Statistical adjustment of network degree in respondent-driven sampling estimators: venue attendance as a proxy for network size among young MSM

Abstract

We introduce a new venue-informed network degree measure, which we applied to respondent-driven sampling (RDS) estimators. Using data collected from 746 young MSM in 2014-2016 in Chicago, IL, and Houston, TX, we estimated the population seroprevalence of HIV and syphilis and risk/protective behaviors, using RDS estimates with self-reported network size as a standard degree measure as well as our proposed venue-informed degree measure. The results indicate that the venue-informed degree measure tended to be more efficient (smaller variance) and less biased than the other measure in both cities sampled. Venue attendance-adjusted network size may provide a more reliable and accurate degree measure for RDS estimates of the outcomes of interest.

Keywords: HIV; Respondent-driven sampling method (RDS); pre-exposure prophylaxis (PrEP); risky sexual behavior; sexually transmitted infections; two-mode affiliation networks; venue affiliation; young men who have sex with men (MSM).

Figure 1

RDS Chains among YMSM for Chicago (left) and Houston (right), excluding non-sprouted seeds.

Figure 1

RDS Chains among YMSM for Chicago (left) and Houston (right), excluding non-sprouted seeds.

Figure 2

Histogram of self-reported network size and model-based venue-predicted degree for Chicago (left) and Houston (right). The degree distributions of self-reported network size tend to be right-skewed with some outliers and have wider variance. In contrast, the degree distributions of the venue-predicted/fitted measure tend to be more symmetric around the mean and have narrower variance. This suggests that the model adjustment appears to assist in right-shifting and/or left-shifting the outliers in the portion of population.

Figure 3

Unadjusted sample estimates and V-H estimators (RDS-II) and 95% confidence intervals, weighted by self-reported degree (Self) and venue-predicted degree (Venue) of five outcome variables for Chicago and Houston. In general, the venue-predicted degrees tended to have narrower CIs compared to the self-reported degrees for all outcomes in both cities. For Chicago, the ratios of variance (Venue vs. Self) were 0.51, 0.51, 0.50, 0.67, and 0.49 for HIV-seropositive, syphilis-seropositive, inconsistent condom use, cannabis use, and PrEP use (HIV-seronegative participants only, N = 246 for Chicago, N = 235 for Houston), respectively. For Houston, the equivalent ratios of variance (Venue vs. Self) were 0.44, 0.44, 0.44, 0.62, and 0.55, respectively.

Figure 4

Impact (relative change) in regard to model adjustment for the RDS-II estimates for right-shifting lower self-reported degrees and left-shifting higher self-reported degrees for Chicago and Houston. The impact in regard to model adjustment, using estimated population degree, on each RDS estimate (RDS-II) of seroprevalences of HIV infection (red), syphilis infection (dark), inconsistent condom use (light green), cannabis use (blue), and PrEP use (purple) by both right-shifting versus left-shifting was empirically assessed by computing the relative change in the RDS-II estimate. This involved the use of varying percentiles of degrees below the lower degree values (for right-shifting) and above the higher degree values’ (for left-shifting) being set to self-reported degree values, while the rest was set to model-adjusted fitted degrees, relative to the use of all self-reported degree values. The results indicate that the right-shifting lower degrees tend to have a larger impact (i.e., greater changes) on the RDS-II estimates than do the left-shifting higher degrees, especially for higher percentiles below the lower degree values’ being set to the self-reported degree values. Conversely, left-shifting higher degree values tend to have a larger impact (i.e., greater changes) on RDS-II estimates for lower percentiles as compared to the higher degrees’ being set to the self-reported degree. In addition, the right-shifting lower degrees (dashed line) and left-shifting higher degrees (solid line) have different magnitudes of change in RDS-II estimates, depending on the different prevalences of interest for both cities.