Machine-learning and AI are transforming every industry. Recently, Alphabet’s DeepMind demonstrated the power of machine-learning by beating the world Go Champion. While games prove a point, dollars drive innovation. Medicine is a major focus. Congestive Heart Failure (Heart failure) treatment is a prime target. It is labor intensive, costly, and afflicts 120 million worldwide. In the US, the costs exceed 100+ billion.
Heart failure is a chronic disease of the circulatory system. A weakened pump deprives the body of oxygen; blood, pooled in the lungs, suffocates while a diminished oxygen supply destroys organs.
Improving outcomes is the key to decreasing costs. Improving diagnoses improves outcomes. Diagnoses relies on diagnosticians possessing wide variation in skills and training. As one might expect, skill variations produce outcome variations. The diagnostic methodology is a complex assimilation of dozens of patient measurements into specific medication recommendations. The objective of these treatments is reduced strain on the heart to increase blood flow.
A system dependent on individual skill is a flawed system. Machine Learning is the answer.
This simplicity of structure gives confidence in this design and solution. It is the design capable of accommodating automated diagnoses.
Cardiolert Systems (Cardiolert) designed a methodology and data structure for capturing diagnostic logic. A single data variable – the patient measurement - drives the diagnostic logic. Simplicity is the handmaiden of feasibility giving confidence in the design. The method defines the diagnosticians’ reasoning in detail and stores the thinking with the results (outcomes). The captured logic is available to process against future sets of patient measurements for advising diagnosticians and eliminating manual “re-processing” of defined reasoning. This empirical cycle of care, “test-treat-retest”, data is the experience feeding deep learning software. Processing past and present datasets identifies correlatives to positive and negative outcomes generating evidence-based inferences for a variety of qualitative uses including practitioner assessment, treatment evaluation, and outcome quality validation.
The data structures support application of machine-learning logic. Cardiolert reviewed the design with a machine learning consulting practice’s chief scientist who concluded “…this problem is “shaped” like a typical predictive machine-learning problem…. With sufficient diagnoses and treatment cycles to feed a machine-learning engine, this design automates significant portions (if not all) of the diagnostic effort.” The Bio-Informatics Department at Arizona State University has validated the design and is collaborating on construction of the machine-learning engine.
The is an “Uber” “Platform” containing the diagnostic methodology and necessary medical data linking the “Crowd” – patient and diagnostician. Access to the “Crowd” solicits diagnostic insights from anyone worldwide. Patients will access for the best care and diagnosticians will access accumulated knowledge of the Crowd and deliver diagnostic insights. The “test-treat-retest” data will fuel a machine learning engine for automating the diagnostic process. Rapid accumulation of treatment histories will accelerate the precision of machine-learning diagnoses.
This implementation is beyond anything conceived today - a monolithic method for defining diagnoses and opening diagnostic methods and outcomes worldwide. Concentrating the best diagnostic logic into a single repository creates the capability to offer the best treatment available at that moment to the next patient seeking care worldwide. (Think about the implication.)
The majority of heart failure victims live in less developed countries where skilled diagnoses are nonexistent. Delivery of the very best care to deprived populations is a philanthropic and financial benefit. India, for example, has the highest heart failure prevalence of any country.
Affordable, improved outcomes will drive adoption.
Cardiolert is seeking funding to complete the proof-of-concept and build a deployable model.
Cardiolert is seeking collaborators and funding to build and deploy this solution.