Association of sleep traits with risk of hypertensive disorders of pregnancy: a mendelian randomization study

Abstract

Background:

Unhealthy sleep patterns are common during pregnancy and have been associated with an increased risk of developing hypertensive disorders of pregnancy (HDPs) in observational studies. However, the causality underlying these associations remains uncertain. This study aimed to evaluate the potential causal association between seven sleep traits and the risk of HDPs using a two-sample Mendelian randomization study.

Methods:

Genome-wide association study (GWAS) summary statistics were obtained from the FinnGen consortium, UK Biobank, and other prominent consortia, with a focus on individuals of European ancestry. The primary analysis utilized an inverse-variance-weighted MR approach supplemented by sensitivity analyses to mitigate potential biases introduced by pleiotropy. Furthermore, a two-step MR framework was employed for mediation analyses.

Results:

The data analyzed included 200 000–500 000 individuals for each sleep trait, along with approximately 15 000 cases of HDPs. Genetically predicted excessive daytime sleepiness (EDS) exhibited a significant association with an increased risk of HDPs [odds ratio (OR) 2.96, 95% confidence interval (95% CI) 1.40–6.26], and the specific subtype of preeclampsia/eclampsia (OR 2.97, 95% CI 1.06–8.3). Similarly, genetically predicted obstructive sleep apnea (OSA) was associated with a higher risk of HDPs (OR 1.27, 95% CI 1.09–1.47). Sensitivity analysis validated the robustness of these associations. Mediation analysis showed that BMI mediated approximately 25% of the association between EDS and HDPs, while mediating up to approximately 60% of the association between OSA and the outcomes. No statistically significant associations were observed between other genetically predicted sleep traits, such as chronotype, daytime napping, sleep duration, insomnia, snoring, and the risk of HDPs.

Conclusion:

Our findings suggest a causal association between two sleep disorders, EDS and OSA, and the risk of HDPs, with BMI acting as a crucial mediator. EDS and OSA demonstrate promise as potentially preventable risk factors for HDPs, and targeting BMI may represent an alternative treatment strategy to mitigate the adverse impact of sleep disorders.

INTRODUCTION

Hypertensive disorders of pregnancy (HDPs) affect nearly 10% of pregnancies and encompass clinical phenotypes, such as gestational hypertension, preeclampsia, and eclampsia. This syndrome ranks second among the leading causes of global maternal mortality following maternal hemorrhage, and remains a major cause of short-term and long-term maternal and fetal morbidity [1,2]. The recognition and evaluation of potential causal risk factors are crucial for the prevention and management of HDPs.

Numerous observational studies have consistently reported associations between various unhealthy sleep characteristics and an increased risk of HDPs [3–7]. A comprehensive meta-analysis, incorporating 120 observational studies comprising approximately 60 million pregnant women, evidenced a significant positive correlation between overall sleep disturbances and the incidence of preeclampsia and gestational hypertension [3]. Specific sleep disturbances, including short sleep duration [3], obstructive sleep apnea (OSA) [8], insomnia [9], excessive daytime sleepiness (EDS) [10], and chronotype [11], have been identified in observational studies as factors associated with HDPs. However, due to the limitations inherent in observational studies, causal relationships cannot be inferred, as confounding factors such as cardiometabolic risk factors may impact the observed associations.

Mendelian randomization, employing genetic instrumental variables in its fundamental design, enables the assessment of causal effects between risk factors and diseases while minimizing the impact of confounders or reverse causation [12]. In this study, we conducted a two-sample Mendelian randomization analysis to investigate the causal relationship between seven sleep traits and the risk of HDPs, as well as the specific subtypes of gestational hypertension and preeclampsia/eclampsia. Furthermore, we estimated the proportion of the significant associations mediated by established cardiometabolic risk factors, including BMI, smoking, and type 2 diabetes (T2D) (Fig. 1).

FIGURE 1.

Framework for two-sample Mendelian randomization in this study. The objective is to uncover the causal association between sleep traits and hypertensive disorders of pregnancy, with potential mediation by cardiometabolic factors.

MATERIALS AND METHODS

Study design

This study utilized publicly available, deidentified summary data from previous genome-wide association studies (GWAS). All data sources cited in this study obtained the participants’ informed consent and relevant ethical approval, with detailed citations provided. This study was conducted from October to November 2023, and is reported in accordance with the recommendations of the Strengthening the Reporting of Observational Studies in Epidemiology Using Mendelian Randomization (STROBE-MR) reporting guideline [13].

This two-sample Mendelian randomization study focused on seven specific sleep traits as exposures: chronotype, daytime napping, sleep duration, EDS, insomnia, snoring, and OSA. Additionally, given the observed association of both shorter and longer sleep durations with adverse outcomes [14,15], separate investigations were conducted to clarify their relationships with the outcomes, alongside exploring the impact of overall sleep duration. The outcomes of interest encompassed HDPs as a whole and the two subtypes of gestational hypertension and preeclampsia/eclampsia. Various sensitivity analyses, including multivariable Mendelian randomization and replication Mendelian randomization, were performed to test the robustness of the primary findings and assess the key assumptions underlying Mendelian randomization. Finally, a mediation Mendelian randomization analysis was conducted to investigate the potential causal role of intermediate factors, such as BMI, in the pathway between sleep traits and HDPs.

Data sources

The GWAS summary-level data used in this study were obtained from the FinnGen consortium [16], UK Biobank, GIANT consortium, deCODE, and DIAMANTE consortium. To ensure robustness and minimize confounding due to the genetic population structure, this study exclusively included individuals with European ancestry. Table 1 and Supplementary Table 1 provides information on the data sources, sample sizes, and detailed definitions of the phenotypes.

TABLE 1.

Overview of genome-wide association studies used

Trait	Study	Consortium	Participants	Phenotype definitiona
Exposure
Chronotype	Jones et al.[17]	UK Biobank	449 734 individuals	Self-reported chronotype.
Daytime napping	Dashti et al.[18]	UK Biobank	452 633 individuals	Self-reported the frequency of daytime napping.
Sleep duration	Dashti et al.[19]	UK Biobank	446 118 individuals	Sleep duration was self-reported and treated as a continuous variable, or alternatively, categorized into three groups: short (less than 7 h of sleep), normal (between 7 and 8 h of sleep), and long (more than 8 h of sleep).
Short sleep duration			Cases: 106 192; Controls: 305 742
Long sleep duration			Cases: 34 184; Controls: 305 742
EDS	Wang et al.[20]	UK Biobank	452 071 individuals	Self-reported the frequency of daytime sleepiness.
Insomnia	Watanabe et al.[21]	UK Biobank	Cases: 66 976; Controls: 141 982	Cases: Self-reported insomnia. Controls: not identified as cases.
Snoring	Campos et al.[22]	UK Biobank	Cases: 61 792; Controls: 156 554	Cases: Self-reported snoring. Controls: not identified as cases.
OSA	NA	FinnGen release 9	Cases: 38 998; Controls: 336 659	Cases: FinnGen code G6_SLEEPAPNO Controls: not classified as case
Outcome
HDPs	NA	FinnGen release 9	Cases: 14 727; Controls: 196 143	Cases: FinnGen code O15_HYPTENSPREG Controls: not classified as case
Gestational hypertension		FinnGen release 9	Cases: 8502; Controls: 194 266	Cases: FinnGen code O15_GESTAT_HYPERT Controls: not classified as case
Preeclampsia/eclampsia		FinnGen release 9	Cases: 7212; Controls: 194 266	Cases: FinnGen code O15_PRE_OR_ECLAMPSIA Controls: not classified as case
Covariate
BMI	Pulit et al.[26]	UK Biobank and GIANT consortium	434 749 female individuals	BMI
Smoking	Liu et al.[28]	deCODE and UK Biobank	Cases: 557 337; Controls: 674 754	Cases: self-reported smoking behavior Controls: never smoker
T2D	Mahajan et al.[27]	DIAMANTE Consortium	Cases: 80 154; Controls: 853 816	Cases: T2D was defined based on the criteria outlined in the previous study [27] Controls: not classified as case

DIAMANTE, DIAbetes Meta-ANalysis of Trans-Ethnic association studies; EDS, excessive daytime sleepiness; GIANT, Genetic Investigation of ANthropometric Traits; HDPs, hypertensive disorders of pregnancy; OSA, obstructive sleep apnea; T2D, Type 2 diabetes.

^aA detailed phenotype definition, including the relevant question and the response options in UK Biobank, can be found in supplemental table 1.

Sleep traits

Genetic associations for the following sleep traits were extracted from GWAS conducted in the UK Biobank: chronotype (n = 449 734 individuals) [17], daytime napping (n = 452 633 individuals) [18], sleep duration (n = 446 118 individuals) [19], long sleep duration (n = 339 926 individuals, 34 184 cases and 305 742 controls) [19], short sleep duration (n = 411 934 individuals, 106 192 cases and 305 742 controls) [19], EDS (n = 452 071 individuals) [20], insomnia (n = 208 958 individuals, including 66 976 cases and 141 982 controls) [21], snoring (n = 218 346 individuals, including 61 792 cases and 156 554 controls) [22].

Genetic association estimates for OSA (n = 375 657 individuals, including 38 998 cases and 336 659 controls) were obtained from the FinnGen consortium's nineth data release of GWAS summary data, released on May 11th, 2023 [16].

Hypertensive disorders of pregnancy

GWAS summary statistics for HDPs and the two distinct subtypes were obtained from the ninth data release of the FinnGen consortium [16]. The dataset consisted of 14 727 cases and 196 143 controls for HDPs, 8502 cases and 194 266 controls for gestational hypertension, and 7212 cases and 194 266 controls for preeclampsia/eclampsia.

Potential confounders and mediators

We considered several cardiometabolic factors, namely, BMI, smoking, and T2D, as potential confounders or mediators in the association between sleep traits and the risk of HDPs. These factors were selected based on their established associations (although causal relationships remain uncertain) with both sleep characteristics and HDPs [23–25].

Summary-level data on BMI were obtained from a GWAS meta-analysis that included 434 794 female individuals of European ancestry from the UK Biobank and Genetic Investigation of ANthropometric Traits (GIANT) consortium [26]. Summary-level data on T2D, consisting of 80 154 cases and 853 816 controls, were obtained from a GWAS meta-analysis of DIAMANTE (DIAbetes Meta-ANalysis of Trans-Ethnic association studies) Consortium [27]. Genetic association estimates for smoking initiation were extracted from a GWAS that included 557 337 cases and 674 754 controls from the deCODE and UK Biobank [28].

Selection of the genetic instrumental variables

We selected instrumental variables associated with each exposure by applying a genome-wide significance threshold of P value less than  5 × 10⁻⁸. To identify independent SNPs, we employed the PLINK clumping method in the TwoSampleMR package, setting a linkage disequilibrium threshold of R² less than 0.001 and utilizing a window of 10 000 kb. The 1000 Genomes European data were used as the reference panel [29]. The effect estimates of instrumental SNPs for each exposure were harmonized with the effect estimates of SNPs for the corresponding outcomes. Ambiguous and palindromic alleles with a minor allele frequency greater than 0.42 were excluded from the analysis, resulting in a refined set of SNPs for subsequent MR analysis. Supplementary Table 2 provides detailed information on the instrumental SNPs utilized in this study.

Testing instrument strength and statistical power

To evaluate the robustness of the instruments employed, we computed the F statistic for each SNP that exhibited associations with the exposures. This calculation was performed using the following formula: F = R²(N−k−1)/k(1−R²) [30]. An F statistic surpassing the predefined empirical threshold of 10 indicates a decreased likelihood of potential bias arising from weak instruments in the ensuing Mendelian randomization analysis. Post hoc power calculations were conducted for univariable primary analyses using Stephen Burgess's calculator [31], which accounts for sample size, case-to-control ratio, and the variance explained by genetic instruments for the exposure.

Statistical analysis

The random-effects multiplicative inverse variance weighted (IVW) method was used as the primary Mendelian randomization method [32] to estimate the associations of genetically predicted sleep traits with risk of HDPs and the subtypes. Cochran's Q statistic was computed to examine heterogeneity of SNPs’ estimates in each Mendelian randomization association. The I² statistic was also calculated to evaluate the heterogeneity. A leave-one-out analysis was performed to evaluate the influence of each individual variant on the observed associations.

To examine the robustness of the primary findings and assess the key assumptions underlying Mendelian randomization, we conducted several sensitivity analyses, including the weighted median, MR-Egger, and MR-PRESSO methods. The weighted median method provides valid Mendelian randomization estimates by taking the median of the ratio estimates and assuming that more than 50% of the weight comes from valid SNPs, which is different from the IVW method that take the weighted mean. MR-Egger regression enables the assessment of horizontal pleiotropy by intercept test and provides estimates after correcting for potential pleiotropic effects. MR-PRESSO can identify outlier SNPs and provides effect estimates after adjusting for these potential outliers.

Furthermore, we conducted a replication Mendelian randomization analysis to validate the association between sleep traits and HDPs. For this analysis, we utilized summary statistics data from the UK Biobank, specifically the phenotype of ‘noncancer illness code, self-reported: gestational hypertension/preeclampsia’ (ukb-b-13535). This dataset includes individuals of European ancestry, with 1864 cases and 461 069 controls, and it does not overlap with the individuals included in the initial analysis. Moreover, to account for pleiotropy from well established cardiometabolic risk factors such as BMI, smoking, and T2D, we conducted additional sensitivity analyses by performing a multivariable Mendelian randomization analysis to adjust for these genetically predicted confounders.

Mediation analysis

To evaluate mediation in the significant Mendelian randomization associations, a two-step Mendelian randomization analysis was employed. First, we estimated the causal effect of the exposure on the potential mediator, and only considered putative mediators that exhibited a significant association with exposure. Second, a multivariable MR analysis was conducted to determine the direct effects of both potential mediators and exposure on the outcome. In cases wherein there was evidence of sleep traits influencing the mediator, which subsequently impacted the risk of HDPs, we utilized the ‘product of coefficients’ method to assess the indirect effect of sleep traits on HDPs risk through each potential mediator. Standard errors for the indirect effects were derived using the delta method.

Results are presented as odds ratios (ORs) with 95% confidence intervals (95% CIs). All analyses were two-sided and performed using the packages TwoSampleMR (version 0.4.25) and MR-PRESSO (version 1.0) in R (version 4.1.3).

RESULTS

Main analysis

All selected instrumental SNPs are listed in Supplementary Table 2. All F statistics of the genetic variants proxying sleep traits were greater than 10, supporting the strength of these instruments for use in Mendelian randomization analyses. The Mendelian randomization IVW analysis showed that a genetically predicted EDS was associated with a higher risk of HDPs (OR 2.96; 95% CI, 1.40–6.26; P = 0.005) and preeclampsia/eclampsia (OR 2.97; 95% CI, 1.06–8.3; P = 0.04). Genetically predicted OSA was associated with a higher risk of HDPs (OR, 1.27; 95% CI, 1.09–1.47; P = 0.002), gestational hypertension (OR, 1.22; 95% CI, 1.01–1.47; P = 0.04), and preeclampsia/eclampsia (OR, 1.26; 95% CI, 1.05–1.52; P = 0.01) (Fig. 2 and Supplementary Table 3). No significant associations were observed between the genetically predicted chronotype, daytime napping, sleep duration, insomnia, and snoring with either HDPs or the subgroups (Fig. 2 and Supplementary Table 3). Similarly, no significant associations were found when sleep duration was categorized into binary phenotypes, such as short and normal sleep duration or long and normal sleep duration) (Supplementary Table 3).

FIGURE 2.

Associations of genetically predicted sleep traits with the risk of hypertensive disorders of pregnancy and its subtypes. CI, confidence interval; EDS, excessive daytime sleepiness; HDPs, hypertensive disorders of pregnancy; OR, odds ratio; OSA, obstructive sleep apnea.

Sensitivity analysis

The weighted mode, weighted median, and MR-Egger analyses consistently corroborated the main analysis of the significant associations (Supplementary Table 4). No horizontal pleiotropy was observed in any of the Mendelian randomization analyses (MR-Egger intercept P > 0.05), although some heterogeneity was noted (Supplementary Table 4). The MR-PRESSO test identified no outliers in most associations, except for the relationships between chronotype, sleep duration, OSA, and the outcomes, where one to four outliers were detected. After removing the outliers, the associations remained robust, though with slightly attenuated effect estimates compared to the initial findings (Supplementary Table 5).

To mitigate the potential bias stemming from cardiometabolic risk factors (e.g., BMI, smoking, and T2D) that may confound the associations, we conducted a multivariable analysis to adjust for these risk factors. The findings consistently demonstrated a significant association between EDS and an increased risk of HDPs (OR 2.37; 95% CI 1.35–4.16; P = 0.003), as well as preeclampsia/eclampsia (OR 2.16; 95% CI 1.06–4.41; P = 0.03). Similarly, a higher risk of HDPs (OR 1.18; 95% CI 1.08–1.29; P = 2.59 × 10^–4), gestational hypertension (OR 1.22; 95% CI 1.08–1.37; P = 0.001), and preeclampsia/eclampsia (OR 1.16; 95% CI 1.04–1.30; P = 0.01) was consistently observed in association with OSA following adjustment (Fig. 3 and Supplementary Table 6).

FIGURE 3.

Associations of excessive daytime sleepiness and obstructive sleep apnea with hypertensive disorders of pregnancy in multivariable analyses, adjusting for three cardiometabolic factors (BMI, smoking, and T2D). CI, confidence interval; EDS, excessive daytime sleepiness; HDPs, hypertensive disorders of pregnancy; OR, odds ratio; OSA, obstructive sleep apnea.

To strengthen the robustness of the aforementioned significant association, we conducted a replication Mendelian randomization analysis using additional GWAS summary statistics on HDPs from the UK Biobank with no overlap with individuals included in the initial analysis. The supplementary findings were consistent in direction with our primary analysis, providing further support to our main findings (Supplementary Table 7).

Mediation analysis

Given the relationship between sleep disorders and heightened cardiometabolic risks, we explored the mediating role of three potential factors (BMI, smoking, and T2D) on the causal association between EDS or OSA and HDPs through mediation analysis. First, utilizing a two-sample Mendelian randomization analysis, we estimated the causal effect of EDS and OSA on each potential mediator. Among the three examined cardiometabolic risk factors, we found that BMI was causally affected by both EDS (OR 1.76; 95% CI 1.21–2.58; P = 0.003) and OSA (OR 1.38; 95% CI 1.20–1.59; P = 6.32 × 10^–6), while T2D was causally affected by OSA (OR 1.58; 95% CI 1.21–2.08; P = 8.76 × 10^–4) (Supplementary Table 8). Hence, we focused on BMI and T2D as potential mediators for further analyses.

Second, we performed a multivariable Mendelian randomization analysis to evaluate the relationship between the potential mediators and HDPs. Our analysis revealed causal evidence supporting the direct impact of both BMI and T2D on the outcomes (Table 2).

TABLE 2.

Mediation effect of EDS and OSA on HDPs through cardiometabolic risk factors (BMI, smoking, and T2D)

Exposure	Outcome	Mediator	Total effect β (95% CI)	Direct effect (exposure to mediator) β (95% CI)	Direct effect (mediator to outcome) β (95% CI)	Mediation effect β (95% CI)	Mediated proportion (%) (95% CI)	P
Excessive daytime sleepiness (EDS)	HDP		1.09 (0.34–1.83)		0.47 (0.39–0.55)	0.27 (0.16–0.38)	24.69 (14.42–34.96)	0.02
	Preeclampsia/eclampsia	BMI	1.09 (0.06–2.12)	0.57 (0.19–0.95)	0.48 (0.38–0.59)	0.27 (0.16–0.39)	25.15 (14.82–35.49)	0.01
Obstructive sleep apnea (OSA)	HDP		0.24 (0.08–0.39)		0.42 (0.32–0.52)	0.14 (0.10–0.17)	57.96 (44.46–71.47)	1.78 × 10−5
	Gestational hypertension	BMI	0.20 (0.01–0.39)	0.33 (0.18–0.47)	0.39 (0.26–0.52)	0.13 (0.09–0.16)	63.46 (45.89–81.02)	3.04 × 10−4
	Preeclampsia/eclampsia		0.23 (0.05–0.42)		0.44 (0.31–0.56)	0.14 (0.11–0.18)	60.9 (45.12–76.68)	1.14 × 10−4
Obstructive sleep apnea (OSA)	HDP		0.24 (0.08–0.39)		0.09 (0.06–0.13)	0.04 (-0.02–0.11)	17.97 (-8.99–44.93)	0.51
	Gestational hypertension	T2D	0.20 (0.01–0.39)	0.46 (0.19–0.73)	0.08 (0.04–0.13)	0.04 (-0.03–0.1)	19.21 (-12.67–51.1)	0.55
	Preeclampsia/eclampsia		0.23 (0.05–0.42)		0.10 (0.05–0.15)	0.05 (-0.02–0.11)	19.72 (-7.63–47.08)	0.47

Finally, we estimated the indirect effects of the mediating role of BMI and T2D. Our analysis revealed that BMI mediated 24.69% (95% CI, 14.42–34.96; P = 0.02) of the association between EDS and HDPs; 25.15% (95% CI, 14.82–35.49; P = 0.01) of the association between EDS and preeclampsia/eclampsia; 57.96% (95% CI, 44.46–71.47; P = 1.78 × 10⁻⁵) of the association between OSA and HDPs; 63.46% (95% CI, 45.89–81.02; P = 3.04 × 10⁻⁴) of the association between OSA and gestational hypertension; and 60.90% (95% CI, 45.12–76.68; P = 1.14 × 10⁻⁴) of the association between OSA and preeclampsia/eclampsia (Table 2). Although a significant association was observed between T2D and both OSA and HDPs, we did not find any significant evidence supporting its role as a mediator in this particular context (Table 2).

DISCUSSION

In this two-sample Mendelian randomization study, we systematically assessed the associations of genetically predicted sleep traits with the risk of HDPs and their two subtypes. Our findings support a potential causal relationship between genetically predicted sleep disorders, EDS and OSA, and the risk of HDPs. Importantly, our results remained robust across several sensitivity analyses, addressing different assumptions regarding horizontal pleiotropy. These findings suggest that horizontal pleiotropy is unlikely to explain the observed associations. Furthermore, our mediation analysis revealed that BMI mediated approximately 25% of the association between EDS and HDPs, as well as preeclampsia/eclampsia, while mediating up to approximately 60% of the association between OSA and the outcomes. In contrast, we did not identify any statistically significant associations between the genetically predicted chronotype, daytime napping, sleep duration, insomnia, snoring, and the risk of HDPs, gestational hypertension, or preeclampsia/eclampsia.

Intricate biological mechanisms have been proposed to elucidate the potential impact of sleep disruption on human pregnancy and its associated disorders. This causal connection can be attributed to the dysregulation of neurotransmitter systems, wherein dysfunction of dopamine promotes oxidative stress through the activation of monoamine oxidase, prostaglandin H synthase, and mitochondrial proteins [33]. Moreover, sleep disorders can be a consequence of circadian rhythm disturbance induced by behavioral factors or genetic predisposition, which has detrimental effects on metabolism and endocrine function [34–36]. Notably, an abundance of evidence suggests an overrepresentation of circadian pathways in HDPs, particularly preeclampsia [37,38].

Research indicates a 33% prevalence of EDS among US adults, which is linked to personal and occupational hazards that can pose risks to public safety [39,40]. Additionally, emerging evidence highlights EDS as an independent pathophysiological entity and a novel indicator of cardiovascular risk [41]. Several observational studies have suggested a potential association between EDS and HDPs [6,10,42]. Our Mendelian randomization analysis further strengthens this evidence by establishing a compelling causal association, highlighting the EDS as a possible preventable risk factor for HDPs.

Our mediation analysis demonstrated that approximately 25% of the relationship between EDS and HDPs was mediated through an increase in BMI. This highlights the significance of effectively managing BMI as a crucial strategy for mitigating the risk of HDPs in individuals with EDS. However, these findings also imply the existence of additional mediating mechanisms that necessitate further investigation. One potential mechanism involves EDS-related systemic inflammation triggered by gut dysbiosis and adipose tissue dysfunction [43,44], which may impact the fetomaternal immune crosstalk and in turn lead to pregnancy complications. Alternatively, these systemic inflammatory processes can contribute to maternal endothelial dysfunction [41], which has long been accepted with respect to the pathophysiology of preeclampsia. Further research in this field is warranted, as elucidating these mechanisms is an essential step towards developing targeted clinical interventions.

Our findings revealed a causal link between OSA and HDPs, which is consistent with the results of observational studies and meta-analyses demonstrating an elevated risk of preeclampsia and gestational hypertension in individuals with OSA [3,4]. The prevalence of OSA is estimated to affect nearly one billion individuals globally, with a significantly higher incidence observed among pregnant women, especially those with a high BMI, compared to the general population. The biological mechanisms underlying the association between OSA and HDPs may involve repetitive cycles of hypoxemia and reoxygenation, which can cause oxidative stress and vascular endothelial dysfunction in pregnant women. Our findings provide further evidence of the association between OSA and HDPs, while also highlighting the importance of targeting BMI as a treatment strategy to mitigate the diverse impact of OSA, as BMI mediates up to approximately 60% of the association between OSA and the outcomes.

Although our Mendelian randomization study effectively addressed multiple limitations inherent in conventional observational studies, it remains crucial to carefully interpret the findings with considering the limitations specific to this study and MR in general. First, we were unable to completely rule out the possibility that genetic instruments were exclusively associated with the exposure and independent of potential confounding factors, resulting in an impact on the outcome solely through the exposures rather than pleiotropy. To minimize this bias, we not only implemented sensitivity analyses within the framework of the Mendelian randomization analysis, and also excluded SNPs linked to traditional cardiometabolic risk factors and performed multivariable Mendelian randomization with adjustment for these factors. Second, we cannot determine whether the null findings observed between several sleep traits and the risk of HDPs represent objective facts or are a consequence of potential limitations in the statistical power of our study to detect a significant relationship. Although previous observational studies have suggested an association between sleep disorders, such as insomnia [9,42] and short sleep duration [3,45], and increased rates of preeclampsia, it is important to note that the potential impact of residual confounding inherent in the observational setting may hinder the reliability of the drawn conclusions. This is particularly true in the case of preeclampsia, which is known to be influenced by various socioeconomic and behavioral factors [46,47]. Another plausible explanation is that our study had limited statistical power to detect a significant association between sleep disorders and HDPs. Our post hoc power calculations (Supplementary Table 9) revealed that we had 80% power to detect relatively larger effect sizes for insomnia and long sleep duration, indicating that the true impact of sleep disorders on HDPs may be smaller than what this study had the ability to identify. Third, the GWAS summary data included only individuals of European ancestry, which restricts the generalizability of our findings to other ancestral populations. Fourth, we cannot completely eliminate the potential influence of sex-related bias. Although we made efforts to utilize female-specific GWAS data, it was necessary to extract GWAS summary statistics for several exposures from datasets encompassing both men and women. Nevertheless, genetic variants located on the sex chromosomes were excluded and sex as a covariate was adjusted.

CONCLUSION

The findings from this two-sample Mendelian randomization study support a potential causal association between two sleep disorders, EDS and OSA, and the risk of HDPs, with BMI playing an essential role as a mediator in these associations. EDS and OSA may serve as indicators for the development of HDPs, and should be further investigated as potential preventable risk factors.

Our Mendelian randomization findings, consistent with those of previous observational studies, highlight the importance of screening for sleep disruptions, including EDS and OSA, in pregnant women. Early identification of at-risk individuals may enable an opportunity to ameliorate not only sleep quality but also maternal blood pressure. Further randomized controlled trials (RCTs) evaluating the impact of interventions, such as behavioral education or pharmacological approaches, on maternal and fetal outcomes with both HDPs and sleep disturbances are warranted.

ACKNOWLEDGEMENTS

The authors want to acknowledge the participants and investigators of the FinnGen study. This study was conducted using data from the UK Biobank, a major biomedical database (www.ukbiobank.ac.uk). They also acknowledge the participants and investigators of Genetic Investigation of ANthropometric Traits (GIANT) Consortium, Trans-Ethnic Association Studies (DIAMANTE) Consortium, and deCODE.

Drs. Xiaotian Li and Yu Xiong had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Xiaotian Li, Yu Xiong, and Huanqiang Zhao.

Acquisition, analysis, or interpretation of data: Huanqiang Zhao and Ping Wen.

Drafting of the manuscript: Huanqiang Zhao, Ping Wen, and Qixin Xu.

Critical revision of the manuscript for important intellectual content: Xiaotian Li, Yu Xiong, Huanqiang Zhao, Yang Zi, Yueyuan Qin, Xiujie Zheng, Shuyi Shao, Xinzhi Tu, and Zheng Zheng.

Statistical analysis: Huanqiang Zhao, Ping Wen, Yang Zi, and Yueyuan Qin.

Obtained funding: Xiaotian Li and Huanqiang Zhao.

Study supervision: Xiaotian Li and Yu Xiong.

This study was supported by the National Key Research and Development Program (2021YFC2701600, 2021YFC2701601, 2021YFC2701602), the National Natural Science Foundation of China (82101785 and 81971411), Shanghai Sailing Program (21YF1403800), and Shenzhen Medical Research Fund (A2303073).

The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

All GWAS summary statistics analyzed in this study are publicly available for download in the FinnGen consortium, UK Biobank, GIANT consortium, deCODE, and DIAMANTE consortium, which are listed in Supplementary table 1. All data generated in the current study can be obtained from the Supplementary Information.

Conflicts of interest

The authors declare no competing interests.