Decreasing Variability in Health Care HST950 Decision Systems Group, Brigham & Women’s Hospital Harvard Medical School Harvard-MIT Division of Health Sciences and TechnologyHST.950J: Medical ComputingPeter Szolovits, PhD Isaac Kohane, MD, PhD Lucila Ohno-Machado, MD, PhDVariability in Health Care Decision support systems Integration of guidelines into practice Decrease variability, homogenize Knowledge discovery in biomedical data Increase variability, customize Support for clinical trialsGuidelines and clinical protocols What are they? Why computerize? Knowledge representation Application in breast cancer protocol eligibility with uncertain informationDecreasing practice variation Studies demonstrate huge variability in practicesWhat are clinical guidelines? Institute of Medicine definition systematically developed statements to assist practitioner and patient decisions about appropriate healthcare for specific clinical circumstances A recommended strategy for management of a medical problem in order to Reduce inappropriate use of resources $$$$$$ Reduce practice variation Improve outcomesConventional publication Guidelines can be developed and published by A medical institution, to be used locally National and international organizations, used by many medical institutions Conventional publication In journals and textbooks Booklets or guideline summaries Compilations of guidelines for referenceTypes of guidelines Risk assessment Chronic disease management Diabetes, asthma, hypertension Screening Diagnosis and workup Protocol-based care (clinical trials)Clinical Trial Protocols Goal is to intervene in a random part of the eligible patient and leave the other part with current standard of care Carefully selected population, with few comorbidities (other diseases) Homogeneous care in each arm to investigate statistical significance of differences Select patients Randomize into - intervention arm -control arm Compare outcomesWhere do the recommendations come from? Panel of experts (most common) Hard to get experts to agree on anything Decision analysis models (least common) Difficult to obtain probabilities and utilities Observational studies Small numbers may lead to wrong recommendations Clinical trials Controlled populations, strict eligibility criteria A major problem is to match the patient in front of you with carefully selected patient population used in the trialsWays of helping implement guidelines/clinical trials Help authors to create guidelines that make sense (verify the “logic”) Eligibility determination for a variety of competing guidelines/protocols Assistance in implementing the prescribed actionsEligibility determination There are hundreds of guidelines and clinical trials out there Automated eligibility could warn providers of guidelines/protocols that match the patient MAJOR problem: uncertainty in patient status (tests to be done, info not available)Increase versus decrease variability Recommendations are based on “average” or “mode” patient “Mode” patient may not exist If more info is available, why not use it?Example Consent forms for interventional cardiology procedures: Acknowledgement that risk of death in hospital is about 2% Who is at 2% risk?Why people want to computerize guidelines Provide automatic decision support Applied to individual patients During the clinical encounter Ambiguities in guidelines may be reduced Software tools and guideline models can promote specifying logic precisely Can integrate guidelines into workflow Patient-specific guideline knowledge available at point of care $$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$Computer-interpretable guidelines Interactive guidelines Enter patient parameters to traverse guideline Guidelines embedded in EPR Systems Automated reminders/alerts Decision support and task management… Why people want to computerize guidelines Can be used for quality assurance Guideline defines gold-standard of care Perform retrospective analysis to test if patients were treated appropriately Allows for interactive visualization of guideline logic e.g., allows one to focus on relevant sections of flowchart $$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$Why share guidelines? Provide consistency in guideline interpretation Reduce cost of guideline development Minimize misinterpretations and errors through the process of public reviewChallenges in sharing guidelines Local adaptation of guidelines Must allow care sites flexibility in modifying guidelines for Availability of resources and expertise Local workflow issues Practice preferences Differences in patient populationPatient and Provider Preferences Who cares? Who elicits preferences for a particular patient? How does this get taken into account?Patient and Clinician Vocabulary: How Different Are They?…Challenges in sharing guidelines Integration with information systems Match patient data in EPR to terms in guideline Match recommendations in guideline to actions in order entry systemGuideline models Guideline models make explicit Knowledge concepts contained in a guideline Structure of the concepts and relationships among them Scope of the model Types of guidelines, e.g. alerts vs. multi-encounter guidelines Level of detail, e.g. structured or text specificationModels for guidelines and rules Individual decision rules (single step) Arden Syntax Multi-step guidelines, modeled as sets of guideline tasks that are connected in a graph nestedArden Medical Logic Modules Format for representation and sharing of single medical decision Each medical decision (rule) is called a medical logic module (MLM) Suitable for alerts and reminders A guideline may be represented by a chained set of MLMs…Arden MLM Simplified example data: potassium_storage := event {‘1730’}; potassium:= read last { ‘32471’}; evoke: potassium_storage (to EPR) logic: potassium > 5 mmol/L action: write “Potassium is significantly elevated”;…Arden Syntax Standard published by ANSI Part of HL7 activity Supported by many commercially-available hospital information systems…Models for multi-step guidelines Multi-step guidelines, modeled