Insulin Therapy in Patients with Type 2 Diabetes and High Insulin Resistance is Associated with Increased Risk of Mortality, Major Adverse Cardiovascular Events, and Diabetic Kidney Disease

Abstract

Objective:

To investigate the relationship between insulin use and clinical outcomes in patients with type 2 diabetes stratified by level of insulin resistance (IR).

Methods:

Cross sectional analysis of the NHANES database from 2001 to 2010. Sample was comprised of 3,124 individuals with diabetes, representing a US population of 16,713,593. Insulin use was self-reported. Fasting glucose and insulin levels were used to assess IR by HOMA-IR determination. Subjects were allocated within High or Low HOMA-IR groups based on the sample median. Outcome variables were mortality, major adverse cardiovascular events (MACE), and diabetic kidney disease (DKD). Logistic regression adjusting for covariates including glycemic control and comorbidities were performed.

Results:

In the adjusted model, insulin use was significantly associated with increased risk of mortality (OR: 2.39, 95% CI: 1.136 – 5.010) having a MACE (OR: 2.45, 95% CI: 1.137 – 4.550), and developing DKD (OR: 1.89, 95% CI: 1.119 – 3.198) in the high HOMA-IR group. The association between insulin use and the outcome variables was not statistically significant in patients within the low HOMA-IR group.

Conclusions:

Insulin use was associated with increased risk of mortality, MACE, and DKD in patients within the high IR group, but the association was not significant within the low IR group. Our findings indicate that insulin therapy could be less beneficial in patients with high IR. Prospective studies are needed to identify subsets of individuals with type 2 diabetes who would benefit the most from insulin therapy, and for which patients, insulin should be avoided.

Introduction

Type 2 diabetes is a progressive condition in which an insulin secretory defect develops, frequently on the setting of an underlying component of insulin resistance. [1]It is widely accepted that the presence of background insulin resistance, attributed mainly to excess nutrients and energy balance, will lead to relative insulin deficiency, which will play a key role in disease progression of type 2 diabetes [2]. In clinical practice however, diversity in the presentation of patients with type 2 diabetes is common. Some patients may present with features of significant insulin deficiency and high insulin sensitivity, whereas for others with severe insulin resistance, large insulin doses may be necessary [3]. These differences in presentation could be due to the fact that insulin resistance and beta cell dysfunction vary widely in patients with type 2 diabetes [4, 5], and each of these defects has been identified as independent predictor of progression of the disease [6].

Insulin therapy is broadly recommended and used to achieve glycemic control in patients with type 2 diabetes who are not meeting recommended goals [7] The efficacy and safety of insulin therapy has been evaluated in large prospective studies such as the United Kingdom Prospective Diabetes Study (UKPDS) [8], and more recently in Outcome Reduction with Initial Glargine Intervention trial (Origin trial) [9]. Despite effectively improving glycemic control, insulin therapy has not been shown to significantly decrease mortality or provide cardiovascular benefit [8, 9]. On the contrary, insulin therapy can result in weight gain and hypoglycemia, both risk factors for increased morbidity and mortality [10–13]. Hyperinsulinemia has also been shown to be associated with negative outcomes, such as increased cardiovascular risk [14], either as an independent factor or as part of the metabolic syndrome [15, 16]. Thus, it has been proposed that the use of additional exogenous insulin in the setting of hyperinsulinemia may increase negative effects in patients with type 2 diabetes [17, 18].

Given that decreased insulin sensitivity and reduced beta cell function can co-exist at varying degrees, and that effective non-insulin agents are now available, therapeutic targets tailored to the prevalent defect may be the most appropriate approach of treating patients with type 2 diabetes [19]. Particularly, given that the benefits of insulin therapy in patients with evidence of marked insulin resistance remain unclear [17, 20]. The aim of this study was to understand the relationship between insulin use and clinical outcomes in patients with type 2 diabetes stratified by level of insulin resistance. We performed the analysis using the nationally representative National Health and Nutrition Examination Survey (NHANES) collected from non-institutionalized adults in the United States, and used three outcomes to investigate the relationships: mortality, major adverse cardiovascular events (MACE), and diabetic kidney disease (DKD). We hypothesized that insulin use in individuals with type 2 diabetes and high insulin resistance will be associated with poor clinical outcomes.

Methods

Sample Data

This analysis used 5 survey waves of the National Health and Nutrition Examination survey (NHANES 2001 to 2010) [21]. NHANES is a cross-sectional survey that was developed to help study the effects of nutrition on health and is designed to be representative of the civilian, non-institutionalized U.S. population. Descriptions of survey design for NHANES have been published elsewhere [22]. Each NHANES cycle uses a complex, multistage probability sampling scheme which allows analyses to be representative of the US population. We limited our analysis to a subpopulation of individuals who reported being diagnosed with diabetes, reported whether they used insulin, and had laboratory information on both plasma insulin and glucose levels. Diagnosis of diabetes was determined using the question “Have you ever been told by a doctor or healthcare professional that you have diabetes or sugar diabetes?” Of the 3,124 individuals who reported they had been diagnosed with diabetes, 1,470 had information on whether they used insulin, and laboratory values to allow calculation of HOMA-IR. Demographic and laboratory data was collected for each individual. We then created three analysis samples based on the outcome of interest (see outcomes below).

Insulin Usage

NHANES asked patients if they were currently taking insulin using the question “{Is SP / Are you} now taking insulin”. This was coded as a dichotomous variable where a ‘yes’ response was represented with a 1 and a ‘no’ response was represented with a 0. All other response options (i.e. ‘Refused’) were coded as missing. NHANES does not report the type or dosage of exogenous insulin.

Homeostatic Model Assessment-Insulin Resistance (HOMA-IR)

Laboratory procedures for insulin and glucose collection are presented elsewhere and provided in the NHANES laboratory subsample [23]. Plasma fasting glucose and basal insulin data is reported based on measures taken in the morning. Insulin resistance was calculated using HOMA-IR, and it was calculated by taking the basal glucose multiplied by the basal insulin dividing by 22.5 as described in prior publications [24]. We then dichotomized HOMA-IR as “high” or “low” by using the median of those with diabetes as a cut point (median = 4.55, interquartile range = [2.39–7.91]). NHANES does not report if patients taking insulin made any special modifications to their regimen for fasting samples of glucose or insulin levels.

Covariates

Covariates included gender (dichotomized as male or female), age, race/ethnicity (grouped as white or minority), education (dichotomized as higher education vs. high school or less), marital status (dichotomized as married or living with a partner vs. not married/widowed/divorced), poverty status (dichotomized as low income where the family income to poverty ratio was 128% or lower vs. higher income), and glycosylated hemoglobin (HbA1c) levels (continuous variable). A continuous variable of the count of comorbidities was created. Comorbidities that were included in this count were cancer, arthritis, liver disease, chronic bronchitis, asthma, thyroid disease, hypertension, and obesity. Hypertension was defined as having an average systolic blood pressure equal to or above 140, an average diastolic blood pressure equal to or above 90, or self-reported hypertension. Obesity was defined as having a body mass index (BMI) of 30 kg/m² or higher. All other comorbidities were self-reported.

Outcome Variables

Mortality:

NHANES collects mortality status from the National Death Index (NDI). It restricts the public use data to adults, aged 18 and up, only. The participants were followed for mortality status from baseline until December 31, 2011. For the purpose of this analysis, mortality status was coded as either 1 for having passed away or 0 for being assumed to be still alive. For the analysis of mortality, 1,450 individuals with laboratory data were available for analysis, representing a US population of 16,503,787.

Major Adverse Cardiovascular Events (MACE): A major adverse cardiovascular event (MACE) was coded as a dichotomous variable where a 1 represented answering ‘yes’ to having a previous diagnosis of congestive heart failure, coronary heart disease, angina, heart attack, or a stroke. If the respondent answered ‘no’ to all of those questions, they were coded as 0. For the analysis of MACE, 1,420 individuals with laboratory data were available for analysis, representing a US population of 16,250,875.

Diabetic Kidney Disease:

Participants with DKD were identified by using eGFR and the urine albumin to creatinine ratio (UACR). The eGFR was calculated using the Modification of Diet in Renal Disease (MDRD) 4 variable formula which accounts for race, gender, urine albumin, and urine creatinine and that has been validated in NHANES participants with diabetes in the past [25]. The formula follows the form:

$$\mathit{\text{eGFR}}= 175\ast c{r}^{-1.154}\ast ag{e}^{-0.203}\ast \left\{I\left(\mathit{\text{black}}=1\right)\ast 1.212\right\}\ast \left\{I\left(\mathit{\text{female}}=1\right)\ast 0.742\right\}$$

Where eGFR is measured in ml/min/1.73m² is urine creatinine in μm/L. UACR was calculated by dividing the urine albumin by the urine creatinine levels and was in units of mg/g. DKD was identified by having an eGFR less than 60 ml/min/1.73m² and/or by a UACR of greater than or equal to 30 mg/g. For the analysis of DKD, 1,470 individuals were available for analysis, representing a US population of 16,627,737.

Statistical Analysis

All analyses was done in Stata/SE 15 (StataCorp LLC. Release 15. College Station, TX). Analysis accounted for the cluster sampling design and oversampling for specific subpopulations across all cycles included in the sample and was adjusted for NHANES’ complex survey design using the appropriate weights [26]. Significance of results was determined based on p<0.05.

First, means and frequencies for demographics and laboratory values in each analytic subgroup were determined. Second, the proportion of individuals with and without insulin use were compared by high vs. low HOMA-IR using chi2 for each outcome. Lastly, stratified multiple logistic regression using the maximum likelihood approach were run to examine the relationship between insulin use, insulin resistance, and clinical outcomes. Six unadjusted and six adjusted models were run, two for each outcome. The outcome of the first two models was mortality, the first model was run on individuals categorized as having high HOMA-IR and the second on individuals categorized as having low HOMA-IR. Insulin use was the primary independent variable with adjustment for age, gender, race, marital status, poverty status, education, A1c, and comorbidities. The outcome for the second two models was MACE, the first model was run on individuals categorized as having high HOMA-IR and the second on individuals categorized as having low HOMA-IR. Insulin use was the primary independent variable with adjustment for age, gender, race, marital status, poverty status, education, A1c, and comorbidities. The outcome for the final two models was DKD, the first model was run on individuals categorized as having high HOMA-IR and the second on individuals categorized as having low HOMA-IR. Insulin use was the primary independent variable with adjustment for age, gender, race, marital status, poverty status, education, A1c, and comorbidities. Given the analysis incorporated existing data and the sample size was limited, we ran post-hoc power analyses on non-significant models using the margins of adjusted models with covariates held at the mean, and assuming a p-value of 0.05 to determine if we had sufficient power (80%) to detect a difference in stratified models.

Results

Table 1 provides the weighted sample demographics by outcome for each analytic sample. Mean age was 58, gender was evenly split between men and women, approximately 61% of each sample was white, 63% were married, and 23% had a family income to poverty ratio 128% or lower. Education varied slightly across the three groups with 44.7% having less than a high school degree in the mortality cohort, 45.3% in the MACE cohort, and 55.5% in the CKD cohort. Mean HbA1c was 7.1% across all three groups, comorbidity count was on average 2.4 and approximately 21.1% used insulin.

Table 1:

Sample demographics for each cohort as defined by outcome (mean or proportion with 95% standard error)

	Mortality	MACE	DKD
n	1,450	1,420	1,470
population	16,503,787	16,250,875	16,627,737
Mean Age	58.48 (0.57)	58.62 (0.56)	58.19 (0.58)
Gender
Male	48.24% (1.93)	47.99% (1.93)	48.31% (1.93)
Female	51.76% (1.93)	52.01% (1.93)	51.69% (1.93)
Race
White	61.75% (2.67)	61.76% (2.68)	61.77% (2.64)
Minority	38.25% (2.67)	38.24% (2.68)	38.23% (2.64)
Marital Status
Married	63.40% (1.67)	63.49% (1.72)	63.29% (1.68)
Not Married	36.60% (1.67)	36.51% (1.72)	36.71% (1.68)
Poverty Status
Poor	23.74% (1.55)	23.53% (1.54)	23.68% (1.53)
Not Poor	76.26 % (1.55)	76.47% (1.54)	76.32% (1.53)
Education
> High School	44.73% (2.15)	45.35% (2.18)	55.53% (2.13)
<= High School	55.27% (2.15)	54.65% (2.18)	44.47% (2.13)
Mean HbA1c (% / mmol/mol)	7.18 (0.06) / 55 (0.7)	7.17 (0.06) / 55 (0.7)	7.18 (0.06) / 55 (0.7)
Insulin Usage
% Uses Insulin	21.18% (1.53)	20.94% (1.49)	21.36% (1.54)
% No Insulin	78.82% (1.53)	79.06% (1.49)	78.64% (1.54)
Median HOMA-IR [IQR]	4.55 [2.41–7.90]	4.55 [2.41–7.84]	4.55 [2.39–7.91]
Comorbidity Count	2.43 (0.06)	2.44 (0.07)	2.41 (0.06)

HbA1c = glycosylated hemoglobin, HOMA-IR = insulin resistance

Table 2 shows the proportions of individuals in each insulin use category for high vs. low insulin resistance. There was a significant difference in mortality between insulin users and non-insulin users within the high HOMA-IR group (χ²: p = 0.04), however there was no significant difference between insulin users and non-insulin users within the low HOMA-IR group (χ²: p = 0.22). A similar pattern was seen in MACE status with a significant difference between insulin users and non-insulin users (χ²: p< 0.01) within the high HOMA-IR group, but no significant difference between insulin users and non-insulin users (χ²: p = 0.30) within the low HOMA-IR group. There were no significant differences between insulin users and non-insulin users for DKD in the low HOMA-IR group but there were significant differences in the high HOMA-IR group (χ² for high HOMA-IR: p< 0.01; χ² for low HOMA-IR: p = 0.25).

Table 2:

Proportions (with 95% CI) of individuals in each insulin use category with high vs. low insulin resistance across three outcomes

		Alive	Dead	MACE	No MACE	Has DKD	No DKD
High HOMA		p=0.04		p < 0.01		p < 0.01
	Insulin	16.95% (13.68 – 20.82)	3.86% (2.32 – 6.36)	8.46% (6.33 – 11.22)	12.34% (9.20 – 16.37)	11.28% (8.35 – 15.08)	9.51% (7.28 – 12.33)
	No Insulin	70.28% (65.92 – 74.29)	8.91% (7.06 – 11.20)	15.99% (12.89 – 19.68)	63.21% (57.86 – 68.26)	28.21% (24.33 – 32.45)	51.00% (46.48 – 55.49)
Low HOMA		p=0.22		p = 0.30		p = 0.25
	Insulin	17.49% (13.31 – 22.64)	4.20% (2.46 – 7.10)	7.77% (5.07 – 11.73)	13.35% (9.10 – 19.16)	10.01% (7.32 – 13.58)	12.11% (7.99 – 17.94)
	No Insulin	67.89% (63.02 – 72.41)	10.41% (7.61 – 14.09)	21.88% (17.39 – 27.15)	57.00% (50.58 – 63.19)	28.12% (24.04 – 32.59)	49.75% (43.85 – 55.66)

Bold text indicates significance at p<0.05 level

Table 3 presents the logistic regression results for the relationship between insulin use and mortality, MACE, and DKD stratified by high and low HOMA-IR. Individuals with diabetes and high HOMA-IR who used insulin were 2.39 times more likely to die compared to those who did not use insulin after adjusting for age, gender, race, marital status, poverty status, education, A1c and comorbidities (OR: 2.39, 95% CI: 1.136 – 5.010). The odds ratio for insulin usage was positive in the low HOMA-IR group, but it was not statistically significant (OR: 2.06, 95% CI: 0.643 – 6.586). Individuals with diabetes and high HOMA-IR who used insulin were 2.45 times more likely to have a MACE compared to those who did not use insulin after adjusting for age, gender, race, marital status, poverty status, education, A1c and comorbidities (OR: 2.45, 95% CI: 1.137 – 4.550). The odds ratio for insulin usage was positive in the low HOMA-IR group, but it was not statistically significant (OR: 2.13, 95% CI: 0.819 – 5.525). Individuals with diabetes and high HOMA-IR who used insulin were 2.45 times more likely to have DKD compared to those who did not use insulin after adjusting for age, gender, race, marital status, poverty status, education, A1c and comorbidities (OR: 1.89, 95% CI: 1.119 – 3.198). The odds ratio for insulin usage was positive in the low HOMA-IR group, but it was not statistically significant (OR: 2.01, 95% CI: 0.991 – 4.096). In post-hoc analyses, the low HOMA-IR models for MACE and DKD each had 80% power, however, the low HOMA-IR model for mortality had 41% power to detect a difference.

Table 3:

Adjusted Odds Ratios for influence of insulin use on Mortality, MACE, and DKD stratified by insulin resistance levels

	Mortality		MACE		DKD
	High HOMA-IR	Low HOMA-IR	High HOMA-IR	Low HOMA-IR	High HOMA-IR	Low HOMA-IR
OR for Insulin Use	2.39*	2.06	2.45**	2.13	1.89*	2.01
95% CI	(1.14 – 5.01)	(0.64 – 6.59)	(1.32 – 4.55)	(0.82 – 5.53)	(1.12 – 3.20)	(0.99 – 4.10)

OR = odds ratio against the reference of no insulin use in the same insulin resistance category

CI = confidence interval,

^*= p<0.05,

^**= p<0.01

Models adjusted for age, gender, race, marital status, poverty status, education, HbA1c, and comorbidities

Discussion

In this study, we aimed to investigate the relationship between insulin use and insulin resistance on clinical outcomes using a nationally representative sample of patients with type 2 diabetes. Results showed that in patients with diabetes, the self-reported use of insulin therapy was significantly associated with increased likelihood of mortality, cardiovascular events, and DKD only in subjects with high insulin resistance (based on the median for patients with diabetes in the sample). The association remained significant after adjusting for demographics, comorbidities, and level of glycemic control (A1c).

To our knowledge, this is the first population based report suggesting that insulin therapy is associated with poor outcomes in patients with type 2 diabetes and high insulin resistance. Insulin use was associated with a greater than two-fold increased risk of mortality, having a MACE, and having DKD in the subgroup of patients with high HOMA-IR, while these associations were not significant in the subjects using insulin within the low HOMA-IR group. While analyses in the MACE and DKD populations had significant power to detect a difference in the low HOMA-IR group, power was low in the mortality population and therefore a difference between the high and low HOMA-IR populations should be taken with caution for that outcome.

Our results are in alignment with other studies suggesting worse outcomes with exogenous insulin use as the association of insulin therapy and increased cardiovascular risk and mortality has previously been described in real-world studies. In a retrospective cohort trial examining The Health Information Network (THIN) data system, Margolis et al. reported an association between exogenous insulin and increased myocardial infarction [27]. Colayco et al. in a nested case-control study of the Kaiser Permanente Southern California (KPSC) data also concluded that patients treated with insulin with or without oral medications had an increased risk (>2.5 folds) of cardiovascular events [28]. Similarly, in a retrospective cohort trial of the UK General Practice Research Database by Currie et al., an association between exogenous insulin (with or without metformin) and increased risk for the primary outcome of that trial (first major adverse cardiac event, first cancer, or mortality) was reported [29].

Despite significantly improving glycemic control, insulin has not consistently shown mortality or cardiovascular benefits in large prospective studies [8, 9]. Conversely, insulin therapy is unquestionably associated with hypoglycemia which is a robust predictor of morbidity and mortality in patients with type 2 diabetes [30–32]. In the UKPDS trial, increased weight gain and risk of hypoglycemia were reported in patients treated with insulin and or sulfonylureas compared to metformin [8]. In a sub-study (UKPDS-13), the use of sulfonylureas and insulin were reported to be significantly associated with increased fasting plasma insulin concentrations and body weight, which was not seen on the diet-controlled group. More hypoglycemic episodes occurred with sulfonylurea or insulin than with diet or metformin [33]. More recently, in the Origin trial, patients with high cardiovascular risk were randomized to receive insulin glargine vs. standard care for a mean duration of 6.2 years. In this study, rates of cardiovascular outcomes were not significantly increased in the insulin-glargine compared to the standard-care groups (hazard ratio, 1.02; 95% confidence interval [CI], 0.94 to 1.11; P=0.63). However, rates of severe hypoglycemia were roughly tripled (1.00 versus 0.31 per 100 person-years) and weight increased by 1.6 kg in the insulin-glargine group, while it decreased by 0.5 kg in the standard-care group.

Several mechanisms supporting the association between hyperinsulinemia, insulin resistance and increased risk of cardiovascular events and all-cause mortality have been proposed [14, 19, 34]. Insulin resistance increases oxidative stress, de-novo formation of dense low-density lipoproteins, and endothelial dysfunction, all contributors of atherosclerosis [35]. While exogenous insulin can reverse glucotoxicity, it can also worsen hyperinsulinemia [19]. Insulin therapy has also been shown to not provide significant reduction of inflammatory markers when compared to metformin or placebo in patients with type 2 diabetes [36]. It has thus been suggested that insulin therapy used to improve glycemic control does not necessarily translate into decreased cardiovascular burden, as hyperglycemia may be more of a marker or a mediator of inflammation rather than the leading factor causing cardiovascular disease [37].

Our study has some important clinical implications. The fact that insulin use was significantly associated with poor outcomes only in the high insulin resistance, but not significantly in the low insulin resistance group, suggests that insulin therapy may not always be beneficial and could be potentially harmful in subjects with already elevated insulin levels; but for subjects with a predominantly insulin secretory defect (presumably lower insulin resistance), insulin use could possibly be safe and appropriate. Thus, amongst patients with type 2 diabetes, further classification to better identify patients and guide clinicians in the management of this complex disease may be necessary as previously suggested [19].

This study has important strengths such as the use of the NHANES dataset which provides estimates representative of the U.S. population. Additionally, we were able to investigate multiple outcomes and covariates within the same population of individuals with diabetes giving consistency to the results. Lastly, the availability of fasting glucose and insulin levels allowed for insulin resistance stratification which was critical to the conception of our study design. However, we acknowledge several limitations. First, data is cross-sectional and thus we cannot conclude on causality. Secondly, while we adjusted for a number of possible confounders, additional unmeasured confounders may exist, such as duration of diabetes, hypoglycemia, insulin type, and insulin dosage as well as medications that impact insulin resistance. Third, we used HOMA-IR – instead of euglycemic insulin clamp – to evaluate insulin resistance. Euglycemic insulin clamp remains the gold standard for evaluating insulin resistance. Fourth, we divided patients by HOMA-IR as having “high” or “low” insulin resistance based on the HOMA-IR median of the sample, resulting in a cut-off higher than those described for subjects without diabetes [38]. Fifth, the sample size was limited despite using multiple waves of data from NHANES. As indicated in the results, the sample size was sufficient for power to find a difference in the MACE and DKD models, however, was limited in the mortality models and therefore should be interpreted with caution. Finally, although information regarding the use of non-insulin medications was available, when separated by drug class, the sample size for individual groups prohibited use in the models.

In conclusion, in this study of a nationally representative sample, we found that that self-reported insulin use was associated with increased risk of mortality, cardiovascular events, and DKD in patients with type 2 diabetes categorized as having high insulin resistance. This association was not significant in patients within the low insulin resistance group. These findings suggest that insulin therapy may be less beneficial and potentially harmful for patients with high insulin resistance, while its use could be adequate and safe in those patients with a predominantly insulin secretory defect. In this era of newer non-insulin therapies, HOMA-IR may be an underutilized test that could help better identify which patients with type 2 diabetes would benefit the most from insulin therapy, and for which patients, insulin should be avoided if possible. Future prospective studies evaluating the use of non-insulin therapies in patients with high insulin resistance are needed to confirm that these therapies can safely be used without resulting in deterioration of glycemic control.