Identification of Patients with Low Life Expectancy (Massachusetts)

Project Final Report (PDF, 445.1 KB) Disclaimer

This project does not have any related annual summary.

Identification of Patients with Low Life Expectancy - Final Report

Citation:
Turchin A. Identification of Patients with Low Life Expectancy - Final Report. (Prepared by Cincinnati Children's Hospital Medical Center under Grant No. R21 HS024977). Rockville, MD: Agency for Healthcare Research and Quality, 2019. (PDF, 445.1 KB)

The findings and conclusions in this document are those of the author(s), who are responsible for its content, and do not necessarily represent the views of AHRQ. No statement in this report should be construed as an official position of AHRQ or of the U.S. Department of Health and Human Services.
Principal Investigator: 
Document Type: 
Research Method: 
Population: 
This project does not have any related resource.
This project does not have any related survey.
This project does not have any related project spotlight.
This project does not have any related survey.
This project does not have any related story.
This project does not have any related emerging lesson.
The inclusion of information from free-text notes into risk of death prediction models significantly improves the ability to predict the probability of death, with the potential that clinicians will be better able to incorporate life expectancy into shared decision making about medical interventions.

Project Details - Ended

Summary:

Clinical decision support (CDS) has been used to bring guidelines to clinicians at the point of care. However, CDS typically employs a one-size-fits-all methodology that may consider only a patient's age and gender, rarely taking into consideration a patient's comorbidities, medications, patient preferences, and importantly, life expectancy. Knowing a patient's life expectancy is crucial in caring for individuals, since benefits of many medical interventions, such as cancer screening or treatment of chronic illness, are predicated on patients living long enough for the interventions to have an impact. While there are many published mortality prediction tools, they are infrequently integrated into electronic medical records (EMRs); when they are, data on functional status needed for the tools are rarely electronically available since these functional assessments are typically documented as part of narrative provider notes.

With a goal of identifying patients at high risk of death, this research integrated the artificial intelligence technique Dynamic Logic in conjunction with natural language processing (NLP) of provider narrative notes. This model of risk of death was created, tested, and analyzed against benchmark statistical and machine learning algorithms.

The specific aims of the research were as follows:

  • Determine whether a combination of artificial intelligence technology Dynamic Logic and NLP improves accuracy of identification of patients with low life expectancy (likely to die over the next 12 months) compared to the currently used methods. 
  • Determine whether the low life expectancy model developed on the general patient population is equally accurate in patients with chronic conditions on the example of patients with a) diabetes, b) hypertension, and c) osteoporosis. 
  • Develop open source software that will integrate Dynamic Logic and NLP to allow users to easily assess patients' life expectancy based on a combination of structured and narrative EMR data. 

A retrospective cohort analysis was completed. The Dynamic Logic combined with NLP model was tested on a study population of 630,000 patients 40 years and older who were treated at Partners Healthcare in Eastern Massachusetts between 2000 and 2014. Data were obtained from the existing EMR and included patient demographics, diagnoses, procedures, vital signs, laboratory tests, and data from provider notes. The second aim was not able to be completed because patients with diabetes, hypertension, and osteoporosis were only five to 15 percent of the data sample, insufficient for analysis.

The researchers found that inclusion of NLP-generated data improved estimates of the probability of death over traditional methods. Accuracy of the algorithm was significantly improved with algorithm optimization methods and data normalization. Pulling information from provider notes significantly increased accuracy.