Using Data to Improve the Management of Diabetes: The Tayside Experience

The Saint Vincent Declaration in October 1988 was a prescient development in diabetes care (1). The global diabetes community, from government health departments and patients’ organizations in Europe to the World Health Organization in Europe and the International Diabetes Federation Europe, came together to agree to programmatic targets for diabetes care in relation to diabetes outcomes and complications.

It was clear, however, that identification of all individuals with diabetes in the population was essential if diabetes care was to be effective and measurable and if the targets of Saint Vincent were to be met. In the late 1980s, registries of patients with type 1 diabetes were relatively common, but there were few if any comprehensive registries of patients with type 2 diabetes in the U.K. At that time the impact of type 2 diabetes had been grossly underestimated.

To address this challenge, an interdisciplinary collaboration of general practitioners, diabetologists, and senior health service managers convened in 1994 to establish a region-wide population-based diabetes monitoring control system. The aim was to develop state-of-the-art information technology to achieve quality assurance of the provision of health care for patients with diabetes in Tayside, Scotland (current population 416,000) (2). The secondary purpose was to create a trustworthy approach to population-wide diabetes research—supporting clinical trials, epidemiology, health service research, and the molecular understanding of diabetes care and its complications.

In 1996 the diabetes audit and research in Tayside Scotland (DARTS) study was launched (3). It was a collaboration between 278 practitioners, 78 practices, and three health trusts or hospitals (Ninewells Hospital and Medical School, Perth Royal Infirmary in Perth, and Stracathro Hospital in Brechin)—all in Tayside. Arguably, it was the first time in the U.K. that an entire population subscribed to a joint single strategy for diabetes care and research that spanned primary, secondary, and tertiary care and a care community that worked with patients to agree to comprehensive data sharing for care and research.

In the first phase, eight independent data sources were used to maximize complete ascertainment of cases of diabetes. This was enabled by the fact that in Scotland every person registered with a general practitioner is given a unique identifier called the Community Health Index (CHI) number. This enabled linkage of diabetes prescriptions, hospital diabetes clinics, data from a mobile diabetes retinopathy screening unit, a biochemistry database, and the Scottish Morbidity Record (SMR01) (which is a list of discharges from hospital coded according to the ICD-9 and -10 ontologies).

This enabled the DARTS Team to identify 7,596 patients living with diabetes, with methodology validated through annual validation of case records of a random selection of general practices using linked prescription, laboratory testing, and eye screening services (Fig. 1). In this initial record linkage study a population-based diabetes registry was created that acted as the fulcrum of >25 years of diabetes research. The clinical, audit, and research utility of DARTS in Tayside led to its adoption by the Scottish government, with rollout across Scotland initially as Scottish Care Information – Diabetes Collaboration (SCI-DC) in 2002 and more latterly as SCI-Diabetes (4). SCI-Diabetes achieved full national coverage during 2006.

The SCI-Diabetes clinical database now includes electronic health care records for all patients with diabetes in Scotland, except for <0.5% who have opted out of the system, allowing joined up care for patients with diabetes wherever they are in Scotland. This is the source of a comprehensive annual audit of diabetes care (see the Scottish Diabetes Survey 2020 [5]) and is used for research by the Scottish Diabetes Research Network Epidemiology (SDRN) Epidemiology Group (6). These studies comply with the Declaration of Helsinki.

However, in this article we focus specifically on Tayside—where we have the longest follow-up for patients with diabetes and comprehensive data linkage to a large breadth of clinical data and linkage to a large biorepository. Leveraging these data, the Tayside Diabetes Managed Clinical Network publish annual reports and information on the status of diabetes and diabetes care for professionals and patients of this health board (7). For research purposes, the Tayside diabetes and associated linked data are managed by the Health Informatics Centre (HIC) based at the University of Dundee, which acts on behalf of the NHS Tayside health board as a Scottish regional Safe Haven (8). HIC provides secure, managed access to health data for researchers (https://www.dundee.ac.uk/hic/). All data are available and analyzed in an ISO27001-accredited Safe Haven environment, and research projects undergo a strict governance approval process. The data are normally processed under GDPR lawful basis of public interest. Data provision and linkage were carried out by the HIC (https://www.dundee.ac.uk/hic), with analysis of anonymized data performed in the ISO27001-accredited and Scottish government–accredited Safe Haven. HIC standard operating procedures were reviewed and approved by the NHS East of Scotland Research Ethics Service, and consent for this study was obtained from the NHS Fife Caldicott Guardians.

Routinely collected clinical data are collected and presented in the SCI-Diabetes electronic medical record through automated linkages with a series of electronic systems. Using the CHI master patient index as a unique identifier and a clearly defined clinical data set, records are collected and linked from primary care systems, laboratory systems, retinopathy screening, the Scottish Ambulance Service, hospital inpatient systems, ward-based blood glucose systems, and others (as seen in Fig. 1). These data are updated at least once every 24 h, with laboratory data updated every 4 h. Hospital clinicians across Scotland use SCI-Diabetes as their sole electronic medical record for diabetes care, contributing routine data and screenings directly into the system front end. Key features of the platform include specialist functionality for primary and secondary care, podiatry, dietetics, diabetes specialist nursing, and inpatient management. A flowchart demonstrating the combination of diabetes care data from different sources is shown in Fig. 1.

Patients are added to SCI-Diabetes directly through patient registration forms where diabetes type and date of diagnosis are collected. In parallel, as diagnoses are recorded in primary care systems, these records will also be collected and processed to ensure that no patients are missed. As soon as an individual is registered with a general practitioner in Scotland and a diabetes diagnosis is recorded, their data will be passed to SCI-Diabetes. The active population is maintained through linkage to the national CHI, which provides notifications as people move around Scotland, leave the country completely, or pass away. Consequently, there are very limited missing data. As such, if the data are collected, they will be available in the SCI-Diabetes system.

For research purposes, many other national and local data sets can be linked, including records of births and deaths, prescribing, primary care, laboratory testing, hospitalizations, surgeries and procedures, echocardiography, vascular laboratory, renal disease, cancer, mental health registries, Scottish Ambulance Service, and accidents and emergencies (or emergency room) admissions. These are detailed in Supplementary Material. Additionally, a flowchart detailing important stages in the development of this resource can be found in Supplementary Fig. 6.

In Fig. 2 we show the number of individuals living with type 1 and type 2 diabetes in the Tayside area in each calendar year beginning in 1996 through to 2020. The data show an increase in the number of people with type 2 diabetes over the years until 2020. The stagnation observed in 2020 is most likely reflective of the impact of the coronavirus disease 2019 pandemic on the detection and diagnosis of diabetes. For type 1 diabetes the prevalence has been largely constant from 2006; the increase prior to this could reflect improved ascertainment and diagnosis.

Prescribing history is available for all individuals residing in Tayside. With use of these prescribing data, Fig. 3 shows that the use of insulin has remained constant over the last 20 years, with a shift to increase in the use of triple therapy with fewer diet-treated, monotherapy-treated, and dual therapy–treated individuals. Figure 4 highlights the major polypharmacy experienced by patients: in 2019, over and above glucose-lowering drugs, 70% of patients with type 2 diabetes were on ≥5 drugs and 30% were on ≥10 drugs. The average number of medications used by this population of individuals is provided in Fig. 4B. As observed, from 2005 to present, the average number of drugs used by individuals with type 2 diabetes was 5.

The SCI-Diabetes system incorporates a guideline-driven clinical decision support, which was developed and trialed in Tayside prior to wider rollout across Scotland (9). This clinical decision support tool prompts for tests (laboratory or screening) to be requested if they have not been recorded in the last 15 months and prompts regarding compliance with national guidelines. For example, if a patient has established microalbuminuria but is not treated with an ACE inhibitor or angiotensin receptor blocker (ARB), a pop up will appear as the patient’s records are accessed prompting the clinician to consider an ACE inhibitor or ARB. More simply, an alert will appear reminding the clinician that foot or eye screening is overdue. This simple support tool doubled the likelihood of a patient being screened for most complications and resulted in improved HbA_1c.

My Diabetes My Way (MDMW) is the NHS Scotland online portal supporting self-management for people living with diabetes across Scotland (10). Development began in Tayside in 2006 following funding from the Scottish Diabetes Group, with subsequent launch of the service in October 2008. While initially launched as a Web-based collection of interactive information resources, the service has evolved with the implementation of online records access for patients (December 2010), the introduction of a dedicated mobile app in August 2018, and online structured eLearning courses throughout 2019. Online records access has been a major eHealth objective for the Scottish government, with MDMW offering patients access to a focused set of their SCI-Diabetes data alongside information to explain clinical measurements and links to information that is tailored to their condition (e.g., personalized information based on their current foot screening status). Patients can also contribute to their record by supplying home-recorded results (6,724 patients; 436,379 records by the end of 2021), patient-reported outcome measures (2,475 patients; 4,828 forms), and personal goals (2,956 patients; 11,399 records). Many diabetes devices also connect to the platform, including blood glucose meters and continuous glucose monitors, while activity can be tracked through a collaboration with Fitbit (1,216 users by end of 2021). The service reported 32,743 active users at the end of 2021, with evaluation showing improved patient satisfaction and a more recent health economic analysis demonstrating significant cost savings and improvements in quality of life associated with use of the service (11,12). Example screenshots from MDMW can be found in Supplementary Figs. 1–5.

Historically, diagnosis of type 1 diabetes has largely been made on clinical grounds, which has led to some patients being misdiagnosed as having type 1 diabetes. In the Using pharmacogeNetics to Improve Treatment in Early-onset Diabetes (UNITED) study (13), patients with diabetes diagnosed before age 30 years, still currently under the age of 50 years, from Tayside and the Exeter region, were systematically screened with use of a pipeline that included urinary C-peptide–to–creatinine ratio, pancreatic autoantibodies, and a monogenic next-generation sequencing panel. We showed that of 1,365 patients, 386 had measurable insulin secretion; of these 216 were pancreatic autoantibody negative, and of these 17 had monogenic diabetes (13). The remaining 199 included patients with type 1 diabetes, type 2 diabetes, and other forms of diabetes. Some patients with monogenic diabetes and type 2 diabetes were able to stop insulin—some of whom had been on insulin for many years. The clinical utility and importance of C-peptide testing were highlighted by research from Strachan and colleagues in Lothian, Scotland, who tested a random serum C-peptide in 859 individuals with a clinical diagnosis of type 1 diabetes. They showed that 114 (13.2%) had C-peptide ≥200 pmol/L and reclassified the diagnosis to non–type 1 diabetes for 58 individuals (6.7% of the tested cohort), with most reclassified to type 2 diabetes. Overall, 1.5% successfully discontinued insulin, while a further 1.9% had improved glycemic control following the addition of cotherapies (14).

This work, among others, has established the clinical utility of C-peptide testing in patients with type 1 diabetes and has led to a national diagnostic pipeline for type 1 diabetes—we believe a first in the world. Importantly, the testing pipeline has been embedded within the SCI-Diabetes system to prompt C-peptide testing in those diagnosed >3 years previously, antibody testing in those who have retained C-peptide secretion, and the use of monogenic diabetes gene sequencing and derivation of a type 1 diabetes genetic risk score. In this way we hope that all patients with diabetes in Scotland will be accurately diagnosed and appropriately treated.

Glycemic trends have improved over the last decade in Scotland based on an analysis of 30,717 people with type 1 diabetes, registered anytime between 2004 and 2016 in the national diabetes database. Overall, we reported that median (interquartile range [IQR]) HbA_1c decreased from 72 (21) mmol/mol in 2004 to 68 (21) mmol/mol in 2016. The largest reductions in HbA_1c in this period were seen in children, from 69 (16) mmol/mol to 63 (14) mmol/mol, and adolescents, from 75 (25) mmol/mol to 70 (23) mmol/mol (15). In this population, improvement in HbA_1c appears to be the result of greater use of technology. Among the 4,684 individuals with type 1 diabetes who commenced continuous subcutaneous insulin infusion (CSII) between 2004 and 2019, HbA_1c was shown to decrease after CSII initiation, with a median within-person change of −5.5 mmol/mol (IQR −12.0, 0.0), while the crude diabetic ketoacidosis event rate was also significantly lower in post-CSII person-time compared with pre-CSII person-time: 49.6 events (95% CI 46.3, 53.1) per 1,000 person-years vs. 67.9 (64.1, 71.9) (16).

More recently, we also documented population-level improvements in glycemic control following the widespread introduction of flash glucose monitoring (FGM) (17). At the time of analysis, the prevalence of ever FGM use had increased to 45.9% of the population with type 1 diabetes. Use varied between 64.3% among children aged <13 years and 32.7% among those aged ≥65 years and between 54.4% and 36.2% in the least deprived versus most deprived quintile. Overall, median within-person change in HbA_1c in the year following FGM initiation was −2.5 mmol/mol (IQR −9.0, 2.5), with this change varying markedly by preusage HbA_1c: −15.5 mmol/mol (−31.0, −4.0) in those with HbA_1c > 84 mmol/mol (9.8%). Interestingly, benefit of FGM was found in all age bands, both sexes, and all socioeconomic strata, and there were major reductions in rates of diabetic ketoacidosis and severe hypoglycemia (17). Taken together these findings show population benefit in real-world clinical practice of technologies in type 1 diabetes and that, if their use was more widespread, it might be possible to address some of the health inequalities that have emerged in health care delivery.

The ability to collect and link data on large numbers of patients from different sources permitted the evaluation of retinal photography when it was introduced as an alternative to direct ophthalmoscopy. In an early study on community-based retinal screening in 2,112 patients, investigators using nonmydriatic photography identified ∼5% of people with diabetes as having retinopathy unrecognized in previous checks using direct ophthalmoscopy (18), with increased prevalence especially in rural areas (19). The “new” process also seemed acceptable to people with diabetes (18). These data helped accelerate the adoption of retinal photography, which in turn accelerated the development of diabetes clinical registers due to the need for accurate patient information at a practical patient interfacing level. Eventually, these data improved patient research (e.g. in DARTS and SCI-Diabetes) and improved general diabetes care.

As techniques in retinal photography developed and became digitalized, regional retinal screening programs became established. Linked Tayside SCI-Diabetes data demonstrated that rates of retinopathy declined with time from 28% to 24% and referral rates to ophthalmology decreased from 5.9% to 3.1% from 2002 to 2004 (20). This was achieved through screening previously unreached patients despite an increasing incidence of diabetes. The rates of retinopathy were similar to those of other parts of the U.K. such as Liverpool (21) and showed a lower incidence than seen in the landmark Wisconsin Epidemiologic Study of Diabetic Retinopathy (WESDR), which was one of the earliest epidemiological studies on retinopathy, from 20 years earlier (22). During the interim, diabetes care had improved. From 2001 to 2006 there was a 2.5-fold reduction in the percentage of people with diabetes receiving either any laser or incident laser treatment in Tayside (23), reflecting improved diabetes management and new retinal screening modalities. These data were some of the first in the world to show a reduction in the use of retinal laser in a real-world population setting and were before anti-VEGF treatment became generally used in the U.K. This was followed by reductions in blindness rates (24), resulting in diabetes no longer being the most common cause of blindness in the working age-group in the U.K.

Demonstrating reductions in use of laser and blindness using the new digital retinal screening resulted in the program becoming a recognized national screening program in 2006. Population-based data helped identify novel risk factors such as social deprivation (25) and that one missed appointment was associated with a threefold increased risk of needing laser photocoagulation, emphasizing the need for good population coverage (25). Ongoing work linking data across all Scotland, Wales, Northern Ireland, and seven regions in England reviewed outcomes in 354,549 people with diabetes, which helped refine the screening program by showing that it was safe for patients with no retinopathy over two successive years of screening to have their screening interval increased to two yearly (26). This policy has now been introduced in Scotland to make the program more efficient at a time when the prevalence of diabetes continues to increase. This was particularly valuable in the recent coronavirus disease 2019 pandemic, when retinal screening capacity was greatly reduced.

Thus, the Tayside and Scottish population-based data have helped validate and implement nonmydriatic retinal photography, which is now a national screening program in many nations. Data have shown a decreasing rate of referral to the eye clinic and a decreasing need for laser photocoagulation as a result of screening. Data have also helped in adapting the program to offer screening every two years for a select group of patients.

One of the earliest population-based studies on amputation estimated rates across Tayside to be ∼2.4 per 1,000 people with diabetes, which was 12-fold higher than among the population without diabetes (27). As SCI-Diabetes came into use across Scotland, it became possible to monitor national amputation rates using linked data sets such as diabetes records, hospital admissions, primary care records, and others as previously described. Scotland was one of the first nations to report a falling amputation rate from a full national data set, with a decrease from 3.04 to 2.13 per 1,000 people with diabetes undergoing any amputation between 2004 and 2008 (28). This represented a 29.8% decrease in all diabetes-related amputations with a 40.7% decrease in major amputations. Similar data were reported from the U.S. about the same time (29).

Collection of foot ulcer data is more challenging than for amputation data, as foot ulcer care is delivered by so many diverse clinical groups across primary and secondary care. One of the best studies examining ulcer incidence in North West England involved telephone follow-up of 9,710 patients previously selected to be screened in the community (30). Of the two-thirds of patients who could be contacted for follow-up, 2.2% self-reported developing an ulcer each year. It is possible that patients with less severe diabetes, or shorter duration of diabetes, may have been less likely to be recruited, and so it is possible this may have represented an overestimate. Recently reported Scotland-wide data, with data linkage to identify all ulcers from multiple sources, showed an annual foot ulcer incidence of 1.1%, with 0.7% being first time ulcers (31). Although this may be an underestimate, it represents one of the first population-based national estimates of foot ulceration. These data also showed that patients with foot ulcers are at much greater risk of death than amputation (5.3-fold higher in type 2 diabetes and 2.4-fold higher in type 1 diabetes [30]), indicating that a focus on treating cardiovascular risk factors for people with foot ulcers is essential.

Foot risk stratification has become increasingly used in health care settings to help target the use of scarce resources such as podiatry to where it is needed the most. Early Tayside-based SCI-Diabetes database studies enabled foot risk stratification to be validated. In Tayside, 3,526 patients were followed up and patients deemed to be at high risk according to a simple clinical algorithm were 80-fold more likely to develop an ulcer during 1.7 years of follow-up (32). Possibly more importantly, low-risk patients had a 99.6% (95% CI 99.5–99.7) chance of remaining ulcer free, indicating that it should be safe to direct resources to patients at higher risk. A follow-up study of 7,184 patients confirmed the findings and showed that the risk stratification tool also predicted healing in those who did develop an ulcer (33). Findings of these studies have been confirmed by others with more detailed (34) and less detailed (35) clinical assessments. Like patients with ulcers, those with high-risk feet are ninefold more likely to die than undergo an amputation according to population-based Scottish data (36).

People with foot disease tend to be treated in many different locations, such as specialist diabetes, vascular, and orthopedic clinics, as well as outpatient general and specialist podiatry services. This makes collecting reliable epidemiological data particularly challenging. In contrast, SCI-Diabetes links data from multiple sources such as diabetes clinic encounters, outpatient podiatry, surgical procedures, prescribing, and laboratory data. Data linkage enables the SCI-Diabetes data set to be continuously supplemented and cross validated.

Thus, Tayside and Scottish population-based data have helped validate foot risk stratification systems that are now widely used in practice, helping patients with the greatest need have priority for scarce resources such as skilled podiatrists. The data have also helped define ulceration and amputation rates and document declining amputation rates.

Tayside diabetes health data have been used extensively over the last 20 years to help provide insight into the causes and consequences of diabetes, the risks of micro- and macrovascular complications, and pharmacoepidemiology. Here we highlight just a few examples of published research with use of these population data, linked prescribing, electronic health records, and genetic studies from bioresources.

Capitalizing on the retinal screening efforts in Tayside, epidemiological studies have explored the role of blood pressure and glycemic exposure in the risk of progression of retinopathy (37), as well as novel biomarkers of retinopathy progression (38). Recently, retinal images from the screening program have been subjected to a semiautomated artificial intelligence platform, VAMPIRE (Vascular Assessment and Measurement Platform for Images of the REtina). This helps classify features of morphology such as vessel tortuosity and fractal dimensions. Researchers in Tayside have demonstrated that according to features of retinal morphology one can predict cardiovascular risk independent of polygenic risk scores and traditional clinical risk factors (39). Additionally, biomarker studies for diabetic nephropathy have also been carried out in Tayside (40). Recently, these data were used to establish the higher rate of acute kidney injury in individuals with type 2 diabetes compared with those without diabetes. However, it was found that decline in renal function following an acute kidney injury was slower in those with diabetes (41).

The rate of glycemic deterioration from diagnosis of type 2 diabetes (T2D) in a real-world cohort was modeled using a linear mixed model with adjustment for drug exposure over time. We showed the rate of glycemic deterioration was a 1.4 mmol/mol increase in HbA_1c per year (95% CI 1.3, 1.4) (42), but with little progression in the elderly (those >70 years old at diagnosis) and more rapid in those with younger age and low HDL cholesterol. These data have been used to demonstrate how a younger age of onset and a poor lipid profile are predictors of glycemic deterioration (42,43). More recently, these data have been used to demonstrate how >60% glycemic variability (HbA_1c variability score [HVS], defined as visit-to-visit changes >5.5 mmol/mol) is associated with increased risk of cardiovascular and microvascular complications in people living with type 2 diabetes (44).

As of January 2022, >24,000 individuals (comprised of 9,000 individuals who do not have diabetes and 15,000 individuals with T2D) have genome-wide array data available as part of the Genetics of Diabetes Audit and Research in Tayside Scotland (GoDARTS) and the Scottish Health Research Register and Biobank (SHARE). (45,46). While GoDARTS is limited to individuals living in the Tayside and Fife regions of Scotland, SHARE is a Scotland-wide resource. All individuals with diabetes in GoDARTS or SHARE are part of the SCI-Diabetes system, while those without diabetes are members of the local population. All participants have linked prescribing, electronic health, and specialist registry data. Genetic data are available from genome-wide arrays with imputation performed against both 1000 Genomes and the Haplotype Reference Consortium. With longitudinal linkage of >20 years of prescribing, biochemistry, and electronic health data, this is an unparalleled resource for the study of the genetics of diabetes, heart disease, chronic kidney disease, etc. GoDARTS has been the substrate of >135 original research publications and has an h-index of 70. A comprehensive list of publications using GoDARTS as either the primary or contributory data source is available from www.researcherid.com/rid/K-9448-2016 (researcher identifier K-9448-2016) (45).

GoDARTS is part of global consortia in the study of type 2 diabetes: including Wellcome Trust Case Control Consortium (WTCCC) and Meta-Analyses of Glucose and Insulin-related traits Consortium (MAGIC). Through these consortia, GoDARTS has contributed to key studies including the discovery of risk loci for type 2 diabetes (47–52), glucose homeostasis (53), and glycemic traits (54). GoDARTS is now part of the Accelerated Medicines Partnership Type 2 Diabetes Knowledge Portal, which is an open-source repository for genome-wide data from all major type 2 diabetes studies (55). This resource allows researchers to look up phenotypes (e.g., fasting insulin) and find all genetic association results and, conversely, allows the investigation of all known phenotype associations for a given genetic variant or gene. Data from the Tayside bioresource have also contributed to the trans-ethnicity discovery of the greater genetic burden of β-cell dysfunction in Asian Indians compared with White Europeans as part of the INdia-Scotland Partnership for pRecision mEdicine in Diabetes (INSPIRED) (56).

Due to linkage to prescription encashment (meaning the collection of a prescription from a pharmacy by the patient) data, GoDARTS has also enabled research into the genetics of drug response, i.e., pharmacogenomics. GoDARTS has been central to the genome-wide discovery of variants associated with response to metformin (ATM and GLUT2-encoded by SLC2A2) (57,58). Using GoDARTS data, researchers observed that the heritability of metformin response was 34% (95 CI 1–68) (58,59). Recently, the study led a multicenter GWAS of glycemic response to sulfonylureas identifying variants in SLCO1B1 and GXYLT1 (60–62). Results of this study showed that heritability of response to sulfonylureas was between 26% and 48%. Interestingly, the study demonstrated that a variant in SLCO1B1 was the substrate of an interaction between statin use and response to sulfonylurea therapy, where variant carriers who were also on statin therapy had limited response to sulfonylureas. Such studies have played a key role in laying the groundwork for precision medicine in diabetes care. Additionally, GoDARTS has been used for candidate gene studies of response to metformin (SLC29A4 and SLC29A1) (63) and thiazolidinediones (CYP2C8 and SLCO1B1) (64). Genetic variants associated with statin response (e.g., LPA, HMGCR) and adverse drug reactions to statins have been discovered and validated in GoDARTS (SLCO1B1 and LILRB5) (65–68), and the study is a contributor to the Genomic Investigation of Statin Therapy (GIST) consortium (69). Exome sequencing of GoDARTS samples led to the discovery of common and rare variants in the gene F5 being associated with increased risk of ACE inhibitor– or ARB-induced adverse drug reactions. The recent availability of exome sequencing data allows for the investigation of the role of rare variants in pharmacogenomics and other disease areas.

As outlined, in Tayside we have a comprehensive electronic medical record system that is used for all diabetes health care encounters and is accessible by patients. The available data are anonymized and used for research and linked to a large variety of other clinical and biobank data. However, the potential for these data to inform on and improve clinical care is huge and largely untapped. One such area is in clinical decision support and the use of artificial intelligence methods to develop powerful prediction models. For example, the extensive longitudinal record data for prescribed drugs, BMI, and HbA_1c can be used to develop prediction models for “best drug” based on predicted HbA_1c reduction, weight gain/loss, and side effects, and these predictions can be used by the clinician and patient to inform on clinical care. Given the low cost of genotyping arrays (less than the cost of a chest X-ray), soon genetic data will be embedded in the medical records and support pharmacogenetics-based prescribing in diabetes. This vision is about to become reality in Tayside, where, with £2.8 million of funding from the Scottish government (Chief Scientist Office), precision diabetes care will soon be implemented. We are developing an intelligent Diabetes (iDiabetes) platform that will include enhanced patient phenotyping including the use of genetic risk scores and measurements of N-terminal prohormone brain natriuretic peptide (NT-proBNP), high-sensitivity troponin, and liver fibrosis markers as well as potential drug response prediction tools to support more precise prescribing. We will also be using the existing clinical information intelligently—to flag to clinicians and patients about risks of hypoglycemia, poor adherence to medication, and declining renal function. All precision recommendations and risks will be made available to people with diabetes via an iDiabetes page on the MDMW patient portal.

In this article we have described a 25-year journey, from a diabetes audit and research database in Tayside to a comprehensive clinical system covering all of Scotland and enabling joined up care, for patients (via MDMW) and primary and secondary care wherever someone with diabetes is cared for in Scotland. This has resulted in improved diabetes care and improved diabetes outcomes. This comprehensive clinical tool has also been the foundation for extensive contributions to our understanding of the causes and consequences of diabetes. Yet with the increasing availability of data and the exponential growth of computational performance and data science, it seems likely that we are just at the beginning of the journey.

Acknowledgments. The authors acknowledge Magalie Guignard-Duff and Dr. Louise Donnelly for providing data aggregates. The authors acknowledge the residents of Tayside who have actively participated in diabetes research. The authors are grateful to the general practitioners, the Scottish School of Primary Care for their help in recruiting participants, and the whole team, which includes interviewers, computer and laboratory technicians, clerical workers, research scientists, volunteers, managers, receptionists, and nurses. The authors also acknowledge NHS Tayside, the original data owner.

Funding. GoDARTS is funded and supported by the WTCCC (072960/Z/03/Z, 084726/Z/08/Z, 084727/Z/08/Z, 085475/Z/08/Z, 085475/B/08/Z) and as part of the European Union Innovative Medicines Initiative SUMMIT program. GoDARTS is supported by Tenovus Scotland and Diabetes UK grants. SHARE is NHS Scotland Research infrastructure initiative and is funded by the Chief Scientist Office of the Scottish government. Additional Funding and initiation of the spare blood retention at NHS Tayside was supported by the Wellcome Trust Biomedical Resource Award (099177/Z/12/Z). Additional genome-wide array data were collected with funding from the National Institute for Health Research (INSPIRED [16/136/102]) with use of U.K. aid from the U.K. government to support global health research.

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

Data and Resource Availability. Restrictions apply to data sets. The data sets presented in this article are not readily available as they contain individual-level identifiable information. Data requests can be initiated by contacting the corresponding author.

Prior Presentation. Parts of this study were presented at the 82nd Scientific Sessions of the American Diabetes Association, New Orleans, LA, 3–7 June 2022.