Association Between Glycemic Control and Adverse Outcomes in People With Diabetes Mellitus and Chronic Kidney Disease:  A Population-Based Cohort Study FREE

Diabetes mellitus (DM) and chronic kidney disease (CKD) are potent independent risk factors for cardiovascular (CV) events and progression to end-stage renal disease (ESRD).^1- 2 Patients with both conditions are therefore at exceedingly high risk of adverse events, and diabetic nephropathy is the most common cause of ESRD in North America, accounting for approximately 40% of patients undergoing incident dialysis.^3- 4 Given projected increases in the prevalence of DM in developing countries, the global burden of diabetic kidney disease is expected to increase dramatically in the coming decades.⁵

Targeting hemoglobin A_1c (HbA_1c) values lower than 7% slows progression of diabetic kidney disease, including both the onset of microalbuminuria and progression to overt nephropathy.^6- 7 (To convert HbA_1c to a proportion of total hemoglobin, multiply by 0.01.) However, the link between intensive glycemic control and CV events or all-cause mortality is more complex and still debated.⁸ For example, one recent trial demonstrated that targeting HbA_1c levels lower than 6% increased mortality in higher-risk patients with type 2 DM.⁹

Despite the importance of diabetic kidney disease, the impact of glycemic control on outcomes in patients with DM and CKD is unknown because most trials of glycemic control have excluded those with reduced glomerular filtration rate (GFR). Indeed, while current National Kidney Foundation's Kidney Disease Outcomes Quality Initiative guidelines^10(pS62) suggest a target HbA_1c level of 7% “for all diabetic patients with or without chronic kidney disease,” very little evidence supports this recommendation.

We designed this study to determine whether HbA_1c level is independently associated with important clinical outcomes, such as all-cause mortality, CV events, hospitalizations, and kidney failure, in people with DM and stage 3 to 4 CKD. We hypothesized that an increased HbA_1c level would be associated with an increased risk of all adverse outcomes.

Data from the Alberta Kidney Disease Network¹¹ (Canada) and the provincial health ministry (Alberta Health and Wellness) were used for this study. From all outpatients older than 18 years who had their serum creatinine level measured in Alberta at least once between January 1, 2005, and December 31, 2006, we selected those with an eGFR of 15.0 to 59.9 mL/min/1.73 m² and DM. We estimated GFR using the Modification of Diet in Renal Disease (MDRD) study equation because it is the most widely used formula and is recommended by current guidelines.¹⁰ We identified DM using validated algorithms (2 physician billing claims in a 2-year period or 1 hospital discharge ever with a diagnosis of DM, excluding gestational DM).¹² Of 32 555 people with DM and an eGFR of 15.0 to 59.9 mL/min/1.73 m², we excluded 436 people (1%) with ESRD receiving hemodialysis or transplantation before their first creatinine measurement in 2005 or 2006 (their index date), those without HbA_1c measurements during the 6-month period after the first eGFR index date (8041 [25%]), and those of First Nations origin (782 [2%]); because we did not have complete data for this population. Patients were classified into 3 groups based on their first HbA_1c measurement during the study period: HbA_1c level lower than 7%; HbA_1c level of 7% to 9%; and HbA_1c level greater than 9%. Comorbidity was assessed by using physician claims and hospitalization data together with validated algorithms¹³ for the variables listed in Table 1. The median household income for each postal code was obtained using data from the 2006 Canadian census.¹⁴

The primary outcome for this study was all-cause mortality. All-cause mortality, dates of hospitalization, first hospitalization for CV events (myocardial infarction, stroke, coronary revascularization, heart failure requiring hospitalization), and the date of first renal replacement therapy for people who developed ESRD were determined by linkage to the provincial health ministry and the provincial renal databases. Coronary revascularization (percutaneous coronary intervention or coronary artery bypass graft surgery) was identified by procedure codes from records of hospitalizations; myocardial infarction, stroke, and heart failure were based on most responsible diagnosis for hospitalizations using validated algorithms.^15- 17 Progression of kidney disease was based on a sustained doubling in serum creatinine value¹⁸ and defined only for participants with at least 2 serum creatinine values; participants whose serum creatinine value doubled from baseline but then declined to less than twice the original value on a subsequent measurement were not considered to have experienced disease progression.

Baseline characteristics stratified by HbA_1c level were given as means or proportions. Poisson regression models (including the number of days at risk as offset) were used for hospitalization rate analyses. Cox proportional hazards models were used for time-to-event analyses, stratified by health service regions. The proportional hazards assumption was tested using log-negative-log survival plots. All models to assess association between HbA_1c levels and outcomes were adjusted for the following potential confounders: age, sex, index eGFR, individual health insurance premium level (a marker of individual-level income), median neighborhood income, comorbidity, and residence location. Because services available to Albertans with CKD may differ by residence location, we did Poisson regression using robust estimates of variance, defining the cluster (grouping) variable by the health region. Because remote residence location was associated with adverse outcomes in patients with CKD, and to further account for any effect of residence location on outcomes, we adjusted both Cox and Poisson models for distance between each patient's residence and the closest nephrologist as previously described.^19- 20 We used a restricted cubic spline function with 3 knots to allow for a nonlinear association between the clinical outcomes and HbA_1c level. Statistical significance was set at P = .05, and all statistical tests were 2 sided. Censoring occurred with death, first event of interest in analyses of nonfatal events, disenrollment from the health plan, or end-of-study date (March 31, 2009).

We also performed several sensitivity analyses. First, we explored the potential effect of misclassification of HbA_1c level by using the mean value of all measurements made during the 6-month exposure period to reclassify HbA_1c categories. Second, we considered HbA_1c level as a continuous rather than a categorical variable. Third, we considered baseline eGFR as a continuous variable in a similar manner. Fourth, we considered HbA_1c level as a time-dependent covariate using all available HbA_1c values prior to the date of each adverse outcome.

The institutional review boards for the Universities of Alberta and Calgary approved the study. Analyses were performed using SAS (version 9.2; SAS Institute Inc, Cary, North Carolina) and Stata SE (version 10.1; StataCorp LP, College Station, Texas) software.

Characteristics of the 23 296 participants by HbA_1c level are shown in Table 1; a comparison between included and excluded participants is shown in eTable 1. Overall, 21 155 participants had stage 3 CKD (eGFR, 30.0-59.9 mL/min/1.73 m²), and 2141 had stage 4 CKD (eGFR, 15.0-29.9 mL/min/1.73 m²). As shown in Table 1, people with a higher HbA_1c level were younger, more likely to be male, and had lower socioeconomic status.

The distribution of HbA_1c levels in study participants with stage 3 and stage 4 CKD is shown in Figure 1 and as a function of eGFR in the eFigure. During the median follow-up period (3.8 years [range, 1-51 months]), 16% of patients died, 49% were hospitalized, 16% had any CV event (3% stroke, 5% myocardial infarction, 8% acute heart failure), 6% had sustained doubling of serum creatinine value, and 2% developed ESRD.

For both stage 3 and stage 4 CKD, there was an increased risk of death associated with higher levels of HbA_1c (P value for trend <.001 for both comparisons). Compared with an HbA_1c level lower than 7%, an HbA_1c level higher than 9% was associated with significantly higher mortality among people with stage 3 and 4 CKD (adjusted hazard ratio [HR], 1.35; 95% CI, 1.21-1.50); the magnitude of the increased risk was similar for stage 3 and stage 4 CKD when considered separately (Table 2 and Table 3). A test for interaction between stage of CKD and HbA_1c level and the risk of mortality was nonsignificant (P = .72).

Similar findings were observed for each of myocardial infarction, stroke, and heart failure, as well as CV events in aggregate and all-cause hospitalization: a graded independent increase in risk was observed at higher levels of HbA_1c in all analyses and remained significant for stage 3 and stage 4 CKD separately in 4 stratified analyses (Table 2 and Table 3). Tests for interaction between stage of CKD and the risk of these events were all nonsignificant (P = .49, .89, .07, .30, and .34, respectively).

We also observed independent and statistically significant relationships between higher HbA_1c level and the risk of a sustained doubling of serum creatinine or the development of ESRD (P value for trend <.001 for both comparisons). For doubling of creatinine, there was no significant interaction according to stage of CKD (P = .65). The magnitude of the increased risk for ESRD associated with higher HbA_1c levels seemed lower for those with stage 4 CKD at baseline, compared with those with stage 3 CKD (P value for interaction <.001): among those with stage 4 CKD, the risk of ESRD was increased by 3% and 13% in patients with HbA_1c levels of 7% to 9% and higher than 9%, respectively, compared with patients with HbA_1c levels lower than 7% (P < .001; Table 3); corresponding findings for those in stage 3 were 22% and 152%, respectively (Table 2).

Because of suggestions that both high and low values of HbA_1c might be harmful,^9,21 we performed additional analyses that used nonlinear splines to examine the association between adverse outcomes and HbA_1c level modeled as a continuous variable. We observed a U-shaped relation between HbA_1c level and mortality, and visual inspection of these plotted data suggested 2 inflection points: there was a significant increase in risk when HbA_1c values were higher than 8% and when they were lower than 6.5% (Figure 1). There was no evidence of increased risk for myocardial infarction, stroke, heart failure, or ESRD at lower levels of HbA_1c (data not shown).

We tested for interactions between baseline HbA_1c level and baseline eGFR (as a continuous variable) on the risk of the clinical outcomes. None were statistically significant except for analyses related to ESRD (P = .02). Therefore, we modeled the association between HbA_1c level and the risk of ESRD as a continuous variable and found a direct relation between the excess risk associated with higher HbA_1c level and baseline eGFR (Figure 2).

In primary analyses, only the first HbA_1c value during the study period was included to represent overall glycemic control. To reduce the risk that our results were influenced by misclassification, we repeated all analyses using the mean of all measurements made during the exposure period to classify participants with respect to HbA_1c level. Results of these analyses were very similar to those of the primary analyses; the magnitude of the association between higher HbA_1c levels and the risk of adverse outcomes seemed similar or stronger, and all tests for trend remained significant (P < .001 for all comparisons). Finally, we repeated analyses using HbA_1c level as a time-varying covariate, and all results were very similar to those in the primary analysis (eTable 2 and eTable 3).

We studied data from almost 24 000 adults with DM and stage 3 to 4 CKD (eGFR, 15.0-59.9 mL/min/1.73 m²) treated in a universal health care system within a single Canadian province. In contrast to findings from patients with kidney failure,^22- 24 we found strong and independent associations between higher levels of HbA_1c and multiple clinically relevant outcomes, including mortality, CV events, hospitalization, and progression to kidney failure. These relations remained significant after controlling for multiple potential confounders, were observed in both stage 3 and stage 4 CKD, and were robust to a variety of sensitivity analyses. Consistent with findings from trials in the general population with DM,^9,25 we also found that levels of HbA_1c greater than 8.0% as well as levels lower than 6.5% were associated with increased mortality. Furthermore, our results suggest that many previous analyses that have considered HbA_1c values of less than 6% to 7% as a homogenous reference group may have systematically underestimated the strength of association between elevated HbA_1c levels and adverse events.^26- 27 Thus, our results have both clinical and scientific implications with respect to studies of glycemic control and adverse events.

Better glycemic control does help to prevent nephropathy and other microvascular complications of DM in people without CKD. For example, in patients with type 1 DM, the Diabetes Control and Complications Trial (DCCT) showed that a target HbA_1c level of 7% (vs 9%) over 9 years reduced the risk of microalbuminuria and macroalbuminuria by 34% and 56%, respectively.⁶ The DCCT-EDIC (Epidemiology of Diabetes Interventions and Complications) study followed patients for another 8 years after DCCT closeout and showed evidence of so-called legacy effects with respect to glycemic control: ongoing benefits in terms of developing microalbuminuria or macroalbuminuria as well as a significant 75% reduction in the risk of incident “renal failure” (creatinine level >177 μmol/L; to convert to conventional units, divide by 88.4).²⁸ Similarly, the UK Prospective Diabetes Study (UKPDS) showed that more intensive glycemic control (HbA_1c level of 7.0% vs 7.9%) in patients newly diagnosed as having type 2 DM was associated with significant and sustained reductions in the risk of microalbuminuria, macroalbuminuria, and doubling of serum creatinine level over a 10-year period.⁷ Based on trials such as these, most guidelines have extrapolated broadly and recommend an HbA_1c target level of 7.0% for patients with CKD, even though this population has been specifically excluded from the seminal trials of glycemic control for type 1 or type 2 DM.^10,29

Since the risk of adverse events such as hypoglycemia may be greater at lower levels of kidney function,³⁰ some have suggested that more liberal target levels for glycemic control may be appropriate for people with CKD,³¹ especially since microvascular damage has already occurred. Although better glycemic control was associated with lower risk of all-cause mortality irrespective of baseline kidney function, we found that the association between better glycemic control and decreased risk of ESRD was attenuated at lower levels of baseline eGFR. We speculate that this finding may represent a “point of no return” for kidney function—beyond which better glycemic control may simply not be enough to prevent progressive kidney function loss.

In contrast to data on microvascular outcomes, the link between better glycemic control and macrovascular outcomes in the general population of patients with DM is less clear. Long-term follow-up from DCCT and UKPDS suggest that more intensive glycemic control may reduce the risk of CV events and death.^25,28 However, 2 recent trials found that HbA_1c target levels below 6.5% led to increased CV mortality⁹ or no reduction in CV events³² in people with type 2 DM. Our findings of higher risk associated with both lower and higher levels of HbA_1c are broadly consistent with these results and suggest that there is little evidence of benefit with respect to macrovascular events when HbA_1c levels are much below 7.0%. That said, to our knowledge, there are no prior studies addressing the potential benefits or harms of better glycemic control on macrovascular outcomes in people with stage 3 or 4 CKD. As with participants in the Action to Control Cardiovascular Risk in Diabetes (ACCORD) trial, it is plausible that patients with DM and CKD who are treated to an HbA_1c level lower than 6.5% might experience iatrogenic harm owing to serious hypoglycemic events or too precipitous a fall in average glucose.⁹

Our current findings contrast with those of previous studies in patients undergoing hemodialysis, several of which show little or no association between HbA_1c level and mortality.^21-23 While these studies were not designed to explain the mechanism mediating the association between glycemic control and outcomes, one might speculate that competing causes of mortality in patients undergoing hemodialysis may attenuate any such association. In addition, the largest such study found that higher levels of HbA_1c were associated with increased mortality in patients undergoing hemodialysis³³ after adjustment for markers of malnutrition or chronic inflammation. Although it is possible that residual confounding by these characteristics have influenced our findings, the malnutrition-inflammation syndrome is substantially less common in patients with milder forms of CKD, such as those in the current study.^34- 35 Perhaps for these reasons, our findings seem more similar to the association between HbA_1c level and outcomes in the general population than in people with hemodialysis-dependent kidney failure.

Strengths of our study include the inclusion of a population-based cohort treated in a universal health care system. We used validated techniques to select patients with diagnosed DM and CKD, and we studied clinically relevant outcomes. Our findings were robust to multiple sensitivity analyses, and the excess risk was both statistically and clinically significant. Despite its novelty and strengths, our study also has several important limitations that deserve mention. First, this was a retrospective observation study that cannot confirm the benefit (or harm) of more intensive glycemic control or the manner in which this control is achieved. Second, while we controlled for many important clinical and demographic factors, we could not control for certain potential confounders, such as use of insulin, epoetin, or other medications; overall intensity of DM care; blood pressure control; and laboratory markers, such as hemoglobin or markers of inflammation or malnutrition. Third, we included only patients with HbA_1c measured as part of their routine clinical care and not part of a study protocol. Although all patients had access to medical services in Alberta's publicly funded health care system, we may not have captured data for individuals who chose not to be tested or who were unaware of their DM status. Fourth, we used HbA_1c level as our index of glycemic control. Although HbA_1c level may not correlate well with measured glycemic control in the setting of kidney failure (especially when epoetin is used), this does not seem to apply to earlier stages of CKD,³¹ suggesting that HbA_1c level is an appropriate index of glycemic control for our study population. Fifth, our definition of CKD was based on a single measurement of serum creatinine, which may have led to misclassification in some individuals. At the least, our sensitivity analyses suggest this misclassification is likely either nondifferential or at worst would tend to bias our findings to the null because of regression to the mean. Finally, we could not distinguish type 1 from type 2 DM, nor did we have any measures of the clinical severity of DM itself.

In summary, we found that both higher and lower levels of HbA_1c were associated with adverse events in a large population of patients with DM and stage 3 to 4 CKD. Our findings are consistent with the hypothesis that (as in the general population of patients with DM) better glycemic control in patients with stage 3 to 4 CKD tends to improve clinical outcomes, but that overly intensive therapy (ie, HbA_1c target level lower than 7%) may be harmful. This speculation requires confirmation in an adequately powered randomized trial.