Examining the feasibility of using general practice records to determine excess mortality among people with diabetes

Data sources

We used data extracted from records of people with diabetes (type 1 and type 2) aged 18 years and older enrolled in practices participating in the Midland Diabetes Structured Care Programme. Data were collected as part of routine audit when the programme was first initiated in 1999⁹, and 2016. As patients are enrolled on the programme they provide written informed consent. In 2016 the original cohort were audited together with a systematic sample of the population enrolled at this time, specifically alphabetical order followed by sampling every third person¹⁰. Data were collected on demographics, lifestyle factors, clinical parameters, and complications (Table 1).

Vital status for the cohort was also collected in 2016, and people were classified as survivors or decedents. Name, date of birth, date, and cause of death (if available) and GP practice location (town and county) of decedents were obtained from GP records. National death certificates were accessed by FR from the General Registrar Office (GRO) database. The database was searched initially using decedent name and date of death if available. Following this, date of birth and GP location were used to verify the correct match. Data matching was conducted in 2017. Once the record was identified, cause of death, underlying cause of death and date of death were extracted from the death certificate. Primary cause was the condition directly leading to death, underlying cause of death was coded as the last listed antecedent cause on the death certificate. Cause of death was manually coded according to the International Classification of Diseases 10^th revision (ICD-10). Mortality data from the national population and four counties in the Midlands of Ireland covered by the structured programme (Offaly, Longford, Westmeath and Laois) for corresponding age groups and calendar years 1999–2013 were obtained from the Central Statistics Office (CSO)⁵. Cause of death extracted from GP records were compared with that extracted from national death certificates.

Data from the whole sample (original cohort and systematic sample) was used to examine the reliability of mortality data from GP records (Figure 1). Data from the original cohort was used to determine standardised mortality ratios. Previous work using these data was undertaken as part of a doctoral thesis (Chapter 5)¹². Specifically, a survival analysis was undertaken using the full of cohort of people with type 1 and type 2 diabetes.

Figure 1. Sample used for main analysis and comparison of GP records vs. national certificates.

Analysis

Only people with type 2 diabetes were included the analysis. Current age rather than baseline age (age at cohort entry in 1999) was used for the assignment to 5-year age groups. Lexis expansion¹², or ‘episode splitting’ was performed to split the data by calendar year periods, effectively into a series of ‘snapshots’ changing the dataset from one observation per individual to one observation for each time interval (i.e., calendar years) per individual. That is, in each calendar year, the participant contributed to one of the corresponding age groups. Sex-stratified standardised mortality ratios (SMRs) were used to compare mortality rates among people with diabetes, between 1999 and 2013 (when national data on deaths were available from the CSO), with the national population. SMRs were also calculated using the Midland counties as a comparator. All analysis was carried out in Stata v12 for windows (StataCorp, College Station, TX, USA).

Comparison of GP records with national certificates

Among the full cohort sampled in 2016 (n = 1475) there were 226 decedents. National death certificates were located for 198 individuals. Table 2 shows cause of death recorded in national death certificates. In death certificates, diabetes was recorded as the underlying cause of death for 16% (n=31) of decedents. None of the certificates had diabetes recorded as the first cause. Less than half of decedents (n= 103, 46%) had diabetes (unspecified, type 1, type 2, or maturity onset diabetes of the young) recorded somewhere on their death certificate.

In GP records, cause of death was recorded for less than one-third of decedents (n = 60, 27%). These 60 records listed only one cause of death and none recorded diabetes as a cause of death.

Of the 55 GP records which could be compared to national certificates (5/60 did not have corresponding national certificates), 75% (n=41) had recorded a cause of death which was also present somewhere on the national certificate, either the direct cause of death present on the certificate (n=24, 59%), the only cause listed on the certificate (n=7, 17%) or an underlying cause on the certificate (n=10, 24%).

Mortality rates among original cohort

Among the original cohort of 328 people with type 2 diabetes enrolled at baseline (1999), 165 (50.3%) had died in the period from 1999 to 2016 (Table 3). Of those who died, date of death could be established for 150 (91%) and cause of death for 146 (88.5%) national death certificates. Of the 165 (50.3%) who died over the 16-year period, the crude mortality rate was 38.94 (95% Confidence Interval (CI): 33.14 - 45.74) deaths per 1000 person-years; 36.75 (29.07 – 46.45) among men and 41.13 (32.95 – 51.35) among women.

Rates in the Midland counties (12.00 (11.85 - 12.14) deaths per 1000 person-years) were similar to the national population (11.80 (11.76 - 11.83)). SMRs, which allowed for differences in the age structure of the populations, indicated that mortality among people with type 2 diabetes was greater than in the general population (Overall SMR = 1.17 (95% CI: 1.00-1.38)), significant among women (1.30 (1.04-1.63)) but not men (1.05 (0.82-1.34)).

Summary

This study examined whether the ‘untapped resource’ of routinely collected general practice data from a structured diabetes programme could be used to determine excess mortality among people with type 2 diabetes. GP records currently do not appear to be a reliable source of death data. Cause of death was recorded in less than one-third of records. Linking clinical data from a fixed cohort in general practice to national records to establish excess mortality was feasible, albeit labour-intensive. Only half of national death certificates recorded diabetes as a cause.

Though mortality was greater than the national population, the excess mortality was considerably lower than a number of recent international studies reporting excess mortality among people with type 2 diabetes two to threefold that observed in the national population^6,13–17. One reason for the low excess mortality may be the fact that cardiovascular risk factors were well-managed as part of this programme^9,18,19

Strengths and limitations

A strength of the study is the long duration of follow up of ‘fixed’ cohort, with date of death (91%) and cause of death (89%), established for most of the 165 who died over the 16-year period. However, our data is limited to a single region within Ireland and to general practices participating in a structured care initiative. As such it may not be generalisable to other regions or practices. Though we have set our estimates of excess mortality in the international context, caution should be taken when comparing excess mortality between countries. Differences may reflect the age structures of the study populations and differences in background mortality rates. The age structure of people with diabetes can be different between countries. Limited sample size, few individuals in younger age groups, require larger sample for greater precision estimates. A final limitation is that we cannot account for errors in the recording of individual death certificates.

Comparison with existing literature

Previous studies have shown that practices are not routinely informed following a patient’s death^20,21, and highlight the challenges GPs experience when trying to obtain information on their patients who die²². Importantly our findings suggest this is an issue even within a structured care programme dedicated to monitoring quality of care and outcomes, and which has demonstrated improvements in recording of important clinical indicators over time⁹. We identified discrepancies between GP records and certificates with respect to the cause recorded, in line with previous studies using primary care databases. For example, a study utilising The Health Improvement Network (THIN) found that while death and dates were reliably recorded (correct in 99% of cases), cause of death recording was unreliable²³.

Implications for research and practice

Date and cause of death were available from national records for most people enrolled in the programme. We illustrated that linking clinical data from a fixed cohort in general practice to national records to establish excess mortality is feasible in Ireland, albeit labour-intensive. A data linkage exercise involving death registrations and longitudinal survey data, also highlighted the requirement for largely manual matching and ICD-10 coding²⁴. In contrast, countries like the UK and Sweden the existence of more robust infrastructure and national identifiers has facilitated electronic linkage of general practice data (e.g., UK Clinical Practice Research Datalink) with mortality data (e.g. Hospital Episode Statistics)^25,26. However, electronic linkage still demands time and resources, and does not circumvent an number of continuing challenges, including limitations in the completeness of general practice data, discrepancies between general practice and secondary care data sources, and reduced sample sizes and follow-up²⁵. Since these data were collected, the General Data Protection Regulations (GDPR) have been introduced. While these new regulations present additional challenges in terms of accessing routinely collected data from GP records for research purposes, they do not apply to data concerning deceased persons^27,28.

Only half of national death certificates recorded diabetes as a cause (antecedent, underlying), similar to previous studies of death certification among people with diabetes²⁹, including an Irish study highlighting issues with reporting of co-morbidities³⁰. The latter may prove challenging for certifying physicians given that people with diabetes often have several coexisting micro- or macrovascular conditions³¹. These issues further reinforce how relying solely on national certificates, may underestimate the impact of diabetes. Where the certificates are completed in hospital, the certifying physician may not have been aware that the patient had diabetes. Existing research indicates recording of diabetes may be more likely where the certifying physician is the primary care physician³². GPs have ongoing contact with their patients, knowledge, and potentially more complete records of patient comorbidities and underlying conditions²⁵. Within general practice, death registers have been used to drive research and QI and to understand the healthcare needs of the practice population^20,33,34. However, improving the quality of mortality data in general practice would require changes to how practices are notified about deaths to make this process more systematic. Previous studies have suggested the need for national statistics centres to provide death certification information to practices, facilitated via a GP identifier on the certificate³⁵. Providing incentives or creating specific functionality within the EHR (i.e. dedicated data entry for ICD-10 codes) may also be required to support death recording. Further work is needed to ascertain the reasons for the discrepancies between the cause of death recorded in GP and national certificates.

Other countries, for example, Scotland, have successfully linked diabetes registers to death records³⁶. This approach may also be viable in Ireland given work is underway to establish a national diabetes register^37,38 and there have been recent efforts to make data more accessible for health research; in 2024, the Central Statistics Office (CSO) in Ireland launched the Health Research Data Centre (RDC), providing a safe space through which researchers can access the data collected by the CSO. This includes mortality data linked by ethnicity, age, gender and cause of death. A diabetes register is further justification for supporting improvements in the quality and usability of general practice data as some register data could be derived from general practice³⁹. However, potential sources of bias would be important to consider here, specifically; the quality of the data extraction tool, use of free text (the quality of which depends on resources or guidance on EHR software), the availability of coding systems in primary care software, frequency of measurements as stipulated in clinical guidelines, the presence of incentives for recording, including awareness of the possibility of future sharing².

Overall, our study reinforces the potential need for a national infrastructure to facilitate linkage between data sources, including national death records and general practice data. Currently there is limited capacity to fully examine the relationship between quality of diabetes management and outcomes.