Presentation is loading. Please wait.

Presentation is loading. Please wait.

Temporal Data Analysis and Electronic Health Records Ben Shneiderman, Catherine Plaisant, Taowei David Wang, Kris Wongsuphasawat {ben,plaisant,tw7,kristw}@cs.umd.edu.

Similar presentations


Presentation on theme: "Temporal Data Analysis and Electronic Health Records Ben Shneiderman, Catherine Plaisant, Taowei David Wang, Kris Wongsuphasawat {ben,plaisant,tw7,kristw}@cs.umd.edu."— Presentation transcript:

1 Temporal Data Analysis and Electronic Health Records Ben Shneiderman, Catherine Plaisant, Taowei David Wang, Kris Wongsuphasawat

2 LifeLines: Patient Histories

3 Lifelines and Improvements
Plaisant CHI 96, AMIA 98 I2b2 (Murphy AMIA 07)

4 Lifelines and Improvements
Plaisant CHI 96, AMIA 98 Bade CHI 2004 I2b2 (Murphy AMIA 07)

5 Specification of Temporal Abstractions
Shahar 1999 Powsner & Tufte, 1994 Post 2007

6 PatternFinder in Amalga

7 1) LifeLines2: Align-Rank-Filter
Multiple records simultaneously visible Align by sentinel events Rank by frequency Filter by events

8 LifeLines2: Contrast+Creatinine
Contrast and Creatinine dataset In some diagnostic radiology procedures, patients are injected contrast material. However, some patients develop adverse side effects to the contrast material. One serious side effect is renal failure, which is detected by high creatinine levels in a patient's blood. This adverse effect usually occur within two weeks after the radiology contrast. WHC is interested in finding the proportion of patients who exhibit this condition in historical records. Screenshots 1-aligned-ranked.png: We align by the 1st occurrence of radiology contrast and rank by the number of creatinine high (CREAT-H) events to bring the most severe patients to the top. We realize two things: (1) some patients have more than 1 "Radiology Contrast" events, and (2), some patients have consistently high creatinine readings (chronic kidney failure). 2-aligned(all)-distribution-selected.png We align by all occurrences of raiology contrast, and then show the temporal summary of CREAT-H events. The patients are presented in 4 exclusive sets in the summary: those who have CREAT-H only before alignment, only after alignment, both before and after, and neither. We then select from the "only after" summary the patients who have at least one CREAT-H event within 2 weeks of any "Radiology Contrast" event. There are 421 patients.

9 LifeLines2: Contrast+Creatinine
VIDEO

10 LifeLines2: Heparin-Induced Thrombocytopenia
Heparin induced thrombocytopenia dataset Heparin is used for anticoagulation when abnormal blood clotting occurrs in a patient. However, Heparin induced thrombocytopenia can occur, usually within 4-14 days after heparin administration. WHC wishes to find how many patients may exhibit these signs. Screenshots 1-ranked.png: The dataset is loaded and ranked by the number of platelet "Low or Critical" events. First, we notice that many patients have more than 1 administrations of heparin. We also see that, serendipitously, that some patients, after being admitted to trauma center, some patients stayed in the hospital for months. 2-aligned(all)-ranked-selected.png We then align by all occurrences of "Heparin Administration", and then select, from the temporal summary, patients who have at least one "Low or Critical" platelet event within 4-14 days after any heparin administration. We see that there are 1094 patients selected.

11 LifeLines2: Heparin-Induced Thrombocytopenia

12 2) Similan: Find similar patients
In terms of ? Not in terms of demographic information Age, gender, etc. But in terms of patient histories Events time Patient#1 Patient#1 Stomach Pain Vomitting Diarrhea

13 Similan: Find similar patients
How? Similarity Measure Between pairs of patients Score ranges from 0 to 1. (Most similar = 1) Users select one target patient

14 Similan: Find similar patients
(target) Patient #2 0.80 0.63 0.96 0.77 Patient #3 Patient #4 Patient #5

15 Similan: Find similar patients
(target) similar Patient #4 0.96 Patient #2 0.80 Patient #5 0.77 Patient #3 0.63 15

16 Challenge What is “similar”? depends on users/tasks Patient#1 (target)
B C Patient#2 swapping A C B B missing Patient#3 A C Patient#4 Time difference A B C time

17 } Mix&Match Measure User customizable similarity measure Total Score
Missing Event Patient#1 (target) Matched Events A C B C Extra Event Patient#2 A B C B } Total Score 0 to 1 Match Score = F(∆ time) Number of Mismatches = Missing + Extra Number of Swapping*

18 Similan: Find Similar Patients
VIDEO Trauma Admission-hematocrit dataset Hematocrit is the proportion of blood volume that is occupied by red blood cells. Low or critical levels of hematocrit indicate significant hemorrhaging. WHC is interested in what proportion of trauma patients who have low or critical levels of hematocrit within 24 hours trauma admission. Screenshots 1-defualt.png: This screen shows the data that is loaded into Lifelines2. The patients are scattered on the calendar time, forcing users to scan horizontally as well as scrolling vertically. 2-aligned(first)-ranked-selected.png We align the data by the first Trauma Admission, and rank the patients by the number hematocrit low or critical events to bring the most 'severe' patients on top. Finally, we show the distribution of "HCT Low or Critical", and select the patients who have "HCT Low or Critical" within 24 hours of Trauma Admission. The blue number 4514 on top shows how many patients satisfy the selection.

19 Take Away Messages Medical Record & other temporal event
exploration is possible for: 1) pattern discovery 2) comparison of groups 3) similar event records Thanks for collaboration with Washington Hospital Center


Download ppt "Temporal Data Analysis and Electronic Health Records Ben Shneiderman, Catherine Plaisant, Taowei David Wang, Kris Wongsuphasawat {ben,plaisant,tw7,kristw}@cs.umd.edu."

Similar presentations


Ads by Google