Relation Between Different Measures of Glycemic Exposure and Microvascular and Macrovascular Complications in Patients with Type 2 Diabetes Mellitus: An Observational Cohort Study

For this observational cohort study, data were obtained from the PHARMO Database Network. This population-based network of Dutch electronic healthcare databases combines rich, patient-centric data from different primary and secondary healthcare settings. Data from the General Practitioner (GP) Database were used, which cover approximately 2.5 million patients from 2005 to 2014. The GP Database comprises data from electronic patient records registered by GPs. The health records include information on diagnoses and symptoms, laboratory test results, referrals to specialists and healthcare product/drug prescriptions. The prescription records include information on the type of product, prescription date, strength, dosage regimen, quantity and route of administration. Drug prescriptions are coded according to the WHO Anatomical Therapeutic Chemical (ATC) Classification System. Diagnoses and symptoms are coded according to the International Classification of Primary Care (ICPC), which can be mapped to ICD codes, but can also be entered as free text. The study period was defined as 1 January 2006–31 December 2014.

As this was an observational study using retrospective data, patients were not involved in the study design or conduct.

To investigate the relation between multiple measures of glycemic exposure and macro- and microvascular complications, it was important to have regularly recorded HbA1c measurements. Diabetes care in The Netherlands is mainly managed in the primary care setting and guided by the Dutch GP guideline for T2DM [14]. It is well organized, including regular checkups. For our study, only general practices adhering to these regular checkups were selected, which was defined as having at least 50% of its T2DM patients with at least three HbA1c measurements per year. This selection was done at the practice level instead of patient level to avoid biased selection as well-managed patients with T2DM may be monitored less frequently.

Within the selected practices, T2DM patients were identified as having either a diagnosis code (ICPC code T90.02) or at least two consecutive prescriptions of oral antihyperglycemic agents (AHA; ATC code A10B) within a 6-month period at any time in the available medication records. The index date was defined as the first oral AHA prescription after at least 1 year of enrollment. Patients without any oral AHA prescriptions during the study period were excluded as well as patients with diagnoses for T1DM, gestational diabetes or polycystic ovary syndrome (PCOS). Only patients between 40 and 84 years of age at the index date were included. The most recent HbA1c measurement within 1 year prior to or at the index date was defined as the index HbA1c measurement. If an index HbA1c measurement was not available, the patient was excluded. Further exclusions were based on the event of interest. Patients with a prior event were excluded for the analysis of that specific event but were possibly included in other analyses. Finally, patients were required to have at least one HbA1c measurement during follow-up. Again, this could differ per event of interest, since the end of follow-up was determined as either the first occurrence of an event, death, loss to follow-up in the PHARMO Database Network or end of the study period, whichever occurred first.

The following characteristics were determined: age at index date, gender, HbA1c at index date, the proportion of prior complications at index date, newly treated, follow-up, number of HbA1c measurements during follow-up and BMI at the index date. Newly treated patients were defined as those without any AHA prescription in the year prior to the index date.

Diabetic foot, retinopathy and renal complications were the microvascular events of interest, and coronary artery disease (CAD) and cerebrovascular disease were the macrovascular events of interest. All complications were based on GP-recorded information.

A Cox proportional hazards regression analysis was performed for each of the micro- and macrovascular complications with each of the four glycemic exposure measures. The PROC PHREG procedure from SAS 9.4 was used.

For the model, using HbA1c at index, the Cox proportional hazards regression model was adjusted for age, gender and the year of index. For illustration, the HbA1c value at index was categorized as <53 mmol/mol (7%), 53 to <64 mmol/mol (7% to <8%) and ≥64 mmol/mol (8%). For the other models, repeated measurements of HbA1c were included, and a time-dependent Cox proportional hazards regression analysis was performed. Age and the measure of exposure (HbA1c, EMA or GBY) were used as time-dependent variables, and the model was corrected for age, gender, the year of index and time (in years) since the first measurement above 53 mmol/mol (7%) and the time (in years) since the last measurement below 53 mmol/mol (7%). The time-dependent HbA1c and the EMA were categorized as follows: <53 mmol/mol (7%), 53 to <64 mmol/mol (7% to <8%) and ≥64 mmol/mol (8%). The categorization of glycemic burden was based on its distribution: 0 (reference), >0 to ≤1, >1 to ≤3 and >3 GBYs.

This article does not contain any new studies with human or animal subjects performed by any of the authors. No ethics committee approval or informed consent was required for this anonymized database study.

The flow chart describing the process of patient selection is shown in Fig. 2. After applying all inclusion and exclusion criteria, approximately 30,000 patients were included per outcome. The characteristics of the total population, without exclusions of patients with prior events, are shown in Table 1. Median (IQR) age was 65 (57–73) years, and approximately half of the patients were female (47%). At index date, 50% of the patients were on target [HbA1c < 53 mmol/mol (7%)], and only 18% of the patients had an HbA1c at index ≥ 64 mmol/mol (8%).

In total, 20 analyses were performed, and the results are presented in Fig. 3 and summarized in Supplemental Table S1. The results of the different events are plotted separately, with the hazard ratios (HRs) on the vertical axis. Within each plot, the four different measures of glycemic exposure are displayed with the HR per categorized variable.

The results for the first model, using only the HbA1c at index, suggested an increased risk for higher HbA1c values for the microvascular complications as well as CAD. However, the results were only significant for the highest category [HbA1c ≥64 mmol/mol (8%)] for renal complications [HR (95% CI) of 1.10 (1.02–1.18)] and CAD [HR (95% CI) of 1.18 (1.04–1.34)] compared to the reference category [HbA1c <53 mmol/mol (7%)].

The results for the time-dependent HbA1c model showed an increased risk for both diabetic foot and retinopathy. For diabetic foot, the HRs (95% CI) for the middle and highest HbA1c categories were 1.09 (1.00–1.19) and 1.36 (1.21–1.52), respectively. For retinopathy, this was 1.19 (1.10–1.30) and 1.55 (1.40–1.73), respectively. For renal complications, only the highest HbA1c category showed a significant association [HR (95% CI): 1.20 (1.10–1.31)]. The associations between time-dependent HbA1c and macrovascular complications were not significant.

The results for the third model, using the EMA, were very similar to the results of the previous model with a few notable exceptions. Specifically, the HRs for retinopathy were markedly higher for both the middle and highest HbA1c categories [i.e., HRs (95% CI) of 1.29 (1.18–1.41) and 1.81 (1.63–2.02), respectively]. Also, the results for CAD were significant for the highest category [HbA1c ≥64 mmol/mol (8%); HR (95% CI) of 1.29 (1.10–1.50)].

The results for the glycemic burden model showed significantly increased risks associated with higher GBY levels for the microvascular complications. For diabetic foot, the HRs were 1.30, 1.46 and 2.15 for >0 to ≤1, >1 to ≤3 and >3 GBYs, respectively, compared to the reference category of no glycemic burden. For retinopathy, the HRs ranged from 1.41 to 2.60, and for renal complications, the HRs ranged from 1.20 to 1.61. The macrovascular complications showed similar results. However, for coronary artery disease GBY >1 to ≤3, the HR was not significantly higher. Cerebrovascular disease on the other hand did show a dose response relation where the HRs ranged from 1.20 to 1.52.

Additional results of the models, including the covariates, are included in Online Resource 1 (Supplemental Table S2).

All micro- and macrovascular complications were weakly associated with exposure to high HbA1c [HbA1c ≥64 mmol/mol (8%)] at index to some extent, but this was only significant for renal complications [HR (95% CI) of 1.10 (1.02–1.18)] and CAD [HR (95% CI) of 1.18 (1.04–1.34)]. When modeling glycemic exposure as the time-dependent observed HbA1c, a significantly increased risk for all microvascular complications was observed for HbA1c ≥64 mmol/mol (8%). For the macrovascular events, this model was inconclusive, and the results were not significant. These results are largely in line with other findings. For renal complications, Gerstein et al. [17] found an association between an increase in HbA1c at index and nephropathy, and Klein [3] found an association between an increase in HbA1c at index and both gross proteinuria and renal failure. For diabetic foot, an increased risk was expected for the model using the HbA1c value at index in concordance with earlier findings [3, 18]; however, these studies found associations with either amputation [3] or sensory neuropathy [18], while we defined a composite endpoint covering a broad spectrum of peripheral arterial problems. However, our findings do suggest that the most recent HbA1c measurement (i.e., time-dependent HbA1c) is a better predictor for diabetic foot than the HbA1c at index. For CAD, the results regarding HbA1c at index are in concordance with the literature [10] where an HR (95% CI) of 1.05 (1.00–1.10) was reported for a 1% increase in HbA1c also using the HbA1c at index. On the other hand, no significant effect was found for the time-dependent HbA1c, suggesting that the HbA1c at index is a better predictor for CAD than the most recent HbA1c measurement. The results for cerebrovascular disease were inconclusive, which is in line with other findings [10] where an HR (95% CI) of 0.96 (0.88–1.06) for a 1% increase of HbA1c at index was reported. The EMA-based results for the microvascular events showed similar results as the time-dependent HbA1c. Retinopathy specifically showed a clear dose response for EMA, and associations were stronger than in the time-dependent HbA1c model. For the macrovascular event of coronary artery disease, a significantly increased risk was observed in the highest risk category [EMA ≥64 mmol/mol (8%)]. Again, the results for cerebrovascular disease were inconclusive. An association between the increase in updated mean HbA1c and the risk has been reported for diabetic foot [1], retinopathy [8, 19], renal failure [17], CAD [1, 2] and cerebrovascular disease [1]. With the exception of cerebrovascular disease, our results reflected previously reported findings. Absence of an association between EMA and the risk of cerebrovascular disease may be explained by the fact that we defined cerebrovascular disease as a composite endpoint including stroke, TIA, CVA and cerebral infarction, while Stratton et al. [1] reported an HR (95% CI) of 1.14 (1.01–1.27) for a 1% increase in updated mean HbA1c for stroke only. For the microvascular events, the analyses using glycemic burden showed the clearest dose response. Overall, the HRs were higher, but the categories used were different from the previous models and not directly comparable as the GBY categories of this cumulative outcome were based on the distribution of the number of patients rather than clinically recognizable cutoff points for HbA1c. Again, the effects for retinopathy were stronger than for the other microvascular events: the HRs ranged from 1.41 to 2.60 against 1.30–2.15 for diabetic foot and 1.20–1.61 for renal failure. Maple-Brown et al. [13] used the AUC as a measure for chronic glycemic exposure and found that it was a good predictor for microvascular complications. For the macrovascular events, CAD (GBY >0 to ≤1 and GBY >3) showed a significant association. Cerebrovascular disease showed a significant, albeit small, dose response with HRs ranging from 1.20 to 1.51.

A strength of the study is the use of a large population in a real-world setting with long follow-up. Data from GPs in The Netherlands are especially suited, since diabetes care in The Netherlands is highly organized, and the majority (90%–95%) of T2DM patients are treated within the primary care setting. This means that we were able to validate our models in actual clinical practice.

A limitation is that the duration of diabetes was not included in the model because it is not always explicitly recorded as a clinical parameter. We also did not differentiate between patients who were and were not newly treated. Both have resulted in underestimation of the glycemic burden and are likely to have resulted in an underestimation of the associations. However, available data reflect the situation in the daily practice of any physician, who will not always know the glycemic history of a patient and will act on the data available at the time of a visit. Furthermore, we performed post hoc analyses, correcting for newly initiated or prevalent use of AHA at the time of the index date, to test whether underestimation of the glycemic burden of patients that were already treated at the start of their recorded history might have caused underestimation of the associations with complications. The results were very similar to the main analysis and therefore could not support this assumption. This means that even with incomplete records, all available history of a patient may provide valuable information to a physician regarding the build-up of glycemic memory. One of the limitations of real-world evidence in general is that information is not collected or recorded for the purpose of doing research. Changes in disease management have had a major impact on the way complications were recorded during the study period, leading to more detailed outcome recording over time. To allow analysis of more or less similar outcomes over time, composite endpoints had to be used. For example, in the case of diabetic foot, this resulted in a mix of micro- and macrovascular causes. This loss of specificity may have weakened the association of specific contributing factors within the composite. Future studies may be able to clarify specific associations by using more specific outcome definitions. In 2006, an integrated diabetes care program was introduced, with structured management and regular checkups of the patient by an interdisciplinary team of GPs, diabetes nurses, podiatrists and ophthalmologists. Check-ups include assessment of HbA1c, renal function, retinopathy, (early) signs of diabetic foot and history taking with regard to macrovascular complications. The gradual rollout of this integrated care was boosted by financial incentives linked to developing performance indicators offered by healthcare insurance companies to GPs from mid-2007 onward. This is reflected in the data by a steady increase in the recorded number of complications, which was about fivefold over the whole study period. As 40% of our population had an index date before 2008, this means that a significant number of patients may have been misclassified as having no complications at the index date and that some complications now registered as newly developed may have been pre-existing. Incidence rates may therefore have been overestimated. As this possible misclassification was most likely irrespective of HbA1c levels, we do not believe that this non-differential misclassification has influenced our results to a large extent. Finally, we have not taken treatment into account, while the type of treatment may influence outcomes through other mechanisms than glucose lowering. For future research, it would be very valuable to take the type of treatment and other potential risk factors into account.

The first two models used a single point measurement as a predictor for outcomes, where this single point can either be static (i.e., HbA1c at index) or time updating (i.e., time-dependent HbA1c). The static model is highly practical for database studies, while the time-updating model better reflects the interpretation a physician might apply in daily practice (i.e., use of the last known value for risk assessment). However, it should be noted that the results of the current study show that neither of these models is necessarily a good representation of the risk to which an individual is exposed. The EMA, on the other hand, shows a stronger relation with diabetic foot, retinopathy and CAD, if not with the other outcomes, which does suggest that glycemic memory plays a role in these outcomes. This is further corroborated by the results of the glycemic burden model where these associations are more pronounced and even become apparent for the macrovascular complications. Given that the glycemic burden model gives a better discrimination between patients that do or do not develop complications, it is a very promising method for further research into the long term effects of hyperglycemia. This could contribute valuable evidence to influence clinical decision making. All in all, the results from this study show that longer exposure to HbA1c levels above target is associated with increased risk of both micro- and macrovascular complications. Furthermore, the results suggest that glycemic memory plays a role in the development of diabetic complications and that treatment inertia increases the risk of these outcomes in broad populations of patients with T2DM who are not on target [20]. From the individual patient perspective, however, it is important to note that intensive glucose-lowering treatment aiming at a low target HbA1c may lead to increased risk of hypoglycemia or adverse drug reactions and therefore is not recommended for patients of older age or with short life expectancy, important comorbidities or poor self-care capabilities [6]. This is why the most recent revision of the diabetes guidelines for T2DM [21] has included higher targets for patients over 70 years of age who cannot be managed with metformin alone. Our finding that glycemic memory plays an important role in the development of complications emphasizes the importance of physicians keeping in mind that lower targets are probably more beneficial for a patient, provided that they are attainable without the trade-off of increased immediate risk. This underlines the need for frequent and active monitoring of the diabetes patient.