Associations between attainment of incentivized primary care indicators and incident sight‐threatening diabetic retinopathy in England: A population‐based historical cohort study

Abstract Aim To examine the impact of attainment of primary care diabetes clinical indicators on progression to sight‐threatening diabetic retinopathy (STDR) among those with mild non‐proliferative diabetic retinopathy (NPDR). Materials and Methods An historical cohort study of 18,978 adults (43.63% female) diagnosed with type 2 diabetes before 1 April 2010 and mild NPDR before 1 April 2011 was conducted. The data were obtained from the UK Clinical Practice Research Datalink during 2010‐2017, provided by 330 primary care practices in England. Exposures included attainment of the Quality and Outcomes Framework HbA1c (≤59 mmol/mol [≤7.5%]), blood pressure (≤140/80 mmHg) and cholesterol (≤5 mmol/L) indicators in the financial year 2010‐2011, as well as the number of National Diabetes Audit processes completed in 2010‐2011. The outcome was time to incident STDR. Nearest neighbour propensity score matching was undertaken, and univariable and multivariable Cox proportional hazards models were then fitted using the matched samples. Concordance statistics were calculated for each model. Results A total of 1037 (5.5%) STDR diagnoses were observed over a mean follow‐up of 3.6 (SD 2.0) years. HbA1c, blood pressure and cholesterol indicator attainment were associated with lower rates of STDR (adjusted hazard ratios [95% CI] 0.64 [0.55‐0.74; p < .001], 0.83 [0.72‐0.94; p = .005] and 0.80 [0.66‐0.96; p = .015], respectively). Conclusions Our findings provide support for meeting appropriate indicators for the management of type 2 diabetes in primary care to bring a range of benefits, including improved health outcomes—such as a reduction in the risk of STDR—for people with type 2 diabetes.


| INTRODUCTION
Sight-threatening diabetic retinopathy (STDR) is a common cause of preventable visual impairment in people with type 2 diabetes. 1 Although advances in the treatment of STDR have resulted in better outcomes over the last decade, [2][3][4][5] preventing STDR remains the cornerstone for reducing visual morbidity. [6][7][8] Optimal control of the three known modifiable risk factors, hyperglycaemia, hypertension and hyperlipidaemia, is key to preventing the development and progression of diabetic retinopathy (DR) to STDR. [6][7][8][9][10] Clinical trials that have evaluated the role of lipid-lowering therapies in the progression to STDR have shown that fenofibrate alone, or in combination with statins, slows the progression to STDR, but the protective effect of statins alone is less convincing. [9][10][11] National screening programmes have been established in several countries worldwide to enable early identification and prompt treatment of STDR. 12,13 As screening every person with diabetes annually is challenging and not cost-effective, 14 various risk prediction models have been developed to predict the individual risk of STDR in order to plan risk-based screening. 15,16 A strong predictor of STDR is the presence of DR. 15,16 The severity of DR progresses from mild, moderate or severe non-proliferative DR (NPDR) to proliferative DR (PDR). Eyes with any severity of DR can also develop diabetic macular oedema (DMO). STDR is defined as any evidence of severe NPDR, PDR or DMO. Time to STDR differs with the severity level of DR. For example, the rate of progression to PDR in eyes with mild DR is 6.2%, while 54.8% of cases of severe NPDR progress to PDR in 1 year. Recent evidence suggests that the incidence and progression of STDR has reduced by approximately 2-3-fold over the last 3 decades. 1 Although the UK Prospective Diabetes Study (UKPDS) and the The Action to Control Cardiovascular Risk in Diabetes (ACCORD) study were conducted approximately 15 years apart, both trials showed that controlling systemic risk factors after the development of DR could avert or delay the onset of STDR. 6,8,17 However, the impact of control of individual risk factors and their combinations on the development of STDR in people with type 2 diabetes with DR, at the population level, has not been clearly elucidated recently. The Quality and Outcomes Framework (QOF) is a pay for performance scheme that was introduced in general practice in the UK in 2004, with the aim of improving clinical outcomes for a diverse group of conditions. 18 The framework rewards maintenance of a register of adult patients with diabetes and achieving recommended targets for control of HbA1c (≤59 mmol/mol [≤7.5%]), blood pressure (BP; ≤140/80 mmHg) and total cholesterol (≤5 mmol/L). 19 An additional national programme, the National Diabetes Audit (NDA), compares diabetes care in England with standards set by the National Institute of Health and Care Excellence (NICE) clinical guidelines and NICE quality standards. Among other things, it monitors provision of nine care processes recommended for all patients with diabetes annually.
These consist of HbA1c, BP, serum cholesterol, serum creatinine, urine albumin-to-creatinine ratio (ACR), foot examination, retinal screening, body mass index (BMI) and smoking review. 20 Therefore, attainment of these QOF indicators and NDA standards is a proxy measure of optimal control of the risk factors of STDR. This study aims to provide evidence of the impact of control of risk factors on the incidence of STDR among those with mild NPDR in England.

| Study design and data sources
The UK Clinical Practice Research Datalink (CPRD) GOLD database, which contains longitudinal patient data collected in routine general practice and dating back to 1987, was used to obtain this retrospective cohort. Consisting of more than 18 million patients (3 million of whom are currently registered), CPRD GOLD is representative of the UK primary care-registered population. Linked Hospital Episode Statistics (HES) and Office for National Statistics (ONS) mortality data are available for most of the CPRD participants located in England. The database has been used to investigate diabetes care processes and outcomes. 21 23 Individuals with a type 1 diabetes or other specified non-type 2 diabetes diagnosis occurring at any time were excluded. Those prescribed insulin within 6 months of a diagnosis made at an age younger than 35 years, or within 3 months of a diabetes diagnosis at the age of 35 years or older, were also excluded. Cohort exit occurred at the earliest of transfer out of database, last CPRD data upload, death, or 31 December 2017 (end of study period). The code lists used in deriving the cohort are available in Table S1.

| Exposures
The exposures of interest included attainment of the QOF HbA1c

| Outcome variable
The outcome was time to first CPRD or HES record of incident STDR (from 1 April 2011), as defined by the DESP classification. 23 STDR included severe NPDR graded as R2, PDR (active or stable) as R3 with or without retinal photocoagulation (P1) and the presence of any sign of diabetic maculopathy recorded as M1. The grade R2 equated to ETDRS severity level 43-57 and R3 included grades of 61 or higher. 23

| Covariates
The study covariates (measured at baseline before 1 April 2011) com-

| Statistical analysis
Cohort characteristics at baseline were summarized, as well as missing data. Missing IMD patient-level values were imputed using IMD practice-level data. Missing lifestyle and ethnicity information was imputed from the remaining variables by employing the mice package in RStudio 3.5.1 24 using five imputations. The matchit package was applied to conduct nearest neighbour propensity score matching using a 0.2 caliper for each exposure definition. 25 Matched samples for each exposure were used to fit univariate and multivariate Cox proportional hazards models with the corresponding exposure included as another covariate. Concordance statistics were obtained for each of the multivariate models. Sensitivity analyses were conducted to examine the effect of QOF indicator attainment among (a) participants meeting the other two indicators (i.e. those not being considered as the exposure); (b) those not meeting either of the other two indicators; (c) those attaining lower targets such as an Hb1Ac of less than 7.0% or of even less than 6.5%, BP of less than 130/80 mmHg and cholesterol of less than 4 mmol/L; and T A B L E 1 Baseline characteristics of those with a non-sightthreatening diabetic retinopathy diagnosis (N = 18,978)  Figure S1 for a flowchart).
The baseline prevalence of any DR in our cohort was 28.8%, and mild (but not sight-threatening) retinopathy among the total who met all non-reti-    Tables S2, S4, S5, and S7-S11. Figure 1 presents Kaplan-Meier survival curve estimates across exposure definitions.

| Associations between NDA process completion exposures and sight-threatening retinopathy
Summaries of associations between NDA care process categories and incident STDR are shown in tively [Tables S10 and S11; Figure S3]). A third sensitivity analysis using lower targets such as Hb1Ac of less than 7.0% or of even less than 6.5%, BP less than 130/80 mmHg and cholesterol less than 4 mmol/L, showed that the multivariate HRs were not overly sensitive to minor changes in exposure definitions, with all 95% CIs overlapping by exposure between the primary analyses and the corresponding sensitivity analyses. Table S12, which includes HRs and 95% CIs, summarizes these comparisons, with additional details included in Tables S13 and S14 and Figure S4. The results of the fourth sensitivity analysis showed that achieving any two QOF indicators was associated with significantly better outcomes than achieving fewer than two QOF indicators (adjusted HR 0.67; 95% CI 0.57-0.78; p < .0001) (Tables S15 and S16; Figure S5).

| Summary of principal findings
Our study provides contemporary evidence of the incidence of STDR

| Implications for policy and practice
While tight glycaemic control was defined as an HbA1c of less than 47.5 mmol/mol (6.5%) in the clinical trials, our study shows that even achieving an HbA1c of 59 mmol/mol (7.5%) is sufficient to lower the risk of progression to STDR among those with mild NPDR in whom all risk factors are suboptimally controlled. In clinical practice, it is challenging to ensure that a mean of less than 47.5 mmol/mol (6.5%) can be achieved at a population level. However, optimal control of all three risk factors is encouraged at an individual level to reduce the incidence of all complications of diabetes. Moreover, it is important that there is scope to enhance coverage of the achievement of QOF indicators, given their strong association with lower STDR incidence.

| Strengths and limitations
Our study included a large sample, which was reasonably representative of the English population. acknowledged. Not all patients with DR or STDR may be coded on their medical record, leading to some underascertainment of cases.
While we accounted for major confounders in the analysis, there is still a possiblity of residual confounding. While a large list of covariates were included within this study, they still comprise a closed set, and controlling for other covariates may provide further filtering of the exposure estimates. For some of the categories within covariates, there were low numbers of individuals who developed STDR, which resulted in wider confidence intervals, hence more uncertainty regarding the reliability of results for these covariates.
STDR cases were comparatively few, which could have limited the statistical power for the sensitivity analysis. We did not consider the change in exposure attainment over time in the analysis, which may have potentially overestimated or underestimated the true effect estimates.
In conclusion, this study provides new information regarding the benefit of control of modifiable risk factors in reducing the incidence of STDR among those with mild NPDR at the population level. Our findings provide support for using appropriate indicators for the management of type 2 diabetes in primary care. Investing in such support for these management strategies could bring a range of benefits, including improved health outcomes-such as a reduction in the risk of STDR-for people with type 2 diabetes.