A retrospective quantitative implementation evaluation of Safer Opioid Prescribing, a Canadian continuing education program

Continuing health professions education (CHPE) is an important policy intervention for the opioid epidemic. Besides effectiveness or impact, health policy implementation should be studied to understand how an intervention was delivered within complex environments. Implementation outcomes can be used to help interpret CHPE effects and impacts, help answer questions of “how” and “why” programs work, and inform transferability. We evaluated Safer Opioid Prescribing (SOP), a national CHPE program, using implementation outcomes of reach, dose, fidelity, and participant responsiveness. We conducted a retrospective quantitative implementation evaluation of the 2014–2017 cohorts of SOP. To measure reach and dose, we examined participation and completion data. We used Ontario physician demographic data, including regulatory status with respect to controlled substances, to examine relevant trends. To measure fidelity and participant responsiveness, we analyzed participant-provided evaluations of bias, active learning, and relevance to practice. We used descriptive statistics and measures of association for both continuous and categorical variables. We used logistic regression to determine predictors of workshop participation and analysis of covariance to examine variation in satisfaction across different-sized sessions. Reach: In total, there were 472 unique participants, 84.0% of whom were family physicians. Among Ontario physician participants, 90.0% were family physicians with characteristics representative of province-wide demographics. Dose: Webinar completion rate was 86.2% with no differences in completion based on rurality, gender, or controlled substance prescribing status with medical regulatory authorities. Fidelity and participant responsiveness: Nearly all participants rated the three webinars and workshop as balanced, and each element of SOP was also rated as highly relevant to clinical practice. This evaluation demonstrates that Safer Opioid Prescribing was implemented as intended. Over a short period and without any external funding, the program reached more than 1% of the Ontario physician workforce. This suggests that the program may be a good model for using virtual CHPE to reach a critical mass of prescribers. This study represents a methodological advance of adapting evaluation methods from health policy and complex interventions for continuing health professions education. Future studies will assess effectiveness and impact on opioid prescribing and utilization within evaluation models of complex interventions.


Introduction
Health policies are often characterized as complex interventions and continuing health professions education (CHPE) programs share many similar distinguishing features: they involve the actions of people, include a complex chain of steps, are embedded in social systems shaped by context, and are dynamic, open systems subject to change [1,2]. This suggests that such CHPE programs can and should be subjected to the same rigorous evaluations as health policies whenever possible. Studying the implementation of CHPE programs is an important step in not only better understanding if a program was delivered as intended but also in offering further insight into 'how' it works to create intended and also unintended effects.
Opioid-related harms such as overdose and death continue to mount throughout high-income North America. In the United States more than 232,000 people died between 1999 and 2018 from opioid overdoses involving prescription opioids, with death rates more than four times higher in 2018 compared to 1999 [3]. Likewise, in Canada the rate of opioid-related deaths has remained persistently elevated with the most recent pre-pandemic rate estimated at 10.7 per 100,000 [4]. While illicit opioids such as fentanyl and its analogues are important drivers of ongoing harms, prescription opioids continue to be important contributors [5]. Similar to the U.S., Canada has recently seen a leveling of national life expectancy for the first time since the Second World War, which has been attributed primarily to opioid-related deaths [6,7]. Indications are that in many jurisdictions, opioid-related harms have risen significantly in the context of the COVID-19 pandemic [8,9].
Continuing health professional education (CHPE) without industry involvement has been considered an important rectifying intervention as identified by a variety of national and regional policy documents [10][11][12], editorialists [13,14], and national media [15]. The Canadian government has made continued major investments in chronic pain and opioid CHPE as part of its opioid crisis response strategy [16]. This is because education about chronic pain and opioid analgesic prescribing has been considered deficient in terms of both quantity and quality across the medical education continuum [17,18]. In a recent scoping review of contemporary opioid analgesic CHPE programs [19] we identified that the majority of programs identified the opioid crisis as a motivating factor for developing and delivering their interventions. Likewise, opioid manufacturer promotions couched as educational activities have been prominent drivers of inappropriate prescribing and opioid-related harms [20]. Our review also identified an ongoing influence of opioid prescribing manufacturers in contemporary CHPE programs, an issue which has also been identified in critiques of the Food and Drug Administration's Opioid Analgesic Risk Evaluation and Mitigation Strategy [21][22][23].
In a recent systematic review of the effectiveness of interventions to reduce opioid-related harms, Furlan et al. [24] identified CHPE as a promising strategy to improve inappropriate use of opioids and reduce abuse, misuse, and addiction. However, this review did not assess the educational strategies of the CHPE programs that demonstrated these outcomes. At least one systematic review of opioid prescribing interventions has characterized education using an evaluative framework for complex interventions [25]. This review classified CHPE and other prescribing interventions such as prescription drug monitoring programs in terms of implementation, effectiveness and impact outcomes. The study authors, however, did not relate such outcome categorizations back to any kinds of frameworks, conceptual models or theories of educational development, delivery or evaluation.
Moore et al.'s CHPE outcomes framework [26,27] provides a useful bridge between outcomes for educational programs and complex intervention. The first two levels of participation and satisfaction map clearly to implementation outcomes, the third through fifth levels (knowledge, competence and performance) map to effectiveness outcomes, and the sixth and seventh levels (patient health and population health) can be categorized as typical impact outcomes (Appendix 1). Moore et al. characterize these outcomes as a hierarchy, with the highest quality programs aiming and achieving demonstrable changes in terms of patient or community health outcomes. Evaluations focused on these higherlevel outcomes can answer questions such as, "did this program have the intended impacts?" or more bluntly "did this program work?". However, if program implementation is not studied concurrently with effectiveness and impact outcomes, then such evaluations cannot answer questions of "how" and "why" the programs did or did not have the intended outcomes [28]. Methods of program evaluation which study implementation can aid in opening this so-called "black box problem" [29]. For example, by examining participation and satisfaction, one can learn not only if a program intervention reached its target audience and met its learning objectives, but one can also garner valuable data related to how a program has been implemented in a specific context or how it may be sustained in that context. These data are essential in informing transferability of findings into other contexts [30].
With this framework of opioid prescribing CHPE as a health policy intervention, we aimed to conduct a comprehensive implementation evaluation of a national Canadian program called Safer Opioid Prescribing (SOP). SOP was fully redeveloped from a previous program that has been delivered by the College of Physicians and Surgeons of Ontario (CPSO) to address remediation needs for physicians for whom this medical regulator had identified instances of inappropriate controlled substance prescribing. SOP continued to address this remediation need for the CPSO, as well as medical regulators in other Canadian provinces, but aimed primarily to provide high quality continuing education for health professionals without any regulatory involvement. In this study, we specifically assessed the implementation outcomes of reach, dose, fidelity and participant responsiveness. Study questions, implementation outcomes and corresponding CPD outcome level are summarized in Table 1.

Methods
We conducted a quantitative retrospective cohort evaluation of SOP between 2014 and 2017 using registration and participant evaluation data. Ethical approval for this study was granted by the University of Toronto Research Ethics Board.

Intervention: Safer Opioid Prescribing development and description
Starting in 2012, we developed SOP using Kern's model for curriculum development [31] with two particular adaptations for CHPE. First, for needs assessment and program planning, we used the PRECEDE-PROCEED model [32,33], a comprehensive framework for program design, implementation and evaluation that is commonly used in the fields of public health and health promotion. In using this model, we formalized our conception of education as a health policy intervention as has been done elsewhere [34]. Using PRECEDE-PROCEED allowed us to: a) contextualize CHPE within the specific circumstances of the Canadian contemporary opioid epidemic and the range of other policy options for addressing it; b) involve the target audience for the intervention in program planning; and, c) conceptualize and categorize specific implementation and effectiveness outcomes during the initial design stages ( Table 2; see also Appendix 2 for Program Logic Model).
Our second adaptation to CHPE was to use multiple systematic reviews of continuing medical education (CME) effectiveness to identify and incorporate best practices in education for achieving practice change and improvement in patient outcomes [44][45][46][47], including for internet-based CHPE [48]. SOP utilized multiple interventions (13 distinct interventions), was of substantial duration (3-4 months), utilized a blended-learning approach, was interactive, and identified links between clinical practice and serious health outcomes [49]. The program was split into two componentsa series of three synchronous evening webinars followed by a oneday in-person workshop to create a flipped classroom (see Appendix 3: Safer Opioid Prescribing Program Components and Description). Besides addressing accessibility, the virtual format also aimed to make the program scalable to reach a large number of participants simultaneously. The webinars were made synchronous and interactive to help create a virtual community of learning [50], which we hypothesized would help normalize a challenging area of practice and also drive higher levels of completiona known challenge for online learning programs [51,52]. The first of the three webinars focused on the multimodal management of chronic pain; the second on the details of opioid prescribing (e.g. patient risk assessment, medication selection, initiation and titration); and the third on situations in which prescribing can be more challenging (e.g. with the elderly, in pregnancy, with people living with opioid use disorder). The workshop addressed challenging cases and communication issues, focusing on skills and competencies particularly suited for a live workshop as compared to a synchronous webinar. Webinar participation was a pre-requisite for workshop participation. Each webinar and the workshop had specific pre-work and post-work to prime learning and to facilitate integration into practice, respectively. The program was accredited for a total of 27 credits of learning: 9 credits for the webinars and an additional 18 credits for the workshop.

Study population
The study population included all SOP participants from January 1, 2014 through June 14, 2017. This study period was chosen because the program had fully launched in its current form by January 2014 and the content of the program was substantially redeveloped after June 2017 based on the release of new Canadian guidelines for opioid prescribing [53]. We included all participants in this study period, regardless of profession, specialty, location, completion status and whether or not they had any identified medical regulator involvement with respect to controlled substance prescribing. We excluded medical residents and trainees, participants for whom there was Prescribed opioids were identified as an important contributor to opioid related harms and family physicians were identified as the most common prescribers of opioids [35].
➔ The scientific planning committee included family physicians from a diversity of backgrounds (primary care, chronic pain care, addictions medicine, anesthesia, pharmacology and inner-city medicine).
The opioid epidemic was growing in scale and was linked to the practices of the majority of family physicians [36].
➔ The program targeted family physicians, though it was designed to also be relevant to specialist prescribers as well as other professionals involved in opioid prescribing (e.g. pharmacists). Nurse practitioners were not identified as a primary target at the time of development since they were not eligible to prescribe opioids in our jurisdiction until early 2017.
There was an inequitable distribution of harms, with greater rates of overdoses and deaths from opioids in rural and remote communitiesplaces where there might be less access to practice supports and high quality CHPE programs [37].
➔ The program was to be delivered virtually and in the evenings, outside of typical practice times, to increase accessibility for rural and remote health professionals.
Chronic pain was a major learning priority for family physicians [38,39] and there were important knowledge gaps with respect to opioid prescribing [40].
➔ SOP content focused on opioid prescribing but was contextualized within models of the management of chronic pain as a complex medical condition.
There was a persistent influence of the pharmaceutical industry on prescribing practices and thus a growing skepticism of opioid educational programs because of possible pharmaceutical industry involvement [20].
➔ Faculty for the program during the study period of interest did not have any history of involvement with opioid or other pharmaceutical manufacturers. The program received no funding from industry for either development or delivery. It was funded entirely by participant registration fees to ensure sustainability. Fees for the program for physician participants were C$450 for the webinars and $650 for the workshops. A reduced rate for non-physician and resident participants was C$150 for the webinars and C$200 for the workshops.
Existing CHPE programs in the field tended to be based on expert opinion rather than the best available evidence, for example, from systematic reviews or clinical practice guidelines.
➔ Foundational documents included a national clinical practice guideline [41] and tool that was developed to support the implementation of the guideline [42].
The provincial medical regulator had an active and substantial influence on opioid prescribing behaviour, which in some cases could be an even stronger driver of prescribing behaviour than certain kinds of educational interventions [43].
➔ Participants in the program were sometimes required or suggested to attend by their medical regulator due to the identification of possible inappropriate controlled substance prescribing; however, program administration and faculty were blinded to participants' regulatory status.
substantial missing participation data, and participants who participated only in the workshop but not in the webinars.

Data sources
Participation data was collected from the registration system of Continuing Professional Development at the University of Toronto which administers SOP. Registration data included dates of participation, practice location, profession and specialty. For Ontario physicians, these registration data were linked with gender, graduating medical school and dates of Ontario licensure from the public register of the CPSO. Prior authorization from the CPSO to access these data was obtained. The Ontario Medical Association's (OMA) Rurality Index of Ontario was used to determine a rurality score based on practice postal code. This Index has been used in other program evaluations as a measure of rurality for Ontario physicians [54]. For data pertaining to the rurality of the Ontario family physician population (as a comparator to our participant group), we combined the OMA-generated RIO score at the Census Sub-Division level [55] with the Ontario Physician Human Resources Data Center's list of Physician Counts by Census Sub-Division for 2017 [56].
For satisfaction data, we used anonymous program evaluations collected immediately post program.

Outcome measures
We collected both participation and satisfaction data to assess for the implementation measures of reach, dose, fidelity and participant responsiveness [57]. We defined program reach as the total number of participants in any webinar. We further characterized this outcome with information about their profession, specialty and province of practice. Specifically, for Ontario physician participants, we collected data about gender, graduating medical school (international versus domestic), number of years of practice since Ontario licensure to first participation in SOP, medical specialty and rurality. This aligns with Durlak and DuPre's [57] definition of reach as, "the percentage of the eligible population who took part in the intervention, and their characteristics" (p 329). Given the known influence of medical regulation on opioid prescribing [43], we also recorded the status of the Ontario physician participants with respect to controlled substance prescribing and the provincial medical regulatory college (CPSO). We defined those who had a public record of controlled substance prescribing restrictions or a public record of a regulatory hearing regarding their controlled substance prescribing as having medical regulatory involvement.
Hewing closely to the definition of dose as "how much of the original program has been delivered" ( [57] , pp. 329), we measured program dose using attendance information for each of the webinars and workshop. Attendance at each of the webinars and workshop was coded as a binary attended / did not attend outcome.
Since reducing bias, and specifically bias relating to opioid manufacturer involvement was a key element of the intended program design, we interpreted ratings of program bias as a measure of program fidelitynamely "the extent to which the innovation corresponds to the originally intended program" ( [57 ], pp.329). Participants responded "yes" or "no" to the question "Was the presentation balanced and unbiased?". Likewise, the program was also intentionally designed to directly address identified clinical needs of practicing physicians which centered around appropriate chronic pain management, alongside opioid prescribing. Thus, we interpreted participant-provided relevance to practice ratings as a measure of program fidelity. This was measured using a 5-point Likert scale where 1 is 'not at all relevant' and 5 is 'very relevant' for the webinars and a 7-point Likert scale where 1 is 'not at all relevant' and 7 is 'very relevant' for the workshops.
Finally, we used anonymized, participant-provided ratings of adequacy of active learning time as a measure of participant responsiveness, which can be defined as "the degree to which the program stimulates the interest or holds the attention of participants" ( [57] , pp. 329). This was measured using a 5-point Likert scale where 1 is 'none' and 5 is 'just enough' for the webinars and a 7point Likert scale where 1 is 'none' and 7 is 'just enough' for the workshops.
These last three measures for fidelity and participant responsiveness were collected anonymously from participants post-intervention and so could not be linked to participant demographics.

Data analysis
We used descriptive statistics mean, median, standard deviation, minimum, and maximum for continuous measures, and frequency and percentage for categorical measures to describe the sample. We assessed the association between categorical variables using Chi-squared and Fisher's exact tests. We assessed the association between binary and continuous measures using two sample t-test. We used logistic regression to assess association between participant factors including gender, years in practice, webinar completion, regulatory college status, setting, and international medical graduate (IMG) or Canadian medical graduate (CMG) status and the likelihood of workshop participation. We used the Hosmer-Lemeshow test to assess the goodness of fit for the logistic regression model and the outcomes are reported as odds ratios. We used analysis of covariance to assess for variability in adequacy of active learning time and clinical relevance across different size groups and program types (webinar or workshop). All tests were two-sided and p < 0.05 was considered statistically significant. We used the statistical software SAS 9.4 for data manipulation and statistical analysis.

Results
During this study period, the SOP series of three webinars was offered 11 times. The workshop was offered 8 times as a standalone program in Toronto, Ontario, Canada at a University-affiliated conference centre. The number of participants per webinar ranged from 26 to 74 (mean = 47.5). Workshop participation ranged from 14 to 26 (mean = 24.2). The response rate for anonymous post webinar evaluations was between 51 to 53% and for post workshop evaluations was 91 to 99%.

Participant characteristics
There were 517 unique registrants for this program. Participants who only participated in the workshop (n = 10) were excluded from this analysis. We combined data for registrants who participated in the program more than once (n = 20) into a single record. We excluded 18 medical residents (3.5%) and also excluded an additional 17 registrants (3.3%) due to incomplete registration or participation data.
In total, there were 472 unique participants (Table 3). One hundred and sixty-four (34.7%) participants were female. The large majority (88.1%) were from Ontario while the remainder were from each of the other Canadian provinces but none of the northern territories. There were three participants from a US state adjacent to Ontario. Three hundred and ninety-eight (84.3%) were family physicians which included general practitioners as well as those with focused practices in emergency medicine, addiction medicine, anesthesiology, community medicine, geriatrics, palliative care, occupational medicine and psychotherapy. Fifty-two (11.0%) were other medical specialists with the majority being from emergency medicine and anesthesiology. Twenty-one (4.4%) were other health professionals, including dentists, pharmacists, registered nurses and nurse practitioners, all of whom were from Ontario.

Ontario physician sample
Among the Ontario physician participants, family physicians were clearly over-represented in this sample at 90.0% compared to 50.2% of the Ontario physician workforce (p < .0001). Two hundred and fifty-nine (64.8%) were Canadian medical graduates, which was significantly lower than the proportion of 70.5% for physicians in the entire province (p < .0001). The mean number of years of practice was 19.2 (SD = 14.0) and ranged from 0.0 to 50.0. There was a clear bimodal distribution with a peak in the 0-10 year range consisting of nearly even female and male participants and another peak in the 30-35 years in practice range consisting mostly of male participants (Fig. 1).
While the overall sample was less urban than was the Ontario physician workforce, this difference was not significant when the setting distribution was compared by physician specialty. The rural to urban distribution of Ontario family physician and medical specialist Nursing, Dentistry, Pharmacy, Unknown < 5 each < 5 each participants were both reflective of the distribution of all physicians in the province (not shown).

Regulatory college status
We analyzed the Ontario physician participants with respect to their status regarding controlled substance prescribing with the provincial medical regulatory college (Table 4). We found that SOP participants with regulatory involvement were much more likely to be male and to have been in practice for longer. There were no differences in the rural to urban distribution, country of medical school graduation or profession type.

Webinar completion
Of the 472 webinar participants, 407 completed all three webinars for a completion rate of 86.2% (Table 5). There was no difference in completion based on gender, rural to urban setting in Ontario, country of graduation, or status with the regulatory college. The completion rate for family physicians (88.4%) and other health professionals (85.7%) was significantly higher than for other medical specialists (69.2%) (p = 0.001). Family physicians with a focused practice in emergency medicine (n = 25) had a completion rate of 68.0%, which was closer to   other medical specialists than to other family physicians (not shown). Likewise, Ontario participants had a significantly higher completion rate of 87.7% compared to the non-Ontario rate of 75.0% (p = .009). However, participants from jurisdictions below a two-hour time difference from Ontario (Quebec, Manitoba, Nova Scotia, New Brunswick, Newfoundland and US) had completion rates similar to Ontario at 85.2% (p = .753), while those at a two-hour or greater time difference from Ontario (Saskatchewan, Alberta, British Columbia) had a significantly lower completion rate of 65.5% (p = .0005).

Workshop participation
Of the 472 webinar participants, 177 (37.5%) participated in the workshop. Ontario participants were more likely to participate in the workshop than were participants from other provinces (participation rate 39.9% versus 19.6%, p = .003). We conducted a multivariate logistic regression to determine predictors of workshop participation for Ontario physician participants. Webinar completers were 5.1 times more likely to participate in the workshop (p < .001; 95% CI = 2.1 to 12.1) than noncompleters. Those who had medical regulatory involvement were 4.1 times more likely than those with no involvement to participate in the workshop (p < .001, 95% CI = 2.3 to 7.2). Urban practitioners were 2.2 times more likely than rural practitioners to participate in the workshop, but this finding was not statistically significant (p = .083, 95% CI = 0.9 to 5.1). There was no difference between urban and non-major urban participants (p = .32, 95% CI = 0.6 to 1.9). Gender, years in practice, or country of medical school graduation also did not predict workshop participation.

Fidelity
Close to 100% of the participants rated the three webinars and workshop as balanced and unbiased (Table 6). Each element of SOP was also rated as highly relevant to clinical practice. An analysis of covariance showed that the effect of group size on relevance to practice when controlling for program type (webinar or workshop) was not significant (p = .514).

Participant responsiveness
All elements of the program were rated as having adequate time for active learning (Table 6). An analysis of covariance showed that the effect of group size on adequacy of active learning time when controlling for program type (webinar or workshop) was not significant (p = .178).

Interpretation of results
SOP was designed as a policy intervention for the Canadian opioid crisis, using best practices in CHPE as a means for driving meaningful change at multiple levels of outcomes. This evaluation strongly suggests that through 2014-2017, SOP was delivered as intended along multiple implementation outcomes including reach, dose, fidelity and participant responsiveness. Family physicians, who are responsible for the majority of long-term and high-dose opioid prescribing, were disproportionately represented in the program. Likewise, participation in the program was representative of the geographic spread of physicians in Ontario, despite the program being delivered virtually from a major urban academic medical centre. This was accomplished in the context of Ontario's vast geography and known issues of poor access to high-speed internet in rural and remote communities [58]. Overall, physician participants were not reflective of the gender mix of Ontario physicians. However, this was mostly driven by participants with medical regulatory involvement, who very much skewed male and also as having more years in practice. This cohort of participants with medical regulatory involvement is reflective of patterns in Ontario that have been identified with respect to potentially problematic opioid prescribing [35] and also patterns of medical regulatory referral to opioid education programs [43]. The program was rated as highly relevant to practice, which may partly explain the very high engagement and completion rates. The slightly lower completion rates amongst specialist physicians and family physicians with focused practices, such as in emergency medicine, may in part be driven by lower relevance to practice ratings. It should be noted, however, that completion rates amongst these groups was still very high compared to internet-delivered CHPE norms [59][60][61]. Another reason for lower completion rates might also be less predictable clinical schedules for certain specialists. Importantly, medical regulator involvement was not an important driver of webinar completion.
Out of province completion rates were also high at 75.0%, though not as high as Ontario participants. This may be driven by lower relevance to practice as some of the program content is focused on epidemiological and practice issues specific to Ontario. However, a more practical reason for this difference may be because of time zone differences making it challenging for working professionals to participate in the webinars.
While not formally assessed as part of this implementation evaluation, scalability of the program [62] is suggested by the variable size of the webinars, with the largest including 74 simultaneous learners, and no systematic variation in time for active learning or relevance to practice based on webinar size. This suggests that fidelity to the program was maintained even with large numbers of participants. Overall, over a short period, the program was able to reach more than 1% of the 32,055 strong Ontario physician workforce. The demonstrated geographic reach of the program and the potential for scalability suggest that this program may be good model for reaching a critical mass of prescribers to drive population level changes in opioid utilization.

Implications
As CHPE evaluation has moved increasingly towards outcome-based approaches [28] with a preference for "higher-level" outcomes such as patient-level and population-level outcomes, there has been a related discounting of implementation outcomes such as participation and satisfaction [63]. Indeed Moore et al.'s updated framework [27] refers to three different kinds of assessment, namely summative, performance and impact. Importantly, however, this framework ignores assessment of implementation. This may be because educational interventions are not commonly conceptualized as complex interventions that are delivered in complex and dynamic health system and policy contextsall of which can affect program delivery and structure (implementation) and thus program effects. Thus, rigorous implementation evaluations can be used to determine how the program was actually carried out and whether it was carried out as intended. While implementation of programs as intended is no guarantee of effectiveness, these data are key to then informing subsequent effectiveness and impact evaluations and also to assessing program theory. Having conducted this implementation evaluation, further evaluation of SOP is now called for to assess effectiveness and impact.
To our knowledge, this study is one of few examples of an opioid prescribing CHPE evaluation that has formally assessed implementation outcomes using an evaluation framework for complex interventions together with a CHPE outcome model. Barth et al. [64] describe the use of the Medical Research Council complex intervention framework to develop and evaluate, in a stepwise manner, an academic detailing intervention to improve use of a prescription drug monitoring program (PDMP). A subsequent study evaluated physician selfreports of PDMP utilizationnamely a performance (education) or effectiveness (complex intervention) outcome [65]. Other opioid prescribing CHPE programs have assessed implementation measures of participation and satisfaction [66][67][68][69][70] but have not directly related these measures to program theory, nor have they used these implementation measures to then inform effectiveness or impact outcomes [19].
Overall this implementation evaluation adds further support to the feasibility of delivering multicomponent CHPE programs virtually to increase reach, scalability and thus potentially effectiveness and impact [71][72][73]. This is particularly relevant for population health problems directly linked to clinical practices of which the opioid epidemic is a pertinent example. Likewise, in the context of the COVID-19 pandemic and related physical-distancing measures, this evaluation provides additional confidence that virtual programs that adhere to evidence-based CHPE practices can provide good models to meet the ongoing learning needs of clinicians, including in complex areas of practice.

Limitations
There are several important limitations to this study. First, this study was of a single program and was conducted retrospectively using data that were collected for both evaluative purposes but also for administrative purposes, such as for tracking participation for accreditation reporting. We did have to exclude 17 participants (3.2%) due to incomplete participation data. This was a small enough number that it was unlikely to significantly bias results. Likewise, demographic data was for the most part complete. We could not, for example, determine rurality for only two of the 400 Ontario physician participants. Also, it is important to note that the evaluative data collected (e.g. relevance to practice and amount of interactivity) was defined prior to the delivery of the program and did reflect attempts to assess underlying program theory. The second data limitation relates to the anonymous nature of the evaluative data. These were kept anonymous as per norms in CHPE to allow participants to freely share their evaluative assessments. However, this did not allow us to link evaluative statements to particular participants and then analyze by demographic factors. This could be rectified in future evaluations by using, for example, a linking identifier that blinds the scientific planning committee to the identity of participants but allows evaluators to link evaluations to the demographic characteristics of de-identified program participants. Likewise, the webinar evaluation data response rates were moderate at 51% which would introduce an unknown bias to these data. Since these responses were anonymous, it is not possible to further assess the nature of this possible bias. However, the consistency of these evaluative responses between the webinars and the consistency of the responses with the workshop data which had an excellent response rate provides confidence that these data are reflective of the entire participant population. Third, the available data did not allow for a direct inquiry into the posited mechanism for change, namely that the SOP structure facilitates the creation of a virtual community of learning and practice. Qualitative inquiry using interviews or focus groups of program participants and facilitators would be well suited to better assess this aspect of the program. As noted above, this evaluation did not study patient-or population-level outcomes, which will be a priority for future study.

Conclusions
This evaluation demonstrates that Safer Opioid Prescribing was implemented as intended. Over a short period and without any external funding, the program reached more than 1% of the Ontario physician workforce. This suggests that Safer Opioid Prescribing may be a good model for using virtual continuing health professions education to reach a critical mass of prescribers. This study represents a methodological advance of adapting evaluation methods from health policy and complex interventions for continuing health professions education. This implementation evaluation will be used to inform further effectiveness and impact evaluations of Safer Opioid Prescribing.