Perception and correlates of opioid overdose risk among overdose survivors who use non-prescribed opioids in San Francisco and Boston

Abstract

Background:

Understanding opioid overdose risk perception may inform overdose prevention strategies.

Methods:

We used baseline data from a randomized overdose prevention trial, in San Francisco, CA and Boston, MA, among people who used non-prescribed opioids, survived an overdose in the past three years, and had received naloxone. Participants were asked how likely they were to overdose in the next four months. We combined “extremely likely” and “likely” (higher risk perception) and “neutral,” “unlikely,” and “extremely unlikely” (lower risk perception). We performed bivariate analyses and separate multivariable logistic regression models of risk perception across (1) sociodemographic, (2) substance use, and (3) overdose risk behavior measures. Covariates were selected a priori or significant in bivariate analyses.

Results:

Among 268 participants, 88% reported at least one overdose risk behavior, however only 21% reported higher risk perception. The adjusted odds ratio (AOR) of higher risk perception was 2.41 (95% CI: 1.10–5.30) among those unhoused in the past 4 months, 2.06 (95% CI: 1.05–4.05) among those using opioids in a new place, and 5.61 (95% CI: 2.82–11.16) among those who had overdosed in the past four months. Living in Boston was associated with higher risk perception in all three models (AOR= 2.00–2.46, 95% CI: 1.04–4.88).

Conclusions:

Despite prevalent risk behaviors, a minority of participants perceived themselves to be at higher risk of overdose. Nonetheless, some known risk factors for overdose were appropriately associated with risk perception. Fentanyl has been prevalent in Boston for longer than San Francisco, which may explain the higher risk perception there.

INTRODUCTION

Opioid overdose deaths have been increasing since 1999 in the United States.¹ In 2021, 75% of drug overdose deaths in the United States involved opioids.² There are many well-studied risk factors for opioid overdose such as unreliability of the drug supply, a history of past overdoses, changes in opioid tolerance, and polysubstance use.² Studies have shown that people reporting overdose risk behaviors such as high average daily opioid dosages, using opioids concurrently with benzodiazepines, and past month alcohol use had a low perceived overdose risk.^3–5 However, existing literature on risk perception is limited to heroin and prescribed opioids, and may not reflect the current era of fentanyl as the predominant non-prescribed opioid.

Previous research on overdose risk perception has identified associations between greater risk perception and chronic pain, more frequent injection, having an opioid use disorder, previous overdose, heroin use, and the rising prevalence of fentanyl.^3–6 Additional studies have found that individuals who survived a previous opioid overdose perceived a greater risk of overdose than those who had not experienced a previous overdose.^2,5,7 There have been mixed results regarding age and perception of overdose risk, with a study conducted by this team finding older age to be associated with lower perceived risk of opioid overdose,⁶ but a more recent qualitative study finding that older individuals were more concerned with fentanyl overdose.⁸ Furthermore, a study found differences in risk perception across sex, gender, and race where identifying as female or Black race were associated with increased risk perception.⁹

We sought to explore overdose risk perception and correlates among a cohort of persons who had experienced overdose and used non-prescribed opioids at a time when fentanyl dominated the opioid supply. While this is distinct from identifying risk factors for overdose per se, we did incorporate known risk factors as covariates, such as stimulant use¹⁰, housing status^11–17, and use of medications for opioid use disorder (MOUD)¹⁸. We used data from a study with two geographically distant sites: San Francisco CA and Boston MA to describe findings in different US regions.

METHODS

Study Subjects and Data Source

We used baseline data from the REBOOT 2.0 study. REBOOT 2.0 is a phase III randomized controlled trial assessing the effectiveness of a repeated counseling intervention to reduce opioid overdose risk among individuals who use non-prescribed opioids, have a history of overdose within the past three years, and had previously received naloxone. Participants from San Francisco, CA and Boston, MA were recruited through both active (community locations such as syringe access sites, naloxone distribution sites, local community-based organizations, and substance use disorder treatment centers) and passive (distribution of posters, cards, and flyers) means and enrolled between April 2019 and June 2022. Participants completed baseline assessments (interviewer-administered surveys using REDCap^19,20) and were scheduled for follow-up visits every four months for 16 months. Participants were compensated between $40 and $100 for study visits depending on the visit type and $3 when they updated their contact information. For this analysis, we used data from the baseline assessment collected prior to any intervention. The study was approved by [identifying information].

Overdose Risk Perception Measures

The primary outcome of this analysis was derived from the question: “In the next 4 months, how likely is it that you will have an opioid overdose?” (Supplementary Appendix A). Likert scale responses were collapsed into a binary measure with “extremely likely” and “likely” combined as “likely” (higher risk perception) and “neutral,” “unlikely,” and “extremely unlikely” combined as “unlikely” (lower risk perception).

Covariates

Sociodemographic Characteristics

Sociodemographic covariates included study site (Boston, San Francisco), age, gender (male, female, transgender or other), race and ethnicity (combined into four mutually exclusive categories: “non-Hispanic White”, “non-Hispanic Black”, “Hispanic Latinx”, and “Other” which includes all participants not in the previous three categories), education (measured as the highest level of education completed and categorized as “less than graduating high school”, “graduating high school”, “some college”, and “graduating college or postgraduate education”), monthly income in the past year (from all legal and non-legal sources, after taxes), sleeping in a shelter or on the streets in the last four months, and relationship status (assessed using three categories: “single”, “with a partner/married”, “divorced/separated/widowed”).

Substance Use Covariates

Substance use covariates included past 30-day stimulant and opioid use, years since first non-prescribed opioid use, injection as primary route of opioid use, number of witnessed opioid overdoses in the past four months, receipt of MOUD in the past four months, severity of opioid dependence (SDS, a five-item scale that measures individuals’ dependence on different substances)²¹, and confidence in determining if one’s drugs contained fentanyl. For past 30-day stimulant use we combined responses of using either cocaine or methamphetamine in the past 30 days. For stimulant and opioid use (fentanyl, heroin, and other non-prescribed opioids) in the past 30 days, a question of any use and, among those reporting use, of how many days used, were combined to derive a categorical measure of “daily use”, “some use”, and “no use”. Years since first non-prescribed opioid use was calculated by subtracting age of first non-prescribed opioid use from age at enrollment. Confidence level in determining whether one’s drugs contained fentanyl before using was asked among people who reported using fentanyl in the past 30 days and categorized by Likert scale responses of “not at all”, “slightly”, “somewhat”, “moderately”, and “extremely”. MOUD was defined as reporting receiving methadone, buprenorphine, or naltrexone in the past four months.

Overdose Risk Behavior Covariates

The REBOOT baseline assessment asked questions about behaviors associated with both the risk of overdose occurrence (e.g., doing a tester shot) and overdose fatality (e.g., using alone). Our research question in this study was focused on the perception of risk of overdose occurrence, not on the perception of the risk of overdose death. Using alone or unmonitored is a mediator of overdose death, but not overdose occurrence, so we did not include it in this analysis. We included if the participant tested the strength of their opioids (e.g., tester shot), used opioids on the same day as alcohol or benzodiazepines, bought from a different dealer, or used in a new place that they had never used in before (i.e., different house, apartment, or public space) all in the past 4 months. These variables had Likert scale responses of “never”, “rarely”, “sometimes”, “often”, “always” which were combined into binary measures of “never/rarely” and “sometimes/often/always.” We categorized “sometimes” with “often” and “always” for risk behaviors because that response implied that the risk behavior was engaged in with some frequency. While participants were also asked about their use of fentanyl test strips in the past 4 months, due to the majority (88%) reporting never or rarely using them, we chose not to include this measure in the analysis.

Statistical Analyses

We performed bivariate analyses of perceived likelihood of overdose in the next four months (higher vs lower) across sociodemographic, substance use, and overdose risk behavior measures. We assessed skewness and kurtosis of continuous variables using normality tests. For normally distributed continuous data, we reported mean and standard deviation, and used t-tests for comparisons. When data were skewed, we reported medians and interquartile ranges (IQR) and used Wilcoxon rank-sum tests for comparisons. We reported frequencies and proportions for categorial data, using Pearson chi-square and Fisher’s exact (for expected cell counts less than 5) for comparisons.

We then conducted the following three separate multivariable logistic regression models, which grouped conceptually similar measures and had increasing levels of modifiability: (1) sociodemographic characteristics (site, age, gender, race and ethnicity, housing status, and witnessing an opioid overdose in the past four months), (2) substance use (site, past 30-day stimulant use, duration of illicit opioid use, past 4-month overdose, injection as primary route of opioid use, and severity of dependence), and (3) overdose risk behaviors (site, testing the strength of opioids, using opioids the same day as alcohol or benzodiazepines, buying from a different dealer, and using in a new place).

Given the exploratory nature of this analysis, we included variables in each model that were either selected a priori or were statistically significant at the p <0.05 level in the bivariate analyses. Site was selected a priori as a covariate in all multivariable models due to the drug market differences between Boston and San Francisco.^22–24 Age was included due to its known association with overdose risk perception.^6,8 Gender, race, reporting a recent overdose, recent injection drug use, and concurrent use of opioids and alcohol were included in the multivariable models due to their association with higher risk perception.^2,5,6,9 We also included cocaine and methamphetamine use as existing research has suggested that individuals who use cocaine, methamphetamine, or other stimulants perceive themselves as being susceptible to opioid overdose.²⁶ We decided to include years since first non-prescribed opioid use in model 2 and testing the strength of opioids in model 3 as they may be associated with risk perception based on formative work.²⁵

We did not include confidence level in determining whether one’s drugs contained fentanyl in our main multivariable models because this assessment was added after we began participant accrual and was only asked of a subset of participants. However, after we performed the main multivariable models, we conducted an additional model among this subgroup that included study site and confidence level to detect fentanyl in one’s drugs as independent variables to further explore how site may be related to opioid overdose risk perception. Effect measures were summarized using adjusted odds ratios (AOR). We conducted Hosmer-Lemeshow tests to assess the goodness of fit of each multivariable model. P-values <0.05 were considered statistically significant. All analyses were performed using Stata 17 statistical software (StataCorp, College Station, TX).

RESULTS

We analyzed data from 268 participants, 59% from San Francisco and 41% from Boston. Mean age was 43 years (SD = 10) and 62% reported male gender. The majority of participants were non-Hispanic White (61%) and 63% reported experiencing homelessness in the last four months. Most reported using fentanyl (87%) or heroin (81%) in the past 30 days and 94% reported stimulant use. Most (88%) participants reported at least one known overdose risk behavior in the past four months (using with alcohol or benzodiazepines, buying from a different dealer, using in a new place, or never or rarely testing the strength of opioids).

A minority of participants (21%) thought it was likely they would experience an opioid overdose in the next four months and 76% thought they would witness someone else overdose during that period. When asked about the previous four months, 47% reported often or always being worried about experiencing an overdose.

Bivariate Results by Overdose Risk Perception

Younger average age (42 years old vs. 43 years old, p <0.001), having witnessed more overdoses in the past four months (median 5 vs. 2, p <0.001), and higher SDS (median 11 vs 9, p <0.001) were associated with higher overdose risk perception [Table 1]. A higher proportion of Boston participants reported higher risk perception compared to San Francisco participants (31% vs. 13%, p = 0.001). A higher proportion of participants who had experienced homelessness (27% vs 10%, p <0.01) and those who experienced an opioid overdose (38% vs 10%, p <0.001) in the past four months also reported higher risk perception [Table 2]. Confidence in detecting fentanyl in one’s drugs before using them was inversely associated with risk perception (p <0.01).

Table 1.

Baseline sociodemographic characteristics among REBOOT study participants by opioid overdose risk perception. (N=268)

		Total (N=268)	Lower Risk Perception (n=213)		Higher Risk Perception (n=55)		p-value
		n	n	%	n	%
Site							<0.01
	San Francisco	157	136	87%	21	13%
	Boston	111	77	69%	34	31%
Age (years) (mean (SD))		43 (10)	43	10	42	9.8	<0.001
Gender							0.62
	Male	165	130	79%	35	21%
	Female	100	81	81%	19	19%
	Transgender or other	3	2	67%	1	33%
Race and ethnicity 1							0.72
	Non-Hispanic White	162	132	81%	30	19%
	Non-Hispanic Black	35	26	74%	9	26%
	Hispanic Latinx	39	30	77%	9	23%
	Other	29	22	76%	7	14%
Highest level of education							0.39
	Less than graduating high school	78	60	77%	18	23%
	Graduating high school	97	76	78%	21	22%
	Some college	67	53	79%	14	21%
	Graduating college or postgraduate education	26	24	92%	2	8%
Monthly income in the past year (US dollars) (median (IQR))		900 (400–1500)	906	423–1500	813	240–1100	0.32
Homeless in last 4 months							<0.01
	Yes	169	124	73%	45	27%
	No	99	89	90%	10	10%
Relationship status							1.00
	Single	138	110	80%	28	20%
	With a Partner/Married	82	65	79%	17	21%
	Divorced/Separated/Widowed	48	38	78%	10	22%

^1.Three participants declined to report race or ethnicity.

Table 2.

Baseline substance use and opioid overdose risk behaviors among REBOOT study participants by opioid overdose risk perception. (N=268)

		Total (N=268)	Lower Risk Perception (n=213)		Higher Risk Perception (n=55)		p-value
		n	n	%	n	%
Cocaine or methamphetamine/amphetamine use in past 30 days							0.51
	Daily use	86	67	78%	19	22%
	Some use	165	134	81%	31	19%
	No use	17	12	71%	5	29%
Years since first non-prescribed opioid use (median (IQR))		19 (12–27)	20	13–27	19	11–26	0.26
Opioid overdose in past 4 months							<0.001
	Yes	104	65	63%	39	38%
	No	164	148	90%	16	10%
Injection heroin, fentanyl, or other opioids as primary route of use							0.24
	Yes	182	141	77%	41	23%
	No	86	72	84%	14	16%
Confidence in determining whether product contains fentanyl before using it (n=191) 1							<0.01
	Not at all	21	10	48%	11	52%
	Slightly	23	18	78%	5	22%
	Somewhat	28	25	89%	3	11%
	Moderately	48	39	81%	9	19%
	Extremely	71	60	85%	11	15%
Fentanyl use in past 30 days							0.64
	Daily use	103	84	82%	19	18%
	Some use	131	101	77%	30	23%
	No use	34	28	82%	6	18%
Heroin use in past 30 days							0.11
	Daily use	79	60	76%	19	24%
	Some use	138	107	78%	31	22%
	No use	51	46	90%	5	10%
Other opioid use in past 30 days							0.51
	Daily use	6	4	67%	2	33%
	Some use	75	58	77%	17	23%
	No use	187	151	81%	36	19%
Number of witnessed opioid overdoses in the past 4 months, (median (IQR))		3 (1–6)	2	1–5	5	2–10	<0.001
MOUD receipt in past 4 months							0.64
	Yes	183	144	79%	39	21%
	No	85	69	81%	16	19%
Severity of Dependence score, (median (IQR))		9 (6–12)	9	6–11	11	8–13	<0.001
Tested the strength of drugs
	Never/Rarely	141	113	80%	28	20%	0.78
	Sometimes/Often/Always	127	100	79%	27	21%
Used the same day as using alcohol/benzodiazepines							0.14
	Never/Rarely	116	97	84%	19	16%
	Sometimes/Often/Always	152	116	76%	36	24%
Bought from a different dealer							0.03
	Never/Rarely	152	128	84%	24	16%
	Sometimes/Often/Always	116	85	73%	31	27%
Used in a new place (n=267) 2							<0.01
	Never/Rarely	156	135	87%	21	13%
	Sometimes/Often/Always	111	77	69%	34	31%

^1.57 missing responses and 20 participants did not report using fentanyl in the past 30 days.

^2.One missing response.

Multivariable Results

Participants at the Boston site compared to participants at the San Francisco site had more than two times the adjusted odds of reporting higher risk perception in all three multivariable models (model 1: AOR= 2.00, 95% CI: 1.04–3.86; model 2: AOR= 2.43, 95% CI: 1.21–4.88; model 3: AOR= 2.46, 95% CI: 1.29–4.69). In Model 1 (sociodemographic characteristics), those who experienced homelessness in the past four months had 2.41 times higher adjusted odds (95% CI: 1.10–5.30) of reporting higher risk perception [Table 3]. In Model 2 (substance use), experiencing an opioid overdose in the past four months and an increase of 1 on the SDS had 5.61 (95% CI: 2.83–11.12) and 1.49 times (95% CI: 1.04–1.28) the adjusted odds, respectively, of reporting higher risk perception [Table 3]. In Model 3 (overdose risk behaviors), using in a new place had 2.06 times the adjusted odds (95% CI: 1.05–4.05) of having higher risk perception [Table 3]. In a separate model assessing the association between site and risk perception (n=191), which included confidence level in detecting fentanyl in one’s drugs, participating at the Boston site was associated with 3.24 times (95% CI: 1.49–7.05) the odds of higher risk perception (Supplementary Appendix A).

Table 3.

Multivariable logistic regression of the odds of opioid overdose risk perception (lower risk perception vs. higher risk perception) in the next 4 months among REBOOT participants who reported illicit opioid use in the past 30 days at baseline. (N=268)

Characteristics	Adjusted Odds Ratio	95% Confidence Interval	p-value
Model 1: Sociodemographic Characteristics (n=260) 1
Boston site 2	2.00	1.04 – 3.86	0.04
Increase in 10 years of age 2	0.93	0.67 – 1.31	0.69
Female 2	0.90	0.46 – 1.73	0.74
Race and ethnicity 2
Non-Hispanic Black	1.83	0.72 – 4.66	0.20
Hispanic Latinx	1.38	0.57 – 3.33	0.47
Other	1.31	0.47 – 3.65	0.61
Spent any nights on the street or in a homeless shelter in the past 4 months	2.41	1.10 – 5.30	0.03
Witnessing an opioid overdose in the past 4 months (for every 1 increase in overdose witnessed)	1.01	0.99 – 1.02	0.28
Model 2: Substance Use
Boston site 2	2.43	1.21 – 4.88	0.01
Cocaine or methamphetamine/amphetamine use in the past 30 days 2	1.21	0.65 – 2.24	0.55
Years since first non-prescribed opioid use (for every 1-year increase) 2	1.00	0.97 – 1.03	0.91
Opioid overdose in the past 4 months 2	5.61	2.83 – 11.12	0.00
Injection heroin, fentanyl, or other opioids as primary route of use 2	1.49	0.70 – 3.15	0.30
Severity of dependence (for every 1-point increase on the scale)	1.16	1.04 – 1.28	<0.01
Model 3: Opioid Overdose Risk Behaviors
Boston site 2	2.46	1.29 – 4.69	<0.01
Tested the strength of drugs 2	1.14	0.61 – 2.14	0.67
Used the same day as using alcohol/benzodiazepines 2	1.11	0.57 – 2.16	0.75
Bought from a different dealer	1.62	0.84 – 3.13	0.15
Used in a new place	2.06	1.05 – 4.05	0.04

^1.Eight participants either declined to report race or ethnicity, reported a gender other than male or female, or did not report the number of overdoses witnessed.

^2.Variables selected a priori; all other variables were selected based on p <0.05 on bivariate analysis.

DISCUSSION

Among our sample of opioid overdose survivors, less than a quarter believed they would likely experience an overdose in the next four months, despite most reporting at least one overdose risk behavior. This aligns with the few studies on overdose risk perception concluding that overall risk perception was low despite involvement in behaviors known to elevate overdose risk.^3–5 These findings highlight an ongoing gap between overdose risk perception and known risk factors. We also identified correlates of risk perception which include participating at the Boston site, recent homelessness, using opioids in a new place, and recent opioid overdose which can inform overdose prevention practices.

Participants at the Boston site had increased adjusted odds of having higher perceived overdose risk across all multivariable models. Since no Boston participants who used fentanyl in the past 30 days reported primarily smoking it and the majority of San Francisco participants who had used fentanyl recently reported smoking as their primary route (59%), we thought this may be a differential perceived risk related to route of fentanyl use. However, once we adjusted for opioid injection in the second model, the Boston site still had significantly higher adjusted odds of higher risk perception. The difference between these sites may reflect the longer-term fentanyl exposure in that city compared to San Francisco, where fentanyl arrived about five years later.^23,24 We were able to measure years since first non-prescribed opioid use, however, our dataset did not include a specific measure of time since first fentanyl use.

A third possible explanation is the difference in the fentanyl markets in San Francisco and Boston, resulting in differential confidence in determining the presence of fentanyl in the drug supply. On the East Coast, it may be harder to differentiate fentanyl from heroin since white powder heroin, which is more commonly used, looks similar to fentanyl. In San Francisco, fentanyl is often easier to distinguish because it is a lighter color than black tar heroin, which is predominantly used on the West Coast.^22,23 In our study, participants who were less confident in determining if fentanyl was in their drugs before using them reported higher overdose risk perception. Based on the separate model assessing the association between site and risk perception adjusting for fentanyl confidence, the Boston site continued to have increased adjusted odds of higher risk perception suggesting that other unmeasured factors also contribute to this finding (Supplementary Appendix A).

Participants who reported being homeless in the past four months had increased adjusted odds of higher overdose risk perception. This finding adds to previous studies of overdose risk perception which either did not include homelessness in their analyses or did not find homelessness to be associated with risk perception.^3–5,7 Further, this finding aligns with prior studies that have demonstrated that homelessness is a risk factor associated with an increased risk of overdose.^11–17 Considering the increase in homelessness in both San Francisco (35% increase from 2019–2022²⁷) and Boston (42% increase from 2022–2023²⁸), housing status is a critical measure in urban areas. We also found using opioids in a new place to be associated with higher risk perception, consistent with rodent studies suggesting using opioids in a novel setting lowers baseline tolerance to opioids.²⁹ On the other hand, people who are homeless and then transition into individual housing may be at increased risk of fatal overdose due to increased likelihood of solitary drug use.³⁰ To confirm that we were assessing different constructs when measuring homelessness and using in a new place, we assessed the correlation between these variables and found only a weak positive correlation (r = 0.27). Harm reduction approaches such as the implementation of safe consumption sites have been associated with a reduction in fatal overdose^31–33 by providing a consistent, safe environment for individuals to use pre-obtained drugs,^31,34 an intervention that may benefit people who use opioids with regard to both overdose risk and risk perception.

We also found that participants who experienced an opioid overdose in the past four months had increased adjusted odds of having higher perceived overdose risk. Other studies on overdose risk perception reported similar findings.^2,5,7 Given the extensive data demonstrating that a prior overdose is associated with future overdose, these results may be a point of leverage for behavior change as the increased perception of risk may be paired with an increased willingness to adjust substance use behaviors in order to address that risk. In fact, the counseling intervention developed for the parent study of this analysis was designed with this premise in mind, including a review of participants’ witnessed and personal overdose history to identify potential overdose risks and inform risk reduction strategies.²⁵

In contrast to our previous analysis on correlates of opioid overdose risk perception⁶, injecting opioids was not a correlate of risk perception in this study. It is important to note that there may be inconsistencies across the measures of injection that may contribute to this difference. Our earlier study assessed an association with increased days of injection among people who use opioids, whereas the current study compared injection to other primary routes of use. It is possible that a relative increase in injection frequency is also related to overdose risk perception in our sample, which we would not detect comparing people who primarily inject opioids to others who primarily use other routes. In addition, a primary route of opioid use does not mean that an individual does not use opioids via other routes. In fact, someone who may primarily inject opioids may utilize other routes specifically in contexts where they anticipate increased overdose risk (e.g., a positive fentanyl test or unknown supplier), thus perceiving lower risk for overdose despite primarily injecting opioids. Since we only measured primary route of opioid use, we were unable to explore these nuances in more detail.

Limitations

This study has limitations that must be considered. First, as our inclusion criteria required participants to be naloxone recipients who use non-prescribed opioids and have had a prior opioid overdose, the results may not be generalizable to other populations, such as those who have not had a prior overdose or have not been exposed to overdose prevention programming. Though this population of overdose survivors who already have received naloxone and continue to use opioids is a key population at risk. Our findings may be influenced by recall or social desirability bias as the assessment was administered by research staff and responses were self-reported, relying on the participants’ memory. Lastly, due to the cross-sectional design of our study, temporal relationships could not be assessed and we did not explore whether risk perception was correlated with subsequent overdose occurrence.³⁵

Conclusion

Understanding opioid overdose risk perception can inform overdose prevention efforts. Our findings suggest that being unhoused and using opioids in new settings may be associated with greater perceived overdose risk, and that being in a site with longstanding fentanyl presence in the drug supply may be associated with a higher perceived overdose risk. A stable place in which to use substances under supervision may alleviate some of these perceived risks and the actual risk of overdose fatality.

HIGHLIGHTS.

• Among 268 participants, 88% reported at least one opioid overdose risk behavior.

• Only 21% perceived themselves to be at higher risk of overdose.

• Homelessness and using in a new place were associated with higher risk perception.

• Participating at the Boston site was associated with higher risk perception.