Identifying and validating the earliest acceptable prediction window for 14‑week weight loss outcome: a nationwide lifestyle intervention
4 affiliations
- 1National Center for Chronic and Noncommunicable Disease Control and Prevention, Chinese Center for Disease Control and Prevention, Beijing, China
- 2Chinese Center for Disease Control and Prevention, Beijing, China
- 3Inner Mongolia Medical University, Hohhot, China
- 4Chinese Academy of Sciences, Beijing, China
Abstract
Background: Early identification of individuals at high risk of weight loss failure may support risk stratification and timely intervention adjustment. However, evidence remains limited regarding the earliest time window at which weight loss failure can be predicted with acceptable performance in real-world walking interventions.
Methods: Data from participants in the 2025 10,000 Steps Challenge were randomly split into training and test sets (7:3). Using landmark datasets from weeks 2–7, we identified the earliest prediction window achieving prespecified acceptable discrimination (AUC > 0.75). We then compared five prediction models and evaluated discrimination, calibration, and the incremental value of baseline covariates.
Results: A total of 12,773 participants were included, of whom 23.14% achieved weight loss ≥ 5% at 14 weeks. In the landmark analysis, the area under the curve increased from 0.646 at week 2 to 0.832 at week 7. Week 4 was the earliest landmark window achieving the prespecified acceptable discrimination (AUC = 0.752), after which predictive performance continued to improve with progressively smaller gains. All five models developed using week 4 data showed good discrimination and calibration in test set, with Brier scores ranging from 0.150 to 0.156, calibration intercepts close to 0, calibration slopes close to 1, and no statistically significant differences in AUC between models. Adding baseline covariates to percentage weight change at week 4 increased the AUC by only 0.006. Sensitivity analyses yielded consistent results.
Conclusions: Week 4 was the earliest acceptable time window for predicting the risk of weight loss failure at 14 weeks, serving as a practical checkpoint to identify non‑responders and inform targeted intervention while sufficient time remains.
Supplementary Information: The online version contains supplementary material available at https://doi.org/10.1186/s13690-026-02037-4.
Related
- Developing and validating a hand hygiene skills intervention for Primary Health Care nurses
- Paths to Participation with Black adults: Identifying Recruitment Preferences and Interest in a Digital Nutrition Intervention to Reduce Hypertension
- Percutaneous Coronary Intervention in Dialysis Patients: Insights from a Nationwide Registry Analysis
- Identifying the active ingredients of a behavioral activation-based digital single-session intervention
- Prognostic Impact of COVID-19 on Patients With Acute Coronary Syndrome Undergoing Emergent Percutaneous Coronary Intervention ― Insights From the J-PCI Nationwide Registry ―
BibTeX
@article{li2026identifying,
title = {{Identifying and validating the earliest acceptable prediction window for 14‑week weight loss outcome: a nationwide lifestyle intervention}},
author = {Maoshu Li and Rui Zhang and Jiayi Xu and Yitong Sun and Xiaoning Cai and Fang Liu and Wei Sun and Jing Wu},
journal = {Archives of Public Health},
year = {2026},
month = aug,
doi = {10.1186/s13690-026-02037-4},
url = {https://doi.org/10.1186/s13690-026-02037-4},
note = {CC BY-NC-ND 4.0},
}
RIS
TY - JOUR
TI - Identifying and validating the earliest acceptable prediction window for 14‑week weight loss outcome: a nationwide lifestyle intervention
AU - Maoshu Li
AU - Rui Zhang
AU - Jiayi Xu
AU - Yitong Sun
AU - Xiaoning Cai
AU - Fang Liu
AU - Wei Sun
AU - Jing Wu
PY - 2026
DA - 2026/08/10
JO - Archives of Public Health
DO - 10.1186/s13690-026-02037-4
UR - https://doi.org/10.1186/s13690-026-02037-4
AB - Background: Early identification of individuals at high risk of weight loss failure may support risk stratification and timely intervention adjustment. However, evidence remains limited regarding the earliest time window at which weight loss failure can be predicted with acceptable performance in real-world walking interventions. Methods: Data from participants in the 2025 10,000 Steps Challenge were randomly split into training and test sets (7:3). Using landmark datasets from weeks 2–7, we identified the earliest prediction window achieving prespecified acceptable discrimination (AUC > 0.75). We then compared five prediction models and evaluated discrimination, calibration, and the incremental value of baseline covariates. Results: A total of 12,773 participants were included, of whom 23.14% achieved weight loss ≥ 5% at 14 weeks. In the landmark analysis, the area under the curve increased from 0.646 at week 2 to 0.832 at week 7. Week 4 was the earliest landmark window achieving the prespecified acceptable discrimination (AUC = 0.752), after which predictive performance continued to improve with progressively smaller gains. All five models developed using week 4 data showed good discrimination and calibration in test set, with Brier scores ranging from 0.150 to 0.156, calibration intercepts close to 0, calibration slopes close to 1, and no statistically significant differences in AUC between models. Adding baseline covariates to percentage weight change at week 4 increased the AUC by only 0.006. Sensitivity analyses yielded consistent results. Conclusions: Week 4 was the earliest acceptable time window for predicting the risk of weight loss failure at 14 weeks, serving as a practical checkpoint to identify non‑responders and inform targeted intervention while sufficient time remains. Supplementary Information: The online version contains supplementary material available at https://doi.org/10.1186/s13690-026-02037-4.
SN - 0778-7367
SN - 2049-3258
ER -