Identifying Ontarians with Type 2 Diabetes Mellitus in Administrative Data: A Comparison of Two Case Definitions

Diabetes poses a substantial burden on the Canadian population [1]. In 2020, about 3,772,000 people in Canada were living with diabetes, with an estimated direct cost of $3.8 billion to the healthcare system. By 2030, both prevalence and direct costs are projected to increase by 30% [1]. People living with diabetes are at an increased risk of comorbidities, hospitalizations, premature death and all-cause mortality [1]. As the incidence and prevalence of diabetes is increasing, accurate estimates of the disease burden on the population can inform diabetes prevention and management efforts. Population-based research plays an important role in understanding the trends of diabetes and impacts of interventions [2, 3]. Additionally, it can help assess the economic impact of diabetes on individuals and the healthcare system, guide the optimal allocation of resources, provide valuable guidance for clinical practice, and identify research priorities [4]. National surveys, health systems data, registries, and administrative data are some of the methods used to measure the current estimates of the burden of diabetes [3]. Although population-based research is an important approach in addressing this increased burden of diabetes, their accuracy may be limited by underrepresentation during sampling, selection bias, confounding factors, and misclassification bias [3]. The accuracy of diabetes prevalence estimates can be improved by the development of accurate case definitions to identify people with diabetes among various sources (e.g., ambulatory and hospitalization records) of administrative data [3]. In Ontario, Canada’s most populous province, the ICES data repository captures comprehensive, census-level data of all publicly covered health care services since 1986 [5]. As such, the ICES data reflect day-to-day interactions of Ontarians with the health care system through the government-run health plan known as the Ontario Health Insurance Plan (OHIP), which provides medically necessary services to most residents of Ontario [5]. The ICES data repository currently includes health service records for over 14 million people [6].

The ICES Ontario Diabetes Database (ODD) is a cohort of Ontario residents with non-gestational diabetes identified through a clinically validated algorithm [7]. Previous work has been done using ICES data to identify and validate case definitions for the identification of people with diabetes using the ODD, using various combinations of hospitalization and physician visit codes. These case definitions were validated against diagnoses of diabetes among adults (20 years of age and older) in electronic medical records (EMRs), with varying levels of sensitivity and specificity [2]. In administrative data research, case definitions can be adopted to increase the sensitivity or specificity of the disease depending on the research objectives of the study. A sensitive case definition detects as many true cases as possible but can also include individuals who do not have the disease (i.e., false positives). On the other hand, a specific case definition detects only those individuals who truly have the disease and may incorrectly classify some who have the disease as not having the disease of interest (false negatives) [2].

This study was conducted to identify patient populations in Ontario administrative data using previously validated sensitive and specific diabetes case definitions, and to compare the demographic and clinical characteristics of these patient populations to understand their impact on characterizing the disease.

In this study, based on their most recent HbA1c values, 1,265,329 patients were identified using sensitive case definition and 938,952 patients using specific case definition from the ODD (Supplementary File Table S3). Of these patients, 150,461 and 145,872 patients with type 1 diabetes were excluded from the sensitive and specific cohorts respectively. Further, patients aged < 15 years at ODD case detection, those aged < 30 years at index date, those who were OHIP ineligible at index or any of previous four quarters, or those who died before index were excluded. The final sensitive cohort included 1,093,812 patients and the final specific cohort included 783,228 patients, with an estimated prevalence of 7.9% and 5.7% of Ontarians living with type 2 diabetes mellitus, respectively.

This study was a large, population-level analysis of patients with type 2 diabetes mellitus using administrative data in Ontario, Canada. The results from this study provide a comprehensive look at the demographic and clinical characteristics of patients with type 2 diabetes mellitus in Ontario. For this study, two case definitions for diabetes were considered sensitive definition (at least two physician visits with diabetes OR at least one drug claim for diabetes OR at least one hospitalization with diabetes within 1 year) and the specific definition (at least three physician visits with diabetes within 1 year). In this study, more patients with type 2 diabetes mellitus were identified by the sensitive case definition compared with specific case definition; however, the demographic and clinical characteristics were mostly comparable between the cohorts. The majority of patients presented with comorbidities in both the cohorts.

In this study, the prevalence of type 2 diabetes in Ontario was estimated to be 7.9% based on the sensitive case definition and 5.7% based on the specific case definition. As per 2016–17 estimates from the Canadian Chronic Disease Surveillance System, the prevalence of diabetes in Canada was 8.8% [12] and about 90% of diabetes cases in Canada are attributed to type 2 diabetes [13]. Hence, the estimated prevalence of type 2 diabetes based on the sensitive definition from our study is in line with that of the data observed from Canada. Patients in both the cohorts had a mean age of approximately 64 years, and there were slightly higher proportion of male patients compared with female patients. Nearly one-fourth of patients (23%) belonged to quintile 1 (lowest income quintile) and 16% to quintile 5 (highest income quintile). The demographics observed in our study are comparable with the demographics of patients with diabetes included in a large population-based retrospective study whose diabetes status was identified from ODD [14]. The patient population had a mean age of ~ 58 years and were more likely males. About 24% of patients were under quintile 1 and 16% of patients under quintile 5 [14]. In our study, while there were very little differences on most demographic characteristics, the specific cohort definition had a slightly lower proportion of patients with a 1-year mean HbA1c of < 7% (53 mmol/mol) at 56% compared to 64% in the specific cohort. Although patients with HbA1c greater than 7% (53 mmol/mol) are at an increased risk of comorbidities [15, 16], the proportion of patients with comorbidities in both cohorts were comparable, which suggests that both case definitions are identifying similar patient populations.

Within our study cohorts, many patients with type 2 diabetes mellitus were identified with comorbid conditions. The most commonly reported comorbidities were hypertension followed by dyslipidemia, CIHD, microalbuminuria and angina. Thus, most of the patients with type 2 diabetes present with cardiovascular comorbidities, which is in line with the published literature [17, 18]. Notably, hypertension was reported in 77% of patients in the sensitive cohort and 80% of patients in the specific cohort. The 2006–2007 data from Canadian Chronic Disease Surveillance System (CCDSS) reported that 63% of adult patients with diabetes had hypertension as a comorbidity [19]. In a retrospective study conducted during July 2014 to June 2015, the prevalence of hypertension in patients with type 2 diabetes was 82% [20]. This reflects that the prevalence of hypertension observed in this study is comparable with the data from other studies. Obesity, on the other hand, was reported in only around 21% of patients with type 2 diabetes in this study, whereas in another study, prevalence of obesity/overweight in this patient population was about 78% [20], implying that obesity may have been underreported in our study.

Our study has applicability to other administrative databases where case definitions may be used to identify patients based on interactions with the healthcare system due to the lack of diagnosis information in an outpatient setting. Our findings show that the patients identified in both of our case definitions are similar on demographic and clinical characteristics, supporting the use of both definitions for research purposes. However, our research would recommend using sensitive definitions where estimates of prevalence and population burden are required due to the larger sample size captured.

This study has several strengths. As Ontario is a single payer system for medical claims, the data used in this study is near census. Hence, the study does not risk missing a recorded diagnosis of diabetes mellitus if a patient requires care from several uncoordinated sources, while incomplete medical history or utilization from a chart audit may face this risk. Moreover, the long study duration helped retrieve comprehensive data and better reflect real-world conditions, thereby resulting in an increased external validity. In addition, the study showed a consistency in coding over time and used validated sensitive and specific case definitions.

The study also has a few limitations. Firstly, the study is limited by ambiguity in the OHIP diabetes diagnosis codes, which do not specifically identify type 1 or type 2 diabetes mellitus within the community setting. To overcome this, we used several inclusion/exclusion criteria to eliminate from the cohort those that had a high probability of having type 1 diabetes mellitus. Secondly, the administrative databases include only publicly reimbursed medical claims and therefore, cash and private prescription drug claims are not captured in this study. Moreover, individuals aged at least 65 years or those receiving professional home and community care services or receiving benefits from Ontario Works or Ontario Disability Support Program or enrolled in Trillium Drug Program are only considered eligible for Ontario Drug Benefit program. Thirdly, case definitions still have a risk of misclassification, but this is minimized through high sensitivity/specificity [2]. Fourthly, individuals who are more effectively managed may have a lower probability of undergoing HbA1c testing, potentially leading to the introduction of selection bias. Additionally, the comorbidities were derived using ICD-10-CA codes, some of which may be under-reported (e.g., obesity). Finally, the data for this study was collected before the widespread use of sodium–glucose cotransporter 2 (SGLT-2) inhibitors and glucagon-like peptide 1 (GLP-1) receptor agonists for treatment of type 2 diabetes mellitus, which may have impacted the prevalence of cardiovascular comorbidities.

The prevalence of diabetes in Canada is increasing over time, and about 90% of all diabetes cases are type 2. Striking the right balance between sensitivity and specificity in defining diabetes in administrative is important to ensure accurate capture of patient populations to better understand interventions and health outcomes for individuals with type 2 diabetes mellitus. Our results also highlighted that individuals with type 2 diabetes commonly experience comorbidities, regardless of the case definition used. This emphasizes the need for an interdisciplinary approach to diabetes care and associated comorbid conditions.