Medical decision support using machine learning for early detection of late onset neonatal sepsis

Creating Medical decision support using machine learning for early detection of late onset neonatal sepsis

Abstract

The objective for the study was to develop models for late-onset for neonatal sepsis. This study used machine learning (ML) which is a subfield in the artificial intelligence. Earlier work showed feasibility of building predictive models with clinical potential.

Design

299 Infants admitted to neonatal intensive care unit in the Monroe Carell JR. Childrens hosptial at Vanderbilt Developed Nine Machine Learning (ML) algorithms

• Study of 18 month period

• Data

1826 total admission and sample size was 299.

Definition of late onset was defined as neonatal sepsis occurring over 72 hours after birth.

Measurement

Compared blood work

1. sensitivity
2. specificity
3. Positive value
4. negative value

Results

Treatment sensitivity

• Top three predictive variables

1. hemotocrit
2. packed cell volume
3. Chorioamnionitis and Respiratory reate

Conclusion

The study showed that with the use of the electronic health record (EHR) and the machine learning (ML) algorithms exceeds the sensitivity and specificity of clinicians.

Second Review

Introduction

Machine learning, a sub field of artificial intelligence, has been shown to have practical medical use. This article seeks to see if that use can be applied to early detection of late onset neonatal sepsis.

Methods

Participants collected data from a local hospital and created an algorithm to detect onset neonatal sepsis.

Results

Machine learning's sensitivity to signs of neonatal sepsis greatly exceeded that of a physician.

Conclusion

Machine learning was concluded to be a viable form of neonatal sepsis detection.

Comments

This was an interesting article that brought attention to the benefits of using computer based monitoring in addition to physician study.

Related Concepts

Natural language processing (NLP)

Related Articles

Visualizing unstructured patient data for assessing diagnostic and therapeutic history

Costs and benefits of health information technology