Tree-based machine learning models such as random forests, decision trees, and gradient boosted trees are popular non-linear predictive models, yet comparatively little attention has been paid to ...explaining their predictions. Here, we improve the interpretability of tree-based models through three main contributions: 1) The first polynomial time algorithm to compute optimal explanations based on game theory. 2) A new type of explanation that directly measures local feature interaction effects. 3) A new set of tools for understanding global model structure based on combining many local explanations of each prediction. We apply these tools to three medical machine learning problems and show how combining many high-quality local explanations allows us to represent global structure while retaining local faithfulness to the original model. These tools enable us to i) identify high magnitude but low frequency non-linear mortality risk factors in the US population, ii) highlight distinct population sub-groups with shared risk characteristics, iii) identify non-linear interaction effects among risk factors for chronic kidney disease, and iv) monitor a machine learning model deployed in a hospital by identifying which features are degrading the model's performance over time. Given the popularity of tree-based machine learning models, these improvements to their interpretability have implications across a broad set of domains.
To determine the incidence and etiology of sudden cardiac arrest and death (SCA/D) in US high school athletes.
A prospective media database of SCA/D was queried for cases aged 14 to 18 years from 7 ...states over 6 school years (September 1, 2007, to August 30, 2013). Event details were investigated to determine participation on a high school athletic team, sex, sport, and occurrence during school-sponsored activity or exertion. National sports participation numbers were used and a conversion factor was applied to account for multisport athletes. Autopsy reports were reviewed and cause of death was adjudicated by an expert panel.
A total of 16,390,409 million athlete-seasons representing 6,974,640 athlete-years (AY) were examined, encompassing 36% of the total US high school athlete population. A total of 104 cases of SCA/D were identified (35 SCA with survival and 69 sudden cardiac deaths SCDs). The rate of SCD was 1:101,082 AY and of SCA/D 1:67,064 AY. Eighty-eight percent (92) of events occurred in male athletes. The rate of SCA/D in male athletes was 1:44,832 AY and in female athletes 1:237,510 AY (incidence rate ratio, 5.3; 95% CI, 2.9-10.6; P<.001). Men's basketball was the highest risk sport with an SCA/D incidence of 1:37,087 AY followed by men's football at 1:86,494 AY. Men's basketball and football athletes accounted for 57% (39) of deaths. Eighty percent of SCDs (55 of 69) were exertional and 55% (38 of 69) occurred while playing for a school-sponsored team. Autopsy reports were obtained in 73% (50) of cases. The most common findings of autopsy were idiopathic left ventricular hypertrophy or possible cardiomyopathy (13 of 50 26%), autopsy-negative sudden unexplained death (9 of 50 18%), hypertrophic cardiomyopathy (7 of 50 14%), and myocarditis (7 of 50 14%).
The rate of SCA/D in male high school athletes was 1:44,832 AY, with almost half due to possible or confirmed cardiomyopathy disease. It is likely that many cases were not identified because of reliance on media reports, and these numbers represent a minimum estimate.
The number of cardiac implantable electronic device (CIED) implantations has increased over recent years as a result of population growth, increasing life expectancy, adoption of guidelines, and ...better access to healthcare. Transvenous lead extraction (TLE), as a part of an overall lead management strategy, has also been increasing, not only as a consequence of this growth, but also because of increasing rates of infection, lead failure, awareness of indications for lead management, and development of extraction tools. Clinical research is essential for understanding efficacy and risks of TLE, which has important implications regarding decision-making and therapeutic strategies in patients who are candidates for this procedure. Data on TLE have mainly come from retrospective series, with variable reporting of endpoints. Recently, the ELECTRa registry conducted by the European Heart Rhythm Association (EHRA), has reported the largest prospective experience on lead extraction published to date in 3555 patients recruited from 19 European countries. There remain unresolved issues, which is a strong incentive for conducting further specifically-designed clinical trials to answer important questions in this area. In addition to clinical studies, national registries are potentially useful for evaluating epidemiology of TLE as well as for quality control and understanding resource implications. Standardization of definitions and reporting of parameters are paramount in order to analyse, compare, and pool data for scientific purposes. Expert consensus statements on lead extraction have been published by the Heart Rhythm Society (HRS) in 2009 and 2017, and by EHRA in 2012. Experience from the ELECTRa registry has been valuable for identifying challenges faced with conducting scientific studies in this field, and provides a framework for future endeavours.
This writing group has been commissioned by EHRA to provide recommendations for designing scientific studies, reports and registries relating to lead extraction.
Sudden cardiac death (SCD) is the leading cause of death in athletes during sport. Whether obtained for screening or diagnostic purposes, an ECG increases the ability to detect underlying ...cardiovascular conditions that may increase the risk for SCD. In most countries, there is a shortage of physician expertise in the interpretation of an athlete's ECG. A critical need exists for physician education in modern ECG interpretation that distinguishes normal physiological adaptations in athletes from abnormal findings suggestive of pathology. On 13-14 February 2012, an international group of experts in sports cardiology and sports medicine convened in Seattle, Washington, to define contemporary standards for ECG interpretation in athletes. The objective of the meeting was to develop a comprehensive training resource to help physicians distinguish normal ECG alterations in athletes from abnormal ECG findings that require additional evaluation for conditions associated with SCD.
The incidence and cause of sudden cardiac death (SCD) in athletes is debated with hypertrophic cardiomyopathy often reported as the most common cause.
A database of all National Collegiate Athletic ...Association deaths (2003-2013) was developed. Additional information and autopsy reports were obtained when possible. Cause of death was adjudicated by an expert panel. There were 4 242 519 athlete-years (AY) and 514 total student athlete deaths. Accidents were the most common cause of death (257, 50%, 1:16 508 AY) followed by medical causes (147, 29%, 1:28 861 AY). The most common medical cause of death was SCD (79, 15%, 1:53 703 AY). Males were at higher risk than females 1:37 790 AY versus 1:121 593 AY (incidence rate ratio, 3.2; 95% confidence interval, 1.9-5.5; P<0.00001), and black athletes were at higher risk than white athletes 1:21491 AY versus 1:68 354 AY (incidence rate ratio, 3.2; 95% confidence interval, 1.9-5.2; P<0.00001). The incidence of SCD in Division 1 male basketball athletes was 1:5200 AY. The most common findings at autopsy were autopsy-negative sudden unexplained death in 16 (25%), and definitive evidence for hypertrophic cardiomyopathy was seen in 5 (8%). Media reports identified more deaths in higher divisions (87%, 61%, and 44%), whereas the percentages from the internal database did not vary (87%, 83%, and 89%). Insurance claims identified only 11% of SCDs.
The rate of SCD in National Collegiate Athletic Association athletes is high, with males, black athletes, and basketball players at substantially higher risk. The most common finding at autopsy is autopsy-negative sudden unexplained death. Media reports are more likely to capture high-profile deaths, and insurance claims are not a reliable method for case identification.
The subcutaneous implantable cardioverter-defibrillator (S-ICD) was developed to reduce short- and long-term complications associated with transvenous ICD leads. Early multicenter studies included ...younger patients with less left ventricular systolic dysfunction and fewer comorbidities than cohorts with traditional ICD.
The purpose of this study was to characterize patient selection and the acute performance of the S-ICD in a contemporary real-world setting.
The S-ICD Post-Approval Study is a prospective registry involving 86 US centers. Patients were enrolled if they met criteria for S-ICD implantation, passed an electrocardiogram screening test, and had a life expectancy of >1 year. Analyses of descriptive statistics, Kaplan-Meier time to event, and multivariate logistic regression were performed.
The study includes 1637 patients who underwent S-ICD implantation. The cohort included 68.6% (1123/1637) male patients, and 13.4% (220/1636) were receiving dialysis for end-stage renal disease. The mean age was 52 ± 15 years, with a mean left ventricular ejection fraction of 32.0% ± 14.6%. Electrocardiogram screening was successful for at least 1, 2, or 3 vectors in 100%, 93.8%, and 51.4% of patients, respectively. Medical imaging (65.1%, 1065/1636) and general anesthesia (64.1%, 1048/16) were used in a majority of patients, and 52.2% (855/1637) were implanted with the 2-incision technique. Induced ventricular tachycardia/ventricular tachycardia was successfully converted in 98.7% (1394/1412) of patients. The 30-day complication-free rate was 96.2%. Predictors of complications included diabetes, younger age, and higher body mass index.
Contemporary US patients with S-ICD have more comorbidities than do previous cohorts with S-ICD, but they are younger with more end-stage renal disease than do patients with transvenous ICD. Implantation success is high, and short-term complication rates are acceptable.
The rate of implantable cardioverter-defibrillator (ICD) infections has been increasing faster than that of implantation. We sought to determine the rate and predictors of ICD infection in a large ...cohort of Medicare patients.
Cases submitted to the ICD Registry from 2006 to 2009 were matched to Medicare fee-for-service claims data using indirect patient identifiers. ICD infections occurring within 6 months of hospital discharge after implantation were identified by ICD-9 codes. Logistic regression was used to examine factors associated with risk of ICD infection. Of 200 909 implants, 3390 patients (1.7%) developed an ICD infection. The infection rate was 1.4%, 1.5%, and 2.0% for single, dual, and biventricular ICDs, respectively (P<0.001). Generator replacement had a higher rate compared with initial implant (1.9% versus 1.6%, P<0.001). The factors associated with infection were adverse event during implant requiring reintervention (odds ratio OR, 2.692; 95% confidence interval CI, 2.304-3.145), previous valvular surgery (OR, 1.525; 95% CI, 1.375-1.692), reimplantation for device upgrade, malfunction, or manufacturer advisory (OR, 1.354; 95% CI, 1.196-1.533), renal failure on dialysis (OR, 1.342; 95% CI, 1.123-1.604), chronic lung disease (OR, 1.215; 95% CI, 1.125-1.312), cerebrovascular disease (OR, 1.172; 95% CI, 1.076-1.276), and warfarin (OR, 1.155; 95% CI, 1.060-1.257).
Patients who developed an ICD infection were more likely to have had peri-ICD implant complications requiring early reintervention, previous valve surgery, device replacement for reasons other than battery depletion, and increased comorbidity burden. Efforts should be made to carefully consider when to reenter the pocket at any time other than battery replacement.
BACKGROUND—The pathogenesis of sudden cardiac death in college athletes has not been defined by systematic case identification.
METHODS AND RESULTS—A total of 45 cases of sudden cardiac death were ...identified in National Collegiate Athletic Association (NCAA) athletes from 2004 to 2008 based on an internal reporting system and review of media reports. Autopsy reports were reviewed and adjudicated by a multidisciplinary panel. Cause of death could be reasonably determined in 36 cases; 3 athletes had no autopsy, 5 autopsy reports could not be obtained, and 1 autopsy had insufficient information to determine cause of death. The most common finding at death was a structurally normal heart or autopsy-negative sudden unexplained death (11, 31%), followed by coronary artery abnormalities (5, 14%), dilated cardiomyopathy (3, 8%), myocarditis related (3, 8%), aortic dissection (3, 8%), and idiopathic left ventricular hypertrophy/possible hypertrophic cardiomyopathy (HCM; 3, 8%). There was 1 case each (3%) of hypertrophic cardiomyopathy, arrhythmogenic right ventricular cardiomyopathy, long QT syndrome, commotio cordis, and Kawasaki disease. There was 1 case of death in a sickle cell positive athlete who also had left ventricular hypertrophy. The adjudicated diagnosis agreed with the official pathology report in only 59% of cases.
CONCLUSIONS—Unexplained death with a structurally normal heart is the most common finding after suspected sudden cardiac death in NCAA athletes. Hypertrophic cardiomyopathy is infrequently seen, and conclusions in autopsy reports may not accurately reflect the pathological findings. Standardized protocols for cardiovascular autopsies in athletes are needed, including postmortem genetic testing, particularly in autopsy-negative cases.
Background
The subcutaneous implantable cardioverter defibrillator (S‐ICD) is a novel implanted defibrillator for the prevention of sudden cardiac death that avoids intravascular access. Use of this ...device is limited by its inability to provide backup pacing. Combined use of the S‐ICD with a permanent pacemaker may be the optimal choice in certain situations though experience with the use of both devices together remains limited.
Methods
We reviewed our single‐center experience with the S‐ICD from March 2011 to November 2015. Four patients with concomitant use of the S‐ICD and a permanent pacemaker were identified. Clinical indication for device therapy, operative details, and subsequent follow‐up was reviewed. During implantation, S‐ICD sensing of paced morphologies was evaluated at maximal voltage output. After S‐ICD implant, if feasible, the upper rate was adjusted to ≤50% of the S‐ICD tachycardia zone to minimize risk of inappropriate shocks.
Results
After a mean follow‐up of over 1 year, no adverse events occurred, including no inappropriate shock, lead malfunction, or device infection. One patient had a total of eight appropriate shocks, while another individual had no inappropriate shocks despite having a unipolar pacing lead.
Conclusions
In unique situations, combined use of the S‐ICD and a permanent pacemaker may be preferable to alternative options. In our experience, this approach was successful in varying conditions including complex congenital heart disease, recurrent device infection, and limited vascular access.