The present meta-analysis was performed in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses [PRISMA; 58] guidelines for systematic reviews and meta-analysis. Our protocol has been registered on the PROSPERO platform under registration number CRD42023446724.

We searched the PubMed, Cochrane Library, and Embase databases up to July 15, 2023. We employed Medical Subject Headings (MeSH) and relevant keywords. We drew upon the previous methodologies of meta-analysis. (57, 59–64). We included case–control and cohort studies without any language restrictions. The search terms included “Breast Neoplasms,” “Malignant Neoplasm of Breast,” “Light exposure,” and “Risk.” The complete search strategy for each database can be found in Supplementary Tables 1–3. Additionally, we scrutinized the reference lists of the included cohort studies and other published meta-analyses to identify relevant observational studies.

The study selection process was carried out by two reviewers (LZN and LZL) in accordance with the predetermined inclusion and exclusion criteria. Both reviewers independently screened the literature and removed duplicate and irrelevant articles based on titles and abstracts. Subsequently, the full texts of potentially eligible articles were obtained and thoroughly reviewed to ensure the inclusion of all relevant studies. Any disagreements between the two reviewers were resolved through discussion until a consensus was reached.

The inclusion criteria were as follows: (a) inclusion of an independent exposure variable pertaining to LAN exposure; (b) consideration of an outcome variable related to breast cancer; (c) incorporation of case–control or cohort study designs. The exclusion criteria were as follows: (a) reviews, conference proceedings, or commentaries; (b) duplicate literature (retaining the variant with the most comprehensive information).

The data extraction process was conducted independently by the first reviewer (LZN), and upon its completion, cross-verification and consultation was performed with the second reviewer (LZL). A predesigned data extraction form was utilized, and the following data were extracted: study type (case–control or cohort), first author, publication year, source of cases, source of controls, number of cases, number of controls, exposure definition, breast cancer definition, classification of exposure, information collection period and scope, baseline age, and values of outcome variables (odds ratio or hazard ratio) determined based on various confounding factors. Any potential discrepancies were addressed through discussions with LZL to achieve a consensus.

The quality of our studies was evaluated using the Newcastle-Ottawa Scale (NOS) (65), which employs a star system ranging from 0 to 9 to evaluate participant selection and exposure measurement, with 2 stars allocated for result comparison and 3 stars designated for outcome assessment and follow-up adequacy. A higher score reflects superior study quality. Scores ranging from 0 to 3 indicate low quality, scores ranging from 4 to 6 indicate moderate quality, and scores ranging from 7 to 9 indicate high quality.

The adjusted relative ratios (RRs) and 95% confidence intervals (CIs) for each observational study were used to assess the association between breast cancer incidence and LAN exposure. Heterogeneity was evaluated using I² values (66) and Cochran’s Q heterogeneity test (67). Furthermore, we utilized Z-tests to assess whether the effect size (RRs) is equal among different subgroups (67). Based on the precedent of previous meta-analyses (33–35), we used a random effects model in each study to address confounding factors and minimize study errors. Sensitivity analysis was conducted by predefining the exclusion of studies that did not pass the sensitivity analysis to ensure the robustness of the overall effect. A funnel plot was examined for symmetry to detect publication bias, and Egger’s regression test was used for statistical assessment (68). The primary data analysis was conducted using Stata statistical software version 17.0 (Stata Corp, College Station, Texas), while the Z-tests were performed using R language (version 4.2.2).

A total of 2,964 observational studies were initially identified after excluding duplicates. After conducting title and abstract screening, 2,871 articles were subsequently excluded. An in-depth review was conducted on 93 articles, and after the further exclusion of 72 articles, 21 studies were eligible for the meta-analysis (Figure 1). The primary reasons for exclusion were irrelevant research topics, inappropriate study types, missing data, inappropriate study subjects, availability of only abstracts, studies sharing the same data, and unclear definitions within the LAN. Among the 21 included studies, 8 were cohort studies (36, 39, 41, 45–47, 50, 52) and 13 were case–control studies (32–35, 37, 38, 40, 42–44, 48, 49, 51).

The 21 studies included in this study were published between 2001 and 2023 and involved a total of 734,372 participants. Of these studies, 11 originated from North America (32, 39–41, 43, 46–50, 52), 3 from Europe (36, 37, 45), 5 from Asia (33–35, 38, 42), and an additional 2 from Australia (44, 51). The average follow-up time for these studies was 12.8 years.

Among the various confounding factors examined in these studies, the proportion of studies focused on estrogen receptor status was 42.9% (n = 9) (36, 37, 41, 42, 45–47, 50, 52), while studies on menopausal status accounted for 47.6% (n = 10) (36, 39–42, 45–48, 51). Additionally, studies investigating indoor LAN accounted for 66.7% (n = 14) (33–44, 50, 51), and studies on outdoor LAN accounted for 66.7% (n = 14) (32–35, 37, 39–41, 45–48, 50, 52).

Regarding lifestyle habits, the proportions were as follows: 19.0% (n = 4) of studies examined the habit of watching TV while sleeping (35, 40, 41, 50), 23.8% (n = 5) looked at the use of bedroom shutters (33–35, 38, 40), and 14.3% (n = 3) explored the habit of waking up and turning on lights (36, 43, 49).

Each study provided clear definitions for both LAN exposure and breast cancer. Adjusted estimates were available for analysis, despite the presence of varying confounding factors across the studies. The characteristics of the cohort (Table 1) and case–control studies (Table 2) are presented.

In the Newcastle-Ottawa Scale (NOS) assessments conducted on the various studies, the average score for the included research was 7.43 (ranging from 5 to 9). A score of 7 is considered as the cut-off point for high-quality research. Studies with a score of 7 or above are regarded as of higher quality. The scores of the included studies are illustrated in Figure 2.

A comprehensive analysis of 21 studies (32–52) revealed a noteworthy correlation between exposure to LAN and a higher susceptibility to breast cancer (RR = 1.12; 95% CI: 1.06–1.17; I² = 31.3%, p < 0.001; Figure 3). Heterogeneity analysis indicated a slight level of heterogeneity in our study, whereas sensitivity analysis did not uncover any individual study findings that overturned the overall results, thus providing evidence for the reliability of our analytical outcomes (Supplementary Figure A).

We conducted subgroup analyses on the 21 included studies to explore the impact of various factors on the incidence of breast cancer (32–52). Based on existing evidence, we categorized these factors into confounding factors (factors that are associated with LAN exposure and breast cancer incidence but are not mediators of the relationship) and effect-modifying factors (those that modify the relationship under different LAN conditions, breast cancer subtypes, or study types). Specifically, confounding factors included geography, menopausal status, BMI, and ethnicity. Effect-modifying factors encompassed study type, LAN (light at night) type, estrogen and progesterone receptor status, and lifestyle. When analyzed as independent confounding factors, no heterogeneity (I² = 0%) was observed for continents, white race, and postmenopausal individuals, which was considered the main factor contributing to the slight heterogeneity observed in the overall analysis.

Regarding confounding factors, in our subgroup analyses that focused on different continents, we observed a clear disparity in breast cancer incidence rates associated with exposure to the LAN. Specifically, the breast cancer incidence rate in Asia (RR = 1.24; 95% CI: 1.15–1.34; I² = 0.0%, p < 0.001) was significantly higher than that in North America (RR = 1.11; 95% CI: 1.06–1.15; I² = 0.0%, p < 0.001). However, the breast cancer incidence rate in Europe (RR = 0.99; 95% CI: 0.89–1.11; I² = 0.0%, p = 0.898), despite being analyzed, did not exhibit any statistically significant difference. When considering racial differences, the incidence of breast cancer was higher in the white population (RR = 1.12; 95% CI: 1.05–1.20; I² = 0.0%, p < 0.001), with no statistically significant difference observed in the Black population (RR = 1.15; 95% CI: 0.98–1.35; I² = 4.8%, p = 0.080). Additionally, subgroup analysis for menopausal status did not yield statistically significant differences between the premenopausal (RR = 1.09; 95% CI: 0.99–1.19; I² = 17.6%, p = 0.083) and postmenopausal groups (RR = 1.03; 95% CI: 0.97–1.09; I² = 0.0%, p = 0.352).

In terms of effect-modifying factors, similar to the meta-analysis conducted in 2021 (57), we conducted separate analyses to assess the effects of LAN exposure on breast cancer risk based on estrogen receptor status, types of study, and types of LAN exposure. The analysis of different types of LAN exposure showed that outdoor LAN exposure (RR = 1.07; 95% CI: 1.02–1.13; I² = 30.9%, p = 0.004) had a slightly higher promoting effect on breast cancer incidence than indoor LAN exposure (RR = 1.02; 95% CI: 0.92–1.13; I² = 65.6%, p = 0.675). In the analysis of estrogen receptor status, LAN exposure exhibited a slightly stronger impact on breast cancer risk for cases with estrogen receptor positivity (RR = 1.10; 95% CI: 1.03–1.18; I² = 17.0%, p = 0.005) compared to cases with estrogen receptor negativity (RR = 1.07; 95% CI: 0.94–1.21; I² = 0.0%, p = 0.316). Subgroup analyses for different study types revealed that LAN exposure had a higher promoting effect on individual breast cancer incidence in case–control studies (RR = 1.16; 95% CI: 1.06–1.27; I² = 40.4%, p = 0.001) compared to cohort studies (RR = 1.08; 95% CI: 1.04–1.14; I² = 0.0%, p < 0.001).We also conducted analyses on various lifestyle factors summarized in the literature. When considering TV on while sleeping as an effect-modifying factor, the impact of LAN exposure on breast cancer did not show statistical significance (RR = 1.06; 95% CI: 0.97–1.16; I² = 0.0%, p = 0.168). This observation holds true for “Turn on the light when you wake up” (RR = 1.09; 95% CI: 0.87–1.37; I² = 47.5%, p = 0.449) and “Bedroom shutters (open)” (RR = 1.07; 95% CI: 0.84–1.37; I² = 64.4%, p = 0.589). The complete dataset can be found in Table 3. The results of the Z-tests and the forest plot for the subgroups can be viewed in Supplementary Figure C.

Visual inspection of the funnel plot did not reveal any significant publication bias in the occurrence of breast cancer. This observation holds true for the overall analysis funnel plot (Figure 4), which is consistent with the results of our Egger’s regression test (p = 0.746). Similar findings of bias were observed across various subgroups (Supplementary Figure B).

Our meta-analysis included a total of 21 observational studies. The collective data analysis unveiled a significant correlation between exposure to LAN and an escalated susceptibility to breast cancer. Regarding the confounding factors, Jack Sariego first proposed the relationship between geographic correlation and breast cancer (69). This relationship is typically associated with fertility rates and taxation (70). High-quality lifestyles or habits in developed regions are effective in curtailing the incidence of breast cancer (71). We observed that LAN exposure in Asian countries showed the highest increase in breast cancer risk, followed by North America, while in Europe, the breast cancer risk outcome did not reach statistical significance. At the level of menopausal status, many studies have observed a protective effect in premenopausal women in terms of obesity (72), night shift work (73), and physical activity (74). We found that breast cancer risk outcomes did not reach statistical significance, whether in premenopausal or postmenopausal states. We considered additional factors that are associated with LAN exposure and breast cancer incidence but are not mediators of the relationship, such as BMI and race, which could potentially introduce significant bias to the risk outcomes. Multicenter studies have indicated a clear association between higher BMI and breast cancer in women (75), while in terms of race, the annual statistical graphs from the SEER Cancer Statistics Review show consistently higher breast cancer incidence rates in the white population than in the Black population (76). Subgroup analysis for LAN exposure revealed that white individuals have a higher breast cancer risk. Existing evidence supports the notion that cultural dynamics, as well as differences in sociodemographic and behavioral characteristics among different population subgroups, modulate the expression of biological diseases, possibly contributing to the higher breast cancer risk in the white population (14). In terms of modifying factors, previous observational studies have consistently shown a significant increase in breast cancer risk associated with outdoor LAN exposure, particularly among estrogen receptor-positive cases. Our findings align with two prior meta-analyses (57, 60). In a previous meta-analysis, Ka Yan Lai et al. highlighted the impact of indoor LAN exposure in their discussion (60). Lifestyle factors such as window orientation, the use of blinds, and television watching may all contribute to an elevated risk of breast cancer (77). Due to the limited number of studies available, they did not conduct a subgroup analysis in this regard. We have addressed this gap in the literature. Although the five studies included in our analysis did not demonstrate a statistically significant difference, the influence of confounding factors within the lifestyle domain is substantial. Therefore, future, more rigorous research focusing on LAN exposure should be conducted.

To date, a total of 8 related meta-analyses have been identified (57, 59–62, 64, 78). Regarding the types of included studies, the predominant ones include: four studies assessing the association between breast cancer risk and indoor and outdoor LAN (57.1%) (59–61, 78), four studies investigating the relationship between breast cancer risk and menopausal status (57.1%) (60, 61, 64, 78), and two studies examining the association between breast cancer risk and estrogen receptor status (28.6%) (60, 78).

Ka Yan Lai et al. (60) published a meta-analysis on LAN exposure in 2021. Their results showed a significant association between ER+ and outdoor LAN exposure and cancer risk. They also observed some correlation in premenopausal women. However, they did not consider the potential influence of the patients’ life habits on the experimental results. Furthermore, although their study included participants from seven countries, they did not explore differences in the effects of LAN exposure on breast cancer between countries, although there is solid evidence that cancer risks vary due to economic differences between countries (79). In our analysis, we supplemented subgroups with continental differences and lifestyle habits. We did not observe significant differences in LAN exposure and breast cancer risk by different menopausal states. Therefore, our study provides more subgroup analysis and new evidence in this field, which has certain reference value.

Another meta-analysis, conducted by Teresa Urbano et al. (57), encompassed 17 studies. Subgroup analysis was performed based on study type, menopausal status, and estrogen receptor (ER) status. Their findings revealed a noteworthy elevation in the risk of breast cancer attributed to exposure to LAN in cohort studies, premenopausal women, and ER-positive women. In contrast, our study employed a larger sample size (n = 21). Notably, our findings demonstrated a more substantial risk association among the case–control study population, contradicting the results reported by Urbano et al. (57).

The existing three hypotheses attempt to explain the connection between LAN exposure and the risk of various cancers: the direct inhibition of melatonin secretion (80), sleep deprivation affecting cell proliferation and cytokine production (81), and the effect of chrono disruption (82, 83). The function of melatonin was described in detail in 2018 (84). Melatonin is considered a pleiotropic and multitasking molecule (85). In addition to regulating circadian rhythms, it plays a significant role in anticancer effects. It inhibits tumor cell proliferation and invasion, suppresses DNA damage, and modulates the immune system to aid in the elimination of tumor cells. All these factors support its inhibitory effect on breast cancer (55, 81). Additionally, some observational studies have confirmed the suppression of melatonin secretion due to indoor LAN exposure (86). In addition, we should also acknowledge the potential impacts of sleep deprivation and circadian disruption. The sleep–wake cycle is a natural component of human life (87), including waking up during the day and sleeping overnight. The sleep pattern is controlled by static sleep pressure and the circadian rhythm (88), so as the day progresses, the combined effect of these two factors makes it easier for us to fall asleep, while signals from the circadian rhythm can prevent us from falling asleep. Experimental evidence has shown that sleep deprivation can significantly impair learning, memory, judgment, and concentration (89). In terms of health-related risk factors, according to the Nurse’s Health Study, shift work nurses, especially those who work at least three times a month for 15 years or more, have a significantly increased risk of colorectal cancer (90). The risk of breast cancer in nurses who have worked in shift rotations for over 20 years is 1.79 times higher (91). The detrimental effects of LAN exposure on sleep have been confirmed in the experiments conducted by Yu-Xiang Xu et al. (92).

Nonetheless, our meta-analysis has limitations. First, as is typical with observational studies, uncontrolled confounding factors can affect our evaluations. When analyzing lifestyle factors, we did not consider potential influences from factors such as “turn on the light when you wake up” due to insufficient descriptions or sample sizes in observational studies. To control for confounding factors within acceptable limits, we did not assess the potential impact of LAN exposure during night shifts or rotating shift work on breast cancer. The impacts of insufficient sleep duration and sleep quality were not explored, despite prior research showing their significance (93).Other established risk factors for breast cancer, such as noise, air quality, smoking, and alcohol intake, were not adequately examined due to small sample sizes. For instance, observational studies on alcohol accounted for only 4.8% (n = 1) (47), and subgroup studies on smoking were similarly limited at 9.5% (n = 2) (46, 47). While the influence of BMI on breast cancer has been established (94), we could not analyze its association with LAN exposure due to the limited data available. The use of personal electronic devices, lightning, and magnetic fields represent potential sources of exposure that should be considered. Second, due to inconsistent indicators of outdoor LAN exposure data of various materials, DMSP (47) or DNP (48) was used for LAN dose detection in existing experiments, and no updated outdoor LAN exposure dose data were available after 2021. Therefore, we did not conduct a dose analysis on the correlation between LAN exposure and breast cancer. Additionally, it is inevitable that we included retrospective cohort studies and case–control studies, where patients’ subjective descriptions may introduce potential biases and ultimately lead to recall bias. Therefore, the interpretation of the results should be approached with caution. The last factor is the subjective design error. In many indoor LAN exposure studies based on self-reporting (35–39, 43, 50, 51), statistical analysis is conducted to compare the population with the highest LAN exposure or presence of LAN exposure to those without LAN exposure or with the lowest dose. However, due to different questionnaire designs and varying definitions of high and low doses (for example, Davis et al. (43) categorized indoor LAN exposure into six levels, while Johns et al. (36) used only three levels), the subjectivity present in each study cannot be avoided in our selection. Undeniably, our study still has many strengths. (1) Following the suggestion of Ka Yan Lai et al. (60), we have, for the first time, considered continents and lifestyle habits in the meta-analysis of LAN exposure and breast cancer and obtained favorable results. (2) Our study supports the claim that there is no significant difference in breast cancer risk between pre- and postmenopausal women exposed to the LAN, as we did not observe any statistical significance in our meta-analysis, providing new evidence for this factor. (3) Our study found that individuals with high exposure to the LAN, especially among Asian populations, are more susceptible to breast cancer. Therefore, it may be of great importance for relevant public health agencies to implement necessary protective measures and preventive strategies for specific groups. Therefore, future research should incorporate relevant influencing factors, particularly those related to unhealthy lifestyle choices. Additionally, designing standardized indoor LAN exposure assessments will aid in evaluating factors associated with breast cancer risk.