black line). When the difference in K functions is positive, RDS seeds are more spatially clustered than RDS peer recruits; when the difference in K functions is negative, RDS peer recruits are more spatially clustered than RDS seeds. The 95 % confidence envelopes for a null difference in the K functions (H0: K _seeds(h) = K _{peer recruits}(h)) (dotted red lines) were based on 1,000 Monte Carlo simulations and represent the set of confidence intervals over the range of spatial distances examined. At distances where the difference in K functions exceeds the 95 % confidence envelopes, differential spatial clustering is observed.

FIG. 7

Convergence plot showing ${\widehat{p}}_1,{\widehat{p}}_2,\cdots ,{\widehat{p}}_n$ for self-reported HIV status in the TSO sample. The black solid line represents the cumulative estimate of the prevalence, and the red dotted line represents the estimate based on the complete sample, ${\widehat{p}}_n$. As seen in this figure, the prevalence of self-reported HIV in the TSO sample is relatively stable over time.

FIG. 8

Convergence plot showing ${\widehat{p}}_1,{\widehat{p}}_2,\cdots ,{\widehat{p}}_n$ for self-reported HIV status in the RDS sample. The black solid line represents the cumulative estimate of the prevalence, and the red dotted line represents the estimate based on the complete sample, ${\widehat{p}}_n$. As seen in this figure, the prevalence of self-reported HIV in the RDS sample is relatively stable over time.

FIG. 9

The bottleneck plot for self-reported HIV status shows the cumulative proportion of the RDS sample reporting HIV-positive status in New York City, by seed (N = 27). While there are 27 different seeds and consequently 27 different recruitment chains (of varying lengths), some are difficult to distinguish because the prevalence within that chain remains constant at 0 %. The red dotted line represents the estimate of self-reported HIV status based on the complete sample, ${\widehat{p}}_n$.

FIG. 10

Distribution of days between recruiter’s baseline survey and recruit’s baseline survey. This figure displays the number of participants (y-axis) enrolled in the START study according to the number of days he/she enrolled after his/her recruiter (x-axis). Of note, 49 individuals (30 %) were enrolled in the study within 1 week of the person who recruited him/her (range 0–440 days ).

FIG. 11

Scatter plot for the correlation between a recruit’s travel distance to the study office (miles) and the number of days between the recruiter’s baseline survey and his/her recruit’s baseline survey (rho = 0.09864; P value = 0.2103).

FIG. 12

Scatter plot for the correlation between a recruit’s travel time to the study office (minutes) and the number of days between the recruiter’s baseline survey and his/her recruit’s baseline survey (rho = 0.06461; P value = 0.4126).

FIG. 13

Miles traveled by recruits to the study office. This figure displays the distance traveled by recruits to the office (miles) for those recruited more than a week after his/her recruiter (one_week=0) and for those recruited within a week (one_week = 1) of his/her recruiter. When categorized as 1 week or less vs. more than 1 week between recruiter’s baseline visit and recruit’s baseline visit, there was no significant difference in the distance (miles) traveled by the recruit to the study office (P value = 0.9555).

FIG. 14

Minutes traveled by recruits to the study office. This figure displays the time traveled by recruits to the office (minutes) for those recruited more than a week after his/her recruiter (one_week=0) and for those recruited within a week (one_week = 1) of his/her recruiter. When categorized as 1 week or less vs. more than 1 week between recruiter’s baseline visit and recruit’s baseline visit, there was no significant difference in the time (minutes) traveled by the recruit to the study office (P value = 0.8475).

FIG. 15

Scatter plot for the correlation between recruiter’s distance (miles) to the office and days between recruiter’s baseline survey and recruit’s baseline survey (rho = −0.02042; P value = 0.7959).

FIG. 16

Scatter plot for the correlation between recruiter’s time (minutes) to the office and days between recruiter’s baseline survey and recruit’s baseline survey (rho = −0.03607; P value = 0.6476).

FIG. 17

Miles traveled by recruiters to the study office. This figure displays the distance (miles) traveled by recruiters to the office for those who’s recruits enrolled in the study more than a week after him/her (one_week=0) and for those who’s recruits enrolled in the study within a week (one_week = 1) of him/her. When categorized as 1 week or less vs. more than 1 week between recruiter’s baseline visit and recruit’s baseline visit, there was no significant difference in the distance traveled (miles) by the recruiter to the study office (P value = 0.1844).

FIG. 18

Minutes traveled by recruiters to the study office. The above figure displays the time (minutes) traveled by recruiters to the office for those who’s recruits enrolled in the study more than a week after him/her (one_week=0) and for those who’s recruits enrolled in the study within a week (one_week = 1) of him/her. When categorized as 1 week or less vs. more than 1 week between recruiter’s baseline visit and recruit’s baseline visit, there was no significant difference in the time traveled (minutes) by the recruiter to the study office (P value = 0.1930).