Fuzzy Inference Systems for Optimized Drug Dosing in Heart Failure Management

Abstract: The proposed fuzzy inference system uses a number of key clinical variables (i.e. systolic blood pressure, estimated glomerular filtration rate, serum potassium, symptoms of congestive heart failure, overall symptom severity and Nt-Pro BNP burden) to provide an optimal drug dosing solution for patients suffering from heart failure in a logical manner that is understandable to the clinician. It takes various forms of input data and normalizes it; then uses fuzzy logic mapping techniques to map the input data to fuzzy set membership values and apply a Mamdani type fuzzy logic rule base with defuzzification using the centroid method to determine recommended intensity levels for administration of three types of drugs used to treat heart failure (loop diuretics, Angiotensin Receptor-Neprilysin Inhibitors and Mineralocorticoid Receptor Antagonists). Additionally, this system includes a "safety filter" which prevents the generation of potentially unsafe dose recommendations by preventing recommendations that violate certain critical thresholds associated with renal function, elevated serum potassium or hypotension. Thus, although no other studies have been identified that use fuzzy logic for determining optimal drug dosing regimens for treating heart failure, the addition of the safety filter increases the confidence in using this new approach. An example case demonstrating how the fuzzy inference model works is shown by applying it to a previously published clinical case report describing a patient in India who suffered from advanced chronic kidney disease, had experienced multiple episodes of decompensated heart failure and presented with high levels of circulating Nt-Pro BNP. The application of this fuzzy inference system resulted in recommendations to significantly increase the dosage of loop diuretics being administered, but also to begin cautiously administering an Angiotensin Receptor-Nephrilysin Inhibitor at a lower than maximum approved dose due to concerns regarding potential worsening of hyperkalemia and/or hypotension. No recommendation was made to initiate Mineralocorticoid Receptor Antagonist (MRA) therapy. The authors demonstrate that subsequent simulations using follow up data show that if there are improvements in both congestion status and the level of biomarkers (e.g., Nt-Pro BNP), this system may be able to assist clinicians in gradually increasing the dosage of medications while maintaining safety. Ultimately, these results suggest that fuzzy inference systems offer a useful tool for developing clinically meaningful and mathematically flexible approaches to personalize pharmacologic treatment options for individual patients with uncertain conditions.

Keywords: Fuzzy inference system, heart failure management, optimized drug dosing, Mamdani inference, personalized medicine, loop diuretics, ARNI, mineralocorticoid receptor antagonist, clinical decision support, renal safety, NT-proBNP, congestion assessment.