Data source and sample
The present study consists of a secondary analysis of data from two phases of a larger multi-phase study on social networks, HIV, and HCV (hepatitis C virus) among PWID in four rural areas of Puerto Rico (Cidra, Cayey, Comerío, and Aguas Buenas). In the multi-phase parent study, 315 participants were initially recruited via respondent driven sampling (RDS) for phase 1. Analysis of phase 1 RDS recruitment has been previously published, offering additional demographic and sociometric data about PWID in rural Puerto Rico before the arrival of Hurricane Maria [32, 33]. In phase 2 of the project, 33 participants from phase 1 were asked to complete extensive social network interviews, and their social connections were then in turn invited to participate in phase 2, resulting in 110 interviews. More information on the nature of the phase 2 interviews and resulting networks has been published [34]. All participants at phase 1 were at least 18 years old, alert at the time of the interview, and had injected drugs within the prior 30 days, as substantiated via track marks and completion of a questionnaire about injection practices. Track marks were identified by trained research assistants with extensive experience working with this population as syringe exchange providers. In this role, research staff were acquainted with the lesions and track marks left by intravenous drug use. In addition, staff conducted a training session where pictures of injection marks and non-injection marks were identified. Computer-assisted interviews were completed by trained field researchers, using a questionnaire adapted from the National HIV and Behavioral Surveillance (NHBS) Round 3 survey instrument [35]. All participants provided consent for the study and were compensated with up to $60 for participating in the survey and each study phase, including compensation for HIV/HCV tests. In addition, each RDS participant could receive $10 for each eligible participant brought to the study (with an upper limit of no more than three additional subjects per participant). Institutional Review Board (IRB) approval for the data collection in the parent study was provided by the University of Nebraska-Lincoln and the University of Puerto Rico.
The present study uses data from two different phases of the parent study: (1) the December 2015-January 2017 phase 2, hereafter labeled as the “pre-Maria (2016) interview,” and (2) a January-June 2019 phase, hereafter labeled as the “post-Maria (2019) interview.” Data for 110 participants were collected in the pre-Maria interview; the post-Maria interview included a total of 103 participants, 66 of whom had also participated in the pre-Maria (2016) interview. The present study’s main analyses used data from the 66 PWID who participated in both the pre-Maria (2016) and post-Maria (2019) interviews (66 individuals measured at two time points, for a total of 132 observations). A separate analysis included all 103 participants from the post-Maria (2019) interview. In the present study, listwise deletion was used for missing data, which ranged from 0.0 to 3.0% on individual variables examined.
Measures
All measures included in the present study were based on participant self-report and were related to demographic characteristics, socioeconomic circumstances, drug use/injection behaviors, treatment utilization, and overdose. Demographic characteristics were assessed based on responses to the post-Maria (2019) interview, while both pre- and post-Maria responses were examined for socioeconomic circumstances, drug use/injection behaviors, treatment utilization, and overdose. Additional file 1: Table S1 details the timeframes (e.g., past year) for each variable in the pre- and post-Maria interviews.
Demographic characteristics included the area of residence (Cidra, Cayey, Comerío, Aguas Buenas, or other), age (in years), gender (man, woman, transgender), educational attainment (less than 12th grade, 12th grade or GED, beyond high school), marital status (single/never married, married/living as married, separated/divorced/widowed), birthplace (Puerto Rico, continental United States), and whether the participant had ever lived in the continental United States (yes/no). Socioeconomic circumstances included current homelessness (yes/no), monthly job income (none, $1–299, $300–699, $700+), current unemployment or disability (yes/no), current health insurance (yes/no), and inability to receive necessary medical care due to cost or access (yes/no).
Drug use/injection practices included age at first injection (less than 15 years old, 15–18, 19–24, 25 or older), injection frequency (monthly or less, less than daily but more than monthly, 1–3 times daily, 4–7 times daily, 8 or more times daily), new needle use for most or all injections (yes/no), use of a needle someone else had used, half of the time or more (yes/no), and sharing a cooker/cotton/water for at least half of all injections (yes/no). Daily injection (yes/no) was assessed for the following drugs: (a) heroin with cocaine, (b) xylazine with cocaine, (c) heroin alone, (d) powder cocaine alone, (e) crack cocaine, (f) buprenorphine, (g) xylazine alone, (h) methamphetamine, and (i) prescription opioids. Weekly non-injection drug use (yes/no) was assessed for (a) cannabis, (b) crack cocaine, (c) powder cocaine, (d) benzodiazepines, (e) prescription opioids, (f) heroin, (g) buprenorphine (from the street), (h) methamphetamine, (i) xylazine, (j) hallucinogens, and (k) ecstasy. Finally, weekly “binge drinking” (yes/no) was defined as more than five drinks (for males) or four drinks (for females) on one occasion, at least weekly.
Measures related to treatment utilization included participation in drug treatment of any kind in the past year (yes/no) and whether the participant attempted but was unable to enter drug treatment in the past year (yes/no).
The present study’s primary outcomes were overdose-related measures. In the parent study, overdose measures comprised cumulative totals/lifetime measures of the number of overdoses participants reported ever experiencing, rather than measures of overdoses within specific time frames (e.g., past-year/past-month overdose experiences). For the present study, therefore, we constructed a binary (yes/no) variable to capture whether each participant reported experiencing an overdose during the interview year and/or calendar year prior (that is, when assessed during the pre-Maria interview, whether each participant reported experiencing an overdose during 2015 and/or 2016, and, when assessed during the post-Maria interview, whether each participant reported experiencing an overdose during 2018 and/or 2019). This measure represents the closest approximation of a past-year overdose measure available within the existing data, although we chose to include overdoses in two calendar years (both the year of interview and the year before) for a more robust measure given our modest sample size. This variable was constructed based on participants’ responses to a question about the year in which their most recent overdose occurred, as assessed during both the pre-Maria and post-Maria interviews.
Three additional questions (lifetime measures) assessed overdoses within participants’ social networks: (a) the number of overdoses the participant reported ever witnessing, (b) the number of people the participant reported knowing who have experienced overdose, and (c) the number of people the participant reporting knowing who have died from a drug overdose. Finally, in the post-Maria (2019) interview only, participants also answered several questions regarding changes they had experienced after Hurricane Maria. Participants were asked for their perception of the frequency of overdoses among people who use drugs (a) during the time period right after Hurricane Maria, compared to before (options: fewer, same, more) and (b) since Hurricane Maria, relative to before (options: fewer, same, more). While the wording of the first question focused only on the period immediately after the hurricane occurred, the second question was intended to encompass a broader period of time, from the day of the hurricane to the day of the interview.
Analyses
All analyses were completed in Stata/MP 16.1. First, descriptive statistics (for demographic measures and age at first injection) were calculated for the 66 PWID who had participated in both the pre-Maria (2016) and post-Maria (2019) interviews. Next, these 66 participants’ responses in the pre-Maria and post-Maria interviews were compared with respect to socioeconomic circumstances, drug use/injection practices, and treatment utilization. Due to the data’s paired responses (repeated measures for the same participants at two different time points), McNemar tests were used for binary variables, and the paired-samples sign test was used for ordinal variables. In consideration of the modest sample size, the mid-p version of the McNemar test [36] was chosen to maximize statistical power while utilizing a more conservative option than the classic McNemar test (limiting type I error) [36, 37].
Next, mixed-effects binomial logistic regression analysis was used to model time of interview (pre-Maria [2016] or post-Maria [2019]) as a predictor of the overdose-related outcome variable (a yes/no measure of whether the participant reported experiencing an overdose in the interview year and/or prior calendar year), with random intercepts at the participant level due to the non-independence of observations clustered within participants. Results were expressed as odds ratios (ORs) with 95% confidence intervals (CIs). Box plots, plotted with the user-written program stripplot [38], were then utilized to depict the raw distributions of the three other overdose-related variables: the reported number of overdoses witnessed, number of people known who had experienced overdose, and number of people known who had died of an overdose, as assessed in the pre-Maria (2016) and post-Maria (2019) interviews. Descriptive statistics (means and standard deviations [SDs]) were also calculated for each of these measures pre- and post-Maria. Finally, a column chart was used to depict the relative frequency of each response to the two post-Maria interview questions regarding participants’ perceptions of changes in the frequency of overdoses among people who use drugs.