A combination of MeSH terms and free words was used to search for relevant cohort studies in PubMed, Web of Science, Embase, and Cochrane Library databases up to February 15, 2025. The search terms included: “gestational diabetes mellitus”, “gestational diabetes”, “GDM” (related to GDM); “all - cause mortality”, “death” (related to all - cause mortality); “cardiovascular events”, “acute heart failure”, “myocardial infarction”, “ischemic stroke”, “coronary heart disease”, “stroke” (related to cardiovascular events). The specific search strategies for each database are as follows: PubMed: ((“gestational diabetes mellitus”[MeSH Terms] OR “gestational diabetes”[Title/Abstract] OR “GDM”[Title/Abstract]) AND (“all - cause mortality”[Title/Abstract] OR “death”[Title/Abstract] OR “cardiovascular events”[Title/Abstract] OR “acute heart failure”[Title/Abstract] OR “myocardial infarction”[Title/Abstract] OR “ischemic stroke”[Title/Abstract])).The complete search strategies for other databases are detailed in Supplementary Table S1. No language restrictions were imposed during the search to ensure the comprehensiveness of literature inclusion.

Study type: Prospective or retrospective cohort studies; Study participants: Adult women (age > 18 years) diagnosed with GDM, with postnatal follow - up duration ≥ 5 years; a control group of women without GDM was included; Outcome indicators: Included studies were required to provide adjusted hazard ratios (HR) with 95% confidence intervals (CI), or to furnish sufficient raw data to calculate effect sizes, and to report findings on the association between gestational diabetes and all-cause mortality (defined as death from any cause during follow-up), acute heart failure, myocardial infarction, and ischemic stroke.Quality criteria: NOS score ≥ 5 (moderate or high quality).

We considered studies eligible for inclusion if they were observational studies with retrospective or prospective cohort or case-control designs; reported at least one episode of cardiovascular disease or venous thromboembolism in women with a history of gestational diabetes; included a comparison group of women without gestational diabetes; and provided risk ratios with 95% confidence intervals. Studies were excluded if they lacked an eligible control group or relevant data on cardiovascular disease outcomes. We also excluded publications lacking original data, such as reviews, editorials, and commentaries. When studies included overlapping cohorts, we selected the study with the largest cohort or most detailed information for analysis. Potential non-English studies were translated using software assistance, with human translators employed when necessary. Studies were initially screened by title and abstract, followed by full-text review of potentially eligible articles.

Two independent researchers (Hong Zeng, Xiaoping Yin) conducted literature screening and data extraction according to preset criteria. In case of disputes, they were resolved through discussion or consultation with a third party (Danqing Zhao). The extracted content included: Basic study information: First author, publication year, country, study type; Characteristics of the study population: Total sample size (GDM group/control group), age, follow - up duration; Outcome indicators: HR and 95% CI for each outcome, and adjusted confounding factors (e.g., BMI, hypertension, history of diabetes); Quality score: The NOS was used to evaluate the quality of cohort studies. The quality was scored from three dimensions: selection of the study population (4 points), comparability between groups (2 points), and measurement of outcomes (3 points). The total score was 9 points, with ≤ 4 points indicating low quality, 5–6 points indicating moderate quality, and ≥ 7 points indicating high quality (16).

Statistical analysis was performed using Stata 15.0 software: Pooling of effect sizes: HR and 95% CI were used as effect sizes. If a study reported Relative Risk (RR) or Odds Ratio (OR) and the incidence of outcome events was low (< 10%), it was approximately regarded as HR (16). The model was selected based on the results of heterogeneity test: a random - effects model was used when I² > 50% and P < 0.1 (indicating high heterogeneity), otherwise a fixed - effects model was used. Heterogeneity analysis: The I² statistic was used to quantify heterogeneity. Meanwhile, subgroup analysis (by country, follow - up duration) and meta - regression (age, publication year, ethnicity) were conducted to explore the sources of heterogeneity.

Sensitivity analysis: A one - by - one exclusion method of individual studies was used to observe the changes in the total effect size and verify the stability of the results.

Assessment of publication bias: Funnel plots were drawn to visually judge publication bias, and Egger’s test was used to quantify the risk of bias (P < 0.05 indicated significant publication bias).

A total of 12,985 articles were initially retrieved. After removing duplicates (4,075 articles) using EndNote, articles that did not meet the criteria (7,880 articles, such as reviews and studies without long - term outcomes) were excluded by reading the titles and abstracts. The remaining 1030 articles were read in full, and further studies with incomplete data (1010 articles) and duplicate publications (11 articles) were excluded. Finally, 9 cohort studies were included (10–13, 17–21). The literature screening process is shown in Figure 1.

The 9 included studies contained a total of 4,191,840 women, including 189,246 cases in the GDM group (4.5%) and 3,992,594 cases in the control group (95.5%). The follow - up duration ranged from 5 to 22 years, with an average follow - up of 11.2 years. The studies were distributed in the following countries: 3 in the United States (10, 11, 18), 2 in China (20, 21), 1 in the United Kingdom (12), 1 in South Korea (13), 1 in Switzerland (17), and 1 in Australia (19). All studies adjusted for key confounding factors such as BMI and age. The NOS scores of all studies were ≥ 7 points (4 studies scored 9 points, 3 studies scored 8 points, and 2 studies scored 7 points), indicating that all were high - quality studies (Table 1).

Due to the high level of heterogeneity in this study, we explored the source of heterogeneity using meta-regression according to age, year of publication, and ethnicity.Meta-regression (Table 2) suggested that the P to for age, year of publication, and ethnicity were all greater than 0.05, suggesting that age, year of publication, and ethnicity were not sources of heterogeneity.

The long - term absolute event rates of the GDM group and the control group are shown in Table 3. During the follow - up period, the absolute incidences of all - cause mortality, acute heart failure, myocardial infarction, and ischemic stroke in women with GDM were 3.2% (95% CI 2.8% - 3.6%), 1.8% (95% CI 1.5% - 2.1%), 1.5% (95% CI 1.2% - 1.8%), and 1.2% (95% CI 0.9% - 1.5%), respectively, which were all higher than those in the control group (2.5%, 1.1%, 0.9%, and 0.7%, respectively).

The funnel plots of all outcomes were approximately symmetrical. The results of Egger’s test showed that there was no significant publication bias for all - cause mortality (P = 0.326), acute heart failure (P = 0.317), myocardial infarction (P = 0.217), and ischemic stroke (P = 0.594), indicating that the study results were less affected by publication bias (Supplementary Figures S4-S7).

This comprehensive meta-analysis, encompassing data from over 4.1 million women across nine high-quality cohort studies, provides robust evidence that a history of GDM is significantly associated with an increased long-term risk of all-cause mortality and major cardiovascular events, including acute heart failure, myocardial infarction, and ischemic stroke (10–14, 17–22). The consistency of these findings across diverse geographical populations and the robustness demonstrated in sensitivity analyses strengthen the credibility of our conclusions.

The key findings of our study indicate that women with a history of GDM face a 29% higher risk of all-cause mortality (11, 18, 19), a 74% increased risk of acute heart failure (12, 13, 17), a 63% elevated risk of myocardial infarction (10–12), and a 70% greater risk of ischemic stroke (10, 13, 18) compared to their non-GDM counterparts. Notably, the association with acute heart failure appeared to be independent of subsequent Type 2 Diabetes (T2DM), as suggested by subgroup analysis (12, 13). This underscores GDM not merely as a transient condition of pregnancy but as a potent indicator of underlying, long-term cardiovascular vulnerability (2, 4). While the absolute event rates remain low in young and middle-aged populations, the substantially elevated relative risks highlight a significant public health concern (2, 23). Given the large population of women affected by GDM, these relative risk increases translate into a substantial number of excess deaths and cardiovascular events at a population level.

Our findings are largely consistent with and extend previous literature (2, 4, 15). The observed HR of 1.29 for all-cause mortality aligns with earlier reports (10, 11). The robust associations found for specific cardiovascular outcomes, particularly the strong link with heart failure, provide a more granular understanding of the cardiovascular sequelae of GDM (12, 22). The sensitivity analysis for all-cause mortality, which showed a strengthened and more precise association (HR = 1.45) after removing a major source of heterogeneity [the study by Bucci (17)], increases our confidence in the true existence of this risk.

The clinical and public health implications of our study are substantial. They reinforce the concept that a history of GDM should be recognized as a critical marker for elevated long-term cardiovascular risk (2, 4, 24). Current postpartum care for GDM women often focuses on short-term glycemic control and screening for T2DM (2, 9). Our results advocate for a paradigm shift towards long-term, holistic cardiovascular risk assessment and management in this population (25, 26). This should include regular monitoring of blood pressure, lipid profiles, and other cardiovascular risk factors, coupled with targeted lifestyle interventions and, where appropriate, pharmacologic strategies to mitigate risk (9, 25). Future research should focus on elucidating the precise biological mechanisms linking GDM to cardiovascular disease (21, 27, 28) and on developing and testing effective long-term intervention strategies for this high-risk group.

We acknowledge several limitations in this meta-analysis. First, significant heterogeneity was observed for several outcomes (all-cause mortality, acute heart failure, ischemic stroke). Although we employed random-effects models and conducted sensitivity and meta-regression analyses, residual heterogeneity likely persists due to unmeasured or unreported confounders, such as lifestyle, socioeconomic status, and details of postnatal management (4, 27, 29). Second, the number of studies available for meta-analysis of myocardial infarction was relatively small (n=4), despite highly consistent results (10–12, 21). Third, the varying follow-up durations (5–22 years) mean that studies with shorter follow-up may not fully capture the long-term trajectory of risk. Fourth, the generalizability of our findings may be limited, as the included studies primarily involved populations from Europe, North America, and Asia, with underrepresentation from Africa and South America.

Based on these limitations, we propose the following directions for future research: