Iron Deficiency and Vitamin D Deficiency Are Associated with Sleep in Females of Reproductive Age: An Analysis of NHANES 2005–2018 Data

Abstract

Background

Iron and vitamin D deficiencies have been implicated in sleep disturbance. Although females are more susceptible to these deficiencies and frequently report sleep-related issues, few studies have examined these associations in females.

Objective

This study investigates the association of iron and vitamin D deficiencies on sleep in a nationally representative sample of females of reproductive age.

Methods

We used 2 samples of 20–49-y-old non-pregnant females from National Health and Nutrition Examination Survey (NHANES) 2005–2008 (N = 2497) and NHANES 2005–2010 and 2015–2018 (N = 6731) to examine the associations of iron deficiency (ID), iron deficiency anemia (IDA), vitamin D deficiency (VDD), vitamin D inadequacy (VDI), and the joint association of both deficiencies with sleep duration, latency, and quality. Sleep outcomes were measured using a self-reported questionnaire. We used the body iron model based on serum ferritin and serum soluble transferrin receptor to identify ID, along with hemoglobin to identify IDA cases. In addition, 25-hydroxyvitamin D levels were used to determine VDD and VDI cases. Logistic regression was used to evaluate these associations, adjusting for potential confounders. In addition, we assessed the multiplicative and additive interactions of both deficiencies.

Results

ID and IDA were associated with poor sleep quality, with 1.42 [95% confidence interval (CI): 1.02, 2.00)] and 2.08 (95% CI: 1.29, 3.38) higher odds, respectively, whereas VDD and VDI were significantly associated with short sleep duration, with 1.26 (95% CI: 1.02, 1.54) and 1.22 (95% CI: 1.04, 1.44) higher odds, respectively. Subjects with both nutritional deficiencies had significantly higher odds of poorer sleep quality compared with subjects with neither condition. For sleep quality, a significant multiplicative interaction was observed between ID and VDD (P value = 0.0005). No associations were observed between study exposures and sleep latency.

Conclusions

Among females of reproductive age, iron and vitamin D deficiencies are associated with sleep health outcomes. The potential synergistic effect of both deficiencies warrants further assessment.

Introduction

Sleep plays a central role in human development and maintenance of mental and physical health. A plethora of studies have shown that short sleep duration, defined as ≤6 h/d, is independently and significantly associated with increased risk of all-cause mortality, obesity, hypertension, cardiovascular diseases, stroke, diabetes mellitus, and mood disorders such as depression and anxiety [1,2]. The age-adjusted prevalence of adults reporting short sleep duration (<7 h/d) has remained constant from 2013 to 2020 with ∼34.8% of adults reporting getting less than the recommended amount of sleep in 2020 [3]. While the trends have been steady recently, short sleep duration remains a major public health concern. It is only one aspect of healthy sleep behavior, which is a multidimensional concept encompassing sleep duration, quality, efficiency, timing, and alertness [4]. Although these dimensions are correlated, epidemiological evidence suggested that each sleep dimension can be independently associated with certain health outcomes. For instance, a meta-analysis of randomized controlled trials showed that sleep duration, sleep quality, and sleep timing have a detrimental effect on markers of insulin sensitivity [5]. In addition, sleep dimensions can differ by sex. Males and females tend to have a similar sleep pattern until puberty, when sex differences start to become more pronounced, as females frequently report sleeping less, having poorer sleep quality and efficiency, and difficulties falling asleep [6,7]. Throughout their lifespan, females are more prone to sleep disturbance as their sleep can be influenced by the menstrual cycle, pregnancy, and menopause [6].

Iron deficiency (ID) and vitamin D deficiency (VDD) are also more prevalent in females than in males. Females, particularly at childbearing age, are predisposed to ID and iron deficiency anemia (IDA) given the high iron requirements during this age period along with menstrual blood loss and inadequate consumption of dietary iron [8]. Untreated ID with or without IDA can affect physical performance, work productivity, and cognitive function, and can increase risk of morbidity and mortality in mothers and infants as well as people with chronic inflammatory diseases [9]. Aside from these consequences, ID and IDA are proposed as potential risk factors for poor sleep health. ID and IDA are frequently associated with sleep disorders such as restless leg syndrome (RLS), periodic limb movements, and sleep-disordered breathing as well as sleep disturbances observed in children with attention-deficit/ hyperactivity disorder and autism spectrum disorders [10]. It has been estimated that 1 in 4 patients with IDA suffer from clinically significant RLS symptoms that cause shorter sleep duration and poorer sleep quality [11]. However, limited research has been conducted on healthy individuals, as most of the research examining the link between ID/IDA and sleep has focused on people with sleep disorders or on infancy and early childhood.

In contrast to research on ID/IDA and sleep, VDD and vitamin D inadequacy (VDI) have been extensively studied, and a well-established association with sleep health has been examined in different age groups and clinical conditions [12]. Interestingly, it is not uncommon for ID and VDD to co-exist. Emerging evidence from recent studies shows that vitamin D is associated with iron and anemia. While most of the evidence emerged from animal studies, in humans, the results are inconsistent but in favor of a positive association between vitamin D and iron status [13,14]. Despite the inconsistent literature, the link between vitamin D and iron is plausible and warrants further assessment in relation to females’s sleep. The current study aims to examine the association between ID, IDA, VDD, and VDI, and the joint association of both the deficiencies with sleep in a nationally representative sample of the United States females of reproductive age.

Methods

Study population

This study used nationally representative data from the serial cross-sectional NHANES initiated in the early 1960s to provide data about the health and nutritional status of the United States population. NHANES uses a multistage cluster sampling design and collects data via interviews, self-administered questionnaires, physical examinations, and laboratory data. Details of the design and the data from each survey are available online (https://www.cdc.gov/nchs/nhanes.htm). For this study, we used 2 samples from different NHANES cycles that met the inclusion criteria of non-pregnant females ages 20–49 y and included measures of exposures (ID, IDA, VDD, and VDI) and outcomes (sleep duration, sleep quality, and sleep latency) (Figure 1). A longer, more extensive sleep questionnaire was used in NHANES 2005–2008 cycles and a shorter version was used in NHANES 2009–2018 cycles. The first sample included participants from NHANES 2005–2008 cycles only and was used for sleep quality and sleep latency outcomes, due to the lack of these 2 outcomes in the subsequent cycles. The second sample included participants from NHANES 2005–2010 and 2015–2018 cycles, excluding participants from NHANES 2011–2014 cycles due to lack of iron data, and was used to assess sleep duration (Figure 1). The final sample size that met the inclusion criteria was 2497 females for the first sample and 6713 females for the second sample.

FIGURE 1.

Flowchart of available outcomes and exposures from NHANES 2005–2018 cycles. ID, iron deficiency; IDA, iron deficiency anemia; VDD, vitamin D deficiency; VDI, vitamin D inadequacy.

Iron and vitamin D measures

Blood samples were collected at mobile examination centers (MECs) by trained personnel under controlled environmental conditions. Frozen specimens were shipped to the NHANES biorepository or directly to the analyzing lab. Complete blood counts (CBC) were performed in duplicate for all samples using Beckman Coulter MAXM. Hemoglobin was measured as part of the CBC. Serum ferritin and serum soluble transferrin receptors were used as ID indicators, as these tests can be used as an indicative measure for iron storage and transportation in the body, respectively. Different assay methods were used during NHANES 2005–2018 to measure serum ferritin and serum soluble transferrin receptors. In 2005–2008, the Roche/Hitachi 912 clinical analyzer was used to measure serum ferritin concentrations which was changed to the Roche Elecsys 170 clinical analyzer in 2009–1010 and then to Roche e601 instrument in 2015–2018. NHANES conducted crossover studies to compare the data due to method changes and a weighted Deming regression was provided to adjust for method differences. For serum soluble transferrin receptor concentration, Hitachi 912 instrument was used in 2005–2008, then Hitachi Mod P in 2009–2010, and lastly, Roche c501 in 2015–2018; however, the 3 methods give similar results, thus no adjustment was required. As ferritin is a positive acute-phase protein, we used a regression correction approach recommended by the Biomarkers Reflecting Inflammation and Nutritional Determinants of Anemia group to calculate adjusted ferritin levels by C-reactive protein (CRP) as follows [15]:

$${\text{Ferritin}}_{\text{adjusted}}={\text{Ferritin}}_{\text{unadjusted}}–\mathrm{\beta }1\left({\text{CRP}}_{\text{obs}}–{\text{CRP}}_{\text{ref}}\right)$$

where β₁ is the CRP regression coefficient, obs is the observed value, and ref is the reference value used to define low inflammation and is equal to the maximum value of the lowest decile determined form log-transformed CRP. The correction of ferritin was applied only to individuals with ln CRP greater than ln CRP_ref to avoid overadjustment. Next, we used the body iron model to identify subjects with ID, which have been verified and used in previous NHANES research [16]. This model uses the ratio of the serum soluble transferrin receptor to serum ferritin as in the following formula:

$$\text{Body}\text{iron}\left(\text{mg}/\text{kg}\right)=\frac{-\left[\log 10\left(\text{soluble}\text{transferrin}\text{receptor}\left[\text{mg}/\mathrm{L}\right]\right)\times 1000/\text{ferritin}\left[\mathrm{\mu }\mathrm{g}/\mathrm{L}\right]-2.8229\right]}{0.1207}$$

A negative value of <0 mg/kg is indicative of ID. Before we applied this model, we adjusted serum ferritin for inflammation using CRP Anemia was defined according to the WHO criterion as having hemoglobin <12 g/dL, and the body iron was used to determine if anemic subjects have IDA.

As with iron measures, vitamin D was analyzed using different methods. In 2005–2006, serum 25-hydroxyvitamin D [25(OH)D] was measured using the DiaSorin radio immunoassay (RIA) kit, which was then changed to standardized liquid chromatography–tandem mass spectrometry (LC-MS/MS) in 2007–2018. NHANES later converted vitamin D levels measured before 2007 to LC-MS/MS method equivalent levels due to bias and imprecise measurements from the DiaSorin RIA kit [17]. We added an indicator for the vitamin D method in the multivariable models. Given the absence of a universal definition for the cutoff value of VDD, we used 2 cutoff points adapted from the Institute of Medicine [18]. We used vitamin D levels <12 ng/mL as indicative of VDD and we used a higher cutoff point of <20 ng/mL to include subjects with VDI.

Sleep measures

Our main outcomes were sleep duration, sleep latency, and sleep quality. A self-reported questionnaire related to sleep habits and disorders was used to collect sleep data. Sleep duration was assessed by asking “How much sleep do you usually get at night on weekdays or workdays?”. We categorized this variable as short sleep duration (≤6 h/night) or normal sleep duration (>6 h/night). Sleep latency was assessed by asking “How long does it usually take you to fall asleep at bedtime?” and we categorized the responses as normal (≤30 min/night) or prolonged (>30 min/night). For sleep quality, subjects were asked several questions about the past month’s frequency of trouble falling asleep, waking up during the night, waking up too early in the morning, feeling unrested during the day, feeling overly sleepy during the day, and not getting enough sleep. We categorized the variable as poor sleep quality if the response was often or almost always for any question and good sleep quality if otherwise.

Covariates

Covariates selection was based on a priori knowledge of risk factors that affect sleep and include age, race/ethnicity, poverty–income ratio (PIR), education level, marital status, season, smoking status, alcohol status, physical activity, sedentary level, BMI, depression, and health conditions. Self-reported race/ethnicity was categorized as non-Hispanic White, non-Hispanic Black, Mexican American Hispanic, other Hispanic, and other. We used PIR and education level as indicators of family socioeconomic status. PIR was calculated by dividing family income by the poverty guidelines specific to the survey year accounting for inflation and family size. The PIR ranged from 0 to 5 with a ratio of <1 considered below the poverty threshold; we categorized this variable as poor if PIR <1, near poor if PIR 1 to <2, middle income if PIR 2 to <4, and high income if PIR ≥4 [19]. Education level was categorized as less than high school, high school, and more than high school. Marital status was categorized into 3 groups which were married/partner, widowed/divorced/separated, or never married. Seasons refer to the 6-mo period when the examination was performed, where the period from November through April 30 indicates Winter/Spring and the period from May 1 through October 31 indicates Summer/Fall.

Smoking and alcohol status were obtained from associated questionnaires. Smoking was categorized as never smoker (smoke <100 cigarettes ever), former smoker (not currently smoking but has smoked ≥100 cigarettes ever), and current smoker (smoked ≥100 cigarettes ever and currently smoking). Alcohol status was categorized as never drinker (drink < 12 alcohol drinks ever), former drinker (not currently drinking but has drunk ≥12 alcohol drinks ever), and current drinker (drink ≥12 alcohol drinks ever and currently drinking). The Global Physical Activity Questionnaire was used to collect data about sedentary time and physical activity levels [20]. We converted the reported minutes of sedentary activity to hours/per day and then categorized the variable to <4, 4 to <8, 8 to <11, and >11 h of sedentary time. For physical activity, we summed the total metabolic equivalents (METs) from vigorous and moderate physical activity levels. METs were calculated by multiplying the weekly physical activity volume (duration × frequency) of each activity level by its corresponding METs value (8 for vigorous and 4 for moderate). Afterward, we categorized METs as no METPA (0 MET min/wk), insufficient METPA (1–499 MET min/wk), and sufficient METPA (≥500 MET min/wk). Body measure data were taken during the MECs visit by trained health technicians. Height was measured in centimeters using a stadiometer with a fixed vertical backboard and an adjustable headpiece. Weight was measured in kilograms using a digital weight scale. BMI was calculated as weight (kg)/height² (m²) and then categorized as normal weight (<25 kg/m²), overweight (≥25 to <30 kg/m²), and obese (>30 kg/m²).

Depression was assessed by self-reported Patient Health Questionnaires (PHQ-9), a validated screening instrument to assess depression in the past 2 wk. The questionnaire contains 9 questions with a possible frequency response of 0 indicating not at all, 1 indicating several days, 2 indicating more than half the days, and 3 indicating nearly all days with a total possible score ranging from 0 to 27. Depression was categorized as none or minimum (0–4), mild (5–9), moderate (10–14), and moderately severe to severe (15–27) [21]. Health conditions including hypertension, high cholesterol, diabetes, heart disease, and kidney were defined using self-reported questionnaires, physical examinations, or laboratory tests. Hypertension was defined as responding yes to “Have you been told on 2 or more different visits that you had hypertension?”, or “Are you taking prescribed medicine to lower blood pressure” or having a mean diastolic blood pressure ≥80 mmHg, or a mean systolic blood pressure ≥130 mmHg. Participants were considered as having hypercholesterolemia if they were told by a physician that they had high cholesterol, or if the total cholesterol level was ≥240 mg/dL. Diabetes was defined as responding yes to “Have you ever been told by a doctor that you have diabetes?”, or “Are you now taking diabetic pills to lower your blood sugar?”, or a fasting glucose concentration ≥126 mg/dL, or HbA1c ≥6.5%. Subjects responding yes to “Have you ever been told by a doctor that you have prediabetes?” or having a fasting glucose concentration ≥100 to <126 mg/dL, or HbA1c ≥5.7 to <6.5% were considered prediabetics. Participants were categorized as having heart disease if they were told by a doctor or other health professional that they had congestive heart failure, angina, heart attack, or coronary artery disease. Subjects have kidney disease if the calculated glomerular filtration rate (GFR) was <60 mL/min per 1.73 m² or the urinary albumin–creatinine ratio was ≥30 mg/g. GFR was calculated using the CKD-Epidemiology (CKD-EPI) [22].

Statistical analysis

All statistical analyses were performed in SAS 9.4 (Cary) using procedures for sample survey data to account for the complex survey design used in NHANES and using appropriate sampling weights and design variables (clustering and stratification). Before the analysis, we checked for missing values in outcome, exposure, and covariate variables, finding 1590 missing values in sample 1 and 2798 missing values in sample 2. To preserve sample size, improve precision, and increase statistical power, we performed multiple imputations by fully conditional specifications (FCS MI) [23]. The FCS MI is a validated and robust statistical method that involves imputing the dataset multiple times using a sequence of regression models, where each incomplete variable is regressed on the other observed variables, allowing for accurate imputation. Afterward, we assessed the distribution of sleep outcomes, ID/IDA, and VDD/VDI across categories of baseline sociodemographic and lifestyle characteristics. For continuous variables, we reported means and standard deviations and for categorical variables, we reported percentages with Rao–Scott chi-square tests. In the bivariate analysis, we ran logistic regression to estimate odd ratios and 95% confidence intervals (CIs) considering survey data complexity. In the multivariable analysis, we adjusted for potential confounders that were associated with exposures in the bivariate analysis. Given the role of vitamin D on iron metabolism, we evaluated the odds of having both deficiencies in sleep by combining the main exposures into 1 exposure variable as iron deficiency and vitamin D deficiency (IDVDD), and as iron deficiency and vitamin D inadequacy (IDVDI). For each variable, we created 4 categories which are ID and VDD/VDI, non-ID and VDD/VDI, ID and non-VDD/VDI, and non-ID and non-VDD/VDI. Similar categories were created for iron deficiency anemia and vitamin D deficiency (IDAVDD) and for iron deficiency anemia and vitamin D inadequacy (IDAVDI). To test the interaction of ID and VDD/VDI, we assessed the multiplicative interaction by including the product term of ID and VDD/VDI in the model. We assessed additive interactions using relative excess risk due to interaction, attributable proportion due to interaction, and synergy index. Lastly, we conducted several sensitivity analyses to assess the robustness of our results, including the exclusion of subjects who exhibited symptoms of sleep disorders, the exclusion of subjects with missing pregnancy data, the omission of sleep outcomes from the multiple imputation process, and performing a complete case analysis. Diagnostic-checking of the model assumptions was performed to check for any violation that could affect the results.

Results

The sociodemographic characteristics of the 2 samples were comparable. The mean age was 34.9 (0.1) y for the sleep duration sample (NHANES 2005–2010 and 2015–2018) and 35.3 (0.2) y for the sleep latency and quality sample (NHANES 2005–2008). About 12.8% of females had ID, 6.0% had IDA, 7.6% had VDD, and 30.1% had VDI. For sleep outcomes, 25.9% reported short sleep duration, 20.2% reported long sleep latency, and more than half reported poor sleep quality.

Table 1 shows the results of the bivariate analysis between sleep outcomes and baseline sociodemographic and lifestyle characteristics. Black females slept fewer hours and took longer to fall asleep compared with other races, whereas White females had poorer sleep quality even though they slept longer hours. Shorter sleep duration was common among females with obesity, less education and less income, were currently smoking, reported no physical activity, and had depression, high cholesterol, or heart disease. Similar trends were observed among females with long sleep latency and poorer sleep quality. For the exposure–covariate associations (Table 2), racial differences were more pronounced as Black and Mexican American females had a higher prevalence of ID/IDA and VDD/VDI compared to White females. The season was associated with VDD/VDI. Nutrient deficiencies were more common among current drinkers, while females who never smoked seemed to have more ID/IDA. Depression and hypertension were more common among females with VDD/VDI, whereas kidney disease was more prevalent among females with ID/IDA.

TABLE 1.

Associations between sleep measures and sociodemographic and lifestyle characteristics¹

	Sleep duration (N = 6731)			Sleep latency (N = 2497)			Sleep quality (N = 2497)
	≤6 h	>6 h	P value	≤30 min	>30 min	P value	Poor	Good	P value
Age (y)	35.5 (0.2)	34.7 (0.2)		35.1 (0.3)	35.7 (0.4)		35.4 (0.3)	35.2 (0.3)
Race/ethnicity (%)			<0.0001			0.0418			<0.0001
Non-Hispanic White	55.2	63.0		67.2	61.8		70.0	61.4
Non-Hispanic Black	20.5	11.0		12.5	17.0		12.6	14.5
Mexican American	9.3	10.3		9.1	8.7		6.3	12.3
Other Hispanic	7.0	6.5		5.0	6.0		4.7	5.7
Others	8.0	9.2		6.2	6.5		6.4	6.1
PIR (%)			<0.0001			<0.0001			0.0373
Poor	19.9	16.2		13.8	24.3		17.4	14.0
Near poor	23.1	20.5		19.1	23.1		20.4	19.2
Middle income	30.0	30.1		30.5	29.4		29.5	31.4
High income	27.0	33.2		36.6	23.2		32.7	35.4
Education level (%)			<0.0001			<0.0001			0.0090
<High school	18.4	12.8		13.9	23.6		14.4	17.7
High school	23.1	19.3		20.3	25.0		22.5	19.7
>High school	58.5	67.9		65.8	51.4		63.1	62.6
Marital status (%)			<0.0001			<0.0001			0.0073
Married, partner	58.7	63.2		64.9	58.8		62.1	65.6
Widowed, divorced, separated	15.9	11.4		12.2	18.3		15.2	11.2
Never married	25.4	25.4		22.9	22.9		22.7	23.2
Season (%)			0.9128			0.2783			0.3215
Winter∖spring	42.7	42.5		42.3	39.4		40.7	42.7
Summer∖fall	57.3	57.5		57.7	60.6		59.3	57.3
Smoking status (%)			<0.0001			<0.0001			<0.0001
Never smoker	56.2	65.9		63.1	45.6		53.4	67.1
Former smoker	14.5	14.3		15.1	16.0		16.4	14.0
Current smoker	29.3	19.8		21.8	38.4		30.2	18.9
Alcohol status (%)			0.0041			0.1222			0.0077
Never drinker	11.8	12.2		12.4	12.2		10.4	14.8
Former drinker	13.7	10.6		12.2	15.5		13.4	12.2
Current drinker	74.5	77.2		75.4	72.3		76.2	73.0
Physical activity (%)			0.0003			<0.0001			0.0185
No reported MVPA	38.9	34.0		20.4	33.8		25.1	20.7
Insufficient MVPA	24.9	24.0		40.4	29.8		37.6	39.0
Sufficient MVPA	36.2	42.0		39.2	36.4		37.3	40.3
Sedentary time (%)			0.0014			0.1231			0.1251
<4 h	29.0	27.7		29.5	33.2		28.7	32.1
4 to <8 h	42.7	39.8		41.2	41.1		42.2	39.9
8 to <11 h	19.8	24.5		21.2	20.1		21.2	20.7
≥11 h	8.5	8.0		8.1	5.6		7.9	7.3
BMI (%)			<0.0001			0.0001			0.3019
Normal weight	32.7	40.6		42.3	32.3		39.6	41.3
Overweight	26.9	24.9		25.1	27.6		25.2	25.9
Obese	40.4	34.5		32.6	40.1		35.2	32.8
Depression (%)			<0.0001			<0.0001			<0.0001
None or minimum	65.6	74.7		78.0	53.6		63.1	85.2
Mild	19.1	16.9		15.2	24.1		21.5	11.5
Moderate	9.4	5.5		4.2	14.3		9.5	2.3
Severe	5.9	2.9		2.6	8.0		5.9	1.0
Hypertension (%)			0.0174			0.0331			0.1508
No	73.6	77.1		76.9	71.0		74.3	77.6
Yes	26.4	22.9		23.1	29.0		25.7	22.4
High cholesterol (%)			<0.0001			0.0001			0.0079
No	76.2	81.9		79.6	69.7		75.1	80.7
Yes	23.8	18.1		20.4	30.3		24.9	19.3
Diabetes (%)			0.0098			0.0421			0.6579
Not diabetic	70.4	73.7		78.2	73.5		76.8	77.8
Pre-diabetic	25.5	23.1		18.8	22.2		19.6	19.3
Diabetic	4.1	3.2		3.0	4.3		3.6	2.9
Heart disease (%)			<0.0001			<0.0001			0.0015
No	96.3	98.3		98.1	94.4		96.3	98.6
Yes	3.7	1.7		1.9	5.6		3.7	1.4
Kidney disease (%)			0.0566			0.6667			0.2864
No	90.8	92.4		90.8	91.4		91.6	90.0
Yes	9.2	7.6		9.2	8.6		8.4	10.0

Abbreviation: PIR, poverty–income ratio.

¹P value obtained from Rao–Scott chi-square test.

TABLE 2.

Associations between exposure and sociodemographic and lifestyle characteristics (N = 6731)¹

	ID			IDA			VDD			VDI
	No	Yes	P value	No	Yes	P value	No	Yes	P value	No	Yes	P value
Age (y)	34.8 (0.2)	36.2 (0.4)		34.8 (0.2)	37.1 (0.5)		35.0 (0.2)	33.8 (0.5)		35.2 (0.2)	34.4 (0.2)
Race/ethnicity (%)			<0.0001			<0.0001			<0.0001			<0.0001
Non-Hispanic White	63.1	46.7		62.6	35.0		64.6	17.3		73.7	31.4
Non-Hispanic Black	12.3	21.3		12.3	32.4		10.5	49.4		6.2	30.3
Mexican American	9.4	14.4		9.7	15.1		9.6	15.8		6.6	18.0
Other Hispanic	6.4	7.9		6.4	9.5		6.6	6.4		5.7	8.8
Others	8.8	9.7		9.0	8.0		8.7	11.1		7.8	11.5
PIR (%)			<0.0001			0.0001			<0.0001			<0.0001
Poor	16.7	20.1		16.9	21.7		16.5	24.9		14.3	23.8
Near poor	20.4	26.3		20.9	26.2		20.9	25.4		19.3	25.7
Middle income	30.1	30.0		30.0	29.8		29.8	33.4		30.1	30.0
High income	32.8	23.6		32.2	22.1		32.8	16.3		36.3	20.5
Education level (%)			0.0002			0.0001			<0.0001			<0.0001
<High school	13.7	17.5		13.9	19.0		13.8	18.9		11.5	20.3
High school	19.9	23.6		20.0	25.4		19.9	25.9		18.6	24.4
>High school	66.4	58.9		66.1	55.6		66.3	55.2		69.9	55.3
Marital status (%)			0.2246			0.7356			<0.0001			<0.0001
Married, partner	62.2	60.6		62.0	62.7		63.2	47.6		65.4	54.2
Widowed, divorced, separated	12.4	14.0		12.6	12.7		12.5	13.6		12.0	13.9
Never married	25.4	25.4		25.4	24.6		24.3	38.8		22.6	31.9
Season (%)			0.0608			0.0019			<0.0001			<0.0001
Winter∖spring	42.0	46.6		42.0	52.2		40.6	66.2		35.7	58.6
Summer∖fall	58.0	53.4		58.0	47.8		59.4	33.8		64.3	41.4
Smoking status (%)			0.0002			<0.0001			0.0001			<0.0001
Never smoker	62.5	69.8		62.7	74.2		63.0	68.5		61.6	67.6
Former smoker	14.5	12.9		14.7	8.8		14.9	7.7		16.5	9.3
Current smoker	23.0	17.3		22.6	17.0		22.1	23.8		21.9	23.1
Alcohol status (%)			<0.0001			<0.0001			<0.0001			<0.0001
Never drinker	11.4	17.5		11.6	20.6		11.6	18.7		9.9	17.3
Former drinker	10.8	15.2		11.2	14.4		11.3	12.2		10.1	14.4
Current drinker	77.8	67.3		77.2	65.0		77.1	69.1		80.0	68.3
Physical activity (%)			0.0002			<0.0001			<0.0001			<0.0001
No reported MVPA	34.3	41.6		34.5	47.5		33.9	51.2		30.7	45.8
Insufficient MVPA	24.9	19.4		24.7	16.2		24.9	16.5		25.2	22.0
Sufficient MVPA	40.7	39.0		40.8	36.3		41.2	32.3		44.1	32.2
Sedentary time (%)			0.0197			0.0160			0.0755			0.0081
<4 h	27.5	31.6		27.8	31.8		28.3	24.8		28.0	28.1
4 to <8 h	40.6	40.0		40.5	42.0		40.4	42.5		41.3	38.7
8 to <11 h	23.7	20.9		23.7	18.0		23.4	22.2		23.2	23.6
≥11 h	8.2	7.4		8.0	8.2		7.9	10.5		7.5	9.6
BMI (%)			0.0024			0.0090			<0.0001			<0.0001
Normal weight	39.3	33.5		38.8	34.2		39.6	26.1		43.7	26.6
Overweight	25.3	26.1		25.6	23.0		26.0	18.0		26.0	24.0
Obese	35.4	40.4		35.6	42.8		34.4	55.9		30.3	49.4
Depression (%)			0.2045			0.2928			0.0469			0.0002
None or minimum	72.6	70.3		72.4	70.8		72.6	69.2		73.8	69.1
Mild	17.3	18.4		17.4	18.7		17.4	18.4		16.6	19.4
Moderate	6.4	7.2		6.6	6.1		6.4	7.6		6.0	7.7
Severe	3.7	4.1		3.6	4.4		3.6	4.8		3.6	3.8
Hypertension (%)			0.0065			0.0302			0.0002			<0.0001
No	76.8	72.4		76.5	71.2		76.8	68.6		78.0	72.0
Yes	23.2	27.6		23.5	28.8		23.2	31.4		22.0	28.0
High cholesterol (%)			0.9102			0.4321			0.7321			0.0605
No	80.4	80.3		80.3	82.2		80.4	80.0		79.8	81.8
Yes	19.6	19.7		19.7	17.8		19.6	20.0		20.2	18.2
Diabetes (%)			<0.0001			<0.0001			0.0004			<0.0001
Not diabetic	73.8	66.5		73.8	57.3		73.4	65.3		75.3	67.1
Pre-diabetic	23.0	29.0		22.9	37.0		23.3	29.4		22.0	27.8
Diabetic	3.2	4.5		3.3	5.7		3.3	5.3		2.7	5.1
Heart disease (%)			0.2252			0.4245			0.1565			0.0008
No	97.9	97.3		97.8	97.2		97.9	96.9		98.2	96.9
Yes	2.1	2.7		2.2	2.8		2.1	3.1		1.8	3.1
Kidney disease (%)			0.0041			0.0019			0.2465			0.0140
No	92.4	89.2		92.3	87.8		92.1	90.7		92.6	90.5
Yes	7.6	10.8		7.7	12.2		7.9	9.3		7.4	9.5

Abbreviations: ID, iron deficiency; IDA, iron deficiency anemia; VDD, vitamin D deficiency; VDI, vitamin D inadequacy; PIR, poverty–income ratio.

¹P value obtained from Rao–Scott chi-square test.

The results of the multivariable analysis are shown in TABLE 3, TABLE 4, TABLE 5. The odds of short sleep duration were 1.22 (95% CI: 1.01, 1.49) times significantly higher among those with ID compared with those who did not have ID. However, adjusting for confounders attenuated the association to 1.13 (95% CI: 0.92, 1.38), such that it was no longer significant. In contrast, significant associations were detected between VDD and VDI with shorter sleep duration that persisted after adjustment for confounders: 1.26 (95% CI: 1.02, 1.54) and 1.22 (95% CI: 1.04, 1.44), respectively. This association was consistent with the results shown in TABLE 4, TABLE 5, where we assessed the joint association of iron and vitamin D. Moreover, females who had both deficiencies, had significantly higher odds of sleeping fewer hours compared to females who had neither deficiency. However, this association attenuated after adjustment.

TABLE 3.

OR (95% CI) from crude and adjusted associations between sleep measures and exposures (ID/IDA and VDD/VDI)

	ID1	IDA1	VDD2	VDI2
Short sleep duration
Crude	1.22 (1.01, 1.49)	1.15 (0.89, 1.48)	1.72 (1.41, 2.09)	1.55 (1.35, 1.79)
Adjusted	1.13 (0.92, 1.38)	0.95 (0.73, 1.24)	1.26 (1.02, 1.54)	1.22 (1.04, 1.44)
Prolonged sleep latency
Crude	1.12 (0.77, 1.64)	1.13 (0.69, 1.86)	1.21 (0.95, 1.55)	1.22 (0.98, 1.50)
Adjusted	1.04 (0.71, 1.55)	1.08 (0.64, 1.84)	0.91 (0.68, 1.22)	0.89 (0.65, 1.23)
Poor sleep quality
Crude	1.23 (0.90, 1.70)	1.62 (1.02, 2.56)	0.96 (0.75, 1.22)	0.93 (0.77, 1.14)
Adjusted	1.42 (1.02, 2.00)	2.08 (1.29, 3.38)	1.03 (0.78, 1.37)	1.02 (0.78, 1.34)

Abbreviations: CI, confidence interval; ID, iron deficiency; IDA, iron deficiency anemia; VDD, vitamin D deficiency; VDI, vitamin D inadequacy; OR: odds ratio.

¹Adjusted for age, race/ethnicity, poverty–income ratio, educational level, BMI, smoking, alcohol, physical activity, sedentary time, diabetes, hypertension, and kidney disease.

²Further adjusted for season, marital status, depression, heart disease, and vitamin D method indicator.

TABLE 4.

OR (95% CI) from crude and adjusted associations between sleep measures and exposures (IDVDD and IDVDI)¹

	IDVDD				IDVDI
	Both	Only VDD	Only ID	None	Both	Only VDI	Only ID	None
Short sleep duration
Crude	1.84 (1.34, 2.54)	1.73 (1.36, 2.20)	1.20 (0.96, 1.50)	Ref	1.81 (1.40, 2.34)	1.52 (1.30, 1.76)	1.11 (0.82, 1,49)	Ref
Adjusted	1.27 (0.92, 1.75)	1.29 (1.01, 1.64)	1.13 (0.90, 1.42)	Ref	1.31 (1.00, 1.75)	1.22 (1.04, 1.45)	1.12 (0.83, 1.52)	Ref
Prolonged sleep latency
Crude	1.19 (0.55, 2.60)	1.24 (0.94, 1.64)	1.13 (0.75, 1.69)	Ref	1.21 (0.74, 1.99)	1.24 (0.97, 1.59)	1.18 (0.65, 2.15)	Ref
Adjusted	0.82 (0.38, 1.77)	0.95 (0.66, 1.35)	1.03 (0.69, 1.56)	Ref	0.79 (0.44, 1.43)	0.94 (0.69, 1.28)	1.19 (0.65, 2.21)	Ref
Poor sleep quality
Crude	1.85 (1.09, 3.14)	0.80 (0.59, 1.09)	1.14 (0.82, 1.57)	Ref	1.32 (0.91, 1.92)	0.87 (0.70, 1.08)	1.08 (0.69, 1.68)	Ref
Adjusted	1.98 (1.04, 3.77)	0.90 (0.64, 1.25)	1.31 (0.95, 1.83)	Ref	1.53 (0.96, 2.43)	0.96 (0.73, 1.28)	1.27 (0.80, 2.00)	Ref

Abbreviations: CI, confidence interval; ID, iron deficiency; IDA, iron deficiency anemia; IDVDD, iron deficiency and vitamin D deficiency; IDVDI, iron deficiency and vitamin D inadequacy; OR: odds ratio.

¹Adjusted for age, race/ethnicity, poverty–income ratio, educational level, smoking, alcohol, physical activity, sedentary time, diabetes, kidney disease, season, marital status, body mass index, depression, hypertension, and vitamin D method indicator.

TABLE 5.

OR (95% CI) from crude and adjusted associations between sleep measures and exposures (IDAVDD and IDAVDI)¹

	IDAVDD				IDAVDI
	Both	Only VDD	Only IDA	None	Both	Only VDI	Only IDA	None
Short sleep duration
Crude	1.70 (1.20, 2.41)	1.73 (1.39, 2.15)	1.10 (0.82, 1.46)	Ref	1.68 (1.24, 2.28)	1.53 (1.32, 1.77)	0.95 (0.61, 1.48)	Ref
Adjusted	1.11 (0.75, 1.63)	1.28 (1.02, 1.60)	0.97 (0.71, 1.32)	Ref	1.17 (0.82, 1.67)	1.22 (1.04, 1.44)	0.90 (0.57, 1.43)	Ref
Prolonged sleep latency
Crude	1.00 (0.47, 2.12)	1.26 (0.98, 1.62)	1.19 (0.66, 2.12)	Ref	1.05 (0.57, 1.95)	1.25 (0.99, 1.58)	1.42 (0.61, 3.30)	Ref
Adjusted	0.79 (0.38, 1.64)	0.94 (0.68, 1.30)	1.11 (0.60, 2.05)	Ref	0.72 (0.38, 1.35)	0.93 (0.67, 1.28)	1.53 (0.67, 3.50)	Ref
Poor sleep quality
Crude	1.99 (0.88, 4.52)	0.88 (0.65, 1.18)	1.52 (0.91, 2.56)	Ref	1.71 (1.04, 2.80)	0.89 (0.73, 1.10)	1.41 (0.71, 2.79)	Ref
Adjusted	2.47 (1.08, 5.66)	0.97 (0.69, 1.36)	1.94 (1.15, 3.29)	Ref	2.23 (1.26, 4.04)	0.98 (0.74, 1.31)	1.78 (0.88, 3.59)	Ref

Abbreviations: CI, confidence interval; IDAVDD, iron deficiency anemia and vitamin D deficiency; IDAVDI, iron deficiency anemia and vitamin D inadequacy; OR: odds ratio.

¹Adjusted for age, race/ethnicity, poverty–income ratio, educational level, smoking, alcohol, physical activity, sedentary time, diabetes, kidney disease, season, marital status, body mass index, depression, hypertension, and vitamin D method indicator.

None of the exposures were significantly associated with sleep latency. However, we found significant adjusted associations between ID/IDA with sleep quality. The odds of having poor sleep quality were 1.42 (95% CI: 1.02, 2.00) times higher among females with ID compared with not having ID and 2.08 (95% CI: 1.29, 3.38) times higher among females with IDA compared to not having IDA (Table 3). The odds significantly increased by 39% when females had both iron and vitamin D deficiencies (1.42 compared with 1.98) and increased by 19% when females had IDAVDD (2.08 compared with 2.47) (TABLE 4, TABLE 5). Females with both IDA and VDI had significantly higher odds of poor sleep quality compared with those with neither condition (Table 5). However, this association was not significant among females who had ID and VDI (Table 4). The post hoc analysis indicated that not getting enough sleep was the main characteristic driving the association between ID/IDA and sleep quality. Notably, results from sensitivity analyses closely aligned with the initial findings, indicating that our results were robust. Moreover, for sleep quality, we found a significant multiplicative interaction between ID and VDD (P value = 0.0005). No statistical multiplicative or additive interactions for other sleep outcomes were observed with ID and VDI.

Discussion

In this study, we examined the cross-sectional association of iron and vitamin D deficiencies and inadequacies with sleep duration, latency, and quality in a representative sample of the United States females of reproductive age. We found that ID/IDA was associated with sleep quality while VDD/VDI was associated with sleep duration, which emphasizes the importance of assessing different sleep dimensions when studying sleep health. Moreover, having both deficiencies may increase the odds of poorer sleep quality more than the expected multiplicative association, suggesting a synergistic effect of ID and VDD, which warrants further investigation. These results contribute to the literature on sleep and health, specifically targeting the 2 most common dietary deficiencies worldwide that are predominant in females of reproductive age.

Although the research on females’s sleep and ID/IDA is scarce, our results are in line with some previous studies in infants and young children. An analysis of sleeping patterns using electroencephalographic (EEG) in 6-mo-old infants with IDA showed alteration in sleep spindle patterns, a key characteristic of non-rapid eye movement (NREM) sleep stages that is involved in memory consolidation and normal brain function [24]. Moreover, a longitudinal cohort study assessing sleep patterns in infants at 6, 12, and 18 mo of age revealed that IDA infants fell asleep faster at night but spent more time awake and less time in quiet sleep compared to non-IDA infants; and these differences disappeared after iron treatment [25]. Subjective assessment of sleep showed similar results in infants. Maternal reports of sleep in infants aged 6–18 mo show that infants with IDA had higher night waking and shorter total sleep duration compared to infants without IDA [26]. In addition, sleep improved in the infant group randomized to 12 mo of iron–folic acid supplements compared with the control group, suggesting that iron and/or folate can improve sleep in infants with ID and IDA [27].

In adults, large cross-sectional studies have shown that short sleep and sleep disturbance were associated with lower hemoglobin levels, and the odds of having anemia was higher among adults with sleep problems [28,29]. However, given the cross-sectional design, it is uncertain whether poor sleep is a risk factor for anemia or a consequence of anemia. Moreover, these studies did not differentiate between IDA and other types of anemia, though it is estimated that 50% of anemia cases are related to ID. To our knowledge, only 1 study specifically examined the association between IDA and sleep in adults clinically diagnosed with IDA, but healthy otherwise, and found that adults with IDA reported worse sleep quality compared to age and gender-matched controls [30]. However, they did not find an association between blood biomarkers of IDA, such as hemoglobin and ferritin, with sleep quality. In addition, iron therapy was shown to be effective in treating sleep disorder symptoms, in particular RLS. A randomized, double-blind, placebo-controlled trial in adults with RLS and low to normal ferritin levels (15–75 μg/L) found a significant improvement in RLS symptoms in the treatment group (325 mg ferrous sulfate twice daily for 12 wk) compared with controls [31]. In a more recent trial, IV iron therapy significantly improved sleep quality in patients with RLS and IDA [32]. Regardless of having RLS or other sleep disorder symptoms, our study demonstrates a significant association between ID/IDA and sleep quality in healthy females, suggesting that adequate iron intake may be an important component in females’s sleep health.

Vitamin D, in contrast, is well known to contribute to different dimensions of sleep health. A meta-analysis of observational studies suggested that VDD is associated with short sleep duration, poor sleep quality, daytime sleepiness, and increased risk of sleep disorders [33]. In a large cross-sectional study of 21,083 individuals, sleep duration was positively associated with serum vitamin D levels in females, but not in males [34]. Moreover, females with insufficient vitamin D have significantly shorter sleep duration compared to females with sufficient vitamin D levels. In a double-blind control trial, participants aged 20–50 y with sleep disorders were randomized to receive either a placebo or 50,000 IU vitamin D for 8 wk [35]. The authors reported a significant improvement in sleep duration, sleep quality, and sleep latency in the group receiving vitamin D treatment. However, clinical trials in people without sleep disorders are scarce and the results from available trials are inconsistent, which warrants further investigations [36,37]. Larsen et al. [36], reported no significant effect of vitamin D supplementation on sleep duration and daytime sleepiness in individuals with vitamin D insufficiency (defined as <42 nmol/L) [36]. Although the underlying mechanisms between iron, vitamin D, and sleep are not well understood, evidence from animal studies indicates that both low iron and vitamin D levels can significantly affect circadian gene expression [[38], [39], [40]]. Moreover, iron is a vital cofactor in the synthesis of neurotransmitters, dopamine and serotonin, which are involved in the sleep–wake cycle [41]. Likewise, vitamin D is vital in the synthesis of the sleep hormone melatonin, which initiates sleep through synchronization of the sleep–wake cycle with night and day [40]. Another possible mechanism is the exposure to natural light, as both vitamin D and the circadian rhythm are regulated by sunlight [40]. The present study suggests a synergistic effect between ID and VDD/VDI that can increase the odds of poor sleep quality. This could be due to the role of vitamin D in iron metabolism, as emerging evidence from in vitro and in vivo studies suggests that vitamin D can directly and indirectly affect iron homeostasis. Vitamin D plays a direct role in the transcriptional regulation of hepcidin through suppression of the hepcidin antimicrobial peptide (HAMP) gene [42]. Hepcidin is a hormone encoded by the HAMP gene that functions as the master regulator of iron homeostasis, as it can inhibit iron absorption from the gut and sequester iron in the macrophages, preventing iron recycling. To the contrary, vitamin D can indirectly restore iron recycling and promote erythropoiesis by down-regulating pro-inflammatory cytokines which can increase iron availability for erythropoiesis and hemoglobin synthesis [42,43]. This association between vitamin D and iron can provide new insights into the mechanisms that could explain the relationship between sleep and these nutrient deficiencies.

This study has several strengths, including the use of NHANES data which is based on a representative sample of the United States population, accounting for various potential confounders, and using biomarkers to assess iron and vitamin D status. However, the cross-sectional design of this study limits its ability to assess temporality and causal inferences. Moreover, the self-reported sleep data without further objective or clinical assessment of sleep outcomes may cause recall/reporting bias. While females reported their habitual sleep, their nutritional deficiencies were measured at one point in time. Yet, it is possible that transient iron deficiencies due to factors such as the menstrual cycle could result in temporary changes in sleep. Therefore, since the study's sleep questions only assessed the usual or typical sleep patterns, they may not have captured the true relationship between sleep and ID during the time of the deficiency. This limitation may have led to an underestimation of the association between sleep and ID. Moreover, we have limited information regarding the intake of dietary supplements for both sleep and nutrient deficiencies which could affect our findings. Although we adjusted serum ferritin for inflammation using CRP, we could not adjust the serum soluble transferrin receptor due to a lack of α1 acid glycoprotein (AGP) measures [44]. The variation in methods used to measure vitamin D in NHANES can introduce inconsistencies in the data. We address this issue by including a vitamin D method indicator in the multivariable models. Finally, although this study controls for various confounders, residual confounding, the effects of unmeasured confounders—such as genetic predisposition or other unknown factors associated with sleep and study exposures—cannot be excluded.

In summary, we have demonstrated a significant association between ID/IDA and sleep quality as well as VDD/VDI and sleep duration among females of reproductive age. The noteworthy combined association of ID and VDD with sleep quality warrants further investigation. These findings hold potential implications for clinicians providing care to females of reproductive age, as they may consider screening for VDD when caring for females with ID/IDA and sleep problems.

Author contributions

The authors’ responsibilities were as follows – MAH, AB: designed the study and conducted the research; MAH analyzed the data; MAH wrote the article; MAH, ECJ, PXKS, KEP, AB: provided critical feedback and editorial review; MAH, AB: had primary responsibility for final content; and all authors: read and approved the final manuscript.

Conflict of interest

The authors report no conflicts of interest.

Funding

The authors reported no funding received for this study.

Data availability

Data described in the manuscript, code book, and analytic code will be made available upon request.