Association of long-term benzodiazepine hypnotic use and prediabetes in US population: A cross-sectional analysis of national health and nutrition examination survey data

Abstract

Benzodiazepine hypnotics’ effects on glucose metabolism are seldom reported, and the association between long-term (≥4 weeks) benzodiazepine usage and prediabetes has not been studied. This study was aimed to investigate the association between benzodiazepine hypnotic usage for ≥ 3 months and the prevalence of prediabetes. We analyzed cross-sectional data from the National Health and Nutrition Examination Survey (NHANES) during 2005 to 2008, selecting adult participants without diabetes who used benzodiazepine hypnotics for at least 3 months or did not take any hypnotics. Individuals taking other hypnotics, antipsychotics, glucocorticoids, or hypoglycemic drugs were excluded. We defined prediabetes as an hemoglobin A1C (HbA1C) 5.7–6.4%, as suggested by the American Diabetes Association. Prescribed drug information was self-reported and checked by official interviewers, and HbA1C data in NHANES was recognized by the National Glycohemoglobin Standardization Program. We calculated the propensity score according to the covariates and adjusted it using multivariate logistic regression. Lower thresholds of HbA1C ≥ 5.5% or ≥ 5.3% were also analyzed. Among 4694 eligible participants, 38 received benzodiazepine hypnotics; using these hypnotics for ≥ 3 months was not significantly associated with the prevalence of prediabetes, as well as HbA1C ≥ 5.5% or ≥ 5.3%. Adjusted for propensity score, the respective odds ratios for prediabetes, HbA1C ≥ 5.5%, and HbA1C ≥ 5.3% were 1.09 (95% confidence interval [CI] 0.19–6.32), 0.83 (95% CI 0.22–3.13), and 1.22 (95% CI 0.3–4.93). No significant association was found between benzodiazepine hypnotic usage ≥ 3 months and the prevalence of prediabetes.

1. Introduction

Benzodiazepine (BZ) hypnotics approved by the US Food and Drug Administration (FDA) include alprazolam, diazepam, temazepam, and others. They maintain their therapeutic edge due to their stable and potent effects when compared to newer hypnotics, such as non-BZ hypnotics (e.g., eszopiclone, zolpidem as so-called “Z drugs”), orexin receptor antagonists (e.g., suvorexant), and melatonin receptor agonists (e.g., ramelteon).^[1] Despite this, the usage of BZ hypnotics is associated with more severe side effects, including their effect on glucose metabolism,^[2] and evidence of harm that arises from their use remains insufficient, with many trials being limited to 4 weeks or lacking harm-related reports.^[1] This situation raises concerns regarding long-term use and may limit certain necessary prescriptions involving BZ hypnotics.

A 2014 study reported that the use of brotizolam for 2 weeks results in increased oral glucose tolerance test (OGTT) values without significant insulin level changes, indicating a potential effect on insulin sensitivity and/or secretion.^[3] Additional research is necessary to confirm whether long-term use of BZ hypnotics increases the risk of developing prediabetes.

Hemoglobin A1C (HbA1C) reflects average blood glucose levels over a period of 3 months and can reliably reflect long-term cumulative changes involving glucose metabolism. Prediabetes is a condition that occurs when individuals have higher glucose levels than normal but do not meet the diagnostic criteria for diabetes. HbA1C-defined prediabetes has been reported to predict the subsequent incidence of diabetes. The American Diabetes Association suggests defining prediabetes as having an HbA1C level of 5.7% to 6.4%.^[4]

As individuals with diabetes are more likely to develop sleep disorders and take hypnotics, which may introduce reverse causation into our study, we focused on individuals without diabetes.

In this cross-sectional study, we aimed to investigate the association between using BZ hypnotics for >≥3 months and the prevalence of prediabetes in the US population using data from the National Health and Nutrition Examination Survey (NHANES).

2. Methods

2.1. Data sources

NHANES employs stratified multistage probability cluster sampling to assess the health and nutritional status of the noninstitutionalized US population through a nationally representative survey conducted by the National Center for Health Statistics.(Supplementary file “Link to NHANES data,” http://links.lww.com/MD/K622) Only the NHANES 2005 to 2008 cycles included information concerning the self-reported frequency of hypnotic medication usage and sleep disorders. Deidentified data for participants aged > 20 years were extracted from these cycles, as the youngest BZ hypnotic user was 27 years old. The NHANES study protocol was approved by the National Center for Health Statistics Research Ethics Review Board and all participants provided written informed consent at enrollment.

2.2. Study design and population

Our analyses were based on cross-sectional data collected from participants during 2 2-year NHANES cycles (2005–2006 and 2007–2008). To restrict our samples to individuals without diabetes, we excluded patients with diabetes recognized according to self-reporting or diagnosed based on HbA1C ≥ 6.5%. As we only aimed to include participants using BZ hypnotics for ≥ 3 months or not using any hypnotics, we excluded candidates with answers of < 5 times per month to the question: “How often do you take pills to help you sleep,” and answers of < 90 days to “Number of days taken medicine.” To minimize potential interference of concurrent medications on blood glucose levels, we excluded participants reporting the use of drugs mainly affecting the central nervous system (CNS), systemic glucocorticoids, or hypoglycemic drugs in the previous month. The excluded CNS drugs included hypnotics apart from BZs, BZs which were not approved as hypnotics by the FDA, antipsychotics, antidepressants, antianxiety drugs, antiepileptics, CNS stimulants, and antiparkinsonian drugs in the previous month. Names of excluded drugs are listed in Supplementary Table1, http://links.lww.com/MD/K479. Prescription data were recorded and confirmed by official interviewers. A flowchart of participant enrollment is presented in Figure 1.

Figure 1.

Flow diagram of the screening and enrollment of study participants. HbA1C = hemoglobin A1C, HDL = high-density lipoprotein.

2.3. Assessment of prediabetes

We defined prediabetes as HbA1C within 5.7–6.4%, as suggested by the American Diabetes Association.^[4] Blood samples for HbA1C analysis were drawn in the mobile examination center. HbA1C data from NHANES were recognized by the National Glycohemoglobin Standardization Program. As Gramaglia study reported that elevated OGTT results do not reach the criterion for impaired glucose tolerance, we also analyzed lower thresholds of HbA1C ≥ 5.5% and ≥ 5.3%.

2.4. Covariates

Based on the literature, the following covariates were included: age; sex; ethnicity; educational level; family income–poverty ratio; marital status; smoking exposure; alcohol usage in the previous year; body mass index; physical activity; sleep disorder; sleep duration; energy intake per day; sugar intake per day; total fat intake per day; hypertension; major depressive disorders; family history of diabetes; asthma; cardio–cerebrovascular disease; and high-density lipoprotein cholesterol (HDL-C) < 35 mg/dL.^[4–7] Age, sex, ethnicity, educational level, family income–poverty ratio, and marital status were self-reported. In NHANES, the family income–poverty ratio was recorded as “5” if > 5. Similarly, age was recorded as “80” if > 80 years. Marital status was categorized into 3 groups: married/living with a partner, widowed/divorced/separated, and never married. Educational level was divided into 4 categories (less than high school, high school or equivalent, college or Associate of Arts degree, and college graduate or above). Smoking exposure was divided into the following 3 groups according to serum cotinine levels: nonsmokers (<0.015 ng/mL), passive smokers (0.015 ≤ level < 10 ng/mL), and active smokers (≤10 ng/mL).^[8] Alcohol drinking status was determined by the question, “Have at least 12 alcohol drinks in a year?” Participants who answered “yes” were classified as alcohol drinkers. Hypertension was defined as using hypertensive drugs or systolic blood pressure ≥ 140 mm Hg or diastolic pressure ≥ 90 mm Hg. Body mass index was calculated based on height and weight measurements; self-reported height and weight were used when measurements were not available. Physical activity was categorized into the following 2 groups: yes or no, derived from questions, “Vigorous activity over past 30 days” and “Moderate activity over past 30 days” in the 2005 to 2006 cycle, and “Vigorous work activity,” “Moderate work activity,” “Walk or bicycle,” and “Vigorous recreational activities” and “Moderate recreational activities” in the 2007 to 2008 cycle (affirmative answers to any of the former questions were treated as “yes”). Major depressive disorder was defined as a score of Patient Health Questionnaire-9 ≥ 10.^[9] Family history of diabetes was acquired from the question, “Close relative had diabetes?” Cardio-cerebrovascular disease was defined as an affirmative answer to any of the following questions, “Ever told you had coronary heart disease,” “Ever told you had angina/angina pectoris,” “Ever told you had a heart attack,” and “Ever told you had a stroke?” Asthma was defined as affirmative answers to the following 2 questions: “Ever been told you have asthma?” and “Still have asthma?” Sleep duration was self-reported. Sleep disorder was defined as an affirmative answer to the question, “Have you been told by a doctor about a sleep disorder? (sleep apnea, insomnia, restless legs)”; answer “never/rarely” to the questions, “How often do you snore,” “How often do you snort/stop breathing?”; or answer “never/rarely/seldom” for the following questions, “How often have trouble falling asleep,” “How often do you wake up during the night,” “How often wake up too early in the morning,” and “How often feel unrested during the day?” Total energy/fat/sugar intake was determined from the first 24-hour dietary recall interview.

2.5. Statistical analyses

According to NHANES analytical guidelines, our analyses accounted for a complex sampling design and sampling weights. The sampling weight was calculated using the following formula: Dietary day 1 sample 4-year weight = fasting subsample 2-year mobile examination center weight/2. Characteristics of participants were described as means (standard error, SE) for continuous variables with normal distribution, median (interquartile range) for continuous variables with nonnormal distribution, and weighted frequencies (percentages, %) for categorical variables. Continuous data with normal and non-normal distributions were compared using Student t test and Kruskal–Wallis tests, respectively. Categorical data were compared using the Chi-squared test. As the percentage of missing data was small (the missing rate varied from 0% to 8.9%) for any variable, no imputation method was used.

Odds ratios (ORs) and 95% confidence intervals (CIs) were calculated to assess the association between BZ hypnotic usage and prediabetes (as well as HbA1C ≥ 5.5% and HbA1C ≥ 5.3%) using a logistic regression model. Considering that the sample size of the BZ user group was small (n = 38), we used a propensity score to adjust the regression model instead of inputting all covariates directly. The propensity score for taking BZ hypnotics was estimated using a multivariable logistic regression model containing all the aforementioned covariates. Propensity score distributions according to BZ hypnotic usage are shown in Supplementary Materials, http://links.lww.com/MD/K621.

All statistical analyses were performed with R, v.4.2.3 (R Project for Statistical Computing, Vienna, Austria) using the survey package, v.4.1–1, and Free Statistics analysis platform, v.1.8 (Free Software Foundation, Beijing, China). For all tests, P < .05 (2-sided) was considered to indicate statistical significance.

2.6. Sensitivity analyses

Given that patients prescribed BZ were more likely to be older and have more comorbidities, a sensitivity analysis was performed with overlapping propensity score weighting to identify potential confounding factors. In overlap weighting, each patient weight is the probability that the patient is assigned to the opposite group so that the distribution of covariates is approximately equal among both groups.

As the frequency of BZ usage varied from 5 to 30 times per month, we restricted the analyses to BZ users reporting a frequency of 16–30 times per month compared with unchanged controls, and propensity score-adjusted regression analysis was performed.

Given that the propensity score was derived simultaneously from all covariates, we also applied a multimodel multivariate logistic regression analysis, which input the covariates step-by-step.

As long-acting BZs of clonazepam, diazepam and lorazepam may be used as antidepressants rather than hypnotics, we excluded the individuals using the aforementioned long-acting BZs and performed a propensity score-adjusted regression analysis.

3. Results

3.1. Characteristics of participants

Of the 10,914 participants aged > 20 years, 3860 were excluded due to taking sleeping pills and their frequency ≤ 5 times per month, taking other hypnotics (including BZs not approved as hypnotics by the FDA), antipsychotics, antidepressants, antianxiety drugs, antiepileptics, CNS stimulants, antiparkinsonian drugs, systemic glucocorticoids, or hypoglycemic drugs in previous month, or having self-reported diabetes, or diabetes diagnosed with HbA1C, and 2363 were excluded for missing data, leaving 4691 participants for analyses. The details are shown in Figure 1.

Overall, 38 participants reported BZ hypnotic usage for ≥ 3 months and > 5 times per month.

Among BZ hypnotic users, 17 used alprazolam, 1 used triazolam, 3 used diazepam, 5 used temazepam, 4 used clonazepam, 2 used oxazepam, and 6 used lorazepam. The duration and frequency are shown in Supplementary Table 2, http://links.lww.com/MD/K480.

A total of 38 participants had used BZ hypnotics in the previous month (0.81%, weighted), and 4653 had not (99.19%, weighted). Based on the weighted analyses, the mean age of the 4691 participants was 43.53 years (SE, 0.50 years) and 2209 were female (47.09%, weighted). Based on HbA1C ≥ 5.7%, 1060 (22.60%, weighted) were classified as having prediabetes. The prevalence of prediabetes was higher among participants who used BZ hypnotics (16 [41.93%]) than among those who did not (1044 [18.42%]) (Table 1). BZ hypnotic users were older (60.73 [SE, 2.16] years) than nonusers (43.39 [SE, 0.50] years), more likely to be female (25 [75.91%] vs 2184 [46.12%]), divorced/separated/widowed (14 [43.51%] vs 756 [14.44%]), have hypertension (22 [56.41%] vs 1258 [24.53%]), have major depressive disorder (7 [19.87%] vs 216 [3.90%]), have lower energy intake per day (1651 [1253–2733] vs 2083 [1524–2831]) and more likely to have cardio-cerebrovascular diseases (8 [19.12%] vs 272 [4.49%]).

Table 1.

Weighted contrast of the study population characteristics across groups divided by benzodiazepine (BZ) hypnotics use.

Characteristic	Participants*			P value
	Total	Non-BZ hypnotics user	BZ hypnotics user
	(N = 4694)	(N = 4653)	(N = 38)
Age (yr), mean (SE)	43.54 (0.51)	43.39 (0.50)	60.73 (2.16)	.002
Female (%)	2211 (47.1)	2184 (46.12)	25 (75.91)	<.001
Ethnicity, No. (%)				.24
Non-Hispanic White	2262 (48.19)	2230 (70.89)	30 (88.23)
Non-Hispanic Black	950 (20.24)	946 (10.50)	4 (7.27)
Mexican American	951 (20.26)	949 (9.26)	2 (2.10)
Other Hispanic	342 (7.29)	339 (3.93)	2 (2.40)
Other Race - Including Multi-Racial	189 (4.03)	189 (5.42)	0 (0.00)
Educational Level, No. (%)				.45
Less than high school	1236 (26.33)	1229 (26.41)	7 (10.26)
High School or Equivalent	1138 (24.24)	1126 (24.18)	12 (33.70)
Some College or AA degree	1290 (27.48)	1275 (29.66)	14 (35.25)
College Graduate or above	1030 (21.94)	1023 (28.80)	5 (20.79)
Marital Status, No. (%)				<.001
Married/Living with Partner	3109 (66.23)	3087 (67.27)	20 (48.01)
Divorced/Separated/Widowed	771 (16.43)	756 (14.44)	14 (43.51)
Never Married	814 (17.34)	810 (18.29)	4 (8.48)
Family Income-poverty Ratio: median (IQR)	3.26 (1.68,5.00)	3.26 (1.67,5.00)	2.14 (1.96,4.76)	.55
Smoking Exposure, No (%)				.91
Non-smoker	803 (17.11)	794 (16.24)	9 (19.57)
Passive Smoker	2667 (56.82)	2647 (56.19)	18 (54.32)
Active Smoker	1224 (26.08)	1212 (27.57)	11 (26.10)
Alcohol Use Last Year, No. (%)	3387 (72.16)	3355 (76.99)	30 (86.21)	.11
BMI, median (IQR)	26.98 (23.90,31.23)	26.98 (23.89,31.24)	26.89 (23.35,31.91)	.91
Physical Active, No. (%)	3320 (70.73)	3292 (75.85)	26 (68.96)	.37
Sleep Disorder, No. (%)	2716 (57.86)	2686 (57.81)	27 (66.84)	.44
Sleep Duration (hr), mean (SE)	6.91 (0.03)	6.90 (0.03)	6.94 (0.27)	.89
Energy Intake per day (kCal), median (IQR)	2076.00 (1524.00,2831.00)	2083.00 (1530.00,2831.00)	1651.00 (1253.00,2733.00)	.10
Sugar Intake per day (g), median (IQR)	106.72 (68.37,159.79)	106.72 (68.37,159.80)	105.70 (66.98,139.39)	.45
Total Fat per day (g), median (IQR)	76.91 (52.37,110.21)	77.07 (52.46,110.21)	61.14 (40.52, 88.90)	.18
Hypertension, No(%)	1281 (27.29)	1258 (24.53)	22 (56.41)	.004
Major Depressive Disorder, No (%)	223 (4.75)	216 (3.90)	7 (19.87)	.004
Family History of Diabetes, No (%)	1777 (37.86)	1763 (35.29)	14 (29.68)	.44
Asthma	285 (6.07)	281 (5.92)	4 (9.87)	.2
Cardio-cerebrovascular Disease, No (%)	280 (5.97)	272 (4.49)	8 (19.82)	<.001
HDL < 35 mg/dL, No (%)	365 (7.78)	361 (7.73)	4 (8.17)	.92
Prediabetes†, No (%)	1060 (22.58)	1044 (18.42)	16 (41.93)	.03

BMI = body mass index (calculated as weight in kilograms divided by height in meters squared), BZ = benzodiazepine, HDL = high-density lipoprotein, IQR = interquartile range, SE = standard error.

^*Categorical data are presented as unweighted frequency (weighted percentage), and continuous data are presented as weighted mean (SE) for normal distribution and weighted median (interquartile range) for non-normal distribution.

^†Prediabetes was defined as HbA1C ≥ 5.7%.

3.2. Propensity score-adjusted regression analyses

The results of propensity score-adjusted regression analysis are shown in Figure 2. Association between BZ hypnotics and prediabetes was found to be insignificant (OR, 1.28; 95% CI, 0.21–7.88), as well as the association between BZ hypnotics and HbA1C ≥ 5.5% or HbA1C ≥ 5.3% (OR 1.08 [95% CI, 0.29–3.95] and OR 1.36 [95% CI, 0.34–5.45] respectively).

Figure 2.

Association between benzodiazepine hypnotics use (>3 mo) and prediabetes (or HbA1C ≥ 5.5%, or HbA1C ≥ 5.3%) in propensity score-adjusted logistic regression analyses in NHANES 2005 to 2008 cycles. The propensity score was calculated according to age, sex, ethnicity, educational level, family income-poverty ratio, marital status, smoking exposure, Alcohol use last year, BMI, physical activity, sleep disorder, sleep duration, energy intake per day, sugar intake per day, total fat intake per day, hypertension, major depressive disorder, close relatives with diabetes, asthma, cardio-cerebrovascular disease, HDL. BMI = body mass index, NHANES = National Health and Nutrition Examination Survey.

3.3. Sensitivity analyses

Supplementary Figure1, http://links.lww.com/MD/K476 shows the results of overlap propensity score-weighting analyses in which the association of BZ hypnotics and prediabetes (or HbA1C ≥ 5.5% or ≥ 5.3%) was also insignificant (OR 1.16 [95% CI, 0.45–2.99], OR 1.05 [95% CI, 0.41–2.69], and OR 1.20 [95% CI, 0.40–3.61] respectively).

The results of analyses restricted to BZ users reporting frequency of 16 to 30 times per month (n = 28) remained consistent with primary analyses (see Supplementary Figure2, http://links.lww.com/MD/K477, OR 1.55 [95% CI, 0.29–8.19], OR 0.83 [95% CI, 0.22–3.13], and OR 1.22 [95% CI, 0.30–4.93], respectively, for HbA1C ≥ 5.7%, ≥5.5%, and ≥ 5.3%).

After excluding the individuals using long-acting BZs, we found the association between BZ hypnotics and prediabetes still insignificant (n = 25, see Supplementary Figure3, http://links.lww.com/MD/K478).

Multimodel multivariate regression analysis also revealed an insignificant association between BZ hypnotics and prediabetes (see Supplementary Table3, http://links.lww.com/MD/K486).

4. Discussion

In this cross-sectional study based on nationwide sampling, the association between long-term BZ hypnotic usage and prediabetes in the US population in the propensity score-adjusted model was found to be insignificant. The findings remained reliable after sensitivity analyses.

To our knowledge, Gramaglia study in 2014 showed short-term (2 weeks) usage of brotizolam (0.25 mg/day) in healthy adults (mean age 38.3 years, n = 12) increased the glucose delta area under the curve in OGTT.^[3] However, the association between long-term usage of BZ hypnotics and glucose metabolism impairment has not been investigated to date. This is the first study focused on the blank area. Specifically, we included BZ hypnotic users who took the drug for ≥ 3 months.

The negative results of our study appear to contradict those of the aforementioned short-term study, which indicated the long-term safety of BZ toward the pancreatic islets. Although the acute administration of BZ hypnotics increased OGTT results in healthy adults in Gramaglia study, the elevated levels did not meet the criterion of impaired glucose tolerance.^[3]

Additionally, previous studies concerning BZ use in healthy adults reported a low risk of hyperglycemia. Short and intermediate-acting benzodiazepines (BZs) such as alprazolam and oxazepam have not been associated with elevated blood glucose levels according to related reports. Alprazolam has been found to blunt physiologic counterregulatory responses to hypoglycemia, including neuroendocrine (growth hormone, glucagon), autonomic (mainly epinephrine), and metabolic (lipolysis, glycogenolysis) responses without affecting that result of blood glucose regulation.^[10–13] In a randomized controlled trial, alprazolam was found to improve glycemic control in diabetic patients with anxiety, an effect not mediated by its anxiolytic actions.^[14] Temazepam was found not to alter blood glucose levels in rats across a variety of dosages.^[15]

In previous clinical studies on long-acting benzodiazepines (BZs), reports of significant effects on glucose metabolism have been rare, with only 1 case of elevated blood glucose associated with hepatic injury noted,^[16] and no impact reported on glycemic control when used for diabetic neuropathy such as tardive dyskinesia, stiff leg syndrome, and nonketotic hyperglycemic chorea.^[17–19] In a clinical trial, Chevassus infused clonazepam to healthy adults and applied an intravenous glucose tolerance test between 15 and 60 minutes after infusion, reporting an insignificant difference in results of glucose tolerance in contrast with placebo.^[20] Animal studies have also shown that diazepam and clonazepam have protective effects against oxidative stress in diabetic depression models.^[21–23] High doses of clonazepam (5 mg/kg/day) can increase fasting blood glucose in experimental animals in a manner dependent on hyperglycemic hormones including epinephrine, similar to diazepam.^[24,25] However, diazepam at a lower dose (0.6mg/kg/day, still above therapeutic range) did not affect glucose when administered to rabbits for 1 month.^[26]

Additionally, while long-acting BZs are not specifically used to treat insomnia but rather anxiety, they are often utilized for insomnia associated with anxious states.^[27] Previous human studies employed short-acting BZs, which may differ in side effect profiles from long-acting BZs. Therefore, a sensitivity analysis was performed in this study excluding long-acting BZ users, and the results remained robust.

Similar scenarios can be observed in the literature, in which a psychotic drug was reported to be harmful to glucose metabolism in short-term studies and safe in long-term studies. An example of this is iloperidone.^[28,29]

BZs are GABA-A receptor-positive allosteric modulators that enhance the effects of agonists. Existing studies surrounding GABA-A receptors (GABAAR) have demonstrated certain benefits of BZ for pancreatic islets, although they are still inadequate to clarify the effects of BZ on islet cells.

Previous studies have shown that GABAAR exerts protective effects in terms of islet function. Oral muscimol solution or subcutaneous microinjection of muscimol (GABAAR agonist) solution for 14 days in streptozotocin-induced apoptotic islet model mice increased the survival rate of pancreatic β cells compared with the blank control, and enhanced their cell replication and functional recovery. Similar results were replicated with human islet β cells.^[30] Alprazolam also enhances the inhibitory effects of GABA on islet apoptosis and promotes islet cell replication.^[31]

Although some researchers have inferred from the high concentration of GABA and GABAAR in islets that the GABA–GABAAR system plays a role in regulating endocrine function within the islets.^[32] it is difficult to speculate definitively on the direction of regulation of GABAAR based on current knowledge.^[33] The results of various studies differ owing to different experimental conditions (e.g., human vs other mammals, in vitro vs in vivo, organs vs cells, and drug dosage), leading to varied conclusions regarding GABAAR effects on islet hormone secretion.^[32,34–37]

Hence, the antiapoptotic and recovery-promoting effects of BZ on islets may neutralize its harmful effects in terms of glucose tolerance in long-term usage.

The strengths of this study lie in the efforts to exclude confounding factors as much as possible, such as excluding individuals taking medications affecting blood glucose and the CNS, adjusting for risk factors like depression, hypertension, sleep disorders, and obstructive sleep apnea syndrome.^[38–41] Since many patients take non-BZ hypnotics (“Z-drugs”) concurrently or alternating with BZs,^[42] and these people may be taking medications for pain-associated sleep disturbances,^[43,44] where chronic non-neuropathic pain has been found to be associated with diabetes, this can introduce confounding.^[45] As Z-drugs have previously been observed to increase diabetes risk in people with non-apnea sleep disorders,^[42] this study only included patients taking 1 type of benzodiazepine in the past month to avoid confounding.

4.1. Limitations

This study had certain limitations. First, the frequency of BZ use among the included BZ users ranged from 5–30 times per month, which may have undermined the robustness of our results. In sensitivity analyses restricted to those using BZ 16 to 30 times/month compared to controls, the results remained consistent with the primary analysis.

Second, the data were collected between 2005 and 2008. We chose to analyze the data from 2005 to 2008 because the frequency of BZ use and information concerning sleep disorders were only recorded in investigation cycles of that period. To ensure the present-day generalizability of our conclusions, we examined prescription data from the 2017 to 2020 cycles of NHANES and found that the prescribed BZ hypnotics consisted of clonazepam, lorazepam, oxazepam, temazepam, and alprazolam, which highly overlapped with those of the 2005 to 2008 cycles.

Last, the findings need asserting by longitudinal data, which show dose adjustments and drug discontinuation related fluctuations. Well-designed cohort studies are warranted.

5. Conclusions

No significant association was found between long-term (≥3 months) BZ use and the prevalence of prediabetes (HbA1C ≥ 5.7%) or lower thresholds (HbA1C ≥ 5.5%, HbA1C ≥ 5.3%). Our finding may be important for clinicians to consider whether to prescribe long-term BZ hypnotics for patients without diabetes. Further longitudinal research is warranted to confirm the causal effect of BZ hypnotics on prediabetes.

Acknowledgments

We thank Jie Liu, M.D. (Department of Vascular and Endovascular Surgery, Chinese PLA General Hospital) for his helpful comments regarding the manuscript.

Author contributions

Conceptualization: Weizhen Wu.

Data curation: Yizhuo Qiao, Zhe Chen.

Formal analysis: Weizhen Wu.

Investigation: Weizhen Wu, Junning Zhang.

Methodology: Junning Zhang, Yan Fu.

Project administration: Zhixu Yang.

Resources: Zhe Chen.

Supervision: Zhixu Yang.

Software: Junning Zhang, Lijiang Ren.

Validation: Yan Fu.

Visualization: Yizhuo Qiao, Lijiang Ren.

Writing – original draft: Weizhen Wu.

Writing – review & editing: Junning Zhang.