Differences in Benefits of Office Based Addiction Treatment: Secondary Analyses Across Sub-groups in the PROUD Randomized Controlled Implementation Trial

Abstract

Background and Aims:

Female, Hispanic, and Black patients with opioid use disorder (OUD) are less likely to receive OUD medication treatment. The PROUD trial demonstrated that implementation of primary care (PC) nurse care management increases OUD medication treatment compared to usual care (UC). This study assessed whether the PROUD intervention’s effect differed across sex, race and ethnicity.

Design:

Secondary analyses of cluster-randomized implementation trial.

Setting.

12 PC clinics (2 per health system) randomized to UC or intervention, stratified by health system.

Participants.

PC patients 16–90 years old.

Intervention.

Three strategies to implement office-based addiction treatment (OBAT) by nurse care managers: (1) full-time nurse salary; (2) nurse training and technical assistance from expert nurses at Boston Medical Center; (3) ≥3 PC providers willing to prescribe buprenorphine. Nurses were trained in the Massachusetts model of OBAT which includes lowering barriers to OUD treatment, assessing and educating patients, supporting initiation of medications for OUD, and providing ongoing medical management, in collaboration with PC providers.

Measurements.

The primary outcome was a clinic-level measure of OUD treatment defined as patient-years of OUD treatment per 10,000 PC patients based on orders and procedures for buprenorphine or extended-release naltrexone from electronic health records and insurance claims (hereafter “OUD treatment”).

Findings.

The mean numbers of patients seen by intervention and UC clinics at baseline were 18,485 and 22,557, respectively. Female patients comprised 60% of the total PC population in intervention clinics and 64% in UC clinics; Asian, Black, Hispanic, or smaller racial groups comprised 61% of the PC population in intervention clinics, and 70% in UC clinics. Compared with UC, the intervention increased OUD treatment for male patients (adjusted difference: 13.7 patient-years; 95% CI: 5.8, 21.7), but not female patients (2.9; −4.3, 10.2); effect modification test, F(1,14)=4.77; p=0.046. Exploratory analyses suggest that differences in the intervention’s effect on receipt of any OUD treatment in female and male patients, rather than differences in the duration of OUD treatment, may account for findings. There was no significant effect modification by race or ethnic group (effect modification test F(4,44)=1.50; p=0.218).

Conclusions.

PC clinics that implemented OBAT by nurses increased patient-years of OUD treatment in male, but not female patients. Exploratory findings suggest that differences in the proportion of patients treated for OUD, rather than differences in the duration of OUD treatment, accounted for observed differences across groups.

INTRODUCTION

Medication treatment of opioid use disorder (OUD) is critical for addressing the opioid epidemic (1, 2), can be lifesaving (3–5), and can be provided in primary care (PC). However, the majority of PC patients with OUD do not receive treatment (6). In the United States (US), female (7–11), Black, and Hispanic patients with OUD (12–17) are less likely to receive OUD medication treatment (18–21), reflecting structural, interpersonal and/or internalized sexism and racism (18, 22).

Efforts to increase OUD treatment often focus on PC (23–25). One such program—the Massachusetts model of office-based addiction treatment (OBAT)—uses a nurse to lower barriers to OUD medication treatment in PC (26–28). However, little is known about whether such efforts to increase OUD treatment in PC can help overcome disparities in OUD treatment related to sex, race and ethnicity (28–33). No research to our knowledge has evaluated whether implementation of OBAT or other PC OUD treatment programs increases OUD treatment equally across sex, race and ethnicity.

The PROUD trial (29) demonstrated that the Massachusetts model of OBAT increased a broad clinic-level measure of PC OUD treatment (34)—days of OUD treatment with buprenorphine or long-acting naltrexone per 10,000 PC patients—which reflects both the number of patients who received any OUD treatment (“reach”) as well as the number of days of treatment per patient. This report presents pre-planned secondary analyses of the PROUD trial, testing whether the PROUD intervention had differential effects on the primary trial outcome across sex, race and ethnicity. Based on observational studies from Massachusetts, our a priori hypothesis, as outlined in the PROUD protocol statistical analysis plan (Supplemental Text A) was that receiving care in a clinic randomized to the PROUD intervention, compared to a clinic offering usual PC, would increase OUD treatment in female more than male patients, and in patients of Hispanic ethnicity or Black race less than others (26–29). A secondary aim was to conduct exploratory patient-level analyses to examine additional measures of OUD treatment across sex, race and ethnicity to provide context for clinic-level primary findings.

METHODS

Study design

PROUD was a hybrid type-III pragmatic cluster-randomized implementation trial, with sample characteristics and results after 2 years (primary outcome) and 3 years reported previously (29, 34–37). PROUD was conducted in six health systems across 5 states (NY, FL, MI, TX, 2 in WA). The primary trial and the secondary outcomes presented here were evaluated from March 1, 2018-February 29, 2020 (prior to the COVID-19 pandemic). Each system selected two clinics, or clusters of nearby clinics, with over 10,000 PC patients (hereafter “clinics”), that were randomly assigned to receive the PROUD intervention or continue usual care (UC). The trial sample and all outcome measures were obtained from secondary electronic health record (EHR) data, including insurance claims in two systems. The study was approved by a single institutional review board with waivers of consent and HIPAA authorization.

PROUD intervention: Implementation of the Massachusetts model of OBAT

The PROUD trial implementation intervention was designed to test whether the Massachusetts model of OBAT would increase OUD treatment in diverse PC clinics outside Massachusetts, and the strategies to implement the model were purposely designed to be simple: (1) providing funding to hire a full-time nurse care manager to support OUD treatment (“nurse” hereafter), (2) providing training and support for the nurses, and (3) having ≥3 PC providers agree to prescribe buprenorphine. Nurses were trained in delivery of OBAT and received support and technical assistance weekly in joint virtual meetings from expert nurses at Boston Medical Center (BMC) (29). The OBAT model implemented in PROUD (38) includes collaboration between a nurse and PC providers. The nurse screens, assesses and educates patients, supports initiation of medications for OUD, and provides ongoing management (e.g., monitoring adherence, substance use and recovery, placing refills for providers to sign, and alerting providers to changes in status). The provider confirms diagnoses, prescribes buprenorphine or long-acting naltrexone, and medically evaluates patients. This allows PC providers to treat OUD in the normal flow of PC. While not designed specifically to overcome disparities, the model and nurse training were designed to improve reach of and retention in OUD treatment. Nurses were trained to address stigma, educate staff (e.g., pharmacy, nursing, other clinical services), be easily accessible (e.g., dedicated OBAT phone), build connections to referral sources within and outside health systems (e.g., women’s clinics, emergency departments, criminal-legal organizations), assess craving, collaborate with providers to ensure adequate buprenorphine doses and timely refills, and provide linkage to resources (e.g., for transportation, housing).

Sample and data collection

PROUD was an open-cohort trial, and the trial sample included patients 16–90 years old who visited one of the 12 trial PC clinics from 3 years before to 2 years after randomization (February 28, 2018, except one system delayed six months) (29, 34). Data from care received anywhere in the participating health systems were collected for baseline measures during 2 years before randomization, and for follow-up measures during 2 years after randomization in 5 health systems (1.5 years in the system that randomized 6 months late) (34).

Measures

Sex, race and ethnicity

Sex, race and ethnicity were obtained from EHR data. Patients were categorized into 5 mutually exclusive subgroups based on ethnicity or race. Some health systems did not document race if Hispanic ethnicity was documented (making race data unavailable for those patients), so race and ethnicity could not be reported separately. Therefore, first, patients with documented Hispanic ethnicity were categorized as Hispanic, then all non-Hispanic patients were categorized as their documented race if a single race was documented, and otherwise “multiple race.” Our primary race and ethnicity measure included Hispanic, Asian, Black, White, and smaller racial groups combined (American Indian/Alaska Native, Native Hawaiian/Pacific Islander, multiple race, other).

Primary clinic-level outcome for OUD treatment: Patient-years of OUD treatment

The primary outcome for the PROUD trial and these secondary analyses was a clinic-level measure of patient-years of OUD treatment with buprenorphine or injectable naltrexone (“OUD treatment” hereafter), defined—within each subgroup—as the number of patient-days of OUD medication treatment at the clinic level, scaled per 10,000 PC patients seen post-randomization, and converted to patient-years of treatment for interpretability. This measure was obtained from medication orders or procedure codes in EHR or claims over 2 years follow-up; treatment with naltrexone also required documented OUD and/or opioid overdose diagnosis to exclude alcohol use disorder treatment (39).

Secondary patient-level outcome measures of different dimensions of OUD treatment

Patient-level outcomes reflecting different dimensions of OUD treatment were used to help interpret findings. These included a binary measure of reach: receipt of any OUD treatment during follow-up; two measures related to the duration of treatment: ≥80% adherence (i.e., having medications available ≥80% of days after initiation) (40, 41) and 6-month retention (i.e., having medications for ≥180 days); and a binary measure of whether the highest buprenorphine dose received was at least 16 mg: ≥16 mg/day buprenorphine, since higher doses have been associated with improved outcomes (34, 42).

The Massachusetts model of OBAT attracts new patients into PC (27, 34). To explore whether OBAT attracted patients into PC clinics for OUD treatment differently across sex, race and ethnicity, all patients seen in a PC clinic and treated for OUD post-randomization were categorized into one of four mutually-exclusive groups based on whether they received new or ongoing OUD treatment, and whether they were new to the PC clinic or health system post-randomization: Ongoing OUD Treatment, patients seen in health system pre-randomization and received OUD treatment pre-randomization (with no gap in OUD treatment >365 days); New OUD Treatment & Existing Patient, patients seen in PROUD trial clinics pre-randomization and received new OUD treatment post-randomization (none in the health system in the prior 365 days); New OUD Treatment & New to Clinic, patients new to PROUD trial clinics post-randomization but not new to health system, and received new OUD treatment; and OUD Treatment and New to Health System, OUD treatment in patients new to the health system post-randomization (i.e., not seen in system pre-randomization, making it unknowable if the treatment was new or ongoing).

Descriptive characteristics

Baseline sample characteristics included age in years, insurance status (Medicare, Medicaid, otherwise insured [e.g., commercial, private], uninsured, unknown), housing instability based on International Classification of Disease (ICD) billing codes (e.g., Z59.81), and neighborhood-level socioeconomic measures based on patients’ residential zip code (median household income, percent unemployed, and percent below federal poverty level) (43).

Analyses

Baseline sample characteristics of all PC patients in intervention and UC clinics were described at the patient level and across groups based on sex, race and ethnicity. Clinic-level analyses estimated mean patient-years of OUD treatment at baseline and follow-up by study arm (6 intervention and 6 UC clinics) in each subgroup defined by sex, race and ethnicity. This was especially of interest given main trial results had revealed UC clinics provided more OUD treatment at baseline (mean 10.6 vs 6.8 patient-years per 10,000 patients) (34).

Main effect modification analyses, conducted at the clinic level consistent with the main trial, were pre-specified and tested whether sex (binary), and race and ethnicity (5 subgroups) significantly interacted with the intervention’s effect on the primary outcome, including stratified analyses comparing intervention and UC clinics (Supplemental Text A). Analyses of interaction were anticipated to be under-powered (Supplemental Text A), but included per National Drug Abuse Treatment Clinical Trials Network (NIDA CTN) policy (44). As above, the a priori hypothesis was that the intervention would increase the primary outcome measure of OUD treatment more in female patients than male patients and less in Hispanic and Black patients than White patients based on observational studies (two-sided test, 0.05 level) (28, 29).

Consistent with the trial protocol (29, 34), main interaction analyses were conducted in all PC patients (rather than those diagnosed with OUD) to avoid bias due to potential differences in OUD diagnosis across trial arms or sociodemographic subgroups, and reported per 10,000 PC patients. Linear mixed-effects regression models were used, as in main trial analyses (34), adding an interaction term between (1) sex or (2) race and ethnicity and the intervention, excluding patients who were missing sex (0.001%) or race and ethnicity (4.7% intervention clinics and 6.1% UC clinics) from analyses, respectively. Models included a site-specific random intercept to account for correlation within health system, and adjusted for the baseline value of the outcome in the subgroup of interest (e.g., female, male); no other covariates were included, based on the approach for the main trial (which implemented a variable selection approach using baseline data to identify covariates prognostic for the outcome; see details in Supplemental Text B).

Two sensitivity analyses were conducted post hoc: the first resulted from an interest in investigating the difference in results when a clinic-specific random intercept was used in addition to the initial health system-specific random-intercept to account for further correlation (Model 2); the second added a second interaction term to Model 2 (interaction of demographic subgroup with the baseline value of the outcome) to account for observed imbalance of the baseline measure across trial arms within subgroups (Model 3).

Patient-level Descriptive Exploratory Analyses.

Post-hoc patient-level exploratory analyses were conducted to describe differences in secondary patient-level measures of OUD treatment across trial arms in subgroups of female and male patients and the 3 largest racial or ethnic groups. These analyses did not test for statistical significance and are instead presented descriptively for the information they provide about different dimensions of OUD treatment across the subgroups. The subsample of patients with documented OUD diagnoses was used to examine if observed differences persisted in those with documented recognition of OUD (34, 42). Other analyses of outcomes used to evaluate quality of care (e.g., ≥ 80% adherence and buprenorphine dose ≥16mg) were restricted to those with new treatment to avoid bias due to the greater numbers of patients with treatment before randomization in UC clinics. Analyses also describe—among those who received OUD treatment—whether treatment was new (if known) and whether they were new to the health system or PC clinic (four mutually-exclusive groups defined above). For these descriptive patient-level analyses, because subgroups were much smaller than 10,000, results are reported as percents (rather than per 10,000 PC patients).

Analyses were conducted in R version 4.3.1.

Role of the funding source and trial registration

The NIDA CTN funded the PROUD trial. Dr Liu, NIDA CTN scientific officer, collaborated on all aspects of the trial and contributed to this manuscript, consistent with his role. The trial was registered at Clinicatrials.gov on January 16, 2018 (#NCT03407638).

RESULTS

Table 1 describes patient characteristics at baseline, with 18,485 and 22,557 patients seen on average in intervention and UC clinics, respectively. The populations were demographically similar. Descriptive clinic-level analyses of the primary outcome showed that days of OUD treatment per 10,000 patients were lower for female than male patients at baseline, with OUD treatment imbalanced across arms (Supplemental Table A). Similarly, descriptive clinic-level analyses revealed marked differences in OUD treatment across race and ethnicity at baseline, with Hispanic, Asian, Black, and smaller racial groups receiving less OUD treatment than White patients, again with imbalance across arms (Supplemental Table A).

Table 1:

Baseline characteristics of primary care patients in PROUD intervention and usual care clinics

	PROUD Intervention	Usual Care
	(n = 109,196)a	(n = 133,170)a
	n (%)	n (%)
Age, years
16–25	12489 (11.4)	16715 (12.6)
26–45	31862 (29.2)	43625 (32.8)
46–65	43964 (40.3)	51075 (38.4)
66–90	20881 (19.1)	21755 (16.3)
Femaleb	65566 (60.0)	84564 (63.5)
Race and ethnicity
Hispanic	30566 (28.0)	44415 (33.4)
Asian	5347 (4.9)	7544 (5.7)
Black or African American	21045 (19.3)	27237 (20.5)
White	42981 (39.4)	39952 (30.0)
Smaller racial groups
American Indian or Alaska Native	557 (0.5)	598 (0.4)
Native Hawaiian or Pacific Islander	435 (0.4)	937 (0.7)
Multiple races	369 (0.3)	721 (0.5)
Other	2730 (2.5)	3609 (2.7)
Unknown	5166 (4.7)	8157 (6.1)
Insurance status closest to randomization
Medicare	23830 (21.8)	24327 (18.3)
Medicaid	31719 (29.0)	50947 (38.3)
Otherwise insured (e.g., commercial, private)	61691 (56.5)	67876 (51.0)
Uninsured	5297 (4.9)	6211 (4.7)
Unknown	811 (0.7)	1121 (0.8)
Patients’ neighborhood-level measuresc
Median household income (in $1,000)	57.4 (37.0, 82.1)	54.3 (39.8, 69.8)
Percent unemployed	5.7 (4.4, 8.8)	6.2 (5.1, 8.2)
Percent below federal poverty level	14.7 (7.5, 26.6)	16.7 (10.6, 25.2)
Housing instabilityd	447 (0.4)	994 (0.7)

^aTotal N of patients at baseline includes patients ages 16–90 years with an eligible PC visit to PROUD or usual care clinics in the 3 years prior to randomization

^bPatients with missing sex were less than 0.001% for PROUD and usual care clinics

^cMedian (interquartile range); we mapped patients’ residential zip code closest to randomization date to neighborhood-level measures from the American Community Survey. There were 1.5% and 1.8% of patients missing a zip code for PROUD and usual care clinics, respectively.

^dNumber (%); based on International Classification of Diseases diagnostic codes in the 2 years prior to randomization.

Effect of the PROUD intervention across sex and race and ethnicity

Sex

Sex significantly modified the effect of the intervention on days of OUD treatment in the primary Model (F(1,14)=4.77; p=0.046; Figure 1 and Table 2), with female patients benefiting from PROUD OBAT implementation less than male patients. Results were similar in Model 2 (F(1, 9)=4.27; p=0.069) and Model 3 (F(1, 12)=5.15; p=0.043) in sensitivity analyses (Table 2). In stratified analyses (Table 2), for female patients, there was a non-significant increase of 2.9 patient-years of OUD treatment per 10,000 in intervention compared to UC clinics, whereas for male patients there was a statistically significant increase of 13.7 patient-years of OUD treatment per 10,000.

Figure 1.

Mean difference between PROUD intervention and usual care clinics in patient-years of OUD treatment among primary care patients across subgroups based on sex, race and ethnicity.

Adjusted mean differences in patient-years of opioid use disorder (OUD) treatment between PROUD and usual care by subgroup (with 95% confidence intervals) and p values for interaction (primary model); “smaller groups” does not include non-Hispanic individuals with unknown race.

Table 2.

Three models of adjusted difference between PROUD intervention and usual care clinics in the primary outcome of patient years of OUD treatment post-randomization across demographic subgroups.

		Model 1: Primary Model: adjusted for baseline with random intercept for health system		Model 2: Model 1 plus random intercept for clinic to account for within-clinic correlation		Model 3: Model 2 plus interaction between categorical demographic sub-group variable (age, sex or race-ethnicity) and baseline value of the outcome
Subgroup	Adjusted Difference (2-sided 95% CI)	P valuea	Adjusted Difference (2-sided 95% CI)	P valuea	Adjusted Difference (2-sided 95% CI)	P valuea
Sex
Interactionb	F (1,14)b=4.77	0.046	F(1,9)b=4.27	0.069	F(1,12)b=5.15	0.043
Female	2.9 (−4.3, 10.2)	0.399	2.9 (−5.8, 11.6)	0.472	2.5 (−6.3, 11.4)	0.546
Male	13.7 (5.8, 21.7)	0.002	13.5 (4.0, 23.0)	0.011	13.5 (3.7, 23.3)	0.011
Race and ethnicity
Interactionb	F(4,44)b=1.50	0.218	F(4,39)b=1.50	0.221	F(4,39)b=1.65	0.182
Hispanic	1.8 (−23.9, 27.4)		1.8 (−23.9, 27.4)		1.2 (−24.3, 26.6)
Asian	−1.0 (−26.6, 24.6)		−1.0 (−26.7, 24.6)		−0.9 (−27.2, 25.3)
Black	0.0 (−25.6, 25.7)		0.0 (−25.7, 25.8)		3.5 (−25.4, 32.5)
White	35.6 (9.4, 61.8)		35.6 (9.3, 61.9)		38.4 (12.2, 64.6)
Smaller groups	−0.3 (−26.2, 25.5)		−0.3 (−26.3, 25.6)		7.2 (−19.4, 33.8)

^aTwo-sided p-value

^bOmnibus test for any heterogeneity of intervention effects across subgroups; F-test with degrees of freedom for numerator (first number in parentheses) and denominator (second number in parentheses); Bold=statistically significant findings.

Race and ethnicity

Race and ethnicity did not statistically significantly modify the effect of the intervention on days of OUD treatment in primary analyses (F(4, 44) =1.50; p=0.218; Figure 1 and Table 2). Sensitivity analyses had similar results (Table 2). In pre-planned stratified analyses, the estimated benefit of the intervention for Hispanic, Asian, and Black patients and smaller racial groups ranged from −1.0 to 1.8 patient years of OUD treatment per 10,000 (Figure 1; Table 2), whereas the estimated benefit of the intervention on days of OUD treatment for White patients was 35.6 patient-years of OUD treatment per 10,000 (Figure 1; Table 2).

Patient-level exploratory analyses of differences across sex, race, and ethnicity

Receipt of Any OUD Treatment.

Table 3 describes OUD treatment at the patient level across sex and the three largest race and ethnic groups at baseline and follow-up. Table 3 also shows the difference between baseline and follow-up in each subgroup, and the difference between the trial arms in each subgroup—in all PC patients (top) and those with documented OUD (bottom), with smaller subgroups in Supplemental Tables B–C. The absolute difference in the proportion of PC patients who received any OUD treatment, in intervention compared to UC clinics, was less than half as great in female as in male patients (0.11% vs. 0.29%) and less than a third as high in Hispanic and Black patients as in White patients (0.09% and 0.08% vs. 0.30%). When descriptive analyses were restricted to those with documented OUD diagnoses (lower half of Table 3), the two-fold difference in the proportion of patients who received any OUD treatment between female and male patients persisted (11.80% vs. 24.98%), whereas the difference between Hispanic or Black patients and White patients was smaller (16.85% and 16.61% vs. 19.20%, respectively). Of note, a higher proportion of White patients were treated without a documented OUD diagnosis during follow-up (23%) than Hispanic and Black patients (11–12%; Table 3). Among patients treated for OUD, the proportion of male (31%), Hispanic (38%) and Black (38%) patients who were new to the health system, appeared higher than for female (19%) and White patients (23%; Figure 2).

Table 3.

Patient-level Exploratory Analyses of Any OUD Treatment

	Total Primary Care N		Any OUD Treatment in primary care patients n (%)		Change Baseline to Follow-up %		Difference %
Total	PROUD	Usual Care	PROUD	Usual Care	PROUD	Usual Care	PROUD > Usual Care
Baseline	109196	133170	77 (0.07)	140 (0.11)
Follow-up	111269	129230	318 (0.29)	176 (0.14)	0.22	0.03	0.18
Female a
Baseline	65566	84564	38 (0.06)	63 (0.07)
Follow-up	65957	82082	141 (0.21)	97 (0.12)	0.16	0.04	0.11
Male a
Baseline	43629	48603	39 (0.09)	77 (0.16)
Follow-up	45309	47145	177 (0.39)	79 (0.17)	0.30	0.01	0.29
Hispanic b
Baseline	30566	44415	9 (0.03)	17 (0.04)
Follow-up	29874	40852	42 (0.14)	24 (0.06)	0.11	0.02	0.09
Black b
Baseline	21045	27237	4 (0.02)	18 (0.07)
Follow-up	19958	26228	26 (0.13)	26 (0.10)	0.11	0.03	0.08
White b
Baseline	42981	39952	57 (0.13)	83 (0.21)
Follow-up	45459	39630	225 (0.49)	106 (0.27)	0.36	0.06	0.30
	OUD Diagnosis n (%)		Any OUD Treatment in patients with OUD n (%)		Change Baseline to Follow-up %		Difference %
Total	PROUD	Usual Care	PROUD	Usual Care	PROUD	Usual Care	PROUD > Usual Care
Baseline	644 (0.59)	730 (0.55)	65 (10.09)	118 (16.16)
Follow-up	806 (0.72)	765 (0.59)	251 (31.14)	145 (18.95)	21.05	2.79	18.26
Female a
Baseline	335 (0.51)	385 (0.46)	32 (9.55)	52 (13.51)
Follow-up	408 (0.62)	419 (0.51)	107 (26.23)	77 (18.38)	16.67	4.87	11.80
Male a
Baseline	309 (0.71)	345 (0.71)	33 (10.68)	66 (19.13)
Follow-up	398 (0.88)	346 (0.73)	144 (36.18)	68 (19.65)	25.50	0.52	24.98
Hispanic b
Baseline	108 (0.35)	104 (0.23)	9 (8.33)	15 (14.42)
Follow-up	122 (0.41)	103 (0.25)	38 (31.15)	21 (20.39)	22.81	5.97	16.85
Black b
Baseline	81 (0.38)	151 (0.55)	4 (4.94)	15 (9.93)
Follow-up	104 (0.52)	169 (0.64)	25 (24.04)	21 (12.43)	19.10	2.49	16.61
White b
Baseline	395 (0.92)	390 (0.98)	47 (11.90)	71 (18.21)
Follow-up	505 (1.11)	427 (1.08)	168 (33.27)	87 (20.37)	21.37	2.17	19.20

OUD = opioid use disorder

Top: Percent of primary care patients with any OUD treatment at baseline and follow-up across trial arms and demographic groups at baseline and follow-up (Smaller race and ethnic groups Supplement Table B).

Bottom: Among patients with a documented OUD diagnosis, percent who received OUD treatment across trial arms and demographic groups at baseline and follow-up (Smaller race and ethnic groups Supplement Table C).

^aTotal % of patients missing sex (baseline or follow-up) < 0.001%.

^bTotal % of patients r missing race and ethnicity, 4.7% and 6.1% for PROUD and usual care clinics, respectively;

Figure 2.

New and ongoing treatment and movement of patients into health system and primary care, among patients who were treated in trial clinics during follow-up (post-randomization), by sex (Panel A) and race and ethnicity (Panel B)

Among patients treated for opioid use disorder (OUD) post-randomization, the proportion in each of 4 mutually exclusive categories based on whether or not patients newly initiated OUD treatment post-randomization (defined as none in prior 365 days), and whether they had been in the PROUD intervention or usual care (UC) clinics or they had been seen elsewhere in the health system prior to randomization: Ongoing OUD Treatment: patients seen in PROUD trial clinics pre-randomization and received OUD treatment pre-randomization (with no gap in OUD treatment >365 days); New OUD Treatment & Existing Patient: patients seen in PROUD trial clinics pre-randomization and received new OUD treatment in a PROUD trial clinic post-randomization (none in the health system in the prior 365 days); New OUD Treatment & New to Clinic: Patient new to PROUD trial clinics post-randomization but not new to health system, and received new OUD treatment; and OUD Treatment and New to Health System: OUD treatment in patients new to the health system post-randomization (i.e., not seen in system pre-randomization, making it unknowable if the treatment was new or ongoing).

When measures of duration were evaluated in descriptive analyses restricted to those who were newly treated for OUD post-randomization (to avoid bias due to differences across trial arms in OUD treatment pre-randomization), there was an absolute increase of 20% in the proportion of female patients with ≥80% adherence, in intervention compared to UC clinics, but no increase in male patients (Table 4). Similarly, there were absolute increases of 18% and 20% in the proportion of Hispanic and Black patients with ≥80% adherence, respectively, in intervention compared to UC clinics, but only 4% in White patients. There were only small differences between female and male patients in the absolute increase in the proportion of patients with 6-month retention, in intervention compared to UC clinics: 6% female and 1% male patients (Table 4). In contrast, the absolute increase in the proportion of Hispanic and Black patients with 6-month retention, in intervention compared to UC clinics, was 10% and 24%, respectively, while there was no increase observed in White patients (Table 4). Similarly, there appeared to be larger increases in the proportion of Hispanic and Black patients treated with at least 16 mg of buprenorphine, in intervention compared to UC clinics, than in White patients (46% and 33% vs. 22%, respectively; Table 4).

Table 4.

Among primary care patients who were newly treated for OUD post-randomization (none in prior 365 days), measures of retention and dose in female and male patients and Hispanic, Black, and White patients^a

	Total Treated		≥80% adherence: % of patients with OUD treatment for ≥80% days after initiation			6-month retention: % of patients treated for ≥ 180 days			≥ 16 mg/day buprenorphine: % of patients who received dose ≥ 16 mg
	PROUD	UC	PROUD	UC	Difference	PROUD	UC	Difference	PROUD	UC	Difference
			%b	%b	%c	%b	%b	%c	%b	%b	%c
Total	286	125	46.15	36.80		34.97	30.40		70.98	44.80
Female	122	75	53.28	33.33	20	34.43	28.00	6	72.13	45.33	27
Male	164	50	40.85	42.00	−1	35.37	34.00	1	70.12	44.00	26
Hispanic	38	14	47.37	28.57	18	31.58	21.43	10	81.58	35.71	46
Black	23	17	43.48	23.53	20	47.83	23.53	24	73.91	41.18	33
White	203	80	47.29	42.50	4	33.99	35.00	−1	69.95	47.50	22

^aThese measures were evaluated post-hoc, in those who initiated OUD treatment post-randomization, to explore whether differences in the impact of the intervention on these measures might have contributed to observed findings in clinic-level primary interaction analyses.and pre-planned subgroup analyses The sample was limited to patients who newly initiated treatment post-randomization due to imbalance in the proportion of patients treated at baseline in PROUD and usual care clinics and the possibility that differences in OUD treatment at baseline could bias these measures if not limited to those who newly initiated.

^bRow percents

^cAbsolute difference in the percent of patients in PROUD and UC clinics with each outcome (among patients who initiated OUD treatment post-randomization)

OUD = Opioid use disorder

UC = Usual care clinics

DISCUSSION

The PROUD trial previously showed that implementation of the Massachusetts model of OBAT increased OUD treatment in PC (34). The present study showed significant effect modification by sex with the PROUD intervention increasing days of OUD treatment more in male than female patients. In contrast, results show no significant effect modification of the PROUD intervention by race and ethnicity, although pre-planned subgroup analyses suggest that the intervention might increase OUD treatment predominantly in White patients.

Observed differences in the intervention effect in female and male PC patients were in the opposite direction as expected. Our a priori hypothesis that implementation of OBAT would increase days of OUD treatment for female more than male patients was based on prior research showing that among patients who received OUD treatment in the Massachusetts model of OBAT, female patients were more likely than male patients to remain in OUD treatment (26, 30, 33). Exploratory analyses in the present study suggest that implementation of OBAT may have increased the reach of OUD treatment more in male than in female patients. Further, this apparent difference in reach was present even when analyses were restricted to those with OUD diagnoses documented in medical records, suggesting that it did not solely reflect the higher US prevalence of OUD in male individuals (45–47). These potential differences in reach are consistent with extensive prior research showing that women in the US and Canada experience greater barriers to OUD treatment, often due to stigma, having fewer economic resources, having more family or caregiving responsibilities and potential concerns about child protective services, and other intrapersonal, interpersonal, and structural constraints, many reflecting societal sexism (7, 9, 48–50). In contrast to the potentially greater reach of OUD treatment due to the intervention in male than in female patients, among those who received new OUD treatment post-randomization, findings suggest greater increases in adherence and 6-month retention due to the intervention in female compared to male patients, consistent with prior studies (26, 30, 33).

Ultimately, despite potential benefits of the intervention for retention of female patients in OUD treatment, the small number who received any treatment limited their benefit from the PROUD intervention. Research on the Massachusetts model of OBAT—including the PROUD trial (34)—has consistently shown that about 70% of patients treated for OUD are new to the PC clinic, suggesting that the model of low barrier care might attract patients to PC. Exploratory analyses among patients treated for OUD suggest that a higher proportion of male than female patients were new to the health system (Figure 2A), raising the possibility that male patients might be more likely to: seek OUD treatment in PC, be referred into PC for OUD treatment from external organizations, or be able to follow through on referrals due to encountering fewer logistical and socioeconomic barriers (7, 9, 48, 49). The OBAT model implemented in PROUD includes nurse outreach to potential referral sources in the health system or community but not outreach to individual PC patients with OUD. Outreach specifically targeting female individuals, both inside and outside health systems, may be necessary. Efforts to increase reach among female patients within the health system might include direct outreach to those with OUD documented in their EHRs or women’s health clinics. Efforts in the community could target social service agencies that support female individuals (e.g., child welfare services, Medicaid offices, women’s shelters). Additionally, nurses may need added resources (e.g., childcare vouchers, transportation) to help female patients overcome logistic and economic barriers. Finally, OBAT is expected to lower the stigma of OUD in PC over time, which could lead female patients with OUD to self-identify; future analyses of PROUD could assess whether sex differences persisted 3 years after implementation given marked increases in days of OUD treatment (37).

Findings of pre-planned subgroup analyses across race and ethnicity, combined with descriptive exploratory analyses, are important in that they generate hypotheses for future research. In the absence of significant interaction effects, effects of interventions evaluated within subgroups, like those in stratified analyses of race and ethnic groups, have a high risk of being falsely positive or falsely negative (51, 52). However, the observed large differences in intervention effect across sub-groups in preplanned stratified analyses (Figure 1) were consistent with our a priori hypotheses and with other research in the US and Canada (17, 53). Further, post hoc analyses suggest the PROUD intervention may have led to large differences in reach of OUD treatment, with 3-fold greater increase in receipt of any OUD treatment for White compared to Hispanic and Black patients (Table 3). When these analyses were restricted to those with documented OUD, the reach of OUD treatment appeared similar for Hispanic, Black and White patients (in contrast to findings in female and male patients). If confirmed in future research, these differences in reach may reflect under-diagnosis in Hispanic and Black compared to White patients in PC (16). Similar to female patients, among patients who received new OUD treatment, exploratory analyses suggest Hispanic and Black patients may benefit more from OBAT than White patients in terms of increased adherence, 6-month retention, and dose of buprenorphine, consistent with prior research (32). These findings are especially important to test in future research given prior observations of lower retention in OUD treatment among Black patients (16) and disparities in buprenorphine dose (12, 16, 28, 54, 55). If future research shows increased retention in Hispanic and Black patients, it would be a critically important benefit of the OBAT model.

Effective approaches to increasing the reach of and retention in OUD treatment in Hispanic and Black PC patients are urgently needed in light of the increasing prevalence of OUD in Hispanic and Black individuals (56, 57) and increasing rates of overdose among Black and American Indian/Alaska Native individuals (58, 59). The small proportion of Hispanic and Black patients treated for OUD in intervention clinics highlights potential limitations of efforts based solely in health systems for overcoming racial and ethnic disparities, although some PC interventions have lessened disparities (30, 33). Due to structural and interpersonal racism embedded in U.S. society and health care, the history of criminalization of drug use and racist drug policies, differences in access to resources (Supplement Table D), and the segregated system of OUD treatment, patients from minoritized racial or ethnic groups, understandably, may distrust health systems and medical providers, contributing to under-recognition, under-diagnosis, and under treatment of OUD (12, 13, 18, 19, 21, 60–65). Integration of nurse models of care into communities deserves testing, for example placing OBAT-trained nurse practitioners in pharmacies, schools, and organizations that serve minoritized communities where they could offer OUD treatment outside health systems (66). Such models will also need to address social determinants of health and concrete elements of structural racism (e.g., unstable housing, inadequate health insurance) (18, 19, 56, 61, 62).

Several limitations merit emphasis. As above, subgroup analyses in randomized controlled trials have a high risk of being erroneous, especially in the absence of significant interaction. UC clinics provided about 50% more OUD treatment than intervention clinics pre-randomization (34), and disparities present prior to randomization likely persisted given the 4–15 months before nurses started seeing patients post-randomization (34). This evaluation also may have been too early to observe meaningful reductions in OUD-related stigma in PC since nurses only practiced in intervention clinics 9–20 months, and observing successful treatment is thought to decrease stigma. The intervention was not equally effective across health systems (34) so findings might have reflected unique characteristics of systems that were successful relatively early. More systems were successful after 3 years (37), and, as noted above, further research is needed to assess disparities after a longer duration of implementation. The Massachusetts model was not designed to specifically overcome disparities, and nurses were hired and trained for clinical skills, not culturally specific outreach or care. The optimal approach to analyze interactions in small-cluster trials is unknown, and the statistical significance of interactions was sensitive to the modeling approach (Table 2). Patients missing sex or race and ethnicity in their EHRs were excluded from analyses of those subgroups; as such, the complete case analyses we conducted could introduce bias if these data are not missing completely at random (67). Moreover, our outcome could be subject to misclassification, for example due to incomplete capture of OUD treatment in the EHR or patients not taking medications ordered, which—if differential across intervention arms or subgroups—could also introduce bias. Additionally, analyses combined all groups except Hispanic, Asian, Black, and White patients in main analyses due to small numbers; future larger trials are needed to assess intersectionality. This paper did not analyze differences across age. Pre-planned analyses of youth and young adults have been presented separately (68), but future secondary analyses across other age groups will be important. Finally, despite variation in types of health systems (e.g., community, integrated insurance and care delivery) (29, 34), generalizability of the 6 health systems willing to participate in PROUD is unknown (36).

CONCLUSIONS

The PROUD trial was unique in its ability to test the impact of OBAT implementation on OUD treatment across sex, race and ethnicity in real world PC clinics. While implementation of the Massachusetts model of OBAT increased days of OUD treatment among PC patients overall, this study suggests that those benefits were not experienced equally by female and male patients, and possibly not across subgroups based on race and ethnicity. Post-hoc exploratory analyses suggested that implementation of OBAT may have increased the reach of OUD treatment less in female than male patients, and less in Hispanic and Black than in White patients. However, among those who received OUD treatment, results also suggested OBAT implementation may have improved retention more in female, Hispanic and Black patients compared to male and White patients. These findings highlight the importance of evaluating reach and duration of treatment separately in efforts to increase OUD treatment, and the importance of future research to develop and test adaptations of the Massachusetts model of OBAT designed to improve reach among female, Black, Hispanic patients.