MOTHER_CARE2: Clinical Maternal and Neonatal Dataset for High-Risk Pregnancy Assessment in Bangladesh

Description

This dataset contains 111 de-identified maternal and neonatal clinical records retrospectively compiled from routine antenatal, intrapartum, and postnatal hospital documentation at a healthcare facility in Bangladesh. It covers maternal demographics, obstetric history, clinical measurements (blood pressure, hemoglobin, gestational age), delivery information, neonatal outcomes, five derived clinical risk indicators, and a composite binary high-risk pregnancy label. Research hypothesis: routinely collected maternal and neonatal clinical measurements can support pregnancy-risk stratification in resource-limited healthcare settings, where advanced diagnostic tools are often unavailable. Findings: of 111 records, 89 (80.2%) are labeled high-risk and 22 (19.8%) low-risk. The composite high_risk label is strongly associated with five derived indicators (age_risk, anemia_flag, hypertension_flag, preterm_risk, low_birth_weight): 88/89 high-risk cases show at least one positive flag, and all 22 low-risk cases show none. Because low_birth_weight and other neonatal variables are only known after delivery, high_risk should be interpreted as a retrospective composite classification rather than a strictly antenatal (pre-delivery) prediction outcome. Interpretation and use: researchers building antenatal prediction models should define a clear prediction time point and use only pre-delivery variables as predictors. The five derived risk-flag variables should generally be excluded when modeling high_risk, since including them causes target leakage rather than genuine predictive learning; they remain useful for descriptive or label-composition analysis. Both original (_raw) and cleaned (_filled) versions of gravida and parity are provided to distinguish source values from processed values. Direct identifiers were removed before publication; patient_id values are study-specific and non-identifying. The accompanying README details label derivation, missing-data patterns, known data-quality anomalies, and target-leakage guidance. The Data Dictionary lists variable descriptions, data types, units, value ranges, and missing-value counts. Intended for research, education, and epidemiological analysis only; not validated for direct clinical decision-making without independent clinical review.

Steps to reproduce

1. Data source: Maternal and neonatal records were retrospectively identified from routine antenatal, intrapartum, and postnatal hospital documentation maintained at a healthcare facility in Bangladesh. 2. Data extraction: Eligible records (n = 111) were manually reviewed and transcribed from paper-based hospital records into a structured digital dataset, covering maternal demographics, obstetric history, clinical measurements (blood pressure, hemoglobin, gestational age), delivery information, and neonatal outcomes. 3. De-identification: Direct personal identifiers were removed during transcription; original hospital identifiers were replaced with study-specific patient_id values. 4. Data preparation: Original recorded values were preserved in *_raw variables. Cleaned (*_filled) versions of gravida and parity were generated to improve completeness. Five binary risk indicators (age_risk, anemia_flag, hypertension_flag, preterm_risk, low_birth_weight) and the composite outcome (high_risk) were derived from the extracted clinical values. 5. Data quality and ethics: Internal consistency checks and missing-value assessment were performed using Microsoft Excel and Python (pandas); known issues were retained and documented rather than silently corrected. Retrospective use of de-identified records was conducted with administrative permission from the hospital's Civil Surgeon. 6. Reproducing analyses: Import `clinical_maternal_dataset.csv` into a statistical/ML environment (Python, R, SPSS, Stata, SAS, or Excel). Refer to `readme.md` and `data_dictionary.pdf` for variable definitions, missing-value handling, and data-quality notes. 7. Modeling guidance and reporting: When predicting high_risk, exclude the five derived risk flags to avoid target leakage; for antenatal-only prediction, also exclude post-delivery variables (delivery_mode, baby_sex, birth_weight_grams, apgar_1min, apgar_5min). Document all preprocessing steps, inclusion/exclusion criteria, and evaluation procedures to ensure reproducibility.

Institutions

• Daffodil International University

  Dhaka Division, Dhaka

Categories

Licence