Optimizing Healthcare

The science of medicine creeps forward in its petty pace almost entirely through peer-reviewed studies, based predominantly upon controlled clinical research…with a little help from retrospective analyses and anecdotal evidence.?The controlled studies are expensive, narrowly focused, inadequate to encompass all clinical conditions, and typically targeted by the perspectives or whims of individual researchers.?Since the practice of medicine relies on the science of medicine, this approach results in suboptimal healthcare.

There is a wealth of valuable data, on all medical problems, lurking in electronic health records (EHRs) across the world.?These big data are not being systematically analyzed.?Therefore, we are not learning from ongoing clinical experiences.?This is unfortunate, at the very least, and it is arguably the primary reason that many healthcare interventions are not as effective as they could be.?Cost is out of control, and the quality is in no way commensurate with these excessive and inexorably increasing costs. ?In 1960, healthcare accounted for 5% of our GDP; in 2020, it was 19.7%.?

The only way to reverse this disturbing trend, so that we can bring about the quality and cost effectiveness that we would all like to see, is to launch a Continuous Quality Improvement (CQI) process.?This starts off by analyzing real-life clinical encounters.?Figure out what are the early signs of serious diagnoses.?Rigorously determine what treatments have worked best for various conditions and under what circumstances.?Then we can combine this new knowledge with what medical science currently knows, to develop a Clinical Guidance System.?The CGS would be functional at the point of care, providing diagnostic and treatment advice to caregivers – based largely upon the analysis of how things turned out for similar patient conditions in the past.

The CGS is more than a clinical decision support system.?It provides proactive advice at each decision point, with supporting data.?The clinicians still make the decisions, overriding CGS guidance if their “art of medicine” so dictates.?But the results of all decisions – whether in concert or conflict with the CGS – are tracked, so that the decision algorithms can be refined on an ongoing basis.?Machine learning.?CQI for healthcare.?It is a pretty picture…and so desperately needed.

How would we develop the CGS??Artificial Intelligence now provides the capability to construct and continually refine this powerful healthcare tool.?We can apply AI to an enormous database of symptoms, findings, diagnoses, management plans, and outcomes.?When a patient presents with a clinical problem, the history-of-present-illness and physical-exam data are routinely entered into the EHR.?The AI system will search the database to identify other patients whose presenting information was most similar.?It would also find what the ultimate, definitive diagnosis was for those patients – displaying those diagnoses that thus seem most likely for the current patient, based upon the available clinical data. ?It would also show what lab, imaging, and other tests were pivotal in arriving at those diagnoses.?

Once the diagnosis is established for a patient, the CGS transitions to assisting with the management plan.?It extracts from the database the past patients with the same diagnosis and with the most similar associated characteristics – age, gender, symptoms, findings, other medical problems, etc.?The CGS segments them by the various treatment plans that were administered, and it assesses the outcome for each of those episodes.?It would then display the relative efficacy for each of the alternative treatments in the population of “similar” patients and circumstances.?That will help the clinician determine what plan is most likely to be successful for the current patient.

Let us get a glimpse of one way the diagnostic-assistance component of the CGS could work, with just a single example.?We might look at all the presenting data for patients with the chief complaint of “headache.”?Separate out those patients who were eventually and definitively diagnosed with brain tumor (or, maybe more specifically, glioblastoma).?Examine the differential in symptoms and signs between that group and all the other patients with headache as their chief complaint.?AI could determine the coefficient on each symptom/sign that would maximize the discriminatory power of the resulting algorithm to identify a brain tumor.?Seizures might get a large positive coefficient, since they are highly associated with brain tumors.?Throbbing aggravated by motion might get a negative coefficient, since that is associated more with migraines and would make brain tumor less likely.?We would also include coefficients for the symptoms/signs that are not present, since they need to factor into the equation.

When a new patient presents with a headache, the CGS applies the AI-determined coefficients for brain tumor to each of that patient’s symptoms/signs, running the equation to come up with a total number.?The higher the number, the more likely it is a brain tumor.?The equation would be processed for migraines and other headache-related diagnoses, as well.?The clinician would be presented with these results, perhaps just for the several highest totals, providing empirically developed guidance for enhancing diagnostic expertise.?A similar operation would take place for the guidance related to treatments, by correlating them to outcomes.

Note that this description of the CGS is overly simplistic; it is presented just to provide an overall sense of how a system like this could be constructed and utilized.?

The clinicians, empowered by this tool, will still have the final say at all decision points.?What they actually decide is naturally added to the database, along with the eventual results, to enable dynamic enhancement in effectiveness of the AI-driven CGS.?That is the beauty of continuous quality improvement:?Enabling patient care to embody higher quality with each passing day.

The data and technology are out there.?Let us put them together.?The time has come to make healthcare all that it can be. #healthcare

During his several decades in the healthcare industry, Joe Weber has been engaged in medical research, hospital administration, consulting, and marketing.?He has published and presented extensively on a variety of healthcare topics, mostly related to electronic health records.?He has been CEO or Chief Marketing Officer for health systems and medical device companies, having served as Associate Director for Ambulatory Care at Cook County Hospital early in his career.?Joe is the holder of 3 U.S. patents, including the invention of predictive typing, also known as automatic word completion (Patent No. 5,305,205, filed in 1990).?He has a BA in Biology from Brandeis, an MS in Biostatistics from Columbia, and an MS in Management from M.I.T.?His master’s thesis at MIT was “Computerized Medical Records.”?