Lifelines2: Discovering Temporal Categorical Patterns Across Multiple Records

 

Project Description

Electronic health records (EHRs) contain a wealth of information.  Categorical event data such as complaints, diagnoses, treatments, etc., are important, and play important roles in health providers decision making.  However, past research efforts have been focused on numerical data and single-record visualization techniques.  Discovering patterns of categorical events across multiple records are supported in limited ways.

Lifelines2 is an interactive visualization tool for visualizing temporal categorical data across multiple records.  The goal of the project is to enable discovery and exploration of patterns across these records to support hypothesis generation, and finding cause-and-effect relationships in a population.  These tasks are not  specific to the medical domain, but we were first motivated by EHRs.

Features

In Lifelines2, we advocate a framework of simple operators to allow users to manipulate multiple records simultaneously to understand relative temporal relationships across records.  The three operators are Align, Rank, and Filter, and we affectionately call it the ARF framework.  Alignment forces every record to be aligned by a certain feature (e.g. 3rd Heart Attack) so the events that occur prior to and after the feature can be compared easily.  Rank and Filter are traditional operators analysts are familiar with, and they augment the Align offers.  In addition, analysts can use temporal summaries to view distribution of multiple event types over time.  Temporal summaries augment filter by allowing temporal constraints to be specified.  Temporal summaries also allow multiple groups of records to be compared.

1.  377 patient records are shown in the main visualization.  These patients had been administered radiology contrast.  Physicians are interested in patients that have high creatinine readings within 2 weeks after the administration of contrast.	2. Every patient's 1st Radiology Contrast event (yellow triangles) is brought on the same vertical line, allowing analysts to view creatinine readings before/after the alignment quickly and perform relative comparisons.	3.  Rank is applied to bring patients with the most number of creatinine high (CREAT-H) readings up top for analyses.

4.  Temporal summaries can be brought forth to see distribution of multiple events over time.	5.  A final zoom into the timeline allows analysts to examine the events in detail.

 

Lifeliens2 is primarily designed for categorical point events, but analysts can add intervals of validity to visually remind them of implicit durations to aid their decision-making process.

 

Video Demonstrations

Title Screen	Description	Format/Resolution
(click to view in 1024x768)	Medical Scenario: Some small percentage of population (1%-7%) experience reduced renal function after infusion of radiographic contrast material such as iodine-based agents.  To monitor patients' renal status, serum creatinine value in blood is tested.  High creatinine readings indicates reduction in renal function.   Demo Summary: This demonstration shows how to use the main features of Lifelines2 to find patients who experience reduced renal function as a result of adverse radiology contrast procedures.	4 minutes 58 seconds
		Flash (.swf)	1024x768
			800x600
(click to view in 1024x768)	Medical Scenario: Heparin-induced thrombocytopenia (HIT) is characterized by >50% of platelet counts within 5-9 days after exposure to heparin.  This is a dangerous condition.  Physicians are interested in two questions: (1) Do HIT patients tend to stay in ICU longer?  (2) How effective is the drug Argatroaan  for HIT patients?   Demo Summary: This demonstration shows how to flexibly create groups and compare them using temporal summaries.  We assume viewers are already familiar with the basic features of Lifelines2.	8 minutes 06 seconds
		Flash (.swf)	1024x768
			800x600

 

 Participants

• Taowei David Wang, Ph.D. Student, Computer Science
• Catherine Plaisant, Associate Research Scientist, UMIACS
• Ben Shneiderman, Professor, Computer Science

Publications

Taowei David Wang, Catherine Plaisant, Ben Shneiderman, Neil Spring, David Roseman, Greg Marchand, Vikramjit Mukherjee, and Mark Smith.  Temporal Summaries: Supporting Temporal Categorical Searching, Aggregation and Comparison.  HCIL technical report(HCIL 2009-09). March, 2009.

Taowei David Wang, Catherine Plaisant, Alex Quinn, Roman Stanchak, Ben Shneiderman, and Shawn Murphy.  Aligning Temporal Data by Sentinel Events: Discovering Patterns in Electronic Health Records.  SIGCHI Conference on Human Factors in Computing Systems (CHI 2008).

LifeLines2 Links

 

LifeLines2 Tutorial

LifeLines2 File Format

CategoryMapper Tutorial

 

Sponsors and Partners

This project is supported in part by the Washington Hospital Center and Harvard Medical School - Partners HealthCare

Other Related Projects from HCIL

Lifelines:  Visualizing patient records, criminal records, and personal histories.

PatternFinder:  integrated interface for visual query and result-set visualization for search and discovery of temporal patterns within multivariate and categorical data sets.

PatternFinder in Azyxxi:  Temporal Query Formulation and Result Visualization in Action

Related Workshops from HCIL

Personal Medical Devices Workshop: Increasing Patient Healthcare Participation (June 3, 2004)

Interactive Visual Exploration of Electronic Health Records (May 30, 2008)