Mortality in the Glycemia Reduction Approaches in Diabetes: A Comparative Effectiveness Study (GRADE)

Type 2 diabetes is common, occurring in over 10% of adults (1). Diabetes-related mortality rate, while decreasing, remains higher than the mortality rate for the general population (2). Among people with type 2 diabetes and known cardiovascular disease (CVD), there have been decreased morbidity and mortality with treatment with newer diabetes medications (3). However, the effect of glucose-lowering medications in populations at lower cardiorenal risk is unknown. Glycemia Reduction Approaches in Diabetes: A Comparative Effectiveness Study (GRADE) compared four glycemia-lowering medicines when added to metformin, with reporting of primary metabolic and vascular outcomes (4,5). GRADE offers an opportunity to assess the determinants of mortality in a cohort of participants with brief duration diabetes.

The study design of GRADE and procedures have previously been published (6). Consenting participants with type 2 diabetes diagnosed within 10 years, treated with metformin alone, and with HbA_1c 6.8%–8.5% (50.8–69.4 mmol/mol) were randomized to one of four glycemia-lowering medications (insulin glargine U-100, sulfonylurea glimepiride, glucagon-like peptide 1 receptor agonist [GLP-1RA] liraglutide, or dipeptidyl peptidase-4 inhibitor sitagliptin) and followed quarterly for a mean of 5 years. Exclusion criteria included major cardiovascular events in the past year, cancer other than nonmelanoma skin cancer or cancer treatment within 5 years, history of heart failure (New York Heart Association functional class III-IV), and estimated glomerular filtration rate <30 mL/min/1.73 m².

The primary outcome for this post hoc analysis was underlying cause of death. “Cardiovascular-cause” of death was a composite of underlying causes including major adverse cardiovascular event (MACE) death (due to known myocardial infarction [MI] or stroke), presumed CVD death, sudden death without a specific noncardiac cause, and unknown death, according to established trial conventions (7) (Supplementary Table 1). Immediate and underlying causes of death were determined from de-identified medical records, death certificates, and autopsy reports, as available, by an internal adjudication committee and external expert cardiologist masked to treatment assignment. Two adjudicators and a tiebreaker (as needed) reviewed and classified events using standard definitions (7,8) (Supplementary Methods 1). Date and cause of death were available for 137 deaths, with 16 unknown. Baseline covariates included age, sex, race, ethnicity, BMI, HbA_1c, hypertension, urine albumin-to-creatinine ratio (ACR), hyperlipidemia or statin treatment (hyperlipidemia), and the Framingham Risk Score (FRS) for CVD (9) (Supplementary Table 2). All analyses are descriptive. Tables 1 and 2 report numbers and percentages for all-cause mortality, cardiovascular-cause mortality, cancer mortality, and mortality from other causes.

To examine possible joint effects of the covariates, we fit a classification and regression tree (CART) model with underlying cause of death as the response and the variables above as the initial set of covariates. For analyses we used R, version 4.2.1 (10).

This article is based on follow-up data and outcome assessments from the 5,047 participants enrolled into GRADE. This database will be available in the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) Central Repository in 2024.

GRADE randomized 5,047 participants with type 2 diabetes (36.4% women, 65.7% White, 19.8% Black, 18.6% Hispanic, 3.6% Asian, 2.7% Native American/Alaska Native, and 0.6% Native Hawaiian/Pacific Islander). Mean ± SD age was 57 ± 10 years and diabetes duration 4.2 ± 2.7 years. Baseline prevalence of MI or stroke was 6.4%, hypertension or taking antihypertensive medication 73%, and hyperlipidemia or lipid-lowering medication 96% (11).

Study enrollment and overall mortality by treatment group are presented (Supplementary Fig. 1). During the 5.0-year mean follow-up (interquartile range 4.2–6.0, range 0–7.6), 153 (3%) participants died (0.59 per 100 participant-years [PY] [95% CI 0.50–0.68]). During the coronavirus disease 2019 (COVID-19) pandemic (after 1 March 2020), mortality was 1.02 per 100 PY (95% CI 0.71–1.33).

The most common underlying cause of death was cancer (26.8%) followed by presumed CVD (23.5%), unknown (10.5%), other (7.8%), infection (7.2%), and MACE (known MI or stroke) (3.9%) (Fig. 1A). COVID-19 was involved in 6.5% of overall deaths and 63% of deaths from pneumonia and infection (Supplementary Fig. 2). The most common causes of death did not differ among the four treatment groups. Combination of MACE, presumed CVD, sudden, and unknown death into one cardiovascular-cause category made cardiovascular-cause the leading underlying cause of death (Fig. 1B). In the case of the 16 deaths adjudicated as “unknown” there was insufficient information available for adjudicators to determine the underlying cause of death. The COVID-19 epidemic could have affected these results, but no analyses were conducted to address its impact. Most (58.5%) cancer deaths were from cancer of the lung and pancreas (Supplementary Table 3).

Table 1 presents the proportion of major underlying causes of death by treatment group with cardiovascular-cause and cancer being most common. Although mortality rate was numerically lowest in the liraglutide group versus the other three groups (rate ratio 0.64 [95% CI 0.42–0.69]), in the overall four-group analysis this was not significant as previously reported (5).

Table 2 shows expected differences in all-cause and disease-specific mortality according to baseline characteristics. Participants who died were older, male, and non-Hispanic, with hypertension, hyperlipidemia, higher ACR, and higher FRS. The pattern was the same for cardiovascular-cause, cancer, and other causes. The lower mortality among Hispanic participants was because they were younger (analyses not shown). Notably, HbA_1c and BMI were not higher and diabetes duration was not longer for participants who died (Supplementary Table 2).

CART analyses identified simple sets of participant characteristics associated with mortality. In these analyses, only the FRS was required in the final tree models for all-cause, cardiovascular-cause, and cancer mortality (Supplementary Fig. 3). For deaths other than cardiovascular-cause or cancer, the final model contained just age.

All-cause mortality in GRADE was relatively low at 3% over 5 years (0.59 deaths per 100 PY) and comparable with 0.93 deaths per 100 PY for a U.S. epidemiological cohort without diabetes (2). GRADE participants had recently diagnosed diabetes and well-managed risk factors, and among the cohort there were low rates of underlying CVD. All-cause mortality rates for type 2 diabetes in U.S. and U.K. epidemiological studies are 1.52 and 2.8–2.9 deaths per 100 PY, or 2.6- to ∼4.8-fold higher than those in GRADE (2,12), respectively. In the U.S. Action to Control Cardiovascular Risk in Diabetes (ACCORD) and the Anglo-Danish-Dutch Study of Intensive Treatment in People with Screen-Detected Diabetes in Primary Care (ADDITION-Europe), investigators tested intensive versus usual glycemic control for CVD outcomes in populations at high CVD risk; all-cause mortality rates were 1.43 vs. 1.14 per 100 PY in ACCORD and 1.43 vs. 1.56 per 100 PY in ADDITION (13,14), or two- to threefold higher than in GRADE.

While the combined cardiovascular-cause of mortality in GRADE was commonest (38.6%), cancer was also a common underlying cause (26.8%). These data parallel contemporaneous epidemiological reports showing a decrease in all-cause and cardiovascular-cause mortality (12) and variable absolute changes in cancer mortality. As a cancer diagnosis within 5 years was exclusionary in GRADE, these cancer deaths are largely incident cases. Notable other causes were infection (7.2%), which was almost twice as common as MACE; chronic liver disease; pneumonia; and accidents. Suicide was as frequent as end-stage renal disease death (2% each) in this group with recently diagnosed diabetes. Clinicians’ awareness of these less frequent causes should ensure appropriate screening, with diligence similar to that for cardio-renal risk.

The numerically lower all-cause mortality in the liraglutide treatment group (5), while not statistically significant, suggests the need for adequately powered studies testing whether the glucagon-like peptide 1 receptor agonist reduces mortality in populations with lower cardiovascular risk as demonstrated for those with high risk (3).

As a simple predictive tool in common clinical usage, the FRS should be investigated further in populations such as GRADE for predicting both cardiovascular-cause and all-cause mortality risk. The strong association with cardiovascular-cause mortality is unsurprising, but the association of FRS with cancer mortality is less obvious (Supplementary Fig. 3). This association is consistent with reports linking cardiovascular and cancer deaths in observational studies of diabetes and in the Framingham Heart Study (3). The basis of this shared risk is unknown.

HbA_1c was not a risk factor for mortality in GRADE. The relationship of HbA_1c with mortality and cardiovascular events is nonlinear, with the nadir for mortality varying from 6.0% to 7.9% (42–63 mmol/mol) in observational studies (15,16). Thus, the range of HbA_1c at baseline in GRADE (6.8%–8.5% [51–69 mmol/mol]) was perhaps too narrow to demonstrate a mortality effect of baseline or on-treatment glycemic control.

Based on the CART analysis, nonglycemic variables were more important than HbA_1c in predicting mortality. Recent diabetes care trends show a disturbing decline in use of blood pressure-, lipid-, and glucose-lowering treatment with worsening of risk factors. These trends predict future increases in diabetes mortality (17).

Together with guideline-directed glycemic control and nonglycemic risk factor management, including for cancer, infection, depression, accidents, smoking, and suicide, there are opportunities to narrow the mortality gap between diabetes and nondiabetes, which is increasingly related to diverse causes (2). That cancer was a leading cause of death in GRADE and other studies (12,17) suggests the importance of increased cancer prevention efforts (18,19).

A limitation of the study was a high proportion of unknown cause of death. This is frequently an issue in trials. Personal device technology might identify hypoglycemia and arrhythmia as contributory in a future study. GRADE was neither designed nor statistically powered for mortality testing, as the expected number of deaths was small, and thus subanalyses of risks and causes require confirmation. The study population was limited to the U.S.

In comparison with mortality in other longer-term diabetes trials, mortality over 5 years was low in the GRADE cohort with <10 years’ duration of diabetes, low rates of established cardiorenal disease, well-controlled traditional risk factors, and baseline HbA_1c 6.8%–8.5% (50.8–69.4 mmol/mol). The data suggest that the FRS predicts cardiovascular and cancer deaths. Whereas the focus in diabetes care has historically been on cardiovascular risk reduction, future studies should examine other public health interventions (20) intended to improve health and reduce mortality such as cancer screening, smoking cessation, infection control, and accident and suicide prevention.

Acknowledgments. The GRADE Research Group is deeply grateful to the participants, whose loyal dedication made GRADE possible.

Funding. GRADE was supported by a grant from the NIDDK of the National Institutes of Health (NIH) under award no. U01DK098246. The planning of GRADE was supported by a U34 planning grant from the NIDDK (U34-DK-088043). The American Diabetes Association (ADA) supported the initial planning meeting for the U34 proposal. The National Heart, Lung, and Blood Institute and the Centers for Disease Control and Prevention (CDC) also provided funding support. The Department of Veterans Affairs provided resources and facilities. Additional support was provided by NIH grants P30 DK017047, P30 DK020541-44, P30 DK020572, P30 DK072476, P30 DK079626, P30 DK092926, U54 GM104940, UL1 TR000439, UL1 TR000445, UL1 TR001108, UL1 TR001409, 2UL1TR001425, UL1 TR001449, UL1 TR002243, UL1 TR002345, UL1 TR002378, UL1 TR002489, UL1 TR002529, UL1 TR002535, UL1 TR002537, 2UL1 TR001425, and UL1 TR002548. Educational materials were provided by the National Diabetes Education Program. Material support in the form of donated medications and supplies has been provided by Becton, Dickinson and Company, Bristol-Myers Squibb, Merck & Co., Novo Nordisk, Roche Diagnostics, and Sanofi.

The content of this manuscript is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Duality of Interest. M.A.B. reports grant support paid to her institution from the NIH, NY Community Trust, and Eli Lilly, data and safety monitoring board (DSMB) consulting fees from Oramed, payment or honoraria for local Medical Grand Rounds, and a leadership role (unpaid board member) for the Lower New York Chapter of the American College of Clinical Endocrinology outside the submitted work. J.B.B. reports grant support paid to his institution from Dexcom, NovaTarg, Novo Nordisk, Sanofi, Tolerion, and vTv Therapeutics; consulting fees paid to his institution from Novo Nordisk; personal consulting fees from Alkahest, Altimmune, Anji Pharmaceuticals, AstraZeneca, Bayer, Biomea Fusion, Boehringer Ingelheim, CeQur, Cirius Therapeutics, Corcept Therapeutics, Eli Lilly, Fortress Biotech, GentiBio, Glycadia, Glyscend, Janssen, MannKind, Mediflix, Medscape, Mellitus Health, Moderna, Pendulum Therapeutics, Praetego, ReachMD, Sanofi, Stability Health, Terns Pharmaceuticals, Valo Health, and Zealand Pharma; personal payment for expert testimony from Medtronic; personal payment for participation on a DSMB or advisory board from Alkahest, Altimmune, and AstraZeneca; and stock or stock options for Glyscend, Mellitus Health, Pendulum Therapeutics, PhaseBio Pharmaceuticals, Praetego, and Stability Health outside the submitted work. R.P.-B. reports grant support paid to her institution from NIDDK, Juvenile Diabetes Research Foundation, Novo Nordisk, and Medtronic, consulting fees from Averitas Pharma, Bayer, Lexicon Pharmaceuticals, Nevro, Novo Nordisk, Procter & Gamble, Reata Pharmaceuticals, and Roche, and leadership roles (member, Board of Directors, and President, Medicine and Science) for the ADA outside the submitted work. S.P.F. reports grants or contracts paid to his institution from the NIH, CDC, and Pfizer outside the submitted work. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. All authors affirmed that authorship was merited based on the International Committee of Medical Journal Editors (ICMJE) authorship criteria. J.B.B., H.K.-S., and R.P.-B. contributed to the conception and design of the research. M.A.B., J.B.B., H.K.-S., C.U., and S.P.F. contributed to acquisition of data. N.Y. contributed to statistical analysis of data. All authors contributed to interpretation of data and results. M.A.B., J.B.B., and H.K.-S. contributed to acquisition of funding. M.A.B., J.B.B., and H.K.-S. contributed to the supervision and management of research. M.A.B., N.Y., and H.K.-S. contributed to the drafting of the manuscript. All authors contributed to the critical review of the manuscript. M.A.B. and N.Y. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

J.B.B. and R.P.-B. are editors of Diabetes Care but were not involved in any of the decisions regarding review of the manuscript or its acceptance.