clinical decision support systems (CDSS)

see also:

• Emergency Department Information Systems (EDIS) for Australia
• clinical decision making

• clinical decision support systems (CDSS) are systems which assist clinicians in making clinical decisions.
• whilst in the past, such systems relied on paper-based systems such as books, policy manuals, etc or micro-film such as PoisIndex poisons management database, the majority are now electronic.
• historically, Emergency Department Information Systems (EDIS) for Australia were designed primarily to capture data for governments to monitor patient throughput and complexity and thus funding arrangements. In Australia in the mid-1990's, these systems evolved to provide more patient flow management assistance to ED staff by allowing staff to more readily track patient care, determine who should be seen next, allocate staff to see them, and discharge them from the ED. Gradually paper-based hospital policy and guidelines became available on hospital intranets to make them more readily available, although the were not always easily searchable, relevant or kept up to date as business rules and clinical management changed over time.
• many attempts have been made to create computerised clinical diagnostic software with the hopeful aim of junior staff being able to enter various patient characteristics, symptoms, signs and investigations, and out pops the most likely diagnosis. Unfortunately, given the sheer complexity of variation in disease prevalence depending upon race, age, sex, co-morbidities, culture, and our relatively poor accuracy of our epidemiologic data for sub-groups of populations, such tools have not been of great value for the care of individual patients. It probably does not help that a program will give the final diagnosis in its top 5 likely diagnoses 50% of the time¹⁾ - any reasonable clinician could probably achieve the same without the software. Nevertheless, for certain conditions, such tools can be of value.
• furthermore, if the CDSS is complex, requires substantial clinician data entry, or is not readily accessible, it is unlikely to be used.
• despite this, there are many levels upon which CDSS can be of use to make the clinician more efficient and perhaps more accurate, and with less adverse decisions:
  • rapid access to relevant guidelines (this wiki for example) which can act as an aide memoire to jog our memory to re-confirm what we think we know and to provide us with a checklist for other things to watch out for, a data repository or even just as links to other resources.
  • online searchable atlases or “visual diagnostic decision support systems (VDDSS)” to assist in diagnoses of skin conditions such as VisualDx
  • rapid access to pharmacologic information for drug doses, etc.
  • clinical calculators - particularly important for paediatrics, drug infusions and the renal impaired
  • rapid access to important patient information such as past discharge summaries, management plans, previous investigation results including ECG (eg. the classic question - is this LBBB old or new?), documented allergies, medications and past illnesses.
  • electronic prescribing which cross checks drug groups for drug interactions or known allergies, whilst ensuring appropriate drug dosing, particularly if it has access to renal function results, patient age and weight.

• “genius diagnosticians make great stories, but they don't make great healthcare - the idea is to make diagnosis accuracy reliable, not heroic” - Don Beswick.

• this is a rules based probabilistic approach using probability trees based on Baye's Theorem and can provide a rank of probabilities of provisional diagnoses IF the underlying prevalence and probability data required in each branch of the tree is known.
• such decision trees can be effective in limited domains but will not detect new disease patterns and may not be transferable from one culture or race to another, and the amount of knowledge required for them to work in wider domains such as general medicine conditions is generally too onerous and complex.

• general approach:
  1. for each possible illness (hypothetical diagnosis) a response that the patient presents is identified (symptom, sign or findings of complementary examination), and it is evaluated whether it is part of the disease;
  2. scores are established for each illness, according to the number of symptoms that are the same as those exhibited by the patient;
  3. the illnesses are classified according to these scores;
  4. inquiries are made about whether after findings of the illnesses with the highest score are also found in the patients;
  5. repeat steps 1 and 2;
  6. repeat the procedure for following illnesses
• problems with the above approach:
  • how strongly is a finding a reflection of a certain disease
  • how strongly does the absence of a finding eliminate the possibility of a particular illness?
  • which is the prevailing rate of each of the hypotheses in the population under study
  • how strongly does a finding in the patient but not present in the hypothesis invalidate the diagnosis
  • the patient presents with only one or more than one illness
  • if there is more than one illness, would the illness be related
• Isabel system:
  • online system available for annual subscription for doctors - currently $A450pa
  • you input age, sex, pregnancy status, race, and clinical features and it returns conditions that match those inputs but NOT in probability of likelihood of actual diagnosis, thus it is more to jog your memory and provide you with an optimised mechanism of gaining more data.
  • the authors²⁾ of the Isabel system claim that in a large adult ER study of 494 cases, it displayed the final diagnosis in 95% of cases and “must-not-miss” diagnoses 90% of the time, although it is not clear how many false diagnoses were presented for each case.
  • “The Isabel system is comprised of two components: the Isabel Diagnosis Reminder System (IDRS) and the Isabel Knowledge Mobilizing System (IKMS). Given a patient’s clinical features, Isabel searches a database of more than 11,000 diagnoses and 4,000 drugs to provide clinicians with a checklist of likely diagnoses and/or drugs that may be causing a patient’s symptoms, as well as additional disease specific knowledge.”

• examples:
  • Internist, Qmr, Dx Plain, Iliad, Consultor ³⁾
• these still only give the actual diagnosis in the top 10 diagnoses 50% of the time, and this probably does not make that much difference to clinical decision making for an individual patient, although it may alert the clinician to some differentials not thought of.

• software that is able to “learn” from past patterns and can incorporate new data elements
• often use “neural networks” - computer architecture designed to simulate the way the brain works