International Conference of the European Federation for Medical Informatics Sarajevo, Bosnia & Herzegovina August 30 – September 2, 2009 Intelligent Consumer-Centric Electronic Medical Record Gang Luo Selena B. Thomas Chunqiang Tang IBM T.J. Watson Research Center {luog, selenat, ctang}@us.ibm.com
Consumer-Centric Medical Search 2
Consumer-Centric Medical Search 3
Consumer-Centric Medical Search 4
Consumer-Centric Medical Search 5
Consumer-Centric Electronic Medical Record • ! In 2004, US President Bush announced plans for most Americans to have electronic medical records by 2014 • ! Google, Microsoft, and Revolution Health launched their Web-based electronic medical record services for ordinary consumers • ! Regarded as one of the most effective methods of addressing the incoming healthcare crisis • ! Existing ones have little intelligence (e.g., checking drug- drug interactions) and can fulfill only a small portion of consumers’ needs 6
Our Proposal: Intelligent Consumer- Centric Electronic Medical Record • ! Add more intelligence into consumer-centric electronic medical record to make it more powerful and serve consumers better • ! Introduce and extend expert system and Web search technology into the consumer-centric electronic medical record domain 7
System Architecture medical knowledge base electronic healthcare search search expert medical information guide system system record information 8
Example Usage • ! Based on people’ medical situations and healthcare needs, automatically recommend – ! Home medical products (focus of this work) • ! Bendable utensils for people with arthritis to aid self-feeding • ! Autodrop Eyedropper Aid for people taking eye drop medication • ! Physicians receive little training on such products in medical school and often cannot help much – ! Retirement place & Residence place • ! Old man with gout, family with a deaf child – ! Local preventative health initiatives • ! Depend on living place, income level 9
Outline • ! User interface • ! Search techniques • ! Some experiments 10
Main Web Page Recommend home medical products consumer-centric Function 2 electronic medical record Function 3 … 11
Input Interface Diseases high importance Viral upper respiratory infection ! G ! Pneumonia Diabetes mellitus Congestive heart failure ◄ ► Previous Next 12
Input Interface – Cont. • ! Symptoms – ! List all possible symptoms of selected diseases • ! Other information – ! Occupation – ! Recent surgeries – ! Pregnancy – ! Computer usage • ! Obtain a list of topics reflecting the consumer’s medical situation and healthcare needs – ! Diseases, symptoms, and other information 13
Output Interface • ! Sequential order presentation (default) – ! Home medical products recommended for various topics are mixed together • ! Hierarchical presentation (alternative) second level first level product 1 topic 1 product 2 … product 1 topic 2 product 2 … … 14
Outline • ! User interface • ! Search techniques • ! Some experiments 15
Step 1: Obtaining Search Guide Information • ! Straightforward approach: For each topic, use its name as a query to retrieve the products for recommendation – ! Frequently improper – ! Topics have their underlying medical meanings – ! Consumer with symptom “weight loss” would like to gain weight • ! Solution: Use medical knowledge to provide semantic translation from topics to their underlying medical meanings • ! For each disease or symptom s , a medical professional pre-compiles a set of phrases as search guide information – ! Each phrase provides one common way of representing the medical meaning of treating s 16
Step 2: Finding Relevant Home Medical Products • ! Use a vertical search engine that crawls Web pages from a few selected, high-quality home medical product shopping Web sites • ! For each topic, perform search using its search guide information S g – ! Use each phrase in S g to retrieve some relevant products – ! Combine together the retrieved products for all the phrases in S g • ! Combine together the retrieved products for all the topics 17
Step 3: Ranking Home Medical Products • ! Traditional ranking methods are used for a single query and unsuitable for our purpose – ! Home medical product Web page are retrieved by different phrases in various topics’ search guide information – ! The number of contained terms can vary significantly from one phrase to another – ! Traditionally computed relevance scores are the product of multiple numbers and on different orders of magnitude for various phrases 18
Our Solution: Extended Language Modeling Method • ! Consider the special properties of our application scenario and fold all the relevant factors into a single big formula – ! Use summation form by treating all the topics as a conceptual disjunctive query – ! Treat each topic rather than each term as a semantic unit – ! Each topic is represented by the phrases in its search guide information – ! Larger weights are given to more important topics – ! Equally important topics are treated equally – ! Count topics (phrases) rather than terms 19
Step 4: Diversifying Search Results • ! Users prefer that the returned first few Web pages mention various topics and provide much new information • ! Penalize Web pages whose contents are similar to those of previously returned Web pages • ! Penalize Web pages mentioning the same topics as previously returned Web pages 20
Outline • ! User interface • ! Search techniques • ! Some experiments 21
Medical Scenario • ! Weight loss • ! Hand weakness The patient is a 61 years old male complaining of fatigue and weakness. The patient notes that the fatigue and weakness started 6 months ago. He feels tired all day. He has poor appetite and lost 8 pounds in the last 6 months. He also complains of occasional nausea and of a vague, deep epigastric discomfort that radiates to the back. He feels sad sometimes, has lost interest in things that he used to enjoy ... 22
Some Returned Relevant Web Pages rank purpose product name 1 weight Vitol Incredible gain Quick Weight Gain 2 weight Genesis Nutrition gain Weight Gain Vanilla 3 weak Boxtopper with Built- hand Up Handle 5 weak T-Bar Cockup hand 23
Conclusions • ! Intelligent consumer-centric electronic medical record – ! Use medical knowledge to help consumers obtain personalized, relevant healthcare information to facilitate their daily lives – ! Require no special user training – ! Form queries automatically – ! Provide diversified search results – ! Effective and efficient at recommending home medical products 24
Recommend
More recommend
