# Clinical Informatics

#### 2.1 Clinical Decision Support

##### 2.1.1 The nature and cognitive Aspects of human decision making
 utility vs value utility is a function of value and also risk aversion, personal preferences / circumstances

###### 2.1.1.1 General
 Availability bias overestimating the probability of unusual events because of recent or memorable instances Omission bias tendency to avoid interventions that may have serious side effects (vs comission bias - inaction felt to be less desirable) Visceral bias tendency for emotions to influence our decisions Confirmation bias tendency to corroborate rather than refute Outcome bias tendency to judge decisions based on past outcome rather than the quality of the decision Representativeness heuristic (bias) overestimate rare diseases by matching patients to 'typical picture' of that disease. Anchoring Failure to adjust probability of disease/outcome based on new information, akin to 'premature closure' Value-induced bias Overestimate probability of an outcome based on value associated with that outcome. Conjunction fallacy assumed that specific conditions are more probable than a single general one (linda is a bank teller and feminist > linda is a bank teller)

###### 2.1.1.2 Medical
 PROACTIVE framework PROblems and objectives (define the Problem, Reframe from different perspectives, identify Objectives for action); consider Alternatives (one of wait-and-see, initiate intervention, obtain more information), model the Consequences, estimate the value of Trade-offs; Integrate the evidence and optimize Values, Explore assumptions and evaluate uncertainty Markov Decision Process MDP a discrete time stochastic control process. Decision making framework for situations where outcomes are partly random and partly under control of decision maker; useful for progression of chronic disease and simulation with Monte Carlo methods Heuristic Decision Making process of making decisions based upon using simple and efficient rules Blois' funnel Breadth of diagnostic considerations are refined, restricted over course of interaction between patient and practitioner

##### 2.1.2 Decision Science
 Decision Science provides structure and guidance for systematic thinking; based on logical principles, informed by what we know about the limitations of human judgment and decision-making; allows logical and consistent analysis of complex decisions

###### 2.1.2.1 Decision analysis
 Decision analysis the formal application of philosophy, methodology, and professional practice to understand and aid in decision-making Influence diagram ID nodes: decision-rectangle, uncertainty-oval (RVs), value-octagon/diamondl & arcs (functional, conditional, informational); emphasizes probabilitstic relationships among variables; an ID composed only of chance nodes is a Bayseian belief network Decision tree Decision node (decision maker in control) is square, chance node (probabilities) is circle, outcome node is triangle What if / Sensitivity Analysis Use a range of values at the nodes to see how the model changes. Plot Expected value vs estimated probability. Lines intersect when standard of care = med x

###### 2.1.2.2 Probability theory
 Bayes theorem P(A|B) = P(B|A)P(A)/P(B)

###### 2.1.2.3 Utility and preference assessment
 Patient preferences the individual's evaluation of dimensions of health outcomes; factored into utility (value). Direct elicitation vs indirect assessment (prescored classification instruments, eg EQ-5D, SF-6D, 15D, Quality of Well-being Scale, HUI) Utility standard gamble (choose between X time in illness vs therapy with a known risk of cure/death), TTO (time in state of illness vs time in state of perfect health), visual analogue (rate health states on a scale 0-100) Time Trade-off TTO years of perfect health willing to live / 10 years of current health state. TTO * # of years in current health = QALYs Standard Gamble SG utility of x years when having x years for sure equivalent to a gamble for 10 years vs 0 years with probability p = U Quality Adjusted Life Year QALY 1 QALY is a year of perfect health; thus QALYs are expressed in terms of years lived in perfect health; weight values between 0 and 1 determined by TTO or SG

###### 2.1.2.4 Cost effectiveness analysis
 Cost effectiveness analysis Health benefit (e.g. QALY, lives saves, cases prevented) vs cost in dollars. Set a cost effectiveness threshold, eg. \$50,000/QALY saved Cost benefit analysis Compares both cost and benefit in dollars. Incremental Cost/Effectiveness Ratio ICER one way to determine if a therapy is cost effective or not. Change in cost with change in a unit of effectiveness.

###### 2.1.2.5 Test characteristics
 Receiver Operating Characteristic Curve ROC plot sensitivity (y) vs 1-specificity (x), i.e. true positives vs false positives. AUC higher with better test. Relative Risk p(event|exposed) / p(event|no exposure) False positive rate FPR 1-Sp False negative rate FNR 1-Sn Likelihood Ratio reflects both sensitivity and specificity; LR+ = Sn/1-Sp, i.e. TPR/FPR; LR- = 1-Sn/Sp, i.e. FNR/TNR Accuracy (TP + TN)/ Total population

##### 2.1.3 Applications of Clinical Decision Support
 Applications of Clinical Decision Support CDS the process that provides clinicians, staff, patients or other individuals with knowledge and person-specific information, intelligently filtered or presented at appropriate times, to enhance health and health care. Components of CDS Knoweldge base, Patient-specific information, Mode of communication: "active knowledge systems which uses two or more items of patient data to generate case-specific advice"

###### 2.1.3.1 Types of decision support
 Functions of CDS Alerting, Reminding, Critiquing, Intepreting, Predicting, Diagnosing, Assisting, Suggesting Rights of CDS (Osheroff) Right information, right person, right format, right channel, right time Modes of delivery of CDS templated data-collection, suggestion, summarization, reminder, information, correct errors, recommend change in plan Interruptiveness On demand (e.g link to fomulary); in-line or background (modeless, e.g. 'unread lab result'), popup, interruptive (modal, e.g. alerts, reminders requiring aknolwedgement) Categories of CDS (leapfrog) 1. may use information about the current clinical context to retrieve pertinent online documents (infobutton) 2. may provide patient-specific, situation-specific alerts, reminders, physician order sets or other recommendations for direct action (classic CDS) 3. may organize information in ways that facilitate decision making and action

###### 2.1.3.2 Users of decision support
 Users of decision support not just doctors, can be patients/families. Level of control: Pre-emptive, suppressible, hard-stop, interruptive

###### 2.1.3.3 Implementing, evalution, maintaining decision support tools
 Implementing, evalution, maintaining decision support tools 10 Commandments for Effective CDS (Bates, 2003) Evaluation of CDS Limitation of current literature, evidence of effectiveness limited. More evidence for process of care than patient outcomes. Diagnostic test ordering did well.

##### 2.1.4 Transformation of knowledge into CDS tools

###### 2.1.4.1 Knowledge generation
 Knowledge generation Agreeing upon what constitutes relevant medical knowledge to be represented

###### 2.1.4.2 Knowledge aquisition
 Knowledge aquisition The process of importing knowledge into the CDS system (not the exact logic but the general flowchart of decision points and actions) Knowledge acquisition tools facilitate entry of knowledge in a structured format directly by domain experts (eg Prot_g_, GEODE)

###### 2.1.4.3 Knowedge modeling
 Knowedge modeling Representing on a granular level each piece of logic and resulting alert (from high-level format) Ambiguity term can reasonably be interpreted in more than one way. Syntactic ambiguity (eg A or B and C). Semantic ambiguity (classic form, different interpretations of same word). Pragmatic ambiguity: usage (eg See you next Wednesday) Vagueness boundaries of word's meaning are not well defined; word or phrase reduces level of information contained in a statement (eg 'may be appropriate'). Underspecification is vagueness that occurs when terms are used with insufficient details for definitive interpretation. Temporal vagueness (range from never to always). Probabilistic terms (impossible ... certain with vague terms in between (unlikely, probable). Passive voice (eg 'should be performed'). These uncertainties are established causes of decreased adherence to CPGs and practice variations despite apparent adherence. Strategies for vagueness include creating a controlled vocabulary or limited structured vocabulary, ordinal scale based on ranking, 'black-list' (eg possibly probably, should) Tri-axial model for ambiguity and vagueness (Codish) 1. Semantics: pragmatic ambiguity when two or more recs are inconsistent or conflict; underspecification (eg moderate, adequate); temporal vagueness, probabilistic terms. 2. Source and rationale: inadvertant ambiguity reflects insufficient editing; deliberate vagueness poses a problem; quality of evidence and strength of rationale help but fail to provide clinicians with range of acceptable interpretations. 3. Recommendation component affected: what should be done, when it should be done, why it should be done Computer Intepretable Guidelines must lend themselves to computation; representation format must allow for clinical expressivity (e.g. tempeoral dependencies, complex rules, strength of evidence, imperative/optional actions); ideally: standard vocabularies/semantics; interoperable, portable DIKW Pyramid Data, Information, Knowledge, Wisdom: a class of models for representing puported structural and functional relationships

###### 2.1.4.4 Knowledge representation
 Knowledge representation Clinical algorithms, Bayesian statistics, Production rules, Scoring and heuristics Clinical Algorithm The codification of evidence-based information into clinical pathways, practice guidelines, and decision rules. Branching algorithm; Path through flow chart; information nodes (gather), decision nodes; benefits: knowledge is explicit, easy to encode; limitations: must follow pathway, cant account for prior results, inability to pursue new etiologies; forerunner of modern CPG Probabilistic Systems Use a naive Bayesian model and a sequential approach to calculate posterior probability = f(prior probability, new information). Assumes conditional independence of findings and mutual exclusivity of conditions, eg Leeds abdominal pain system (1975). Leeds Abdominal Pain System de Dombal; Used sensitivity, specificity, disease-prevalent data for various signs, symptoms and test results to calculate using Bayes' theorem, the probability of seven etiologies of acute abdominal pain Belief networks Conditional dependencies are modeled explicity; more modern approach Production Rules Rule-based system; Knowledge encloded as IF-THEN rules. Require a formal language for encoding the rules plus an interpreter (inference engine) that operates on the rules. Limitations: depth first could lead to focus in wrong area; difficult to maintain rule bases; systems slow and time-consuming Inference engines JESS (java-based); CLIPS (NASA); Drools (open-source, developed by JBoss community) Backward chaining system pursues goal and asks questions to reach goal; inference engine determines whether the premise (left) of a given rule is true by invoking other rules that can conclude the values of variables that are currently unknown and that are referenced in the premise of the given rule MYCIN Shortliffe (1975) first rule-based expert system; never used clinically; diagnosed infectious diseases (meningitis/bacteremia), backward chaining approach (unusual now) Forward chaining similar to clinical algorithms. Scoring and Heuristics Knowledge is represented as profiles of findings that occur in diseases; there are measures of importance and frequency for each finding in each disease; most scalable approach for comprehensive DSS Scoring and Heuristics (examples) INTERNIST-1/QMR, DXplain, Iliad. Modern day: Isabel INTERNIST-1 aimed to develop an expert diagnostician; evolved into Quick Medical Reference (QMR) where goal changed to using knowledge base explicity (did not succeed); DxPlain used principles of INTERNIST-1/QMR but developed more disease coverage. Evoking strength = given patient with this finding, how strongly should i consider the diagnosis, i.e. PPV; frequency: how often patients with the disease have the finding, i.e. sensitivity HELP Salt Lake City (1970-80s); generated automated alerts; beginning of event-driven generation of specialized warnings - stored in HELP sectors, evolved into Arden Syntax Arden Syntax HL7 standard representation format for decision support rules. International standard for MLMs endorsed by HL7 and ANSI. Maintenance, Library (eg purpose), Knowledge (mechanics), Resources (eg language) Medical Logic Module MLM An individual decision rule in Arden Syntax. A specialized form of an ECA rule Event-Condition-Action Rule ECA evaluation of situation-specific conditional expression logic is triggered by an external event - if true then action performed. Eg An abnormal resut is a condition in an event monitor with the event being the ordering of the test Curly Braces Problem in Arden Syntax, area within curly braces, eg {get systolic blood pressure}. Local reference data contained within curly braces (fetch data out of target system); must translate into local implementation: a limitation in sharing MLM/CDS across systems Guideline Interchange Format GLIF focuses on more complex multi-part guidelines, including complex clinical pathways GELLO HL7 standard abstract expression language for specifying database queries (from Arden Syntax or others). Based on Object Constraint Language. Requires v3 RIM which has been slow to adopt Shareable Active Guideline Environment SAGE First effort to separate CIS and CDS and recombine using an API: a vMR approach (solves the vocabulary problem) Virtual Medical Record vMR HL7 standard based on the v3.0 RIM specifies canonical kinds of data one might find in EHR so that MLMs written in terms of GELLO queries on the vMR can be translated programmatically into actual queries on patient data at a local institution SEBASTIAN places a standardized interface infront of CDS modules (on internet) ATHENA-CDS Ontology-driven approach that separates static knowledge from problem solving knowledge (integrated into VistA). Driven by EON; ATHENA-CDS created an ontology of CPGs (eligibility criteria, clinical algorithm, guideline drugs) using Prot_g_, like a hierachy of classes in Ruby; the classes are instantiated to define the particular clinical algorithm, the particular guideline drugs etc. Other examples include GLIF, GUIDE, PRODIGY, Proforma, Asbru, GLARE Controlled Terminology a list of terms that are the canonical representations of the concepts; contain relationships between the terms ontology like a controlled terminology: enumerates the important entities and relationships among those entities (taxonomic relationships) as well as partitive relationships.

###### 2.1.4.5 Knowledge management and maintenance
 Knowledge management and maintenance Reliance on EHR patient data; guideline authorship so you can review/update, review patterns of use, commercial vendors, service-oriented architechure (SOA) for plug and play CDS Decision Support Service DSS HL7 standard specification for a service-oriented architecture (SOA) to drive CDS OpenCDS Uses DSS standard, open software tools; robust authoring environment for rules, integration with standard terminologies, can be integrated with other CDS tools such as HL7 infobuttons standard CDS Workgroup reconcile various competing approaches to CDS; vMR is one project, a data model for representing input and output to CDS systems); Terminology management model (CTS2); Formalism for data manipulation in CDS (GELLO); formal method for describing process and work flow; taxonomy of services or actions evoked by guidelines

##### 2.1.5 Legal, ethical and regulatory issues
 Legal, ethical and regulatory issues as long as CDS viewed as open loop (clinician between patient and system), not very regulated. CDS Meaningful Use Requirements Objective 2: must satisfy measure 1 (implement 5 CDS interventions related to 4 or more CQMs) and measure 2 (drug-drug and drug-allergy interaction checks (excluded if EP writes fewer than 100 med orders))

##### 2.1.6 Quality and safety issues
 Quality and safety issues recognized need to view health IT as potentially dangerous

##### 2.1.7 Supporting decisions for populations of patients
 Supporting decisions for populations of patients including used in personal health records

#### 2.2 Evidenced-based Patient Care

 Evidenced-based Patient Care set of tools and disciplined approach to informing clinical decision-making

##### 2.2.1 Evidence sources
 Evidence sources Hierarchy of Evidence (6S, bottom to top): single studies (MEDLINE) - synopses of single studies - syntheses (systematic reviews) - synopses of syntheses - summaries (textbooks, CPGs) - systems/actionable knowledge (guidelines, decision rules, order sets; integrate the lower levels with individual patients) Level of evidence pyramid meta-analysis/systematic review > RCT > cohort > case-control > case reports/series > animal studies Foreground questions asks about specific questions. Essential components (PICO --> PICOTS (time/setting)). Categories of questions : Intervention, Diagnosis, Harm, Prognosis. Results valid? important? applicable to patient care? subquestions? Background questions ask about general knowledge metanalysis and summary statistics metanalysis combines results of multiple similar studies; Odds Ratio for discrete variables, (weighted) Mean Difference for continuous variables; significance level = effect size * size of study Cochrane Collaboration Intl collaboration produces systematic reviews/metanalyses

 Evidence grading ACP smart medicine uses A, B, C (low-quality/anecdotal evidence) (there are many grading schemata) United States Preventive Services Task Froce USPSTF AHRQ program for preventive services, grades A, B, C (selective), D (against), I (insufficient) with certainty levels (high, moderate, low)

##### 2.2.3 Clinical guidelines
 Clinical Practice Guidelines CPG Systematically developed statements to assist practictioner and patient decisions about health care for specific clinical circumstances. "Systems" are evidence used in a CDS Evidenced-based Practice Center EPC AHRQ program, reviews all relevant literature to produce evidence reports including HIT topics National Guideline Clearinghouse NGC AHRQ initiative, bibliographic database provides information on CPGs to further their disemination and implementation; a carefully curated compilation of evidenced-based guidelines Centre for Evidenced-Based Medicine CEBM U. Oxford EBM resource National Institute for Health and Clinical Excellence NICE British system, approximately 100 CPGs are posted, summaries available Digital Guidelines Library DeGeL facilitates gradual conversion of free text guidelines to a formal machine readable language Guideline Element Model GEM ANSI standard; XML-based shared ontology for CPGs (for authors to annotate and identify key elements)

##### 2.2.5 Information retrieval and analysis

###### 2.2.5.1 Search Skills
 Search Skills Google performs an AND search of all terms

#### 2.3 Clinical workflow analysis, Process Resdesign, and Quality Improvement

 Workflow Sequence of physical and mental tasks performed by various people within and between work environments; can occur at several levels: one person; between/among people; across departments, facilities, and/or organizations. Can occur sequentially or simultaneously. Workflow example: medication ordering Communication between provider and patient; clinical provider's thought process; physical action by the provider of writing a paper prescription and having the patient take the rx to the pharmacy, or the physical action by the provider of entering an electronic rx into an EHR and trasmitting the order electronically Workflow analysis evaluation or assessment of a workflow process. An evaluation or assessment of the mental and physical steps of a workflow process, inclduing the order of the steps and the steps that comprise interaction among organizations involved. Evolved from the Scientific Management Movement (Taylor) in the late 1800s Workflow analysis example: medication ordering 1. identify all types of communication in current use 2. analyze the steps and sequence that occur in each type of communication 3. thought processes : identify steps and sequence of steps that provider goes through; ask provider to share information with you 4. identify steps and sequence of steps to execute these processes(prescription), start to finish Recurring Theme 1. Current state 2. Desired future state 3. Make a plan, execute 4. Evaluate outcome and consequences of being there Principles of Workflow Analsysis 1. Goals of the work being studied must be understood 2. purpose and goals of the analysis must be clear 3. analysis of workflow must relate to the work as it is actually performed