INTRODUCTION
Reducing premature births has been an important public health agenda in
the United States at least since the 1980s. Prematurity of infants is
one of the critical causes of fetal death and medical complications in
infants, e.g. , low oxygen level at birth and nervous system
problems (1). Russell and colleagues estimated that the costs of
hospitalization for infants with low birth weight (LBW) or preterm birth
(PTB) would be six times higher than that of other infants without PTB
or LBW (2). Moreover, the physical and mental health of newborn babies
reproduce socioeconomic disadvantages throughout their lifetime (3).
As the premature birth rate has continuously been high despite advanced
medical technologies and prenatal care,(4,5) several Prenatal Care Home
Visiting (PCHV) programs have been implemented.
A PCHV generally refers to a
service in which nurses or trained counselors visit families with a
pregnant woman to provide advice for antepartum care treatments or
counseling services (6,7). The services mainly provide monitoring to
keep mothers receiving appropriate care services and counseling focused
on parenting curricula and reducing behavioral risk factors, such as
smoking, alcohol/drug use, and mental health problems (7,8).
Furthermore, receiving prenatal care on a regular basis is considered as
a primary measure of preventing premature births because it becomes a
starting point to detect risks at an earlier stage of gestation.(7) PCHV
programs specifically target to improve the accessibility to prenatal
care for mothers in underserved communities (6,7). Rich evidence finds
that infants from mothers in a socially and economically disadvantaged
group are exposed to a higher risk of prematurity (9–11).
With ambitions to reduce disparities in birth outcomes, several PCHV
programs have been implemented in the United States, however, the
outcomes have been mixed. Two well-known large-scale randomized
experiments examined the effectiveness of the home visiting approach to
providing prenatal care: Nurse-Family Partnership (NFP) and the Mothers
and Infant Home Visiting Program Evaluation-Strong Start (MIHOPE-Strong
Start). As an earlier attempt, NFP, as a nonprofit organization, has
provided PCHV services for mothers in disadvantaged groups and
established evidence-based practices built upon its foundational
randomized trials in Elmira, Memphis, and Denver (12). Olds the founder
of the program, reported that nurse-visited mothers in the group that
received the PCHV service through the program showed fewer rates of
preterm births, infections, diseases (e.g., pregnancy-induced
hypertension and atrial blood pressures), and higher average birth
weight, compared the other group that did not receive the service (8).
On the other hand, the MIHOPE-Strong Start is a relatively recent
attempt initiated by the Office of Planning, Research, and Evaluation
(OPRE) of the Administration for Children and Families (ACF) in 2012
(7). Different from findings of Olds, this program has not produced
statistically significant improvements in families’ prenatal behaviors
and birth outcomes (7).
At its core, the implementation of PCHV stands on the assumption that
mothers would seek out the service voluntarily if they need it,i.e. , self-efficacy theory (8). However, this may not hold true
because individuals’ care demands are often not connected to the actual
use of care services due to mothers’ preferences or lack of knowledge
and information. For example, mothers who work full-time might be less
likely to have a limited chance to use prenatal care services than
mothers who work part-time. Although existing programs often narrow the
target based on sociodemographic profiles of mothers, such an approach
is still exposed to poor accuracy in screening mothers with high risks.
Prematurity at birth is a complex clinical problem that occurs by
various sociodemographic and behavioral risk factors (13,14). Therefore,
identifying high-risk individuals with only a couple of sociodemographic
factors may not be sufficient to account for the causal complexity of
birth prematurity.
PCHV interventions may have a higher potential if the agency can
selectively encourage mothers for whom a premature birth is highly
likely, with higher precision that allows identifying high-risk
individuals. To achieve this, this study explores the potential of
applying machine learning (ML) techniques in predicting premature births
by using the birth data from the National Center for Health Statistics
(NCHS) of the Centers for Disease Control and Prevention (CDC). More
specifically, this study compares the performances of ordinary linear
regression (OLS) and Deep Neural Network (DNN). The application of ML
has been increasing in the fields of medicine, including mortality (15),
cancer (16), opioids (17), influenzas (18), and mental diseases (19).
Different from conventional statistical models that express the outcome
as a linear combination of inputs and weights, a DNN classifier improves
predictive performance by employing non-linear functions and multiple
layers that bridge inputs and outputs (20). By combining this approach
with the US birth data that is publicly accessible to most of public
agencies, this study shows that attempts to use a home visiting approach
for improving birth outcomes among mothers with high risk can benefit
from a predictive approach improved by ML.