Considerations for addressing bias in artificial intelligence for health equity.

Health equity is a primary goal of healthcare stakeholders: patients and their advocacy groups, clinicians, other providers and their professional societies, bioethicists, payors and value based care organizations, regulatory agencies, legislators, and creators of artificial intelligence/machine learning (AI/ML)-enabled medical devices. Lack of equitable access to diagnosis and treatment may be improved through new digital health technologies, especially AI/ML, but these may also exacerbate disparities, depending on how bias is addressed. We propose an expanded Total Product Lifecycle (TPLC) framework for healthcare AI/ML, describing the sources and impacts of undesirable bias in AI/ML systems in each phase, how these can be analyzed using appropriate metrics, and how they can be potentially mitigated. The goal of these "Considerations" is to educate stakeholders on how potential AI/ML bias may impact healthcare outcomes and how to identify and mitigate inequities; to initiate a discussion between stakeholders on these issues, in order to ensure health equity along the expanded AI/ML TPLC framework, and ultimately, better health outcomes for all.

Longevity of Sprint Fidelis implantable cardioverter-defibrillator leads and risk factors for failure: implications for patient management.

BACKGROUND: Sprint Fidelis (Fidelis) implantable cardioverter-defibrillator leads are prone to fractures that have caused adverse events, primarily inappropriate shocks, and a few reported deaths. More than 100 000 patients have Fidelis leads. No independent multicenter long-term performance information exists for this lead, and single-center studies suggest that certain patients are at increased risk for Fidelis failure. Our aim was to assess the longevity of Fidelis leads and to evaluate clinical variables and adverse events associated with Fidelis failure. The results were compared with like data for Quattro Secure (Quattro) implantable cardioverter-defibrillator leads implanted at our centers. METHODS AND RESULTS: This 3-center study included adults ≥18 years of age who received Fidelis or Quattro leads for the prevention of sudden cardiac death. From November 2001 to January 2009,1023 Fidelis and 1668 Quattro leads were implanted and followed up. The failure rate for Fidelis leads was 2.81%/y compared with 0.43%/y for Quattro leads (P<0.0001). No deaths or injuries occurred as a result of lead failure, but 42% of fractures caused inappropriate shocks. The survival of Fidelis leads at 4 years was 87.0% (95% confidence interval, 83.6 to 90.1) compared with 98.7% (95% confidence interval, 97.9 to 99.4) for Quattro leads (P<0.0001). Multivariate predictors of Fidelis failure were younger age (hazard ratio, 0.98; 95% confidence interval, 0.96 to 0.99), female gender (hazard ratio, 0.61; 95% confidence interval, 0.40 to 1.00), and cardiac disease (P=0.041). CONCLUSIONS: Compared with Quattro leads, the survival of Fidelis leads continues to decline, and Fidelis failure is notably higher in younger patients, women, individuals with hypertrophic cardiomyopathy, and patients with arrhythmogenic right ventricular dysplasia or channelopathies. These findings have significant implications for the management of patients who have Fidelis leads, and they demonstrate the importance of weighing clinical variables in assessments of implantable cardioverter-defibrillator lead performance.

An unusual cause of abnormal intrathoracic impedance in a patient with arrhythmogenic right ventricular cardiomyopathy.

A device interrogation for a 32-year-old woman with arrhythmogenic right ventricular cardiomyopathy and an implantable cardioverter-defibrillator capable of monitoring intrathoracic impedance demonstrated a dramatic increase in her OptiVol Fluid index levels in the absence of clinical heart failure. The timing of her intrathoracic impedance changes correlated with plasma volume changes during an uneventful pregnancy without cardiovascular complications. The strengths and limitations of intrathoracic impedance are described in the context of the physiology of pregnancy.

Managing financial and nonfinancial conflicts of interest in healthcare delivery.

The issue of conflicts of interest held by physicians and other caregivers has drawn increased attention in recent years. In response, some commentators and institutions have proposed policies intended to protect patients' rights, including disclosure of relationships with for-profit companies. Conflicts of interest policies must balance the need for transparency and patient safety, but must also take into account the positive aspects of such relationships, including innovative product development. This article discusses various types of conflicts that can and do occur in healthcare delivery and potential policy options that may address these issues. The case history, device, physician, and patient described below do not represent actual people or events, but are intended to be emblematic of the challenges faced in the current medical marketplace.

Premarket clinical evaluation of novel cardiovascular devices: quality analysis of premarket clinical studies submitted to the Food and Drug Administration 2000-2007.

The quality of clinical data submitted by manufacturers to support Food and Drug Administration cardiovascular device premarket approval (PMA) applications varies widely and formal quality assessment has not been previously performed. This study evaluated all original cardiovascular device PMAs with Food and Drug Administration decisions between January 1, 2000, and December 31, 2007, to assess the quality of clinical investigations submitted by manufacturers. Effectiveness and safety end points were judged high quality if they were clearly defined and associated with a specific time point for analysis. Subject accounting was high quality if 90% or greater of the original cohort was accounted for at study conclusion. In total, 88 cardiovascular device PMAs (77.3% permanent implants), 132 clinical studies, 37,328 study subjects (age 61.0 +/- 14.5 years, 33.9% women, 86.3% white), and 29,408 device recipients were analyzed. All PMAs contained clinical data. Primary effectiveness end points, primary safety end points, and subject accounting were deemed high quality in 81.8%, 60.2%, and 77.3% of pivotal studies, respectively. Key cardiovascular comorbidities (coronary artery disease 51.1%, diabetes 36.6%, hypertension 35.2%, heart failure 37.5%, tobacco use 31.8%) and race (14.8%) were infrequently reported, and studies rarely included patients younger than 18 years of age (10.2% of studies). Poorly defined safety and effectiveness end points, poor patient accounting, and incomplete collection of important patient comorbidities make device safety and effectiveness assessments more challenging. Women, pediatric, and nonwhite populations are underrepresented in premarket cardiovascular clinical trials. Manufacturers, regulators, and the clinical community should collaborate to address these study shortcomings to ensure that patients are treated with reliable, safe, and clinically useful medical devices.

Association of electrocardiographic morphology of exercise-induced ventricular arrhythmia with mortality.

BACKGROUND: The prognostic importance of exercise-induced ventricular arrhythmia (EIVA) may be confounded by the presence of lower-risk idiopathic right ventricular outflow tract arrhythmias with left bundle-branch block (LBBB) morphology. OBJECTIVE: To determine whether right bundle-branch block (RBBB)-morphology EIVA was associated with increased mortality. DESIGN: Retrospective cohort. SETTING: Academic medical center. PATIENTS: 585 unique patients with EIVA and 2340 patients without EIVA, matched by age, sex, and risk factor, who were referred for exercise testing in an academic medical center. MEASUREMENTS: Deaths and ischemia and infarction found on perfusion scan. RESULTS: During a mean follow-up of 24 months (SD, 13), 31 deaths occurred in the EIVA group compared with 43 deaths in the group without EIVA (5.3% vs. 1.8%; P < 0.001). Worse survival in patients with RBBB-morphology or multiple-morphology EIVA (6.9%) than in patients without EIVA caused this difference. Patients with LBBB-morphology EIVAs had a mortality rate (2.5%) similar to that of patients without EIVA (P = 0.93, log-rank test). Among patients without known atherosclerotic coronary artery disease, any RBBB-morphology EIVA was associated with death (hazard ratio, 2.73 [95% CI, 1.78 to 4.13]; P < 0.001), but LBBB-morphology EIVA was not (hazard ratio, 0.82 [CI, 0.18 to 2.04]; P = 0.72). LIMITATIONS: Not all LBBB-morphology EIVA can be dismissed, and not all RBBB-morphology EIVA is high risk. Further evaluation of patients for structural heart disease was clinically driven, not protocol-driven. CONCLUSION: Right bundle-branch block- or multiple-morphology EIVA is associated with increased mortality. Inclusion of patients with isolated LBBB-morphology EIVA, which often is idiopathic, may contribute to differences in the prognostic importance of EIVA in previous studies.

Predicting early mortality after implantable defibrillator implantation: a clinical risk score for optimal patient selection.

BACKGROUND: Patients with advanced heart disease are at risk from sudden death; however, benefit from implantable cardioverter defibrillators (ICDs) may be limited as a result of early mortality from other causes. The objective of this study was to develop a model to predict mortality within the first year after ICD implantation. METHODS AND RESULTS: A retrospective analysis was performed of 469 consecutive patients who underwent ICD implantation at a single tertiary-care center from 1999 to 2002. Vital status was determined from the Social Security Death Index. Patients were randomized into prediction and validation cohorts. A risk score was derived from the prediction cohort by multivariate logistic regression and applied to the validation cohort. One point was assigned for each variable in the risk score (age > 80 years, history of atrial fibrillation, creatinine > 1.8 mg/dL, New York Heart Association class III or IV). One-year mortality significantly increased with increasing risk score in both the prediction and validation cohorts. Validation cohort mortality was 3.4% for 0 points, 4.3% for 1 point, 17% for 2 points, and 33% for > or = 3 points (P for trend < .0001). A risk score > or = 2 predicted a 1-year mortality rate of 21%, whereas a risk score < 2 predicted a mortality rate of 4% at 1 year (P < .0001). CONCLUSION: A risk score using simple clinical criteria may identify patients at high risk of early mortality after ICD implantation. This may be helpful in consideration of ICD risk/benefit for individual patients. Further studies conducted in a prospective manner using these clinical criteria are warranted.

Mortality following defibrillator implantation in patients with renal insufficiency.

INTRODUCTION: Patients with renal insufficiency have an increased risk of atherosclerotic coronary artery disease, cardiovascular events, and sudden cardiac death. Due to under-representation of patients with renal disease in large clinical trials, outcomes of implantable cardioverter defibrillator (ICD) implantation in this group remain unclear. METHODS AND RESULTS: Inpatient and ambulatory records were reviewed for 741 consecutive patients undergoing 947 defibrillator implants or replacements at Department of Defense Medical Facilities. Demographics, medical history, and mortality were reviewed. The mean age of the cohort was 64 +/- 14 years and 599 (80.8%) were male. There were 173 patients (23.3%) with chronic renal insufficiency, 22 (3.0%) undergoing hemodialysis, and 546 (73.7%) without reported renal disease. The mean number of annual hospital admissions for heart failure among patients with and without renal failure was 3.8 +/- 4.0 versus 1.2 +/- 1.9 (P < 0.0001), respectively. The 1-year survival for those without renal insufficiency was 96.6%, compared to 87.8% for those with chronic renal insufficiency, and 88.7% for those undergoing hemodialysis. Multivariate analysis demonstrated a significant association between mortality among ICD patients and renal insufficiency, independent of coexisting congestive heart failure, ischemic cardiomyopathy, and diabetes mellitus (P < 0.0001). CONCLUSIONS: Among ICD recipients, those with renal insufficiency have a significantly higher mortality rate than those without renal insufficiency. Among a cohort of patients with ICDs, those with known renal insufficiency have higher rates of health care resource utilization and more heart failure admissions. Development of a national registry for ICDs should include data with regard to renal function.

Pacemaker and ICD generator reliability: meta-analysis of device registries.

CONTEXT: Despite there being millions of pacemaker and implantable cardioverter-defibrillator (ICD) generator implants worldwide, little is known about device reliability. OBJECTIVES: To perform a meta-analysis of prospective pacemaker and ICD registries to determine annual rates of pacemaker and ICD malfunction and to identify trends in these rates. DATA SOURCES: MEDLINE (January 1966 to April 2005), the Cochrane Central Register of Controlled Trials (through second quarter 2005), the Cochrane Database of Systematic Reviews (through second quarter 2005), and a bibliographic review of secondary sources. Search terms included pacemaker, artificial; defibrillators, implantable; registries; performance; and malfunction. STUDY SELECTION: Eligible registries were prospective; reported the number of patients with pacemaker, ICD, or both; and allowed determination of the annual number of device malfunctions. Of 1007 references screened, 3 registries meeting the selection criteria were identified and included 2.1 million pacemaker person-years and 14,821 ICD person-years of observation. DATA EXTRACTION: Included the annual number of patients with pacemakers and ICDs at risk of device failure and the annual number of generator malfunctions (1983-2004 for pacemakers, 1988-2004 for ICDs). A device malfunction was defined as an integral component failure that required device explantation prior to reaching elective replacement. Failures of pacemaker and ICD electrodes were not included in the study. DATA SYNTHESIS: There were 2981 pacemaker and 384 ICD generator malfunctions. Pacemaker reliability improved markedly during the 1980s (P for trend <.001) and the pacemaker malfunction rate remained low during the remainder of the study. Implantable cardioverter-defibrillator reliability improved during the first 10 study years (P for trend <.001). From 1998-2002, however, the ICD malfunction rate increased more than 4-fold (P for trend <.001), before decreasing substantially in the latter 2 years of the study. Overall, the mean (SE) annual ICD malfunction rate was about 20-fold higher than the pacemaker malfunction rate (26.5 [3.8] vs 1.3 [0.1] malfunctions per 1000 person-years, P<.001). Battery malfunctions were the most common cause of device failure. CONCLUSIONS: Pacemaker reliability has improved markedly. In contrast, after more than a decade of improving device reliability, the ICD malfunction rate transiently increased before experiencing substantial reductions in the latter 2 study years. Whether increasing device sophistication accounts for the observed decrease in reliability is not known. Continued monitoring of pacemaker and ICD performance is required.

Recalls and safety alerts affecting automated external defibrillators.

CONTEXT: Automated external defibrillators (AEDs) play a key role in the community resuscitation of persons with cardiac arrest and are of proven clinical benefit. Although AEDs are complex medical devices designed to function during life-threatening emergencies, little is known about their reliability. OBJECTIVES: To determine the number and rate of AED recalls and safety alerts, to identify trends in these rates, and to identify the types of malfunctions prompting AED and AED accessory advisories. DESIGN AND SETTING: Analysis of weekly US Food and Drug Administration (FDA) Enforcement Reports between January 1996 and December 2005 was performed to identify all recalls and safety alerts (collectively referred to as "advisories") involving AEDs and AED accessories. Confirmed AED device malfunctions were identified by reviewing AED-related adverse events reported to the FDA. MAIN OUTCOME MEASURES: Number of AEDs and AED accessories subject to FDA recall or safety alert between January 1996 and December 2005; annual AED advisory rates; and number of confirmed fatal AED-related device malfunctions reported to the FDA. RESULTS: During 2.78 million AED device-years of observation, 52 advisories (median [25th and 75th percentiles], 4.5 [3.0 and 5.0] per year) affecting 385,922 AEDs and AED accessories were issued. The mean (SE) annual number of AEDs affected by advisories was 5.1 (1.5) devices per 100 AED device-years. Overall, 21.2% of AEDs distributed during the study period were recalled, most often because of electrical or software problems. The AED advisory rate did not significantly increase during the study period, although the annual number of AED advisories (P for trend =.02) and AED advisory devices (P for trend = .01) did increase. Confirmed fatal AED-related device malfunctions occurred in 370 patients. CONCLUSIONS: Automated external defibrillators and AED accessory advisories occur frequently and affect many devices. Actual AED malfunctions do occur occasionally, although the number of observed malfunctions is small compared with the number of lives saved by these important devices. As the prevalence of AEDs continues to increase, the number of devices affected by advisories can also be expected to increase. Efforts should be directed at developing a reliable system to locate and repair potentially defective devices in a timely fashion.

Pacemaker and ICD generator malfunctions: analysis of Food and Drug Administration annual reports.

CONTEXT: Pacemakers and implantable cardioverter-defibrillators (ICDs) are complex medical devices proven to reduce mortality in specific high-risk patient populations. It is not known if increasing device complexity is associated with decreased reliability. OBJECTIVES: To analyze postapproval annual reports submitted to the US Food and Drug Administration (FDA) by manufacturers of pacemakers and ICDs to determine the reported number and rate of pacemaker and ICD malfunctions and to assess trends in device performance. DESIGN AND SETTING: Pacemaker and ICD annual reports submitted to the FDA for the years 1990-2002 were reviewed. A pacemaker or ICD generator was defined as having malfunctioned if it was explanted due to an observed malfunction, returned to the manufacturer, and confirmed by the manufacturer to be functioning inappropriately. Leads and biventricular devices were not included in the study. Deaths were attributed to device malfunction only if they were witnessed, the malfunction immediately led to the death, and the malfunction was confirmed by the manufacturer. MAIN OUTCOME MEASURES: Number of implanted pacemaker and ICD generators; number of reported malfunctions; and annual malfunction replacement rates. Generator malfunction replacement rates were defined as the annual number of replacements due to confirmed malfunction divided by the annual number of implants. RESULTS: During the study period, 2.25 million pacemakers and 415,780 ICDs were implanted in the United States. Overall, 17,323 devices (8834 pacemakers and 8489 ICDs) were explanted due to confirmed malfunction. Battery/capacitor abnormalities (4085 malfunctions [23.6%]) and electrical issues (4708 malfunctions [27.1%]) accounted for half of the total device failures. The annual pacemaker malfunction replacement rate per 1000 implants decreased significantly during the study, from a peak of 9.0 in 1993 to a low of 1.4 in 2002 (P = .006 for trend). In contrast, the ICD malfunction replacement rate per 1000 implants, after decreasing from 38.6 in 1993 to 7.9 in 1996, increased markedly during the latter half of the study, peaking in 2001 at 36.4 (P = .04 for trend). More than half of the reported ICD malfunctions occurred in the last 3 years of the study. Overall, the annual ICD malfunction replacement rate was significantly higher than the pacemaker malfunction replacement rate (mean [SD], 20.7 [11.6] vs 4.6 [2.2] replacements per 1000 implants; P<.001; rate ratio, 5.9 [95% confidence interval, 2.7-9.1]). Sixty-one deaths (30 pacemaker patients, 31 ICD patients) were attributable to device malfunction. CONCLUSIONS: This study demonstrates that thousands of patients have been affected by pacemaker and ICD malfunctions, the pacemaker malfunction replacement rate has decreased, the ICD malfunction replacement rate increased during the latter half of the study, and the ICD malfunction replacement rate is significantly higher than that for pacemakers. Although pacemakers and ICDs are important life-sustaining devices that have saved many lives, careful monitoring of device performance is still required.

Identification of the ventricular tachycardia isthmus after infarction by pace mapping.

BACKGROUND: Ventricular tachycardia (VT) isthmuses can be defined by fixed or functional block. During sinus rhythm, pace mapping near the exit of an isthmus should produce a QRS similar to that of VT. Pace mapping at sites proximal to the exit may produce a similar QRS with a longer stimulus-to-QRS interval (S-QRS). The aim of the study was to determine whether a VT isthmus could be identified and followed by pace mapping. METHODS AND RESULTS: Left ventricular pace mapping during sinus rhythm was performed at 819 sites in 11 patients with VT late after infarction, and corresponding CARTO maps were reconstructed. An isthmus site was defined by entrainment and/or VT termination by ablation. Pace-mapping data were analyzed from the identified isthmus site and from sites at progressively increasing distances from this initial isthmus site. Sites where pace mapping produced the same QRS with different S-QRS delays were identified to attempt to trace the course of the isthmus. In 11 patients, 13 confluent low-voltage infarct regions were present. In all these regions, parts of VT isthmuses were identified by pace mapping. In 11 of 13 of the identified isthmus parts, the QRS morphology of the pace map matched a VT QRS. In 10 of 11 patients, radiofrequency ablation rendered clinical VTs noninducible. Successful ablation sites were localized within an isthmus identified by pace mapping in all of these 10 patients. CONCLUSIONS: VT isthmuses can be identified and part of their course delineated by pace mapping during sinus rhythm. This method could help target isthmus sites for ablation during stable sinus rhythm.

Electrically unexcitable scar mapping based on pacing threshold for identification of the reentry circuit isthmus: feasibility for guiding ventricular tachycardia ablation.

BACKGROUND: We hypothesized that delineating electrically unexcitable scar (EUS) within low-voltage infarct regions will locate reentry circuit isthmuses by defining their borders. The pacing threshold and electrogram amplitude that best determines EUS is unknown. METHODS AND RESULTS: The change in dimension of the virtual electrode was estimated in 11 patients and observed to increase by 4.4+/-2.5 mm as stimulus strength increases from threshold (2.9+/-1.8 mA) to 10 mA. EUS was defined as a threshold >10 mA. In 14 consecutive patients, mapping and ablation of ventricular tachycardia (VT) were performed using an electroanatomic mapping system. During sinus rhythm, unipolar pacing was performed at sites with bipolar electrogram amplitude <1.5 mV. EUS regions were marked on the maps. Reentry circuit isthmuses were identified by entrainment mapping or pace mapping, and ablation was performed. EUS was identified in the infarct in all 14 patients (11.8+/-13.9 cm2). All 20 VT circuit isthmuses identified were adjacent to EUS. Although electrogram amplitude correlated with pacing threshold (r=0.64, P<0.0001), many isthmuses had very low-amplitude electrograms, and EUS could not be identified from electrogram amplitude alone. RF ablation lines connecting selected EUS regions abolished all inducible VTs in 10 patients (71%); spontaneous VT was markedly reduced during follow-up (from 142+/-360 to 0.9+/-2.0 episodes per month, P=0.002). CONCLUSIONS: This new method of identifying EUS provides complimentary information to the electrogram amplitude in delineating potential reentry circuit paths, potentially facilitating ablation during sinus rhythm.

Development and validation of a simple risk score to predict the need for permanent pacing after cardiac valve surgery.

OBJECTIVES: The study objective was to develop and validate a simple risk score to predict postoperative permanent pacing (PPM) after valve surgery. BACKGROUND: Our ability to identify patients preoperatively that will require PPM is poor. A simple preoperative risk score to predict PPM after valve surgery could assist both clinical practice and research. METHODS: All valve surgery patients at our institution from 1992 to 2002 were included (n = 4,694). Two-thirds of the patients were randomly selected to form a risk score prediction group (PG), and the score was then applied to the remaining patients (validation group [VG]). RESULTS: Preoperative right bundle branch block (odds ratio [OR], 3.6; 95% confidence interval [CI], 2.3 to 5.7) and multivalve surgery that included the tricuspid valve (OR, 3.7; 95% CI, 2.3 to 6.1) were the strongest independent predictors of PPM, while multivalve surgery that did not include the tricuspid valve (OR, 2.1; 95% CI, 1.3 to 3.3), preoperative left bundle branch block (OR, 2.0; 95% CI, 1.3 to 2.9), preoperative PR interval >200 ms (OR, 1.9; 95% CI, 1.3 to 3.0), prior valve surgery (OR, 1.8, 95% CI, 1.2 to 2.7), and age >70 years (OR, 1.4; 95% CI, 1.04 to 2.0) also predicted PPM. A risk score from 0 to 6 identified patients in the VG with incidences of PPM of 1.9%, 5.2%, 8.7%, 11.5%, 21%, 36%, and 50%, respectively. CONCLUSIONS: A simple risk score incorporating preoperative conduction, age, prior valve surgery, and surgery type predicts PPM after valve surgery. This score may be useful in the perioperative management of valve surgery patients.

Recognition of far-field electrograms during entrainment mapping of ventricular tachycardia.

OBJECTIVES: The goal of this study was to assess entrainment for distinguishing far-field potentials (FFP) due to depolarization of tissue at a distance from the mapping catheter from the local potential (LP) due to depolarization of tissue at the catheter electrode during mapping of ventricular tachycardia (VT). BACKGROUND: Electrograms with multiple peaks commonly complicate mapping and identification of catheter ablation targets in infarcts. METHODS: Retrospective analysis of catheter mapping data from eight patients with prior infarction was performed to evaluate multipotential electrograms at sites where pacing entrained VT. Potentials that were visible and not altered during pacing were defined as FFP. Potentials obscured by the pacing stimulus were designated possible LPs. The criteria for FFP were then assessed in a second cohort of five patients. RESULTS: At 32 of 39 (82%) sites with multiple potentials, entrainment identified one of the potentials as an FFP. Radiofrequency ablation, assessed at 15 sites, reduced the amplitude of LPs by 62%, without significant effect on FFP amplitude. At 56% of sites with multiple potentials, measuring the postpacing interval to an FFP would lead to erroneous classification of the site location relative to the reentry circuit. In prospective evaluation, double potentials were identified at 77 sites in infarcts; entrainment demonstrated an FFP at 66 (86%) sites. CONCLUSIONS: Far-field potentials are common during mapping in infarcts. Many can be distinguished from local potentials by entrainment, improving the accuracy of mapping.