This study utilized a cluster randomized controlled trial design to investigate the 18-month health impact of ILSM among 320 women with a history of GDM in rural China. The study was approved by the local ethics committee and registered at the Chinese Clinical Trial Registry. The details of the rationale, study and results of the 6-month follow-up have been published elsewhere [22, 23].

The study was conducted in Yongding County and You County, Hunan Province in south-central China. Yongding County comprising 17 towns, has a large ethnic minority population (~ 50%, mainly Tujia and Miao population) in western Hunan Province; and You County comprising 14 towns, has a large ethnic Han population in eastern Hunan Province. The diversity (i.e., application to different geographic locations, ethnic groups, health care systems) of the included populations was used to test whether the ILSM program could be generalized to diverse underserved populations.

The town (cluster) was the unit of randomization, and a randomization protocol available on the internet (http://stattrek.com/statistics/random-number-generator.aspx) was used. No specific eligibility criteria for towns were identified. Eight towns were randomly selected from each county, yielding 16 towns. Following recruitment, the 16 towns were randomly allocated in a 1:1 ratio to either the intervention or the control group with stratification for the two counties, and each group included 8 towns (4 towns per county). Considering the realistic resource restriction in these towns (e.g., lack of resources in town-level hospitals, shortage of health care providers), two local county-level hospitals with the highest number of pregnant and birth-giving women (one per county), located in the center of the counties within a close distance from these towns, was chosen as the research sites to implement the study.

Inclusion criteria were: 1) adult women with a history of GDM; 2) 6 weeks to 10 years postpartum; 3) living in the allocated towns and intending to live in these towns for at least 3 years; 4) having telephone access, and 5) able to read and speak in Mandarin Chinese. Exclusion criteria were: 1) women who were pregnant or planned to be pregnant within the next 3 years; 2) a diagnosis of diabetes; and 3) other serious health problems, such as physical or cognitive disability.

Eligible women who delivered babies within the past 10 years at the two research sites were identified through a medical record review. Local registered nurses who received standardized research protocol training contacted potential participants via telephone or in-person at a clinical visit. Nurses explained the research study to interested women, confirmed their eligibility, and obtained consent.

The timeline of the intervention activities and measurements for ILSM group and control group is showed in Table 1. Both groups received usual care based on current clinical guidelines, including general education about their T2D risk as well as a T2D prevention education brochure provided to each participant.

The intervention group also received the ILSM program from eight trained local nurses following the ILSM protocol reported elsewhere [23]. Before the ILSM program, the nurses received a structured five-day training from the research team using the Train the Trainer Model, which engages master trainers (research team members) in coaching new trainers (eight local nurses) to be competent to carry out the ILSM intervention. The ILSM training entailed self-study, class sessions, and live practice, with homework and practice between training. The nurses were required to pass a final evaluation held by our research team, which included a scenario simulation test and a personal interview to ensure they were equipped with essential intervention delivery skills.

Each nurse conducted the intervention for a group of ~ 20 participants from the same town at the research-designated hospitals. Participants were invited to attend six biweekly group seminars (90 minutes) and eight telephone consultation sessions (20 minutes each). The topics of group seminars included orientation and goal-setting, healthy eating patterns, physical activity, stress management, family support, family lifestyle patterns, and relapse prevention and farewell [23]. All content in the ILSM program was tailored to the context of rural women with a history of GDM. During the intervention period, a research assistant acted as a resource person at each research site and attended all sessions to assess intervention fidelity via a checklist (including evaluation concerning four domains: adherence, exposure, quality of delivery, and participant responsiveness). More details about the fidelity checklist are provided in Additional file 1: Appendix I.

To promote the long-term efficacy of the intervention, a series of strategies such as setting recursion goals and providing a environmental support, was used. For example, participants were required to set various small and easy-to-implement goals concerning health behavior at the early stage of intervention to ensure their early behavioral wins. These early wins help convince participants that behavior change is possible, thus induced into a recursive process and eventually sparking a positive and continual behavior change loop. In addition, participants were equipped with resources of health behavior change during the intervention, and these resources create supportive physical and psychological environment after intervention. By interacting with these essential elements and repeatedly reinforcing the desired behavior via environment, long-term benefits will tend to persist and reinforce the behavior change.

Due to the nature of educational and behavioral interventions, it was not feasible to blind participants and local nurses (investigators), though data assessors were blinded. Nurses and study participants were asked to sign an agreement that they would not share the intervention materials or protocol with others before the completion of the study.

Data were collected at baseline, 6-month, and 18-month. At each visit, all participants were invited to complete questionnaires on T2D risk assessment, lifestyle behavior, and psychological outcomes. At baseline, all participants also completed a demographic and clinical questionnaire. Data were collected in a quiet room at the research site by trained research assistants. At the same time, local nurses collected blood samples and completed physical examinations for physiological data. Data for women in intervention and control groups were collected on separate days to avoid contamination.

The demographic and clinical data included age, ethnicity, education, occupation, family income, and the number of months after delivery.

Primary outcome: T2D risk score was the primary outcome, as our target population was high T2D risk groups with normal glycemia rather than populations with impaired glucose tolerance. Adults with a high risk for T2D and normal glycemia have limited room to improve glycemic outcomes; thus, glycemic outcomes are unsuitable as a primary outcome. The T2D risk score was developed based on specific modifiable diabetes risk factors (such as BMI, waist circumference, physical activity, and dietary intake), and also includes some unmodifiable risk factors, which is in line with the theoretical mechanism of most diabetes prevention programs [24].

The T2D risk score was measured by the Chinese Diabetes Risk Scale (CHINARISK) [25], adapted from the Canadian Diabetes Risk Questionnaire [26]. This scale systematically combines modifiable and unmodifiable diabetes risk factors in order to identify people who may develop T2D in the next 10 years. Total scores range from 0 to 88; higher scores represent a greater risk of T2D. The questionnaire has a positive predictive value of 57% and a negative predictive value of 78%, with a sensitivity of 73% [25].

Secondary outcome: Glycemic outcomes included fasting blood glucose (FBG) and 2 h Oral Glucose Tolerance Test (OGTT) [27]. Venous blood samples were collected after overnight fasting, followed by blood samples taken 2 hours after consuming 75 g of glucose.

Weight-related outcomes included BMI and waist circumference. BMI was calculated by dividing body weight (kilograms) by height squared (meters); Waist circumference was measured at the midpoint between the highest point of the iliac crest and the lowest rib according to WHO standard [28].

Behavioral outcomes: Physical activity was assessed by the validated Chinese version of the International Physical Activity Questionnaire (Short Form) [29]; Fruit/vegetable consumption was measured by an item on the CHINARISK scale; The Intention to Eat Low Glycemic-index Foods was assessed with a 24-item questionnaire that uses a 7-point Likert scale, on which higher scores indicate a greater intention to eat low glycemic-index foods [30].

Psychological outcomes: Perceived stress was measured by the 14-item perceived stress scale [31]. Quality of life was assessed using the WHOQOL-BREF questionnaire, which includes 26 items and evaluates physiological, psychological, social relations, and environmental domains of quality of life. General self-efficacy was measured using the 10-item general self-efficacy questionnaires [32]. Social support was measured using the 10-item social support rating scale (SSRS) [33]. All the measurements used in this study with well-documented psychometric properties and have been used in Chinese populations in China [34,35,36,37].

The analyses were done at the individual level in SPSS (Version 22.0; Armonk, NY, United States). The double-entry data method was adopted to ensure data accuracy via the EpiData 3.0 software (EpiData Association, Odense, Denmark). All randomly assigned participants (N = 320) entered the intention-to-treat analysis, which means that the randomization groups were used in this analysis, irrespective of any protocol violations. We ensured that the outcomes for participants who withdrew from the trial prior to the 6- or 18-month were retained in the analysis.

The data were presented as means with SDs or as counts with percentages. Descriptive statistics were used to describe demographic and clinical characteristics. The demographic and clinical data of the intervention and control groups were compared using two independent samples t-tests and Chi-square tests. Repeated measure analyses (three points) were conducted using a generalized estimating equation (GEE) model to compare the differences between two groups in T2D risk scores, glycemic, weight-related, behavioral, and psychological variables from baseline to 18-month follow-up. GEEs were developed as an extension of the general linear model to analyze longitudinal and other correlated data. GEE models take into account the correlation between repeated measurements in the same subject; models do not require complete data and can be fitted even when individuals do not have observations at all time points [38]. We added an interaction term (group by time) to each model to investigate the interactive effects of intervention and time. Models were performed unadjusted and adjusted for age, months after delivery, ethnicity, education, marriage, occupation, and family income. To assess the potential effects of clustering, we calculated the eta2 coefficient for each of the three outcomes and three time periods. The eta2 values for FBG ranged from 0.319 to 0.491. Coefficients for OGTT-2 h and the T2D risk score were substantially smaller, ranging from 0.049 to 0.114. In general, larger eta2 values are associated with larger standard errors, wider confidence intervals, and more conservative p-values. In order to adjust for such effects, we used fixed effects GEE regression models where the cluster itself is included as a factor within the model.

To assess if the intervention effect was statistically different between subgroups, we conducted several subgroup analyses for T2D risk scores: research site (You County, Yongding County), ethnicity (Han, minority), BMI (≤24 kg/ m2, > 24 kg/ m2), waist circumference (≤80 cm, > 80 cm), glucose dysregulation (FBG ≤6.1 mmol/L or OGTT-2 h ≤7.8 mmol/L, FBG > 6.1 mmol/L or OGTT-2 h > 7.8 mmol/L) and months after delivery (≤12, > 12). We included interaction terms (group by subgroup) in the GEE models to assess differences between subgroups.