Weight change mediates the effect of lifestyle intervention on postpartum diabetes in women with prior gestational diabetes: a secondary analysis of a randomized clinical trial

Abstract

Background

Women with a history of gestational diabetes mellitus (GDM) have a substantial 20–50% risk of developing diabetes postpartum. Understanding the precise mechanisms by which lifestyle interventions prevent diabetes is crucial for optimizing clinical care. This study aimed to quantify the extent to which weight change mediates the observed 46% reduction in postpartum diabetes risk achieved by an intensive lifestyle intervention in women with prior GDM.

Methods

This secondary mediation analysis utilized data from the Tianjin GDM Prevention Program (NCT01554358), a large-scale, 4.5-year randomized controlled trial conducted in China. A total of 1180 women with a recent history of GDM diagnosed according to the 1999 WHO criteria (mean age 32.4 ± 3.50 years; mean BMI 23.9 ± 3.83 kg/m²) were randomized to either an intensive lifestyle intervention (n = 592) or usual care (n = 588) 1–5 years postpartum. Body weight indices (including weight, body mass index (BMI), waist circumference, and body fat percentage) and diabetes status were assessed annually. Cox regression models combined with mediation analysis were employed to estimate the proportion of the intervention’s effect mediated by candidate factors.

Results

Weight reduction demonstrated a continuous, dose-dependent association with a lower incidence of postpartum diabetes, without evidence of a threshold effect. Overall, changes in weight indices mediated 13.0–18.8% of the intervention’s protective effect. This mediation proportion was significantly higher among women who were overweight at baseline, accounting for 24.3–34.7% of the intervention’s benefit. Within this overweight subgroup, changes in waist circumference and body fat individually accounted for over 30% of the total mediated effect.

Conclusions

Weight loss is a quantifiable and clinically relevant, though partial, mechanism by which lifestyle intervention reduces postpartum diabetes risk, particularly in overweight women with a history of GDM. These findings underscore the importance of continuous monitoring and management of even modest weight changes as a key component of diabetes prevention strategies for this high-risk population.

Trial registration

NCT01554358. Retrospectively registered clinical trial.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12916-026-04644-y.

Research in context

What is already known about this subject?

Women with prior gestational diabetes mellitus (GDM) have a 20–50% risk of developing postpartum diabetes. Intensive lifestyle interventions effectively reduce this risk, but the precise mechanisms underlying this success remain elusive.

What is the key question?

To what extent does weight change mediate the reduction in postpartum diabetes risk achieved by intensive lifestyle intervention in women with a history of GDM?

What are the new findings?

Weight reduction is a quantifiable, dose-dependent mediator of the intervention’s protective effect against postpartum diabetes. Weight changes accounted for 13.0–18.8% of the overall benefit, with a significantly higher mediation proportion (24.3–34.7%) observed in overweight women at baseline, where waist circumference and body fat were key contributors.

How might this impact on clinical practice in the foreseeable future?

Clinicians should prioritize continuous monitoring and management of even modest weight changes in women with a history of GDM, as this represents a key, clinically relevant pathway for diabetes prevention.

Background

The global surge in diabetes incidence, reaching 10.2% worldwide and 12.4% in China, presents a significant public health challenge [1, 2]. Gestational diabetes mellitus (GDM) serves as a critical precursor to the development of diabetes, affecting about 14% of pregnant women worldwide [3]. GDM not only leads to short-term adverse outcomes for both mothers and fetuses, such as preterm birth, macrosomia, and neonatal morbidities [4], but also substantially increases the long-term risk of diabetes in mothers during the postpartum period. Studies have shown that among women with a history of GDM, approximately 20–50% will develop diabetes within 5 years after childbirth [5–7], depending on the diagnosis method of GDM, screening criteria, and ethnicity [8].

While lifestyle interventions have proven effective in delaying diabetes onset in general populations with impaired glucose tolerance (IGT) [9–14], the evidence regarding their efficacy in post-GDM women has been mixed. Although a post hoc analysis of the Diabetes Prevention Program demonstrated a significant reduction in diabetes incidence among women with a history of GDM [15, 16], other dedicated randomized controlled trials (RCTs) in this population have often yielded inconsistent results, frequently attributed to methodological limitations like small sample sizes and short follow-up periods [17–21]. Recently, our large-scale Tianjin GDM Prevention Program (Clinicaltrials.gov Identifier: NCT01554358), a randomized clinical trial, has made significant contributions by providing compelling evidence that a long-term, individually designed intensive diet and physical activity program could successfully reduce the incidence of postpartum diabetes by 46% in young women with a history of GDM [22].

However, the precise mechanisms underlying this success remain elusive. Although weight management is a cornerstone of diabetes prevention, the extent to which it mediates the effect of lifestyle interventions specifically in post-GDM women has not been rigorously quantified. This study aims to translate the Tianjin GDM Prevention Program’s success into actionable knowledge by investigating the mechanistic pathways of lifestyle intervention in preventing postpartum diabetes among women with prior GDM. We specifically focus on quantifying, for the first time, the extent to which weight change mediates the observed reduction in diabetes risk, providing essential evidence to guide the development of more targeted and effective clinical strategies for diabetes prevention in this high-risk group.

Methods

Study design and participants

The Tianjin GDM Prevention Program is a 4-year, two-arm randomized controlled trial conducted at Tianjin Women and Children’s Health Center, Tianjin, China. The detailed study design has been described elsewhere [23, 24]. The study protocol (Additional file 1: Trial Proposal) was approved by the Human Subjects Committee of Tianjin Women’s and Children’s Health Center (the approval numbers are 2009–01, 2013–03-01, and 2017–03-01) and all study participants provided written informed consent [23, 24].

Between August 2009 and July 2011, women diagnosed with GDM between 2005 and 2009 in six urban districts of Tianjin, China (N = 4644) were recruited 1–5 years postpartum. The GDM diagnoses, using the 1999 World Health Organization (WHO) criteria [25], encompassed both diabetes (fasting glucose ≥ 7.0 mmol/L or 2-h glucose ≥ 11.1 mmol/L) and IGT (2-h glucose ≥ 7.8 mmol/L and < 11.1 mmol/L) after a 75-g oral glucose tolerance test (OGTT). These diagnoses were made within a near-universal screening program that achieved a participation rate exceeding 91% during the diagnostic period [26]. This recruitment was facilitated by a comprehensive health care registration system that provided health and contact information for women with GDM in Tianjin. Of these women, 1263 with GDM, aged 20 to 49 years, completed the baseline survey. Crucially, this baseline survey identified two distinct subgroups. A subgroup of 83 women were newly diagnosed with diabetes at baseline and subsequently participated in a 9-month lifestyle intervention program. The primary focus of this manuscript is the remaining 1180 eligible women with a history of GDM but without diabetes at baseline. These 1180 women, with an average postpartum duration of 2.26 years, were randomly assigned to receive either a 4-year lifestyle intervention (n = 586) or usual care (n = 594).

The intervention methods have been previously described [23, 24]. The main components of the intervention included six face-to-face meetings with study nutritionists in the first year, followed by two additional sessions and two telephone conversations each subsequent year. Participants in the intervention group received specific guidance to achieve the intervention goals: (1) a reduction of 5 to 10% of initial body weight in overweight women [body mass index (BMI) ≥ 24 kg/m²], with no weight reduction recommended for normal-weight women (BMI < 24 kg/m²) [27]; (2) limiting fat intake to less than 30% of total energy consumed; (3) ensuring carbohydrate intake constitutes between 55 and 65% of total energy consumed; (4) consuming 20 to 30 g of fiber per day; and (5) engaging in moderate or vigorous physical activity for at least 30 min per day, 7 days a week [23, 24].

The women in the usual care group were given general oral and written information about diabetes awareness, dietary modification, and physical activity increase at baseline and subsequent annual visits, but no specific individualized programs were offered [23, 24].

Measurement

All participants completed a questionnaire and underwent a physical examination, including anthropometric measurements and blood pressure assessments at baseline and during annual visits [23]. The questionnaire collected information on socio-demographics, family history, dietary habits (through a self-administered food frequency questionnaire, and 3-day 24-h food records), alcohol intake, smoking habits, and physical activity. The 3-day 24-h food records, the self-administered food frequency questionnaire, and the physical activity questionnaire had all been validated in the China National Nutrition and Health Survey in 2002 [28–30].

Body weight, height, waist circumference, and body fat were measured for all participants using a standardized protocol by specially trained research doctors. Height was measured without shoes but wearing light clothing, and recorded to the nearest centimeter. Weight and body fat were assessed using a body composition analyzer (SC-240, Tanita, Tokyo, Japan), which has been validated for acceptable accuracy in estimating body fat compared to dual-energy X-ray absorptiometry [31]. Waist circumference was measured midway between the lower rib margin and the iliac crest, and recorded to the nearest half centimeter. BMI was calculated by dividing current weight in kilograms by the square of height in meters. Percent change in body weight was calculated by dividing the change in weight by the initial body weight.

At baseline and the annual visits, a standard 2-h 75-g glucose tolerance test was conducted and blood samples were collected from all participants after an overnight fast of at least 12 h. Plasma glucose was measured using an automatic analyzer (TBA-120FR; Toshiba, Tokyo, Japan) at 0 and 2 h after administration during the OGTT.

Assessment of outcomes

Participants were followed until the date of the diagnosis of diabetes, the last date of an OGTT, or death, whichever occurred first, up to December 2020. Incident diabetes was diagnosed based on fasting glucose concentration ≥ 7.0 mmol/L and/or 2-h glucose ≥ 11.1 mmol/L, according to the 1999 WHO criteria [25]. Diagnosis was performed using an annual 2-h 75-g OGTT, with the second test conducted at least a week after the first OGTT [23]. For participants who missed any OGTTs during follow-up, information on medical history including physician-diagnosed diabetes and medications was collected at the end of the study.

Statistical analysis

Differences between the lifestyle intervention and control groups were presented as means (standard deviation, SD) for continuous variables and frequencies (percentages) for categorical variables, with comparisons made using T-tests and chi-square (χ²) tests, respectively. The Cox proportional hazards model was used to estimate the hazard ratio (HR) for the development of postpartum diabetes between two study groups over the study period. Weight change indices from baseline to years 1–4 across different groups were analyzed using generalized linear models.

To assess the dose–response relationship between weight change indices and the risk of postpartum diabetes, we employed restricted cubic splines (RCS) by constructing Cox regression models. In these models, weight change indices were entered as continuous variables to visualize their association with the hazard of postpartum diabetes.

Mediation analysis was performed to evaluate and quantify the potential mediating role of weight change indices in the effect of lifestyle intervention on postpartum diabetes among women with previous GDM. Theoretically, if the independent variable X influences the dependent variable Y through a certain variable M, then M is considered a mediator between variables X and Y. The mediation analysis equations using linear regressions were as follows: Y = cX + e₁; M = aX + e₂; Y′ = c′X + bM + e₃. A three-step approach was used to test the mediation hypotheses [32]. Step 1, the total effect of lifestyle intervention on postpartum diabetes was examined without considering weight change indices. Step 2, the indirect/mediating effect was assessed by examining the relationships between weight change indices and lifestyle intervention, as well as between weight change indices and postpartum diabetes. Step 3, the direct effect was tested by including weight change indices in the model from step 1. The mediation package in R software was used to automatically obtain the indirect/mediating effect, with the deviation correction self-help method employed to calculate the 95% confidence interval (CI) of the coefficients. The proportion of total effect mediated by the indirect/mediating effect was also calculated using these estimates. Potential influencing factors (age, baseline BMI, and baseline fasting plasma glucose) were adjusted in the mediation analysis to assess the independent mediating role of weight change indices in the effect of lifestyle intervention on postpartum diabetes.

All statistical analyses were performed with SPSS Statistics V.25.0 (IBM SPSS, Chicago, IL, USA) and R 4.0 (R Foundation for Statistical Computing, Vienna, Austria) software programs. A two-sided P value < 0.05 was considered statistically significant.

Results

Baseline characteristics were comparable between participants in the intervention and control groups (Additional file 2: Table S1). The number of participants completing the years 1–4 follow-up surveys and the number of newly diagnosed postpartum diabetes were detailed in Fig. 1. Over a mean follow-up of 4.5 years, the intervention group had a 46% lower cumulative incidence of postpartum diabetes than the control group (HR 0.54, 95% CI: 0.11–0.67).

Fig. 1.

Participant flow chart

The values of changes in various weight indices between the intervention and control groups were shown in Table 1 and Fig. 2. During years 1–4 follow-up, participants in the lifestyle intervention group experienced greater decreases or smaller increases in body weight, waist circumference, BMI, and body fat compared to the control group. When stratified by baseline BMI, the weight indices all showed more pronounced differences between the intervention and control groups among overweight women at baseline (Fig. 2). The relationship of changes in weight indices with postpartum diabetes were shown in Table 1. Women who developed postpartum diabetes had significantly larger weight increase or smaller weight decrease compared to women who were free of postpartum diabetes.

Table 1.

The relationship of change in weight indices with lifestyle intervention and incident postpartum diabetes mellitus risk

	Relationship with lifestyle intervention			Relationship with incident postpartum diabetes mellitus
	Intervention group (N = 586)	Control group (N = 594)	P value	Without diabetes (N = 1115)	With diabetes (N = 65)	P value
Change in body weight, kg
At year 1	− 0.79 (− 1.04 to − 0.54)	− 0.19 (− 0.43 to 0.06)	0.001	− 0.53 (− 0.71 to − 0.35)	0.31 (− 0.44 to 1.06)	0.032
At year 2	− 0.29 (− 0.57 to − 0.01)	0.18 (− 0.09 to 0.45)	0.018	− 0.10 (− 0.30 to 0.10)	0.78 (− 0.05 to 1.61)	0.043
At year 3	0.30 (0.01 to 0.59)	0.69 (0.40 to 0.98)	0.059	0.42 (0.21 to 0.63)	1.90 (1.02 to 2.77)	0.001
At years 4–6	1.25 (0.92 to 1.58)	1.23 (0.90 to 1.55)	0.923	1.22 (0.98 to 1.46)	1.58 (0.60 to 2.56)	0.482
Mean value	0.12 (− 0.12 to 0.36)	0.48 (0.24 to 0.72)	0.036	0.25 (0.08 to 0.42)	1.14 (0.42 to 1.86)	0.018
Percent change in body weight, %
At year 1	− 1.07 (− 1.44 to − 0.69)	− 0.16 (− 0.54 to 0.21)	0.001	− 0.69 (− 0.96 to − 0.41)	0.69 (− 0.44 to 1.83)	0.021
At year 2	− 0.24 (− 0.67 to 0.18)	0.42 (− 0.01 to 0.84)	0.031	0.02 (− 0.29 to 0.32)	1.32 (0.04 to 2.59)	0.052
At year 3	0.80 (0.35 to 1.25)	1.35 (0.91 to 1.79)	0.087	0.96 (0.64 to 1.29)	2.99 (1.66 to 4.33)	0.004
At years 4–6	2.51 (2.01 to 3.02)	2.39 (1.89 to 2.89)	0.739	2.44 (2.08 to 2.81)	2.65 (1.14 to 4.16)	0.790
Mean value	0.50 (0.13 to 0.87)	1.00 (0.63 to 1.36)	0.060	0.68 (0.42 to 0.95)	1.92 (0.81 to 3.02)	0.034
Change in body mass index, kg/m2
At year 1	− 0.31 (− 0.40 to − 0.21)	− 0.07 (− 0.17 to 0.02)	0.001	− 0.21 (− 0.28 to − 0.14)	0.13 (− 0.16 to 0.42)	0.026
At year 2	− 0.11 (− 0.22 to 0.00)	0.07 (− 0.04 to 0.18)	0.020	− 0.04 (− 0.12 to 0.04)	0.32 (0.00 to 0.64)	0.033
At year 3	0.12 (0.01 to 0.23)	0.27 (0.16 to 0.38)	0.068	0.16 (0.08 to 0.24)	0.75 (0.41 to 1.08)	0.001
At years 4–6	0.49 (0.36 to 0.62)	0.48 (0.36 to 0.61)	0.931	0.48 (0.39 to 0.57)	0.64 (0.25 to 1.02)	0.430
Mean value	0.05 (− 0.04 to 0.14)	0.19 (0.09 to 0.28)	0.038	0.10 (0.03 to 0.17)	0.46 (0.18 to 0.74)	0.014
Change in waist circumference, cm
At year 1	− 1.70 (− 2.09 to − 1.32)	− 0.85 (− 1.23 to − 0.46)	0.002	− 1.30 (− 1.58 to − 1.02)	− 0.81 (− 1.98 to 0.35)	0.426
At year 2	− 1.14 (− 1.54 to − 0.75)	− 0.59 (− 0.98 to − 0.20)	0.050	− 0.91 (− 1.19 to − 0.63)	− 0.12 (− 1.30 to 1.06)	0.200
At year 3	− 0.27 (− 0.64 to 0.11)	0.12 (− 0.25 to 0.49)	0.153	− 0.16 (− 0.42 to 0.11)	1.31 (0.19 to 2.42)	0.013
At years 4–6	− 0.70 (− 1.10 to − 0.30)	− 0.60 (− 1.00 to − 0.19)	0.719	− 0.71 (− 1.01 to − 0.42)	0.49 (− 0.72 to 1.70)	0.058
Mean value	− 0.95 (− 1.28 to − 0.63)	− 0.48 (− 0.80 to − 0.16)	0.043	− 0.77 (− 1.01 to − 0.53)	0.22 (− 0.76 to 1.19)	0.054
Change in body fat percent, %
At year 1	− 0.48 (− 0.65 to − 0.30)	− 0.02 (− 0.19 to 0.15)	< 0.001	− 0.28 (− 0.40 to − 0.15)	0.24 (− 0.29 to 0.76)	0.062
At year 2	− 0.20 (− 0.39 to − 0.01)	0.11 (− 0.08 to 0.30)	0.021	− 0.07 (− 0.21 to 0.07)	0.41 (− 0.15 to 0.97)	0.104
At year 3	0.33 (0.13 to 0.52)	0.58 (0.39 to 0.77)	0.068	0.41 (0.28 to 0.55)	1.10 (0.53 to 1.67)	0.023
At years 4–6	0.95 (0.74 to 1.16)	0.85 (0.64 to 1.06)	0.487	0.89 (0.74 to 1.05)	1.00 (0.37 to 1.63)	0.746
Mean value	0.15 (− 0.01 to 0.31)	0.38 (0.22 to 0.54)	0.048	0.24 (0.12 to 0.36)	0.69 (0.21 to 1.17)	0.076

Values are mean (95% confidence interval). Percent change in body weight was calculated by dividing the change in weight by the initial body weight

Fig. 2.

Weight change indices over 4 years by lifestyle intervention and baseline body mass index

The dose–response relationships between various weight change indices and the risk of postpartum diabetes mellitus were elucidated by RCS analyses, as illustrated in Fig. 3. The RCS analysis demonstrated a significant positive association between weight change indices and the risk of developing postpartum diabetes. Specifically, weight gain was associated with an elevated risk of diabetes, whereas weight loss was linked to a decreased risk. Crucially, the analysis did not identify a distinct threshold; rather, risk modification appeared continuous with weight changes in both gaining and losing directions. When mean values of these weight change indices were analyzed as continuous variables, multivariable-adjusted hazard ratios (HRs) for postpartum diabetes associated with each unit increase in body weight change (kg), percentage change in body weight (%), BMI (kg/m²), waist circumference change (cm), and body fat change (%) were 1.10 (95% CI: 1.03–1.18), 1.08 (95% CI: 1.03–1.14), 1.30 (95% CI: 1.08–1.55), 1.06 (95% CI: 1.00–1.13), and 1.23 (95% CI: 1.09–1.40), respectively. These findings underscore the graded impact of postpartum weight modifications on diabetes susceptibility.

Fig. 3.

Restricted cubic spline analysis of weight change and postpartum diabetes mellitus risk

The mediation analysis results (illustrated in Fig. 4) revealed that lifestyle intervention’s effect on reduced postpartum diabetes incidence was partially mediated by changes in body weight (years 1–3), BMI (years 1–3), waist circumference (year 2), and body fat (years 1–3) (all P < 0.05). These effects were independent of age, baseline BMI, and baseline fasting plasma glucose. Specifically, the mediating effects of mean changes in body weight, percentage body weight, waist circumference, BMI, and body fat on postpartum diabetes risk were 16.0%, 16.1%, 16.3%, 13.0%, and 18.8%, respectively. The mediation was more pronounced among women who were overweight at baseline. In this subgroup, the mediating effects of changes in body weight, percentage body weight change, waist circumference, BMI, and body fat on postpartum diabetes risk were 24.3%, 25.4%, 24.4%, 34.7%, and 30.7%, respectively. Notably, changes in waist circumference and body fat individually accounted for over 30% of this mediation within the overweight group. Conversely, this mediation was not statistically significant among women with normal weight at baseline.

Fig. 4.

Effects and proportions mediated by change in weight change indices over time between lifestyle intervention and postpartum diabetes mellitus. Adjusted for baseline variables (age, body mass index, and fasting plasma glucose)

Discussion

Drawing insights from the Tianjin GDM Prevention Program, this prespecified post hoc analysis confirmed that participants in the lifestyle intervention group experienced greater decreases or smaller increases in body weight, waist circumference, BMI, and body fat compared to the control group. Weight change indices demonstrated a continuous, dose-dependent positive association with postpartum diabetes risk across both gaining and losing directions, without a specific threshold. Crucially, we provide novel evidence that the effect of lifestyle intervention is at least partially mediated by changes in these weight indices. Specifically, these mediators accounted for 13.0 to 18.8% of the intervention’s effect in the overall cohort, and a substantially greater proportion, 24.3 to 34.7%, among women who were overweight at baseline. Importantly, these mediating effects were robust, remaining significant and independent of age, baseline BMI, and baseline fasting plasma glucose.

Our findings align with and extend evidence from landmark diabetes prevention trials. Studies including the Da Qing [12, 14], Finnish [9, 10], and Diabetes Prevention Program [11, 13] trials have demonstrated the efficacy of lifestyle interventions, with the Diabetes Prevention Program notably showing benefits in women with prior GDM [15, 16]. While weight loss is a common goal in these interventions, our study moves beyond establishing efficacy to elucidating the underlying mechanism. The established mechanism for weight loss-induced diabetes prevention is largely explained by the “twin cycle hypothesis” [33], which posits that chronic caloric excess leads to fat accumulation (particularly in the liver), thereby promoting insulin resistance and hyperinsulinemia—this excess fat can then impair insulin production. Weight loss, by reducing body fat (including visceral fat), helps restore insulin sensitivity and glucose metabolism, thereby lowering diabetes risk [34, 35].

While a substantial body of research has confirmed the general association between weight loss and a reduced risk of type 2 diabetes [36], our study offers a unique contribution: we have quantified the specific mediating role of body weight change within the context of a structured, intensive lifestyle intervention aimed at preventing postpartum diabetes in women with a history of GDM. Unlike broader cohort studies evaluating weight changes [37], our data originated from a large-scale RCT—the gold standard for evaluating health efficacy. This allowed us to precisely estimate the proportion of the intervention’s overall protective effect attributable to weight loss. Furthermore, we have revealed the significantly enhanced importance of this mediating pathway in overweight women and detailed the contributions of specific weight indices such as waist circumference and body fat percentage, adding a more granular mechanistic layer to existing findings.

Although weight loss is widely acknowledged as a cornerstone of diabetes prevention in overweight and obese individuals, the precise quantitative contribution of weight change to the efficacy of lifestyle interventions in reducing diabetes risk has remained less defined. A UK study (1984–2007) found that BMI accounted for only 26% of the increase in diabetes cases [38]. This indicates that other lifestyle factors also play a critical role in the rising diabetes rates, such as physical inactivity, exposure to noise or fine dust, short/disturbed sleep, smoking, stress, depression, and low socioeconomic status [39]. Furthermore, research on the impact of weight reduction on visceral fat has shown diminishing returns beyond a 5–10% weight loss [35]. Based on a prespecified post hoc analysis of a large-scale RCT, we demonstrate that weight change mediates a substantial, previously unmeasured portion (13.0–18.8%) of the intervention’s effect, with this proportion being markedly higher (24.3–34.7%) in overweight women. This not only confirms weight management as a critical pathway but also highlights that a larger share of the benefit—especially in normal-weight women—stems from non-weight-related mechanisms (e.g., improved insulin sensitivity or cardiorespiratory fitness via increased physical activity), offering a new direction for optimizing future interventions.

The intervention specifically prescribed a 5–10% weight-loss target for overweight women, thereby creating a more direct and potent pathway for reducing diabetes risk through weight change. Our 1-year findings showed significantly higher weight-loss goal attainment in the intervention group compared to controls among overweight women (30.2% vs. 17.5%, P = 0.003) [24]. For normal-weight women, who were not assigned a weight-loss target, the benefits of the intervention likely operated through alternative, non-weight-related physiological mechanisms, with a marked increase in physical activity goal achievement (26.6% vs. 13.0%, P < 0.001) [24]. This contrast explains why weight change contributed to a much greater extent to risk reduction in the overweight subgroup. Overall, our results emphasize that the mediating role of weight change is not solely a biological phenomenon but is also profoundly influenced by the intervention’s behavioral objectives—specifically, targeting weight loss in overweight women, while activating non-weight-related behavioral mechanisms (such as physical activity) in individuals with normal weight.

Additionally, the observed continuous dose–response relationship suggests that proactive management of weight change—even changes not meeting a strict threshold—confers benefits. Furthermore, identifying weight as a partial mediator indicates the existence of other crucial, non-weight-related pathways through which lifestyle interventions impact diabetes risk. This underscores a vital area for future research: developing more effective “diabetes-protective lifestyles.”

Interestingly, while participants in the intervention group generally experienced weight loss in the first year, both groups showed a weight gain trend from the second year onward. Despite this, the intervention group consistently exhibited smaller weight increases compared to the control group—a pattern particularly evident among overweight women. Our RCS analysis revealed a continuous, dose-dependent positive association between weight change indices and postpartum diabetes risk, extending across both weight gain and loss directions and notably lacking a specific threshold. These findings not only strongly confirm the critical role of lifestyle interventions in postpartum weight management but also unveil a complex, bidirectional relationship between weight change and diabetes risk. Crucially, the absence of a defined “safe” range for weight change emphasizes that clinical practice should prioritize encouraging postpartum women to adopt sustained healthy lifestyles and closely monitor weight trends, rather than solely focusing on achieving specific weight targets [40].

Our trial confirms the effectiveness of the Tianjin GDM Prevention Program in reducing GDM incidence and innovatively quantifies the mediating role of weight change. However, the generalizability of the findings remains a key limitation that must be carefully considered. The study population included only Chinese women with GDM diagnosed according to the 1999 WHO criteria. Therefore, the observed effects may differ in populations diagnosed using updated criteria—such as the 2013 WHO standards that are likely to identify a larger number of GDM cases [41, 42]. Moreover, the 1999 WHO criteria define GDM as a state analogous to IGT in non-pregnant individuals. This similarity suggests that the Tianjin GDM Prevention Program may work in a way that is comparable to well-known diabetes prevention studies in people with IGT [9–14], which can help in understanding its long-term effects. However, it also means that the results may not apply as broadly to other groups or situations. Additionally, the efficacy of lifestyle interventions can vary significantly across ethnic groups [8, 43]. Differences in genetic susceptibility, lifestyle patterns, and metabolic responses among populations necessitate caution when extrapolating results from Chinese women to other groups. Future research should validate such interventions in diverse, multi-ethnic cohorts and across a broader baseline BMI range. Finally, it should be noted that diabetes diagnosis in this study has required confirmation through two positive plasma glucose tests during 75-g OGTTs or a clinical diagnosis by a physician. Although this approach ensures diagnostic reliability, it may yield a slightly lower observed incidence of diabetes compared with studies using a single OGTT. Further research is still needed (especially utilizing glycated hemoglobin (HbA1c) as a primary diagnostic tool and conducting long-term follow-up in diverse populations) to fully elucidate long-term diabetes risk and intervention effectiveness within the evolving landscape of diabetes diagnostics. In summary, the observed reduction in GDM incidence mediated by weight change in our study must be interpreted in the context of the specific study population and the diagnostic criteria employed. Another limitation was that the proportion of participants completing all assessments declined over time, which may have affected the precision of estimates in later follow-up years. To mitigate potential biases from participant attrition, we performed an intention-to-treat analysis that included all randomized participants, ensuring comprehensive results.

Conclusions

In conclusion, this study provides the first quantitative evidence that the success of lifestyle interventions in preventing postpartum diabetes in women with prior GDM is partially mediated by body weight change—this pathway is particularly important for overweight individuals. These findings strongly emphasize the critical importance of monitoring weight change as a key indicator of lifestyle intervention efficacy in this population. Future research should focus on elucidating the non-weight-related mechanisms and developing more comprehensive diabetes-protective lifestyle strategies.

Abbreviations

BMI

Body mass index

CI

Confidence interval

GDM

Gestational diabetes mellitus

HR

Hazard ratio

IGT

Impaired glucose tolerance

OGTT

Oral glucose tolerance test

RCS

Restricted cubic splines

SD

Standard deviation

Authors’ contributions

WQL and GH were responsible for accessing all study data, ensuring its integrity, and performing accurate data analysis. The concept and design of the study were contributed by WQL, HL, LF, LW, SZ, and GH. Data acquisition, analysis, and interpretation involved WL, GL, JL, YS, RG, XY, and ZY. WQL was responsible for drafting the manuscript, with critical revision for important intellectual content provided by YS, RG, YZ, XY, ZY, and GH. Statistical analysis was performed by WQL. Funding was obtained by GH and WQL. Administrative, technical, and material support was provided by HK L, LF, LW, SZ, WL, GL, and JL. All authors read and approved the final manuscript.

Funding

This study is supported by the grant from the European Foundation for the Study of Diabetes (EFSD)/Chinese Diabetes Society (CDS)/Lilly program for Collaborative Research between China and Europe, the Natural Science Foundation of Tianjin, China (23JCYBJC00960), and the Tianjin Health Research Project, China (TJWJ2025QN096). Dr. Gang Hu is partly supported by grants from the National Institute of Diabetes and Digestive and Kidney Diseases (R01DK132011) and the National Institute of General Medical Sciences (U54GM104940) of the National Institutes of Health. The funders had no role in the trial design, conduct, data analysis, or reporting of this study.

Data availability

The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

The study was conducted in accordance with the ethical standards of institutional and/or national research committees and following the principles of the 1964 Declaration of Helsinki and later amendments. The study protocol was approved by the Human Subjects Committee of Tianjin Women and Children’s Health Center (the approval numbers are 2009–01, 2013–03-01, and 2017–03-01). All study participants provided written informed consent prior to their enrollment in the study.

Consent for publication

Not applicable. This manuscript does not contain any individual person’s data in any form.

Competing interests

The authors declare no competing interests.