Access to Primary Care for People With Opioid Use Disorder

Study context

According to the World Health Organization (WHO) primary health care is a "highly effective and efficient way to address the main causes and risks of poor health and wellbeing today" (1). Primary care is particularly important for vulnerable populations as it appears to mitigate effects from factors like low socioeconomic status (2-4). This appears to be true for those with opioid use disorder (OUD): in previous work, enrolment to a primary care physician was associated with higher rates of cancer screening and diabetes monitoring for those in treatment for OUD (5).

In universal health care systems, like Canada's, everyone is entitled to primary health care, regardless of health conditions or financial status. Our previous work, however, indicates that Ontarians in treatment for OUD are less likely to have access to primary care than others. Only 43% of people in treatment for OUD were enrolled with a primary care physician compared to 73% of matched controls (5). Our findings are consistent with an American study that reported poor access to primary care for those in treatment for OUD (6). These findings are concerning. Not only are those with OUD a vulnerable population with complex health needs, their numbers are also growing. OUD currently affects 15.5 million people worldwide (7), with Canada and the United States (U.S.) having the highest burden of disease (8).

Barriers to primary care for those with OUD have been poorly studied. Patient factors, such as difficulty attending appointments, and system barriers, like transportation costs, may affect access (9). Provider factors appear to play a role as well. A qualitative interview study in Ontario found that provider discrimination and stigma towards those with substance use disorder presented a barrier to accessing primary care (9). Other studies support this finding, showing that many providers, just like the general public, have stigma towards those with substance use disorder (10,11). It is possible that negative attitudes lead providers to discriminate against people with OUD when deciding whether to offer a new patient appointment.

Study rationale and goal

Our study goal is to determine if discrimination by physicians is a barrier to accessing primary care for those with OUD.

Research questions

1. Are people in treatment for OUD less likely to be offered a new patient appointment compared to people in treatment for diabetes?
2. What is the impact of physician gender, years in practice, rurality and model of care on offers of a new patient appointment?

Methods

The investigators will conduct a randomized controlled trial (RCT), using a similar approach to Olah et al that looked at access to primary care for those of low socio-economic status (14). They will make unannounced phone calls to primary care physicians in Ontario asking for a new patient appointment. Physicians will be randomly assigned to one of two clinical scenarios: a patient with diabetes treated by an endocrinologist, or a patient with OUD prescribed methadone by an addiction physician. In the secondary exploratory analyses, the investigators will also assess the impact of the model of care (team vs not), gender, rural vs urban, and years in practice.

Population The investigators will use information on the College of Physicians and Surgeon of Ontario (CPSO) website to identify all physicians in Ontario who have an active, independent, unrestricted practice with a self-reported speciality in family medicine (20).

Sample size calculation The objective of our study is to compare the proportion of patients offered a new patient appointment in each of the arms of the study. Our team consensus was that a 10% difference in the proportion of new patient appointments would be clinically meaningful. In a similar study design, Olah et al found that 23.5 % of those with diabetes were offered an appointment from a physician in Ontario (14). For our sample size calculation, the investigators assumed that 23% of those with diabetes and 13% of those in treatment for OUD would be offered a new appointment. At 80% power, 5% type-1 error level, to detect 10% difference in the proportion of new appointments offered between opioid (13%) and diabetes (23%) groups, would require 231 physicians with each scenario. However, previous studies using this approach found that up to one third of physicians were excluded at the time of making the phone call (because physician worked in a walk-in clinic or had a practice that was not primary care; or because of no response after five phone calls) (14,18). Therefore, the study should include 308 physicians in each arm.

To create our sample, the investigators will create a database using publicly-available information on the CPSO website. They will include all physicians who have an active, unrestricted, independent practice with speciality in Family Medicine. They will also record the physician's address(es), postal code, phone number(s), date of registration for independent practice and gender in our database. Physicians will only be identified in the database by a unique identifier. All these data items are collected by the CPSO. They will then group the physicians by primary practice address and then randomly select (using a random number generator), one physician from the list of physicians who work at the same site. The investigators will exclude other physicians from the same practice address.They will then use a random number generator to randomly select a statistical determined sample of physicians. We will randomly assign them to one of two clinical scenarios: a patient with diabetes treated by an endocrinologist, or a patient with OUD prescribed methadone by an addiction physician.

Scenarios The investigators will train a research assistant (RA) in the clinical scenario and pilot-test the interaction with a simulated receptionist giving different responses (see Appendix 1: Sample Script). To reduce confounding, they will have the same RA make all the phone calls. The RA will contact each physician to request an appointment, for one of the two clinical scenarios. The investigators will exclude physicians at the time of making the phone call if the physician's receptionist reports that the clinic is a walk-in, or that the physician does not offer primary care (14). They will also exclude practices at the time of making the call if they request an in person visit or health card number prior to making an offer of a new patient appointment. The investigators will attempt to reach a physician up to five times over a six-week period. For offices with voicemail, they will leave a message with the script but continue to make up to five phone calls to speak with someone in person. The investigators will accept call-backs up to six weeks after the initial phone call. They will exclude practices where they are not able to speak to anyone or leave a voicemail after five phone calls. They will only request an appointment once from a physician. If the physician's clinic offers an appointment, they will call the next day and cancel it. The investigators will continue sampling until they reach our target number of physicians.

Data analysis The investigators will use descriptive statistics to report on the population. To answer our research question, they will compare the proportion of patients offered a primary care appointment (a binary outcome) across the randomized groups. They will use chi-squared test and logistic regression to determine if there is a statistically and clinically significant difference in the proportions of offering a new patient appointment between the two clinical scenarios.

The investigators expect groups to be balanced on measured and unmeasured confounding variables. Nonetheless, they will investigate the distribution of measured variables across the two groups. If they observe imbalances, they will estimate the adjusted impact on the likelihood of being offered a new appointment, after controlling for possible confounding factors.

For our secondary analyses the investigators will use Fisher Exact test and two sample Wilcoxon test to assess if pre-specified subgroups: physician gender (male vs. female); practice location (rural < 50,000 vs. urban practice setting > 50,000); years in clinical practice, and practice model (team-based vs. not) affected the likelihood that a patient would receive an unconditional offer of a new patient appointment.

The investigators expect groups to be balanced on measured and unmeasured confounding variables. Nonetheless, they will investigate the distribution of measured variables across the two groups. If they observe imbalances, they will estimate the adjusted impact on the likelihood of being offered a new appointment, after controlling for possible confounding factors.

For our secondary analyses the investigators will use Fisher Exact test and two sample Wilcoxon test to assess if pre-specified subgroups: physician gender (male vs. female); practice location (rural < 50,000 vs. urban practice setting > 50,000); years in clinical practice, and practice model (team-based vs. not) affected the likelihood that a patient would receive an unconditional offer of a new patient appointment.