2009 Implementers Group Meeting Program Decision Support

Owned by Former user (Deleted)
Last updated: Aug 18, 2010Version comment
5 min read

Venue: Tree Tops

Extremely Rough Notes

These notes may or may not get edited.

large space based on logic service (ARDEN Syntax) which became usable in that last here. used by Paul's CHICA (?) System Martin (AMPATH) is working on taking critical summarires and driving them through the logic service and provide clinicians with information Paul: Clinical reminders are transformations of data eg: Hemogloben of 8, MCV of 5(?) for a child means iron deficient Reminders can be string of text, other values or a combination of both Burke: Goals are to allow the system to treat Birdge the gap between computers and clinicians/researchers Answers to real questions are almost always derived concepts Is asmathic? answers: asmatic diagnosis? been to astma clinic? are they on treatment? peak flow below threashold? Idea: define a derived concept which encapsulates the rules Make derived concepts as available as standard concpets Make it possible for non-programmers to define rules Diagnosis suggestions Burke: Decision support can be thought of broader as just logic service Aim to help the users along the way HTML form entry validation is a form of decision support Conflicting prescriptions (counter indicaitons) alerts Prescription suggestions is a form of decision support (makes things easier) e.g. In Burke's hospital the changed a default dosage of Antibiotics from twice to once a day when the literature changed, and had the desired effect Decision support also means prividing the correct options to guide care practices Need to think about how we maximize our effect Martin (clinician) about changing behaviour in Eldoret: Use systems to collect data Has not happened: clinicians getting the data back -> They want to see the information as a flow sheet or clinical summary Challenge: Show data to clinical !! (The report vertically fine) Next step to use data to improve care Look at standards and protocols and use system data to provide alerts e.g. Haven't got CD4 could, prescription contra indications, various logic rules REASON: Mostly clinical officers and NOT clinitians, so decision support WILL improve care AMPATH: Coded deision rules in java Found -> rules aren't followed e.g. no CD4 count for many patients (or old CD4 counts) Privided clinical summaries based on hard coded rules Included alerts when from 35% complience went up to 60-70% bring providers to the same leve Easier to hard code for a few rules for 200 rules not so easy? and there are many rules that could be applied to improve care Thinks can prove to the world to increase patient outcomes, and reduce the cost of care and increase quality of care even for non-clinitians Need to decouple from concept_id's to make more robust and reusable Paul: Martins work is part of clinical trial Now have full-time developer (Win?) working on logic service Challenges: when clinicians approach developer to write a rule they think about coarse idea's seems on the surface to be simple, but defining things depend on how ppl practive medicing e.g. finding ppl who are HIV+ may be difficult could have a diagnosis, could be based on treatment, might be assumed for everyone (e.g. in an HIV clinic) necessary to consider the complexities need a way to document within the system Paul has a grant and other funding to improve logic service, with is being used in production Making it easy to write the rules is difficult Don't want clinicians to have to learn Java ARDEN is a language specifically used to represent medical logic rules Collegue made a parser that converts ARDEN to Java Roger Friedman: [CDC] Used to having well-defined cases and dont want fuzzy case logic Paul: Need to match logic to vagueries of how people document medical care AMPATH: Clinitians taking care of 60 patients in a half-day Difficult to document acurately always : high burden Have to make a assumptions have vagueries Trying to extract clinical knowledge is where rubber hits the road Similar problem for indicators -> rule is written as OR's (e.g. for ppl on antiretrovals) Would like to use rules to compensate for human recording bias Burke: "Life is hard" - Burke's mentor Clem Simple rule: every woman over 45 should have a mamogram has complexities e.g. had mamogram in external system, had bilateral masectomy, only has 3 months to live Have to accept things don't happen according to plan So we are trying to define the rules and apply as best as possible and realising that sometimes there are going to be exceptions with the goal of providing clinicians with useful information Loop will be self-correcting because if rules are broken, hopefully the pattern on care wil be correct in the next iteration Chris: Interested in how treatment failure will be affected by this Based on clinicians decisions to change regimens Requesting whether decision support speaks to this (genotyping)? Martin: Thinks logic service should help by alowing you to write rules to define criteria for treatment failure Logic should be generic framework Can use every piece of data stored in the system (adherence, drugs, labs, etc) You can use this data to write the ruels Paul: Key for genotypic inforation: get into the database like other data Then write rules taking this into account Chris: Can you do data mining? Paul: Maching Learning? Chris: Ben building SVM to look at outcomes and genotypic information Paul: Dont have machin learning, but can use ouput of something like that Isaac: Asked about statistical approach, maching learning? Evan: Was asked about Boabab and Mateme for (NDH and PIH) Knew workflow of creating a new treatment takes long So they asked clinicans to link treatments to diagnosis Show list or treatments ordered by frequency Martin: Frequency is good in some cases -- unless a lot of them are doing it wrong Evan: Want to be able to remove things from list Want to be able to move to something that is editorialised ?: OMRS focused on person level Epidimiology looks are person, place, time, agent Family doctor able to make decision while also consider context (e.g. context) Asked how context information can be used to assist decision support to make brige between family doctor style and mass care Burke: First step is to get the necessary data interview with social worker or nurse to get context PRojects in place to get aggregate data to systems that are good at epidemiological analysis ?: So can logic serive do this stuff with the data? Paul: Local context has large influence on how health-care is practiced Local variance is key to how things are done Rules are difficult to generalize outright Local tweaks will NEED to be done for treatment success Thats why local capacity is important Hamish: 1. Machine LEarning Working on collecting data on TB drug resistance Will be doing machine learning Will include geographical and sociological factors Illustrates overlap of clinical treatment and other factors Challenging to collect data thats why nurese or social workers are used 2. When buiding DS systems Subtle art to scale up from a few rules to many rules ppl get annoyed with lots of rules and then ignore them ?: If we have default concepts can we get nice trusted rules by default Paul: Thought about this if ppl share conpepts, derived concepts will work higher level rules will be human understood (hiv positive) will still have to add local context will have to define where customization needs to be done even if tweaking is needed, still worth sharing sharing is good because it improve the generality of the ruels MAtrin: Problem of sharing rules and reminders is pretty international There are funded projects which try to come up with ways to share ruels and reminders we would like to use those standed carol(?): we need a holistic view contextual factors are critical to have usefule DS system OMRS value: access to the knowledge need transfer across languages transfer data through language and DS we will see clinical care improvement DS can be confusing will be nice to get help from DS from developed world Darius: CAn we hide the complexity? can WHO (or others) help define a few crucial rules Burke: We are trying to do that, and also with concept (OCC) and other standardisation attempts using collaboration and crowd sourcing sharing is needed when creating forms or concepts, first see what exists and try and use it rules should folllow Andy: this discusiion is about symantic interoperability OCC allows us to define internal standard Using reference maps (SNOMED) need to be used for external interoperability ?: Philedelphia uses DS to provide clinical feedback can be aggregated up chris bailey: last year talked about convening a meeting about clinical ontologies should consider it this year to make sure what we're doing is aligned with other efforts