Our study successfully constructed and validated a logistic regression-based interpregnancy weight management-focused prediction model that exhibited high accuracy in forecasting HDP in subsequent pregnancies. The existing prediction models are designed for assessment at the post-conception stage [15, 16, 25], and this is the first prediction model for interpregnancy weight management, a type of pre-conception care. Unlike typical supervised learning assumptions that expect similar characteristics across derivation and validation cohorts [26], our model maintained reliability even with divergent cohort characteristics, underscoring its robustness and generality. Remarkably, this highly accurate prediction model comprises a simplified set of five key variables, which greatly enhances its practicality in clinical use.

“Interpretability” and “Accuracy” are important in constructing a prediction model [27]. “Interpretability” is a passive property of a model, which indicates the degree to which a given model can be interpreted by a human observer. “Accuracy” refers to the degree of the model performance. Interpretability and accuracy generally have a trade-off relationship [28]. Logistic regression, a conventional method, has higher interpretability and lower accuracy, whereas machine learning has lower interpretability and higher accuracy [27].

It is widely recognized that individuals are reluctant to adopt tools that cannot be directly interpreted. Thus, there are concerns about the abundance of research that emphasizes new algorithms without focusing on user-friendliness, practical interpretability, or efficacy for end users [29, 30]. Furthermore, when implementing machine learning-based prediction models in clinical settings, the operational environment of programming languages used in model construction, such as R or Python, is necessary. In contrast, prediction models developed using logistic regression analysis can be replicated using any software capable of logarithmic calculations. Therefore, the results of the present study, obtained with reasonable accuracy using logistic regression analysis, are advantageous for clinical applications.

A recent meta-analysis on externally validated prediction models for HDP concluded that models based solely on maternal information exhibit equivalent predictive accuracy to those augmented with various biomarkers [25]. This suggests that maternal information plays a significantly crucial role in predicting the occurrence of HDP. Consequently, the high predictive accuracy achieved in our study using only clinical data aligns logically with these insights. This capability indicates the potential for high-accuracy predictions in low-resource countries and regions.

Our prediction model makes women aware of their own HDP risks before pregnancy and visualize that one can reduce risk through one’s efforts. Using the outputs, establishing weight management goals through bidirectional communication between healthcare providers and women is beneficial for the implementation of interpregnancy weight management. This interactive process allows the integration of each individual’s risk tolerance and commitment to lifestyle modifications, thereby enabling the setting of personalized and realistic targets. That is, bidirectional communication enables women to consider their past experiences and current environment to ascertain the degree of attainable or appropriate weight loss. Concurrently, healthcare providers can evaluate whether the goals are reasonable.

The primary strength of this study is its novel approach, which focuses on pragmatic applications for interpregnancy weight management, a part of IPC. Secondary, the prediction model, which sets weight management objectives based on annual BMI changes, ensures consistency and reliability irrespective of variations in pregnancy intervals. Additionally, the model design, utilizing only five optimal covariates, not only enhances user-friendliness but also reduces the risk of inaccuracy due to missing data in clinical use.

This study had some limitations. First, it was retrospective, and the annual BMI changes were not intervention-induced. Further studies are required to determine whether intervention-induced weight reduction reduces the risk of HDP in subsequent pregnancies. Accordingly, we plan to conduct a prospective study on weight management as a part of IPC using the prediction model developed in this study. Second, self-reported body weight was used to calculate the pre-pregnancy BMI. However, as almost all participants weighed themselves during antenatal checkups in the first trimester of pregnancy, the difference between their self-reported and actual weights was likely to be minimal. Third, all validations in this study were conducted using pregnancy data from within Japan. To guarantee its generalizability, geographic validation is necessary. Nevertheless, the prediction model developed herein utilizes solely clinical information, making it applicable across countries and regions regardless of their economic status.

The global rise in obesity is a growing issue, and Asia is no exception. In particular, Asians are more sensitive to the health risks of obesity compared to Western populations [31]. However, a major problem is that no standardized methods for weight management during IPC have been established [32]. Additionally, cultural, racial, and economic factors have been reported to influence postpartum weight loss programs [32, 33].

This study succeeded in developing a prediction model for setting weight management goals during IPC, using a cohort predominantly composed of Asians. We believe this represents a significant step towards more personalized medicine, in contrast to the uniform goal-setting recommended by conventional guidelines [4, 5]. Moving forward, we plan to conduct further research to establish a practical management system based on this prediction model in Asian population. Our future research will focus on how effective management can improve perinatal outcomes and, ultimately, extend healthy life expectancy.