Human and machine learning of prognostic prediction for prelabor rupture of membranes and the time of delivery: a nationwide development, validation, and deployment using medical history
Sufriyana, H.; Wu, Y.-W.; Su, E. C.-Y.
Show abstract
ImportancePrognostic predictions of prelabor rupture of membranes lack proper sample sizes and external validation. ObjectiveTo develop, validate, and deploy statistical and/or machine learning prediction models using medical histories for prelabor rupture of membranes and the time of delivery. DesignA retrospective cohort design within 2-year period (2015 to 2016) of a single-payer, government-owned health insurance database covering 75.8% individuals in a country SettingNationwide healthcare providers (n=22,024) at primary, secondary, and tertiary levels Participants12-to-55-year-old women that visit healthcare providers using the insurance from [~]1% random sample of insurance holders stratified by healthcare provider and category of family: (1) never visit; (2) visit only primary care; and (3) visit all levels of care PredictorsMedical histories of diagnosis and procedure (International Classification of Disease version 10) before the latest visit of outcome within the database period Main Outcomes and MeasuresPrelabor rupture of membranes prognostication (area under curve, with sensitivity, specificity, and likelihood ratio), the time of delivery estimation (root mean square error), and inference time (minutes), with 95% confidence interval ResultsWe selected 219,272 women aged 33 {+/-} 12 years. The best prognostication achieved area under curve 0.73 (0.72 to 0.75), sensitivity 0.494 (0.489 to 0.500), specificity 0.816 (0.814 to 0.818), and likelihood ratio being positive 2.68 (2.63 to 2.75) and negative 0.62 (0.61 to 0.63). This outperformed models from previous studies according to area under curve of an external validation set, including one using a biomarker (area under curve 0.641; sensitivity 0.419; sensitivity 0.863; positive likelihood ratio 3.06; negative likelihood ratio 0.67; n=1177). Meanwhile, the best estimation achieved {+/-} 2.2 and 2.6 weeks respectively for predicted events and non-events. Our web application only took 5.14 minutes (5.11 to 5.18) per prediction. Conclusions and RelevancePrelabor rupture of membranes and the time of delivery were predicted by medical histories; but, an impact study is required before clinical application. Key PointsO_ST_ABSQuestionC_ST_ABSCan we use medical histories of diagnosis and procedure in electronic health records to predict prelabor rupture of membranes and the time of delivery before the day in nationwide insured women? FindingsIn this prognostic study applying retrospective cohort paradigm, a significant predictive performance was achieved and validated. The area under receiver operating characteristics curve was 0.73 with the estimation errors of {+/-} 2.2 and 2.6 weeks for the time of delivery. MeaningPreliminary prediction can be conducted in a wide population of insured women to predict prelabor rupture of membranes and estimate the time of delivery.
Matching journals
The top 2 journals account for 50% of the predicted probability mass.
Similar papers in this journal
- Improving Pre-eclampsia Risk Prediction by Modeling Individualized Pregnancy Trajectories Derived from Routinely Collected Electronic Medical Record Data 96%
- Zero-shot Interpretable Phenotyping of Postpartum Hemorrhage Using Large Language Models 94%
- Continuous-Time and Dynamic Suicide Attempt Risk Prediction with Neural Ordinary Differential Equations 94%
Similar papers in this journal
- Widely accessible prognostication using medical history for fetal growth restriction and small for gestational age in nationwide insured women 97%
- Large Language Models Improve the Identification of Emergency Department Visits for Symptomatic Kidney Stones 93%
- Machine learning approach to dynamic risk modeling of mortality in COVID-19: a UK Biobank study 93%
Similar papers in this journal
- Deep representation learning for clustering longitudinal survival data from electronic health records 94%
- Artificial intelligence-based histopathology image analysis identifies a novel subset of endometrial cancers with distinct genomic features and unfavourable outcome 92%
- Deep transfer learning for reducing health care disparities arising from biomedical data inequality 92%
Similar papers in this journal
- Large-language-model-based 10-year risk prediction of cardiovascular disease: insight from the UK biobank data 93%
- Benchmarking transformer-based models for medical record deidentification: A single centre, multi-specialty evaluation 92%
- A Chinese host genetic study discovered type I interferons and causality of cholesterol levels and WBC counts on COVID-19 severity 91%
Similar papers in this journal
- APPLAUSE: Automatic Prediction of PLAcental health via U-net Segmentation and statistical Evaluation 93%
- Spatial Transcriptomics Expression Prediction from Histopathology Based on Cross-Modal Mask Reconstruction and Contrastive Learning 91%
- Encrypted federated learning for secure decentralized collaboration in cancer image analysis 89%
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.