Topics of Interest Part IV

1
Topics of InterestPart IV

• Vasileios Hatzivassiloglou
• University of Texas at Dallas

2
Medical Informatics

• Use of computer systems to support medicine
• A broad field, with some very engineering,
  support tasks and some challenging AI-related
  tasks
• PDAs for doctors accessing lab results
• security / privacy concerns
• analysis of images, lab results, and text

3
Scrubbing medical records

• Privacy an important concern (for legal reasons
  as well)
• Patient records are maintained online and are a
  treasure trove for data mining
• insurance companies
• pharmaceutical companies
• Text analysis (e.g., Scrub developed at CMU) can
  locate personal information and remove it

4
Predicting hospital stay

• Given measurements on a patient taken upon
  admission to a hospital / ICU, predict
• whether the patient will live
• how long will the patient stay in the unit
• Important for scheduling and cost assessments
• Mortality can be predicted at the 95 level
• Morbidity can be predicted within 20 at the 85
  level

5
Customized information retrieval

• Doctors want to access journal articles about
  their patients medical condition
• There are too many articles
• Doctors have little time
• A large Digital Library system was built to
  address these needs in cardiac intensive care
  (cost 8 million over six years)

6
Challenges

• Can we personalize the retrieved results to the
  patient?
• Can we present only the relevant portions of
  selected articles?
• Can we combine information from many articles?

7
The online patient record

• Serves as a user model
• Contains many types of information
• pharmacy report (list of drugs)
• catheterization lab report (tables of numeric
  values)
• EKG (stream of numeric values across time)
• echocardiogram and chest x-ray (images)
• pre-op history, physical, admission and discharge
  summary (text)

8
(No Transcript)
9
Understanding the patient record

• Extract important terms
• Calculate values for each term, if applicable
• Ischemic heart failure
• Base heart rate 80
• Left ejection fraction low

10
Matching patients and articles

• Both patients and articles are represented as
  vectors of term-value pairs
• Terms are weighted by rarity and type
• Match is based on how well both the terms and the
  corresponding values match

11
Scenario
12
Articles

• Clinical Predictors of In-Hospital Prognosis in
  Unstable Angina ECLA3
• Risks and Benefits of Combined Maze Procedure for
  Atrial Fibrillation Associated With Organic Heart
  Disease
• Prognostic Value of Cardiac Troponin T After
  Noncardiac Surgery 6-Month Follow-Up Data
• Primary Pulmonary Hypertension Improved
  Long-Term Effects and Survival With Continuous
  Intravenous Epoprostenol Infusion
• Implantable Left Ventricular Assist Devices
  Provide an Excellent Outpatient Bridge to
  Transplantation and Recovery
• Myocardial Viability in Patients with Chronic
  Coronary Artery Disease and Previous Myocardial
  Infarction Comparison of Myocardial Contrast
  Echocardiography and Myocardial Perfusion
  Scintigraphy
• Match Patient A Match Patient B Match both

13
A Matching Patient Record and Journal Article

• Patient Record
• This is a 44 year old female past medical history
  of coronary artery disease, status post
  myocardial infarction in 1983, status post CABG
  in 1989 The patient was admitted to New York
  Presbyterian Hospital on 12/3/99 sic a
  worsening CHF and unstable angina for evaluation
  for heart transplant.
• Medical Article
• This was a multicenter prospective study of
  consecutive patients admitted to coronary care
  units with unstable angina. Baseline
  characteristics were age 60.18-16 years, history
  of prior myocardial infarction in 336 patients
  (32) In-hospital treatment consisted of
  angioplasty or coronary artery bypass grafting
  (CABG) in 25.1.

14
Summarizing the results

• Identify portions of the article that convey
  results
• Text categorization using training based on key
  phrases, position within Results Section
• Within results select sentences and phrases that
  refer to characteristics of the patient
• Identify and omit repeated information
• Highlight any conflicts

15
Sample output The patient

• 44-year old female, post-infarction, post-bypass,
  congestive heart failure, ischemic etiology of
  heart disease, base heart rate of 96 beats/min,
  considered for a pacemaker.

16
Sample output Summary

• The risk of death was similar for the subset of
  men and women with ischemic heart disease of
  heart failure 1. A significant reduction in
  sudden death was observed in patients with a BHR
  greater than 90 beats/min treated with amiodarone
  4. ... One article 8 found that pacing with a
  100-msec AV delay fails to improve hemodynamic
  function in patients with severe chronic CHF
  while another 9 showed a significantly better
  quality of life in patients treated with pacing.

17
Bioinformatics

• The application of computer science methods and
  tools to the domain of biology
• Primarily, molecular biology (proteins and genes)

18
Why bioinformatics?

• Proteins and genes can be viewed as (long)
  sequences of letters from a fixed alphabet
• allows discrete methods for studying strings
• They carry information
• relates to coding and information theory
• A large number of interacting agents
• modeled as networks
• machine learning of interactions