Evaluating Ethnic Variations in the Risk of Infections in People With Prediabetes and Type 2 Diabetes: A Matched Cohort Study

The proportion of adults in England estimated to have type 2 diabetes (T2D) has been increasing over recent decades (1) due to a combination of improvements in life expectancy, rising obesity levels (2), and declines in case-fatality (3). Additionally, a growing proportion of U.K. adults are thought to have nondiabetic hyperglycemia or “prediabetes” (4,5). Due to improved cardiovascular disease risk factor management, cardiovascular disease mortality among people with T2D has declined substantially in recent decades, resulting in a larger proportional increase in the burden of other conditions among people with diabetes (6–8). One of these is infectious diseases, which are common in people with diabetes (9), resulting in significant health service use, especially in primary care (10), and which have substantial negative impacts on quality of life (11).

We previously showed that 15% of people with T2D had a serious infection requiring hospitalization over a 5.5-year follow-up period, a doubling of risk compared with age- and sex-matched patients without diabetes (12). However, we and other studies were unable to assess whether this risk of infection varied by ethnic group. This might arise due to differences in age structure, obesity levels, comorbidities, other risk factors (such as smoking), or potentially, socioeconomic status (13). Although few studies have estimated risk by ethnicity among people with diabetes, a higher risk has been reported in the U.S. Black population (14). Assessing the risk among non-White ethnic groups is important since their prevalence of T2D is markedly higher and the onset is at younger ages (15). Furthermore, little is known about infection risk in the large population of people with prediabetes, where the proportion at younger ages may be higher among non-White ethnic groups (16).

This study, therefore, aims to extend our earlier work using data from the Clinical Practice Research Datalink (CPRD) in two important ways. Firstly, we take advantage of improved ethnicity recording in the data to compare within different ethnic groups how similar the risk of infection is between people with and without T2D. Secondly, we investigate for the first time whether an association with infection is also found in people with prediabetes compared with those without diabetes or prediabetes. Finally, for both of these aims, we additionally provide a picture of the attributable risks of infection in primary care and for hospitalizations, due to prediabetes and T2D, both among those with these conditions and across the wider adult population. In doing so, we have chosen to evaluate risks in a period ending just before the coronavirus disease 2019 (COVID-19) pandemic, before severe disruptions to health service use occurred.

CPRD is a primary care database in the U.K. jointly sponsored by the Medicines and Healthcare Products Regulatory Agency and the National Institute for Health and Care Research (17). It provides a pseudoanonymized longitudinal medical record for all registered patients (>99% of the U.K. population are registered with a general practitioner), with diagnoses and other clinical information recorded using Read codes. The database recently expanded (CPRD Aurum) to include 16 million currently registered patients (17), with >80% having their ethnicity recorded (18). More than 90% of contributing CPRD practices in England have consented to their data being linked to external sources; researchers have no access to geographical identifiers such as residential postcode (19). These data sources include Hospital Episodes Statistics (HES), which records every National Health Service hospital admission in England (20), and the Index of Multiple Deprivation (IMD), a composite small-area (∼1,500 people) measure used in England for allocation of resources (21). Within CPRD, the distribution of IMD is comparable to the national distribution and provides researchers with a good proxy for individual socioeconomic deprivation (22).

This retrospective matched cohort study design included all patients aged 18–90 alive on 1 January 2015 and actively registered for at least 1 year from practices where HES linkage was available. A total of 8,722,348 patients from 1,447 practices in England were eligible (Supplementary Fig. 1). Ethical approval for the study was granted by CPRD’s Research Data Governance (protocol no. 21_000592).

We classified patients with prediabetes or T2D based on information recorded up to 1 January 2015. Diabetes was first identified from Read codes indicating the patient had been previously diagnosed with diabetes (Supplementary Table 1) and then classified into type 1 or 2 using a strategy developed previously (Supplementary Fig. 2) (12). A total of 527,151 T2D patients were selected (6.0% prevalence). Patients with type 1 (n = 33,851) were not included in this analysis, as we did not anticipate identifying differences in infection risk by ethnicity in this group. Patients with prediabetes were identified from the remaining population if they fulfilled any of 1) Read code for “prediabetes” before 2015; 2) HbA_1c ≥42 mmol/mol (or ≥6%) during 2013–2014; or 3) Read code suggesting impaired glucose tolerance during 2013–2014. We excluded any patients with prediabetes (n = 738) if they had received any antidiabetes medication (except biguanides) before 2015. A total of 273,216 patients with prediabetes were selected (3.1% prevalence).

Patients were grouped into five broad ethnicity categories (White, South Asian, Black, mixed/other, and missing) based on recorded Read codes (Supplementary Table 1) (18). In the U.K., ethnicity is predominantly self-reported in primary care records. In our data, we were able to classify ethnicity for ∼90% of patients with prediabetes or T2D.

For each patient with prediabetes or T2D we created two distinct sets of patients without prediabetes or diabetes matched on 1) age, sex, and practice, and 2) age, sex, and ethnicity. For each of the four sets produced, patients were randomly selected without replacement from the set of all suitable matches. Thus, it was possible for a patient without prediabetes or diabetes to be selected in each of the four matched sets. More than 99% of prediabetes/T2D patients were matched (Supplementary Fig. 1), and, overall, at least one match was found for >98% of patients within each ethnic group. All patients were followed up to the earliest date of patient death or deregistration, practice leaving CPRD, or 31 December 2019. We also conducted a sensitivity analyses for patients with prediabetes who received a diagnosis of diabetes during the study by censoring their follow-up time on the day of the first diabetes diagnosis.

We classified and grouped infections broadly along the same lines as our previous study (12). First, we updated an extensive list of Read codes (primary care) and ICD-10 codes (hospital data) for all infection diagnoses (Supplementary Table 1). Second, we searched electronically in the data over a 5-year period (2015–2019) for the following: 1) any infection with a prescription in primary care for an antibiotic, antifungal, or antiviral within ±14 days of the diagnosis; 2) any new hospital episode where an infection was the primary diagnosis. In the U.K., hospital data are organized into finished consultant episodes and assigned a primary diagnosis (20). Subsequent episodes can be assigned to the same admission, with a different primary diagnosis (e.g., a hospital-acquired infection). For each summary group, only one event was counted within a 90-day period, with multiple codes assumed to be the same event. Additionally, we conducted an analysis with each of the individual infection groups, again restricting to one event per group within a 90-day period.

We also extracted patient information on smoking history, BMI, and comorbidities as of 1 January 2015. We selected 12 chronic conditions routinely collected as part of the Quality and Outcomes Framework (QOF), a U.K.-wide system for performance management and payment of general practitioners in primary care (23). These were atrial fibrillation, cancer, chronic obstructive pulmonary disease, coronary heart disease, chronic kidney disease, dementia, epilepsy, heart failure, hypertension, peripheral vascular disease, serious mental illness, and stroke.

Conditional Poisson regression compared infection rates during follow-up between patients with prediabetes or T2D to those without prediabetes/diabetes, with an offset fitted for total days of follow-up time in the study (Stata 15 software). These were conditioned on the match sets, which implicitly controls for the matching factors (age, sex, and practice/ethnicity). These were initially fitted without any further adjustment, but we also fitted models that adjusted for socioeconomic status (IMD quintile, with quintile 1 representing the most deprived 20% small areas in England), smoking, and a count of comorbidities. To assess the impact of ethnicity as a confounder, we compared results from the ethnicity-matched with the non-ethnicity–matched analysis. To explore the impact of age and ethnicity as effect modifiers, we fitted stratified models by age-group (18–50, 51–70, 71–90) and by ethnicity separately as well as together. For these analyses, we present and compare unadjusted models in the main analysis because between–ethnic group differences in key confounders, such as socioeconomic deprivation, will be indirectly controlled as each model only compares within ethnic group. However, we also provide adjusted estimates (by ethnicity) in the Supplementary Material. Sensitivity analyses explored the impact of additionally adjusting for obesity and censoring follow-up time for patients with prediabetes who received a diagnosis of diabetes during the study period. Finally, attributable risks for infection in people with prediabetes and T2D and population-attributable risks were estimated for each ethnic group, assuming prediabetes or T2D is the direct cause of any observed infection risk. These were estimated using models stratified by 10-year age-group (18–29, 30–39, and so on to 80–89) and summed using a weighted average (24). We provide calculations derived from the unadjusted and adjusted models described above.

The study funder had no role in study design, collection, analysis, and interpretation of data; writing of the report or decision to submit the paper for publication. The corresponding author had full access to all the data in the study and final responsibility for the decision to submit for publication.

The data that support the findings of this study are available from CPRD obtained under license from the U.K. Medicines and Healthcare Products Regulatory Agency, but restrictions apply to the availability of these data, which were used under license for the current study and therefore are not publicly available. CPRD data governance and the license to use CPRD data do not allow distribution of patient data directly to other parties. Researchers must apply directly to CPRD for data access (https://www.cprd.com). However, code lists generated during the current study are available in the St George’s Research Data Repository (https://10.24376/rd.sgul.21565557).

Table 1 summarizes the baseline characteristics of patients with prediabetes and T2D by ethnicity. Approximately 70% of patients were classified as White ethnicity, followed by ∼10% as South Asian ethnicity. In the wider population, crude prevalence was highest for South Asian ethnicity (5.5% prediabetes, 11.1% T2D), with this difference more striking at younger ages (Supplementary Fig. 3 and Supplementary Table 2). Patients with prediabetes were more likely to be women for all ethnicities, whereas people with T2D were more likely to be men (except for Black ethnicity). The White ethnicity group had a mean age of 67 years for both prediabetes and T2D. Non-White ethnicities were on average 10–13 years younger for prediabetes and 5–7 years younger for T2D, and consequently, had fewer comorbidities. For both prediabetes and T2D, non-White ethnicities were more likely to live in deprived areas, with 4 in 10 in the Black ethnicity group residing in the most deprived quintile. Among those with prediabetes, Black patients had the highest average recorded BMI, with 48.3% being >30 kg/m², while for those with T2D, White people had the highest BMI levels (54.3% >30 kg/m²).

We also compared baseline characteristic differences between patients with prediabetes or T2D and age-, sex-, and ethnicity-matched patients without prediabetes/diabetes (Supplementary Table 3). Patients with prediabetes and T2D were both more likely to live in deprived areas, be more obese, have a history of smoking, and have comorbidities than patients without prediabetes or diabetes. The relationship with socioeconomic deprivation was further explored within ethnic groups (Supplementary Fig. 4) and showed that the association of greater deprivation with prediabetes or T2D, compared with patients without prediabetes/diabetes, is maintained despite differences in overall deprivation between ethnic groups.

There were significant increases in infection risk for both patients with T2D and prediabetes compared with patients without prediabetes/diabetes (Table 2). Comparisons using the age-, sex-, and ethnicity-matched versus the age-, sex-, and practice-matched comparison group yielded similar results, and all analyses from this point use the age-, sex-, and ethnicity-matched group. For T2D, the (unadjusted) relative risks of infection were higher for hospitalizations (incidence rate ratio [IRR] 1.91, 95% CI 1.90–1.93) than primary care infections (IRR 1.51, 95% CI 1.51–1.52). The relative association with primary care infections was similar by sex, but slightly higher in women for hospitalizations (IRR 2.02, 95% CI 1.99–2.05 vs. IRR 1.83, 95% CI 1.80–1.85). Larger relative associations were found at younger ages for both outcomes (e.g., IRR 2.96 95% CI 2.85–3.08 for hospitalizations and T2D for ages 18–50), where the infection rates among patients without diabetes were comparatively lower. Even after accounting for the greater number of comorbidities in people with T2D (as well as differences in smoking and socioeconomic deprivation), associations with both infection outcomes were still observed among people with T2D. Adjusting for obesity only explained a small proportion of the observed association and was less influential than comorbidity in the models (Supplementary Table 4).

For people with prediabetes, (unadjusted) infection risks were smaller overall and similar for primary care (IRR 1.35, 95% CI 1.34–1.36) and hospitalizations (IRR 1.33, 95% CI 1.31–1.35) and showed similar gradation with age. After adjustment for confounders, the association with hospitalization was reduced further, especially among older ages. Censoring patients with prediabetes diagnosed with diabetes during follow-up had minimal impact on these associations (Supplementary Table 4).

Figure 1 plots crude infection rates (primary care, hospitalizations) in patients with prediabetes and T2D patients by ethnicity. These are compared with matched patients without diabetes/prediabetes. Among patients with T2D, infections in primary care were highest for South Asian people (235.1 per 1,000 per year), while hospitalization infections were highest for the White group (68.0 per 1,000 per year). However, compared with patients without prediabetes/diabetes in the same ethnic group, the relative increase in risk for both infection outcomes was broadly similar across ethnic groups (e.g., IRRs for hospitalization were South Asian 1.98, 95% CI 1.91–2.06; Black 1.87, 95% CI 1.76–2.00; mixed/other 1.98, 95% CI 1.88–2.08; and White 1.88, 95% CI 1.86–1.98). For prediabetes, the pattern of a similar increase in infection risk by ethnic group was observed for infections in primary care but was less consistent for hospitalizations, where Black people with prediabetes had no statistically significant increase in risk compared with Black people without prediabetes/diabetes (IRR 1.07, 95% CI 0.96–1.18).

As people of non-White ethnicities with prediabetes or T2D were on average younger (and healthier) in our data than White people with the same conditions, we stratified the (unadjusted) ethnic specific IRRs by age (Fig. 2) and also adjusted the IRRs for potential confounders (Supplementary Tables 5 and 6). These reveal that for T2D, there tends to be a higher relative risk among the youngest ages (18–50) in the White ethnic group (IRR 2.12, 95% CI 2.07–2.16 for primary care infections; IRR 3.23, 95% CI 3.08–3.40 for hospitalizations) compared with the risks found in non-White ethnicities. Above age 50, the IRRs are generally similar between ethnicities for both prediabetes and T2D, although the Black ethnicity group with prediabetes showed no increased risk with hospitalizations. Adjusting for confounders within each ethnic group attenuated all associations, but not did alter findings made on comparisons between ethnic groups.

Finally, we investigated associations with specific infection types by ethnicity in prediabetes and T2D (Supplementary Tables 7 and 8). Within each ethnic group, associations were consistently observed for every infection group apart from extremely rare ones where power was low. Upper respiratory tract infections were almost twice as common in South Asian T2D patients compared with other ethnic groups with T2D. However, when compared with South Asian patients without diabetes/prediabetes, the relative increase in risk was more similar (IRR 1.43, 95% CI 1.40–1.46), albeit still higher than the corresponding relative risk within the White ethnic group (IRR 1.21, 95% CI 1.20–1.22).

Attributable risk fractions by ethnic group due to prediabetes and T2D were estimated for primary care and hospital infections (Supplementary Table 9) and then weighted accordingly to create population-wide estimates. The attributable fractions by ethnic group among patients with prediabetes or T2D derived from the IRRs in Fig. 1 were 32–36% for primary care infections and 49–56% for hospitalization infections for T2D, while for prediabetes they were similar for primary care infections (24–32%) but lower for hospitalization infections, especially among Black people with prediabetes (7%). The percentage of infections among all adults in the population attributable to prediabetes or T2D was 5.8% for primary care and 10.0% for hospitalizations. When estimated by ethnic group, these were highest among South Asian people (11.6% and 17.4%, respectively).

Our study has two key findings. Firstly, we have shown that the relative risks for infection associated with T2D appear to be broadly similar within each ethnic group, and this remains true at different ages and after adjustment for potential confounders. As in our previous work (12), compared with people without prediabetes/diabetes of the same age, sex, and now additionally ethnicity, the risk of hospitalization for infection was roughly doubled among people with T2D, and ∼50% higher for infections requiring primary care contact and an associated prescription. Secondly, we have shown that increased risks for infection are also present in people with prediabetes, albeit lower, at approximately a 30% increase for both infection outcomes when compared with people without diabetes/prediabetes.

Overall, a major strength of our study is the extremely large sample size (8 million total adults, 750,000 with prediabetes and T2D). The large number of patients with prediabetes is a likely result of the increased emphasis on both vascular health checks and specifically, diabetes prevention and screening in primary care (25), although it still may be underestimating the true scale of prediabetes in the general population, as this continues to rise globally (26). However, our sample is still likely highly representative of people living with prediabetes or T2D in England before the COVID-19 pandemic (17). Previous CPRD analyses (12,27,28) have been based on an earlier data set (CPRD GOLD), which underrepresented major urban areas with ethnic diversity. By using the newer Aurum database, with higher overall recording of patient ethnicity, we were able to successfully match patients by broad ethnic group unlike previous studies (27), resulting in a stronger design for assessing the pattern of between ethnic differences in risk. In our study, matching was highly successful overall and in each ethnic group: >98% of people living with prediabetes or T2D had at least one patient without diabetes matched on age and sex. Additionally, we selected patients without diabetes or prediabetes matched on practice rather than ethnicity to establish the impact of ethnicity as a confounder in our earlier findings. Although ∼10% of patients with T2D or prediabetes could not be assigned an ethnicity, we retained these patients in the analysis to assess any potential bias. These patients tended to be older and more affluent, largely resembling the characteristics of the White ethnicity group, suggesting that we captured a high proportion of non-White ethnicities with T2D or prediabetes.

Importantly, our analysis was able to consider the impact of age as an effect modifier and show that relative risks were higher in the youngest age-group, where the baseline risks of infection in the population without prediabetes or diabetes tends to be much lower. For example, people <50 with T2D were three times more likely to be hospitalized with an infection than people without prediabetes or diabetes. However, the increase in infection risk with age in the population reference group is such that the absolute risk differences are still greater at older ages despite the lower relative risk. Since non-White ethnic groups in the U.K. are younger on average, we were able to investigate how age was modifying the relationships with the overall ethnic-specific associations. This revealed that younger people with T2D of White ethnicity were at greater risk (both relative and absolute differences), while people with prediabetes from Black and Afro-Caribbean ethnic minorities had little or no excess risk of hospitalization for infection. We further considered the impact of factors more common in people with prediabetes or T2D (socioeconomic deprivation, smoking, comorbidity), even though some of their comorbidity, such as heart disease, may be a result of their diabetes. Although this may represent an overadjustment, our conclusions regarding ethnicity and infection risk were not altered.

Clearly, primary care recording of infection outcomes is pragmatic and imperfect, based on clinical diagnoses. However, the magnitude of infection risk was stronger for infections resulting in a hospitalization, where infection recording is more often supported by laboratory findings. Identification of both diabetes and prediabetes is dependent on general practice consultations; prediabetes, in particular, may be significantly underdiagnosed in primary care (4), resulting in misclassification of exposure and potentially underestimating the association with infection risk. However, the pattern of infection risk identified for this patient group appears similar, albeit lower, than that for people with T2D. Our study is based on retrospective cohort data from a period ending just before the COVID-19 pandemic. Incidence and recording of infections, especially in primary care, has likely been profoundly affected by the disruption to normal health service delivery during the pandemic, so we believe it is more informative to assess non–COVID-19 infection risk prior to this date. Future studies could focus on how this has been affected since 2020.

Ethnic differences in infection risk have been a topic of considerable recent interest since persistent mortality differences by ethnicity were noted from the early stages of the COVID-19 pandemic (29). Subsequently, a higher risk of testing positive for severe acute respiratory syndrome coronavirus 2 among non-White ethnic groups was also confirmed (30). To the best of our knowledge, few studies have considered whether infections in general might differ by ethnicity in people with diabetes. The higher prevalence of tuberculosis infection reported among Hispanic and Asian people in the U.S. has been hypothesized to explain the association between diabetes and tuberculosis (31). While this might reflect poor health care access, socioeconomic deprivation, or potentially, publication bias (since most studies only report on individual infections), a recent U.S. cohort study reported higher risks of infection-related hospitalization associated with diabetes in younger people and in people of Black ethnicity (14). However, the study size (∼1,500 people with diabetes) meant it could not stratify by ethnicity or consider whether differences in age among the Black population might explain ethnic variation in infection risk.

In our study, we were able to report on differences in infection rates by ethnic group at different ages, which suggested that non-White ethnicities were less likely to be hospitalized for an infection. In the U.K., overall hospitalization rates have been shown to vary by ethnicity across different conditions (32), likely due to pervasive inequalities and differences in health-seeking behaviors (33). For example, people from Black African and Caribbean ethnicities have higher reported hospitalizations for diabetes and endocrine disorders relative to the White population, but their overall hospitalization rates are significantly lower, possibly reflecting barriers to access within the health care system (32). By contrast, some specific infection groups we reported on were more common in non-White ethnicities, such as upper respiratory infections in South Asian people. These may reflect generally higher household size found for Bangladeshi and Pakistani people in the U.K. (34), representing a greater risk of transmission among household members. However, the higher or lower infection rates by ethnicity were true irrespective of diabetes, so estimated relative risks of infection for prediabetes or T2D versus people without diabetes tended to be similar within each ethnic group (with the possible exception of a greater infection risk among younger White people with T2D). We did not assess whether diabetes clinical management or severity differed by ethnicity, although a previous study found improved risk factor recording and faster time to starting antidiabetes medications among ethnic minority groups (35).

Management of infections in people living with diabetes is clearly a common problem for patients and doctors but had not always received much attention before the pandemic. While National Institute for Health and Care Excellence guidelines for people with T2D provide little mention of infections (36), we have shown that attributable risks in those living with T2D are substantial (34% for primary care, 50% for hospitalizations) and consistent by ethnic group. Although some of this attributable risk may be explained by their poorer overall health, it does not change the fact that these patients experience a large burden of infections by having T2D. Our analyses also demonstrated that higher risks were also present, although somewhat reduced, in people with prediabetes in both health care settings. This suggests that any infection risk does not begin with a diagnosis of diabetes but is instead on a continuum and already present in this large group of patients. Given the high and rising prevalence of prediabetes and T2D (3,4,26), this is of considerable importance and results in a substantial population burden affecting people with diabetes and prediabetes and health services. We estimate that ∼6% of all adult infections treated in primary care and 10% of hospitalizations due to infection are statistically attributable to prediabetes or T2D and that these will continue to increase with rising prevalence (2). The South Asian population experiences the greatest estimated burden (12% and 17%, respectively) as a result of the higher prevalence of prediabetes and T2D at all ages.

There has been little previous development of interventions to reduce infection risks in people with diabetes; infection outcomes have not been included in most major diabetes management trials (11). Therefore, robust observational data may instead provide an initial path to developing interventions, potentially based on increased self-management of risk factors and awareness by promptly identifying infections and knowing when to seek professional help, which could lead to reduced hospital admissions for some infections. Enhanced glucose monitoring and offering better diabetes control among the highest-risk individuals might improve infection outcomes, but this has not been clearly demonstrated and thus remains a gap in our knowledge. As we have previously shown that poor glycemic control is associated with the risk of serious infections (37), future work could establish whether any association with HbA_1c level is also present in people with prediabetes.

In conclusion, we have used large routine health databases in England to reveal that the increased infection risk associated with T2D is generally consistent across different ethnic groups and also observed in people with prediabetes. The burden of infections attributable to T2D and prediabetes is significant and will continue to have public health implications with their rising prevalence. Efforts to reduce infection risk in people with diabetes remains an important challenge in all ethnic groups.

Acknowledgments. The authors thank their helpful Patient Study Advisory Group for their input, particularly to discussion and interpretation of the results.

Funding. This study is funded by the National Institute for Health and Care Research (NIHR) Research for Patient Benefit Programme (NIHR202213) and supported by the NIHR Applied Research Collaboration South London at King’s College Hospital NHS Foundation Trust.

The views expressed are those of the authors and not necessarily those of the NIHR or the Department of Health and Social Care.

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

Author Contributions. I.M.C. led the data curation and statistical analysis. I.M.C. and J.A.C. wrote the first draft of the manuscript with input from all coauthors. I.M.C., J.A.C., U.A.R.C., S.W., D.G.C., and T.H. were involved in the conception and design of the study. J.A.C. led on funding acquisition. I.M.C, J.A.C., U.A.R.C., S.W., E.S.L., D.G.C., P.H.W., and T.H. were involved in the interpretation of data for the work. I.M.C, J.A.C., U.A.R.C., S.W., E.S.L., D.G.C., P.H.W., and T.H. approved the final version for publication. I.M.C. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.