Relationship between Sleep Duration and Quality with Glycated Hemoglobin, Body Mass Index, and Self-Reported Health in Marshallese Adults

Abstract

Objective

To document sleep duration and sleep quality among a sample of Marshallese adults and to examine if sleep duration and quality are associated with type 2 diabetes (T2DM), body mass index (BMI), and self-reported health in the Marshallese population.

Design

Cross-sectional analysis of a staff-administered survey.

Setting

Thirty Marshallese churches in Arkansas and Oklahoma.

Participants

The study includes 378 Marshallese participants, 56.6% female, with a mean age of 42.4 years (± 11.6). Recruitment was limited to participants who were considered overweight, with a BMI >25kg/m2.

Measures

Staff-administered surveys were used to collect data on sleep duration, sleep quality, and self-reported health. Clinical measures were collected by trained research personnel using standard tools and protocols. Kruskal-Wallis tests, Spearman’s correlations, and nonparametric tests of trends were used to evaluate differences in HbA1c, BMI, and self-reported health by sleep duration and quality. Multivariable analyses were used to test the associations, controlling for sociodemographic factors.

Results

Fifty-four percent of the participants reported something other than normal sleep duration and 52.4% reported at least one night of difficult or interrupted sleep in the previous two-week period. Longer sleep duration was associated with lower HbA1c and poorer sleep quality was associated with higher HbA1c. Poor sleep quality was associated with lower self-reported health. However, neither sleep duration nor quality were associated with BMI. The associations were found independent of sociodemographic factors.

Conclusion

This is the first study to document sleep duration and sleep quality, as well as the first study to examine the relationship between sleep and HbA1c, BMI, and self-reported health in Marshallese adults with a BMI >25kg/m2. This research will be used to help develop sleep interventions to address T2DM health disparities in the Marshallese community.

INTRODUCTION

Native Hawaiian Pacific Islanders (NHPI) are the second fastest growing population in the US, with many southern and western states seeing the largest increase.^{1, 2} From 2000 to 2016, the Marshallese population grew by 197% in Arkansas and the largest concentration of Marshallese in the continental US reside in a community in the northwest region of the state.^{3, 4} Employment, educational opportunities, and familial connections draw many Marshallese from the Republic of the Marshall Islands (RMI) to northwest Arkansas. The RMI is a chain of volcanic islands and coral atolls located in the central Pacific Ocean halfway between Hawaii and New Zealand.⁵ From 1946 to 1958, the US military conducted an extensive nuclear weapons testing program in the RMI with more than 60 nuclear tests; the nuclear tests were equivalent to 7,200 Hiroshima-size bombs.⁶ The nuclear tests contaminated the environment and traditional food sources (e.g., fish, breadfruit, coconut), which resulted in changes in the traditional diet.^7–9 The shift to foods high in fat and processed commodity food have negatively affected the health of the Marshallese.^7–9 The nuclear contamination caused acute radiation exposure and serious health effects for the Marshallese community.^{7, 10} In 1986, the Compact of Free Association (COFA) with the US and the Republic of the Marshall Islands (RMI) was signed. This diplomatic agreement, a result of a history of nuclear weapons testing in the RMI in the 1940s and 50s, provides the US military exclusive access and control over a strategic base of operations in the Pacific and permits Marshallese to freely enter, live, work, and study in the US without a visa.¹¹

The NPHI population in the US face significant health disparities, including a high prevalence of heart disease, obesity, and type 2 diabetes (T2DM).^12–19 The Marshallese have a particularly high prevalence of obesity and T2DM, with estimates of T2DM prevalence ranging from 20% to 50%, compared to 8.6% for US adults.²⁰ The NHPI, and Marshallese specifically, are also underrepresented in research, which perpetuates and further obscures these health disparities impacting this community.^21–24 Further complicating health disparities research with the Marshallese community are several socioeconomic challenges, including low paying jobs without insurance coverage, low levels of educational attainment, and unstable housing.²⁵

Sleep duration and quality are important contributors to health and have been linked to poor health outcomes in previous research.^{26, 27} For example, in the United States (US) 50 to 70 million adults report insufficient sleep and/or suffer from a sleep disorder.²⁷ Meta-analysis studies have demonstrated both short and long durations of sleep are predictors of poor cardiovascular outcomes, and short sleep is significantly associated with a higher mortality rate.^{28, 29} Meta-analysis of the effect of sleep duration on HbA1c showed both short and long sleep duration were associated with increased HbA1c levels.²⁸ Studies have also found short sleep to be a predictor of obesity.^{27, 30–32} Therefore, chronic insufficient sleep has become an increasing concern in most countries, including the United States.

Previous work has explored sleep duration and sleep quality in other underrepresented populations, including in Black, Hispanic/Latinx, and American Indian/Native Alaskan populations.^33–41 One study examining sleep duration in the NHPI population found that NHPIs were more likely to report insufficient sleep (less than seven hours of sleep per night) compared to Whites, and NHPIs were more likely than other population groups to report poor sleep quality, including more difficulty falling asleep than Blacks.³⁴ Very short sleep (less than 5 hours per night) was associated with a higher prevalence of diabetes in NHPIs.³⁴ Although previous work has explored the connection between sleep and health in the NHPI population, it has not examined the connection between sleep duration/quality and T2DM among the Marshallese population in the United States.

Given the high rates of T2DM in the Marshallese population, the lack of research on sleep duration and quality and its effects on the health of the Marshallese population leaves a major gap in the literature, which this study seeks to fill. The first aim of this study is to document sleep duration and sleep quality among a sample of Marshallese adults. The second aim is to examine whether sleep duration is associated with T2DM, body mass index (BMI), and self-reported health in the Marshallese population. The third aim is to examine whether sleep quality is associated with T2DM, BMI, and self-reported health in the Marshallese population. We hypothesize that shorter sleep duration and poor sleep quality will be associated with higher HbA1C levels (an indicator of T2DM), higher BMI, and worse self-reported health among Marshallese adults.

PARTICIPANTS AND METHODS

Participants

The sample was drawn from 30 Marshallese churches in Arkansas and Oklahoma, as part of a community-based participatory research partnership with the Marshallese community. The community-based partnership has been described in several articles published elsewhere.^{19, 42, 43} Church-based recruitment is culturally appropriate and has previously been identified as the preferred recruitment location by community stakeholders.^{44, 45} Participants were recruited to take part in a diabetes prevention and education intervention; participating churches distributed study information, and study staff presented information to entire congregations in both Marshallese and English. Given the goal of diabetes prevention and education in this intervention, individuals considered eligible for the study were at least 18 years of age, self-identified as Marshallese, and had a BMI of 25 kg/m² or higher at the time of the study.⁴⁴ Individuals who were pregnant, breastfeeding an infant, or had a clinically significant medical condition likely to affect weight at the time of data collection were excluded from the study. Three hundred and eighty participants were recruited and informed consent was obtained. This study utilized complete case analysis of the data from baseline surveys administered to 378 of the participants. Given the use of listwise deletion in complete case analysis, participants included in the analysis completed all of the measures of interest included in the model. Therefore, of the 378 participants in the study, 375 completed the sleep duration measure, 377 completed the sleep quality measure, and 376 completed the self-reported health measure (see Table 1). Consent forms and all survey instruments were translated into Marshallese by a certified translator and approved by the study’s Marshallese Community Advisory Board, and were administered by bilingual study staff (Marshallese and English).⁴⁴

Table 1.

Participant Sociodemographic and Health Characteristics (n = 378)

Measure	Mean ± SD or n (%)
Age	42.1 ± 11.6
Sex
Male	164 (43.4)
Female	214 (56.6)
Employment
Employed	206 (54.5)
Unemployed	172 (45.5)
Marital Status
Married or Cohabitating	313 (82.8)
Single	65 (17.2)
Education
Less than a HS Diploma	184 (48.7)
HS Diploma	132 (34.9)
Beyond HS Diploma	62 (16.4)
HbA1c	7.7 ± 2.7
BMI	33.7 ± 5.4
BMI Classification
Overweight	97(25.7)
Obesity Class I	146(38.6)
Obesity Class II	89(23.5)
Obesity Class III	46(12.2)
Self-Reported Health (n = 376)
Excellent	49 (13.0)
Good	251 (66.8)
Fair/Poor	76 (20.2)
Sleep Duration (n = 375)
Very Short (0–4)	20 (5.3)
Short (5–6)	138 (36.8)
Normal (7–8)	173 (46.1)
Long (≥ 9)	44 (11.7)
# of Nights Trouble Sleeping (n = 377)
0	180 (47.8)
1–2	109 (28.9)
3–5	56 (14.9)
6 or more	32 (8.5)

Note: Percentages may not add up to 100% due to rounding. SD= Standard Deviation, HS = High School

Measures

Sleep.

Sleep was assessed using questions utilized in the Behavioral Risk Factor Surveillance System (BRFSS) survey core module and have commonly been used to evaluate sleep.^46–48 Sleep duration was measured using the question “On average, how many hours of sleep do you get in a 24-hour period?” Sleep duration was categorized into four categories; (1) very short sleep (0–4 hours), (2) short sleep (5–6 hours), (3) normal sleep (7–8 hours), and (4) long sleep (≥ 9 hours). Sleep quality was assessed using the question “Over the last two weeks, how many days have you had trouble sleeping or staying asleep?” Insomnia is indicated by difficulty sleeping 3 nights or more per week, however, we were particularly interested in the effects of subclinical levels of difficulty sleeping, therefore sleep quality was categorized as: (1) zero nights, (2) 1 to 2 nights, (3) 3 to 5 nights, and (4) 6 nights or more.⁴⁹

Biometric data.

All biometric data collection was completed by trained research staff and included HbA1c and BMI. To measure HbA1c, study staff used a finger stick blood collection and received results using the Rapid A1c test kit and Siemens DCA Vantage Analyzer. BMI was calculated using height and weight measurements collected by trained research staff, using a calibrated scale. Participants were weighed without shoes and in street clothes.

Self-reported health.

Self-reported health status has been shown to be an indicator of future adverse health events (e.g. myocardial infarction) and mortality.^{50, 51} Self-reported health was measured using the question “Would you say that in general your health is excellent, good, fair, or poor?” Self-reported health was created as a categorical variable of excellent (=2), good (=1), and fair/poor (=0). The self-reported health measure has been validated when combining categories, therefore fair and poor were combined due to the low number of “poor” responses.^{52, 53}

Sociodemographic characteristics.

Sociodemographic characteristics have been linked to both poor sleep and poor health outcomes.^{26, 27} Sociodemographic variables include (1) age; (2) sex of the participant; (3) education; (4) employment status; and (5) marital status.

Analysis

Descriptive statistics, including means and standard deviations for continuous variables and frequencies and proportions for categorical variables, are presented to characterize all participants. Kruskal-Wallis equality-of-populations rank tests, Spearman’s Correlations, and nonparametric tests of trends were used to evaluate differences in HbA1c, BMI, and self-reported health by sleep duration and sleep quality. Multivariable analyses were conducted to test the associations between the variables of interest and sleep, including linear regression for HbA1c and BMI and ordinal regression for self-reported health, controlling for sociodemographic characteristics; multivariate analyses report unstandardized coefficients. The analyses were conducted using STATA version 15.1, and p-values less than .05 were considered statistically significant. All study procedures were approved by the University of Arkansas for Medical Sciences Institutional Review Board (IRB #207034).

RESULTS

Participant Characteristics

Table 1 reports the characteristics of the participants. The mean age of participants was 42.4 years (± 11.6). Fifty-seven percent of the participants were female, and 82.8% of the participants were married or cohabitating. Nearly half (48.7%) of participants had not graduated high school. Overall, participants had a mean HbA1c of 7.7% (± 2.7%) and a mean BMI of 33.7 kg/m² (± 5.4 kg/m²). One-quarter (25.7%) of the participants were classified as overweight, 38.6% were categorized as obese class I, 23.5% were categorized as obese class II, and 12.2% were categorized as obese class III. Thirteen percent of the participants reported their health as excellent, 66.8% reported their health as good, and 20.2% reported their health as fair or poor.

In regard to sleep duration, 5.3% of the participants reported very short sleep (0–4 hours), 36.8% reported short sleep (5–6 hours), 46.1% reported normal sleep (7–8 hours) and 11.7% reported long sleep (≥ 9 hours). For sleep quality, 47.8% of participants reported zero nights of difficulty sleeping or interrupted sleep, 28.9% percent of participants reported having one or two nights of difficulty sleeping or interrupted sleep, 14.9% reported three to five nights of difficulty sleeping or interrupted sleep, and 8.5% reported six or more nights of difficulty sleeping or interrupted sleep in the preceding two-week period.

Biometric Measures and Sleep

Table 2 reports the results of the Kruskal-Wallis equality-of-populations rank tests conducted to determine if the measures of HbA1c and BMI were different for the four sleep duration categories. Results showed a statistically significant difference in HbA1c among the four categories (χ² (3) = 19.98, p < .001). The mean and median for HbA1c were highest among the participants reporting very short sleep (M = 8.6, Mdn = 7.6), and decreased with the lowest mean and median among participants who reported long sleep (M = 6.7, Mdn = 5.9). The nonparametric test for trend showed a negative linear trend between HbA1c and sleep duration (z = −4.45, p < .001).

Table 2.

Kruskal-Wallis equality-of-populations rank test of Biometric Measures by Sleep Duration (n = 375)

	HbA1c				BMI
Sleep Duration	n	Mean	Median	IQR	n	Mean	Median	IQR
Very Short (0–4 hours)	20	8.6	7.6	4.10	20	33.5	33.3	5.40
Short (5–6 hours)	138	8.2	7.0	4.40	138	33.6	32.5	7.00
Normal (7–8 hours)	173	7.4	6.1	2.90	173	33.6	32.9	7.00
Long (≥ 9 hours)	44	6.7	5.9	1.25	44	35.0	34.3	6.00
Kruskal-Wallis Equality-of-Populations Rank Test	X2 = 19.98 p < .001				X2 = 2.60 p = .46
Nonparametric Test for Trend	z = −4.45 p < .001				z = .89 p = .37

Notes: IQR = Interquartile Range, p values meeting significance are bolded

There were no significant differences in BMI when analyzed as a continuous measure. Additional analysis of BMI as a categorical measure (Overweight, Obesity Class I, Obesity Class II, and Obesity Class III) also showed no differences by sleep duration (r_s = .06 p = .27; supplemental Table 1). The Kruskal-Wallis test did not show significant differences for HbA1c by sleep quality (Table 3). However, among the three categories of participants who reported at least one night with trouble sleeping, the nonparametric test of trends indicated median HbA1c increased linearly as the number of days of disrupted sleep increased (z = 2.58, p = .01). No significant differences were found for BMI and sleep quality. Additional analysis of BMI as a categorical measure (Overweight, Obesity Class I, Obesity Class II, and Obesity Class III) also showed no differences by sleep quality (r_s = −.03, p = .57; supplemental Table 2).

Table 3.

Kruskal-Wallis equality-of-populations rank test of Biometrics Measures by Sleep Quality (n = 377)

	HbA1c				BMI
# of Nights of Trouble Sleeping	n	Mean	Median	IQR	n	Mean	Median	IQR
0	180	7.4	6.2	2.45	180	33.9	33.0	6.70
1–2	109	7.6	6.2	3.00	109	33.4	32.9	7.90
3–5	56	8.3	7.3	5.05	56	33.3	33.3	6.30
6+	32	8.5	7.8	4.75	32	34.6	34.0	5.80
Kruskal-Wallis Equality-of-Populations Rank Test	X2 = 7.51 p = .06				X2 = 1.86 p = .60
Nonparametric Test for Trend	z = 2.58 p = .01				z = −.01 p = .99

Notes: IQR = Interquartile Range, p values meeting significance are bolded

Multivariable regressions of biometric measures.

Table 4 presents the results of the regression models controlling for the sociodemographic characteristics of the participants. For HbA1c, sleep duration was negatively associated with HbA1c (β = −.40, p = .03), and sleep quality was positively associated with HbA1c (β = .30, p = .03). HbA1c also positively associated with age (β = .06, p < .001) in both models of HbA1c, and negatively associated with sex in both the sleep duration (β = −.92, p < .001) and sleep quality (β = −1.05, p < .001) models. HbA1c was positively associated with employment in the sleep duration model (β = .58, p = .05). For BMI, neither sleep duration nor sleep quality were significant in the models, however, sex was positively associated with BMI in both the model controlling for sleep duration (β = 2.35, p < .001) and the model controlling for sleep quality (β = 2.32, p < .001; Table 4.). These results reflect the results of the bivariate analyses of the relationship between sleep duration/quality and HbA1c and BMI.

Table 4.

Association between Sleep Duration/Sleep Quality controlling for Sociodemographic Characteristics

	HbA1c and Sleep Duration			HbA1c and Sleep Quality			BMI and Sleep Duration			BMI and Sleep Quality			Self-Reported Health and Sleep Duration			Self-Reported Health and Sleep Quality
	β	SE	p	β	Std. Err.	p	β	SE	p	β	SE	p	β	SE	p	β	SE	p
Age	.06	.01	>.001	.06	.01	>.001	−.05	.02	.06	−.04	.02	.08	−.04	.01	>.001	−.03	.01	.001
Female	−.92	.28	.001	−1.05	.28	>.001	2.34	.58	>.001	2.32	.57	>.001	−.32	.23	.16	−.26	.23	.26
Education	−.32	.18	.08	−.34	.18	.06	.56	.37	.13	.57	.37	.12	.01	.15	.93	.04	.15	.78
Employment	.58	.29	.05	.53	.29	.07	.11	.59	.85	.07	.59	.90	−.19	.24	.42	−.15	.24	.51
Marital Status	−.55	.37	.14	−.54	.36	.14	.14	.75	.86	.08	.74	.91	−.26	.30	.39	−.27	.30	.38
Sleep Duration	−.40	.18	.03	--	--	--	.007	.38	.99	--	--	--	.14	.15	.34	--	--	--
Sleep Quality	--	--	--	.30	.14	.03	--	--	--	−.04	.29	.90	--	--	--	−.35	.12	.003

Self-Reported Health and Sleep

Table 5 reports the results of the Spearman’s correlation conducted to determine if self-reported health was correlated with sleep duration. There was not a significant correlation between self-reported health and sleep duration (r_s = .08, p = .11).

Table 5.

Spearman Correlation test of Self-Reported Health by Sleep Duration (n = 373)

	Self-Reported Health
	Excellent	Good	Fair/Poor
Sleep Duration	n(%)	n(%)	n(%)
Very Short	0(0.0)	12(60.0)	8(40.0)
Short	14(10.2)	93(67.9)	30(21.9)
Normal	31(18.0)	118(68.6)	23(13.4)
Long	4(9.1)	26(59.1)	14(31.7)
Spearman Correlation		rs =.08 p = .11
Nonparametric Test for Trend		z = 1.50 p = .13

Note: Percentages may not add up to 100% due to rounding, p values meeting significance are bolded

Table 6 reports the results of the Spearman’s correlation conducted to determine if self-reported health was correlated with sleep quality. There was a positive correlation between self-reported health and sleep quality (r_s = .21, p < .001). The nonparametric test for trend results indicated a positive trend in self-reported health by sleep quality category (z = 3.99, p < .001), with the proportion of participants reporting good health decreasing and the proportion reporting fair/poor health increasing as more nights of poor sleep quality are reported.

Table 6.

Spearman Correlation test of Self-Reported Health by Number of Night of Troubled Sleep (n = 375)

	Self-Reported Health
	Excellent	Good	Fair/Poor
# of Nights Trouble Sleeping	n(%)	n(%)	n(%)
0	31(17.2)	126(70.0)	23(12.8)
1–2	9(8.3)	74(68.5)	25(23.2)
3–5	6(10.9)	35(63.6)	14(25.5)
6+	3(9.4)	15(46.9)	14 (43.8)
Spearman Correlation	rs = .21 p < .001
Nonparametric Test for Trend	z = 3.99 p < .001

Note: Percentages may not add up to 100% due to rounding; p-values meeting significance are bolded.

Multivariable regressions of self-reported health.

Sleep duration did not have an association with self-reported health, however, sleep quality was negatively associated with self-reported health (β = −.35, p = .003; Table 4.) Age had a negative association with self-reported health in both the sleep duration (β = −.04, p < .001) and sleep quality (β = −.03, p < .001) models (Table 4.). These results reflect the results of the bivariate analyses of the relationship between sleep duration/quality and self-reported health.

DISCUSSION

The results indicate 54% (CI: [.49, .59]) of the Marshallese report something other than normal sleep duration. This proportion is higher than other population based studies, including one utilizing the 2014 National Health Interview Survey (NHIS), which showed 47% of NHPI and 37% of Whites reported something other than normal sleep duration.³⁴ Although the proportion estimated for Marshallese (54%) reporting something other than normal sleep is higher than for Blacks in the NHIS sample (50%), the overlapping confidence interval indicates the disparities in sleep duration are similar, if not higher, for the Marshallese sample comparatively. Furthermore, 52.3% of the Marshallese in the study reported one or more nights of difficulty falling or staying asleep, whereas only one-third of the Black, White, and NHPI respondents in the 2014 NHIS study reported the same. Overall, the Marshallese participants in this study demonstrated a higher proportion of short and long sleep durations and lower sleep quality than reported by many of the other population groups in the 2014 NHIS, although this may be due to the sample consisting of participants with a BMI of 25 kg/m² or higher.^{34, 54, 55}

As hypothesized, the results indicate significant differences in HbA1c by sleep duration in both the bivariate and multivariable analyses. These results are consistent with prior research describing the correlation between sleep duration and glycemic control. A previous meta-analysis of the effect of sleep duration on HbA1c revealed both short and long sleep duration were associated with increased HbA1c levels and provided evidence of a u-shaped dose response.²⁸ For the Marshallese participants in this study, however, the results indicate a linear dose response with HbA1c highest for those with very short sleep and lowest for those in the long sleep category. The lack of a u-shaped dose response may be due to the relatively small number of participants in the long sleep category (n = 44).

Shorter sleep duration has been shown to affect changes in hormone levels, including a decrease in leptin and increase in ghrelin.^{56, 57} Hormonal changes can lead to the dysregulation of the hypothalamus-pituitary-adrenal axis, increasing cortisol levels.^{56, 57} These hormonal changes can increase appetite, raise BMI, and increase insulin resistance resulting in higher HbA1c levels, which in part, may help to explain the linear response of HbA1c to sleep duration.^{56, 57} Moreover, previous research has suggested that the u-shaped dose response may be more pronounced at younger ages, and a more linear response could be expected in older adults.⁵⁸ The participants in the intervention had a mean age of 42, therefore the linear response of HbA1c may be more a function of age, rather than sleep duration itself.

Although there was not a significant difference in HbA1c levels between the sleep quality groups, there was a positive trend with HbA1c increasing as sleep quality decreased. Prior work has connected poor sleep quality and the increase in HbA1c to potentially similar hormonal disruptions as seen in sleep duration. Increases in HbA1c associated with sleep quality have also been indirectly linked to self-care behaviors.⁵⁹ For example, increases in daytime sleepiness may result in difficulties with medication adherence, a decrease in the amount of physical activity, and poor diet choices, all of which may lead to increases in HbA1c levels and difficulty with glycemic control.^{28, 59} Additionally, elevated glucose levels lead to polyuria, polydipsia, and polyphagia, all of which may lead to more sleep interruptions on their own.⁵⁷ Given the disparate health outcomes for the Marshallese population in the United States when compared to other racial and ethnic groups and the lack of research on sleep in this population, the connection between HbA1c and sleep duration/quality, hormonal responses, and self-care behaviors are important areas for future research to explore to better understand disparities in health outcomes for the Marshallese population.

Our hypothesis regarding the association between sleep and BMI was not supported. There were no significant differences in BMI by sleep duration or sleep quality in either the bivariate or multivariable analyses. This is in contrast to prior findings, which have shown suboptimal sleep duration is associated with increasing levels of obesity.^{56, 57} The lack of a positive association may be due to the sample being comprised of people with a BMI of 25 kg/m² or higher.

Contrary to our hypothesis, sleep duration did not have a significant relationship with self-reported health in either the bivariate or multivariable analyses. Prior literature has shown a u-shaped dose response of sleep duration on self-reported health, with lower appraisals of health for people reporting very short sleep and long sleep. However, this is not the case for the Marshallese adults participating in this study. Poorer sleep quality was associated with lower self-reported health in both sets of analyses. Sleep quality demonstrated a linear dose response on self-reported health, with worse sleep quality decreasing the proportion of people reporting good health and increasing the proportion of people reporting fair or poor health. The findings support our hypothesis, as well as findings in previous literature.^{60, 61}

Self-reported health can mean different things to different people at any point in the life course, and there are potential issues with how the self-reported health measure captures perceptions of health in different populations groups.⁶² Previous research has noted that self-reported health predicts mortality more accurately for Whites versus Blacks or Latinx populations. Though self-reported health may appear to be less predictive in certain sociodemographic and cultural groups, it may be that self-reported health is capturing perceived health differently in different populations.^{63, 64} Therefore, unlike sleep quality, sleep duration may not influence self-reported health in the same way for the Marshallese participants in our study as it does in other racial and ethnic groups, and is an additional area in need of future research to better understand these results.

There are limitations to keep in mind when interpreting these results. The sample is not a random sample and was drawn from a purposive sample of overweight and obese adults. In addition, although 48% of the participants had an HbA1c indicative of type 2 diabetes (≥ 6.5%), the survey did not include questions regarding the use of diabetes medications or complications related to diabetes. Diabetes medication use and complications related to diabetes may influence sleep duration and quality and should be considered in future research in this population.

Participants had a mean BMI of 33.7, which may mask differences in BMI by sleep duration and quality. BMI is related to obstructive sleep apnea, which was not measured in this cohort. Further, the data used in the analysis of general health is self-reported by the participant and does carry a risk social desirability bias. This limitation is reduced through use of validated instruments, and although social desirability may play a role, prior work demonstrated limited effects even for sensitive questions (e.g. substance use).⁶⁵ The sleep duration and quality measures are also self-reported and are reliant on the participant’s recollection of their sleep. Moreover, obesity, type 2 diabetes, and high blood pressure develop over time and given cross-sectional nature of the study and the older age of the participants in the intervention, we are not adequately able to account for changes in sleep and health over time.

Finally, the sample only included Marshallese living in Arkansas and Oklahoma, limiting the generalizability of the findings because they may not representative other Marshallese or NHPI populations living outside Arkansas and Oklahoma. The large and diverse sample size from two states improves generalizability, and prior studies conducted among Marshallese in different states have had yielded similar results across populations^{19, 66–68} Despite the limitations, this article adds important information to the literature on a populations that is often excluded from research. This research will provide a basis for future inquiry in a rapidly growing population with significant health disparities.

CONCLUSIONS

The Marshallese community has one of the highest prevalence of T2DM in the world, and this study adds a significant contribution to the literature as the first study to document sleep duration and sleep quality in a sample of Marshallese adults. In addition, it is the first study to examine the relationship between sleep and HbA1c, BMI, and self-reported health in a sample of Marshallese adults. The connections between sleep and physical health are complicated and often cyclical, and is an area in need of future research to address health disparities faced by the Marshallese. This research and future research will be used to develop sleep interventions as a means of addressing T2DM health disparities in the Marshallese community.