The study protocol was registered with Open Science Framework (osf.io/7zx3d). We report our review using the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidance (PRISMA; see online Appendix 1). The study did not deviate from the pre-registered protocol.

We searched four major databases, including PubMed, Embase, PsycInfo, and Cochrane Central Register of Controlled Trials (CENTRAL), with no limits applied (date, language), from database inception to 22 July 2024. We also searched the gray literature and reviewed reference lists of the included studies and related systematic reviews. The literature review was developed based on a preliminary review of the existing literature and consultation with topic and information specialists.

All articles with a potentially relevant title, abstract, or ones for which the relevance was unclear, were evaluated independently by two authors to determine inclusion eligibility. Discrepancies were resolved by deliberation between the two reviewers or with input from a third author. We excluded conference abstracts, case reports, case series, meta-analyses, and studies with duplicate data. Appendix 2 shows the complete search strategies, and Appendix 3 presents the reasons for exclusion. We defined the eligibility criteria for study inclusion based on the PICOS framework: Population (P): We included studies that reported on suicide deaths or self-harm/suicidal behaviors (SHSB) in the general population. No restrictions were applied regarding age, gender, or geographic location. Intervention/Exposure (I): The primary exposure of interest was the occurrence of specific calendar holidays, namely Christmas Eve, Christmas Day, New Year's Day, and Valentine's Day. Comparator (C): Comparator days were defined as non-holiday (regular) days, serving as a baseline reference for evaluating changes in suicide and SHSB rates. Outcomes (O): The primary outcomes were: (1) the proportion of suicide deaths on each holiday (defined as the number of suicides on the holiday divided by the total number of suicides in that year); (2) the risk of suicide (risk ratio or risk difference compared to regular days); and (3) the proportion and risk of SHSB on the specified holidays. SHSB was broadly defined to include suicide attempts, deliberate self-harm, and parasuicide, acknowledging diagnostic and definitional overlap in the literature. Study Design (S): We included observational studies (e.g., time-series, retrospective cohort, cross-sectional designs) that provided quantitative data on suicide or SHSB during at least one of the specified holidays and on non-holiday comparison days.

Two authors independently and in duplicate extracted data from selected articles. WebPlot Digitizer (https://automeris.io/WebPlotDigitizer/) was used to extract numerical data from the figures. We extracted the total number of suicides over the entire study duration, the average annual number of suicides, the average daily number of suicides, and the number of suicides on each of the four holidays. We also extracted the proportion of annual suicides that occur on each of the four holidays. The quality of the included studies was assessed using an outcome specific modified Newcastle-Ottawa scale. The Newcastle-Ottawa scale was selected for its ability to distinguish variations in quality based on number of confounders and covariates and its reliability compared with the ROBINS-I (Risk Of Bias In Non-randomized Studies—of Interventions) tool, The detailed data extraction is showed in Appendix 4.

All analyses were conducted in the statistical software R, version 4.2.0 (R core team, R Foundation for Statical Computing, Vienna, Austria) using the metafor, ggpolt2, dplyr, and robumeta R packages.

The proportion of annual suicides occurring on each of the four holidays was calculated as the number of suicides on that holiday divided by the total number of suicides in that year. We also calculated the risk ratio (RR) and risk difference (RD) for each of the four holidays compared to regular days (i.e., the average daily number of suicides over 365 days). Additionally, we calculated the proportion of annual SHSBs occurring on each of the four holidays, as well as the RR and RD of the four holidays compared to regular days.

For the proportion-based meta-analyses, the Freeman-Tukey double arcsine transformation was applied to stabilize variance. The pooled proportion with 95% confidence interval was calculated using the restricted maximum likelihood random-effects model with the inverse-variance weighting method. Cumulative proportion-based meta-analyses ordered by study year were also performed for RD and RR. We assessed inter-study heterogeneity with the Cochran Q test and quantified it by using the I² statistic. For the Q statistic, we considered a P-value < 0.1 to be a statistically significant indicator of heterogeneity. I² values considered representative of low, moderate, and high heterogeneity were < 25%, 26–50%, and >50%, respectively. We assessed publication bias by using funnel plot analysis, with evaluation of asymmetry by visual inspection followed by Egger's test. Tests were 2-tailed, and P-value < 0.05 was set as statistically significant.

Where at least 10 studies were included in a meta-analysis, meta-regression explored heterogeneity by the following characteristics identified a priori: study design (national vs. local), continent, country, study duration, and study year (not publication year). Regarding holidays that did not conform to the expected proportions, subgroup analysis by country was conducted and presented the results using a world map.

Publication bias or small study effects were examined using funnel plots and the Egger's test when 10 or more studies were included in a meta-analysis. The sensitivity analysis is to perform meta-analysis of non-affirmative studies and generalized linear mixed model with logit transformation for the primary outcome. The meta-analysis of non-affirmative studies assesses the robustness of our estimates and determines whether the pooled estimate would remain in the same direction even under the worst case publication bias.

After searching the databases and excluding duplicate records, we identified 586 unique, potentially eligible articles; 465 were excluded after title and abstract screening, and 93 were excluded after assessment of the full text (Supplementary material 3). Ultimately, 28 studies (n = 2,186,094) were included in our analysis, 18 for suicide deaths and 10 for SHSB (Supplementary Table 1). A PRSIMA flowchart depicting our search strategy is presented in Supplementary Figure 1. The follow-up duration of these studies were from 1 (27) to 35 (21) years. The earliest of study year is 1969 (21) and the most recent is 2019 (28), spanning a range of 50 years. There were 14 studies from Europe, 10 from the North America, two from Asia, one from South America, and one from Pacific.

The quality assessment of the included studies were assessed with the Newcastle-Ottawa score (Supplementary Table 2) (29). Scores of 1, 2, and 3 indicate a high risk of bias, scores of 4, 5, and 6 indicate a moderate risk of bias, and scores of 7, 8, and 9 indicate a low risk of bias. There were 6 of 28 studies scoring 6 points (19, 23, 30–33), 8 studies scoring 7 points (17, 24, 27, 34–38), 12 studies scoring 8 points (18, 20–22, 28, 39–45), and 2 studies scoring 9 points (46, 47).

The proportion of annual suicides (Figure 1) was 0.23% [0.17%, 0.28%; number of studies (k) = 11; I² = 50.8%] on Christmas Eve, 0.24% (0.19%, 0.29%; k = 17; I² = 65.8%) on Christmas Day, and 0.39% (0.31%, 0.48%; k = 16; I² = 80.7%) on New Year's Days, while it is 0.26% (Supplementary Figure 2, 0.24%, 0.29%; k = 17; I² = 0%) on regular days (i.e., the average daily number of suicides over 365 days) (Figure 2). Compared with regular days, the risk was 17% lower on Christmas Day (RR = 0.83; 0.72, 0.96), and it was 33% higher (RR = 1.33; 1.08, 1.65) on New Year's Day. Additionally, the RD (Supplementary Figure 3) was significantly lower on Christmas Eve (−0.06%; −0.10%, −0.01%; k = 10; I² = 0%), and it was significantly higher on New Year's Day (0.10%; 0.02%, 0.17%; k = 14; I² = 60.7%) when compared with regular days. The proportion difference between these three holidays and regular days showed similar findings on meta-regression analysis (Supplementary Table 3).

The temporal trend of the proportion of annual suicides occurring from Christmas Eve to New Year Day are showed in Figure 3. Generally, most countries show a decreased or similar proportion of suicide deaths from Christmas Eve to Christmas Day, except for Mexico, and the largest decrease in the suicide proportion is seen in the Australia and Netherlands. However, most countries show an increased proportion of suicide after Christmas Day, leading up to New Year's Day.

The cumulative meta-analyses are in the Supplementary Data (Supplementary Figures 4–6). For Christmas Eve, the initial studies from 1969 to 1970 showed a lower proportion of suicide; however, after that, there was a sudden increase, but it gradually declined in the end. This temporal trend was similar on the Christmas Day. Importantly, the proportion of suicide on New Year's Day (Supplementary Figure 6) was low between 1969 and 1970, and after that, it is gradually increased, with more recent studies contributing to progressively higher estimates.

The meta-regression analyses (Supplementary Table 4) showed that the proportion of annual suicides occurring on Christmas Eve was lower on studies using national dataset (p < 0.001) and varied with different countries (p = 0.039). The proportion of annual suicides occurring on New Year's Day also varies with different countries (p < 0.001) and was lower on studies with longer duration (p = 0.009). The RD and RR of suicide on Christmas Eve and on New Year's Day compared with regular days showed similar findings with those of proportion of suicide (Supplementary Tables 5, 6). Besides, the RR of suicide on Christmas Day compared with regular days varies with different countries (Supplementary Table 5; p = 0.002).

The proportion of annual suicides occurring on Valentine's Day (Supplementary Figure 7) was 0.27% (0.24%, 0.30%; k = 5; I² = 0%), and the RR (Supplementary Figure 8) and RD (Supplementary Figure 9) were not significant compared with that on regular day. The cumulative meta-analysis did not show any temporal relationship (Supplementary Figure 10).

Stratified by country (Figure 4), most countries showed lower proportion of annual suicides occurring on Christmas Eve, except for Australia (0.43%; 0.31%, 0.57%) and Hungary (0.30%; 0.15%, 0.49%). Additionally, most countries showed lower proportion of annual suicides occurring on Christmas Day (Supplementary Figure 11), except for Colombia (0.66%; 0.37%, 1.04%) and Mexico (0.44%; 0.26%, 0.65%). For New Year's Day (Figure 5), the top three countries with the highest proportion of annual suicides are Columbia (0.93%; 0.57%, 1.37%), Sweden (0.54%; 0.40%, 0.70%), and Mexico (0.50%; 0.32%, 0.73%).

The proportion of annual SHSB occurring on Christmas Eve, Christmas Day, New Year's Day, and Valentine's Day are showed in the supplement (Supplementary Figures 12, 13). Compared with regular days, the risk of SHSB was 26% lower on Christmas Eve (RR = 0.74; 0.57, 0.96; Supplementary Figure 14), and it was 17% higher on New Year's Day (RR = 1.17; 1.03, 1.34). However, the RR of SHSB was not significant on Valentine's day compared with regular days (Supplementary Figure 15). The RD of SHSB (Supplementary Figures 16, 17) showed similar findings as those of suicide deaths.

Visual inspection of the funnel plots for the primary outcomes did not show significant publication bias (Supplementary Figures 18–29). However, the Egger test was significant for proportion of annual suicides occurring on New Year's Day (Supplementary Figure 20; p = 0.011) and RD of suicide (Supplementary Figure 24; p = 0.004) on New Year's Day compared with regular days. We also conducted trim-and-fill analysis for potential publication bias (Supplementary Figures 30–33). For Christmas' eve (Supplementary Figure 30) and Valentine's day (Supplementary Figure 33), the trim-and-fill analysis identified two potentially missing studies for each analysis, but the adjusted overall effect size remained similar to the original estimate (PFT = 0.05, 95% CI: 0.04–0.05) and (PFT = 0.05, 95% CI: 0.05–0.05). The funnel plot showed slight asymmetry, suggesting minimal publication bias without materially altering the pooled results. For Christmas day (Supplementary Figure 31) and New Year's Day (Supplementary Figure 32), the pooled analysis yielded an overall proportion of 0.05 (95% CI: 0.05–0.06) and 0.06 (95% CI: 0.06–0.07) across all studies. The funnel plots revealed a largely symmetrical distribution, indicating minimal risk of publication bias. No additional studies were imputed by the trim-and-fill procedure, suggesting the robustness of the pooled estimate.

When including non-affirmative studies (Supplementary Table 7), the direction of pooled estimates of RR and RD of suicide deaths showed consistent direction with those of original analyses on the four holidays. The generalized linear mixed model with logit transformation (Supplementary Table 8) showed similar findings on the proportion of suicide on the four holidays.

Supplementary Tables 9–12 demonstrated leave-one-out analyses for proportion of annual suicides occurring on Christmas Eve, Christmas Day, New Year's Day, and Valentine's day. For Christmas Eve, after removing Baker et al. (22), the I² would decline from 50.8% to 0%, indicating this study was the source of heterogeneity. For Christmas Day, after removing Jones et al. (36), Fernández-Niño et al. (34), and Arango et al. (40), the I² would decline from 65.8% to 47.0%, 46.7%, and 32.7%, suggesting these three studies contributed slightly more to heterogeneity. For New Year's Day, leave-one-out sensitivity analyses showed that removing any single study produced minimal changes in the pooled estimates (0.0022–0.0025), with all results remaining statistically significant (p < 0.001). Moderate heterogeneity (I² = 60–70%) persisted across all scenarios, and no single study substantially reduced it. For Valentine's day, removing any single study kept the I² at 0, indicating high homogeneity.

Our study has several strengths. We calculated the annual mean proportion of the number of suicide on the major holiday as primary outcomes. In this way, we can avoid the fluctuations in suicide rates throughout the year and overcome the different suicide tendency in different country/study by comparing only with the annual average suicide proportion. Besides, we also calculated RD and RR to represent both absolute and relative risk difference between major holidays and regular day. Previous reviews (15, 48) investigating suicide risks during Christmas/New Year have only conducted systematic reviews due to the heterogeneity in study designs or humanities. We conducted the first meta-analysis and utilized sensitivity tests (including subgroup analysis, meta-regression, and meta-analysis of non-affirmative studies) to further validate our findings.

Our study has several limitations. First, most of the included studies originate from developed countries and predominantly Christian populations, which limits the ability to generalize the findings to developing and least developed countries. In different cultures and regions, the importance of Christmas and New Year to people may vary. In countries where Christianity is not the predominant religion, the suicide risk during Christmas may be similar to that on regular days; however, certain holidays or cultural events in other communities may also be associated with decreased suicidal tendencies, warranting further investigation. Second, we overcame some difficulties by using the annual mean proportion of the number of suicides to calculate the suicide proportion and risk. However, the changes in suicide rates over the years might not be considered, particularly a very long follow-up period. We employed meta-regression to examine whether the study year and the study duration are important moderators, showing no or trivial effect. Third, the included studies span from the 1970s to the 2010s. Over these 40–50 years, environmental factors and suicide mortality rates have also changed in different countries. Therefore, in addition to meta-regression, we also conducted cumulative meta-analyses ordered by study year. We found that the risk of suicide on New Year's Day gradually increased, with more recent studies contributing to progressively higher estimates. Fourth, in the epidemiology of suicide, factors such as age, gender, socioeconomic status, substance use, prior suicidal behavior, and alcohol use are important. However, few included studies provided data on their association with suicide on major holidays, and most also lacked detailed information on suicide methods, limiting further analysis of these risk factors. Fifth, when calculating SHSB, it is not possible to distinguish between suicide attempts accompanied by some deaths or self-harm/parasuicide without death. Therefore, there is a certain degree of heterogeneity in the results of SHSB. Sixth, we found small study effects on the risk of suicide on New Year's Day. However, the meta-analyses of non-affirmative studies suggest the robustness of our study findings. Finally, in South Korea (20), Taiwan, and China, Lunar New Year might be considered a more important new year compared to Solar New Year. However, the data on Lunar New Year's Day were not available.