SushantGautam/nvss-birth-records-usa-2016-2020

🍼 Preterm Birth Prediction Dataset (NVSS 2016–2020)

This dataset is a preprocessed and curated version of the U.S. National Vital Statistics System (NVSS) Natality Birth Data (2016–2020), specifically tailored for use in machine learning tasks related to predicting preterm births.

📚 Background

This dataset preparation and preprocessing pipeline is from the master's thesis:

Predicting Preterm Birth with Machine Learning Methods
by Zélie Dresse (May 2022)
Master’s Programme in Data Analytics and Business Economics, Lund University
📄 Read the full thesis here
💻 Source code on GitHub

The original data files are publicly available from the National Bureau of Economic Research (NBER). This dataset uses the CSV files from 2016–2020, with some additional information parsed from the text files for 2019 and 2020.

🧪 Objective

The dataset is used to build models that can predict spontaneous preterm birth, defined as birth before 37 weeks of gestation. It contains structured features related to maternal demographics, medical history, prenatal care, behavioral risk factors (e.g., smoking), and more.

📊 Summary of Variables Used in the Dataset

This dataset is derived from the U.S. National Vital Statistics System (2016–2020). Below is a detailed description of each variable used in our analysis and modeling pipeline.

Variables

Variable	Description
MAGER	Mother's age in completed years at the time of delivery.
RESTATUS	Maternal residence status: resident, intrastate nonresident, interstate nonresident, foreign resident.
MRACE6	Mother's race recoded into 6 categories: White, Black, American Indian/Alaska Native (AIAN), Asian, Native Hawaiian or Pacific Islander (NHOPI), More than one race.
MHISP_R	Mother's Hispanic origin (Non-Hispanic, Mexican, Puerto Rican, Cuban, Central/South American, Other Hispanic).
MEDUC	Mother's highest education level completed.
PRIORLIVE	Number of prior live births the mother has had.
PRIORDEAD	Number of previous children born alive who later died.
PRIORTERM	Number of other pregnancy outcomes (e.g., stillbirths, abortions).
ILLB_R11	Month prenatal care began, recoded into 11 categories.
PRECARE	Month prenatal care began (1 = first month, 2 = second month, etc.).
WIC	Whether the mother received WIC (Women, Infants, and Children) nutritional assistance during pregnancy.
CIG_0	Average number of cigarettes smoked per day before pregnancy.
CIG_1	Average number of cigarettes smoked per day during the 1st trimester.
CIG_2	Average number of cigarettes smoked per day during the 2nd trimester.
CIG_REC	Whether the mother smoked during pregnancy (derived from CIG_1 and CIG_2).
BMI	Mother's Body Mass Index, calculated from height and pre-pregnancy weight.
RF_PDIAB	Pre-pregnancy diabetes (Yes/No).
RF_GDIAB	Gestational diabetes (Yes/No).
RF_PHYPE	Pre-pregnancy hypertension (Yes/No).
RF_GHYPE	Gestational hypertension (Yes/No).
RF_EHYPE	Eclampsia (Yes/No).
RF_PPTERM	Previous preterm birth (Yes/No).
RF_INFTR	Infertility treatment used (Yes/No).
RF_FEDRG	Fertility enhancing drugs used (Yes/No).
RF_ARTEC	Assisted reproductive technology used (Yes/No).
RF_CESAR	Previous cesarean delivery (Yes/No).
RF_CESARN	Number of previous cesarean deliveries.
IP_GON	Gonorrhea present during pregnancy (Yes/No).
IP_SYPH	Syphilis present during pregnancy (Yes/No).
IP_CHLAM	Chlamydia present during pregnancy (Yes/No).
IP_HEPB	Hepatitis B present during pregnancy (Yes/No).
IP_HEPC	Hepatitis C present during pregnancy (Yes/No).
LD_INDL	Labor was induced (Yes/No).
PAY_REC	Source of payment for delivery (e.g., Medicaid, private insurance, self-pay, other).
OEGest_R10	Clinical estimate of gestation, grouped into weeks.
M_Ht_In	Mother's height in inches.
PWgt_R	Mother's pre-pregnancy weight in pounds.
ME_ROUT	Whether delivery was vaginal (routine) or cesarean.
ME_TRIAL	Whether trial of labor was attempted before cesarean.
FEMALE	Infant sex (1 = Female, 0 = Male).
father_info	Whether father's age or paternity was acknowledged (Boolean).
US_NATIVE	Whether mother was born in the United States (Boolean).
birth_interval	Time since previous birth (e.g., 0–3 months, 4–11 months, ..., or Not applicable for first births).
young_mother	Whether the mother is under 20 years of age (Boolean).
old_mother	Whether the mother is aged 35 years or older (Boolean).
low_educ	Whether the mother has low education (≤ high school graduate).
high_educ	Whether the mother has high education (≥ master's degree).
BMI_low	Whether mother is underweight based on BMI (<18.5).
BMI_overweight	Whether mother is overweight based on BMI (25–29.9).
BMI_obese	Whether mother is obese based on BMI (≥30).
first_preg	Whether this is the mother’s first pregnancy (Boolean).
first_birth	Whether this is the mother’s first live birth (Boolean).
DMAR_U	Marital status: Unknown if paternity acknowledged.
DMAR_Y	Marital status: Yes if paternity acknowledged.
MAR_P_U	Mother's marital status unknown.
MAR_P_Y	Mother was married at time of delivery.
no_precare	No prenatal care received during pregnancy.
precare_1trim	Prenatal care started in the first trimester.
precare_2trim	Prenatal care started in the second trimester.
preterm	Target variable indicating preterm birth (<37 weeks) or full term (>=37 weeks).

🛠️ Usage Example

To prepare the dataset for machine learning, we separate the features (X) and the target (y). Below is a minimal example of how to load and preprocess the data correctly:

from datasets import load_dataset
dataset = load_dataset("SushantGautam/nvss-birth-records-usa-2016-2020", split="train")
df = dataset.to_pandas()

# Columns to drop: either because they are target labels, cause leakage, or are redundant
drop_cols = [
    "preterm",      # Target variable
    "OEGest_R10",   # Gestational age – defines preterm birth (target leakage)
    "ILLB_R11",     # Redundant with derived prenatal care features
    "PWgt_R",       # Used in BMI calculation
    "BMI",          # Removed if using BMI categories (e.g., BMI_low, BMI_obese)
    "ME_ROUT",      # Delivery method – not known at prediction time
    "ME_TRIAL",     # Trial of labor – not available before delivery
    "LD_INDL",      # Labor induction – happens during delivery
    "PRECARE"       # Raw month value – dropped if using derived binary indicators
]
# Define feature matrix X and target vector y
X = df.drop(columns=drop_cols)
y = df["preterm"]

This ensures only pre-delivery information is used to predict preterm birth, avoiding data leakage.

💡 Citation

If you use this dataset or parts of the processing pipeline, please cite the original thesis:

@mastersthesis{dresse2022preterm,
  title={Predicting Preterm Birth with Machine Learning Methods},
  author={Zélie Dresse},
  year={2022},
  school={Lund University},
  url={https://github.com/zeliedresse/master-thesis}
}

🔗 Licensing

This dataset is derived from public domain sources (NVSS via NBER). The processing pipeline is based on open-source code released under the MIT License by Zélie Dresse.