[Paper Review] Infant Mortality Prediction using Birth Certificate Data

Antonia Saravanou, Clemens Noelke+3 more|arXiv (Cornell University)|Jul 21, 2019

Data-Driven Disease Surveillance13 references4 citations

TL;DR

This paper proposes a machine learning approach to predict infant mortality using birth certificate data, leveraging socio-economic and clinical features to improve prediction accuracy. By training race-specific models with XGBoost, it achieves AUC scores up to 0.84, significantly outperforming standard epidemiological methods and highlighting the critical role of maternal education, race, and birth weight in mortality prediction.

ABSTRACT

The Infant Mortality Rate (IMR) is the number of infants per 1000 that do not survive until their first birthday. It is an important metric providing information about infant health but it also measures the society's general health status. Despite the high level of prosperity in the U.S.A., the country's IMR is higher than that of many other developed countries. Additionally, the U.S.A. exhibits persistent inequalities in the IMR across different racial and ethnic groups. In this paper, we study the infant mortality prediction using features extracted from birth certificates. We are interested in training classification models to decide whether an infant will survive or not. We focus on exploring and understanding the importance of features in subsets of the population; we compare models trained for individual races to general models. Our evaluation shows that our methodology outperforms standard classification methods used by epidemiology researchers.

Motivation & Objective

• To improve infant mortality prediction accuracy using comprehensive birth certificate data, including socio-economic and clinical factors.
• To investigate whether modeling by race enhances prediction performance compared to general population models.
• To identify the most influential features in predicting infant mortality using interpretable machine learning techniques.
• To evaluate model performance across different class imbalance scenarios (1:1, 1:10, 1:145) to assess robustness.
• To provide actionable insights for public health policy by identifying high-risk subpopulations and key risk factors.

Proposed method

• Utilizes a dataset of 12 million U.S. birth certificates (2000–2002) with 128 features, including maternal, infant, and obstetric characteristics.
• Applies XGBoost for binary classification of infant survival (yes/no), with hyperparameter tuning to optimize performance.
• Trains separate models for each racial group (White, Black, American Indian) to assess the impact of population-specific modeling.
• Evaluates model performance using AUC and examines feature importance via information gain in gradient-boosted decision trees.
• Tests models under varying class imbalance distributions (1:1, 1:10, 1:145) to simulate real-world data skew and assess generalization.
• Uses external socio-economic features such as maternal education, marital status, and place of residence to enhance predictive power.

Experimental results

Research questions

• RQ1 Can machine learning models trained on birth certificate data outperform traditional epidemiological methods in predicting infant mortality?
• RQ2 Does training race-specific models improve prediction accuracy compared to a single general model?
• RQ3 Which socio-economic and clinical features are most predictive of infant mortality according to the model?
• RQ4 How does class imbalance (1:145) affect model performance, and what strategies mitigate its impact?
• RQ5 To what extent do maternal characteristics such as education, age, and race contribute to predicting infant mortality?

Key findings

• The XGBoost model achieves an AUC of 0.84 on the White population and 0.82 on the general population (1:145 imbalance), significantly outperforming standard epidemiological methods.
• Race-specific models improve prediction accuracy, with the Black population achieving AUC 0.83 and American Indian 0.75, indicating that subgroup modeling enhances performance.
• The top two most important features are birth weight (in grams) and Apgar score, followed by maternal education, age, and race, highlighting the role of socio-economic factors.
• Nine of the top 20 features are maternal-related, including marital status, place of residence, and number of prenatal visits, underscoring the importance of maternal context.
• The model maintains strong performance across different imbalance levels, with AUC decreasing only slightly from 0.84 (1:1) to 0.82 (1:145), indicating robustness.
• Feature importance analysis reveals that clinical factors (e.g., gestation weeks, congenital anomalies) and socio-economic indicators (e.g., maternal education, smoking) are critical for accurate prediction.