The bidirectional causal relationships of insomnia with eleven cancers

Abstract

Insomnia is closely associated with tumors and is recognized as a risk factor for cancer development. This two-sample Mendelian randomization (MR) study investigated the potential causal link between genetically determined insomnia and various cancers, including brain, colorectal, esophageal, gastric, hepatic bile duct, liver, lung, malignant lymphoma, pancreatic, skin, and thyroid cancers. Summary statistics from genome-wide association studies for insomnia and eleven cancer types were examined. instrumental variables (IVs) were selected from single nucleotide polymorphisms that passed strict quality control for minimal linkage disequilibrium and genome-wide significance. All IVs demonstrated strong predictive power for insomnia, with F-statistics exceeding the conventional threshold of 10. The primary analysis was conducted using the inverse variance weighted method. Sensitivity analyses were performed using MR-Egger regression, weighted median, and weighted mode methods to test robustness, pleiotropy, and heterogeneity. MR analysis suggested potential causal associations between insomnia and increased risk of hepatic bile duct cancer (odds ratio [OR] = 4.458, 95% confidence interval [CI]: 1.074–18.493, P = .040), reduced risk of hepatic cancer (OR = 0.312, 95% CI: 0.100–0.973, P = .045), and reduced risk of skin cancers (OR = 0.599, 95% CI: 0.439–0.818, P = .001). However, after Bonferroni correction (adjusted threshold P < .0045), these associations did not retain statistical significance. No notable causal influence of these cancers on insomnia was observed. Sensitivity analyses showed no evidence of reverse causality or significant bias. These findings, though not retaining statistical significance after stringent multiple testing correction, provide suggestive evidence that genetically predicted insomnia may influence the risk of certain cancers, potentially through mechanisms involving genetic predisposition and biological pathways that modulate inflammation, immune function, and circadian rhythms. The study highlights the translational value of understanding insomnia’s role in cancer risk, suggesting that clinicians should consider this association when managing patients with insomnia. Further research with larger sample sizes is warranted to validate these findings.

1. Introduction

Insomnia is characterized by persistent difficulty in starting or maintaining sleep, and it not only impairs quality of life but also poses significant health risks. In recent years, insomnia has received considerable attention for its potential role in the development of various cancers, notably ovarian cancer and breast cancers.^[1–3] The relationship between sleep and the circadian system is closely associated with hormonal production, metabolic processes, and immunoregulatory mechanisms, all of which play a significant role in the initiation and advancement of cancer.^[4] The prevalence of sleep disturbances is notably higher among patients with cancer compared to the general population.^[5,6] Among these sleep issues, insomnia emerges as a predominant condition for those diagnosed with cancer.^[1] The potential connection between insomnia and oncogenesis has garnered substantial interest within the scientific community. Given cancer’s multifaceted nature and complex origins, it is crucial to elucidate how genetic mechanisms may contribute to carcinogenesis through disrupted sleep patterns, thereby advancing preventive and therapeutic strategies.

To address this knowledge gap, our study utilizes the powerful tool of Mendelian randomization (MR). The MR analysis was proposed as a method for precisely assessing the potential causal link between an exposure and an outcome.^[7–9] The MR approach leverages genetic differences as instrumental variables (IVs) to mitigate the effects of unaccounted factors on the linkage between the exposure and the consequent outcome. This approach is supported by the following evidence: Mendel second law stipulates that the stochastic arrangement of genetic variants during the process of gamete formation results in a random distribution of exposure tied to one or multiple genetic variants.

This process occurs independently of environmental risk factors, which precede disease progression.^[10] MR provides a robust framework to overcome the intrinsic challenges of observational research, especially those concerning uncontrolled variables and the directionality of cause and effect. This enables a stringent examination of the causality linking insomnia to malignancies. Furthermore, the MR technique is less prone to issues of confounding, bidirectional causality, and assorted biases.^[11] Consequently, genetic variants are employed as IVs in MR analysis, thereby rendering the method less vulnerable to reverse causation and potential confounders. In addition, data from previous genome-wide association studies (GWAS) enable the investigation of potential links and causal effects of insomnia on cancer development.

In our research, we employed a two-sample MR analysis to investigate the possible causal connections between insomnia and eleven types of cancer, including brain tumors, colorectal, esophageal, gastric, hepatic bile duct, hepatic, and lung cancers, malignant lymphoma, pancreatic, skin, and thyroid cancers, using published data from GWASs.

We hypothesized that genetically determined insomnia may exert a causal influence on the risk of developing specific cancers through mechanisms involving genetic predisposition and biological pathways that modulate inflammation, immune function, and circadian rhythms. Specifically, we anticipated observing a heightened risk for certain cancers, such as hepatic bile duct cancer, resulting from the potential disruption of these pathways by insomnia. Conversely, we also hypothesized that insomnia might be associated with a reduced risk for some cancers, such as skin cancers, potentially due to compensatory mechanisms or differential genetic influences. Elucidating these potential causal relationships and underlying mechanisms is crucial for developing targeted prevention and therapeutic strategies for cancer patients with insomnia.

2. Methods

2.1. Study design

We conducted a MR study to examine the causal relationship between genetically determined insomnia (with single nucleotide polymorphisms [SNPs] as the exposure factor) and cancer risk (with SNPs as the outcome) through bidirectional analyses. The aim of this research is to establish if there exists a correlation between genetic predisposition to insomnia and heightened cancer risk, as inferred from forward MR analysis.

The primary objective of this research was to determine whether a genetic predisposition to insomnia is associated with an increased risk of cancer, as inferred from forward MR analysis. Furthermore, a reverse MR analysis was performed to ascertain whether genetic predisposition to cancer influences the likelihood of developing insomnia. The schematic of the study is depicted in Figure 1.

Figure 1.

The study design of the Mendelian randomization analysis on the causal associations between insomnia and eleven cancers.

2.2. Data source

Genomic analyses for insomnia were conducted using data from the UK Biobank. Summary statistics for insomnia were derived from a study by Ben Elsworth and colleagues.^[12] A total of 462,341 participants were included in the GWAS. Insomnia was assessed by querying the presence of difficulties initiating sleep at night or waking up during the night (data field: 1200) among subjects in the UK Biobank.

Summary data for a comprehensive set of eleven cancers, including brain tumor, colorectal, esophageal, gastric, hepatic bile duct, hepatic, lung cancers, malignant lymphoma, pancreatic, skin, and thyroid cancer, were sourced from GWAS summary data (Sakaue S et al, PubMed ID:34594039).^[12] Table 1 shows further details of the eleven cancers. The selection of these cancers was based on their prevalence and clinical significance in oncology, as well as the availability of high-quality GWAS data (Sakaue S et al, PubMed ID:34594039).^[12] These cancers were chosen to represent a broad spectrum of malignancies with varying etiologies and genetic underpinnings, thereby allowing us to comprehensively investigate the potential causal links between insomnia and diverse cancer types.

Table 1.

Eleven cancers used in this study (PMID: 34594039).

Trait	GWAS ID	Year	Cases	Controls	Sample size	Number of SNPs	Build	Download link
Brain tumor	ebi-a-GCST90018800	2021	833	490,709	491,542	24,198,642	HG19/GRCh37	https://gwas.mrcieu.ac.uk/datasets/ebi-a-GCST90018800/
Colorectal cancer	ebi-a-GCST90018808	2021	6581	463,421	470,002	24,182,361	HG19/GRCh37	https://gwas.mrcieu.ac.uk/datasets/ebi-a-GCST90018808/
Esophageal cancer	ebi-a-GCST90018841	2021	998	475,308	476,306	24,194,380	HG19/GRCh37	https://gwas.mrcieu.ac.uk/datasets/ebi-a-GCST90018841/
Gastric cancer	ebi-a-GCST90018849	2021	1029	475,087	476,116	24,188,662	HG19/GRCh37	https://gwas.mrcieu.ac.uk/datasets/ebi-a-GCST90018849/
Hepatic bile duct cancer	ebi-a-GCST90018803	2021	832	475,259	476,091	24,196,592	HG19/GRCh37	https://gwas.mrcieu.ac.uk/datasets/ebi-a-GCST90018803/
Hepatic cancer	ebi-a-GCST90018858	2021	379	475,259	475,638	24,194,938	HG19/GRCh37	https://gwas.mrcieu.ac.uk/datasets/ebi-a-GCST90018858/
Lung cancer	ebi-a-GCST90018875	2021	3791	489,012	492,803	24,188,684	HG19/GRCh37	https://gwas.mrcieu.ac.uk/datasets/ebi-a-GCST90018875/
Malignant lymphoma	ebi-a-GCST90018878	2021	3546	487,257	490,803	24,193,511	HG19/GRCh37	https://gwas.mrcieu.ac.uk/datasets/ebi-a-GCST90018878/
Pancreatic cancer	ebi-a-GCST90018893	2021	1196	475,049	476,245	24,195,229	HG19/GRCh37	https://gwas.mrcieu.ac.uk/datasets/ebi-a-GCST90018893/
Skin cancer	ebi-a-GCST90018921	2021	25,928	466,275	492,203	24,178,924	HG19/GRCh37	https://gwas.mrcieu.ac.uk/datasets/ebi-a-GCST90018921/
Thyroid cancer	ebi-a-GCST90018929	2021	1054	490,920	491,974	24,198,226	HG19/GRCh37	https://gwas.mrcieu.ac.uk/datasets/ebi-a-GCST90018929/

GWAS = genome-wide association studies, SNP = single-nucleotide polymorphism.

To reduce the risk of a type 1 error, the participant count in the MR analysis for the exposure and outcome groups was strictly limited.

2.3. Using SNPs to select IV

The foundational premise of MR posits that genetic IVs should exhibit a substantial correlation with the exposure.^[13] The criteria for IV selection per trait stipulated that the pertinent SNP surpassed the stringent genome-wide significance level (P < 5E−08) and maintained minimal linkage disequilibrium with proximal SNPs within a 10,000 kb (kilobases) radius (r² < 0.001).

For cancers with a limited number of genome-wide significant SNPs due to smaller sample sizes (e.g., brain, esophageal, hepatic bile duct, and hepatic cancers, as well as malignant lymphoma, pancreatic, and thyroid cancers), we applied a more relaxed instrument selection threshold (P < 5 × 10⁻⁶) to include a sufficient number of genetic variants for a robust MR analysis, following established practices in the field. To ensure consistency, the effects were harmonized to ensure that the impact assessments in both the exposure and outcome GWAS corresponded to identical alleles for each SNP.

2.4. MR analysis and sensitivity analysis

To validate the secondary and tertiary MR assumptions, it is necessary to presume that the IVs in question are irrelevant to confounding factors and that IVs affect outcomes solely through exposure.^[14] The 2 aforementioned assumptions are collectively referred to as the no-pleiotropy assumption.^[15] In the context of MR analysis, we employed a range of analytical methods, including the inverse variance weighting (IVW) method, MR-Egger regression, the weighted median estimator, and the weighted mode estimator. The IVW method was the primary analytical approach used in this MR analysis. MR-Egger regression was conducted to assess the presence of horizontal pleiotropy, which refers to the situation where genetic variants affect the outcome through pathways other than the exposure of interest. Heterogeneity was tested using both MR-Egger regression and the IVW method, with Cochran Q statistics employed to quantify the degree of heterogeneity among the causal estimates from different genetic variants. Outliers were identified and removed using the MR-PRESSO technique, which stands for Mendelian randomization pleiotropy residual sum and outlier. Additionally, we implemented a leave-one-out analysis to evaluate the influence of a particular SNP on causality and to quantify the effect of each SNP on the causal pathway.^[16] A P-value exceeding .05 signifies the lack of heterogeneity and horizontal pleiotropy.

2.5. Statistical method

The R package of TwoSampleMR (v0.6.3), which is part of the R software (v4.4.0) and is installed on the RStudio platform, was employed in the MR analyses. The findings were presented adhering to the STROBE-MR criteria, which stands for the Strengthening the Reporting of Observational Studies in Epidemiology using Mendelian Randomization.^[17] All hypotheses were assessed using a two-tailed test, with statistical significance defined as P < .05.

In the absence of pleiotropy signals for the chosen IVs (with an MR-Egger intercept P-value exceeding .05), the IVW method is deemed the most robust analytical technique. It is possible that the results may be affected by weak IVs. Consequently, the F score was employed to ascertain whether the selected SNPs were influenced by the weak IVs.

To mitigate the possible effects of weak instrument bias on causality assessments, we evaluated the potency of the genetic instruments across all retained SNPs. This assessment was performed by calculating the F statistic for each SNP, approximated using the formula F = (β/SE)², which provides a reliable estimate of instrument strength in large-sample GWAS settings.^[13] An F statistic exceeding 10 is regarded as a robust instrumental variable, indicating a low risk of weak instrument bias in MR studies.^[18] The selected SNPs with strong F statistics were then used in subsequent analyses. The F-statistics for the insomnia exposure are reported in the Results section, and those for the cancer exposures in the reverse MR analysis are provided in Supplementary Table S3, Supplemental Digital Content 1.

Given that eleven cancer outcomes were tested in the forward MR analysis, we applied a Bonferroni correction to account for multiple testing. A P-value <.0045 (0.05/11) was considered statistically significant. Results with a P-value between .0045 and .05 were considered suggestive of an association.

2.6. Ethics approval

This study utilizes publicly accessible data. The summary statistics regarding insomnia and various cancers are devoid of any personal identifying information. Ethical clearance and informed consent were obtained for the initial GWASs.

2.7. Data availability

Summary statistics for insomnia and eleven cancers were downloaded from the Integrative Epidemiology Unit’s Open GWAS database, which is accessible at the provided link (https://gwas.mrcieu.ac.uk/), with the data being current as of June 7, 2024. The GWAS ID for sleepless/insomnia is ukb-b-3957. The GWAS IDs of eleven cancers were detailed in Table 1. The outcomes of the bidirectional MR analysis, including tests for heterogeneity (Q-statistics), the Egger regression intercept, the MR-PRESSO global test, and the leave-one-out sensitivity analyses, were summarized in Tables 2 and 3 and Supplement Table S2, Supplemental Digital Content 2, as well as Figures 2–4. The relevant codes can be provided upon reasonable request.

Table 2.

Causal estimates for insomnia and eleven cancers by MR.

Phenotype		MR results
Exposure	Outcome	N SNPs	(IVW) OR (95% CI)	(IVW) P	P for MR-egger intercept	Heterogeneity test Cochran Q (P)
Insomnia	Brain tumor	39	1.994 (0.467–8.500)	.351	.335	43.057 (.264)
	Colorectal cancer	39	0.738 (0.478–1.137)	.168	.395	32.623 (.716)
	Esophageal cancer	39	0.753 (0.223–2.536)	.647	.074	50.350 (.087)
	Gastric cancer	39	1.057 (0.554–2.019)	.866	.966	42.386 (.287)
	Hepatic bile duct cancer	39	4.458 (1.074–18.493)	.040*	.028	41.477 (.322)
	Hepatic cancer	39	0.312 (0.100–0.973)	.045*	.062	39.911 (.385)
	Lung cancer	39	1.750 (0.994–3.081)	.053	.156	35.103 (.604)
	Malignant lymphoma	39	0.646 (0.310–1.343)	.242	.359	33.739 (.667)
	Pancreatic cancer	39	1.596 (0.501–5.085)	.429	.089	33.664 (.670)
	Skin cancer	36	0.599 (0.439–0.818)	.001*	.107	33.882 (.522)
	Thyroid cancer	39	0.586 (0.167–2.054)	.404	.076	29.682 (.831)

CI = confidence interval, IVW = inverse-variance weighted, MR = Mendelian randomization, N SNPs = number of SNPs retained and used in the MR analysis after filtered by MR-PRESSO outlier analysis, OR = odds ratio, P = P-value for the causal estimates, P for MR-egger intercept: P-value for intercept of MR-egger regression.

^*P < .05.

Table 3.

Causal estimates for eleven cancers and insomnia and by MR.

Phenotype		MR results
Exposure	Outcome	N SNPs	(IVW) OR (95% CI)	(IVW) P	P for MR-egger intercept	Heterogeneity test Cochran Q (P)
Brain tumor	Insomnia	15	1.001 (0.998–1.004)	.660	.557	12.623 (.556)
Colorectal cancer		25	1.001 (0.994–1.007)	.803	.683	25.105 (.400)
Esophageal cancer		23	1.000 (0.995–1.004)	.839	.244	43.614 (.004)
Gastric cancer		6	1.006 (0.992–1.016)	.557	.280	10.374 (.065)
Hepatic bile duct cancer		8	0.998 (0.992–1.004)	.569	.135	13.679 (.057)
Hepatic cancer		23	0.999 (0.996–1.002)	.570	.559	32.689 (.066)
Lung cancer		10	0.994 (0.986–1.004)	.230	.375	4.748 (.856)
Malignant lymphoma		17	1.001 (0.995–1.007)	.806	.646	17.933 (.328)
Pancreatic cancer		18	0.998 (0.993–1.003)	.510	.247	38.776 (.002)
Skin cancer		52	0.997 (0.992–1.002)	.304	.245	63.562 (.111)
Thyroid cancer		13	1.000 (0.997–1.004)	.893	.181	13.451 (.337)

CI = confidence interval, IVW = inverse-variance weighted, MR = Mendelian randomization, N SNPs = number of SNPs retained and used in the GSMR analysis after filtered by HEIDI outlier analysis, OR = odds ratio, P = P-value for the causal estimates; P for MR-egger intercept: P-value for intercept of MR-egger regression.

Figure 2.

Association between insomnia and hepatic bile duct cancer estimated by MR. (A) Scatter plot showing the effect of genetic instruments on insomnia risk against their effect on hepatic bile duct cancer. (B) The funnel plot showed no asymmetry. (C) Forest plot of hepatic bile duct cancer for each 1 SD increase of insomnia risk. (D) There was no substantial change in the IVW causal estimate after removing any of the instrumental SNPs. IVW = inverse variance weighted, MR = Mendelian randomization, MR-PRESSO = MR pleiotropy residual sum and outlier, SNP = single-nucleotide polymorphism.

Figure 4.

Association between insomnia and skin cancer estimated by MR. (A) Scatter plot showing the effect of genetic instruments on insomnia risk against their effect on skin cancer. (B) The funnel plot showed no asymmetry. (C) Forest plot of skin cancer for each 1 SD increase of insomnia risk. (D) There was no substantial change in the IVW causal estimate after removing any of the instrumental SNPs. IVW = inverse variance weighted, MR = Mendelian randomization, MR-PRESSO = MR pleiotropy residual sum and outlier, SNP = single-nucleotide polymorphism.

3. Results

3.1. Causal effects of insomnia on cancers

All selected SNPs had F-statistics > 10 (ranging from 29.85 to 199.35; mean F-statistic = 44.93 for the insomnia exposure), indicating that weak instrument bias is unlikely to substantially influence the causal estimates (see Supplementary Table S1, Supplemental Digital Content 3 for details). In accordance with strict quality control protocols and standard procedures outlined in the Methods section, forty-two SNPs were chosen to serve as IVs (Supplement Table S1, Supplemental Digital Content 3). During the harmonization process, SNPs rs2644128, rs2803296, and rs8180817 were excluded due to their palindromic nature and intermediate allele frequencies. As demonstrated in Table 2, the IVW method suggested a potential causal association between genetically predicted insomnia and an elevated risk of hepatic bile duct cancer (odds ratio [OR] = 4.458, 95% confidence interval [CI]: 1.074–18.493, P = .040) (Fig. 2). However, after applying a Bonferroni correction for multiple testing (significance threshold P < .0045), this association, along with those observed for hepatic cancer (OR = 0.312, 95% CI: 0.100–0.973, P = .045) and skin cancer (OR = 0.599, 95% CI: 0.439–0.818, P = .001), did not reach the stringent significance level and should be interpreted as suggestive findings. Insomnia was not significantly associated with the risk of the other 8 cancers (IVW, P > .05) (Table 3).

Our study aimed to identify extensive pleiotropic effects in the MR-derived causal correlations between insomnia and hepatic bile duct, hepatic, and skin cancers. The MR-Egger intercept test and the MR-PRESSO approach were applied to detect potential pleiotropic influences.

For MR analysis of insomnia’s effect on hepatic bile duct cancer, the global test of MR-PRESSO demonstrated no significant horizontal pleiotropy (P = .322) (Supplement Table S2, Supplemental Digital Content 2). Similarly, the MR-Egger intercept test and the MR-PRESSO test showed no significant horizontal pleiotropy in the analyses of hepatic cancer and skin cancer (P > .05) (Table 2 and Supplement Table S2, Supplemental Digital Content 2).

The leave-one-out method indicated that the causal links were not influenced by any single SNP (Fig. 3). There was no heterogeneity detected in the MR analysis of hepatic bile duct cancer, hepatic cancer, and skin cancer (P > .05) (Supplement Table S2, Supplemental Digital Content 2).

Figure 3.

Association between insomnia and hepatic cancer estimated by MR. (A) Scatter plot showing the effect of genetic instruments on insomnia risk against their effect on hepatic cancer. (B) The funnel plot showed no asymmetry. (C) Forest plot of hepatic cancer for each 1 SD increase of insomnia risk. (D) There was no substantial change in the IVW causal estimate after removing any of the instrumental SNPs. IVW = inverse variance weighted, MR = Mendelian randomization, MR-PRESSO = MR pleiotropy residual sum and outlier, SNP = single-nucleotide polymorphism.

Insomnia was not significantly linked to the development risk of the other 8 cancers, including brain tumor, colorectal, esophageal, gastric, lung cancer, malignant lymphoma, pancreatic, and thyroid cancer (IVW, P > .05) (Table 3).

The final set of IVs exhibited no signs of pleiotropy, as evidenced by a non-significant MR-Egger intercept (P > .05). This indicates that insomnia does not have a causal influence on the risk of brain tumor, colorectal, esophageal, gastric, lung cancer, malignant lymphoma, pancreatic, and thyroid cancers (as detailed in Table 2).

3.2. Causal effects of cancers on insomnia

In the reverse MR analysis, all genetic instruments for each cancer type had F-statistics > 10, with ranges and mean values presented in Supplementary Table S3, Supplemental Digital Content 1, indicating that weak instrument bias was unlikely to affect the causal estimates.

An inverse MR analysis was conducted to investigate whether eleven types of cancer have causal impacts on the development of insomnia. Genetic variants for eleven cancers were used as exposures. The MR analysis showed no significant influence of the eleven cancers on insomnia, with all P-values exceeding .05 (as detailed in Table 3). The MR findings suggest the absence of a causal link between these cancers and insomnia.

MR-Egger regression analyses found no evidence of horizontal pleiotropy in the relationship between the eleven cancers and insomnia, with non-significant intercept P-values (all > .05) (as detailed in Table 3). Therefore, no causality relationship was observed between cancers and insomnia risk (as detailed in Table 3).

4. Discussion

In this study, we employed GWAS summary data to elucidate the bidirectional causal relationships between insomnia and eleven cancer types. Our findings provide suggestive evidence that genetic liability to insomnia may exert a causal influence on the risk of developing hepatic bile duct cancer, hepatic cancer, and skin cancer. Conversely, insomnia risk does not appear to be significantly impacted by these 3 cancers. Furthermore, no significant bidirectional causal relationships were observed between insomnia and the remaining 8 cancers, including brain tumor, colorectal cancer, esophageal cancer, gastric cancer, lung cancer, malignant lymphoma, pancreatic cancer, and thyroid cancer.

Insomnia emerges as a potential risk factor for hepatic bile duct cancer. Given its profound impact on multiple vital physiological processes, the importance of sleep in maintaining human health cannot be overstated.^[19] Sleep regulation is a complex phenomenon, meticulously managed through intricate physiological mechanisms.^[20] For instance, insomnia may lead to immunosuppression, thereby hindering the body’s capacity to eliminate potential carcinogens. Furthermore, persistent insomnia fosters a pro-inflammatory state, which has been identified as a key facilitator of oncogenesis across various malignancies. Additionally, hormonal imbalances associated with insomnia, notably elevated cortisol levels, may contribute to tumor progression. Circadian disruption, which is closely intertwined with insomnia, can influence cancer susceptibility through the regulation of deoxyribonucleic acid damage repair mechanisms and apoptotic processes.^[20] Circadian misalignment has been linked to several cancer types.^[21] These mechanistic insights support the plausibility of the observed positive association between insomnia and hepatic bile duct cancer and suggest that early detection and management of sleep disturbances could have preventive implications.

Our finding that genetic liability to insomnia is associated with a reduced risk of hepatic and skin cancers stands in contrast to much of the observational literature, which generally reports a positive or null association. This discrepancy warrants careful consideration of the underlying mechanisms and potential limitations of the MR framework. One plausible explanation is horizontal pleiotropy, whereby the genetic variants used as instruments for insomnia influence cancer risk through pathways independent of sleep disturbance. For instance, SNPs associated with insomnia are also implicated in immune regulation, cortisol secretion, and metabolic syndrome – all of which play complex and sometimes opposing roles in different cancer types. This pleiotropic effect could potentially explain the divergent direction of effects observed between hepatic bile duct cancer (increased risk) and hepatocellular carcinoma (decreased risk). Although our sensitivity analyses (MR-Egger intercept, MR-PRESSO) did not detect significant horizontal pleiotropy, these methods primarily address directional pleiotropy. The possibility of balanced pleiotropy, where multiple causal pathways cancel each other out, cannot be entirely ruled out and may contribute to the observed null or protective associations.

The opposing causal effects observed for hepatic bile duct cancer (increased risk) and hepatocellular carcinoma (decreased risk) are particularly intriguing given their shared anatomical location. This divergence likely reflects the distinct etiological and genetic architectures of these 2 malignancies. Cholangiocarcinoma arises from the biliary epithelium – a tissue highly susceptible to inflammatory insults driven by bile acid reflux, microbial translocation, and cytokine-mediated injury – whereas hepatocellular carcinoma originates from hepatocytes and is more closely associated with metabolic dysfunction, steatosis, and viral hepatitis, where the tumor microenvironment is shaped more by fibrotic remodeling and immune exhaustion than by direct ductal inflammation.^[22,23] It is plausible that insomnia-related genetic variants exert pleiotropic effects on pathways such as bile acid metabolism or specific inflammatory cytokines, which could differentially influence these 2 cell types or their microenvironments. Specifically, the biliary epithelium’s direct exposure to the gut–liver axis and its inherent sensitivity to inflammatory signals may render it more vulnerable to insomnia-related circadian and immune perturbations, while hepatocytes, embedded within a metabolically complex microenvironment, might be subject to counterbalancing protective mechanisms. Such effects may promote cholangiocarcinogenesis while having a neutral or protective effect on hepatocytes. This hypothesis, while speculative, underscores the complexity of the relationship and warrants further investigation at the molecular level, particularly in experimental models that distinguish between these 2 hepatic cell compartments.

Our finding of an inverse association between genetic liability to insomnia and skin cancer risk contrasts with conventional expectations. Applying the same framework of horizontal pleiotropy, it is plausible that the selected genetic variants influence skin cancer risk through pathways independent of sleep disturbance, such as immune modulation or pigmentation traits. Although our sensitivity analyses (MR-Egger intercept) did not detect significant directional pleiotropy, the possibility of balanced pleiotropy cannot be entirely excluded and may contribute to this observed protective association.

Indeed, sleep disturbances are correlated with an increased likelihood of several types of cancer development, while exhibiting an inverse correlation with the incidence and mortality rates of other cancers.^[24] Recent research suggests a possible link between insomnia and a higher risk for endometrioid epithelial ovarian cancer, contrasting with the inverse relationship noted between high-grade serous and clear cell subtypes of epithelial ovarian cancer.^[3] In our study, we observed an increased risk of cholangiocarcinoma associated with insomnia, yet a contrary association was observed with hepatic and skin cancers. These findings, together with our results, highlight the site-specific nature of the relationship between insomnia and cancer.

Several studies have indicated that insomnia could be a contributing factor to the development of lung cancer.^[25,26] Nevertheless, insomnia does not appear to be a universal risk factor for all cancer types. A two-sample bidirectional MR analysis showed no causal link between insomnia and bladder cancer,^[27] consistent with our null findings for several other malignancies. Our study revealed no causal connection between insomnia and brain tumor, colorectal, esophageal, gastric, lung cancer, malignant lymphoma, pancreatic, or thyroid cancer.

Our study has several limitations. Firstly, two-sample MR strategies examine the effect of genetic predisposition, which may not fully capture the impact of adult-onset or environment-triggered insomnia. Secondly, the current analysis was limited to the European population, which reduced the potential for bias due to population structure but limits the generalizability of our findings to other ancestries. Thirdly, we were unable to examine effects specific to sex or age owing to the absence of stratified data. Future research should consider these aspects once segmented GWAS summary data become accessible. Finally, while our study examined 11 cancer types, we applied a stringent Bonferroni correction (P < .0045) for multiple testing, and none of the observed associations retained statistical significance. This suggests that our findings, particularly for hepatic bile duct and hepatic cancers with borderline P-values, should be considered suggestive and hypothesis-generating rather than definitive. Replication in larger, independent GWAS consortia with greater statistical power is warranted to validate these potential causal relationships.

In summary, our study provides suggestive evidence for a causal link between insomnia and the risk of hepatic bile duct cancer, hepatic cancer, and skin cancer, while finding no significant bidirectional relationships with several other malignancies. This investigation uncovers complex associations between insomnia and different types of cancer, underscoring the clinical relevance of addressing sleep and offering novel insights for both cancer prevention and therapeutic strategies. Longitudinal cohort studies incorporating detailed sleep assessments and biomarker analyses could help validate the temporal relationship and quantify the magnitude of risk. Among future research avenues, the use of new orexin receptor antagonists for the treatment of insomnia warrants particular interest.^[28] These medications, such as suvorexant and lemborexant, have demonstrated promise in improving sleep quality and may have potential applications in cancer patients with sleep complaints. Further research is needed to evaluate the safety and efficacy of orexin receptor antagonists in cancer patients and to explore their potential role in cancer prevention and treatment.

Acknowledgments

The authors acknowledge all those who participated in the collection and contribution of data to the GWAS data set, as well as those who made their GWAS summary statistics publicly available.

Author contributions

Conceptualization: Jiachu Li, Yali Xu.

Data curation: Jiachu Li, Yali Xu.

Formal analysis: Jiachu Li, Yali Xu.

Funding acquisition: Yali Xu.

Investigation: Jiachu Li, Yali Xu.

Methodology: Jiachu Li, Li Cheng, Fei Li, Yali Xu.

Resources: Jiachu Li, Fei Li, Yali Xu.

Software: Jiachu Li, Yali Xu.

Supervision: Jiachu Li, Li Cheng, Yali Xu.

Validation: Jiachu Li, Li Cheng, Yali Xu.

Visualization: Yali Xu.

Writing – original draft: Jiachu Li, Yali Xu.

Writing – review & editing: Jiachu Li, Li Cheng, Fei Li, Yali Xu.