Rethinking Clinical Decision Support

AkēLex’s Approach to Clinical Decision Support and Predictive Analytics: A Primer

You’re familiar with the problem: U.S. health care costs are rising exponentially while wellness, prevention, disease management efforts, and life expectancy trail dozens of nations. Physician shortages, unmanageable volumes of medical knowledge, and limited access to health care resources compound the quality-cost problem.

Expert systems are a natural fit to help address these challenges, with recent breakthroughs showcasing their viability. AkēLex has taken a unique approach to developing a clinical decision support system (CDSS) and predictive analytics technology that—after a decade in development—is working today.

Central to AkēLex’s expert system is the Adaptive Knowledge Engine (AKE℗), a platform agnostic, nonlinear decision engine. With its modular data ontology (Lexicon) and Context Parsing Engine, The AKE supports the creation of a tailored CDSS for any functional need, from care delivery to medical simulation to medical device applications.

An extensive investigation into the characteristics of an ideal CDSS formed the basis of the AKE’s design. That design is intended to meet a dozen pragmatic clinical requirements:

1. Recognize and manage multiple simultaneous problems or conditions.

AkēLex’s Viewpoint: Complex medical histories are not single source problems.

Example: Patients with diabetes often suffer from concurrent kidney or peripheral vascular disease. Recognizing and managing these conditions in combination or complications arising from this combination is vital.

2. Tailor responses and recommendations to the specific needs or medical literacy of the user.

AkēLex’s Viewpoint: Architecture should allow the language and the selection of content to be tailored, based upon user role and skill set.

Example: An elderly man lives with his daughter who is his caregiver. She has no knowledge of dementia diagnostic criteria but is concerned about her father’s declining memory and subtle personality change. She is presented with an evolving set of simple questions in plain English that allows her to identify needed next steps for her father. The same information, if shared via the cloud with a care provider, would be presented in clinical language without translation.

3. Allow users to input real-time data in any order.

AkēLex’s Viewpoint: An advanced CDSS should eliminate data and search order bias and not require users to conduct their assessments using a fixed template.

Example: A patient is referred to a care team based on abnormal outside screening tests. These test results are initially entered and can be used to guide a medical assistant who is now collecting the patient’s history. The history is entered in any order in which it is presented and, where appropriate, can direct the collection of additional tests. These can be ordered and resulted without having to wait for the physician’s physical examination.

4. Perform simultaneous qualitative and quantitative assessments.

AkēLex’s Viewpoint: An advanced CDSS should combine qualitative and quantitative assessments. Pure qualitative assessment tends to over-emphasize rare conditions (zebras) in assessments, which adds static to the average assessment. Pure statistical assessments tend to miss the unusual presentation of common ailments or an unusual case. The combination of methods better mimics expert diagnostic assessment.

Example: A sixty-five year old woman has had increasing fatigue over the past two months. While the tool recognizes that fatigue may be the presenting feature of diabetes or Epstein Barr Virus in a younger patient, it also recognizes that since the patient is older, it is more likely to be heart disease. The system can then combine the qualitative findings of cardiac risk factors, characteristics of current symptoms, and EKG interpretation to provide a quantitative assessment of the likelihood of unstable angina in this specific subset of the general population.

5. Coach a user to identify and gather, without bias, the most useful missing data, and generate refinement questions that aid the user in determining the correct diagnosis or most appropriate next steps.

AkēLex’s Viewpoint: An advanced CDSS should support real-time documentation and generate questions to the user that are presented so that the user does not know the purpose or intent of the question (avoiding bias), or positive/negative predictive value of the question.

Example: A teenager would be led through a series of questions, (mood, drug use, bullying, gender identity, for example) to identify the risk of suicide or depression and, once establishing that as a possible concern, the user would be instructed on next best actions to obtain help and be kept safe.

6. Integrate with an existing electronic health record to utilize a patient’s complete medical history in all assessments and coordinate an evaluation requiring multiple patient encounters.

AkēLex’s Viewpoint: An advanced CDSS should integrate with existing medical records. Lifecom’s system time-date stamps all patient data so every relevant encounter is evaluated in the context of the patient’s total history.

Example: A woman who has not been to the doctor for many years has recently had a positive home pregnancy test. She now develops abdominal pain. Five years ago she had several episodes of pelvic inflammatory disease. The system recognizes this as a risk factor for an ectopic pregnancy and alerts the woman of the need for immediate medical care.

7. Recognize critical clinical patterns that may only be revealed or properly considered when viewed in a temporal context.

AkēLex’s Viewpoint: An advanced CDSS should be configurable to perform any trend analysis including complex multistep analyses that incorporate data from disparate sources.

Example: A single glucometer reading for a Type II diabetes patient can be incorporated into an array of blood glucose values, Hgb A1c’s, patient medications, etc. to test for subtle trends in the data that might suggest an emerging problem.

8. Utilize both positive and negative data, as well as differentiate between questions that are unknown from questions that are unanswered, in performing assessments.

AkēLex’s Viewpoint: Recognizing that negative data is often the most predictive information (and the most often ignored by clinicians), a decision support systems should integrate negative data into clinical pattern recognition to aid in diagnosis and management of disease. Distinguishing the unknown from the unanswered aids in generating the AKE’s refinement questions.

Example: A young woman presents to an urgent care center after returning from a camping trip with a fever and some anorexia. She has some mild low abdominal tenderness and white blood cells in her urine. With fever and urine that looks infected, the clinician is tempted to diagnose the patient with a kidney infection; however the negative presence of flank tenderness makes this diagnosis much less likely than appendicitis with the inflamed appendix lying on the bladder.

9. Update content rapidly and seamlessly in light of new medical knowledge and apply it across the entire patient population to identify those patients whose condition or medical management is affected.

AkēLex’s Viewpoint: Decision support content should be completely modular and not integral to a platform’s architecture. Individual and grouped content should be publishable to the application at any time required.

Example: Resistance patterns to influenza medication can change seasonally and can differ by geographic region. Lifecom can update CDC recommendations for treatment in real time as soon as they are released, and can tailor treatment recommendations to the user’s specific geographic location. Lifecom can modify the content within the database and publish a revised knowledge repository to update all field applications, which would alert users to the new recommendations. The AKE’s data files would also be updated with the new guidelines. Finally, this would provide a mechanism to identify all patients treated with the obsolete guidelines.

10. Provide immediate, actionable recommendations (drug therapy, additional testing, triage to higher level of care, etc.) even when there is no definitive diagnosis.

AkēLex’s Viewpoint: An advanced CDSS should provide diagnostic management, triage, and workflow support.

Example: Parents have a three year old with a high fever. The CDSS-directed interview assesses the potential severity of the child’s illness and alerts the parents of what to do next (ibuprofen and fluids, pediatrician follow up the next day, go to the emergency room now, etc.), regardless of whether a specific diagnosis is made.

11. Provide both CDSS and educational simulation.

AkēLex’s Viewpoint: An advanced CDSS should support both CDSS and simulation training without modification.

Example: The clinical team evaluates a case of an aortic dissection complicated by an acute myocardial infarction. The stored digital record of this case is stripped of patient identifying information and included in a library of educational cases. The file is later loaded into the system in simulation mode and used to test resident physician decision-making.

12. Support process improvement through improving the quality of clinical inputs to the medical record.

AkēLex’s Viewpoint: An advanced CDSS should store assessments in a highly granular format that preserves the precise sequence of assessment, data capture, analysis, and conclusion, making it an ideal process improvement platform.

Example: An improvement team is tasked with improving the efficiency of routine diabetes assessment. The team uses the AKE to review a series of anonymous assessments and compares the workflow in each. Once a superior track is identified, the AKE uses revised content to support the improved process.

To date decision support systems have been utilized with varying success (Watson, PKC, Isabel, etc.). However none is particularly effective in practice because of their common limitations and lack of integration within existing workflow. AkēLex’s AKE was developed to overcome these deficiencies. It was built to reside in a distributed or cloud based system; however, it can be installed as a fully functional native application (PC, tablet, smart phone, etc.) that operates independent of a network connection.

In practice, this enables caregivers at every level to overcome knowledge gaps and elevate the scope of their practice. This kind of expertise delivered to the point of care doesn’t solve every problem with the cost and quality of care, but it’s certainly an enormous step in the right direction.