Visual Abstract
Keywords: clinical epidemiology, diabetes, diabetes mellitus, epidemiology and outcomes, randomized controlled trials
Abstract
Significance Statement
Identifying and quantifying treatment effect variation across patients is the fundamental challenge of precision medicine. Here we quantify heterogeneous treatment effects of intensive glycemic control in the Action to Control Cardiovascular Risk in Diabetes (ACCORD) trial, considering three outcomes of interest—a composite kidney outcome (driven by macroalbuminuria), all-cause mortality, and first assisted hypoglycemic event. We demonstrate that the effects of intensive glycemic control vary with risk of kidney failure, as predicted by the kidney failure risk equation (KFRE). Participants at highest risk of kidney failure gain the largest absolute kidney benefit of intensive glycemic control but also experience the largest absolute risk of death and hypoglycemic events. Our findings illustrate the value of identifying clinically meaningful treatment heterogeneity, particularly when treatments have different effects on multiple end points.
Objective
Clear criteria to individualize glycemic targets in patients with type II diabetes are lacking. In this post hoc analysis of the ACCORD, we evaluate whether the KFRE can identify patients for whom intensive glycemic control confers more benefit in preventing kidney microvascular outcomes.
Research Design and Methods
We divided the ACCORD trial population into quartiles on the basis of 5-year kidney failure risk using the KFRE. We estimated conditional treatment effects within each quartile and compared them with the average treatment effect in the trial. The treatment effects of interest were the 7-year restricted mean survival time (RMST) differences between intensive and standard glycemic control arms on (1) time-to-first development of severely elevated albuminuria or kidney failure and (2) all-cause mortality.
Results
We found evidence that the effect of intensive glycemic control on kidney microvascular outcomes and all-cause mortality varies with baseline risk of kidney failure. Patients with elevated baseline risk of kidney failure derived the most from intensive glycemic control in reducing kidney microvascular outcomes (7-year RMST difference of 114.8 [95% confidence interval 58.1 to 176.4] versus 48.4 [25.3 to 69.6] days in the entire trial population) However, this same patient group also experienced a shorter time to death (7-year RMST difference of −56.7 [−100.2 to −17.5] v. −23.6 [−42.2 to −6.6] days).
Conclusions
We found evidence of heterogenous treatment effects of intensive glycemic control on kidney microvascular outcomes in ACCORD as a function of predicted baseline risk of kidney failure. Patients with higher kidney failure risk experienced the most pronounced reduction in kidney microvascular outcomes but also experienced the highest risk of all-cause mortality.
Introduction
Type 2 diabetes is a widespread noncommunicable disease of increasing prevalence wherein complications contribute to morbidity and mortality, requiring staggering health care expenditures.1 The leading cause of kidney failure, blindness, peripheral neuropathy, and lower limb amputation, type 2 diabetes is also a dominant risk factor for atherosclerotic cardiovascular disease and mortality. Effective glycemic control is fundamental to diabetes management; for patients who require more than nonpharmacological (lifestyle) management strategies, an array of oral and injectable medications have been introduced over the past two decades. Despite an expansion of therapeutic options, there remains uncertainty regarding the benefits and risks of the intensity of glycemic control for patients with type 2 diabetes.2,3
Previously conducted randomized controlled trials have shown that intensive glycemic control reduces microvascular events and may reduce macrovascular events, but increases the risk of serious hypoglycemic events and may increase the risk of mortality.4–7 Although newer pharmacological therapies have reduced the risks of serious hypoglycemia, nearly all clinical diabetes management guidelines advocate personalizing glycemic targets to reduce the potential for harm.8–12 Unfortunately, clear criteria for how to individualize glycemic targets are lacking.13 Intensive glycemic control has consistently been shown to reduce the risk of microvascular disease, explained largely by its effect on albuminuria.14 Therefore, identifying patients who differentially benefit from intensive glycemic control on kidney microvascular outcomes would be a promising approach to individualize glycemic targets.
Heterogeneous treatment effects occur when the effect of treatment depends on patients' baseline covariates. Conventional subgroup analyses (e.g., testing one-variable-at-a-time interactions) in randomized controlled trials of intensive glycemic control have failed to identify meaningful heterogeneity in treatment effects on microvascular outcomes.14 These analyses, however, have only explored heterogeneous treatment effects on the hazard ratio (HR)/relative scale, despite the fact that such effects are best understood on the absolute scale for clinical decision making.15 The lack of heterogeneity of treatment effects on the relative scale does not imply lack of heterogeneity of treatment effects on the absolute scale. Perhaps more importantly, conventional approaches to subgroup analysis are underpowered, prone to spurious false positive results due to multiple testing, and under-represent true clinical heterogeneity, in which patients differ from one another across many variables simultaneously.16 Instead, modern predictive approaches to characterizing heterogeneous treatment effects group patients using many clinically salient variables simultaneously and are better at detecting important treatment effect heterogeneity obscured by conventional analyses.15–18
In this work, we used a risk-based approach to identifying heterogeneous treatment effects, in which a multivariable model that predicts risk for the outcome is used to stratify patients within the trial to quantify risk-based variation in treatment effects. In this post hoc analysis of Action to Control Cardiovascular Risk in Diabetes (ACCORD) trial, we evaluated whether the kidney failure risk equation (KFRE), a validated tool that integrates patient's baseline age, sex, eGFR, and urine albumin-creatinine ratio into a risk score for kidney failure, identified patients who could disproportionately benefit from intensive glycemic control.19,20
Methods
Data
This is a post hoc secondary analysis of the limited-access ACCORD BioLINCC dataset obtained from the US National Institutes of Health. Details of the ACCORD study population, interventions, and study procedures have been previously published.4,7 In brief, 10,251 patients with diabetes, hemoglobin A1c (HbA1c) >7.5% and cardiovascular disease or two or more cardiovascular risk factors, were randomly assigned to intensive glycemic control (HbA1c <6.0%) or standard glycemic control (HbA1c 7.0%–7.9%).
Outcomes
The primary outcomes of interest in this analysis were (1) kidney microvascular events, defined as the composite of the time to first development of severely elevated albuminuria (urine albumin-creatinine ratio ≥300 mg/g) or time to first development of kidney failure, defined as initiation of maintenance dialysis or kidney transplantation or an increase in serum creatinine >3.3 mg/dl in the absence of an acute reversible cause; (2) all-cause mortality; (3) the time to first medically attended hypoglycemic event, a relevant adverse event in the setting of glycemic control. Kidney microvascular events were prespecified outcomes in ACCORD, and the original definitions are used here.7
Treatment Effect of Interest
There are several metrics used to define treatment effects with time-to-event data. Heterogeneous treatment effects are best understood on the absolute risk scale, and as such, the treatment effect of interest here is the 7-year restricted mean survival time (RMST) difference for each of the outcomes under intensive versus standard glycemic control.17,21,22 The RMST captures the average time free from a clinical event, within a specific time window (hence, restricted). The RMST difference, a measure of the treatment effect, is the average time delay in the onset of an event under treatment versus control, within a specific time window. It has been shown that one can make inferences about the RMST up to the largest follow-up time in the study.23 As such, in this analysis, we focus on the 7-year RMST difference between intensive and standard glycemic control, as in the ACCORD study, patients were followed for up to 7 years. The RMST difference has an additional geometric interpretation as the area between the two estimated survival curves of interest.
General Framework for Quantifying Heterogenous Treatment Effects
Heterogeneous treatment effects occur when the effect of treatment varies with baseline covariates in a nonrandom way. One general approach to quantifying heterogeneous treatment effects is to compare how the treatment effect in a subset of the trial population (conditional average treatment effect) compares with the treatment effect in the entire trial population (average treatment effect). If the conditional average treatment effects are significantly different from the average treatment effect, then there is evidence for heterogeneity in the treatment effect. Different methods exist to identify optimal subsets of the trial population, ranging from prespecified subgroup analyses to approaches that stratify patients by baseline risk (“risk-modeling”) to approaches that model treatment-by-covariate interactions explicitly (“effect-modeling”).
In this work, we use the risk-modeling approach, in which a multivariable model that predicts risk for the outcome is applied to stratify patients within the trial to examine risk-based variation in treatment effects. The premise behind using risk-modeling to identify heterogeneous treatment effects is that the effect of treatment will vary with baseline risk of the outcome (also called risk magnification).15,17,18 We used an externally developed risk prediction model to explore heterogenous treatment effects of intensive glycemic control on kidney outcomes.19,20,24 If treatment effects truly varied with patients' baseline risk of the outcome, we would expect a biologically plausible relationship between risk and the treatment effect (e.g., a monotone increasing treatment effect with risk or a U-shaped curve between risk and treatment effects).
Statistical Analysis
We analyzed the data in accordance with the intention-to-treat principle, with the goal of quantifying heterogeneous treatment effects of intensive glycemic control on kidney microvascular outcomes and all-cause mortality.
For participants with available data, we calculated the 5-year risk of kidney failure using the KFRE, which incorporates information on baseline age, sex, eGFR (using the race-free Chronic Kidney Disease Epidemiology Collaboration 2021 equation), and urine albumin-creatinine ratio. We estimated the empirical distribution of 5-year risk of kidney failure in the entire trial population and determined the 25th, 50th, and 75th percentiles of the empirical distribution. We grouped patients into quartiles of their 5-year risk of kidney failure.
Using the overall trial population, we estimated the average treatment effect as the 7-year RMST difference between intensive and standard glycemic control for the composite kidney microvascular outcome and all-cause mortality (average treatment effect). Randomization is not guaranteed to achieve covariate balance within subgroups. Therefore, within each quartile of predicted 5-year risk of kidney failure, before estimating treatment effects, we first quantified the balance of the baseline covariates across the treatment and control arms. Baseline variables with absolute standardized mean differences >0.10 would require adjustment in the RMST estimation, for example, by using an analysis of covariance-type adjustment for these covariates. Within each quartile, we estimated the 7-year RMST difference between intensive and standard glycemic control for the composite kidney microvascular outcome and all-cause mortality (conditional average treatment effects).
We quantified evidence for heterogeneous treatment effects by calculating the difference between the average treatment effect in the entire trial population and the conditional average treatment effects within each subgroup defined by quartiles of predicted 5-year risk of kidney failure. We estimated standard errors and associated 95% confidence intervals (CIs) for the conditional average treatment effects as well as their differences with the average treatment effect using a bootstrap procedure with 1000 replicates. A significant nonzero difference (associated 95% CI does not contain zero) between the average treatment effect and the conditional average treatment effect in a subgroup indicates evidence of treatment effect heterogeneity.
To allow for comparison with conventional metrics of treatment effects in time-to-event data, we also present estimated HRs using Cox proportional hazard models.
Companion Analyses
In companion analyses, we explored treatment effects within the quartiles of 5-year predicted kidney failure risk on (1) the total eGFR slope and (2) time to first sustained 40% reduction in eGFR or kidney failure or an increase in serum creatinine >3.3 mg/dl in the absence of an acute reversible cause. We also present treatment effects within deciles of 5-year predicted kidney failure risk. We also considered alternative risk stratification strategies, including stratifying by quartiles of (1) baseline albuminuria, (2) baseline eGFR, and (3) predicted risk derived from a model that specifically predicts early diabetic kidney disease (defined as eGFR <60 ml/min per 1.73 m2 and/or urine albumin-creatinine ratio (UACR) ≥30 mg/g for 3 or more months, caused by diabetes).24 As with the KFRE, we evaluated for heterogeneous treatment effects within quartiles of predicted diabetic kidney disease risk on the basis of the alternative models. Additional details on the statistical approach used to examine heterogeneous treatment effects on the total eGFR slope are provided in the supplement.
Reproducibility
All codes to reproduce this analysis are available at https://github.com/vivekcharu/ACCORD-JASN-2024.
Results
Of 10,251 participants randomized in ACCORD, 9777 (4904 randomized to intensive and 4873 randomized to standard glycemic control; 95.3%) had available baseline covariate data (age, sex, eGFR, and urine albumin-creatinine ratio) to estimate 5-year predicted kidney failure risk using the KFRE. The average effect of intensive glycemic control in the trial-eligible population was to delay the onset of severely elevated albuminuria or kidney failure by 48.4 days over a 7-year period (7-year RMST difference: 48.4 days [95% CI, 25.3 to 69.6]; corresponding HR: 0.75 [95% CI, 0.65 to 0.86]; Supplemental Figure 1). By contrast, the average effect of intensive glycemic control was to shorten the time to death by 23.6 days over a 7-year period (7-year RMST difference: −23.6 days [95% CI, −42.2 to −6.6]; HR: 1.20 [95% CI, 1.04 to 1.40]) and the time to first medically attended hypoglycemic event by 233 days over a 7-year period (7-year RMST difference: −260 to −208 days; HR: 2.91 [95% CI, 2.56 to 3.30]; Supplemental Figure 1).
Our goal was to assess whether effects of intensive glycemic control on kidney microvascular outcomes and adverse events varied with patients' baseline risk of kidney failure. The distributions of 5-year predicted kidney failure risk on the basis of the KFRE were similar in the intensive and standard glycemic control arms (Supplemental Figure 2). The 25th, 50th, and 75th percentiles of the empirical distribution of 5-year predicted kidney failure risk in the entire trial population were used to divide trial participants into four mutually exclusive subgroups. The 5-year predicted kidney failure risk ranged between zero and 0.004% (mean: 0.002%) in the first quartile; 0.004 and 0.014% (mean: 0.008%) in the second; 0.014 and 0.078% (mean: 0.034%) in the third; and 0.078 and 97% (mean: 0.99%) in the fourth. Table 1 displays the baseline characteristics of patients in each of the four subgroups. Compared with patients with the lowest predicted 5-year risk of kidney failure (quartile 1 [Q1]), those with the highest risk of kidney failure (quartile 4 [Q4]) were more likely to be male, Black, and older in age, have lower fasting plasma glucose, higher urine albumin-creatinine ratio, higher serum creatinine, and lower eGFR (Table 1). We found that baseline covariates were well-balanced between the treatment and control arms within each subgroup; absolute standardized mean differences were uniformly below 0.10, providing evidence that further adjustments were unnecessary, and would be unlikely to materially change the results presented (Supplemental Figure 3).
Table 1.
Baseline characteristics of Action to Control Cardiovascular Risk in Diabetes trial patients, overall, and by quartile of 5-year predicted risk of kidney failure using the kidney failure risk equation
| Baseline Characteristic | 5-yr KFRE Risk Quartiles | ||||
|---|---|---|---|---|---|
| Overall | Quartile 1 | Quartile 2 | Quartile 3 | Quartile 4 | |
| 5-yr predicted risk from KFRE (median %; IQR) | 0.014 (0.075) | 0.002 (0.001) | 0.007 (0.005) | 0.029 (0.026) | 0.28 (0.56) |
| Demographics/anthropomorphic measurements | |||||
| Female sex (yes/no) | 0.38 (0.49) | 0.48 (0.5) | 0.3 (0.46) | 0.34 (0.47) | 0.39 (0.49) |
| White (yes/no) | 0.62 (0.48) | 0.66 (0.47) | 0.67 (0.47) | 0.61 (0.49) | 0.57 (0.5) |
| Black (yes/no) | 0.19 (0.39) | 0.11 (0.31) | 0.14 (0.35) | 0.23 (0.42) | 0.29 (0.45) |
| Hispanic (yes/no) | 0.07 (0.26) | 0.09 (0.28) | 0.07 (0.26) | 0.06 (0.25) | 0.06 (0.24) |
| Other race/ethnicity (yes/no) | 0.11 (0.32) | 0.14 (0.35) | 0.12 (0.33) | 0.1 (0.3) | 0.09 (0.28) |
| Baseline age (yr) | 62.1 (9.5) | 58.6 (6) | 62 (8.6) | 63.6 (9.4) | 65.15 (10.3) |
| Body mass index (kg/m2) | 31.83 (7.71) | 32.27 (8.13) | 31.81 (7.47) | 31.7 (7.64) | 31.54 (7.6) |
| Laboratory data | |||||
| Systolic BP (mm Hg) | 136 (22) | 133 (20) | 135 (21) | 136 (23) | 138 (23) |
| Diastolic BP (mm Hg) | 75 (14) | 76 (13) | 75 (14) | 75 (15) | 73 (16) |
| Pulse pressure (mm Hg) | 60 (19) | 56 (17) | 60 (18) | 61 (18) | 65 (21) |
| Total cholesterol (mg/dl) | 178 (53) | 182 (54) | 177 (53) | 176 (51) | 178 (52) |
| Triglycerides (mg/dl) | 156 (122) | 161 (131) | 156 (122) | 148 (115) | 158 (121) |
| Very low–density lipoprotein (mg/dl) | 31 (25) | 32 (26) | 31 (25) | 30 (23) | 32 (24) |
| Low-density lipoprotein (mg/dl) | 101 (44) | 102 (45) | 99 (43) | 100 (42) | 101 (44.5) |
| High-density lipoprotein (mg/dl) | 40 (14) | 40 (13) | 39 (13) | 40 (13) | 40 (15) |
| Alanine aminotransferase (mg/dl) | 24 (14) | 25 (14) | 25 (15) | 24 (13.12) | 22 (12) |
| Creatine phosphokinase (mg/dl) | 106 (92) | 91 (74) | 102 (83) | 113 (102) | 120 (113) |
| Fasting plasma glucose (mg/dl) | 168 (65) | 173 (67) | 170 (63) | 167 (63.25) | 163 (67) |
| Glycosylated hemoglobin (%) | 8.1 (1.3) | 8.1 (1.3) | 8.1 (1.3) | 8.1 (1.2) | 8.2 (1.3) |
| Urinary albumin (mg/dl) | 1.69 (4.48) | 0.9 (1.09) | 1.94 (4.14) | 1.85 (5.54) | 3.99 (15.82) |
| Urinary creatinine (mg/dl) | 116.9 (77.5) | 112.7 (73.4) | 116.3 (80.1) | 121.3 (78.55) | 116.75 (76.23) |
| Urinary albumin-creatinine ratio (mg/g) | 14 (40) | 8 (9) | 17 (40) | 15 (44.5) | 35 (153) |
| eGFR (ml/min per 1.73 m2) | 87.95 (26.47) | 102.37 (6.11) | 95.91 (9.16) | 81.83 (9.54) | 63.3 (13.24) |
| Serum creatinine (mg/dl) | 0.9 (0.2) | 0.7 (0.2) | 0.8 (0.2) | 1 (0.2) | 1.2 (0.3) |
| Medications | |||||
| Antihypertension medication (yes/no) | 0.81 (0.39) | 0.74 (0.44) | 0.8 (0.4) | 0.83 (0.37) | 0.87 (0.33) |
| Cholesterol-lowering medication (yes/no) | 0.68 (0.47) | 0.65 (0.48) | 0.68 (0.47) | 0.7 (0.46) | 0.71 (0.46) |
| Diuretic medication (yes/no) | 0.37 (0.48) | 0.25 (0.44) | 0.32 (0.46) | 0.39 (0.49) | 0.51 (0.5) |
| Insulin (yes/no) | 0.35 (0.48) | 0.28 (0.45) | 0.3 (0.46) | 0.37 (0.48) | 0.45 (0.5) |
| Oral diabetes medication (yes/no) | 0.83 (0.37) | 0.85 (0.36) | 0.85 (0.35) | 0.84 (0.37) | 0.78 (0.41) |
Data are presented as mean (SD). IQR, interquartile range; KFRE, kidney failure risk equation.
We found evidence that the effect of intensive glycemic control on kidney microvascular outcomes (driven by severely elevated albuminuria) varied with baseline risk of kidney failure (Figures 1A and 2). First, we quantified how the effect of intensive glycemic control within each of the four subgroups differs from the overall average treatment effect in the trial. In the absence of heterogenous treatment effects, the effect of intensive glycemic control in each subgroup would be equivalent to the average treatment effect in the entire trial population, and thus, the difference in treatment effects between each subgroup and the overall trial population would be zero. By contrast, we found that compared with the average treatment effect in the entire trial population, patients in Q1 had lower treatment effects (difference in RMST differences between the overall trial effect and Q1 effect: −38.3 days [95% CI, −71.0 to −6.3]), and patients in Q4 experienced more than double the average treatment effect (difference in RMST differences between the overall trial effect and Q4 effect: 66.4 days [95% CI, 19.5 to 118.1]; Figure 1A, Table 2). Patients in Q2 and Q3 experienced treatment effects similar to the average treatment effect (Figure 1A). More specifically, translating these findings to the original treatment effect scale, we estimate that for patients with 5-year predicted kidney failure risk <0.004% (Q1), intensive glycemic control delayed the onset of the composite kidney outcome by 10.1 days over a 7-year window (95% CI, −23.9 to 43.1), while for patients with a 5-year predicted kidney failure risk >0.08% (Q4), intensive glycemic control delayed this onset by 114.8 days (95% CI, 58.1 to 176.4; Table 2). These results indicate that the average treatment effect observed in the entire trial population was predominantly driven by the effect seen in patients with elevated 5-year kidney failure risk. In companion analyses, we further demonstrated that the effect of intensive glycemic control on the composite kidney outcome in patients in the highest quartile of kidney failure risk is largely driven by its effect on the incident severely elevated albuminuria outcome (Supplemental Figure 4). We also demonstrated consistent, monotone results when considering deciles of predicted 5-year kidney failure risk (rather than quartiles; Supplemental Figure 5). In companion analyses, we also evaluated how the effects of intensive glycemic control varied with baseline 5-year predicted kidney failure risk by KFRE on the total eGFR slope (Supplemental Figures 6–8; Supplemental Table 1). We found that the average effect of intensive glycemic control in the trial-eligible population was to slow the rate of eGFR decline by 0.37 ml/min per 1.73 m2 per year (95% CI, 0.32 to 0.42). There was evidence of heterogeneity in this effect by KFRE quartile, with slightly, but significantly larger effects in patients with higher KFRE (i.e., Q1: 0.30 ml/min per 1.73 m2 per year; 95% CI, 0.20 to 0.39 versus Q4: 0.46 ml/min per 1.73 m2 per year; 95% CI, 0.36 to 0.57).
Figure 1.
Heterogeneous treatment effects of intensive glycemic control by baseline risk of kidney failure in ACCORD. Heterogeneous treatment effects of intensive glycemic control on (A) the composite kidney outcome, (B) all-cause death, and (C) time to first medically assisted hypoglycemic event. The x axis displays each subgroup of patients, defined by quartiles of 5-year predicted risk by the KFRE. The y axis displays the normalized restricted mean survival time (RMST) difference in days, defined as the RMST difference in the subgroup of interest minus the RMST difference in the entire trial. Normalized RMST values of zero indicate that the treatment effect in the subgroup of interest is equivalent to that in the entire trial population (no evidence of heterogeneous treatment effects). Normalized RMST values above zero (including the 95% confidence interval) indicate that the treatment effect in the subgroup of interest is larger than the treatment effect in the entire trial population (more beneficial); values below zero indicate that the treatment effect in the subgroup of interest is below that in the entire trial population (more harmful). KFRE, kidney failure risk equation.
Figure 2.
Heterogeneous treatment effects of intensive glycemic control on the composite kidney outcome by baseline risk of kidney failure in ACCORD. (A–D) Heterogeneous treatment effects of intensive versus standard glycemic control on the composite kidney outcome, by quartile of KFRE 5-year predicted risk at baseline. Plots demonstrate cumulative incidence curves by quartile of KFRE 5-year predicted risk at baseline; hazard ratios (HRs) and the 7-year restricted mean survival time difference between treatment and control arms are presented in days. Notice that subgroup-specific HRs (and associated 95% CIs) overlap, providing no evidence of heterogenous treatment effects on the relative risk scale, as opposed to the absolute scale, where subgroup-specific RMST differences are nonoverlapping, providing evidence of heterogenous treatment effects. CI, confidence interval.
Table 2.
Heterogeneous treatment effects of intensive glycemic control on the composite kidney outcome, all-cause death, and time to first medically assisted hypoglycemic event
| Population/Outcome of Interest | 7-yr RMST Difference (Intensive versus Standard Glycemic Control; 95% CI; d) | Difference Between the 7-yr RMST Difference in Each Quartile and That in the Overall Triala (95% CI; d) | Hazard Ratio (95% CI) |
|---|---|---|---|
| Composite kidney outcome | |||
| Overall trial | 48.4 (25.3 to 69.6)b | Ref | 0.75 (0.65 to 0.86)b |
| KFRE Q1 | 10.1 (−23.9 to 43.1) | −38.3 (−71.0 to −6.3)b | 0.93 (0.63 to 1.37) |
| KFRE Q2 | 40.1 (−4.7 to 84.7) | −8.3 (−46.9 to 31.1) | 0.78 (0.59 to 1.04) |
| KFRE Q3 | 50.08 (7.95 to 92.69)b | 1.68 (−35.00 to 36.96) | 0.73 (0.54 to 0.83)b |
| KFRE Q4 | 114.77 (58.14 to 176.41)b | 66.37 (19.51 to 118.09)b | 0.66 (0.53 to 0.83)b |
| All-cause mortality | |||
| Overall trial | −23.64 (−42.2 to −6.64)b | — | 1.20 (1.04 to 1.40)b |
| KFRE Q1 | −21.03 (−47.7 to 6.66) | 2.61 (−21.95 to 30.27) | 1.37 (0.91 to 2.07) |
| KFRE Q2 | −19.53 (−53.51 to 12.56) | 4.11 (−27.23 to 32.29) | 1.15 (0.84 to 1.58) |
| KFRE Q3 | 11.06 (−20.23 to 42.14) | 34.70 (7.29 to 63.81)b | 0.95 (0.70 to 1.29) |
| KFRE Q4 | −56.7 (−100.19 to −17.50)b | −33.06 (−66.43 to −0.53)b | 1.30 (1.03 to 1.64)b |
| Time to first medically assisted hypoglycemic event | |||
| Overall trial | −229.13 (−259.33 to −204.71)b | Ref | 2.91 (2.56 to 3.30)b |
| KFRE Q1 | −168.20 (−213.64 to −125.15)b | 60.93 (19.02 to 103.34)b | 3.05 (2.22 to 4.17)b |
| KFRE Q2 | −213.13 (−259.38 to −167.22)b | 16.00 (−27.98 to 56.02) | 3.90 (2.85 to 5.35)b |
| KFRE Q3 | −242.62 (−297.18 to −188.44)b | −13.50 (−59.35 to 33.38) | 2.99 (2.32 to 3.84)b |
| KFRE Q4 | −283.93 (−340.50 to −224.35)b | −54.80 (−103.41 to −5.83)b | 2.41 (1.97 to 2.94)b |
CI, confidence interval; KFRE, kidney failure risk equation; Q1–Q4, quartiles 1–4; RMST, restricted mean survival time.
The normalized RMST difference in days, defined as the RMST difference in the subgroup of interest minus the RMST difference in the entire trial. Normalized RMST values of zero indicate that the treatment effect in the subgroup of interest is equivalent to that in the entire trial population (no evidence of heterogeneous treatment effects). Normalized RMST values above zero (including the 95% CI) indicate that the treatment effect in the subgroup of interest is larger than the treatment effect in the entire trial population (more beneficial); values below zero indicate that the treatment effect in the subgroup of interest is below that in the entire trial population (more harmful).
Treatment effects have 95% CI that do not overlap with zero.
We also found evidence that the effect of intensive glycemic control on all-cause mortality varies with baseline risk of kidney failure (Figures 1B and 3). The effect of intensive glycemic control on all-cause mortality for patients in Q1 and Q2 of 5-year predicted kidney failure risk were similar to the overall population; however, patients in Q3 experienced a lower risk of mortality than average (difference in RMST differences between the overall trial effect and Q3 effect: 34.7 days [95% CI, 7.3 to 63.8]; Figure 1B) and patients in Q4 experienced a higher risk of mortality than average (difference in RMST differences between the overall trial effect and Q4 effect: −33.1 days [95% CI, −66.4 to −0.5]; Figure 1B). Translating these comparisons to the original treatment effect scale, we estimate that for patients in Q4 of 5-year predicted kidney failure risk, intensive glycemic control shortens the time to death by 56.7 days over a 7-year time frame (7-year RMST difference: −56.7 days [95% CI, −100.2 to −17.5]; Table 2). Importantly, cardiovascular deaths account for a substantial proportion of all-cause mortality in ACCORD, and a similar trend in treatment effects was observed for cardiovascular death (Supplemental Figure 9).
Figure 3.
Heterogeneous treatment effects of intensive glycemic control on all-cause mortality by baseline risk of kidney failure in ACCORD. (A–D) Heterogeneous treatment effects of intensive versus standard glycemic control on all-cause mortality, by quartile of KFRE 5-year predicted risk at baseline. Plots demonstrate cumulative incidence curves by quartile of KFRE 5-year predicted risk at baseline; hazard ratios (HRs) and the 7-year restricted mean survival time difference between treatment and control arms are presented in days.
In addition to all-cause mortality, we considered medically attended hypoglycemic events as an important adverse event, and we found evidence that the effect of intensive glycemic control on the time-to-first-attended hypoglycemic event also varied with baseline KFRE risk (Figures 1B and 4); patients with the highest risk of kidney failure at baseline experienced the shortest time-to-first-hypoglycemic event on intensive versus standard glycemic control (7-year RMST difference of −284 versus −168 days, Q4 versus Q1; Table 2).
Figure 4.
Heterogeneous treatment effects of intensive glycemic control on time to first medically assisted hypoglycemic event by baseline risk of kidney failure in ACCORD. (A–D) Heterogeneous treatment effects of intensive versus standard glycemic control on time to first medically assisted hypoglycemic event, by quartile of KFRE 5-year predicted risk at baseline. Plots demonstrate cumulative incidence curves by quartile of KFRE 5-year predicted risk at baseline; hazard ratios (HRs) and the 7-year restricted mean survival time difference between treatment and control arms are presented in days.
In companion analyses, we also explored how alternative risk scores perform in identifying heterogeneous treatment effects across different outcomes; we performed analogous analyses stratifying by (1) quartile of baseline urine albumin-creatinine ratio, (2) quartile of baseline eGFR, and (3) an externally developed risk score to predict early diabetic kidney disease24 (Supplemental Figures 10–12). Mirroring results from the primary analyses, we found that baseline urine albuminuria and eGFR (which are strongly correlated with KFRE; Supplemental Table 2) are able to identify heterogenous treatment effects in some outcomes but not others. By contrast, we found no evidence for heterogeneous treatment effects of intensive glycemic control on the composite kidney outcome or all-cause mortality when grouping patients on the basis of the diabetic kidney disease risk score developed by Jiang et al.24 Finally, we attempted to evaluate for heterogeneous treatment effects on the time to first sustained 40% reduction in eGFR or kidney failure; however, we found that intensive glycemic control had no effect on this outcome (7-year RMST difference: −2 days [95% CI, −25 to 20 days], HR: 1.02 [95% CI, 0.9 to 1.15]; Supplemental Figure 13). Given the null average treatment effect for this outcome, further analyses to evaluate for treatment effect heterogeneity were not performed.
Discussion
In this post hoc analysis of ACCORD, we demonstrated heterogeneous absolute treatment effects of intensive glycemic control (target HgbA1C <6.0%) on kidney microvascular outcomes (predominantly severely elevated albuminuria) and all-cause mortality on the basis of baseline 5-year predicted risk of kidney failure using the KFRE. We found that patients in the highest quartile of 5-year risk of kidney failure at baseline (>0.08% by KFRE) benefitted disproportionately from intensive glycemic control on the composite kidney outcome (predominantly severely elevated albuminuria; 7-year RMST difference of 114.8 days [95% CI, 58.1 to 176.4]) but also experienced an increased risk of all-cause death (7-year RMST difference: −56.7 days [95% CI, −100.2 to −17.5]). Absolute treatment effects in the highest quartile of kidney failure risk were approximately two-fold to three-fold those in the entire trial population. Echoing these findings, we also demonstrate that patients in the highest quartile of 5-year risk of kidney failure at baseline also experienced larger treatment effects on the rate of eGFR decline (total eGFR slope) but experienced shorter times to first-medically-attended hypoglycemic events.
There remains debate about the role of intensive glycemic control in the management of patients with type II diabetes. While nearly all guidelines recommend individualized glycemic targets, the American Diabetes Association, International Diabetes Federation, and European Association for the Study of Diabetes recommend an A1C target of <7% for most adults with type II diabetes; the American Association for Clinical Endocrinologists recommends an HgbA1C target of <6.5%. On the basis of the results of several large clinical trials, including ACCORD, the primary benefit of intensive glycemic control is reduced risk of microvascular complications, mostly driven by kidney microvascular outcomes.14 Our analysis demonstrates that the observed effect of intensive glycemic control on kidney microvascular outcomes in ACCORD is almost entirely driven by a subset of patients representing one quarter of the trial-eligible population at elevated risk of kidney failure at baseline. To our knowledge, ours is the first study to demonstrate that the benefit of intensive glycemic control on kidney microvascular outcomes was most pronounced in a subset of patients in ACCORD.
Several studies have explored potential mechanisms for the increase in mortality observed among participants randomized to intensive glycemic control in ACCORD.25–29 Our finding that patients at higher risk of kidney failure at baseline also experienced higher risks of mortality with intensive glycemic control echoes prior work demonstrating increased cardiovascular mortality under intensive glycemic control for patients with CKD in ACCORD.30 Definitive explanations to account for the increased mortality in this group of patients, however, are lacking. Importantly, other large randomized trials of intensive glycemic control, Action in Diabetes and Vascular Disease: Preterax and Diamicron MR Controlled Evaluation and Veterans Affairs Diabetes Trial, did not demonstrate increased mortality among patients receiving intensive glycemic control, and the finding of a mortality benefit among patients in Q3 of baseline KFRE also raises caution about the interpretation of this finding. In the modern era, new pharmacologic strategies may allow for glycemic control with lower risks of hypoglycemia than with insulin and/or sulfonylurea agents used in ACCORD. Whether our findings in ACCORD are reproducible in other randomized trials of intensive glycemic control, Veterans Affairs Diabetes Trial and Action in Diabetes and Vascular Disease: Preterax and Diamicron MR Controlled Evaluation, remains to be seen.5,6
Our study has several strengths. First, we quantify treatment effects on the absolute scale, using the RMST metric. The RMST captures the mean survival time in each treatment arm over a time window, with the difference in RMSTs describing the delay in onset of the outcome of interest between the treatment arms. In contrast to the HR, the RMST has a causal interpretation and a direct interpretation.31–33 Second, we used a risk-modeling approach to identify heterogenous treatment effects of intensive glycemic control on kidney microvascular outcomes, using an externally validated risk score. The risk-modeling approach aims to subgroup patients on the basis of baseline risk of the outcome using multiple clinically meaningful baseline covariates, avoiding potential issues with conventional “one-variable-at-a-time” subgroup identification. While this approach has been successfully used to characterize heterogeneous treatment effects in several cardiovascular trials,18 it has been underutilized in kidney disease trials. To our knowledge, this is the first study exploring heterogenous treatment effects on kidney outcomes in ACCORD and the first study to evaluate whether treatment effects on kidney outcomes might vary with a well-validated risk score for kidney failure. Although a new risk model can be developed using data from the RCT itself, an externally developed prediction model is preferred because overfitting data in the trial population can exaggerate the degree of risk heterogeneity. Externally developed prediction models are also more likely to be generalizable and have clinical utility. In addition, our study highlights several critical methodological considerations when exploring heterogenous treatment effects in randomized trials using similar approaches: (1) It is essential to consider treatment effects on both potential benefits and adverse events—our analysis demonstrates the subgroup that derived the largest absolute benefit on kidney outcomes from intensive glycemic control also experienced the largest absolute risk for hypoglycemic events and all-cause mortality; (2) the choice of risk score is essential in studies using a risk-modeling approach—we were not able to identify meaningful heterogenous treatment effects using an alternative risk score developed for early diabetic kidney disease (see Supplemental).24 Overall, our approach is straightforward and could have promising applications to several newer kidney disease clinical trials, including those of sodium–glucose transport protein 2 inhibitors and glucagon-like-1 receptor agonists. To encourage applications of these methods, we have made code publicly available.
Several limitations must also be mentioned. First, ACCORD enrolled a relatively low kidney risk group, and as such, we lack relevant data on the risks and benefits of intensive glycemic control in a population at high risk for kidney disease. Second, we used the KFRE to estimate patients' 5-year risk of kidney failure at baseline,19,20 which was developed to predict kidney failure in patients with CKD stages 3–5, although it is often used in patients with early-stage CKD (stage 2) as well.34,35 Most of the ACCORD participants did not have CKD at baseline, and our primary kidney outcome was driven almost entirely by albuminuria (kidney failure events were rare, occurring in approximately 4% of the trial population). Despite these limitations, we note that the KFRE is a monotone increasing function in UACR and a monotone decreasing function in eGFR and as such, likely still captures relevant risk for kidney outcomes. We did also assess heterogenous treatment effects using baseline albuminuria, baseline eGFR, and an alternative risk score that specifically predicts early diabetic kidney disease24; these additional analyses demonstrate that different risk stratification approaches yield different results, especially when one considers heterogeneity across multiple outcomes. It is possible that an alternative risk score would provide a more optimal grouping of patients than any of the scores considered. This analysis highlights the critical need for the development novel statistical methods to identify heterogeneous treatment effects across multiple outcomes simultaneously. Third, while risk-modeling is an elegant approach to identifying heterogenous treatment effects, the underlying assumption of the approach is that treatment effects will vary with baseline risk for the outcome. While this is certainly reasonable in many clinical contexts, it is not guaranteed, mathematically or clinically. Fourth, kidney function was only evaluated at scheduled visits causing interval censoring, and the precise time of severely elevated albuminuria or kidney failure was unknown in general, which may reduce the power of the analysis. However, owing to randomization, it is unlikely that interval censoring introduced systematic bias in estimating the treatment effects. Finally, this study does not provide immediate guidance for clinical decisions because ACCORD was conducted almost twenty years ago, and treatment options and monitoring approaches to glycemic control have evolved substantially.
A validated tool to predict kidney failure identified individuals who gain the largest absolute kidney benefit but also experienced the largest absolute risk of death from the ACCORD intensive glycemic control treatments. Our findings illustrate the value of applying modern predictive approaches to uncover clinically meaningful treatment heterogeneity for diabetes and kidney disease therapies, particularly when treatments have effects on multiple end points.
Supplementary Material
Disclosures
I.H. de Boer reports Consultancy: Alnylam, Astra Zeneca, Bayer, Boehringer-Ingelheim, Boehringer-Ingelheim/Lilly, George Clinical, Gilead, Medscape, Otsuka; Research Funding: Dexcom, Novo Nordisk; Honoraria: National Institutes of Health; and Advisory or Leadership Role: Deputy Editor of Clinical Journal of the American Society of Nephrology, Clinical Practice Guideline Co-Chair of Kidney Disease Improving Global Outcomes, and Chair of the American Heart Association Kidney in Heart Disease Science Committee. G.M. Chertow reports Consultancy: Akebia, Ardelyx, AstraZeneca, Calico, Gilead, Miromatrix, Reata, Sanifit, Unicycive, Vertex; Ownership Interest: Ardelyx, CloudCath, Durect, DxNow, Eliaz Therapeutics, Outset, Physiowave, PuraCath, Renibus, Unicycive; Research Funding: CSL Behring, NIAID, NIDDK; Advisory or Leadership Role: Board of Directors, Satellite Healthcare, Co-Editor, Brenner & Rector's The Kidney (Elsevier); and Other Interests or Relationships: DSMB service: Bayer, Gilead, Mineralys, NIDDK, ReCor. P. Geldsetzer reports Advisory or Leadership Role: GSK plc. M.K. Tamura reports Ownership Interest: Recursion; Honoraria: American Federation for Aging Research (nonprofit); and Advisory or Leadership Role: Beeson External Advisory Committee; CJASN Editorial Board; Clin-Star Advisory Board. L. Tian reports Employer: Gilead Inc.; Consultancy: Athersys Inc.; Biogen Inc., Ownership Interest: Gilead Inc.; and Research Funding: Consulting service with Athersys Inc., Biogen Inc., and Eisai Inc. All remaining authors have nothing to disclose.
Funding
This work was supported by KL2TR003143 (V. Charu), P30DK116074 (V. Charu), R01HL089778 (L. Tian), K24DK085446 (G.M. Chertow and M.E. Montez-Rath), R01DK128108 (M.K. Tamura and M.E. Montez-Rath), K24AG073615 (M.K. Tamura and M.E. Montez-Rath). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. V. Charu takes full responsibility for the work as a whole, including the study design, access to data, and the decision to submit and publish the manuscript.
Author Contributions
Conceptualization: Vivek Charu, Manjula Kurella Tamura.
Data curation: June Li, Manjula Kurella Tamura.
Formal analysis: Vivek Charu, Jane W. Liang.
Funding acquisition: Vivek Charu, Manjula Kurella Tamura.
Investigation: Vivek Charu, Manjula Kurella Tamura.
Methodology: Vivek Charu, Jane W. Liang, Maria E. Montez-Rath, Manjula Kurella Tamura, Lu Tian.
Software: Jane W. Liang.
Supervision: Vivek Charu.
Writing – original draft: Vivek Charu, Glenn M. Chertow, Jane W. Liang, Manjula Kurella Tamura, Lu Tian.
Writing – review & editing: Ian H. de Boer, Vivek Charu, Glenn M. Chertow, Pascal Geldsetzer, June Li, Jane W. Liang, Maria E. Montez-Rath, Manjula Kurella Tamura, Lu Tian.
Data Sharing Statement
Previously published data were used for this study.
Supplemental Material
This article contains the following supplemental material online at http://links.lww.com/JSN/E561.
Supplemental Figure 1. Effects of intensive (treatment) versus standard (control) glycemic control in the ACCORD trial population on (A) the composite kidney outcome, (B) all-cause mortality, and (C) the time to first assisted hypoglycemic event. 7-year RMST differences are presented in days.
Supplemental Figure 2. Baseline 5-year predicted kidney failure risk using the kidney failure risk equation among patients randomized to standard (red) and intensive (green) glycemic control.
Supplemental Figure 3. Covariate imbalance between treatment and control arms in ACCORD by quartile of KFRE.
Supplemental Figure 4. Heterogeneous treatment effects of intensive glycemic control on each renal microvascular outcome considered: (A) incidence macroalbuminuria and (B) kidney failure.
Supplemental Figure 5. Heterogeneous treatment effects of intensive glycemic control on (A) the composite kidney outcome and (B) all-cause mortality within deciles of predicted 5-year kidney failure risk.
Supplemental Figure 6. Average treatment effect on total eGFR slope in the entire ACCORD trial.
Supplemental Figure 7. Analysis of heterogenous treatment effects on the total eGFR slope outcome by quartile of baseline KFRE.
Supplemental Figure 8. Analysis of heterogenous treatment effects on the total eGFR slope outcome by quartile of baseline KFRE.
Supplemental Figure 9. Heterogeneous treatment effects of intensive glycemic control time to (A) time to primary outcome, a composite of cardiovascular death, nonfatal myocardial infarction and nonfatal stroke, and (B) time to cardiovascular death, by quartile of 5-year predicted kidney failure risk at baseline by KFRE.
Supplemental Figure 10. Heterogeneous treatment effects across several outcomes by quartile of baseline UACR.
Supplemental Figure 11. Heterogeneous treatment effects across several outcomes by quartile of baseline eGFR.
Supplemental Figure 12. Heterogeneous treatment effects across several outcomes by quartile of baseline risk for early diabetic kidney disease on the basis of the risk score of Jiang et al.
Supplemental Figure 13. Average treatment effect of intensive versus standard glycemic control in the entire trial eligible population on the time to first ESKD or sustained (consecutive) eGFR reduction of 40%.
Supplemental Table 1. Heterogeneous treatment effects in total eGFR slope by KFRE quartile.
Supplemental Table 2. Cross-tabulation of patients by baseline KFRE quartile, baseline UACR quartile, and baseline eGFR quartile.
References
- 1.Khan MAB, Hashim MJ, King JK, Govender RD, Mustafa H, Al Kaabi J. Epidemiology of type 2 diabetes—global burden of disease and forecasted trends. J Epidemiol Glob Health. 2020;10(1):107–111. doi: 10.2991/jegh.k.191028.001 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Rodriguez-Gutierrez R, Gonzalez-Gonzalez JG, Zuñiga-Hernandez JA, McCoy RG. Benefits and harms of intensive glycemic control in patients with type 2 diabetes. BMJ. 2019;367:l5887. doi: 10.1136/bmj.l5887 [DOI] [PubMed] [Google Scholar]
- 3.Rodríguez-Gutiérrez R, Montori VM. Glycemic control for patients with type 2 diabetes mellitus: our evolving faith in the face of evidence. Circ Cardiovasc Qual Outcomes. 2016;9(5):504–512. doi: 10.1161/CIRCOUTCOMES.116.002901 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Action to Control Cardiovascular Risk in Diabetes Study Group, Gerstein HC Miller ME Byington RP, et al. Effects of intensive glucose lowering in type 2 diabetes. N Engl J Med. 2008;358(24):2545–2559. doi: 10.1056/NEJMoa0802743 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.ADVANCE Collaborative Group, Patel A MacMahon S Chalmers J, et al. Intensive blood glucose control and vascular outcomes in patients with type 2 diabetes. N Engl J Med. 2008;358(24):2560–2572. doi: 10.1056/NEJMoa0802987 [DOI] [PubMed] [Google Scholar]
- 6.Duckworth W Abraira C Moritz T, et al. Glucose control and vascular complications in veterans with type 2 diabetes. N Engl J Med. 2009;360(2):129–139. doi: 10.1056/NEJMoa0808431 [DOI] [PubMed] [Google Scholar]
- 7.Ismail-Beigi F Craven T Banerji MA, et al. Effect of intensive treatment of hyperglycaemia on microvascular outcomes in type 2 diabetes: an analysis of the ACCORD randomised trial. Lancet. 2010;376(9739):419–430. doi: 10.1016/S0140-6736(10)60576-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.American Diabetes Association. 6. Glycemic targets: standards of medical care in diabetes—2021. Diabetes Care. 2021;44(suppl 1):S73–S84. doi: 10.2337/dc21-S006 [DOI] [PubMed] [Google Scholar]
- 9.de Boer IH Caramori ML Chan JCN, et al. Executive summary of the 2020 KDIGO Diabetes Management in CKD Guideline: evidence-based advances in monitoring and treatment. Kidney Int. 2020;98(4):839–848. doi: 10.1016/j.kint.2020.06.024 [DOI] [PubMed] [Google Scholar]
- 10.Blonde L Umpierrez GE Reddy SS, et al. American Association of Clinical Endocrinology Clinical Practice Guideline: developing a diabetes mellitus comprehensive care plan—2022 update. Endocr Pract. 2022;28(10):923–1049. doi: 10.1016/j.eprac.2022.08.002 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.de Boer IH Khunti K Sadusky T, et al. Diabetes management in chronic kidney disease: a consensus report by the American Diabetes Association (ADA) and Kidney Disease: Improving Global Outcomes (KDIGO). Diabetes Care. 2022;45(12):3075–3090. doi: 10.2337/dci22-0027 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Kidney Disease Improving Global Outcomes KDIGO Diabetes Work Group. KDIGO 2022 Clinical practice guideline for diabetes management in chronic kidney disease. Kidney Int. 2022;102(5S):S1–S127. doi: 10.1016/j.kint.2022.06.008 [DOI] [PubMed] [Google Scholar]
- 13.Ismail-Beigi F, Moghissi E, Tiktin M, Hirsch IB, Inzucchi SE, Genuth S. Individualizing glycemic targets in type 2 diabetes mellitus: implications of recent clinical trials. Ann Intern Med. 2011;154(8):554–559. doi: 10.7326/0003-4819-154-8-201104190-00007 [DOI] [PubMed] [Google Scholar]
- 14.Zoungas S Arima H Gerstein HC, et al. Effects of intensive glucose control on microvascular outcomes in patients with type 2 diabetes: a meta-analysis of individual participant data from randomised controlled trials. Lancet Diabetes Endocrinol. 2017;5(6):431–437. doi: 10.1016/S2213-8587(17)30104-3 [DOI] [PubMed] [Google Scholar]
- 15.Kent DM Paulus JK van Klaveren D, et al. The predictive approaches to treatment effect heterogeneity (PATH) statement. Ann Intern Med. 2020;172(1):35–45. doi: 10.7326/M18-3667 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Kent DM, Rothwell PM, Ioannidis JP, Altman DG, Hayward RA. Assessing and reporting heterogeneity in treatment effects in clinical trials: a proposal. Trials. 2010;11(1):85. doi: 10.1186/1745-6215-11-85 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Kent DM van Klaveren D Paulus JK, et al. The predictive approaches to treatment effect heterogeneity (PATH) statement: explanation and elaboration. Ann Intern Med. 2020;172(1):W1–W25. doi: 10.7326/M18-3668 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Kent DM, Steyerberg E, van Klaveren D. Personalized evidence based medicine: predictive approaches to heterogeneous treatment effects. BMJ. 2018;363:k4245. doi: 10.1136/bmj.k4245 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Tangri N Stevens LA Griffith J, et al. A predictive model for progression of chronic kidney disease to kidney failure. JAMA. 2011;305(15):1553–1559. doi: 10.1001/jama.2011.451 [DOI] [PubMed] [Google Scholar]
- 20.Tangri N Grams ME Levey AS, et al. Multinational assessment of accuracy of equations for predicting risk of kidney failure: a meta-analysis. JAMA. 2016;315(2):164–174. doi: 10.1001/jama.2015.18202 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Chen PY, Tsiatis AA. Causal inference on the difference of the restricted mean lifetime between two groups. Biometrics. 2001;57(4):1030–1038. doi: 10.1111/j.0006-341x.2001.01030.x [DOI] [PubMed] [Google Scholar]
- 22.Kloecker DE, Davies MJ, Khunti K, Zaccardi F. Uses and limitations of the restricted mean survival time: illustrative examples from cardiovascular outcomes and mortality trials in type 2 diabetes. Ann Intern Med. 2020;172(8):541–552. doi: 10.7326/M19-3286 [DOI] [PubMed] [Google Scholar]
- 23.Tian L Jin H Uno H, et al. On the empirical choice of the time window for restricted mean survival time. Biometrics. 2020;76(4):1157–1166. doi: 10.1111/biom.13237 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Jiang W Wang J Shen X, et al. Establishment and validation of a risk prediction model for early diabetic kidney disease based on a systematic review and meta-analysis of 20 cohorts. Diabetes Care. 2020;43(4):925–933. doi: 10.2337/dc19-1897 [DOI] [PubMed] [Google Scholar]
- 25.Basu S, Raghavan S, Wexler DJ, Berkowitz SA. Characteristics associated with decreased or increased mortality risk from glycemic therapy among patients with type 2 diabetes and high cardiovascular risk: machine learning analysis of the ACCORD trial. Diabetes Care. 2018;41(3):604–612. doi: 10.2337/dc17-2252 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Bonds DE Miller ME Bergenstal RM, et al. The association between symptomatic, severe hypoglycaemia and mortality in type 2 diabetes: retrospective epidemiological analysis of the ACCORD study. BMJ. 2010;340:b4909. doi: 10.1136/bmj.b4909 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Boyko EJ. ACCORD glycemia results continue to puzzle. Diabetes Care. 2010;33(5):1149–1150. doi: 10.2337/dc10-0432 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Riddle MC Ambrosius WT Brillon DJ, et al. Epidemiologic relationships between A1C and all-cause mortality during a median 3.4-year follow-up of glycemic treatment in the ACCORD trial. Diabetes Care. 2010;33(5):983–990. doi: 10.2337/dc09-1278 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Siraj ES Rubin DJ Riddle MC, et al. Insulin dose and cardiovascular mortality in the ACCORD trial. Diabetes Care. 2015;38(11):2000–2008. doi: 10.2337/dc15-0598 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Papademetriou V Lovato L Doumas M, et al. Chronic kidney disease and intensive glycemic control increase cardiovascular risk in patients with type 2 diabetes. Kidney Int. 2015;87(3):649–659. doi: 10.1038/ki.2014.296 [DOI] [PubMed] [Google Scholar]
- 31.Hernán MA. The hazards of hazard ratios. Epidemiology. 2010;21(1):13–15. doi: 10.1097/EDE.0b013e3181c1ea43 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Martinussen T, Vansteelandt S, Andersen PK. Subtleties in the interpretation of hazard contrasts. Lifetime Data Anal. 2020;26(4):833–855. doi: 10.1007/s10985-020-09501-5 [DOI] [PubMed] [Google Scholar]
- 33.Royston P, Parmar MK. Restricted mean survival time: an alternative to the hazard ratio for the design and analysis of randomized trials with a time-to-event outcome. BMC Med Res Methodol. 2013;13(1):152. doi: 10.1186/1471-2288-13-152 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Duggal V, Montez-Rath ME, Thomas IC, Goldstein MK, Tamura MK. Nephrology referral based on laboratory values, kidney failure risk, or both: a study using veterans affairs health system data. Am J Kidney Dis. 2022;79(3):347–353. doi: 10.1053/j.ajkd.2021.06.028 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Bhachu HK Cockwell P Subramanian A, et al. Impact of using risk-based stratification on referral of patients with chronic kidney disease from primary care to specialist care in the United Kingdom. Kidney Int Rep. 2021;6(8):2189–2199. doi: 10.1016/j.ekir.2021.05.031 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
Previously published data were used for this study.