Lp-PLA₂ activity is associated with increased risk of diabetic retinopathy: a longitudinal disease progression study

Diabetes Audit and Research in Tayside Scotland (DARTS) is an integrated clinical management system linking all clinical events of individuals with type 2 diabetes in the Tayside region of Scotland and provides complete continuity of care from general physicians, diabetes specialists and population screening services. GoDARTS (Genetics of DARTS) is a sub-cohort study of individuals from DARTS who have provided a sample of blood for genetic studies and have given consent for linking genetic data to study complications of diabetes, the clinical data for which is continually updated from the electronic medical record [27]. Between 2004 and 2007, participants with type 2 diabetes were recruited (together with non-diabetic controls, who were not included in this study). A serum sample was also collected from every participant at recruitment, and a series of characteristics was measured. Recruitment was treated as the baseline for this study. A cohort study in GoDARTS with a 3 year follow-up period was used to evaluate the primary objective: to test the association between Lp-PLA₂ activity levels (measured at baseline) and incident diagnosis of retinopathy, as well as progression of retinopathy grade.

The exposure of interest was Lp-PLA₂ activity level, measured in serum samples taken at baseline. The biobanked serum samples for this population were analysed for Lp-PLA₂ activity using the CAM colorimetric assay (diaDexus, San Francisco, CA, USA) [28]. Note that Lp-PLA₂ mass was neither measured nor analysed. The activity assay was performed at the Immunoassay Biomarker Core Laboratory, University of Dundee. The diaDexus assay was provided directly from the manufacturer to the laboratory. The intra-assay %CV (n = 36) was 5.8%. The inter-assay (assay to assay) %CV (n = 8) was 7.6%. The duplicate intra-assay %CV of those participants measured in duplicate (n = 134) was 3.0% (range 0%–18.6%).

For retinopathy, data were acquired from the Scottish National Diabetic Eye Screening Service. The severity of retinopathy was coded as per the Scottish Diabetes Retinal Grading Scheme (www.​ndrs-wp.​scot.​nhs.​uk/​wp-content/​uploads/​2013/​04/​Grading-Scheme-2007-v1.​1.​pdf), and are denoted as follows: disease-free (DR0), mild (DR1), observable (DR2), referable (DR3) and proliferative (DR4). Retinal screening in Tayside, Scotland has been undertaken since 1990, initially with Polaroid images, progressing to digital images in 2000 [10]. The Tayside diabetic retinopathy screening protocol has been described previously [29] and the GoDARTS cohort has been used previously by Liu et al to examine risk factors for the progression of diabetic retinopathy [10]. This cohort provides a high-resolution longitudinal dataset for the study of retinal disease progression. Data includes stages of retinopathy, recorded separately for both eyes, and dates. The eye with more advanced retinopathy stage was used for analysis. The screening data used in this study are from the years 1990 to 2011. The most recent screening data prior to baseline was used to establish the prevalent retinopathy stage. The primary outcome variables were time to first occurrence in the record of DR1, DR2, DR3 or DR4 (i.e. incident retinopathy). Time to DR2 was assessed as the time from baseline until first visit where grade DR2 was recorded, and similarly for times to DR3 or DR4. However, this study does not examine the association of Lp-PLA₂ with DME, as the screening data does not contain information specific to DME status.

Previous studies of diabetic and non-diabetic participants have demonstrated that Lp-PLA₂ activity levels are strongly associated with incident coronary heart disease and mortality [23]. In GoDARTS, death during the follow-up period was determined from the administrative records of the General Registrar Office, which holds details of deaths throughout Scotland. Time from baseline to death (for the full cohort) was analysed using the Cox proportional hazards model, with age and sex as covariates, and with quartiles of baseline Lp-PLA₂ activity levels as the exposure (see electronic supplementary material [ESM] Table 1 and ESM Fig. 1).

Exploratory analyses suggested that the modelling assumptions were sensitive to the temporal proximity of death and retinopathy events to the time of serum collection and Lp-PLA₂ measurement. Therefore, a follow-up period of 3 years was chosen to reduce the potential for bias, while maintaining follow-up for enough incident and progression events to accrue. The main analyses reported considered only outcomes that occurred within a 3 year follow-up period after baseline for each participant. Participants with no diabetic retinopathy assessment, and who also did not die within this follow-up period, were excluded completely from the main analyses. The effect of these exclusions (which made the main analysis sample size substantially smaller than the full cohort size) was assessed using sensitivity analyses. Since participants with DR4 at baseline could not be analysed for further disease progression, these participants were also excluded from all analyses. Of the full cohort of n = 6731 participants, n = 684 were excluded because they had DR4 at baseline (Table 1), a further n = 4667 were excluded from the main analyses because they had inadequate follow-up (neither death nor retinopathy assessment within 3 years), and finally 16 individuals were excluded from the analyses as they had inadequate baseline covariate information. Hence 1364 participants were included in the study cohort for analyses.

Lp-PLA₂ measurements were used as a linear trait for association testing with baseline covariates and divided into quartiles for cumulative incidence modelling of survival analysis. Means and SDs of Lp-PLA₂ activity are presented. Diabetic retinopathy grades were the outcome event and were treated as a binary (yes or no) variable for survival analyses. Baseline variables were selected for potential inclusion as covariates, based on known associations with diabetic retinopathy. Variables considered as potential covariates in the analyses were an individual’s retinopathy status prior to baseline (estimated from the last visit at or prior to serum collection date), and participant age, sex, prior duration of diabetes, BMI, calculated LDLc, HbA_1c, smoking status, systolic and diastolic blood pressure (SBP and DBP), eGFR, use of statins and use of diabetes-controlling medication (all measured at baseline). The association with baseline covariates was assessed using univariate linear regression. Variables found to be significantly associated in univariate regressions were then tested in survival analyses, and non-significant variables were eliminated from the final model. Survival analyses adjusted for competing risk of death were conducted with Lp-PLA₂ activity levels (in quartiles) using a cumulative incidence model. Results of survival analyses are presented as the omnibus test for the Lp-PLA₂ variable, HR and 95% CI. The results of the main effects (unadjusted) and final (adjusted) models on Lp-PLA₂ hazards are shown in Table 2. Individuals with missing data were excluded on a per-analysis basis.

The cumulative incidence graphs presented are unadjusted, and demonstrate the main effect of Lp-PLA₂ quartiles on the development and progression of retinopathy. All HRs of Lp-PLA₂ use the lowest quartile of Lp-PLA₂ activity as the reference. All statistical analyses were conducted using SAS 9.4 (SAS institute, Cary, NC, USA).

The Tayside Medical Ethics Committee has approved the GoDARTS study and informed consent was obtained for all participants. The participants have consented to research on their samples and data related to diabetes, its treatment and related conditions. They have also consented to the fact that the studies may involve collaborative studies with commercial companies and that the participants will not benefit financially from such collaborations. Neither the University of Dundee research team nor GlaxoSmithKline had access to personally identifiable information. All event level data is provided to the University of Dundee research team in anonymised fashion by the Data Linkage team of the Health Informatics Centre at the University of Dundee.

All diabetic participants in GoDARTS with available serum were assayed for Lp-PLA₂ activity. Measured Lp-PLA₂ activity levels were approximately normally distributed (presented in Fig. 1a) and were therefore analysed as an untransformed, continuous variable for baseline analysis. The mean (SD) Lp-PLA₂ activity level was 121 (35) nmol min⁻¹ ml⁻¹. The observed quartiles were Q1 ≤ 97.2, Q2 ≤ 117.3, Q3 ≤ 140.7 and Q4 ≤ 377.9 nmol min⁻¹ ml⁻¹, which were used to divide the participants into equal sized groups for survival analyses.

Their sex, age, usage of diabetes-controlling medication, usage of statins, smoking status, duration of diabetes, HbA_1c, LDLc levels, eGFR and SBP were all associated with their Lp-PLA₂ activity levels. The distribution of Lp-PLA₂, and the association of sex, BMI, LDLc and HbA_1c with Lp-PLA₂ in the population is presented in Fig. 1. Lp-PLA₂ levels stratified by sex showed significantly different levels; women have lower levels than men (Fig. 1b). The mean (SD) Lp-PLA₂ levels amongst women was 114 (35) nmol min⁻¹ ml⁻¹ compared with 127 (35) nmol min⁻¹ ml⁻¹ for men. The correlation of Lp-PLA₂ activity levels with BMI, LDLc, HbA_1c and eGFR at baseline are also shown in Fig. 1c–f, respectively. The only notable correlation was with LDLc, which showed a strong linear relationship with Lp-PLA₂, with a correlation coefficient of 0.44.

The association of these potentially confounding variables with Lp-PLA₂ is presented for the study population and the full GoDARTS cohort (Table 1). Variables showed similar associations in the study population compared with the full GoDARTS cohort; however, the study cohort were on average 2 years older, had type 2 diabetes for 2 years longer, a higher proportion were medicated for type 2 diabetes, and a lower proportion were disease-free (DR0) at baseline. This suggests that the study cohort had more severe disease at baseline and were therefore more likely to be screened regularly for progression during follow-up.

In agreement with results from previous large meta-analyses including both diabetic and non-diabetic participants [23], Lp-PLA₂ activity levels were strongly associated with death in the diabetic population of the GoDARTS cohort for whom Lp-PLA₂ activity was measured (n = 6731). This is demonstrated in ESM Fig. 1 and ESM Table 1 as an increased risk of death for participants in the highest two quartiles of Lp-PLA₂ levels compared with the lowest. The hazard of death for those in the highest Lp-PLA₂ quartile was approximately one and a half times that for participants in the lowest quartile (HR 1.45, 95% CI 1.24, 1.68; p < 0.001). Hence, when analysing the association between Lp-PLA₂ activity levels and time to progression, censoring at death would violate the non-informative censoring assumption required for standard survival analyses such as Cox regression [30]. Therefore, a competing risk survival analysis [31] was used to analyse the association with incident diabetic retinopathy and progression to more severe stages.

When considering the longer follow-up period of 5 years, the association with Lp-PLA₂ activity with incident diabetic retinopathy was attenuated (HR 1.17, 1.14 and 1.26 for Lp-PLA₂ quartiles 2, 3 and 4 respectively) with only the highest quartile remaining significant (p = 0.03). Further details are provided in ESM Table 3 and ESM Fig. 2. Therefore, it was decided to limit our analysis to a 3 year follow-up period.

Lp-PLA₂ activity levels are shown to be associated with an individual’s risk of developing diabetic retinopathy. Individuals with Lp-PLA₂ activity levels above the median have a very high risk of progressing to more severe stages. Since this relationship appears to be independent of traditional risk factors and, specifically, independent of baseline LDLc levels, these data are consistent with the hypothesis that systemic inhibition of Lp-PLA₂ activity may be a good therapeutic target in the prevention of this complication of diabetes.

This is a population cohort of white Scottish individuals, with many variables captured using routinely collected health record information. The study population is representative of the diabetic population in Scotland undergoing routine screening [10]. The duration of type 2 diabetes is an important risk factor when considering the extrapolation of these results; diabetic retinopathy is more prevalent when duration of type 2 diabetes is longer than 10 years [2]. This is reflected in the duration of type 2 diabetes in the study population, which, with the addition of 3 years of follow-up, is on average 12 years. Furthermore, the rate of adherence to annual, recommended screens for those with pre-existing retinopathy was reported to be as low as 61%, while the rate of screening for those with long duration of type 2 diabetes was 57% [33]. Overall there is evidence to suggest that adherence to annual screens range between 34% and 65% [3, 9, 33‐35]. This suggests that the drop-off rate noted with the 3 year follow-up exclusion criteria is reflective of real-world consumption of healthcare in a population with type 2 diabetes. Further, association characteristics of Lp-PLA₂ activity are similar to those reported elsewhere [21‐23]. Therefore the generalisability of the study to a population of European descent with routine access to healthcare is likely to be high. On the basis of previous observations in animal and human studies it is unlikely that the ethnicity of the cohort would limit the generalisability of this study.