MOTHER_CARE1: Maternal Health Dataset for Antenatal High-Risk Pregnancy Prediction in Bangladesh

Description

This dataset contains 609 de-identified maternal and neonatal delivery records retrospectively collected from the labor ward register of a government hospital in Bangladesh. It includes maternal demographic characteristics, obstetric history, antenatal care information, admission reason, gestational age, delivery details, neonatal outcomes, and a binary high-risk pregnancy label generated using predefined clinical criteria based on the available maternal clinical information. The dataset was compiled to support research investigating whether routinely available maternal clinical characteristics—including maternal age, parity, gravida, antenatal care attendance, gestational age, and admission reason—can be used to identify high-risk pregnancies in low-resource healthcare settings where advanced diagnostic resources may be limited. Of the 609 records, 379 (62.2%) were classified as high-risk according to the predefined labeling criteria. All data were manually transcribed from routine hospital records without modification of the original recorded values. Personally identifiable information was removed prior to dataset preparation to ensure patient anonymity. The high-risk pregnancy label was generated by the research team using predefined clinical criteria rather than directly recorded in the source register. Users should review the accompanying README file for details of the labeling methodology, data quality considerations, and guidance on avoiding potential data leakage when developing predictive models. A companion Data_Dictionary file provides detailed information on each variable, including data type, description, units, allowed values, and missing values. This dataset is intended to support research in maternal health, obstetric epidemiology, reproductive medicine, public health, and machine learning applications.

Steps to reproduce

1. Data source: Maternal and neonatal records were retrospectively collected from the labor ward register of a government hospital in Bangladesh. The register contained maternal demographic characteristics, obstetric history, antenatal care information, admission reason, gestational age, delivery information, and neonatal outcomes. 2. Data extraction: A total of 609 eligible records were manually transcribed from the paper-based labor ward register into a structured digital dataset. Original recorded values were preserved during transcription without intentional modification. 3. De-identification: All direct personal identifiers were removed before dataset preparation. Study-specific patient_id values were assigned to preserve participant confidentiality. 4. Label derivation: The binary high_risk label was derived during dataset preparation using predefined clinical criteria based on the available maternal clinical information rather than being directly recorded in the source register. Of the 609 records, 379 (62.2%) were classified as high-risk. 5. Data preparation: Variable names were standardized, categorical values were harmonized where appropriate, and missing values were retained as documented in the original register. No undocumented clinical information was added to the published dataset. 6. Documentation: Refer to the accompanying `Data_Dictionary.pdf` for detailed descriptions of all variables, including data types, measurement units, allowable values, and missing-value information. Consult the accompanying `readme.md` for details of the label derivation methodology, data-quality considerations, and guidance on avoiding potential target leakage. 7. Software: Import `maternal_dataset.csv` into a statistical or data analysis environment such as Python (pandas), R, SPSS, Stata, SAS, or Microsoft Excel. 8. Reproducibility: Preserve missing values as documented in the original register, clearly report all preprocessing procedures, inclusion/exclusion criteria, feature engineering, and evaluation methods, and follow the recommendations in the accompanying readme.md to facilitate transparent and reproducible analyses.

Institutions

• Daffodil International University

  Dhaka Division, Dhaka

Categories

Licence