PRESCRIBE TO SAVE LIVES: AN INTERVENTION TO INCREASE NALOXONE PRESCRIBING AMONG HIV CLINICIANS

Abstract

Background:

Overdose is a major cause of preventable death among persons living with HIV. This study aimed to increase HIV clinicians’ naloxone prescribing, which can reduce overdose mortality.

Methods:

We enrolled 22 Ryan White funded HIV practices, and implemented onsite, peer-to-peer training, post-training academic detailing, and pharmacy peer-to-peer contact around naloxone prescribing in a non-randomized stepped wedge design. HIV clinicians completed surveys to assess attitudes towards prescribing naloxone at pre-intervention, 6- and 12-months post-intervention. Aggregated electronic health record (EHR) data measured the number of patients with HIV prescribed and the number of HIV clinicians prescribing naloxone by site over the study period. Models controlled for calendar time and clustering of repeated measures among individuals and sites.

Results:

Of 122 clinicians, 119 (98%) completed a baseline survey, 111 (91%) a 6-month survey, and 93 (76%) a 12-month survey. The intervention was associated with increases in self-reported “high likelihood” to prescribe naloxone [odds ratio (OR) 4.1 (1.7, 9.4), p= 0.001]. Of 22 sites, 18 (82%) provided usable EHR data which demonstrated a post- intervention increase in the total number of clinicians who prescribed naloxone [Incidence rate ratio 2.9 (1.1, 7.6), p = 0.03] and no significant effects on sites having at least one clinician who prescribed naloxone [OR 4.1 (0.7. 23.8), p = 0.11]. The overall proportion of all HIV patients prescribed naloxone modestly increased from 0.97% to 1.6% [OR 2.2 (0.7, 6.8), p = 0.16].

Conclusion:

On-site, practice-based, peer-to-peer training with post-training academic detailing was a modestly effective strategy to increase HIV clinicians’ prescribing of naloxone.

1.0. Introduction

Drug overdose is the leading cause of accidental death in the U.S.,¹ causing approximately 100,000 deaths in 2021,² and it contributes to significant and rising drug- related mortality among people living with HIV (PLWH).³ Among PLWH, approximately 40% have received at least one opioid prescription and commonly engaged in high-risk prescription opioid use.⁴ A meta-analysis of observational and experimental studies has shown that HIV-seropositivity is associated with nearly two times increased risk of opioid overdose mortality.⁵ While AIDS related mortality has been declining among PLWH due to the widespread availability of anti-retroviral treatment, in contrast, drug-related causes of death have increased more than four-fold since 2011.³

When witnessed, opioid overdoses, even with fentanyl, can be effectively reversed with the opioid antagonist naloxone ^6,7; thus, federal advisories have advocated for widespread naloxone education and use⁸ and the Biden-Harris administration have featured naloxone access in its drug policy priorities.⁹. Yet, clinicians have consistently shown gaps in knowledge about naloxone and patient risk assessment,^10,11 and, as a result, naloxone remains grossly under prescribed- with merely 1.5% of commercially-insured, high-risk patients prescribed this lifesaving medication.¹² Clinicians can be trained to prescribe naloxone to at risk patients.¹³ When naloxone is co-prescribed to patients treated with chronic opioid therapy, the patients have reduced emergency department use.¹⁴ Given the high rates of opioid use and high risk of overdose among PLWH, HIV outpatient practices are promising settings in which to implement interventions that target HIV clinicians to treat opioid use disorder (OUD) and to expand naloxone prescribing so as to reduce overdose.

The Prescribe to Save Lives (PtSL) study sought to leverage the high priority that HIV clinicians place on providing lifesaving treatment to persons with HIV, such as achieving viral suppression through adherence to anti-retroviral treatment,¹⁵ to increase the provision of lifesaving interventions around overdose prevention and OUD. Using implementation strategies of peer-to-peer training, proactive expert support, academic detailing with motivational interviewing, and addressing environmental barriers, we sought to increase naloxone prescribing among HIV clinicians as a mechanism to increase their engagement in addressing their patients’ overdose risk and OUD. We hypothesized that HIV clinicians who received the implementation intervention would report greater interest in, confidence in, readiness to and commitment to prescribing naloxone. Furthermore, we hypothesized that, following the intervention roll-out, more HIV clinicians would prescribe naloxone and more patients would receive it.

2.0. Methods

2.1. Trial Design and Site Recruitment

We employed a stepped wedge implementation study design that enrolled 22 Ryan White funded HIV practice sites from 18 states that had 500 or more PLWH and were in the top 30 states nationally for opioid overdose deaths in the four years prior to study initiation (2013-2016). Although a randomized stepped wedge design was originally planned, challenges with the upfront recruitment of all sites necessitated moving to a design in which recruitment order determined implementation order. Site recruitment occurred on an ongoing basis from January 2017 to July 2018. After site recruitment, lead administrators provided a list of clinicians (including physicians and advanced practitioners (Nurse Practitioners and Physician Assistants). Study staff contacted the clinicians to obtain informed consent and to explain the protocol and its procedures, requirements, risks, and potential benefits. Implementation was rolled out in three waves, the first starting in August 2017 and six months between subsequent wave starts. The Institutional Review Boards of Baystate Medical Center and the Lifespan Hospitals reviewed and approved the study.

2.2. Data Collection

Participating clinicians at all sites completed surveys immediately before their initial training visit and again approximately 6- and 12-months post intervention at their site. Participating clinics were asked to provide aggregate naloxone prescribing data from their electronic health record (EHR) for the years 2017 to 2019 at the end of the study.

2.3. Intervention

The intervention included the following elements (Figure 1):

Figure 1.

Schematic of the Prescribe to Save Lives Intervention and Outcomes.

Onsite, Peer-to-Peer Training.

A 1.5-hour interactive training session provided HIV clinicians with the epidemiological rationale and empirical evidence for prescribing naloxone to their patients. The session trained clinicians to identify patients at high risk for overdose, especially those using opioids, and taught them to discuss the patient’s individualized overdose risk and prescribe naloxone, regardless of other substance use. Most sessions were scheduled during lunchtime and catered to maximize clinician participation.

Post-training Support.

At the initial training, the study team’s expert physician offered mentorship support to site clinicians and provided contact information. The study team reached out to the site clinicians within 2 weeks after the training during the implementation period to discuss questions or concerns about prescribing naloxone.

Academic Detailing.

The study team developed detailing manuals, brochures, and patient materials promoting naloxone. Study detailers with training in motivational interviewing visited the clinicians and provided these materials. Detailing visits occurred 1, 3, and 5 months after the initial training either in-person or via telephone.

Pharmacy Peer-to-Peer Contact to Ensure Naloxone Availability.

Prior to a site’s initial training, a study team pharmacist sent a personalized letter to pharmacists at HIV clinic-recommended pharmacies to explain the study and prepare them to fill an increased number of naloxone prescriptions. In addition to the letter, each pharmacy was contacted by phone to recruit pharmacists to the onsite training and to verify naloxone stocking.

2.4. Outcomes

Prescribing intent.

To assess intent to prescribe, we examined items drawn from the Opinions about Medication Assisted Treatment (OAMAT) instrument¹⁶ and four separate attitude scales developed for our survey. Intent to prescribe was derived from the question “how likely are you to prescribe naloxone?” with responses rated on a 5-point Likert scale ranging from 1: Not at all/None to 5: A lot/Very much so. We dichotomized responses into positive (4 or 5 rating) vs. neutral/negative (1-3 rating). For the attitude scales, participants provided unanchored ratings from 0 – 100 of their interest in, confidence in, readiness to, and commitment to prescribing naloxone.

Prescribing uptake.

To assess uptake, we analyzed aggregate EHR data from participating clinics which included total number of patients with HIV, number of patients with HIV who were prescribed naloxone, and number of HIV clinicians prescribing naloxone. Data was provided for 5 time periods spanning the length of the study, with each site providing data from before and after the intervention. These data generated three outcomes: the number of clinics with any naloxone prescriber, the mean number of clinicians in a clinic who prescribed, and the proportion of all HIV patients who were prescribed naloxone.

Clinician characteristics.

Clinicians self-reported their demographics including gender, race, ethnicity, age, and education level. They also reported their specialty, number of years working with patients with HIV, hours/week and percent time performing patient care for patients with HIV, and number of patients with HIV in the clinician’s care.

2.5. Statistical Methods

Descriptive statistics for clinician characteristics, as well as site prescribing numbers, were reported using means and standard deviations or frequencies and percentages. Marginal models were fit to determine intervention effects using both individual prescriber survey and site-aggregated EHR data. Generalized estimating equation (GEE) population-averaged models included both indicator variables for intervention (pre/post) and the five calendar time periods. The calendar time periods were coded as a series of categorical indicators to allow for non-linear trends, with the first coinciding with pre-wave one baseline survey administration and the last coinciding with the third wave’s final follow up. Models included a repeated statement with subject=prescriber for survey data and site for aggregate clinic EHR data. Each model used an appropriate link function for the given outcome: logit for binary, log link for count outcomes and identity for continuous outcomes.

3.0. Results

From the 22 participating HIV practice sites across 18 states in the continental US, this analysis included the 122 clinicians who were able to prescribe medications, (Table 1). The number of clinicians ranged from 2 to 27 per site, with a mean of approximately 5 per site. Most were female (72%) and white (62%), with a mean age of 43. The sample was comprised of 48% physicians and 49% advanced practitioners. On average, clinicians had been working with patients with HIV for almost a decade and had 155 patients with HIV under their direct care. Of the 122 clinicians, 119 (98%) completed the baseline survey, 111 (91%) a 6-month survey, 93 (76%) a 12-month survey, and 89 (73%) all 3 surveys. Only 18 (82%) of 22 sites provided usable EHR data regarding naloxone prescriptions during the study period.

Table 1.

Characteristics of prescribing clinicians at 22 Ryan White funded HIV clinics, August 2017-December 2019 (n=119)

Demographics	N=119
Female, n (%)	86 (72.3)
Hispanic/Latinx, n (%)	12 (10.1)
Race, n (%)
White	74 (62.2)
Black	22 (18.4)
Other	22 (18.4)
Age (years), mean (standard deviation)	43.3 (9.7)
Practitioner type, n (%)
Physician	57 (47.9)
Nurse Practitioner	49 (41.2)
Physician Assistant	9 (7.6)
Missing	4 (3.4)
Practice, mean (standard deviation)
Years working with HIV pts	9 (8)
Hours per week in patient care	35 (12)
Percentage of time in HIV patient care	45 (37)
Number of direct care HIV patient in the panel	155 (173)

3.2. Naloxone prescribing.

Models predicted significant increases on 0-100 scales post-intervention compared to pre-intervention for interest (point estimate of mean difference [diff], 10.2; 95% confidence interval [CI] 2.3 to 18.2), confidence (diff 31.1; 95% CI 22.0 to 40.3), readiness (diff 24.4; 95% CI 14.8 to 34.1) and commitment (diff 13.2; 95% CI 4.6 to 21.8) to prescribing naloxone, all p ≤ 0.01 (Figure 2). Self-reported intent to prescribe naloxone also increased post- intervention. The odds ratio (OR) for a clinician reporting being “highly likely” to prescribe naloxone post vs. pre-intervention was 4.1 (95% confidence interval (CI) 1.7, 9.4), p=0.001.

Figure 2. Pre- vs post-intervention mean change in score of clinician attitudes towards prescribing naloxone.

Derived from continuous 0-100 scales on clinician prescriber surveys. The total N for any survey participation is 122, with 119 clinician prescribers (98%) completing the baseline survey, 111 (91%) the 6-month survey, 93 (76%) the 12-month survey, and 89 (73%) all 3 surveys.

Abbreviations: Confidence Interval- CI

In the raw EHR data, clinicians during during pre-intervention time periods saw 44,607 patients with HIV across all enrolled sites, and 66,756 patients during post-intervention time periods. Among these patients with HIV, 433 (0.97%) were prescribed naloxone pre-intervention, and 1062 (1.6%) were prescribed naloxone post-intervention. The number of sites that had at least one clinician prescribing naloxone was 9 (47%) pre-intervention and 14 (74%) post-intervention. The mean number of clinicians per site prescribing naloxone was 1.89 before the intervention and 4.16 afterwards.

After the intervention, the number of clinicians prescribing naloxone increased almost three-fold (incidence rate ratio [IRR], 2.9, 95% CI 1.1, 7.6, p = 0.03) (Table 2). The intervention had no detectable effect on whether a site had any naloxone prescribers (Table 2). The model-predicted percentages of patients prescribed naloxone remained small, increasing slightly from 0.8% (95% CI 0.2, 3.4) pre-intervention to 1.8% (95% CI 0.7, 4.7) post-intervention, with an odds ratio of 2.2 (95% CI 0.7, 6.8), p = 0.16. No trends by time period were detected.

Table 2.

Naloxone Prescribing per Electronic Health Record (EHR) Data

	Clinic with Any Naloxone Prescriber		Mean Number of Clinicians Who Prescribed		Proportion of all HIV Patients Prescribed Naloxone
	IRR (95% CI)*	p value	Odds ratio (95% CI)*	p value	Odds ratio (95% CI)*	p value
Post- vs. Pre-Intervention	2.9 (1.1, 7.6)†	0.03	4.1 (0.7, 23.8)‡	0.11	2.2 (0.7, 6.8)§	0.16
Time period
01/01/2017 – 08/09/2017	ref		Ref		ref
08/10/2017 – 01/24/2018	0.7 (0.5, 1.1)		0.9 (0.6, 1.5)		0.9 (0.7, 1.2)
01/25/2018 – 08/10/2018	0.6 (0.3, 1.2)		1.3 (0.3, 4.9)		0.59 (0.3, 1.3)
08/11/2018 – 02/10/2019	0.5 (0.2, 1.4)		1.5 (0.3, 8.5)		0.6 (0.2, 1.9)
02/11/2019 – 12/31/2019	0.7 (0.3, 1.8)	0.38	1.2 (0.2, 5.9)	0.78	0.7 (0.2, 2.2)	0.32

^*Incidence rate and odds ratios from generalized estimating equation models.

^†Of 18 clinics with useable EHR data , 7 (39%) had one or more clinician who prescribed naloxone in the pre-intervention time periods versus 13 (72%) clinics post-intervention.

^‡The mean number of naloxone prescribers per clinic was calculated by taking the highest number of prescribers in the pre- and post-intervention time periods for each clinic then dividing by number of clinics. Pre-intervention clinics had a mean of 1.9 naloxone prescribers versus 4.2 post-intervention.

^§Among 44,607 HIV patients seen during pre-intervention time periods, naloxone was prescribed to 433 (0.97%) versus 1062 (1.6%) of 66,756 during post-intervention time periods.

4.0. Discussion

In HIV practice settings, an implementation strategy of on-site, peer-to-peer training with post-training academic detailing was associated with greater interest, readiness, confidence, commitment and intent to prescribe naloxone to patients at risk for opioid overdose. The intervention tripled the total number of clinicians who prescribed naloxone. While the proportions of sites having at least one naloxone prescriber and the number of patients prescribed naloxone increased, these changes were not statistically significant.

The study intervention presupposed that an ethos of lifesaving derived from HIV clinicians’ experience with antiretroviral therapy could extend to naloxone – viz., naloxone’s safety and direct lifesaving effect would provide a meaningful starting point for HIV clinicians to address their patients’ OUD and risk of opioid overdose. Despite the impressive relative increase in prescribing, the numbers of patients with HIV who received naloxone was less than 2% even post-intervention, making the absolute improvement of modest clinical significance. A cross-sectional study from two safety-net hospitals showed that 10% of patients living with HIV on chronic opioid therapy ever received naloxone.¹⁷ While our study did not identify HIV patients who received chronic opioid therapy and lacked information about the proportion of patients at highest risk (e.g. those with a prior overdose or concomitant use of opioids with sedative-hypnotics), our findings are similar and suggest that naloxone receipt among patients with HIV and prescribing by HIV clinicians is uncommon, and opportunities to expand on interventions like ours are critically needed.

Even though the intervention modestly improved EHR-documented naloxone prescribing, clinicians had improved attitudes towards and increased self-reported naloxone prescribing. The reasons why attitudes and knowledge change alone are insufficient to modify clinician prescribing behavior are uncertain, and likely include complex logistical barriers, fear of stigmatizing or offending patients, and unfounded concerns about compensatory increases in risk behaviors by patients.¹⁰ Furthermore, while HIV clinicians in our study had generally favorable attitudes towards prescribing and recognized its value, they have historically been less comfortable engaging in risk/benefit conversations surrounding harm reduction as a result of constrained clinic visit time and their perceived limited role in overdose counseling.¹⁸ Future research should employ qualitative methods to understand the disconnect between HIV clinicians’ reported willingness to prescribe naloxone and actual prescribing. Our lessons-learned from this study suggest that modifications to the intervention should examine system changes to overcome logistical barriers, such as diversifying healthcare clinician support with pharmacist or other ancillary clinical staff ¹⁹ or automated EHR decision support to co-prescribe naloxone with opioids.²⁰

Prior efforts to bolster HIV clinicians’ engagement with their patients’ OUD have had similar modest impacts. One of the lessons of the Health Resources and Services Administration Special Projects of National Significance Program Integrating Buprenorphine Therapy into HIV Primary Care Settings (BHIVES) was that HIV clinicians were reluctant to engage in the care of persons with OUD ^21,22. Barriers included limited training and skills in recognizing and diagnosing OUD²³ and lack of confidence and discomfort addressing drug use.²² Our intervention provided a 1.5-hour introduction to overdose prevention with naloxone and was not a comprehensive review of OUD management; thus, HIV clinicians may have felt uncomfortable knowing that their overdose prevention discussions may necessitate more nuanced counseling and management of OUD with which they were less experienced. Given the dearth of clinician education in management of substance use disorders, another lesson-learned is that interventions such as ours might benefit from coupling with longitudinal exposure to low barrier trainings that can give clinicians tools to support more nuanced management of OUD.

The strengths of this study include its national, multisite scope and replicable initial training and academic detailing visits. Several limitations merit comment. First, this study was conducted in Ryan White funded HIV clinics and may not be generalizable to all outpatient clinics that provide HIV care. Second, the non-randomized implementation rollout design might have introduced bias in that sites recruited earlier contributed less time in the control condition and might have had greater enthusiasm for the intervention than sites enrolled later. Third, HIV clinician surveys were subject to recall and social desirability bias. Fourth, the analyses cannot control for patient-level factors, such as types and severity of substance use, which might have affected whether the clinicians felt naloxone prescribing was necessary. Fifth, we had limited information about community naloxone distribution without a prescription, except that no site distributed naloxone at the time of the initial training; increased community naloxone distribution over the study period might have reduced direct prescribing and biased our findings toward the null. Sixth, electronic health record (EHR) data from four clinics could not be obtained or was unusable; if clinicians from sites that provided usable EHR data had greater response to the intervention, this data loss may have biased results. To examine for possible information bias, we performed analyses examining the interest, confidence, commitment and intent to prescribe of clinicians at sites that provided usable EHR data versus those that did not [data available on request]. Notably, the four sites that did not provide usable EHR data had less favorable attitudes at baseline and greater increases in response to the intervention than did sites that provided usable EHR data. This analysis suggests that sites with usable EHR data did not have greater responsivity to the intervention; if anything it appears possible that this data loss biased EHR-based findings towards the null. Finally, our use of EHR data to evaluate the absolute number of naloxone prescriptions could not capture whether HIV clinicians discussed overdose prevention and naloxone prescribing with the patient and the prescription was declined because the patient already had it, preferred to get it from another community source, was concerned about co-payment costs or the pharmacy did not stock it.²⁴

5.0. Conclusions

Despite these limitations, the implementation strategy of on-site, peer-to-peer training with post-training academic detailing with among HIV clinicians was associated with more functional attitudes toward prescribing naloxone and enhanced intent to prescribe naloxone. Although naloxone prescribing almost tripled, fewer than 2% of HIV patients received it. Although certainly a modest absolute change, this study cannot discern whether such a change is clinically significant because we do not know the characteristics of the patients who received naloxone; even a modest change would be clinically significant if the intervention led clinicians to consider naloxone for patients at the highest risk of overdose. In order to augment naloxone prescribing further, future modifications to the intervention should address clinician and systemic barriers to engagement with persons with OUD, and examine innovations to enhance support for naloxone prescribing in HIV care settings.