Are the Different Diabetes Subgroups Correlated With All-Cause, Cancer-Related, and Cardiovascular-Related Mortality?

Abstract

Context

Numerous studies have shown that cardiovascular disease (CVD) represents the most important cause of mortality among people with diabetes mellitus (DM). However, no studies have evaluated the risk of CVD-related mortality among different DM subgroups.

Objective

We aimed to examine all-cause, CVD-related, and cancer-related mortality for different DM subgroups.

Design, Setting, Patients, and Interventions

We included participants (age ≥ 20 years) from the National Health and Nutrition Examination Survey III (NHANES III) data set. We evaluated the risks of all-cause and cause-specific (CVD and cancer) mortality among 5 previously defined diabetes subgroups: severe autoimmune diabetes (SAID), severe insulin-deficient diabetes (SIDD), severe insulin-resistant diabetes (SIRD), mild obesity-related diabetes (MOD), and mild age-related diabetes (MARD).

Primary Outcome Measure

The hazard ratios (HRs) for all-cause and cause-specific (CVD and cancer) mortality were measured for each of the 5 DM subgroups. We also evaluated the odds ratios (ORs) for retinopathy and nephropathy in each subgroup.

Results

A total of 712 adults were enrolled and the median follow-up time was 12.71 years (range, 0.25-18.08 years). The number of deaths in the 5 subgroups (SAID, SIDD, SIRD, MOD, and MARD) were 50, 75, 64, 7, and 18, respectively, and the number of CVD-related deaths in the 5 subgroups was 29, 30, 26, 2, and 11, respectively. Compared to the MOD subgroup, the adjusted HRs and 95% CIs of CVD-related mortality for the SAID, SIDD, SIRD, and MARD subgroups were 3.23 (95% CI, 0.77-13.61), 2.87 (95% CI, 0.68-12.06), 2.23 (95% CI, 0.53-9.50), and 4.75 (95% CI, 1.05-21.59), respectively (the HR for the MARD subgroup had a P value of .04). In addition, compared to the MARD subgroup, the adjusted ORs and 95% CIs for retinopathy in the SAID and SIDD groups were 2.38 (95% CI, 1.13-5.01, P = .02) and 3.34 (95% CI, 1.17-6.88, P = .001), respectively. The ORs for nephropathy were nonsignificant.

Conclusions

Our study of patients from the NHANES III data set indicated that among the different DM subgroups, the MARD subgroup tended to have a higher CVD-related mortality than the MOD subgroup. The all-cause and cancer-related mortality rates were similar across the different diabetes subgroups. In addition, compared to the MARD subgroup, the SAID and SIDD subgroups had a higher retinopathy risk, but there was no difference in nephropathy among the subgroups.

mortality, severe autoimmune diabetes, severe insulin-deficient diabetes, severe insulin-resistant diabetes, mild obesity-related diabetes, mild age-related diabetes

More than 30.3 million Americans (9.4% of the US population), are now living with diabetes mellitus (DM), and 84.1 million Americans have prediabetes (1). Numerous epidemiological studies have shown that cardiovascular disease (CVD) represents the leading cause of mortality among people with DM (2). Einarson et al demonstrated that among individuals with type 2 DM (T2DM), there was a 32.2% prevalence of macrovascular complications, and coronary heart disease (CHD) was the most common complication (3). Apart from CHD, stroke, and peripheral artery disease, other macrovascular complications in T2DM were also major causes of disability among individuals with T2DM (3). Moreover, CVD and DM have always been among the top 10 causes of mortality in the United States and many other countries (4). These diseases cause enormous burdens not only to individuals but also to society. Therefore, macrovascular complications, especially CHD in T2DM, have become important public health issues in recent decades.

In addition to macrovascular complications, DM is associated with microvascular complications, a principal cause of end-stage renal disease, blindness, and neuropathy (5, 6). According to the traditional definitions of diabetes, approximately 80 % of patients are classified as having T2DM. However, among the same diagnosis and treatment patients, they have heterogeneous phenotypes and disease complications (7). To find a powerful tool to identify different patient characteristics and risk of diabetic microvascular complications, a refined classification of DM from 3 large cohort studies in Sweden was performed (8). Ahlqvist et al identified 5 novel subgroups of patients with newly diagnosed diabetes based on 6 variables: age at diagnosis, body mass index (BMI), glycated hemoglobin (HbA_1c), glutamate decarboxylase antibodies (GAD-65 Abs), and the updated homeostasis model assessment of β-cell function (HOMA2-β) and HOMA2 of insulin resistance (HOMA2-IR) (8). They separated participants from 3 large cohort studies in Sweden into 5 novel DM subgroups, namely, severe insulin-deficient diabetes (SIDD), severe insulin-resistant diabetes (SIRD), severe autoimmune diabetes (SAID), mild obesity-related diabetes (MOD), and mild age-related diabetes (MARD). They reported that the SIRD subgroup had a higher risk of diabetic nephropathy, whereas the SIDD subgroup had a higher risk of retinopathy (8). Zou et al confirmed the possibility of categorizing diabetes patients into subgroups using the China National Diabetes and Metabolic Disorders Study for Chinese patients and the National Health and Nutrition Examination Survey III (NHANES III) for American patients (9). They demonstrated that the distribution of subgroups was similar across different participant samples, but the patients in each subgroup had similar characteristics across all ethnicities (9). In addition, Bancks and colleagues estimated the prevalence of CVD risk factors and microvascular complications among each subgroup but did not measure CVD-related mortality events (10). Furthermore, Dennis et al examined the A Diabetes Outcome Progression Trial (ADOPT) and used the same clusters reported by Ahlqvist et al to estimate glycemic progression and the incidence of chronic kidney disease. They demonstrated a faster progression of renal disease in the SIRD and MARD subgroups (11). Moreover, Zaharia et al examined the same clusters reported by Ahlqvist et al in the German Diabetes Study and confirmed that diabetic polyneuropathy was more prevalent at baseline in the SIDD subgroup than in other subgroups (12). Udler and colleagues identified novel clusters of genetic loci and revealed that clusters 1 and 2 were related to β-cell function, and clusters 3, 4, and 5 were related to the insulin response (13). Tanabe et al used the same clusters reported by Ahlqvist et al to examine a Japanese population. They demonstrated increased risk of diabetic retinopathy (DR) in SAID and SIDD subgroups, whereas the SIRD subgroup had the highest risk of developing diabetic nephropathy (14). Collectively, these studies’ results implied the generalizability of diabetic microvascular complications among the novel diabetic subgroups in Europe, China, and the United States. Some studies have examined the prevalence of CVD risk factors but neglected to examine CVD-related mortality, which is the leading cause of mortality in the DM population across these different DM subgroups.

Therefore, our aim was to analyze the all-cause and cause-specific (CVD and cancer) mortality risks among the 5 different DM subgroups in the United States by using data from the NHANES III. The results of this study may enhance our understanding of the risks of CVD, all-cause, and cancer-related mortality among the 5 DM subgroups.

Materials and Methods

Ethics statement

Data for our study came from the NHANES III database, a nationally representative sample of the civilian, noninstitutionalized US population, which had been approved by the National Center for Health Statistics (NCHS), a part of the Centers for Disease Control and Prevention (CDC).

Data source and participants

Our study participants were individuals who were age 20 years or older from the NHANES III (1988-1994) and who underwent a detailed home interview and a health examination by trained medical staff at a mobile examination center. The total number of participants in the NHANES III was 20 050, and we enrolled 1677 DM participants for our analyses. The exclusion criteria in our study were individuals with missing information on age, sex, smoking history, medication use condition (blood pressure–lowering, antidiabetic and lipid-lowering), glutamate decarboxylase 65 antibody (GAD-65 Ab), and HOMA2. Mortality data were acquired from the National Death Index (NDI) and matched to the NHANES participants (15). The NHANES is a nationally representative, population-based, multistage, cross-sectional study carried out and approved by the US NCHS (15). All participants provided written informed consent during the period of recruitment. Demographic data were collected, and participants underwent detailed home interviews, physical examinations, and a body-composition assessment in addition to providing blood samples.

Definitions of traditional diabetes

The definition of type 1 diabetes was having C-peptide values less than 0.3 nmol/L and being glutamate decarboxylase 65 antibody (GAD-65 Ab) positive, and the definition of latent autoimmune diabetes in adults was being GAD-65 Ab positive and having more than 0.3 nmol/L of C-peptide values (16). Moreover, T2DM was defined as meeting the DM criteria except for being GAD-65 Ab positive; gestational diabetes was defined as the development of DM among women during pregnancy (17).

Definitions of different subgroups of diabetic mellitus

Ahlqvist et al chose 6 variables—GAD-65 Ab, HbA_1c, BMI, age diagnosed with DM, and HOMA2-IR and HOMA2-β based on C-peptide and fasting plasma glucose values estimated by the HOMA2 calculator—to separate the participants into 5 novel subgroups (8, 18). Ahlqvist and colleagues used the K-means clustering method to separate the participants, but we modified the process to provide the clinician with a more convenient method in clinical applications. The positive cutoff value of GAD-65 Ab was defined as greater than the 75th percentile of the population, severe insulin deficiency was defined as a HOMA2-β index of less than 50%, and severe insulin resistance was defined as a HOMA2-IR index of 3 or greater (13, 19).

Outcome assessment

The primary outcomes in our study were time to all-cause and cause-specific (CVD and cancer) mortality among the 5 DM subgroups. CVD-related mortality included myocardial infarction (MI) and cerebrovascular diseases. The NCHS established the NDI and offered the confirmed mortality data (15). Follow-up information was used from the time of the NHANES III survey through December 31, 2006. In addition, we evaluated the odds ratios (ORs) for retinopathy and nephropathy among the subgroups.

Retinopathy and nephropathy definition

Retinopathy was defined as the presence of the following factors on the fundus photograph: nonproliferative DR, intraretinal microvascular abnormalities without microaneurysms, retinal microaneurysms, hemorrhages, soft exudates, hard exudates, and proliferative DR (20). Regarding nephropathy, chronic kidney disease was defined as kidney damage and an estimated glomerular filtration rate (eGFR) of less than 60 mL/min per 1.73 m². eGFR was calculated using equations derived from the Modification of Diet in Renal Disease (MDRD) Study (21).

Measurement: covariates

From the demographic questionnaire data, we investigated the following variables: age, sex, race/ethnicity, smoking history, and type of medication (blood pressure–lowering, antidiabetic, and lipid-lowering). We assessed the type of medication by using codes from the NHANES III data set or by using self-reports from NHANES III questionnaires. The NHANES medical codes are listed in Supplemental Tables 1 to 3 (22). All of the procedures followed CDC guidelines.

Statistical analysis

All statistical analyses were performed using SPSS software (version 18.0, SPSS, Inc). The t test was used to evaluate the individual differences among demographic characteristics and continuous variables. Kruskal-Wallis tests were used to determine whether samples were normally distributed. We presented the quantitative data as the mean ± SD for normally distributed variables and as the median with interquartile ranges for nonnormally distributed variables. We presented qualitative data as frequencies and percentages. Kaplan-Meier plots and the log-rank test were performed to evaluate the time of the occurrence of mortality in the future among the novel subgroups. Individual survival among the novel subgroups was plotted using Kaplan-Meier curves. We used multivariate Cox proportional-hazard models to obtain the HRs of each novel DM subgroup during the follow-up period. We adjusted for age, sex, race/ethnicity, ever smoking, hypertension (HTN) medication use, diabetes medication use, and lipid medicine use. A P value (2-sided) of less than .05 was considered significant. In addition, no previous studies have examined CVD-related mortality across diabetes subgroups. Thus, we chose the MOD subgroup as the reference group for all association analyses because this subgroup had the lowest mortality rate in this study.

Results

Study sample characteristics

The study population included 712 adults from the NHANES III database (Fig. 1). Table 1 shows the demographic data of the 5 different subgroups: SAID (n = 164, 23.0%), SIDD (n = 275, 38.6%), SIRD (185, 25.9%), MOD (n = 32, 4.5%), and MARD (n = 56, 7.9%). As shown in Table 1, the MOD subgroup had the lowest glucose and HbA_1c levels. Compared with other subgroups, the SIRD subgroup had a higher HOMA2-β percentage and the highest HOMA2-IR level. The SIDD subgroup had the highest glucose, HbA_1c, and the lowest HOMA2-β percentage level. The SIDD subgroup also had a higher prevalence of retinopathy, whereas the SIRD subgroup had a higher prevalence of eGFR lower than 60 mL/min per 1.73 m².

Figure 1.

Study flow diagram.

Table 1.

Characteristics of study participants with different subgroups of diabetes

	SAID (n = 164)	SIDD (n = 275)	SIRD (n = 185)	MOD (n = 32)	MARD (n = 56)	P
Continuous variablesa
Exam age, y	66.5 (58.0 to 73.75)	62.0 (53.0 to 71.0)	65.0 (58.0 to 73.5)	61.5 (55.25 to 69.5)	70.5 (60.25 to 79.0)	< .001
Age at diagnosis, y	56.0 (45.0 to 64.0)	50.0 (40.0 to 59.0)	55.0 (47.0 to 64.0)	52.0 (40.0 to 64.25)	61.0 (50.0 to 68.0)	< .001
Glucose, mg/dL	182 (142.0 to 258.5)	247 (197.0 to 298.0)	158 (137.0 to 191.0)	134.5 (128.25 to 151.5)	136 (130.0 to 143.75)	< .001
HOMA2-β, %	49.50 (25.38 to 78.43)	29.30 (19.9 to 40.2)	91.20 (69.5 to 105.9)	72.50 (60.5 to 85.86)	69.50 (59.28 to 77.38)	< .001
HOMA2-IR	3.350 (2.03 to 4.58)	2.90 (1.60 to 4.60)	4.40 (3.60 to 5.60)	2.60 (2.15 to 2.80)	2.30 (2.00 to 2.60)	< .001
C-peptide, nmol/L	1.11 (0.73 to 1.53)	0.85 (0.53 to 1.23)	1.66 (1.39 to 2.08)	1.02 (0.87 to 1.10)	0.91 (0.79 to 1.02)	< .001
HbA1c, %	8.30 (6.70 to 9.90)	9.30 (7.90 to 10.70)	7.20 (6.50 to 8.30)	6.50 (5.90 to 7.40)	6.60 (6.00 to 7.55)	.50
GAD65-Ab, U/mL	0.038 (0.03 to 0.05)	0.001 (–0.01 to 0.01)	0.001 (–0.01 to 0.01)	–0.001 (–0.001 to 0.01)	–0.001 (–0.001 to 0.01)	< .001
LDL, mg/dL	141.5 (120.3 to 179.25)	139.0 (115.0 to 170)	132.5 (114.5 to 172.8)	135.5 (125.5 to 159.0)	147.0 (117.5 to 170.0)	.48
SBP, mm Hg	138.0 (122.5 to 156.0)	136.0 (124.0 to 152)	138.0 (126.0 to 154.0)	132.0 (118.5 to 153.0)	138.0 (124.0 to 156.0)	.68
Categorical variablesb
Male sex	82 (50.0)	135 (49.1)	79 (42.7)	14 (43.8)	27 (48.2)	.63
Non-Hispanic white	66 (40.2)	95 (34.5)	88 (47.6)	13 (40.6)	31 (55.4)	.02
Smokingc	100 (61.0)	153 (55.6)	97 (52.4)	21 (65.6)	29 (51.8)	.37
Retinopathy	72 (43.9)	142 (51.6)	66 (35.7)	10 (31.3)	12 (21.4)	< .001
eGFR < 60d, MDRD	73 (44.5)	95 (34.7)	102 (55.4)	15 (46.9)	30 (53.6)	< .001
HTN medication use	79 (48.2)	116 (42.2)	111 (60.0)	16 (50.0)	25 (44.6)	.94
Lipid medication use	18 (11.0)	29 (10.5)	20 (10.8)	3 (9.4)	7 (12.5)	.99
Oral DM medication use	65 (39.6)	117 (42.5)	82 (44.3)	5 (15.6)	17 (30.4)	< .001
Insulin use	50 (30.5)	101 (36.7)	34 (18.4)	6 (18.8)	6 (10.7)	.02

Abbreviations: DM, diabetes mellitus; eGFR, estimated glomerular filtration rate; GAD6-Ab, glutamic acid decarboxylase antibody; HbA_1c, glycated hemoglobin; HOMA2-β, homeostasis model assessment of β-cell function; HOMA2-IR, homeostasis model assessment of insulin resistance; HTN, hypertension; LDL, low-density lipoprotein; MDRD, Modification of Diet in Renal Disease Study; MOD, mild obesity-related diabetes; MARD, mild age-related diabetes; SAID, severe autoimmune diabetes; SBP, systolic blood pressure; SIDD, severe insulin-deficient diabetes; SIRD, severe insulin-resistant diabetes.

^aContinuous values expressed as median (interquartile range).

^bCategorical variables data are expressed as No. (%).

^cSmoked more than 100 cigarettes in life.

^d The eGFR is estimated by MDRD formula.

Survival analyses of cardiovascular diseases categorized by different diabetes mellitus subgroups

Figure 2 shows the Kaplan-Meier plots of the DM subgroups and survival analysis for CVD mortality. After a total 18.08-year follow-up (mean, 11.29 years), the numbers of CVD-related deaths in each subgroup (SAID, SIDD, SIRD, MOD, and MARD) were 29, 30, 26, 2, and 11, respectively. The HRs of CV mortality and death numbers across DM subgroups are summarized in Table 2. The adjusted HRs and 95% CIs for SAID, SIDD, SIRD, and MARD were 3.23 (95% CI, 0.77-13.6), 2.87 (95% CI, 0.68-12.06), 2.23 (95% CI, 0.53-9.50), and 4.75 (95% CI, 1.05-21.59), respectively (the HR for the MARD subgroup had a P value of .04). Collectively, these findings indicate that compared to the MOD subgroup, the MARD subgroup tended to have a higher CVD mortality.

Figure 2.

Kaplan-Meier curve for cardiovascular mortality categorized by different subgroups of diabetes mellitus.

Table 2.

Hazard ratios and events of cardiovascular disease mortality categorized by different subgroups of diabetes mellitus

Subgroups (n)	Deaths, n (%)	Unadjusted-model HR (95% CI)	P	Adjusted-model HR (95% CI)b	P
MODa (32)	2 (6.2)	(Reference)	–	(Reference)	–
SAID (164)	29 (18.0)	3.76 (0.90-15.77)	.07	3.23 (0.77-13.61)	.11
SIDD (275)	30 (10.9)	2.28 (0.54-9.54)	.26	2.87 (0.68-12.06)	.15
SIRD (185)	26 (14.0)	2.86 (0.68-12.05)	.15	2.23 (0.53-9.50)	.28
MARD (56)	11 (19.6)	5.13 (1.14-23.16)	.03	4.75 (1.05-21.59)	.04

Abbreviations: HR, hazard ratio; MOD, mild obesity-related diabetes; MARD, mild age-related diabetes; SAID, severe autoimmune diabetes; SIDD, severe insulin-deficient diabetes; SIRD, severe insulin-resistant diabetes.

^aMOD as reference group.

^bAdjustment for age, sex, race/ethnicity, ever smoking, hypertension medication use, diabetes medication use, and lipid medicine use.

Survival analyses of all-cause mortality classified by diabetes mellitus subgroups

The numbers of deaths in each DM subgroup (SAID, SIDD, SIRD, MOD, and MARD) after a total 18.08-year follow-up (mean, 11.29 years) was 50, 75, 64, 7, and 18, respectively. In addition, the median follow-up time was 12.71 years, and the median time to death in the 5 subgroups (SAID, SIDD, SIRD, MOD, and MARD) was 9.21, 7.33, 6.50, 8.33, and 7.50 years, respectively. Table 3 lists the HRs of all-cause mortality and death numbers among the DM subgroups. The adjusted HRs and 95% CIs for the SAID, SIDD, SIRD, and MARD subgroups were 1.68 (95% CI, 0.76-3.72), 2.07 (95% CI, 0.95-4.51), 1.88 (95% CI, 0.86-4.15), and 2.35 (95% CI, 0.97-5.66), respectively (all P > .05). These findings indicate there were no differences in all-cause mortality between the MOD subgroup and the other subgroups. Notably, the association of the MARD subgroup with all-cause mortality was attenuated after adjusting for covariates (unadjusted HR, 2.45; 95% CI, 1.02-5.86, P = .05 and adjusted HR, 2.35; 95% CI, 0.97-5.66, P = .06).

Table 3.

Hazard ratios and events of all-cause mortality categorized by different subgroups of diabetes mellitus

Subgroups (n)	Deaths, n (%)	Unadjusted-model HR (95% CI)	P	Adjusted-model HR (95% CI)b	P
MODa (32)	7 (21.9)	(Reference)	–	(Reference)	–
SAID (164)	50 (30.4)	1.88 (0.85-4.14)	.12	1.68 (0.76-3.72)	.20
SIDD (275)	75 (27.2)	1.65 (0.76-3.58)	.21	2.07 (0.95-4.51)	.07
SIRD (185)	64 (34.5)	2.07 (0.95-4.51)	.07	1.88 (0.86-4.15)	.12
MARD (56)	18 (30.5)	2.45 (1.02-5.86)	.05	2.35 (0.97-5.66)	.06

Abbreviations: HR, hazard ratio; MOD, mild obesity-related diabetes; MARD, mild age-related diabetes; SAID, severe autoimmune diabetes; SIDD, severe insulin-deficient diabetes; SIRD, severe insulin-resistant diabetes.

^aMOD as reference group.

^bAdjustment for age, sex, race/ethnicity, ever smoking, hypertension medication use, diabetes medication use, and lipid medicine use.

Survival analyses of cancer-related mortality stratified by different diabetes mellitus subgroups

Table 4 reveals the HRs of cancer-related mortality and the number of deaths among the different DM subgroups. The adjusted HRs and 95% CIs for the SAID, SIDD, SIRD, and MARD subgroups were 0.26 (95% CI, 0.06-1.18), 0.87 (95% CI, 0.25-3.11), 0.66 (95% CI, 0.18-2.40), and 0.70 (95% CI, 0.14-3.51), respectively (all P > .05). These findings indicate there were no differences in cancer-related mortality between the MOD subgroup and the other subgroups.

Table 4.

Hazard ratios and events of cancer mortality categorized by different subgroups of diabetes mellitus

Subgroups (n)	Deaths, n (%)	Unadjusted-model HR (95% CI)	P	Adjusted-model HR (95% CI)b	P
MODa (32)	3 (9.3)	(Reference)	–	(Reference)	–
SAID (164)	4 (2.4)	0.33 (0.07-1.46)	.14	0.26 (0.06-1.18)	.08
SIDD (275)	14 (5.1)	0.70 (0.20-2.45)	.58	0.87 (0.25-3.11)	.84
SIRD (185)	12 (6.5)	0.87 (0.24-3.08)	.83	0.66 (0.18-2.40)	.53
MARD (56)	3 (5.3)	0.86 (0.17-4.29)	.86	0.70 (0.14-3.51)	.66

Abbreviations: HR, hazard ratio; MOD, mild obesity-related diabetes; MARD, mild age-related diabetes; SAID, severe autoimmune diabetes; SIDD, severe insulin-deficient diabetes; SIRD, severe insulin-resistant diabetes.

^aMOD as reference group.

^bAdjustment for age, sex, race/ethnicity, ever smoking, hypertension medication use, diabetes medication use, and lipid medicine use.

Odds ratios of retinopathy and nephropathy categorized by different diabetes mellitus subgroups

Table 5 lists the ORs of retinopathy and nephropathy among the different DM subgroups. Compared to the MARD subgroup, the SAID and SIDD subgroups had a higher retinopathy risk. The adjusted ORs and 95% CIs for the SAID and SIDD subgroups were 2.38 (95% CI, 1.13-5.01; P = .02) and 3.34 (95% CI, 1.17-6.88; P = .001), respectively. There were no differences in nephropathy between the MARD subgroup and the other subgroups.

Table 5.

Odds ratios of retinopathy and nephropathy categorized by different subgroups of diabetes mellitus

Variables	Retinopathy				Nephropathy
Subgroups (n)	Unadjusted-model OR (95% CI)	P	Adjusted-model OR (95% CI)b	P	Unadjusted-model OR (95% CI)	P	Adjusted-model OR (95% CI)b	P
MARDa (56)	(Reference)	–	(Reference)	–	(Reference)	–	(Reference)	–
SAID (164)	2.87 (1.41-5.83)	.004	2.38 (1.13-5.01)	.02	0.70 (0.37-1.28)	.24	0.81 (0.39-1.66)	.56
SIDD (275)	3.92 (1.98-7.73)	< .001	3.34 (1.17-6.88)	.001	0.46 (0.26-0.82)	.01	0.64 (0.32-1.29)	.21
SIRD (185)	2.03 (1.00-4.12)	.05	1.79 (0.86-3.75)	.12	1.08 (0.59-1.97)	.81	1.41 (0.69-2.89)	.35
MOD (32)	1.67 (0.62-4.45)	.31	2.04 (0.73-5.72)	.17	0.77 (0.32-1.83)	.55	1.45 (0.52-4.01)	.48

Abbreviations: HR, hazard ratio; MOD, mild obesity-related diabetes; MARD, mild age-related diabetes; OR, odds ratio; SAID, severe autoimmune diabetes; SIDD, severe insulin-deficient diabetes; SIRD, severe insulin-resistant diabetes.

^aMARD as reference group.

^bAdjustment for age, sex, race/ethnicity, ever smoking, hypertension medication use, diabetes medication use, and lipid medicine use.

Discussion

Our study replicated and modified the previous study reported by Ahlqvist and colleagues to categorize the US diabetic population from the NHANES III data set into 5 subgroups (8). Previous studies found that low titers of GAD-65 Abs are detectable in the serum of approximately 5% of T2DM patients, and approximately 8% of healthy participants older than 50 years are positive for GAD-65 (19). Thus, we choose the 75th percentile of eligible participants to cover this possibility (23). In addition, the U.K. Prospective Diabetes Study (UKPDS) revealed that among T2DM patients, β-cell dysfunction occurs in 50% (24). The individuals in the UKPDS receiving DM medication therapy demonstrated an early increase in β-cell function from 45% to 78% in year 1, followed by a decrease of approximately 5% per year. Thus, we chose the value of 50% of the β-cell function for our threshold. Moreover, Ahmet et al applied the HOMA-IR to examine the relationships between age, sex, and BMI and the prevalence of DM and insulin resistance and found that the cutoff value was 3 (25). Thus, the positive cutoff value of GAD-65 Abs was defined as greater than the 75th percentile of the population, severe insulin deficiency was defined as a HOMA2-β index of less than 50%, and severe insulin resistant was defined as a HOMA2-IR index of 3 or greater (13, 19).

In this cohort study, the most prominent finding was that compared to the MOD subgroup, there was a positive correlation between the MARD subgroup and CVD-related mortality risk (P = .04). Furthermore, none of the DM subgroups had a significant correlation with all-cause or cancer-related mortality risk. To our knowledge, the present study is the first to focus on the relationships among CVD-related, all-cause, and cancer-related mortality in diabetic individuals in the United States categorized into SAID, SIDD, SIRD, MOD, and MARD subgroups. Moreover, compared to the MARD subgroup, the SAID and SIDD subgroups had higher retinopathy risks. The adjusted ORs and 95% CIs for the SAID and SIDD subgroups were 2.38 (95% CI, 1.13-5.01) and 3.34 (95% CI, 1.17-6.88), respectively, and all P values were less than .05. There were no differences in nephropathy between the MARD subgroup and the other subgroups.

First, old age, especially being 75 years and older, is highly correlated with CVD-related mortality. Mozaffarian et al described a report from the American Heart Association stating that almost two-thirds of CVD deaths occur in people aged 75 or older (26). Furthermore, the proportion of individuals aged 70 or older who died within 1 year following a first MI was 2- to 3-fold higher than in those aged 40 to 69 years. Second, apart from MI, the stroke mortality rate also increases with age. A study from Canada showed that 7-day and discharge stroke fatality rates were higher among those older than 80 years than among younger age groups (27). In our study, the mean enrolled age and mean age at DM diagnosis in the MARD subgroup were 68.05 and 60.29 years, respectively, and an increase in age was associated with higher risks for CVD events and CVD-related mortality. After a total 18.08-year follow-up (mean, 11.29 years), almost all patients in the MARD subgroup who were older than 75 years had a high risk for CVD death. Third, the MARD subgroup had lower HbA_1c levels (median, 6.6%; interquartile range, 6.0-7.55) than the SAID, SIDD, and SIRD subgroups. Abraira and colleagues indicated that a lower HbA_1c level among older individuals was correlated with new cardiovascular events (28). These participants were veterans with a mean ± SD age of 60 ± 6 years and an average of 7.8 years since their DM diagnosis, and Abraira et al concluded that it might be reasonable to set a therapeutic goal of HbA_1c levels not much lower than 8% among these older patients (28). In addition, after a 3.5-year follow-up, the Action to Control Cardiovascular Risk in Diabetes (ACCORD) trial showed that intensive therapy among individuals with T2DM, who had a mean ± SD age of 62.2 ± 6.8 years, led to an increase in cardiovascular mortality (HR 1.35; 95% CI, 1.04-1.76; P = .02) (29). Intensive control of blood sugar among older patients with DM might cause an increase in CVD-related mortality, similar to our MARD subgroup, who had a relatively older age and lower HbA_1c levels. Taken together, among elderly DM adults (aged 65 and order), intensive glycemic control (HbA_1c = 6–7%, or even < 6%) did not lead to benefits for CVD mortality.

Interestingly, our study revealed that the MOD subgroup had a relatively lower risk of CVD-related mortality. First, although obesity, especially abdominal obesity, is highly correlated with T2DM, HTN, and hyperlipidemia, a higher CVD risk might exist in patients with severe obesity (BMI ≥ 40 kg/m²) (30, 31). The mean BMI in the MOD subgroup was 34.41 kg/m², which was lower than 40 kg/m², and this subgroup had a lower CVD mortality risk. A similar result was reported by Flegal et al, who found that obese participants with a BMI ranging from 30 to 35 kg/m² did not have a higher risk of mortality (32). Moreover, we separated the severe insulin deficient (HOMA2-β < 50%) and severe insulin resistance (HOMA2-IR > 3) participants into SIDD and SIRD subgroups. Thus, individuals in the MOD subgroup were only moderately obese but without severe insulin deficiency and severe insulin resistance. Hamer and Stamatakis reported that the prevalence of metabolically healthy obese (MHO) among obese participants from the Health Survey for England and Scottish Health Survey was 24% (33). Compared to metabolically healthy nonobese (MHNO) individuals, the MHO individuals were not at increased risk of CVD (HR 1.26; 95% CI, 0.74-2.13) or all-cause mortality (HR 0.91; 95% CI, 0.64-1.29). Moreover, Calori and colleagues reported that there were no increased CVD, cancer-related, or all-cause mortality risks among MHO based on BMI and HOMA, similar to the obese individuals without severe insulin resistance in our study (34).

This study demonstrated that all-cause mortality was similar across the DM subgroups. Individuals with all types of diabetes have higher mortality than the non-DM population, and the well-recognized etiological factor is prolonged exposure to hyperglycemia (35, 36). However, other factors, such as autoimmune antibodies, insulin resistance, and β-cell function, are also important and may lead to an increased risk of mortality for certain subgroups. For instance, Olsson et al reported that although autoimmune diabetes had a lower baseline metabolic risk profile than T2DM, all-cause mortality was equally high as in T2DM. The excess risk was related to poor serum glucose control, and the SAID subgroup had relatively higher HbA_1c levels and an equal risk of mortality compared to other subgroups (37). Regarding the β-cell function factor, Curtis et al indicated that a 20% decrease in HOMA-β was associated with increased CVD events and all-cause death among healthy elderly individuals in the United States (38). Furthermore, regarding insulin resistance, Ausk and colleagues reported that HOMA-IR was positively correlated with all-cause mortality in a nondiabetic US population with a normal BMI, whereas the correlations were inconclusive among individuals with high or low BMIs (39). Another meta-analysis demonstrated that HOMA-IR was independently associated with greater risks of CVD or all-cause mortality in nondiabetic adults (40). The previously mentioned factors might explain why individuals in the MARD subgroup had higher CVD-related mortality but had equal all-cause mortality risks compared to the SAID, SIDD, and SIRD subgroups.

Our study also analyzed the retinopathy and nephropathy risks among the 5 DM subgroups. Compared to the MARD subgroup, the SAID and SIDD subgroups had a higher retinopathy risk, especially the SIDD subgroup, (OR 3.34; 95% CI, 1.17-6.88; P = .001). The median HOMA-β levels in the SAID and SIDD subgroups were 49.5 and 29.3, respectively, which were lower than those in the other 3 subgroups. Ahlqvist et al reported that the SIDD subgroup had the highest risk of retinopathy (8). These results were similar to Chowdhury et al, who found that DR is associated with reduced β-cell function, leading to fasting and postprandial hyperglycemia and hypoinsulinemia rather than insulin resistance (41). Regarding diabetic kidney disease, our results showed that there were no differences in nephropathy between the MARD subgroup and the other subgroups. Dennis and colleagues examined the incidence of diabetic kidney disease in the ADOPT and used the same clusters reported by Ahlqvist et al; they found that the SIRD and MARD subgroups showed a faster progression of renal disease, whereas Ahlqvist reported that only the SIRD subgroup had a significantly higher risk of nephropathy than individuals in the MOD and MARD subgroups (8, 11). Furthermore, Bancks et al found that a young adulthood-onset group had a greater prevalence of eye and renal complications, whereas our study showed that compared to MARD subgroup, there were higher retinopathy risks among the SAID and SIDD subgroups and no differences in the risk for nephropathy among the subgroups (10).

The previous studies reported by Ahlqvist, Zaharia, Dennis, Bancks, and Tanabe and colleagues revealed that the MARD subgroup was the largest subgroup, whereas the SIDD subgroup was the largest in our study (8, 10-12, 14). All of these studies used the same way “K-means” for cluster analysis. The K-means clustering minimizes within-cluster variances and has the tendency to produce equal-sized clusters. On the contrary, we used thresholds for GAD-65 Ab, HOMA2-IR, and HOMA2-β to separate the DM subgroups. This difference compared with the previous studies can separate individuals according to being older but more autoimmune, having severe insulin resistance, and being severe insulin deficiency related from the MARD subgroup. Clinicians might provide the more proper treatment, like using insulin sensitizers or insulin secretagogues, for these individuals instead of ascribing their conditions to age.

However, there were still some limitations in our study. First, there were no absolute thresholds for HbA_1c and age in our study. Our eligibility criteria were participants who were diagnosed with DM and who had an HbA_1c level of at least 6.5%. Although HbA_1c is an objective laboratory parameter, it has been shown to fluctuate from 8.7 to 8.1 mmol/mol (6.8%) over 12 weeks when patients start to take DM medicine (42). Thus, when we enrolled the participants, the cross-sectional HbA_1c value was affected by individual adherence to medication. In addition, HbA_1c levels were more difficult to assess in our subgroups of insulin resistance, insulin deficiency, and autoimmune disease. It is more appropriate to measure individual adherence to medication. Second, we observed that there was a relatively older age in our sample. It was difficult to separate eligible participants by age. Third, our optimal cutoff value for GAD-65 Abs was calculated from the NHANES III data set, but there were some patients for whom these laboratory data were not available. Moreover, there were relatively large number of exclusions because of missing data on age, sex, medication use condition, GAD-65 Abs, and HOMA2. There were also low event numbers for CVD, all-cause, and cancer-related deaths, especially in the MOD subgroups. Individuals in the MOD subgroup were only moderately obese but did not have severe insulin deficiency and severe insulin resistance as assessed by HOMA-IR and HOMA-β. Perhaps a study including GAD-65 Abs, HOMA-IR, and HOMA-β laboratory data for all patients could be conducted to confirm our results in the future. Finally, this is a retrospective study, and further prospective studies are needed to support the findings.

In conclusion, we have demonstrated that compared to the MOD subgroup, the MARD subgroup from the NHANES III data set tended to have higher CVD-related mortality than the other subgroups. On the other hand, all-cause and cancer-related mortality did not differ among the DM subgroups. In addition, compared to the MARD subgroup, the SAID and SIDD subgroups had a higher retinopathy risk, but there was no difference in the risk of nephropathy among the subgroups.

Abbreviations

Abbreviations
 
• ADOPT

  A Diabetes Outcome Progression Trial

 
• BMI

  body mass index

 
• CDC

  Centers for Disease Control and Prevention

 
• CHD

  coronary heart disease

 
• CVD

  cardiovascular disease

 
• DM

  diabetes mellitus

 
• DR

  diabetic retinopathy

 
• eGFR

  estimated glomerular filtration rate

 
• GAD-65 Ab

  glutamate decarboxylase 65 antibody

 
• HbA_1c

  glycated hemoglobin

 
• HOMA-β

  homeostasis model assessment of β-cell function

 
• HOMA-IR

  homeostasis model assessment of insulin resistance

 
• HOMA2-β

  updated homeostasis model assessment of β-cell function

 
• HOMA2-IR

  updated homeostasis model assessment of insulin resistance

 
• HRs

  hazard ratios

 
• MARD

  mild age-related diabetes

 
• MDRD

  Modification of Diet in Renal Disease

 
• MHNO

  metabolically healthy nonobese

 
• MHO

  metabolically healthy obese

 
• MI

  myocardial infarction

 
• MOD

  mild obesity-related diabetes

 
• NCHS

  National Center for Health Statistics

 
• NDI

  National Death Index

 
• NHANES

  National Health and Nutrition Examination Survey

 
• ORs

  odds ratios

 
• SAID

  severe autoimmune diabetes

 
• SIDD

  severe insulin-deficient diabetes

 
• SIRD

  severe insulin-resistant diabetes

 
• T2DM

  type 2 diabetes mellitus

 
• UKPDS

  U.K. Prospective Diabetes Study

Acknowledgments

Financial Support: No financial support was used for this study.

Author Contributions: Peng-Fei Li contributed to the design of the study, was responsible for the management and retrieval of data, contributed to initial data analysis and interpretation, and drafted the initial manuscript. Peng-Fei Li and Wei-Liang Chen decided on the data collection methods and were responsible for the data analysis decisions. Wei-Liang Chen conceptualized and designed the study, supervised all aspects of the study, critically reviewed and revised the manuscript, and approved the final manuscript as submitted. Both authors read and approved the final manuscript.

Additional Information

Disclosure Statement: P.-F. L. and W.-L.C. declare that they have no conflict of interest.

Data Availability

The data that support the findings of this study are available from the corresponding author, upon reasonable request.