Impact of the Diabetes Canada Guideline Dissemination Strategy on the Prescription of Vascular Protective Medications: A Retrospective Cohort Study, 2010–2015

The World Health Organization describes the “rule of halves” whereby half of the people with long-term conditions are not known, half of those conditions known are not treated, and half of those conditions treated are not controlled (1). As a result of these failures, patients with diabetes are at increased risk of developing significant morbidity and mortality related to atherosclerotic cardiovascular complications including coronary heart disease, stroke, and peripheral vascular disease (2). The lifetime 10-year cardiovascular disease (CVD) risk is >20% for patients aged >40 years and living with diabetes (3). Effective risk factor modification with vascular protective medications, including statins, antihypertension agents, and antiplatelet agents, is an essential component of diabetes management to improve both quantity and quality of life-years (3,4). Clinical trial data show a 22–37% risk reduction in CVD for patients age >40 years on statin therapy and a 25% risk reduction for patients age >55 years taking ACE inhibitors (ACEIs) (4–6). Despite evidence-based support for these therapies, a national physician survey in 2012 estimated that 43% of patients with diabetes do not meet guideline-recommended LDL targets (≤2 mmol/L) and 64% fail to meet blood pressure (BP) targets (<130/80 mmHg), suggesting suboptimal vascular protection management (7).

For bridging of this evidence-to-practice gap for vascular protection in patients with diabetes, dissemination tools were launched with the 2013 Diabetes Canada (previously Canadian Diabetes Association) evidence-based guidelines. Compared with previous (2008) guidelines, the 2013 Diabetes Canada guidelines no longer require providers to stratify patients into different risk categories prior to recommending vascular protective therapy, thereby simplifying the assessment for vascular protection (Supplementary Table 1). Statin use is recommended for all patients ≥40 years old and living with diabetes; ACEIs or angiotensin receptor blockers (ARBs) are recommended for patients ≥55 years old with diabetes. Antiplatelet medications are no longer recommended for routine use in the primary prevention of CVD for patients with diabetes (3). Diabetes Canada also expanded on its patient- and provider-directed dissemination and implementation strategy, which was based on the Knowledge-to-Action Cycle and included a variety of new dissemination tools (8). The nationwide dissemination strategy launched in April 2013 and targeted multiple national and provincial systems-level groups (e.g., government agencies, nongovernmental agencies, disease advocacy groups, and professional associations), as well as health care providers and people living with diabetes across Canada via large-scale communications campaigns (e.g., television, radio, digital and print media) (9) (Supplementary Table 2). Interventions including in-person lecture series, conferences, webinars, web-based professional and patient resources such as flow sheets, electronic point of care decision support, a mobile application, and electronic medical record (EMR) templates were rolled out over 24 months. An evaluation of the effectiveness of this dissemination strategy has been published elsewhere (9).

Multiple studies examining vascular protective agents in Canada have shown an increase in statins and ACEIs/ARBs prescribed and used over the past two decades (10–12). Estimates from one Canadian province (Ontario) suggest that statin prescriptions in patients with diabetes aged >65 years have increased 53% from 1996 to 2010 and ACEI/ARB prescriptions have increased by 22% over a 6-year period (1995–2001) (11,12). These studies used data from provincial drug registries, which failed to capture nonprescription medications including antiplatelets and typically involved smaller geographic areas (i.e., city- or province-level data), making it difficult to determine whether dissemination and implementation of the Diabetes Canada guidelines achieved a population-level impact across Canada. Drug registry data also exclude the population of patients with diabetes age <65 years who may be eligible for vascular protective therapy. In addition, there are no studies that examined the prescription of these medications before and after the 2013 Diabetes Canada guidelines were released and disseminated, leaving the impact of the guidelines unknown.

Successful guideline evaluation is critical in assessing whether suggested recommendations have been adopted into practice at a population level. The current study used data from the Canadian Primary Care Sentinel Surveillance Network (CPCSSN), a pan-Canadian EMR surveillance system, to capture longitudinal trends in vascular protective agent prescriptions. The specific aims were to 1) examine changes associated with the launch of the 2013 Diabetes Canada Guidelines and additional dissemination efforts in the rates of vascular protective medications prescribed in primary care for older patients with diabetes and 2) examine differences in the rates of vascular protective medication prescriptions according to provider and patient characteristics.

We used a retrospective cohort design and followed the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) guidelines (13). We included patients living with diabetes whose data were in the CPCSSN database from April 2010 to December 2015. Diabetes was defined according to the CPCSSN-validated case definition (14) (Supplementary Table 3). The study cohort was dynamic, and patients were enrolled quarterly when the following three temporal conditions were met: 1) onset of diabetes recorded in the EMR prior to or during each quarter of interest, 2) first patient visit recorded within a given quarter or any quarter preceding it, and 3) patient met eligibility criteria for the medication being analyzed defined as per the 2013 Diabetes Canada guideline criteria: patients age ≥40 years for statins and age ≥55 years for ACEI/ARBs. For antiplatelet agents, patients age ≥40 years with no cardiovascular event were included, as guideline changes suggest that these medications should no longer be used for primary prevention in this group. Priorities set were based on well-accepted guideline recommendations and framed such that in our defined cohorts, those prescribed the agent are “controlled” and those not prescribed the agent are “uncontrolled.” Achievement of LDL targets (≤2 mmol/L) and BP targets (<130/80 mmHg) was assessed in patients in the following age categories: age ≥40 years for LDL and age ≥55 years for BP. We employed a censoring point whereby patients were removed from the cohort using the last encounter date if they had not seen their family doctor as indicated by an encounter in the EMR in the last 2 years. The study was approved by the University of Toronto’s Research Ethics Board (REB#33127).

CPCSSN, established in 2008, is an EMR-based information system designed for chronic disease surveillance. Every 3 months, EMR data from primary care practices in 10 practice-based research networks (PBRNs) across Canada are extracted, cleaned, and merged into a single database housed at the Centre for Advanced Computing at Queen’s University in Kingston, Ontario, Canada (15). Contributing PBRNs are located in Alberta (two), British Columbia (one), Manitoba (one), Newfoundland and Labrador (one), Nova Scotia (one), Ontario (three), and Quebec (one), Canada. The network is composed of >1,100 family physicians contributing data for >1,500,000 patients. Information contained in the database includes network and provider identifiers, de-identified patient demographic information, date and type of each patient’s encounters, patient health conditions, risk factors, referrals, laboratory investigations, procedures, and medications (15). CPCSSN captures all medications recorded in the EMR, including medications prescribed by primary care providers or specialists and those purchased over the counter.

Data from 1 January 2010 to 31 December 2015 were used for this project. The project data included patient demographic variables (year and month of birth, sex, and neighborhood-level socioeconomic status [SES] indicators derived from residential postal codes), physical measurements (height, weight, BMI, systolic BP, and diastolic BP), diabetes-related laboratory records (i.e., total cholesterol, LDL, HDL, triglycerides, and urine albumin-to-creatinine ratio), comorbidities/risk factors (i.e., hypertension, CVD, dyslipidemia, microalbuminuria, and smoking history), all medications prescribed to patients (i.e., lipid-lowering agents, ACEIs, ARBs, antiplatelets, and proton pump inhibitors [PPIs] as a reference control), provider/site characteristics (i.e., province of site and rurality of site) for each quarter of interest, and CPCSSN patient and provider data table (containing information on hierarchical relationship between patient, care provider, site, and network). Year of medical school (MD) graduation for providers was collected from the University of Toronto Practice-Based Research Network (UTOPIAN) network only owing to limited availability at other PBRNs.

All medications in CPCSSN database were coded using Anatomical Therapeutic Chemical codes; these codes were used to identify the relevant medications for this project (16). Long-term conditions were identified using CPCSSN-validated algorithms where available (i.e., diabetes and hypertension) (Supplementary Table 3). For the rest of the conditions, we used coded and free text diagnosis data from the CPCSSN health conditions table; this table populates data from the Cumulative Patient Profile (14). CVD was defined according to guideline and antiplatelet indications to include ICD-9 codes for coronary heart disease (410–414), cerebrovascular disease (430–438), and peripheral vascular disease (440–445) (17). Dyslipidemia was defined using appropriate medication history and coded and free text diagnosis data and included patients on lipid-lowering therapy or with LDL >2.0 mmol/L. Albuminuria was defined according to Diabetes Canada guidelines, whereby a urine albumin creatinine ratio <2 mg/mmol was categorized as normal, 2–20 mg/mmol as microalbuminuria, and >20 mg/mmol as macroalbuminuria (18). SES was measured in accordance with the Canadian Institute for Health Information method of deriving neighborhood-level deprivation indices on a quintile scale for each patient (19). Rurality was defined according to postal code whereby a second letter of “0” was considered rural and a second letter “1–9” considered urban (20). In accordance with recommendations of the World Health Organization and Health Canada, BMI was classified into categories of underweight (<18.5 kg/m²), normal weight (18.5–24.99 kg/m²), overweight (25–29.99 kg/m²), obese class I (30–34.99 kg/m²), obese class II (35–39.99 kg/m²), and class III (≥40 kg/m²) (21).

This study used an interrupted time series (ITS) design. For the primary analysis, the proportion of eligible patients who had prescriptions for statins, ACEI/ARBs, or antiplatelets in each quarter was computed using a longitudinal data analysis (22). Quarterly intervals were chosen to introduce less fluctuation in the time series curve compared with a monthly approach (23). If patients had multiple visits within a quarter, the most recent record was used for all variables studied. A total of eight quarters before, one quarter during, and nine quarters after the intervention were analyzed. Patients were deemed to be prescribed a statin, ACEI/ARB, or antiplatelet agent if they had any prescription record in the four preceding quarters and one quarter after each quarter of interest (lag4, lead1) (Fig. 1). This approach was used to account for variation in refill protocols and prescription procedures among family physicians (i.e., some provide a prescription every 3 months and others provide multiple repeats) (24). Having four quarters before and one after each quarter of interest would provide a sufficient time frame to capture prescriptions given on a yearly basis. We also examined other approaches, including (lag0, lead0), (lag1, lead1), (lag2, lead2), (lag3, lead3), (lag4, lead4), and (lag4, lead1). We found that prescription rates for the latter three approaches were very similar, further justifying our selection of the (lag4, lead1) approach.

The launch of the Diabetes Canada guidelines and dissemination strategy in April 2013 was considered the intervention point (2013 quarter 2 [Q2]). A segmented regression model was used to assess baseline trend/slope, the level change immediately after intervention, and the trend/gradual change (25). We also measured the prescription rates of statin, ACEI/ARB, and antiplatelets using a cross-sectional study design, which deemed a patient to be using the medication if they had at least one prescription during the entire study period (2010 Q1–2015 Q4). This approach serves as a sensitivity analysis for the (lag4, lead1) approach and allowed us to test for subgroup differences in the proportion prescribed statin, ACEI/ARB, and antiplatelets.

Secondary analyses included the proportion of eligible patients attaining LDL (≤2.0 mmol/L) or BP (<130/80 mmHg) target. A carryover approach was implemented: each patient’s most recent measurement was used in the following quarters until a new measurement was recorded in the EMR. A segmented regression model was used to assess the baseline trend/slope, level change, and trend/gradual change for secondary outcomes (i.e., eligible patients attaining LDL or BP target).

Results were stratified by patient characteristics (age, sex, BMI, and SES), presence of risk factors/comorbidities (smoking, CVD, hypertension, dyslipidemia, and albuminuria), and provider characteristics (province of care and rurality). We assessed the statistical significance across patient, provider, and geographical characteristics for prescription rates for statins, ACEI/ARBs, and antiplatelet agents using χ² test (adjusted for multiple hypothesis testing). We used the method of false discovery rate to control for inflated type I error rates in multiple hypothesis testing procedures (26). PPI prescriptions were used as a reference to control for confounding factors. Using modified hypothesis testing, we look at the temporal difference in prescription rates of PPIs compared with those of all three vascular protective medications (e.g., statin 2011 Q2–2015 Q3 vs. PPI 2011 Q2–2015 Q3) and considered statistical significance at P < 0.05. The analyses were conducted using SAS, version 9.4 (27).

The total number of patients enrolled in each dynamic cohort is outlined in Table 1. Quarterly cohort size ranged from 23,016 to 70,693 patients (Supplementary Table 4). Variation in population size was attributed to increasing recruitment of practices into the CPCSSN database over time. Results of the ITS analysis are presented in Table 2 and graphically in Fig. 2. There were no significant changes in the rate of statin, ACEI/ARB, antiplatelet, or PPI prescriptions prior to the release of the 2013 guidelines. After guideline publication, there was a significant change in slope for statins (−0.52% per quarter, SE 0.15, P < 0.05) and ACEI/ARBs (−0.38% per quarter, SE 0.13, P < 0.05) prescribed. A significant change in slope was also seen for PPI prescriptions (−0.18% per quarter, SE 0.05, P < 0.05). The change in slope was not significant for antiplatelets. The absolute difference in the rates of statin, ACEI/ARB, or antiplatelet prescriptions from the start (2011 Q2) to the end (2015 Q4) of the study period was significantly less than that of the PPI reference control (P < 0.0001).

Prescription rates in eligible patients were significantly higher in urban than rural practices for statins, ACEI/ARBs, and antiplatelets across the study period (Table 1). χ² test revealed significant differences in prescriptions rates between provinces for statins (P < 0.001), ACEI/ARBs (P < 0.001), and antiplatelets (P < 0.001). Statin and ACEI/ARB prescription rates were highest in the province of Quebec (74.2% for statins and 55.9% for ACEI/ARBs) and in the Maritimes provinces (64.8% for stains and 50.5% for ACEI/ARBs) and lowest in Alberta (45.1% for statins and 39.5% ACEI/ARBs) and British Columbia (35.2% for statins and 41.5% ACEI/ARBs).

There were no significant differences between MD graduation groups for statin (P = 0.152) or ACEI/ARB (P = 0.18) prescriptions in the UTOPIAN population (Table 1). However, there were significant differences between MD graduation groups for antiplatelet prescriptions (P < 0.001). Antiplatelet prescription rates were 20.9% for MD graduation years of 1965–1979, 15.3% for 1980–1994, and 18.3% for ≥1995.

Prescription rates according to patient characteristics are seen in Table 1. There were no significant sex-based differences in prescription rates for statins (P = 0.99), ACEI/ARBs (P = 0.71), or antiplatelet agents (P = 0.15). Patients with hypertension had significantly higher statin (P < 0.001), ACEI/ARB (P < 0.001), and antiplatelet (P < 0.001) prescription rates compared with patients without hypertension. Patients with CVD also had higher statin (P < 0.001) and ACEI/ARB (P < 0.001) prescription rates compared with patients without CVD. Patients without dyslipidemia had significantly lower rates of statin prescription (P < 0.001), ACEI/ARB prescriptions (P < 0.001), and antiplatelet prescriptions (P < 0.001) compared with those with dyslipidemia. Significant differences were detected for age groups, SES quintiles, smoking status, hypertension, dyslipidemia, and albuminuria for all three vascular protective medications (P < 0.05). Significant differences between BMI groups were seen with respect to ACEI/ARB prescriptions (P < 0.001).

Across the study period, LDL and BP targets were achieved by 47–57% and 61–66% of patients, respectively (Fig. 2B and C). The release of the 2013 guidelines did not significantly impact the proportion of patients achieving BP targets, as there were no significant changes in this proportion before, during, or after guideline release (Table 2). The release of the guidelines did not impact the proportion of patients achieving LDL targets, as there were no significant changes in this proportion during or after guideline release.

The results of this large study suggest that dissemination of the 2013 Diabetes Canada guidelines was not associated with any further improvements in physician prescribing behavior as measured by EMR prescription estimates for primary vascular protection in patients age ≥40 years with diabetes. While there was no change in prescriptions at the intervention point, a small but statistically significant decrease in slope after guideline publication was observed with statins and ACEI/ARBs, as well as the reference control. A negative change in slope (improved guideline adherence) was also observed with antiplatelets; however, this change was not significant. There were no significant changes with respect to LDL and BP targets at or after guideline intervention. Overall prescription rates were influenced by factors including rurality, province, and patient age and SES. While statistically significant changes were seen across patient/provider characteristics, these do not necessarily represent clinical significance given the large sample size.

Our time series analysis revealed a negative change in slope for statins, ACEI/ARBs, and PPIs. Owing to a lack of sufficient variation in the time series curves shown in Fig. 2A, we caution that this statistical inference can be affected by minuscule changes in prescription rates during any quarter pre– and post–guideline implementation (28). This decrease was also seen with the PPI comparator, suggesting that external factors cannot be excluded. It is possible that there was no actual improvement in vascular protective medication prescribing after release of the 2013 guidelines. The CPCSSN population also involves a dynamic cohort, so a change in composition of patients, physicians, or sites included in the database in each quarter could also influence prescription rates. Entry to the dynamic cohort was random, and there would be no reason to suspect that patients with diabetes from practices joining CPCSSN later would be more likely to have lower prescribing rates. We also did not observe any remarkable shifts with respect to composition of site locations but did observe an increase in patients age 40–49 years from 5.89% to 9.15% across the study period. Patients in this age category had lower statin and ACEI/ARB prescription rates, and thus an increase in the proportion of these patients may have contributed to the measurable decrease seen in prescribing rates over time. It is also possible that an emerging focus on resource stewardship in medical education and practice influenced physicians to prescribe more cautiously, decreasing prescription rates over time (29).

Our prescription estimates are lower than those from 2003 to 2008 Diabetes Canada guideline evaluations (7,30). These studies used physician self-reported survey data subject to selection bias, as physicians willing to participate may be those who prescribe in accordance with guideline recommendations. A validation study is currently ongoing to compare our results with data from the Institute for Clinical and Evaluative Sciences (ICES), which encompasses publicly funded administrative health data for the Ontario population eligible for universal health coverage (31). Preliminary results from this study confirm no improvement in statin, ACEI, and ARB prescriptions at or after guideline intervention (P > 0.05). Statin prescription estimates were similar (62–65%) to those in the current study (52–56%), and increased rates may be attributed to the older population cohort included in ICES (age ≥65 years).

We observed better guideline adherence in urban areas and provinces, including Quebec, Ontario, and the Maritimes. It is possible that dissemination outreach strategies were less effective for removed locations including rural areas and provinces such as Northwest Territories, Alberta, and British Colombia. Increased distance from health care access leading to reduced frequency of physician visits may inhibit physicians in rural areas from adhering to guidelines (32). Physicians in rural areas report increased patient resistance to medical or preventative care as a barrier to guideline adherence (32). We also observed lower prescription rates among patients from lower SES brackets, which is consistent with low SES cohorts being less likely to receive guideline-recommended diabetes care (33).

National survey data from primary care providers across Canada highlight significant gaps in knowledge and behavior change constructs for vascular protection (9). While awareness of the 2013 Diabetes Canada vascular protective guidelines was 71% among survey respondents, only 22% had correct knowledge of guideline content for vascular protection upon questioning and only 21% reported adhering to guideline recommendations (9). It remains possible that despite ongoing dissemination efforts, prescription rates may have reached a ceiling in terms of any further improvements in uptake.

Limitations of this study should be acknowledged. First, the CPCSSN cohort involves a convenient sample of patients, providers, and practices, which limits the generalizability of findings. Second, results must be interpreted carefully, as prescription estimates may not correspond with actual medication usage. Third, the CPCSSN-validated case definition for diabetes differed from the 2013 Diabetes Canada guidelines definition. The CPCSSN definition includes both type 1 and type 2 diabetes, with an HbA_1c ≥7% or fasting glucose ≥7 mmol/L, while Diabetes Canada includes an HbA_1c ≥6.5% or fasting glucose ≥7 mmol/L. We suspect that CPCSSN was unable to capture some patients who would otherwise be considered to have diabetes by the Diabetes Canada guidelines, potentially overestimating the rates of statin and ACEI/ARB prescriptions, as usage of these medications is associated with higher HbA_1c levels (34,35). However, we suspect that this number would have been small, as patients can be captured by other diabetes criteria according to the CPCSSN case definition (14). Fourth, a lack of a validated case definition for CVD, microalbuminuria, and dyslipidemia in CPCSSN may reduce the specificity and sensitivity of multiple estimates. Finally, ITS analyses cannot infer cause-effect relationships, as external factors are not accounted for. An analysis of PPI prescriptions showed trends similar to those of the vascular protective prescriptions under investigation, suggesting that the potential for confounding effects cannot be excluded.

There are limitations unique to using practice-based EMRs, whereby accuracy and completeness can vary across different systems, providers, and sites, in turn affecting the validity of the study (36). It is possible that a full record of patient medications including those requiring manual input (nonprescription, specialist prescriptions) are not recorded in the EMR by all providers. Accordingly, data from nonprescription medications as a group are typically less accurate than prescription estimates (37). Antiplatelet agents carry significant bleeding risk, and we suspect that they would be more likely to be recorded in the EMR than other nonprescription medications with lower-risk profiles. There is also no reason to expect differences in the completeness of data pre– and post–guideline implementation.

Despite these limitations, this study captured trends in vascular protective medication prescriptions in a large sample of Canadian primary care patients with diabetes. Previous studies have primarily used provincial data and are limited in their generalizability at a national level. Use of longitudinal data enabled assessment of trends over time and allowed for effective guideline evaluation at a population level. EMR data allowed for a more valid assessment of prescription rates, as self-reported data often overestimate medication adherence behavior (38). EMR data also allowed us to capture the rates of nonprescription medications including antiplatelet agents and data across greater age ranges (i.e., age <65 years), which are often unavailable with drug registry data.

We carried out statistical inference using an ITS design, which is considered a stronger research design than the traditional before-and-after longitudinal design and remains the recommended quasi-experiment design when a randomized control trial is not feasible at the community level (39,40). The total number of patients prescribed medications per quarter were aggregated into a single population-level proportion, and therefore results correspond with an intervention effect at the population level rather than the individual level. ITS has also been mainly used in conditions applicable to our study: a macro-level intervention (nationwide guideline dissemination strategy), a unidimensional outcome (prescription rates for vascular protective agents), and a large population (of primary care patients with diabetes) (25).

This study highlights a persistent gap in the management of vascular protection among patients with diabetes. Qualitative assessments are needed to assess the collective experience of physicians and elicit factors not captured in our quantitative design (i.e., registries which capture information about why a physician did not prescribe a guideline-recommended treatment). Further dissemination efforts should be targeted at groups with lower adherence rates including physicians from rural areas and those treating patients with lower SES. They may also involve EMR-based prompts to serve as reminders for providers. It is possible that a ceiling level of vascular protective medication uptake has been achieved; however, it is necessary to explore other reasons for guideline nonadherence and revise dissemination strategies accordingly before arriving at this conclusion.

Acknowledgments. The authors would like to thank Carolyn Gall Casey (Diabetes Canada, Toronto, Ontario, Canada) for her continued support and contributions in securing study funding.

Funding. A.V.R. was supported by the Comprehensive Research Experience for Medical Students (CREMS) Stipend. The authors are also grateful to Diabetes Canada for providing funds for a component of data collection and analysis. The Canadian Diabetes Association is the registered owner of the name Diabetes Canada.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. A.V.R. wrote the manuscript. A.V.R., J.N., and C.Y. initially conceived the concept of this study. S.K., B.A., and R.M. performed the statistical analyses. B.A. obtained and managed the data. A.V.R., S.K., J.N., B.A., M.G., R.M., F.S., and C.Y. made substantial contributions to the interpretation of data and revised the manuscript for important intellectual content. C.Y. is the guarantor of this work and as such, had full access to all the data in the study and takes responsibility for the integrity of data and the accuracy of the data analysis.