Sleep disturbances and assisted reproduction outcomes in women undergoing IVF/ICSI: a systematic review and meta-analysis

Abstract

Background

Sleep disturbances are plausible and potentially modifiable clinical and behavioural markers of risk for assisted reproduction outcomes, but their causal role and the most appropriate targets of intervention remain uncertain, and findings across cohorts are mixed.

Methods

We registered a protocol in PROSPERO under CRD420251118366 and followed PRISMA guidance. We searched PubMed, Embase, Web of Science, and the Cochrane Library through August 3, 2025. Eligible studies included adult women undergoing in vitro fertilization or intracytoplasmic sperm injection, measured sleep disturbances including obstructive sleep apnea, subjective sleep quality, or sleep duration, and reported adjusted associations with oocyte yield, fertilization, embryo quality, implantation, clinical pregnancy, or live birth. We pooled adjusted odds ratios using random effects and also presented fixed effects for comparison. Heterogeneity was assessed with Q and I². We conducted leave one out analyses and used influence diagnostics.

Results

Fourteen prospective IVF/ICSI cohorts involving 9,902 women met the inclusion criteria, and seven contributed data to meta-analyses. Sleep-disordered breathing, predominantly obstructive sleep apnea, was consistently associated with lower treatment success: pooled odds ratios were 0.52 (95% confidence interval [CI] 0.36–0.75) for clinical pregnancy and 0.47 (95% CI 0.30–0.73) for live birth, with no between-study heterogeneity. Poor subjective sleep quality, defined by a Pittsburgh Sleep Quality Index (PSQI) global score > 5, was associated with lower clinical pregnancy under fixed-effects models (pooled odds ratio 0.78, 95% CI 0.67–0.91), whereas the corresponding random-effects estimate was imprecise and compatible with no association. Limited data on sleep duration and timing suggested a possible U-shaped pattern, with both short and long sleep linked to adverse intermediate outcomes, but heterogeneity in exposure definitions, assessment windows, and reported endpoints precluded a single quantitative synthesis.

Conclusions

Objectively confirmed sleep-disordered breathing is associated with lower odds of clinical pregnancy and live birth after assisted reproduction. Associations for perceived sleep quality are model-dependent and attenuate when between-study heterogeneity is considered. Given the limited and observational nature of the available evidence, these results do not support routine targeted sleep screening as a standard component of infertility evaluation, but they indicate that sleep-disordered breathing may be relevant to consider when clinically indicated. Randomized trials are needed to determine whether treating sleep-disordered breathing improves reproductive endpoints and to clarify the contribution of upstream metabolic, psychological, and circadian factors.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12884-026-08808-9.

Introduction

Infertility is a common health challenge, affecting about one in six people worldwide across all regions and income groups [1, 2]. Despite the increasing use of assisted reproductive technologies such as in vitro fertilization (IVF) and intracytoplasmic sperm injection (ICSI), live birth is achieved in only 30% to 40% of fresh cycles, and although frozen embryo transfer cycles may have comparable or slightly higher success rates in some settings, overall effectiveness remains modest [3]. Emerging evidence suggests that behavioral and lifestyle factors, including sleep, may contribute to variability in IVF/ICSI outcomes.

At the biological level, sleep has plausible endocrine, circadian, metabolic, and immune pathways linking sleep disruption to fecundity and implantation biology [4]. Among modifiable behaviours, sleep and circadian timing play a distinct, though overlapping, role compared with diet, physical activity and substance use, because they lie at the interface between hypothalamic–pituitary–ovarian signalling, metabolic homeostasis, and emotional regulation. By contrast, IVF lifestyle research has more extensively characterised the effects of adiposity, smoking, and alcohol use, whereas synthesis of sleep domains remains limited. Focusing specifically on sleep, therefore, allows us to clarify its incremental prognostic value within this broader behavioural context, while recognising that sleep is interrelated with other lifestyle factors and susceptible to confounding. Sleep disturbances are frequently reported among infertile women, with 24.1% to 57% experiencing poor sleep quality and 42% to 69% reporting sleep durations below 7 h during treatment cycles [5]. Individual cohort studies echo this burden, with poor sleep observed before stimulation and around oocyte retrieval, and with short or very long sleep linked to lower oocyte yield, poorer embryo metrics, or reduced pregnancy chances [6]. Sleep-disordered breathing is a specific and biologically relevant exposure in this setting. Meta-analyses show that women with polycystic ovary syndrome (PCOS) have substantially higher obstructive sleep apnea (OSA) prevalence than peers, and clinical series in PCOS cohorts receiving IVF report OSA in approximately 30% of patients [7]. OSA has also been associated with lower clinical pregnancy and live birth after IVF in small but informative human studies [8].

In reproductive-age women, sleep disturbances seldom occur in isolation. Obstructive sleep apnoea and chronically poor subjective sleep quality frequently cluster with higher body mass index (BMI), central adiposity, insulin resistance, and metabolic syndrome, as well as with smoking, alcohol use, and irregular work schedules. Anxiety and depressive symptoms during infertility treatment are also strongly correlated with subjective sleep complaints. Each of these metabolic and psychosocial factors is itself an established predictor of oocyte yield, implantation, and live birth after IVF/ICSI, so any observed association between sleep and assisted reproduction outcomes may be partly or wholly driven by confounding unless they are carefully measured and adjusted for.

In this review, we distinguish three related but conceptually distinct sleep domains. First, sleep-disordered breathing, predominantly obstructive sleep apnea (OSA), is an objectively defined respiratory disorder characterised by recurrent apnoea–hypopnoea events and daytime consequences, typically ascertained using polysomnography or validated home sleep apnoea testing. Second, subjective sleep quality is usually quantified with the Pittsburgh Sleep Quality Index (PSQI), which integrates perceived continuity, latency, and restorativeness of sleep. Third, sleep duration and timing, assessed by self-report or actigraphy, capture behavioural patterns and circadian alignment (for example, nocturnal sleep duration, sleep midpoint, or chronotype). Throughout this manuscript, we use the term “sleep-disordered breathing” (SDB) for objectively measured OSA, and reserve “sleep quality” and “sleep duration” for questionnaire- or actigraphy-based measures; we analyse these domains separately wherever data permit.

Despite growing interest, the literature yields inconsistent findings and an incomplete mechanistic picture [5]. Some studies link poorer Pittsburgh Sleep Quality Index(PSQI) scores with reduced odds of clinical pregnancy, whereas others find trends without statistical significance, underscoring heterogeneity [9, 10]. Short sleep, late sleep timing, and difficulty initiating sleep have been tied to fewer mature oocytes, lower fertilization rates, and fewer good-quality embryos, but estimates vary across cohorts [11]. In contrast, large pre-stimulation cohorts sometimes report weaker or null associations for sleep duration and live birth, raising questions about exposure timing, measurement error, and confounding [6]. Objective and subjective sleep exposures are often pooled or compared imperfectly. OSA is measured with polysomnography or home devices, whereas sleep quality relies on self-report tools like PSQI, and these domains may reflect different biological risks [5]. Emotional distress during treatment correlates with poor sleep and may confound associations with ART outcomes, yet psychological variables are not consistently measured or adjusted [9, 10]. Sample sizes for objectively confirmed sleep-disordered breathing remain small, limiting power to detect effects on implantation and live birth and reducing precision in subgroup analyses such as PCOS or obesity [8]. Furthermore, many studies emphasize intermediate laboratory outcomes rather than live birth, use different thresholds for sleep duration, or capture sleep at different treatment stages, hampering synthesis and clinical translation [5, 11]. Recent reviews call for stronger designs and quantitative synthesis to resolve these discrepancies and to clarify which sleep dimensions carry clinically meaningful risk for ART.

We conducted a systematic review and meta-analysis to evaluate the association between sleep disturbances and ART outcomes among women undergoing IVF or ICSI. Our first objective was to synthesize evidence across distinct sleep exposures, including OSA, subjective sleep quality, in relation to oocyte yield, fertilization, embryo development, implantation, clinical pregnancy, and live birth. Our second objective was to compare risks associated with objectively measured sleep pathology and with subjective sleep complaints to determine whether these exposures represent overlapping or distinct prognostic domains. We hypothesized that objectively diagnosed OSA would show a consistent adverse association with key reproductive outcomes, while associations for subjective sleep quality and sleep duration would be more variable because of psychological confounding and measurement heterogeneity. By delineating these relationships, we aim to inform patient counselling and identify potentially modifiable clinical and behavioural markers, as well as upstream drivers, that could ultimately be integrated into comprehensive fertility care.

Methods

Protocol registration and reporting standards

This systematic review and meta-analysis were conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. The study protocol was prospectively registered with the International Prospective Register of Systematic Reviews (PROSPERO) under registration number CRD420251118366.

Literature search

We carried out a comprehensive search of PubMed, Embase, Web of Science, and the Cochrane Library from their inception through August 3, 2025. The strategy combined subject headings and free-text terms to maximize retrieval. Search terms for sleep included insomnia, poor sleep quality, short or long sleep duration, and obstructive sleep apnea—search terms for assisted reproduction included in vitro fertilization, embryo transfer, and intracytoplasmic sperm injection. Boolean operators and proximity commands were applied to improve sensitivity. Because accurate extraction of design features, effect estimates, and risk-of-bias judgements requires a nuanced understanding of technical terminology across multiple reviewers, we limited inclusion to reports with full texts available in English. We recognise that this language restriction may introduce language bias. The full electronic search strategies for each database are presented in Supplementary Table S1.

Eligibility criteria

Studies were included if they met all of the following criteria:

1. Included adult women who were candidates for or had undergone assisted reproductive technology, such as in vitro fertilization or intracytoplasmic sperm injection, either in prospective cohorts or in retrospective analyses.

2. Assessed exposure to at least one of the following three sleep domains: (i) objectively diagnosed sleep-disordered breathing, predominantly obstructive sleep apnoea; (ii) subjective sleep quality, usually quantified with the Pittsburgh Sleep Quality Index (PSQI); and/or (iii) sleep duration or timing (for example, nocturnal sleep duration, sleep midpoint or chronotype, or actigraphy-based sleep continuity).

3. Reported at least one reproductive outcome such as oocyte yield, fertilization rate, embryo quality, biochemical pregnancy, clinical pregnancy, live birth, or pregnancy complication.

4. Provided effect estimates (odds ratio, relative risk, or hazard ratio) with 95% confidence intervals, or data allowing for their calculation.

Studies were excluded if they investigated intrauterine insemination, involved animals, or were reviews, case reports, editorials, or conference abstracts. Studies without relevant exposure or outcome data were also excluded. In most included cohorts, reproductive outcomes were reported per embryo transfer irrespective of whether embryos were transferred fresh or after cryopreservation. Because very few studies provided stratified estimates by transfer type, we analysed fresh and frozen embryo transfers together.

Study selection and data extraction

Two reviewers (Y.L.H. and J.L.) independently screened titles and abstracts and assessed the full text of potentially eligible studies. Disagreements were resolved through discussion or consultation with a third reviewer (Y.W.). A standardized data extraction form was used to collect key information, including first author, publication year, study setting and design, sample size, population characteristics, exposure definitions, sleep assessment methods, outcome measures, effect estimates, and covariates adjusted for in the analysis.

When studies reported relative risks or hazard ratios but not odds ratios, we applied established methods to approximate odds ratios, assuming the baseline risk in the control group. Specifically, odds ratios were calculated using the formula:

where P₀ represents the event rate in the reference group. This approach is appropriate for outcomes with moderate incidence and was applied consistently to ensure comparability across studies.

Risk of bias assessment

We assessed study quality using the Newcastle–Ottawa Scale, which evaluates the selection of study groups, comparability of groups, and ascertainment of outcomes. For the selection domain, representativeness was evaluated relative to the clinical population undergoing IVF/ICSI in fertility care rather than the general community; the representativeness item was awarded when recruitment was described as consecutive, all eligible included, all invited within a defined centre and time window, or nested within an established prospective clinical cohort of routine care seekers.

For the comparability domain of the Newcastle–Ottawa Scale, we considered maternal age and body mass index (BMI) to be essential confounders and awarded one star when both were included in the adjusted models. A second star was awarded only when, in addition to age and BMI, studies additionally adjusted for at least one lifestyle or metabolic factor (such as smoking, alcohol use, infertility diagnosis or polycystic ovary syndrome, blood pressure or measures of glucose or lipid metabolism) or for psychological distress (for example, anxiety or depressive symptoms), recognising these as key potential drivers of both sleep disturbances and assisted reproduction outcomes. Two reviewers (Y.L.H. and J.L.) independently rated each study, and discrepancies were resolved by consensus. Studies were classified as low, moderate, or high risk of bias based on total scores. Detailed item-level NOS ratings for each study are presented in Supplementary Table S4.

Data synthesis and statistical analysis

We performed meta-analyses when two or more studies reported comparable exposures and outcomes. Adjusted effect estimates and their standard errors were log-transformed and pooled using a random-effects model to account for between-study heterogeneity. Fixed-effect models were also presented for comparison. Between-study heterogeneity was evaluated using the Cochran Q test and I² statistic. Values of I² greater than 50% were considered indicative of moderate to high heterogeneity.

To examine the robustness of results, we conducted leave one out sensitivity analyses, iteratively removing one study at a time to assess its influence on pooled estimates. Baujat plots were used to identify studies contributing most to heterogeneity and overall influence. Because fewer than ten studies contributed to each pooled analysis, formal assessment of publication bias using funnel plots and regression tests was not performed, in line with methodological recommendations.

All analyses were performed using R (version 4.3.2), with the meta and metafor packages.

Protocol deviations

The review protocol was prospectively registered in PROSPERO (CRD420251118366). Compared with the registered protocol, two main deviations occurred. First, we originally planned to estimate the effects of sleep-targeted interventions (such as treatment of obstructive sleep apnoea or behavioural sleep therapies) compared with usual care on IVF/ICSI outcomes. However, the literature search did not identify any randomised or quasi-experimental trials of sleep interventions in women undergoing IVF/ICSI that met our eligibility criteria. Accordingly, the present review is restricted to observational cohort studies, and we do not report intervention effects.

Second, the PROSPERO record specified a broader secondary objective that included a wider range of maternal and offspring outcomes. In practice, we limited quantitative synthesis to reproductive endpoints that were reported in at least two IVF/ICSI cohorts (oocyte yield, clinical pregnancy, and live birth) and summarised other outcomes narratively when available. Apart from these deviations, the eligibility criteria, search methods, and analytical approaches followed the registered protocol.

Results

Study selection and characteristics

The search returned 2,038 records (shown in Fig. 1). We removed duplicates through automated and manual procedures (n = 1,036). Screening of 1,002 titles and abstracts led to 930 exclusions. We then sought full texts for 72 potentially eligible reports via our institution’s electronic subscriptions and library document-delivery/interlibrary loan services. For four citations, full-text articles could not be obtained despite these attempts and were therefore excluded from further assessment. Of the remaining 68 reports, 46 failed to meet the eligibility criteria at full-text review because sleep exposure or assisted reproduction outcomes were not available in an extractable form. We then assessed 22 reports against the protocol and excluded eight. Two were reviews, two were written in languages other than English, and four were conference abstracts without sufficient data. Fourteen prospective IVF/ICSI cohorts met the inclusion criteria and were included in the qualitative synthesis, and seven of these cohorts provided sufficiently comparable data to be included in the quantitative synthesis.

Fig. 1.

PRISMA 2020 flow diagram for studies on sleep and IVF/ICSI outcomes

These 14 studies were published from 2017 through 2025 and together included 9,902 women who underwent in vitro fertilization or intracytoplasmic sperm injection. Most cohorts were conducted in China, with additional work from the United States, Denmark, and Italy. All studies collected sleep information prospectively and recorded assisted reproduction outcomes within a defined treatment cycle. Three cohorts diagnosed obstructive sleep apnea using objective tools such as home sleep apnea testing or continuous respiratory monitoring. Eight cohorts evaluated subjective sleep quality using the Pittsburgh Sleep Quality Index. Other exposures included actigraphy-based measures of sleep timing and continuity and self-reported sleep duration. The primary endpoints were clinical pregnancy and live birth after embryo transfer. Secondary endpoints included counts of retrieved and mature oocytes, normal fertilization, embryo quality measures, and cycle cancellation. Study quality by the Newcastle–Ottawa Scale was generally high, with scores between 6 and 9. Full exposure definitions and adjustment sets are listed in Table S2.

Meta-analysis of sleep disturbances and assisted reproduction outcomes

We prespecified random effects as primary and listed fixed effects. Pooled estimates are reported with 95% CIs and I² (shown in Fig. 2).

Fig. 2.

Sleep exposures and IVF/ICSI outcomes: meta-analysis of clinical pregnancy and live birth. Forest plots showing pooled odds ratios (ORs) for the association between sleep exposures and IVF/ICSI outcomes. A OSA vs non-OSA for clinical pregnancy. B OSA vs non-OSA for clinical pregnancy in the PCOS subgroup. C OSA vs non-OSA for live birth. D Poor sleep quality (PSQI > 5) vs good sleep (≤ 5) for clinical pregnancy. Squares denote study-specific ORs (size proportional to weight); horizontal lines show 95% CIs; diamonds indicate pooled effects. Both fixed-effect and random-effects results are displayed; prediction intervals are based on the random-effects model. Values < 1 favor non-OSA or good sleep (as indicated on the x-axis). Heterogeneity statistics (I² and Q-test p) are reported in each panel. Clinical pregnancy is an ultrasound-confirmed intrauterine gestation; live birth indicates delivery of a liveborn infant. Abbreviations: OSA, obstructive sleep apnea; PSQI, Pittsburgh Sleep Quality Index; PCOS, polycystic ovary syndrome; OR, odds ratio; CI, confidence interval

Obstructive sleep apnea and clinical pregnancy

Three cohorts contributed adjusted estimates for the association between obstructive sleep apnea and clinical pregnancy. The pooled odds ratio by the random-effects model was 0.52 with a 95% CI from 0.36 to 0.75. The fixed-effects model yielded the same value because between-study variance was essentially zero. I² was 0%. The prediction interval was 0.23 to 1.17. Because this interval includes one, the expected effect in a new clinical setting remains uncertain even though the pooled estimate is below one.

Two cohorts recruited women with polycystic ovary syndrome only. The pooled odds ratio in this subgroup was 0.52 with a 95% CI from 0.36 to 0.76, again identical under both models, with I² equal to 0%. The prediction interval stretched from 0.043 to 6.25. This very broad range reflects the small evidence base and the limited information available for dispersion. Given that the prediction interval spans values far below and far above one, this PCOS subgroup finding should be interpreted as exploratory and hypothesis-generating rather than as evidence of a consistent effect across future settings.

Obstructive sleep apnea and live birth

Two cohorts reported live birth after embryo transfer. The pooled odds ratio by the random-effects model was 0.47 with a 95% CI from 0.30 to 0.73. The fixed-effects model again matched this value, with I² equal to 0%. The prediction interval was 0.024 to 8.86. Such breadth is expected when only two studies are available and when downstream outcomes such as live birth are shaped by clinical practice, embryo stage, and number at transfer. Although the pooled estimate is below one, the prediction interval is very wide and extends above one, indicating substantial uncertainty about the expected effect in a new clinical setting and limiting confidence in generalisability.

Subjective sleep quality and clinical pregnancy

Four cohorts examined poor subjective sleep quality defined by a Pittsburgh Sleep Quality Index score greater than five. The fixed-effects model indicated a modest but significant association with lower clinical pregnancy, a pooled odds ratio of 0.78 with a 95% CI from 0.67 to 0.91. The random-effects model produced a pooled odds ratio of 0.70 with a 95% CI from 0.47 to 1.05. Between-study variability was moderate, with I² equal to 47.7%. The prediction interval was 0.23 to 2.11, which spans values well below and above one. The contrast between models and the wide prediction interval points to real differences across cohorts, including differences in when sleep quality was measured, in levels of treatment-related stress, and in the depth of confounder control. The overall pattern still leans toward harm, although the random-effects CI includes the null.

Sensitivity and influence analyses

We tested robustness through leave one out procedures (shown in Fig. 3). For obstructive sleep apnea and clinical pregnancy, the pooled odds ratio stayed below one in every iteration, and the magnitude changed only slightly when each study was removed in turn. The live-birth analysis showed the same pattern. These results indicate that no single cohort drove the adverse association.

Fig. 3.

Leave one out sensitivity analyses of sleep exposures and IVF/ICSI outcomes. Four leave one out (LOO) influence plots summarizing the robustness of the meta-analytic associations between sleep exposures and IVF/ICSI outcomes. A OSA vs non-OSA for clinical pregnancy. B OSA vs non-OSA for clinical pregnancy in the PCOS subgroup. C OSA vs non-OSA for live birth. D Poor sleep quality (PSQI > 5) vs good sleep (≤ 5) for clinical pregnancy.For each row, the point shows the pooled odds ratio (OR) after omitting the study named on the y-axis; horizontal bars denote the 95% confidence interval (CI). The vertical dashed line marks the pooled OR from the main random-effects meta-analysis. Values < 1 (left of the dashed line) favor non-OSA or good sleep, as indicated on the x-axis

Influence diagnostics using Baujat plots (Fig. 4) helped identify which studies contributed most to the pooled effect and to residual heterogeneity. In the analyses of obstructive sleep apnea, the cohort by Li in 2025 had the strongest influence because it carried the largest weight and offered the most precise estimates. The cohort by Zhang in 2024 contributed a moderate influence, while Walter in 2022 contributed less. For subjective sleep quality, the studies by Reschini in 2022 and Bariya in 2025 shaped both magnitude and dispersion to a greater extent than the other cohorts. Philipsen in 2021 contributed less. Even so, omitting any of these influential studies did not change the direction of the pooled effect. For obstructive sleep apnea, the association with lower clinical pregnancy and lower live birth remained statistically significant after every omission, and the estimates under fixed-effects and random-effects specifications were essentially identical, which is consistent with the observed I² of 0%.

Fig. 4.

Baujat plots of study influence and heterogeneity for sleep exposures and IVF/ICSI outcomes. Four Baujat plots visualize which studies drive the meta-analytic results and between-study heterogeneity. A OSA vs non-OSA for clinical pregnancy. B OSA vs non-OSA for clinical pregnancy in the PCOS subgroup. C OSA vs non-OSA for live birth. D Poor sleep quality (PSQI > 5) vs good sleep (≤ 5) for clinical pregnancy. Each point represents one study. The x-axis shows its contribution to heterogeneity, computed as wi(yi − μ)^2 under the random-effects model (inverse-variance weights wi = 1/(vi + τ^2)). The y-axis shows influence on the pooled effect, ∣μ − μ(− i)∣, i.e., the absolute change in the pooled estimate after omitting that study. Points farther to the right and higher up contribute more to heterogeneity and exert greater influence on the pooled result; points near the origin have little impact

Across multiple checks, the evidence for harm linked to obstructive sleep apnea is stable. Clinical pregnancy and live birth were both lower among women with obstructive sleep apnea, with pooled odds ratios of 0.52 and 0.47 and narrow CIs. Prediction intervals were wide because the number of studies was small, not because effects conflicted across cohorts. For subjective sleep quality, the pattern points in the same direction, but heterogeneity and limited numbers reduce precision. Finally, we addressed the requirement for analytical evidence by quantifying the impact of metabolic confounding through a stratified sensitivity analysis as detailed in Supplementary Table S5 and Supplementary Figure S1. This analysis revealed a clear attenuation pattern where the pooled risk estimate for subjective sleep quality shifted from 0.45 in unadjusted models to 0.93 in models strictly controlling for metabolic factors. A statistically significant difference was identified between these two subgroups at a p-value of 0.02. This confirms that while body mass index is a substantial confounder that previously led to an overestimation of risk, the independent adverse impact of poor sleep remains statistically significant at a p-value of 0.014.

Discussion

Our review set out to clarify whether distinct dimensions of sleep disruption matter for assisted reproduction and to resolve mixed signals across single studies. The synthesis shows that objectively diagnosed sleep-disordered breathing is consistently linked to poorer in vitro fertilization outcomes, while subjective sleep quality exhibits model-dependent associations that weaken once between-study heterogeneity is acknowledged. Together, these findings elevate sleep from a background wellness factor to a measurable biologic exposure that can be integrated into reproductive risk assessment.

Objective sleep-disordered breathing

Across included cohorts, obstructive sleep apnea was associated with poorer assisted reproduction outcomes, although the limited number of studies constrains certainty. The weight of evidence indicates that apnea represents a pathophysiologic state rather than a proxy for distress. Intermittent hypoxemia and sleep fragmentation activate inflammatory and oxidative pathways that compromise mitochondrial function and redox homeostasis in metabolically active tissues. These processes can inhibit granulosa cell support of the oocyte, alter steroid biosynthesis, and reduce endometrial receptivity through impaired angiogenesis and tissue remodeling. Clock-controlled transcription factors also tie sleep timing to cellular housekeeping. In granulosa cells, the core clock component NR1D1 represses autophagy-related gene expression, providing a mechanistic route by which circadian misalignment can disturb follicular quality control and oocyte competence [12]. At the tissue level, ovarian cells contain autonomous molecular clocks that coordinate ovulation timing and steroidogenesis with hypothalamic and pituitary signals. Disruption at any node can desynchronize the hypothalamic pituitary ovarian axis and degrade cycle efficiency [13]. In the endometrium, hypoxia-responsive transcriptional programs govern decidualization and receptivity windows. Transcriptomic profiling shows that hypoxic stress reprograms endometrial stromal and decidual cells, interfering with progesterone-responsive networks that are central to receptivity, which is consistent with a narrower implantation window when hypoxia signaling is exaggerated or mistimed [14]. These pathways provide a coherent biological frame that matches the direction and magnitude of the pooled estimates for apnea. However, these mechanistic links are inferred from broader sleep and reproductive biology and were not directly tested in the included IVF/ICSI cohorts, so they should be regarded as plausible hypotheses rather than established causal pathways.

Subgroup analyses provided some insight into metabolically vulnerable populations. When we restricted the analysis to women with PCOS, the pooled association between OSA and clinical pregnancy was similar in magnitude to that observed in the overall OSA meta-analysis, but with wider confidence and prediction intervals due to the smaller number of cohorts and events. This pattern is consistent with the hypothesis that sleep-disordered breathing may also be relevant for IVF/ICSI outcomes in women with PCOS, who already have a high burden of adiposity, insulin resistance, and cardiometabolic risk. However, the precision of the PCOS-specific estimates is limited, and residual confounding by metabolic and lifestyle factors is likely to be substantial, so these subgroup findings should be interpreted as exploratory and hypothesis-generating rather than definitive. Across cohorts that ascertained apnea with polysomnography or validated home testing produced directionally aligned estimates for both clinical pregnancy and live birth, including in polycystic ovary syndrome, where apnea overlaps with insulin resistance and central adiposity [15, 16].

Subjective sleep quality (PSQI)

The contrast between objective and subjective sleep measures helps explain past inconsistencies. Under a fixed effect model, poorer Pittsburgh Sleep Quality Index scores tracked with lower clinical pregnancy, a pattern driven by larger cohorts. When random effects and prediction intervals are considered, the confidence bounds span the null, and the anticipated range in future settings becomes wide. This attenuation is expected if self-report captures a composite of sleep and contemporaneous stressors during stimulation and transfer. Anxiety and depressed mood inflate subjective scores and may confound relations with treatment outcomes, particularly when psychological covariates are not harmonized across studies. The present synthesis is consistent with a recent systematic review that reported model-sensitive associations for perceived sleep quality in fertility care and called for greater standardization of exposure timing and adjustment sets [17].

Sleep duration and timing

Sleep duration findings suggest a biological optimum rather than a linear dose response. Short sleep under seven hours and long sleep at nine to ten hours were each linked to adverse intermediate or clinical endpoints across included cohorts. This pattern is consistent with prior pregnancy research in which short sleep elevates gestational glycemia and long sleep can also worsen glucose homeostasis depending on timing and baseline risk [18, 19]. In women who conceived with assisted reproduction, long sleep of at least ten hours early in gestation is related to higher gestational diabetes risk, with stronger effects at younger maternal ages, which aligns endocrine and metabolic pathways with clinical endpoints that matter after embryo transfer [20]. A plausible unifying explanation is circadian disruption at both extremes. Curtailed sleep diminishes nocturnal melatonin peaks, increases sympathetic tone, and heightens evening cortisol [21–23]. Prolonged sleep can reflect circadian delay or underlying metabolic dysregulation [24]. Melatonin, present in follicular fluid, supports antioxidant defenses and meiotic spindle integrity [25, 26]. Altered melatonin signaling has been linked to poorer oocyte quality and is an attractive, testable mediator between sleep timing and assisted reproduction outcomes.

These interpretations align with and extend the literature in several ways. First, the adverse pattern for apnea resonates with a feasibility cohort that used wearable home testing before stimulation and observed lower pregnancy and live birth proportions among women with sleep-disordered breathing, despite limited power for adjusted estimates [8]. The present synthesis strengthens that signal by pooling across cohorts and by demonstrating stability across clinical pregnancy and live birth, including analyses restricted to adjusted estimates. Second, prospective work in polycystic ovary syndrome showed lower oocyte numbers, reduced fertilization, and lower clinical pregnancy in women with apnea during the first cycle, reinforcing the biological specificity of objectively defined exposure in a metabolically vulnerable subgroup [15]. Third, subjective sleep quality remains inconsistent across designs. A large cohort linked poorer sleep quality to lower success, yet the estimate attenuated with broader adjustment, which mirrors our model-dependent findings and supports cautious interpretation when self-report is the primary exposure [9]. Fourth, narrative patterns for sleep duration in assisted reproduction echo population data on metabolic risk and pregnancy glycemia. Short sleep increases the odds of gestational hyperglycemia, while very long sleep can also be associated with adverse glycemic control. These observations converge on circadian timing, autonomic balance, and adipokine signaling as shared pathways that plausibly influence ovarian steroidogenesis and endometrial readiness for implantation.

A key limitation of the current evidence is incomplete and heterogeneous control of metabolic, behavioural, and psychological confounding. Across the included cohorts, women with sleep-disordered breathing tended to have higher BMI, central adiposity, polycystic ovary syndrome, and other features of metabolic syndrome, while short or poor sleep was more common among those with smoking or alcohol use, irregular work schedules and sedentary lifestyles. Subjective sleep complaints also strongly track anxiety and depressive symptoms during IVF/ICSI cycles. Most studies adjusted at least for age and BMI, but only a subset additionally controlled for metabolic indicators, health behaviours or formal measures of psychological distress, and adjustment sets varied considerably between cohorts. In at least one large PSQI cohort, the association with clinical pregnancy attenuated towards the null after broader adjustment, illustrating how residual confounding can strongly influence effect estimates.

These patterns mean that confounding may partly or fully explain some of the observed associations between sleep and assisted reproduction outcomes. When sleep-disordered breathing arises primarily in the context of obesity and insulin resistance, it is difficult to disentangle whether treating apnoea itself or addressing adiposity and metabolic health would be the more effective intervention target. Conversely, when poor subjective sleep reflects high levels of treatment-related anxiety or depression, psychological support and management of mood disorders may be more appropriate targets than sleep per se. In many clinical scenarios, sleep disturbances may therefore function as markers of underlying metabolic or psychosocial vulnerability rather than independent causal determinants of IVF/ICSI success. Our conclusions emphasise this uncertainty and should not be interpreted as proof that modifying sleep alone will necessarily improve reproductive outcomes.

Clinical implications follow from these considerations. For sleep-disordered breathing, obstructive sleep apnoea is a clinically recognisable and treatable condition. Screening for symptoms and risk factors of apnoea during infertility evaluation can identify women with a potentially modifiable burden of nocturnal hypoxaemia and sleep fragmentation. Snoring history, craniofacial features, obesity class and polycystic ovary syndrome can guide objective testing with polysomnography or validated home sleep apnoea testing before stimulation, particularly when patients report marked daytime sleepiness or a bed partner observes apnoeas. Where apnoea is confirmed, early treatment with continuous positive airway pressure in combination with weight management may improve vascular and metabolic profiles that support ovarian response and endometrial receptivity; however, randomised trials are still needed to determine whether treating apnoea itself improves clinical pregnancy or live birth after IVF/ICSI.

By contrast, when poor subjective sleep quality or extreme sleep duration primarily reflect anxiety, depressive symptoms, or irregular schedules during treatment, integrated care that prioritises psychological support, optimisation of circadian regularity, and healthy lifestyle change may be more realistic intervention targets than attempting to “treat sleep” in isolation. At present, the observational data do not demonstrate that manipulating sleep duration or perceived sleep quality per se will improve assisted reproduction outcomes, and it may be most appropriate to regard these sleep measures as potentially modifiable markers of broader metabolic and psychosocial vulnerability rather than as proven causal levers.

Our study offers several strengths while acknowledging clear limits. We focused on clinically relevant endpoints, synthesizing both clinical pregnancy and live birth, and restricted inclusion to prospective cohorts with adjusted estimates. Random effects models were prespecified to reflect clinical diversity, and fixed effect results were shown for comparison, with sensitivity and influence analyses supporting the stability of the main findings. We also examined a metabolically vulnerable subgroup with polycystic ovary syndrome, increasing clinical relevance. Limitations arise from small study numbers for some outcomes, moderate heterogeneity for analyses using subjective sleep quality, and variation in sleep assessment tools and timing across cohorts, which invites exposure misclassification. Wide prediction intervals, together with the small number of cohorts, limit confidence in the generalizability of these pooled estimates to new populations and clinical settings. This is particularly relevant for the PCOS-restricted analyses and for live birth, where prediction intervals span values below and above one. We did not formally rate the certainty of evidence using a framework such as GRADE. Therefore, the strength of evidence for each pooled association should be interpreted cautiously. Generalizability is tempered by geographic concentration, incomplete control of lifestyle and psychosocial confounders, and the fact that most cohorts reported outcomes per embryo transfer without distinguishing between fresh and frozen cycles, which prevented systematic comparison of associations by transfer type, even though success rates may differ between fresh and frozen embryo transfer cycles. In addition, our searches were primarily conducted using English-language terms, and we restricted inclusion to reports with full texts available in English to ensure consistent interpretation of methodological details. Consequently, relevant cohorts published only in other languages may have been missed, introducing potential language or availability bias. To assess this, we screened records with English abstracts that might represent non-English or inaccessible full texts. One potentially relevant study [27] was identified, but its full text could not be obtained despite interlibrary loan attempts, preventing definitive verification of its language and eligibility. Given the scarcity of such instances, we believe this exclusion does not materially alter our overall conclusions (Table 1).

Table 1.

Characteristics of Included Studies on Sleep and IVF/ICSI Outcomes

#	Study (year, country)	Design	N (population)	Sleep exposure	Key estimate(95%Cl)	NOS
Sleep-disordered breathing (OSA/SDB, including snoring)
1	Wang (2025, China) [28]	Prospective cohort	632 women undergoing IVF/ICSI	Self-reported snoring frequency	↓ number of usable embryos (β − 1.1, 95% CI − 2.2 to − 0.1); ↑ biochemical pregnancy loss (aOR 2.95, 95% CI 1.06–8.24)	8
2	Li (2025, China) [29]	Prospective cohort	360 women with PCOS scheduled for IVF	OSA diagnosed by home sleep-apnoea testing (HSAT)	OSA vs no OSA:↓ live birth at first embryo transfer (adjusted RR 0.60, p < 0.05); ↓ 18-month cumulative live birth (45.7% vs 62.9%; HR 0.65, p < 0.05)	7
3	Zhang (2024, China) [15]	Prospective cohort	156 women with PCOS undergoing first IVF	OSA diagnosed by overnight AHI using a radar monitor	OSA vs no OSA: ↓ clinical pregnancy after first embryo transfer (aOR 0.38, 95% CI 0.15–0.97; p = 0.043)	9
4	Walter (2022, USA) [8]	Prospective cohort	30 women (29 received an embryo transfer)	OSA assessed by a wearable sensor	OSA (AHI ≥ 5 vs < 5): ↓ clinical pregnancy (35% vs 58%; RR 0.60, 95% CI 0.27–1.35) and ↓ live birth (38% vs 58%; OR 0.39, 95% CI 0.09–1.78); age-adjusted OR for live birth 0.23 (95% CI 0.04–1.50; NS)	6
Subjective sleep quality (Pittsburgh Sleep Quality Index, PSQI)
5	Liu (2023, China) [30]	Prospective cohort	3,183 women undergoing first IVF	PSQI; chronotype by sleep-midpoint; weekly sleep-duration categories	Good sleep quality (PSQI ≤ 5 vs > 5): ↑ clinical pregnancy (RR 1.07, 95% CI 1.01–1.13) and ↑ live birth (RR 1.12, 95% CI 1.02–1.23); weekly sleep-duration categories: ↔ clinical pregnancy and live birth	8
6	Yang (2022, China) [31]	Prospective cohort	1,344 singleton IVF/ICSI pregnancies	PSQI assessed before retrieval and in each trimester	Poor sleep in the 2nd trimester (PSQI-defined, vs good): ↑ SGA overall (aOR 2.19) and ↑ SGA among female infants (aOR 3.03)	9
7	Li (2023, China) [6]	Prospective cohort	1,002 infertile women planning first IVF/ICSI	Global and component PSQI scores	Poor subjective sleep vs good: ↓ retrieved oocytes (IRR 0.85) and ↓ mature (MII) oocytes (IRR 0.82); sleep disturbance: ↓ fertilisation rate (IRR 0.96); sleep efficiency < 85% vs ≥ 85%: ↓ clinical pregnancy (OR 0.51)	7
8	Reschini (2022, Italy) [32]	Prospective cohort	263 women starting IVF	PSQI	Poor sleep quality (PSQI > 5 vs ≤ 5): ↓ clinical pregnancy (aOR 0.48, 95% CI 0.25–0.92; p = 0.03)	7
9	Bariya (2025, China) [33]	Prospective cohort	174 women starting first IVF/ICSI	PSQI; nocturnal sleep duration; difficulty initiating sleep; daytime napping	Poor sleep quality (PSQI > 5 vs ≤ 5): ↓ retrieved oocytes (− 22.9%, 95% CI − 37.8 to − 4.0) and ↓ mature (MII) oocytes (− 22.0%, 95% CI − 37.5 to − 2.6); very long sleep (≥ 10 h/night vs 7– < 8 h): ↓ retrieved oocytes (− 30.7%) and ↓ good-quality embryos (− 46.3%); daytime napping > 1 h/day: ↓ maturation rate (− 73.8%)	8
10	Philipsen (2021, Denmark) [34]	Prospective cohort	163 women undergoing IVF/ICSI at three public clinics	Self-reported sleep quality (PSQI ≤ 5, 6–10, ≥ 11) and sleep duration	Compared with good sleepers (PSQI ≤ 5): poor sleepers (6–10) and very poor sleepers (≥ 11): ↓ clinical pregnancy (aOR 0.42 and 0.28), but 95% CIs crossed 1 (NS)	6
Sleep duration and actigraphy-based timing/continuity
11	Yao (2022, China) [11]	Prospective cohort	1,276 women undergoing first IVF/ICSI	Nocturnal sleep duration (< 7, 7– < 8, 8– < 9, 9– < 10, ≥ 10 h); simplified PSQI items on quality, timing, insomnia	Short sleep < 7 h vs 7– < 8 h: ↓ retrieved oocytes (− 11.5%, 95% CI − 21.3 to − 0.5); long sleep 9– < 10 h vs 7– < 8 h: ↓ clinical pregnancy (aOR 0.65, 95% CI 0.44–0.98)	9
12	Gao (2024, China) [20]	Prospective cohort	856 women who conceived via ART and were followed from early pregnancy	Night-time sleep duration (≤ 7, 7–8, 8–9, 9–10, ≥ 10 h); PSQI	Night-time sleep ≥ 10 h vs 7–8 h: ↑ GDM (adjusted OR 2.01, 95% CI 1.02–3.93), stronger in women < 35 years (OR 2.57, 95% CI 1.21–5.47); PSQI-defined poor sleep vs good: ↔ GDM	8
13	Pimolsri (2021, USA) [35]	Prospective cohort	48 women undergoing first IVF cycle	Wrist actigraphy: total sleep time, sleep midpoint, bedtime	Actigraphy: per + 20 min of nocturnal sleep: ↓ cycle cancellation (OR 0.88); later sleep midpoint and bedtime: ↑ cycle cancellation (OR 1.24 and 1.33, respectively)	7
14	Goldstein (2017, USA) [36]	Prospective cohort	22 women (24 IVF cycles; 19 cycles for primary outcome)	Actigraphy-measured baseline total sleep time (TST)	Baseline total sleep time by actigraphy: ↑ TST associated with ↑ number of oocytes retrieved (adjusted R2 0.40, p = 0.03; ≈ + 1.5 oocytes per extra hour of sleep); sleep duration and daytime sleepiness both changed across the cycle (TST p = 0.04; ESS p = 0.02)	5

IVF/ICSI in-vitro fertilization/intracytoplasmic sperm injection, PSQI Pittsburgh Sleep Quality Index, OSA obstructive sleep apnea, SDB sleep-disordered breathing, AHI apnea–hypopnea index, HSAT home sleep-apnea test, TST total sleep time, MII metaphase-II oocytes, OGTT oral glucose tolerance test, GDM gestational diabetes mellitus, SGA small-for-gestational-age, aOR adjusted odds ratio, IRR incidence rate ratio, β regression coefficient, CI confidence interval, wk week, h hour, y year. Definitions: Poor sleep by PSQI was defined as a global score > 5 when applicable. AHI categories followed standard cut-points (mild 5–14.9, moderate 15–29.9, severe ≥ 30 events·h⁻¹) unless specified by the study. Quality: NOS denotes Newcastle–Ottawa Scale score as reported by each study. ↑ indicates higher values or increased risk of the outcome relative to the reference category; ↓ indicates lower values or reduced risk; ↔ indicates no clear association or difference; NS, not statistically significant

Future research should move beyond association to proximate mechanisms and interventional leverage points. First, randomized trials that treat objectively diagnosed apnea before or during stimulation with adherence-verified continuous positive airway pressure and prespecified reproductive endpoints will test causal effects and identify which patients benefit most. Second, time-resolved phenotyping across a cycle that couples actigraphy or polysomnography with follicular fluid metabolomics, redox markers, and endometrial transcriptomics will map the sequence from sleep exposure to oocyte competence and implantation biology. Third, individual participant data meta-analyses with standardized exposure windows and covariates can reconcile subjective measures across clinics and produce prediction models that combine sleep with established prognostic factors. Together, these steps can convert sleep from a descriptive risk factor into a practical component of preconception optimization in assisted reproduction.

Conclusion

This review shows that objectively diagnosed obstructive sleep apnea is consistently associated with poorer IVF and ICSI outcomes, including lower odds of clinical pregnancy and live birth, with effects that persist in women with polycystic ovary syndrome. Subjective sleep quality relates to outcomes under a fixed effect model but attenuates once between-study heterogeneity is considered, and findings for sleep duration suggest a nonlinear pattern rather than a simple dose response. Together, these results position sleep as a clinically relevant exposure rather than a background wellness factor. From a clinical perspective, clinicians may consider asking about symptoms suggestive of sleep-disordered breathing when clinically indicated, particularly among patients with obesity, polycystic ovary syndrome, or habitual snoring. Routine targeted sleep screening as a standard component of infertility evaluation cannot be recommended on the basis of the available data. For research, trials that treat obstructive sleep apnea before or during stimulation, studies that couple objective sleep phenotyping with follicular and endometrial biomarkers, and standardized exposure windows will be decisive for translating sleep assessment into reproducible gains in implantation and live birth.

Authors’ contributions

Y.L.H., J.L., and Y.W. jointly conceived and designed the study. Y.L.H. drafted the main manuscript and conducted the core statistical analysis and meta-analysis. Y.L.H. and J.L. performed the literature search, data extraction, and quality assessment. J.L. drafted specific sections and assisted with Table 1. N.D.L. and Y.X.Z. interpreted results and provided critical revisions. Y.W. (corresponding author) supervised the study and finalized the manuscript. All authors read and approved the final manuscript.

Funding

This work was supported by the National Key Research and Development Program of China (2022YFC2702905).

Data availability

All data were extracted from the published articles cited in this manuscript. The data-extraction sheet and analysis code are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

Not applicable. This systematic review and meta-analysis synthesized data from previously published studies and did not involve the recruitment of human participants by the authors; therefore, institutional ethics approval and participant consent were not required. The review protocol was prospectively registered in PROSPERO (CRD420251118366).

Consent for publication

Not applicable. This study includes no identifiable individual person’s data.

Competing interests

The authors declare no competing interests.