Machine Learning Can Unlock Insights Into Mortality

The study of mortality is fundamental in public health research, but our ability to derive detailed insights is often limited by the practical constraints ofthe available data. Although the National Vital Statistics System maintains a national record of death certificate data enabling basic research...

Full description

Saved in:
Bibliographic Details
Published in:American journal of public health (1971) 2021-07, Vol.111 (S2), p.S65-S68
Main Authors: Young, Jessica C., Pack, Cory, Gibson, Teresa B., Yoon, Frank, Irwin, Debra E., Shiv, Shalu, Cooper, Toska, Dasgupta, Nabarun
Format: Article
Language:English
Subjects:
Best practice
Bias
Data
Death
Death & dying
Epidemiology
Health care
Health insurance
Health research
Insurance coverage
Laboratory tests
Learning algorithms
Life expectancy
Life span
Machine learning
Mortality
Opinions, Ideas, & Practice
Other Race/Ethnicity
Other Statistics/Evaluation/Research
Prescription drugs
Privacy
Public health
Statistics
Vital statistics
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The study of mortality is fundamental in public health research, but our ability to derive detailed insights is often limited by the practical constraints ofthe available data. Although the National Vital Statistics System maintains a national record of death certificate data enabling basic research on mortality trends and life expectancy, clinical information on the context of a given death in national death records is limited. When the necessary information is available, the ability to examine clinical details surrounding death and health indicators in the period before death enables research that can meaningfully inform public health strategies and interventions.Insurance claims data can fill gaps in mortality research by providing details about diagnosed health conditions and key health care services a person receives before death, such as surgical procedures, laboratory tests, and drug prescriptions. This level of granularity composed of data that are continuously and prospectively collected for each patient offers insights that are not available with vital statistics alone and opens the door to uncovering how medications, health conditions, and health encounters may be associated with mortality. A major limitation in mortality research is that data sets that have rich longitudinal health information (claims) and those that have recorded death dates (vital statistics) are often separate, and linkage may be prohibitively expensive or prohibited because of data privacy restrictions. We discuss the research implications of having disparate streams of health and mortality data; introduce how machine learning can help overcome these limitations; highlight important considerations for machine learning, including the risk of algorithmic bias; and briefly discuss best practices for applying machine learning to enhance public health research.
ISSN:0090-0036
1541-0048
DOI:10.2105/AJPH.2021.306418