Infant Mortality Prediction using Birth Certificate Data

Antonia Saravanou; Clemens Noelke; Dimitrios Gunopulos; Dolores Acevedo-Garcia; Nicholas Huntington

arxiv: 1907.08968 · v2 · pith:NHU3VPBCnew · submitted 2019-07-21 · 💻 cs.LG · stat.ML

Infant Mortality Prediction using Birth Certificate Data

Antonia Saravanou , Clemens Noelke , Nicholas Huntington , Dolores Acevedo-Garcia , Dimitrios Gunopulos This is my paper

Pith reviewed 2026-05-24 18:40 UTC · model grok-4.3

classification 💻 cs.LG stat.ML

keywords infant mortalitybirth certificate dataclassification modelsracial subgroupsfeature importancemachine learningepidemiology methods

0 comments

The pith

Classification models on birth certificate features outperform standard epidemiology methods for infant mortality prediction and differ across racial groups.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper tries to establish that classifiers trained on features from birth certificates can decide whether an infant will survive to its first birthday more accurately than the classification techniques routinely used in epidemiology research. It pays particular attention to training separate models for individual racial and ethnic groups and comparing them to one general model in order to understand which features matter most in each subset. A sympathetic reader would care because the infant mortality rate measures both newborn health and overall societal conditions, and long-standing racial differences in that rate suggest that group-specific predictions could support more precise responses. If the claim holds, public health efforts could shift toward data-driven identification of risks that vary by population segment.

Core claim

The authors establish that features extracted from birth certificates can be used to train classification models that decide infant survival and that these models outperform the standard classification methods employed by epidemiology researchers. They focus on exploring feature importance within population subsets defined by race by comparing models trained for individual races against a single general model.

What carries the argument

Race-specific versus general classification models trained on birth certificate features to predict infant survival

If this is right

Higher accuracy in predicting infant survival on new birth certificate data than current epidemiology approaches.
Different features emerge as important when models are trained on individual racial subsets rather than the full population.
Race-specific models reveal distinctions in performance and relevant predictors compared with a single general model.
More precise mapping of factors that contribute to observed racial differences in infant mortality rates.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Reliable models of this kind could support the design of health interventions aimed at specific racial groups rather than uniform national strategies.
The same birth certificate features might be tested for predicting related outcomes such as low birth weight or neonatal complications.
Public health agencies could incorporate subgroup modeling as a standard step when analyzing administrative records to reduce bias in risk assessment.

Load-bearing premise

Birth certificate records contain sufficiently complete, unbiased, and predictive features for reliable out-of-sample classification across racial subgroups.

What would settle it

A held-out collection of birth certificate records on which the race-specific and general classification models show no accuracy advantage over standard epidemiology methods.

Figures

Figures reproduced from arXiv: 1907.08968 by Antonia Saravanou, Clemens Noelke, Dimitrios Gunopulos, Dolores Acevedo-Garcia, Nicholas Huntington.

read the original 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.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Standard ML classifiers on birth-certificate data with a race-stratified comparison, but the abstract gives zero numbers, baselines, or validation details to support the performance claim.

read the letter

The paper trains ordinary supervised classifiers on birth-certificate features to predict whether an infant survives its first year, then compares models fit on race-specific subsets against a single pooled model. That comparison is the only concrete thing it adds; everything else is standard feature-importance work on an existing public dataset. The topic matters for public-health tracking of disparities, and restricting the analysis to race subgroups is a reasonable way to surface differences that a pooled model might mask. The authors also note that birth records are already collected at scale, so the data source itself is practical. Beyond that, the manuscript supplies no evidence. The abstract states that the methodology outperforms standard epidemiology methods yet reports no accuracy, AUC, precision-recall, calibration, or cross-validation scheme. Dataset size, class imbalance handling, and how missing fields were treated are also absent. Without those numbers it is impossible to judge whether the race-specific models actually improve anything or simply overfit smaller subgroups. The central modeling assumption—that birth-certificate fields are sufficiently complete and unbiased across racial groups—receives no discussion either. If the full paper contains the missing tables and checks, the work could be a modest incremental application worth a methods note in an epidemiology journal. On the current abstract alone it does not yet justify referee time.

Referee Report

1 major / 0 minor

Summary. The manuscript develops classification models trained on features from U.S. birth certificate records to predict whether an infant survives to age one. It emphasizes analysis of feature importance within racial and ethnic subgroups and compares the performance of race-specific models against a single general model, claiming that the proposed methodology outperforms standard classification approaches used in epidemiology research.

Significance. If the empirical claims are supported by rigorous, reproducible evaluation, the work could inform public-health efforts to reduce racial disparities in infant mortality by identifying actionable predictors from routinely collected administrative data. The subgroup modeling focus aligns with fairness considerations in health ML.

major comments (1)

[Abstract] Abstract: the central claim that 'our methodology outperforms standard classification methods used by epidemiology researchers' is stated without any performance metrics, baseline methods, dataset size, cross-validation procedure, class-imbalance handling, or subgroup-specific results. This absence is load-bearing for the paper's empirical contribution and precludes verification of the stated improvement.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive comment on the abstract. We address it point-by-point below.

read point-by-point responses

Referee: [Abstract] Abstract: the central claim that 'our methodology outperforms standard classification methods used by epidemiology researchers' is stated without any performance metrics, baseline methods, dataset size, cross-validation procedure, class-imbalance handling, or subgroup-specific results. This absence is load-bearing for the paper's empirical contribution and precludes verification of the stated improvement.

Authors: We agree that the abstract as written does not provide the quantitative details needed to substantiate the central claim. In the revised manuscript we will expand the abstract (within length limits) to include: dataset size and source years, the cross-validation procedure, the class-imbalance handling strategy, the specific baseline methods from the epidemiology literature that were compared, key performance metrics (AUC, F1, etc.), and a brief statement of the subgroup-specific findings. This revision will allow readers to assess the claimed improvement directly from the abstract. revision: yes

Circularity Check

0 steps flagged

No significant circularity

full rationale

This is an empirical ML classification paper with no equations, derivations, or parameter-fitting steps presented as predictions. The central claim is an out-of-sample performance comparison against baselines; it does not reduce any result to its own inputs by construction, self-definition, or self-citation chains. The work is self-contained as a standard data-driven evaluation.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The work rests on standard supervised-learning assumptions plus the domain premise that birth-certificate fields are adequate for mortality prediction. No new entities are introduced.

free parameters (1)

model hyperparameters and feature selection thresholds
Typical ML training choices; values not reported.

axioms (1)

domain assumption Birth-certificate features are sufficiently complete and unbiased for mortality classification
Invoked implicitly by training classifiers on these records.

pith-pipeline@v0.9.0 · 5680 in / 1000 out tokens · 19086 ms · 2026-05-24T18:40:19.240875+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

18 extracted references · 18 canonical work pages

[1]

Jason Abrevaya. 2002. The effects of demographics and maternal behavior on the distribution of birth outcomes. In Economic applications of quantile regression

work page 2002
[2]

Soobader, and Lisa Berkman

Dolores Acevedo-Garcia, M. Soobader, and Lisa Berkman. 2007. Low birthweight among U.S. Hispanic/Latino subgroups: The effect of maternal foreign-born status and education. Social Science & Medicine (2007)

work page 2007
[3]

Dolores Acevedo-Garcia, Mah-J Soobader, and Lisa F. Berkman. 2005. The Differ- ential Effect of Foreign-Born Status on Low Birth Weight by Race/Ethnicity and Education. Pediatrics (2005)

work page 2005
[4]

Douglas Almond, Kenneth Y Chay, and David S Lee. 2005. The costs of low birth weight. The Quarterly Journal of Economics (2005)

work page 2005
[5]

Antonia Saravanou, Clemens Noelke, Nick Huntington, Dolores Acevedo-Garcia and Dimitrios Gunopulos. 2019. Predicting Infant Mortality at the Time of Birth. Population Association Annual Meeting, Austin, TX. (2019)

work page 2019
[6]

Brian M Casey, Donald D McIntire, and Kenneth J Leveno. 2001. The continuing value of the Apgar score for the assessment of newborn infants. New England Journal of Medicine (2001)

work page 2001
[7]

Tianqi Chen and Carlos Guestrin. 2016. Xgboost: A scalable tree boosting system. In Proceedings of the 22nd SIGKDD 2016 . ACM

work page 2016
[8]

Jamie Mihoko Doyle, Samuel Echevarria, and W Parker Frisbie. 2003. Race/ ethnicity, Apgar and infant mortality. Population Research and Policy Review

work page 2003
[9]

Thomas Hegyi, Tracy Carbone, Mujahid Anwar, Barbara Ostfeld, Mark Hiatt, Anne Koons, Jennifer Pinto-Martin, and Nigel Paneth. 1998. The Apgar score and its components in the preterm infant. Pediatrics (1998)

work page 1998
[10]

Nancy A Hessol, Elena Fuentes-Afflick, and Peter Bacchetti. 1998. Risk of low birth weight infants among black and white parents. Obstetrics & Gynecology

work page 1998
[11]

Robert A Hummer, Monique Biegler, Peter B De Turk, Douglas Forbes, W Parker Frisbie, Ying Hong, and Starling G Pullum. 1999. Race/ethnicity, nativity, and infant mortality in the United States. Social Forces (1999)

work page 1999
[12]

John and Pat Langley

George H. John and Pat Langley. 1995. Estimating Continuous Distributions in Bayesian Classifiers. In Proceedings of the Eleventh Conference on Uncertainty in Artificial Intelligence (UAI’95)

work page 1995
[13]

Kochanek, Sherry L

Kenneth D. Kochanek, Sherry L. Murphy, Jiaquan Xu, and Betzaida Tejada-Vera

work page
[14]

National Vital Statistics Reports (2006)

Deaths: Final Data for 2014. National Vital Statistics Reports (2006)

work page 2014
[15]

Osypuk and Dolores Acevedo-Garcia

Theresa L. Osypuk and Dolores Acevedo-Garcia. 2008. Are Racial Disparities in Preterm Birth Larger in Hypersegregated Areas? American Journal of Epidemiol- ogy

work page 2008
[16]

Daniel Powers, Frisbie Parker, and other. 2006. Race/Ethnic differences and age-variation in the effects of birth outcomes on infant mortality in the US. Demographic Research (2006)

work page 2006
[17]

Bernhard Schölkopf, Robert C Williamson, Alex J Smola, John Shawe-Taylor, and John C Platt. 2000. Support vector method for novelty detection. In Advances in neural information processing systems . 582–588

work page 2000
[18]

Allen J Wilcox and Rolv Skjaerven. 1992. Birth weight and perinatal mortality: the effect of gestational age. American Journal of Public Health (1992)

work page 1992

[1] [1]

Jason Abrevaya. 2002. The effects of demographics and maternal behavior on the distribution of birth outcomes. In Economic applications of quantile regression

work page 2002

[2] [2]

Soobader, and Lisa Berkman

Dolores Acevedo-Garcia, M. Soobader, and Lisa Berkman. 2007. Low birthweight among U.S. Hispanic/Latino subgroups: The effect of maternal foreign-born status and education. Social Science & Medicine (2007)

work page 2007

[3] [3]

Dolores Acevedo-Garcia, Mah-J Soobader, and Lisa F. Berkman. 2005. The Differ- ential Effect of Foreign-Born Status on Low Birth Weight by Race/Ethnicity and Education. Pediatrics (2005)

work page 2005

[4] [4]

Douglas Almond, Kenneth Y Chay, and David S Lee. 2005. The costs of low birth weight. The Quarterly Journal of Economics (2005)

work page 2005

[5] [5]

Antonia Saravanou, Clemens Noelke, Nick Huntington, Dolores Acevedo-Garcia and Dimitrios Gunopulos. 2019. Predicting Infant Mortality at the Time of Birth. Population Association Annual Meeting, Austin, TX. (2019)

work page 2019

[6] [6]

Brian M Casey, Donald D McIntire, and Kenneth J Leveno. 2001. The continuing value of the Apgar score for the assessment of newborn infants. New England Journal of Medicine (2001)

work page 2001

[7] [7]

Tianqi Chen and Carlos Guestrin. 2016. Xgboost: A scalable tree boosting system. In Proceedings of the 22nd SIGKDD 2016 . ACM

work page 2016

[8] [8]

Jamie Mihoko Doyle, Samuel Echevarria, and W Parker Frisbie. 2003. Race/ ethnicity, Apgar and infant mortality. Population Research and Policy Review

work page 2003

[9] [9]

Thomas Hegyi, Tracy Carbone, Mujahid Anwar, Barbara Ostfeld, Mark Hiatt, Anne Koons, Jennifer Pinto-Martin, and Nigel Paneth. 1998. The Apgar score and its components in the preterm infant. Pediatrics (1998)

work page 1998

[10] [10]

Nancy A Hessol, Elena Fuentes-Afflick, and Peter Bacchetti. 1998. Risk of low birth weight infants among black and white parents. Obstetrics & Gynecology

work page 1998

[11] [11]

Robert A Hummer, Monique Biegler, Peter B De Turk, Douglas Forbes, W Parker Frisbie, Ying Hong, and Starling G Pullum. 1999. Race/ethnicity, nativity, and infant mortality in the United States. Social Forces (1999)

work page 1999

[12] [12]

John and Pat Langley

George H. John and Pat Langley. 1995. Estimating Continuous Distributions in Bayesian Classifiers. In Proceedings of the Eleventh Conference on Uncertainty in Artificial Intelligence (UAI’95)

work page 1995

[13] [13]

Kochanek, Sherry L

Kenneth D. Kochanek, Sherry L. Murphy, Jiaquan Xu, and Betzaida Tejada-Vera

work page

[14] [14]

National Vital Statistics Reports (2006)

Deaths: Final Data for 2014. National Vital Statistics Reports (2006)

work page 2014

[15] [15]

Osypuk and Dolores Acevedo-Garcia

Theresa L. Osypuk and Dolores Acevedo-Garcia. 2008. Are Racial Disparities in Preterm Birth Larger in Hypersegregated Areas? American Journal of Epidemiol- ogy

work page 2008

[16] [16]

Daniel Powers, Frisbie Parker, and other. 2006. Race/Ethnic differences and age-variation in the effects of birth outcomes on infant mortality in the US. Demographic Research (2006)

work page 2006

[17] [17]

Bernhard Schölkopf, Robert C Williamson, Alex J Smola, John Shawe-Taylor, and John C Platt. 2000. Support vector method for novelty detection. In Advances in neural information processing systems . 582–588

work page 2000

[18] [18]

Allen J Wilcox and Rolv Skjaerven. 1992. Birth weight and perinatal mortality: the effect of gestational age. American Journal of Public Health (1992)

work page 1992