Project Grant F31HL189078

PRECISION PHENOGROUPING AND TREATMENT OF ATRIAL FIBRILLATION IN REAL-WORLD SETTINGS - ABSTRACT ATRIAL FIBRILLATION (AF), CHARACTERIZED BY IRREGULAR AND OFTEN RAPID BEATING OF THE ATRIA, IS THE MOST COMMON CARDIAC ARRHYTHMIA AND A MAJOR PUBLIC HEALTH ISSUE, WITH INCREASING PREVALENCE AND INCIDENCE, SUBSTANTIAL MORBIDITY AND MORTALITY, AND SIGNIFICANT STROKE RISK. MANAGEMENT REMAINS CHALLENGING, OWING LARGELY TO AN INCOMPLETE UNDERSTANDING OF ITS COMPLEX PATHOPHYSIOLOGY AND HETEROGENEITY. CURRENT AF CLASSIFICATIONS INTO STAGES (I.E., PAROXYSMAL, PERSISTENT, AND PERMANENT) ARE BASED PRIMARILY ON EPISODE FREQUENCY AND DURATION, CAPTURING A NARROW DIMENSION OF THE DISEASE. THIS APPROACH FAILS TO ACCOUNT FOR THE CLINICAL HETEROGENEITY SEEN IN REAL-WORLD PRACTICE, POTENTIALLY LEADING TO SUBOPTIMAL CARE. PATIENTS WITH AF VARY WIDELY IN COMORBIDITIES, CARDIAC STRUCTURE AND FUNCTION, DISEASE PROGRESSION, OUTCOME RISKS, AND TREATMENT RESPONSES, YET CURRENT GUIDELINES DO NOT INTEGRATE THESE FACTORS IN MANAGEMENT. RECENT EFFORTS HAVE SHIFTED TOWARDS THE IDENTIFICATION OF AF PHENOGROUPS, GROUPS OF PATIENTS WITH SHARED CHARACTERISTICS AND PHENOTYPES, THAT MAY HAVE DIFFERENTIAL PATTERNS OF PROGRESSION AND TREATMENT RESPONSES. ADDITIONALLY, ADVANCEMENTS IN DATA SCIENCE AND THE GROWING AVAILABILITY OF LARGE-SCALE DATA HAVE ENABLED THE APPLICATION OF MACHINE LEARNING (ML) APPROACHES TO REVEAL PREVIOUSLY UNDISCOVERED INSIGHTS. THEREFORE, THIS RESEARCH AIMS TO ADVANCE PRECISION MEDICINE IN AF TOWARD A PHENOGROUP-DRIVEN APPROACH BY INTEGRATING ML WITH REAL-WORLD DATA. THE PROPOSED RESEARCH WILL LEVERAGE THE LARGE, LONGITUDINAL MERATIVE MARKETSCAN DATABASES OF PATIENT-LEVEL MEDICAL CLAIMS WITH >600,000 AF PATIENTS TO ADDRESS THREE SPECIFIC AIMS. IN AIM 1, TO DISCOVER AND CHARACTERIZE DISTINCT, CLINICALLY RELEVANT AF PHENOGROUPS, FOUR UNSUPERVISED ML CLUSTERING TECHNIQUES WILL BE EVALUATED. A SUPERVISED ML APPROACH WILL IDENTIFY THE MOST INFORMATIVE CLINICAL FACTORS ASSOCIATED WITH PHENOGROUP MEMBERSHIP. IN AIM 2, TO ESTABLISH THE DIFFERENTIAL PROGNOSES OF THE DISCOVERED PHENOGROUPS, TIME-TO-EVENT ANALYSES WILL BE EMPLOYED. MULTI-STATE MODELS WILL MODEL TRANSITION HAZARDS FOR TWO DISTINCT PROGRESSION PATHWAYS (I.E., STRUCTURAL/THROMBOEMBOLIC AND ISCHEMIC/ATHEROSCLEROTIC). COX PROPORTIONAL HAZARDS MODELS WILL ESTIMATE DIFFERENCES IN TIME TO FIRST INCIDENT OUTCOMES, AND NEGATIVE BINOMIAL REGRESSION WILL ESTIMATE RATES OF HEALTHCARE UTILIZATION. IN AIM 3, TO EVALUATE THE COMPARATIVE EFFECTIVENESS OF ANTICOAGULANT THERAPIES ON STROKE-RELATED OUTCOMES ACROSS PHENOGROUPS, THE TARGET TRIAL FRAMEWORK WILL BE USED TO COMPARE DIRECT ORAL ANTICOAGULANTS AGAINST VITAMIN K ANTAGONISTS ON STROKE AND STROKE-RELATED OUTCOMES. CAUSAL CONTRASTS WILL BE ESTIMATED USING TARGETED MAXIMUM LIKELIHOOD ESTIMATION WITH A SUPERLEARNER LIBRARY TO PROVIDE ROBUST ESTIMATES OF PERSONALIZED TREATMENT EFFECTS. IT IS HYPOTHESIZED THAT ML WILL REVEAL NOVEL AF PHENOGROUPS THAT EXHIBIT DIFFERENTIAL DISEASE PROGRESSION AND RESPONSES TO ANTICOAGULANT THERAPY. THE RESULTS WILL PROVIDE A DATA-DRIVEN APPROACH THAT CAPTURES THE HETEROGENEITY OF AF AND ESTABLISH EVIDENCE-BASED, PHENOGROUP-SPECIFIC TREATMENT RECOMMENDATIONS, ACCELERATING THE TRANSLATION OF PRECISION MEDICINE TO CARDIOVASCULAR PRACTICE.