If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Address for correspondence: Roseanne O. Yeung MD, MPH, Division of Endocrinology & Metabolism, Department of Medicine, 9th Floor Clinical Sciences Building, 11350–83 Avenue, Edmonton, Alberta T6G 2G3, Canada.
Diabetes, Obesity and Nutrition Strategic Clinical Network, Alberta Health Services, Calgary, Alberta, CanadaDivision of Endocrinology & Metabolism, Department of Medicine, University of Alberta, Edmonton, Alberta, Canada
The National Diabetes Surveillance System (NDSS) case definition, which identifies a case of diabetes using administrative health records as “two physician claims or one hospital discharge abstract record, within a 2-year period for a diagnosis bearing International Classification of Disease codes for diabetes,” was compared with expanded case definitions, including pharmacy (PHARM) and laboratory (LAB) data. The PHARM definition included any therapeutic antihyperglycemic agents, and the LAB definition included thresholds of ≥1 glycated hemoglobin measurement of ≥6.5%, or 2 instances of random glucose ≥11.1 mmol/L or fasting glucose ≥7.0 mmol/L.
Methods
In this retrospective study we used administrative data from the Diabetes Infrastructure for Surveillance, Evaluation, and Research project. Descriptive statistics were used to characterize participants by several subgroups.
Results
The NDSS identified 291,242 diabetes cases, indicating a provincial prevalence of 6.83%. Using LAB plus PHARM identified 52,040 additional cases, so the combination of NDSS or LAB or PHARM identified the largest number of cases (n=343,282), increasing the diabetes prevalence estimate to 8.06%. These 3 sources resulted in 7 unique subsets: NDSS only (n=42,606), PHARM only (n=16,310), LAB only (n=32,202), NDSS+LAB (n=32,582), NDSS+PHARM (n=22,503), LAB+PHARM (n=3,528) and NDSS+LAB+PHARM (n=193,551). Refinement using demographic and clinical characteristics allowed presumptive cases of polycystic ovarian syndrome to be excluded.
Conclusions
The widely used NDSS case definition can be enhanced by the addition of LAB and PHARM data. Including PHARM and LAB data identified subsets of the diabetes population, which can maximize the yield for detection of diabetes cases in Alberta and provide a richer understanding of this population to target interventions to improve health outcomes.
Résumé
Objectifs
Au Canada, les données administratives sur la santé tirées du Système national de surveillance du diabète (SNSD) permettent de recenser les cas de diabète définis par « deux demandes de paiement de médecins ou un dossier d’hospitalisation en deux ans pour un diagnostic avec un code de la Classification internationale des maladies correspondant au diabète ». Nous avons comparé la définition de cas du SNSD aux définitions de cas élargies, qui incluaient les données de pharmacie (PHARM) et les données de laboratoire (LAB). La définition des PHARM englobait toute classe thérapeutique d’hypoglycémiant, et la définition des LAB englobait les seuils de ≥ 1 mesure de l’hémoglobine glyquée de ≥ 6,5 %, ou 2 occurrences de glycémie aléatoire de ≥ 11,1 mmol/L ou de glycémie à jeun ≥ 7,0 mmol/L.
Méthodes
Dans la présente étude rétrospective, nous avons utilisé les données administratives du projet Diabetes Infrastructure for Surveillance, Evaluation and Research. Nous avons utilisé les statistiques descriptives pour caractériser les participants en différents sous-groupes.
Résultats
Le SNSD a permis de relever 291 242 cas, soit une prévalence provinciale du diabète de 6,83 %. Grâce à l’utilisation des LAB plus PHARM, 52 040 cas supplémentaires ont été trouvés. Ainsi, la combinaison du SNSD ou des LAB ou des PHARM a permis de trouver le plus grand nombre de cas (n = 343 282), et a fait accroître l’estimation de la prévalence du diabète à 8,06 %. Ces 3 sources ont mené à 7 sous-ensembles distincts : SNSD seul (n = 42 606), PHARM seules (n = 16 310), LAB seules (n = 32 202), SNSD+LAB (n = 32 582), SNSD+PHARM (n = 22 503), LAB+PHARM (n = 3528) et SNSD+LAB+PHARM (n = 193 551). L’affinement à l’aide des caractéristiques démographiques et cliniques a permis d’exclure les cas présumés de syndrome d’ovaires polykystiques.
Conclusions
La définition de cas largement utilisée du SNSD peut être améliorée par l’ajout des données LAB et des données PHARM. L’inclusion des données PHARM et LAB qui a permis d’établir les sous-ensembles de la population diabétique peut maximiser l’efficacité de détection des cas de diabète en Alberta et favoriser une meilleure compréhension de cette population afin de cibler les interventions qui améliorent les résultats cliniques.
Conventional methods of diabetes surveillance must evolve with increasing access to clinical information.
•
New methods including laboratory and medication administrative data can refine diabetes prevalence estimates.
Introduction
The most recent estimates from the Canadian Chronic Diseases Surveillance System (CCDSS) suggest there are almost 3.2 million Canadians living with diabetes (
Public Health Agency of Canada Canadian Chronic Disease Surveillance System at-a-glance---Twenty years of diabetes surveillance using the Canadian Chronic Disease Surveillance System.
). This number increases by approximately 200,000 new cases each year, representing a 3.3% annual rise in the number of those living with diabetes. Caring for people living with diabetes and its complications contributes substantial costs to Canadian health-care systems. Diabetes Canada estimated that the cost of diabetes rose from $6.3 billion in 2000 to $16.9 billion in 2020 (
). Given the cost of treatment and the implications for quality of life in those living with diabetes, it is crucial that surveillance systems for diabetes optimally identify not only those living with diabetes, but also those at higher risk to enable more precise resource allocation and improve both quality and efficiency of care.
Surveillance and planning for the management of diabetes in Canada is based largely on available administrative health-care data. The National Diabetes Surveillance System (NDSS) (
) was established in 1999, supported by the Public Health Agency of Canada, later evolving into the CCDSS to include other chronic conditions, and provides ongoing national surveillance of diabetes. In the NDSS (and CCDSS), diabetes cases are commonly defined as “two physician claims or one hospital discharge abstract record within a 2-year period for a diagnosis bearing International Classification of Diseases (ICD) code, ninth (ICD-9) or tenth (ICD-10) edition definitions for diabetes” (
). This case definition has been widely used and found to have relatively high sensitivity (82.3%) and specificity (97.9%) for identifying diabetes cases, but up to 20% of cases are missed (
). For example, physician claims are not recorded in a standardized manner, and patients with diabetes commonly have multiple comorbidities, so it is possible that physicians may submit claims for the other more pressing conditions for which the patient has been seen, rather than diabetes (
). We recognize that trained medical coders follow some standardized practices to complete discharge records, but the methods of discharge coding vary (
). These limitations may result in inaccurate estimates of diabetes prevalence.
With increasingly available electronic health data, we have the opportunity to enhance diabetes surveillance in the general population. Specifically, in Alberta, the Pharmaceutical Information Network (PHARM) database includes all prescription drugs dispensed by pharmacies and can identify individual dispensed diabetes medications. Alberta also houses a Consolidated Laboratory Data Repository (LAB) for all laboratory investigations performed in the province, including tests for diagnosis and monitoring of diabetes (blood glucose values and glycated hemoglobin [A1C]). Incorporating PHARM and/or LAB data could overcome some of the limitations of definitions based solely on claims data and also provide the opportunity to stratify risk within the identified population. We hypothesize that a diabetes case definition that includes PHARM and LAB data better identifies cases than a definition based on NDSS alone. In this study, we aimed to identify the number of cases using the NDSS case definition compared with expanded case definitions including PHARM and LAB data, and describe the clinical characteristics of the groups identified by these different definitions.
Methods
Data sources and design
Data for these analyses were derived from the Diabetes Infrastructure for Surveillance, Evaluation and Research (DISER) project. DISER is a province-wide project in Alberta, initiated in 2017, with an aim to enhance the data infrastructure for chronic disease surveillance by expanding use of available secondary administrative health data sources (including Physician Claims, Discharge Abstract Database, Alberta/National Ambulatory Care Reporting System, PHARM and LAB). Administratively, DISER is housed within Alberta Health Services, under the stewardship of the Diabetes Obesity and Nutrition Strategic Clinical Network. DISER is a collaborative project between Alberta Health, Alberta Innovates, Alberta Health Services, Alliance for Canadian Health Outcomes Research in Diabetes, the University of Alberta and Novo Nordisk/industry partners.
An independent team of academic and clinical experts (a Clinical Data and Advisory/Management committee) conceptualized and carried out this study.
Case definitions
Potential case definitions were established based on literature and clinical experience, and defined by consensus after discussion among Clinical Data and Advisory/Management committee members. We included the original NDSS case definition, a PHARM definition and a LAB definition, as shown in Table 1. The PHARM definition was based on the validated definition used by the Canadian Primary Care Sentinel Surveillance Network, shown to have sensitivity and specificity >95% (
Diabetes Canada 2018 clinical practice guidelines for the prevention and management of diabetes in Canada: Definition, classification and diagnosis of diabetes, prediabetes and metabolic syndrome.
). In all 3 sources, NDSS, PHARM and LAB records occurring during periods of possible pregnancy were excluded. Within all the cases identified by the combined definition (NDSS or LAB or PHARM), we identified 7 subgroups: 3 subgroups from each of the unique data sources, and 4 subgroups using different combinations of data from these data sources (“combined case definitions”; see footnote in Table 1).
Combined case definitions: NDSS and LAB = first date both thresholds met; NDSS and PHARM = first date both thresholds met; NDSS and LAB and PHARM = first date all thresholds met; NDSS or LAB or PHARM = first date both thresholds met.
A1C, glycated hemoglobin; DAD, Discharge Abstract Database; ICD-9 and ICD-10, International Classification of Diseases, Ninth and Tenth Edition, respectively; LAB, Consolidated Laboratory Data Repository; NDSS, National Diabetes Surveillance System; PHARM, Pharmaceutical Information Network.
∗ Combined case definitions: NDSS and LAB = first date both thresholds met; NDSS and PHARM = first date both thresholds met; NDSS and LAB and PHARM = first date all thresholds met; NDSS or LAB or PHARM = first date both thresholds met.
The identification of prevalent individuals as of January 1, 2017 was based on incidence identified any time before January 1, 2017. For all 3 sources (NDSS, LAB, PHARM), all individuals considered incident before that date who were still alive and still registered in Alberta were considered prevalent according to the applicable criteria. Once individuals were identified as incident, they were then considered prevalent. These administrative data systems came on in a phased approach; claims data became available starting 2001, Discharge Abstract Database data in 2002, PHARM data in 2002 and LAB data in 2009.
Statistical analysis
Diabetes prevalence estimates for Alberta were calculated using the various case definitions listed in Table 1 using the denominator of 4,261,098, which was the number of Alberta residents registered with Alberta Health Services on January 1, 2017. We calculated the percentage of cases captured by each data source from the largest possible number of diabetes cases captured, that is, NDSS or LAB or PHARM (Table 1). Descriptive statistics were used to characterize case definition subsets including age, sex, mean number of A1C tests in 2016, insulin use, other diabetes medications and proportions with an A1C of <6.5%, individuals taking any diabetes medication and those with an A1C of >8%. All analyses were performed using STATA version 16.0 (StataCorp, College Station, Texas, United States).
Results
Figure 1 shows a graphic representation of the number of cases identified by the 7 unique subsets from the 3 sources (NDSS, PHARM and LAB). The NDSS case definition identified 269,714 cases from physician claims and 21,528 from discharge records, for a total of 291,242 unique cases. The PHARM definition identified 235,892 cases with 2 instances of any diabetes medication. The LAB definition identified a total of 261,863 cases. By combining cases identified using NDSS or LAB or PHARM, a total of 343,282 cases were identified, representing a provincial prevalence of 8.06%. Notably, the NDSS+LAB+PHARM subset identified 193,551 cases, or 56.4% of all the possible cases identified by NDSS or LAB or PHARM. Percentages of all cases identified by each data set was 84.8% for NDSS, 66.8% for PHARM and 61.1% for LAB. The 42,606 cases identified only by the NDSS definition (patients who did not fit LAB or PHARM definitions) were, on average, 62.6 years old with a mean A1C of 5.8%, of whom 48.3% were female (Table 2). By comparison, almost 5-fold more cases were identified in the combined NDSS+LAB+PHARM subset (193,551 cases), who were, on average, 61.5 years old, had a mean A1C of 7.7% and were 42.8% female. The subgroup of cases identified only by the LAB definition contained 32,202 cases, with an average age of 62.1 years and a mean A1C of 6.6%, of whom 45.7% were female. The small subgroup identified only by the PHARM definition (n=16,310) contained the highest proportion of female cases (85.3%), with the youngest mean age of 42.4 years and lowest mean A1C of 5.5% (Table 2).
Figure 1Numbers of 7 unique subsets of diabetes cases using NDSS-, PHARM- and LAB-based definitions. LAB, Consolidated Laboratory Data Repository; NDSS, National Diabetes Surveillance System; PHARM, Pharmaceutical Information Network.
The clinical characteristics of the subgroup of cases identified only by PHARM seemed distinct from the other subgroups, prompting more detailed evaluation and refinement of the PHARM definition. The large proportion of younger females, along with lower mean A1C, in the PHARM-only subset suggests this cohort may have included individuals who do not in fact have diabetes but have metformin prescribed for another indication, the most likely being polycystic ovary syndrome (PCOS) or prediabetes. To refine the PHARM subset and exclude presumptive nondiabetic PCOS cases, the following more stringent criteria were applied:
1.
Female on metformin.
2.
Did not meet NDSS criteria.
3.
Had no diagnostic lab tests for diabetes (i.e. as in Table 1).
4.
Did not have other non-metformin diabetes medications prescribed.
Of the 16,310 individuals in the PHARM subset, 10,933 met all these criteria for possible PCOS. Therefore, the refined PHARM-only subset, excluding likely PCOS cases, was 5,377 individuals (Figure 2).
Figure 2Refinement of PHARM subset to exclude presumed polyscystic ovary syndrome cases. LAB, Consolidated Laboratory Data Repository; PHARM, Pharmaceutical Information Network.
Our study using LAB and PHARM data to augment provincial administrative health-care databases to identify diabetes cases identified 343,282 individuals, suggesting a prevalence of 8.06%, compared with 291,242 (prevalence, 6.83%) identified by NDSS, which represents an 18% increase in case finding. It is striking that only 193,551 individuals (56.4%) were identified by all 3 data sets, suggesting they are complementary. NDSS identified an additional 42,606 individuals who did not have medication or laboratory evidence of diabetes, which may reflect the fact that these were individuals with prediabetes; had diabetes that had gone into remission where their diagnostic lab data were taken before our data were extracted or not available in the LAB system while not requiring medication; or incorrect diagnosis cases, as this group had a mean A1C of 5.8%, well below the diagnostic threshold of 6.5%. Given that these 42,606 individuals that the NDSS-only definition identified represent 12.4% of additional cases without medication or lab evidence of diabetes, one may argue these individuals could be considered as low risk and their omission from a diabetes surveillance system like DISER is of minor impact. Conversely, this population could include individuals whose glycemic control is not being monitored and may have suboptimal diabetes care. We hypothesize that the addition of LAB and PHARM data make the diagnosis of diabetes more certain, suggesting that using NDSS alone overestimates the prevalence of diabetes in the population, consistent with estimates of overestimation in other studies (
Furthermore, although NDSS may be limited because it relies on humans to label diabetes based on clinical judgement, we found an additional 41,107 cases (11.9%) that fulfilled definitions for diabetes based on medication and laboratory definitions, which would have been missed by NDSS alone, even after excluding presumptive PCOS cases. A meta-analysis of the NDSS case definition found that up to 20% of diabetes cases were missed, suggesting that the inclusion of laboratory and medication dispensation data are powerful tools in improving accurate detection of diabetes cases (
). A meta-analysis of 6 studies demonstrated a pooled sensitivity of 82.3% (95% confidence interval [CI], 75.8% to 87.4%) and a specificity of 97.9% (95% CI, 96.5% to 98.8%) (
) showed that sensitivity ranged from 26.9% to 97%, and specificity ranged from 94.3% to 99.4% for the NDSS case definition. The same review assessed estimated positive predictive value, which ranged from 71.4% to 96.2%, and negative predictive value, which ranged from 95% to 99.6%. The high positive predictive value of the NDSS is due to the fact it combines data from multiple sources (i.e. physician claims and hospital discharge records), along with an observation period of >1 year.
Sensitivity and specificity were not calculated in our study. However, other studies have shown that combining several data sources may improve identification of cases. Lipscombe et al (
) found that, among 8 algorithms, the best performance characteristics included hospitalization data, physician claims and prescription databases for identifying diabetes cases. The authors showed physician claims represented the data source that drove better performance for all definitions. In our study, NDSS proved to be the main driver for identifying the most cases, again with physician claims contributing the largest proportion; however, as we now have the ability to corroborate physician claims with medication and laboratory data, it is clear we should evolve our population surveillance methods given the known limitations of NDSS.
The richness of data now allows for better understanding of the population and stratification based on increasingly granular clinical characteristics. When observing the different subsets in our study, the PHARM-only subset had the smallest number of cases (16,310), the youngest mean age and the highest proportion of females. The high proportion of females (85.3%) in the PHARM-only subset pointed to the alternative use of metformin for PCOS rather than diabetes (
Effect of metformin on clinical, metabolic and endocrine outcomes in women with polycystic ovary syndrome: A meta-analysis of randomized controlled trials.
). In previous studies, the exclusion of PCOS was challenging because of the lack of integration of medication and laboratory data. Therefore, in refining the PHARM-only subset by excluding all those who are female and taking metformin and had no LAB record for glucose testing and were not identified by NDSS, up to 67% of those in the PHARM-only subset were excluded as presumptive nondiabetic PCOS cases.
Our study has demonstrated the feasibility of accounting for the important clinical variables of medication use and laboratory investigations to segment the population into higher and lower risk strata for surveillance, and provides considerations for future clinical quality improvement. We hope these findings will expand the conversation about how to act on these enhanced data to promote more personalized medicine and more equitable and targeted resource allocation.
Strengths of our study include its use of a large population-based sample through the administrative health databases at Alberta Health Services, which now include pharmacy dispensation and laboratory data, and a cohort of people living with diabetes whose data were linked to these administrative databases. Although the administrative health-care data are not collected for surveillance purposes, claims data can now be corroborated with medication and laboratory data, which can reduce errors of case finding. Our study’s limitations include its inability to calculate sensitivity or positive predictive value of the expanded case definitions. Without sensitivity analysis, it is not possible to validate the accuracy of the expanded case definitions including PHARM and LAB data. Although we recognize validation is necessary using the NDSS+LAB+PHARM definition, we intend to embark on this in the future using electronic medical chart data, as Alberta is in the midst of deploying its provincial electronic medical record system.
In conclusion, our results suggest that enhancing the NDSS case definition with the addition of medication and laboratory data allows for refinement of the resulting definition, further suggesting that the accuracy of diabetes case finding is improved by the addition of medication and laboratory data. Further validation of this expanded case definition should be undertaken. Our findings may also be useful for understanding prevalence of diabetes in the Canadian context as health systems develop improved collection and management processes for laboratory and pharmaceutical data.
Acknowledgments
This work was supported by Alberta Health Services and Novo Nordisk Canada (Grant No. PRJ33527 ).
Author Disclosures
P.Se. reports volunteer roles with Diabetes Canada as Clinical Practice Guidelines Steering Committee Chair and participation on its board of directors. P.Se. and R.O.Y. report salary support from the Alberta Academic and Medical Health Services Program. R.O.Y. has consulted for Novo Nordisk. No other authors have any conflicts of interest to declare.
Author Contributions
All authors helped to conceptualize the study. A.R. provided the initial data tables, and all authors were involved in data analysis and interpretation. N.M. drafted the initial manuscript; all authors contributed to the final draft.
References
Public Health Agency of Canada
Canadian Chronic Disease Surveillance System at-a-glance---Twenty years of diabetes surveillance using the Canadian Chronic Disease Surveillance System.
Diabetes Canada 2018 clinical practice guidelines for the prevention and management of diabetes in Canada: Definition, classification and diagnosis of diabetes, prediabetes and metabolic syndrome.
Effect of metformin on clinical, metabolic and endocrine outcomes in women with polycystic ovary syndrome: A meta-analysis of randomized controlled trials.
00468-8&id=gr1.jpg){kind=link}
00468-8&id=gr2.jpg){kind=link}