This was a retrospective cohort study; we used the Strengthening the Reporting of Observational Studies in Epidemiology guidelines to report the study [18].
Setting and participants
This study was conducted using data from Ontario, Canada. Ontario’s health care system includes universal, tax-funded coverage for all medically necessary physician services and laboratory tests, with no patient co-payment. About 75% of Ontario’s population is formally registered or enrolled in primary care; each patient is attached to the practice of a family physician. 75% of family physicians practicing comprehensive care use a patient enrollment model. 25% of physicians in patient enrollment models practice in inter-professional Family Health Teams [19].
Two different validated algorithms to define two cohorts of patients living with diabetes were used to examine our data. The first used the CPCSSN (EMR derived) algorithm [20] and the second used the Ontario Diabetes Database (ODD), derived using an administrative data algorithm [14].
EMR cohort
We used EMR data from the Ontario sites of CPCSSN, extracted up to the end of 2014 and linked at ICES [21]. The four sites included were: Toronto and surroundings, eastern Ontario, Hamilton and London. The CPCSSN algorithm for diabetes [12] was used to identify patients with diabetes in that dataset.
Linked CPCSSN-ICES data from January 1st, 2010 to December 31st 2016 were included. We retrieved data on patients that were present in the CPCSSN dataset, had a valid Ontario health card number, were enrolled or virtually enrolled to a primary care physician and were age 66 or more at the beginning of each quarter of interest. Patients were enrolled in the cohort during or after the quarter containing their first patient-physician encounter as recorded in CPCSSN, if they met the CPCSSN definitions for diabetes at any time prior to each quarter of interest.
Administrative cohort
We replicated the study using Ontario population based administrative data. Ontario has a population based publicly funded healthcare system; a unique health care number (the Ontario Health Insurance Plan or OHIP) identifies each patient. We used a validated administrative definition for diabetes, the ODD, which is a population-based registry at ICES [14]. The ODD validated algorithm labels cases as diabetes if there are two physician billing claims for diabetes (ICD9 code 250) or one hospitalization discharge record with a diagnostic code of diabetes within 2 years [14]. Patients likely to have gestational diabetes are excluded by not counting billing claims or hospitalizations that appeared 120 days before or 180 days after a hospital record with a diagnosis of pregnancy care or delivery [14].
Similar to the EMR cohort, ICES data from January 1st, 2010 to December 31st 2016 were included. We retrieved data on patients that had a valid Ontario health card number and were age 66 or more at the beginning of each quarter of interest. Patients were enrolled in the cohort if they met the ODD definitions for diabetes at any time prior to each quarter of interest.
Patients were censored from the cohort during the quarter in which they met at least one of the following conditions 1) deceased; 2) last encounter with the healthcare system prior to study end date; 3) loss of eligibility for Ontario’s publicly funded health insurance plan, OHIP; 4) No longer enrolled or virtually enrolled with a primary care physician.
Variables
The outcome of interest was prescription coverage rates as an indication of whether a patient with diabetes was likely taking an appropriate cardiovascular protective medication in each quarter of interest as per Diabetes Canada guidelines between January 1, 2010 and December 31, 2016. Proton pump inhibitors (PPIs) were used as a comparator for secular trends.
The medications of interest were: 1) Statins; 2) ACE Inhibitors; 3) ARBs; 4) Antiplatelets (only for those with no cardiovascular disease); 5) PPIs.
Data sources
ICES holds databases on OHIP eligibility, publicly funded medications, and encounters with the health care system such as inpatient and outpatient visits. These datasets were linked using unique encoded identifiers and analyzed at ICES. All patients age 65 or more and covered by OHIP are eligible for publicly funded medication coverage through the Ontario Drug Benefit (ODB) program. The ODB database, available at ICES, contains a copy of information on all ODB eligible medications dispensed in Ontario pharmacies including date dispensed, number of days supplied, cost and Drug Identification Number (DIN), .
Data on medications were obtained using the ODB database. We used the index date corresponding to the date when the medication was dispensed by the pharmacist.
Linked databases at ICES were used to define patient eligibility. These included claims data using the OHIP database for dates of encounters and the Registered Persons Database (RPDB) to define OHIP eligibility, basic demographics and date of death. The Client Agency Program Enrollment (CAPE) tables were used to identify enrollment with a primary care physician; for patients not enrolled, virtual enrollment to a physician was defined through their pattern of care (physician with the majority of primary care billings for a patient). The Ontario Myocardial Infarction Dataset (OMID) [22] were used to define the presence of cardiovascular conditions.
The Johns Hopkins Adjusted Clinical Groups (ACG) method was used to measure disease burden. Adjusted diagnostic groups (ADGs) cluster groups of diagnostic codes (ACGs) to categorize illnesses and predict health care utilization. These diagnostic clusters, or conditions have similar clinical criteria and expected use of resources; there are 32 diagnostic clusters. Each individual patient is assigned ADGs based on their health conditions; a patient can be assigned between zero and 32 ADGs, with more ADGs indicating greater health care needs [23].
ACGs are also used to group patients by expected health care utilization, or Resource Use bands (RUBs). RUBs range from zero (non-users of health care) to five (very high users) [24].
Data from the Ontario portion of CPCSSN have been linked at ICES; linkage rates were 98.7% [21]. The population included in the linked data was slightly older, included a greater proportion of women and was more likely to live in urban areas than the Ontario population. The linked, combined data were used for this study.
Quantitative variables
An individual was considered to be covered (likely taking a medication) during a given quarter if they were found to have a supply of this medication with a duration overlapping at least 75% of days in that time period. The medication did not have to be dispensed in the given quarter, as the maximum allowed supply in ODB is 100 days. For early refills, we restricted the maximum leftover supply to be 100 days. We grouped ACEis and ARBs in the analysis as they were considered to be equivalent in terms of cardiovascular protection [25]. ARBs are often substituted if ACE is cause side effects such as coughing [25].
This study used an interrupted time series model to determine the impact of the 2013 guidelines on the prescription rates of statin, ACEi/ARB and antiplatelets on a quarterly basis from 2010 to 2016. In particular, the primary outcome included the proportion of eligible patients who had a supply of statin, ACEi/ARB or antiplatelets medication for at least 67 days (75% of days or more) within a 90-day quarter. A total of 13 and 14 quarters were selected before and after the implementation of the Diabetes Canada guidelines in April 2013 (Quarter 2). The intervention effect is assessed using two parameters: (i) a change in trend (i.e. difference in pre-slope and post-slope) and (ii) a change in level (i.e. difference at the beginning of the intervention) [26].
Since the Diabetes Canada guidelines are readily available as a public resource to clinicians, this interrupted time series model was fitted without a roll-in period of the intervention. As part of the sensitivity analyses, the interrupted time series analysis was repeated by excluding patients with no encounters in the EMR (CPCSSN dataset), where there was therefore no opportunity to prescribe medications in primary care.
The analyses were carried out using SAS version 9.4 (SAS Institute. SAS/STAT Software, Version 9.4. SAS, Cary, NC, 2012) and p-values < 0.05 were considered to be statistically significant.