Individualized Risk Stratification of Sudden Cardiac Death: A New Take on Components of the Decision-Making Process
Sudden cardiac death (SCD) accounts for anywhere from 5 to 15% of the total United States mortality annually, and is among the leading causes of overall mortality worldwide. Other than risk-factor modification primarily aimed at reducing risk of coronary artery disease, which accounts for up to 70% of sudden death, the most effective method of preventing SCD to date is use of Implantable Cardioverter-Defibrillators (ICDs) in high risk individuals. Although guidelines exist regarding selection of patients for implantation of an ICD, much of the process guiding recommendations on an individual basis is based on extrapolation from population level studies. At a population health level, these studies are important because they provide rationale for use of a given medical intervention across a population. However, at an individual level, these studies are generally only indicative of an average treatment effect, applied to the average subject enrolled in the study. There is presently no clear method for describing an individual patient’s risk, and risk reduction after intervention, based on extrapolation from population studies.
The decision-making process for ICD implantation has been well studied and described. A common finding in many of these investigations has been that patients did not feel adequately informed about the indications for ICD, or about possible adverse effects and complications of implantation of an ICD. There is a predictable conflict that can arise when a patient undergoes a procedure and treatment shown to be beneficial at a population level, and then experiences complications individually. In other words, the risk of a complication for a procedure might be 2% in a study conducted on thousands of people, but if you are the patient with the complication, the complication rate is 100%. This process demands individualized approaches to discussions about risk that go beyond extrapolation from population studies, and yet most providers are at a loss when attempting to explain to the patient in front of them why they should or should not undergo an invasive treatment. Furthermore, when it comes to discussions taking place at the level of individual there are factors, such as fear of a devastating outcome, expectations and sense of fairness, and economic incentives, which may be even more relevant to the discussion than clinical guidelines, and yet are not often considered at the level of population studies. These factors may have previously been beyond the scope of investigation, but with increasing use of electronic social media, data mining capabilities, and machine learning analysis, there is reason to be optimistic about incorporation into future studies.
In this review, we will cover four broad components involved in individualized discussions and decision-making about SCD risk and ICD implantation: probability, biology, psychology, and economics (Figure 1). Each of these components plays some role in the decision process for ICD implantation in every patient, and many will find their way into the discussions between a provider and patient about why an ICD may or may not be indicated, either explicitly or implicitly. We will attempt to highlight the challenges with current population-level approaches to these discussions, introduce novel concepts and strategies for individualized approaches, and present opportunities for future investigation or technologies that might provide additional guidance for both patients and providers.
Figure 1. Overlapping components of decisions about sudden cardiac death and defibrillators.
Probability and Statistical Considerations
In medical school epidemiology classes, many of us are taught that one of the best statistical methods to determine the ‘real world’ impact of a treatment is the number needed to treat (NNT). Defined as the reciprocal of the absolute risk reduction, the NNT provides the number of subjects who would need to be treated by a given intervention to prevent one negative outcome. For ICD studies, the NNTs have ranged between 3 over 5 years for high-risk patients up to 18 over 20 months for moderate risk patients. Any yet, while the NNT does indeed provide better visualization of the impact of a given intervention over a measure like relative risk, it remains a population-level statistic, and thus suffers from the flaw of averages, in which a single person’s risk is unlikely to be equal to the population risk, and in some cases may be well above or below. Further, as an individualized approach to explaining the rationale for ICD implantation, the NNT can be challenging in that it leads to the provider attempting to convince a patient to undergo an invasive procedure, with complications, in which they might have only a 5% chance of benefit (NNT of 20). In contrast, if the provider were able to explain the benefit in a tangible manner, such as the expected longevity in years, then patients might have a better understanding of the benefit from ICD implantation. It is with the latter goal in mind that a simple modification of the statistical description can provide additional interpretability for an individual patient.
For an alternative description of individual SCD risk, we can turn to the field of modern cryptography, or code breaking, where like SCD, the overall probability of breaking an encryption system is low, but with potentially devastating consequences if it occurs. Because modern computers are capable of executing millions of lines of code in seconds, the ‘strength’ of an encryption algorithm is often measured in the amount of time required to break it using a brute force method (or in other words, guessing every possible permutation of the password or key). To perform this calculation, one need simply take the probability of guessing the correct key by chance, apply the inverse, and calculate the time until the probability of not guessing correctly is over 50% (Box 1).
When applied to predicting risk of sudden death, and benefit from ICDs, this approach provides two measures of individual risk: 1) How long would I be expected to live before I have a 50% or greater chance of dying suddenly? 2) How many more years would I be expected to live with an ICD than without one? The first measure can be obtained from any observational study or clinical trial directly, or via an individual risk calculator such as the SCA Risk Assessment tool from the Heart Rhythm Society (http://www.scarisk.org). The second requires extrapolation from clinical trials, although with additional analysis using methods based on distance or propensity score matching, a provider could tailor the expected benefit towards an individual’s specific risk factor profile.
Why might this approach be better? After all, it is still simply extrapolating from population studies to the individual. The first reason is that, unlike NNT, individual time to event measures in years provides the patient with a tangible number that can be explained in common-sense terms. “You have an average of 10.5 years before your risk of sudden death is over 50%” is much easier to interpret than “You have a 12% 5-year risk of sudden death.” The second reason, important for the individualized decision process, is that it provides a direct measure with which to compare improvement following an intervention. “You have an average of 10.5 years before your risk of death is 50%, but it is increased to 20 years with an ICD.” This information not only provides a directly interpretable quantitative measure, but it also provides a tangible value for comparing risk of complications. “Are you willing to trade 10 years of increased life expectancy for a risk of complication of 2%?” may not be an easy question to answer, but it’s more straightforward than “This procedure has a 2% complication rate, but an expected improvement in mortality of 17%, which is statistically significant at P < 0.05…”. This explanation takes on even more relevance in elderly subjects; for example, in an 80-year old patient, trading an expected lifetime of 90.5 years for one of 100 years may not be worth an additional risk of complications.
Future innovations in risk prediction of SCD, particularly focused around use of large datasets and machine learning hold even greater potential for development of individualized risk prediction probability models, with the added option of including additional information, such as genetics, biomarkers, and lifestyle/environmental data, into a prediction model (See next section on Biology). Importantly, the approach of training and testing a deep learning model is conducted entirely on an individual basis, and as such, methods to target risk stratification away from the population and towards the individual level will be greatly needed in the future. When combined with an analytical framework based around expected longevity in years rather than population risk, these innovations would have potential to provide a much clearer picture to the patient about what they stand to gain and lose by undergoing the invasive procedure.
Biological Considerations
A complete understanding of the pathophysiology of SCD remains largely elusive in many cases. Although the most common precipitant of SCD is coronary artery disease, many cases of SCD are in individuals without a prior history of SCD or structural cardiac disease. Investigations into genetic mutations, ion channelopathies, and other ‘non-structural’ causes of SCD are ongoing, with new syndromes and conditions noted every year. However, despite these findings, and the information about novel biological mechanisms of SCD, the majority of the current focus on risk stratification of SCD is based on population-wide studies conducted in patients with a history of coronary heart disease and cardiomyopathy.
Mechanistically speaking, SCD is most commonly due to development of ventricular fibrillation (VF), often after onset of ventricular tachycardia (VT), although bradycardic arrests do occur, often with poor outcomes. An ICD can terminate VT through one of two mechanisms: pacing the heart out of the rhythm (using a feature called anti-tachycardia pacing) or shocking the heart with high voltage, which in essence ‘resets’ the resting potential and terminates the arrhythmia (VF is only generally treated with shocking). What specifically causes VT to degenerate into VF is not entirely known; certain characteristics of VT (rate, morphology, underlying structure) have been noted to have an association with an increased risk of degenerating into VF, although validation of these characteristics has been limited. Importantly, as will be illustrated below, many investigations have noted that even patients with a history of VT are not guaranteed to have VF or SCD, indicating that a better understanding of individual risk of SCD, even in patients with VT, is needed.
VT and VF, like all tachyarrhythmias, are generally due to one of three arrhythmic mechanisms: triggered, automatic, or reentry. Triggered arrhythmias are defined by the induction with ventricular pacing, and are generally attributed to calcium overload of the sarcoplasmic reticulum and triggering of afterdepolarizations through the sodium/calcium exchanger. Automatic arrhythmias are clinically defined by spontaneous initiation and termination, and are generally attributed to abnormal phase 4 depolarization of ‘unhealthy’ ventricular cells. Reentry is most commonly (in VT at least) due to the presence of a myocardial scar, often seen following a myocardial infarction. All three mechanisms can be identified when there is structural heart disease, due to either prior infarct or non-infarct-related (i.e., nonischemic) cardiomyopathy, which is among the explanations for why structural heart disease is one of the strongest risk factors for risk of SCD.
Based on these biological characteristics, it would seem that one could develop individualized risk prediction models based on the presence of myocardial scar, inducibility with pacing during electrophysiology study, and/or circulating biomarkers of electrical or functional stress, and indeed a number of studies have been published describing associations between these findings and SCD. However, many of these population-level markers of risk do not reach levels needed to inform individual risk, due to noise or high variability in measurement, lack of appropriate effect size, or both. Increased left ventricular mass on echocardiography and/or ECG has been shown to have an association with SCD in a number of population-level studies, and yet aside from patients with extreme amounts of hypertrophy, and a diagnosis of hypertrophic cardiomyopathy, mild-moderate ventricular hypertrophy is generally not considered in discussions about risk of SCD in individual patients.
In order to reach the level of informing individualized clinical decision making, a biomarker needs to be easy to measure, temporally stable and reliable, and have an effect size that is sufficient to overcome the relative rarity of SCD events in a population. This challenge is further heighted by the fact that the two largest randomized clinical trials that drive a majority of the indications for ICD implantation were based on a very broad set of indications, providing added liability in designing clinical studies at narrowing indications. Nonetheless, there are several reasons to believe that there is room for narrowing indications for ICD implantation, and for designing a more individualized approach to risk stratification of SCD.
For one, the majority of patients in whom ICDs are implanted, including both primary and secondary indications, never receive therapy. In addition, the rate of ‘appropriate’ ICD therapies reported in most registries and clinical trials is 2-3x higher than the rate of sudden death in the control arms of early ICD studies. This indicates that the actual per-individual rate of ICD preventing SCD is much lower than what many patients and providers perceive when they study rate of ‘appropriate ICD therapy’ as a surrogate endpoint to SCD.
In addition, in patients with an ICD placed for secondary prevention purposes, studies have struggled to produce a statistical benefit, with the only one reaching statistical significance identifying no mortality benefit in patients with an EF greater than 35%. This indicates that in the absence of structural heart disease (depressed LVEF), even if patients have a history of VT, it may not necessarily be life-threatening. Such a concept is also be supported by the MADIT-RIT study, which showed that programming that increased the time to VT/VF detection was associated with an improvement in mortality. In other words, programming that minimized the amount the ICD played in treatment was beneficial in the long run, not only in terms of inappropriate therapy, but mortality.
Finally, for many patients at high risk of SCD, there is a significant impact from the competing risk of death from heart failure or other conditions. Several negative ICD studies, including DINAMIT and IRIS, demonstrated that ICD implant early after myocardial infarction was not associated with an improvement in mortality, despite an increased risk of SCD during this period (noted in a sub-study of the VALIANT trial). Any risk model that identifies a risk of SCD, particularly in subjects with heart failure, needs to account for this competing risk, otherwise the patient is merely trading one nature of death (sudden, generally painless) for another.
Despite the inherent challenges in methods for individualizing prediction risk of SCD for ICD implantation, there is much room for improvement with regard to decreasing the number of unneeded ICDs implanted, and identifying additional people outside of the ‘classic’ indications who might actually benefit from implantation. A number of investigations have identified potential markers for risk stratification, and it is likely that inclusion of these factors in an individual prediction model could provide additional guidance about the expected benefits in terms of predicted mortality with or without an ICD (see above). It is also possible that advances in -omics, wearable and implantable monitoring technology, and machine-learning and artificial intelligence will provide the framework to move the field of biological risk stratification for SCD from one focused on extrapolating population-level findings, to one focused on identification of individual risk markers. In addition to furthering our understanding of the mechanisms of SCD, these efforts will also improve the decision-making process for individuals regarding need for ICD implantation.
Human Decision-making
Medical decisions are made by humans. And being humans, the doctor who provides the information and recommendation and the patient who ultimately decides whether to follow the doctor’s advice, will not make these decisions in a vacuum. They will make them using queues from people and environment around them based not on raw statistical probability or biological inference, but on the very human characteristics of fairness and risk-avoidance. The next two sections will describe components of the decision-making process that are seldom described formally, but may ultimately play a larger role in the decision ‘process’ than probability or biology. In the following section, we will describe how theories from behavioral psychology play a role in the decision process. In the subsequent section, we will focus on the role of economics, particularly the role of incentives, in discussions about ICD implantation to reduce risk of SCD.
Behavioral Psychology
The Ultimatum game is a well-known experiment from behavioral economics, in which two subjects are placed in a room and told that they will be given a total of $1000 (or some amount of money) on the condition that the first person, the proposer, gets to decide how to divide up the sum, and the second, the responder, gets to decide whether to agree to the arrangement or not. If the responder agrees, then both get to keep their respective sums. If he does not agree, then neither person gets any money and both walk away empty handed. The Ultimatum game has been examined in a number of situations, including among smokers, twins, and in chimpanzees, with largely reproducible results. If the subjects were making decisions based on logic alone, then the responder would agree to any amount proposed, since something is always better than nothing. However, when the experiment is done in practice, the second person agrees to divisions down to around 30% before refusing to agree to the deal60. In other words, people maintain a certain concept of ‘fairness’, which in some cases overrides their sense of logic.
What does the Ultimatum tell us about decision making for ICDs? A patient walking into a physician’s office is unlikely to have zero expectation about how the encounter will proceed. What advice may be given, what options will be presented, what tone the physician may take in providing the information and recommendations to the patient; all have potential to send a signal to the patient about how invested the physician is in the decision-making process. This subtle, often unspoken, interaction can play a major role in whether the patient follows the physician’s advice, or whether he or she ignores it or seeks an additional opinion. The impression about whether the physician is being ‘unfair’ in not offering an ICD, particularly if the expectation is that an ICD should be provided, can play a major part in the decision-making process, regardless of a patient’s biological or statistical risk of SCD. Fairness has been studied in medical decision making in a number of settings, including obesity counseling61, perinatal decisions62, and decisions made in anesthesiology. While studies of cognitive bias are more recently being applied in decisions around ICD implantation, it is likely that future research will provide additional understanding of the role in fairness in individualized approaches to SCD.
Fairness is not the only ‘illogical’ concept that humans seem to invoke when making decisions. A perhaps more relevant one was described by the Nobel prize winning psychologists Amos Tversky and Daniel Kahneman, in their landmark 1979 paper ‘Prospect Theory: An Analysis of Decision under Risk.’ An example of the application of Prospect theory is described in Box 2.
Prospect theory describes the phenomenon in which people tend to overweight negative outcomes compared with positive outcomes, and make decisions on the basis of avoiding negative outcomes, rather than based on pure probability (Figure 2). Although it has been studied in the decision making around end-stage cancer, risky behavior, and weight gain, prospect theory has not been well-studied in the field of SCD risk stratification. This is unfortunate, as there are few outcomes in which individuals overweight the perception of risk as they do with death—overweighing the risk of death is the basis for the life insurance business. When making decisions about risk of SCD, and ICD implantation, fear of making the wrong decision, and not implanting an ICD in a patient who later succumbs, is a powerful factor.
Figure 2. Perceived vs. actual gains and losses under Prospect theory. The ordinate is the ‘actual’ gain or loss as described in Scenario 1 (Box 2), and the abscissa is the perceived gain or loss (utility). Under Prospect theory, individuals tend to overweight perceived losses to gains. See text for details.
It is possible that future research in SCD risk stratification and decision making around ICD implantation will eventually, like in fields of cancer and drug abuse, move to studying the impact and differential effect of risk aversion and Prospect theory. Perception of risk is a highly individualized metric, and it will likely take well-designed studies to move beyond the larger population studies (with or without individualized probability assessment—see prior) to a fully individualized approach to management of risk in SCD. However, despite the challenges, there is great potential for studies of behavioral psychology and SCD prediction in the form of passive data collection from internet and social media. Data mining from patient forums, comments, and social media provide an opportunity for unique insight into the factors that patients weigh in making a decision. Machine learning algorithms can be applied to these sources, and identification of features (search terms, key words) with potential to guide discussions around risk stratification for SCD and ICD implantation. Queries of Google search patterns have been conducted for a number of medical and nonmedical outcomes, and have provided starkly contrasting information to what had historically been obtained through patient surveys and questionnaires. Although many driving factors in behavioral psychology can be challenging to study in a medical setting, there is much reason to be optimistic about how this information may change the way we talk with patients about risk of SCD in the future.
Economic Incentives
Thus far, we have focused the discussion on the individualized approaches to SCD risk stratification and ICD implantation from the patient’s perspective. In this section, we will focus on a component that plays a much larger role for the provider, particularly in the current fee-for-service model that dominates much of American health care: economic incentives. This topic can be a challenge for most providers to discuss, and nearly all of us like to believe that we only make medical decisions based on what we believe is best for our patients. However, the data has often shown otherwise, with ICD implantation being a particularly striking example of the role of economic incentives in the medical decision-making process (see below).
A review of cases submitted to the National Cardiovascular Data Registry-ICD Registry between January 1, 2006 and June 30, 2009 found that 22.5% of patients in whom ICDs were implanted did not meet evidence-based guidelines. This finding is further troubling in light of recent study by Desai et al., which found that hospitals that reached settlements with the U.S. Department of Justice for inappropriate implantation of ICDs experienced a greater decrease in proportion of implantation of ICDs not meeting Centers for Medicare and Medicaid Services National Coverage Determination criteria. Prior work had found that patients whose ICD implantation was audited by the Department of Justice had worse survival compared with nonaudited controls, even after adjustment for baseline characteristics. In Europe, there is a trend towards increased implantation rates in countries with higher GDP, although the economic differences are not the only driver for use of the technology, with physicians understanding and/or beliefs in guidelines also identified as a key factor. These studies provide a high-level picture of the role of economic incentives in management of SCD, but understanding the role on an individual provider level requires digging deeper.
There are many more subtle economic incentives facing an individual provider faced with the decision of whether or not to implant an ICD to prevent SCD. As shown in Table 1, the number of incentives favoring implantation appear to be greater than those favoring non-intervention, at least in the setting in which many U.S. physicians practice. This table does not necessarily account for the different weights placed on each incentive by an individual, nor does it consider guideline vs. non-guideline indications, and the individual provider’s knowledge about indications. The challenge is that many of these factors that play a role in everyday patient care are quite difficult to study on a population level, partly due to a lack of methods for quantification, but also likely due to a lack of direct consideration by investigators. The challenging aspect of studying incentives is that those providers most capable of providing valuation information and insight about incentives are also the ones most likely to suffer detrimental consequences to raising awareness. As Upton Sinclair noted, “It is difficult to get a man to understand something when his salary depends upon his not understanding it.”
The medical device industry is a multi-billion-dollar industry, in which it does not take much imagination to identify a heavy incentive to encourage providers to implant more ICDs. These companies invest heavily in marketing departments, and despite the enactment of various ‘Sunshine Act’ programs to transparency of money directly supplied to providers, they retain a prominent position in delivery of medical information to providers via various education programs, honoraria, and presence at national scientific conferences. It would be unfair to criticize these companies for performing a task central to any business, marketing, although there has been a move among several large academic institutions to curtail the influence of industry81. Investigations into these effects are ongoing, with the European Heart Society releasing a policy statement in 2012 directed at the issue.
Closer to home for most providers, local incentives cannot be overlooked in the decisions around implanting an ICD for SCD risk. Like many procedures, provider productivity measured as relative value units (RVUs) is heavily weighted toward implantation of ICDs. According to the most recent Medicare physician fee schedule (https://www.cms.gov/Medicare/Medicare-Fee-for-Service-Payment/PhysicianFeeSched/index.html), implantation of a single chamber ICD offers a reimbursement of about 6 work RVUs, plus practice RVUs and additional RVUs if ultrasound access or venography is employed. These factors clearly favor intervention over conservative management, which may require the same or even greater effort in terms of follow-up visits, counseling, and discussion with family members or referring providers who want to know why an ICD is not being implanted. These factors add addition incentives as cardiac electrophysiology is generally a referral-based subspecialty, and many referring physicians make referrals to a proceduralist with the expectation that a procedure will be performed. An electrophysiologist who does not meet the expectations of a referring provider is unlikely to receive future referrals, resulting in additional loss of income, even if future referrals are appropriate and the present one is not.
Underlying many of these pro-intervention incentives is the counterfactual dilemma of intervention. It is easy to know when the decision to implant an ICD for SCD was correct, because the patient suffers from an SCD event that is prevented by the ICD. On the other hand, it is impossible to know that not implanting an ICD was the correct decision since there is no way of knowing the counterfactual situation in which a patient who would have undergone implantation and had a complication or just not needed the ICD benefitted from the decision not to implant. Of course, the worse-case scenario of choosing not to implant and being wrong and having the patient die is not only economically costly in malpractice payments, etc., but it is also traumatizing for the provider, who more likely than not will err on the side of implanting more often in the future. In the end, it is this aspect of SCD that makes it unlike nearly any other medical decision: being wrong results in patient death.
Like the components of behavioral psychology in SCD risk stratification, it is likely that further study of economic incentives will only improve our ability as providers to provide individualized approaches to management and decision-making. In contrast to many other potential areas where Big Data could be applied, cost-analysis is one in which active research and use of machine learning is ongoing, and furthering our understanding of the impact on cost and decision making. Claims databases are generally widely available, and because they often can be more easily de-identified, can be shared and analyzed by many investigators outside of those directly involved in care. How cost incentives fit into the decision process, particularly with regard to ICD implantation, remains largely undescribed at the level of the individual provider, and efforts to explore this factor can lead to animosity among colleagues and hospital administrators. However, like behavioral psychology, there is much reason to be optimistic that the next ten years will provide a great deal of opportunity to uncover these effects. Attention to outliers in billing practices can uncover providers who might be operating on the more extreme edge of implanting ICDs for SCD; in addition, with availability of high-density electronic health record data, investigators and regulators can obtain additional clinical information about patient populations in order to sort out the busy practitioner from the culprit over-implanter. This information can be obtained passively, with publication of data analysis algorithms to allow for close scrutiny of data collection methodology not previously available with use of registries, which are susceptible to data entry inaccuracy. These approaches do not necessarily overcome the more subtle economic incentives listed in Table 1, but the expansion of data available holds potential for novel approaches to investigating these factors.
Conclusion
With development of ICD technology, the past 20-30 years has enabled providers to offer life-saving treatment for a previously devastating outcome in SCD. Through population-level studies, this period has also helped us to learn a great deal about who does and does not benefit from ICD implantation across a population. A key limitation of these studies is that because they are performed at a population-level, they can only indirectly provide guidance about individual management. We believe that in order to take this next step towards individualized risk stratification and decision-making, a greater understanding of the components of this process is needed. In this review, we have outlined some of these components, and hypothesized about how advances in Big Data and machine learning, among other areas, might be able to make clear previously latent or opaque factors. It is exciting to think about where this process will go, and how it will change in the next 20-30 years.
References
1. Zheng ZJ, Croft JB, Giles WH and Mensah GA. Sudden cardiac death in the United States, 1989 to 1998. Circulation. 2001;104:2158-63.
2. Chugh SS, Jui J, Gunson K, Stecker EC, John BT, Thompson B, Ilias N, Vickers C, Dogra V, Daya M, Kron J, Zheng ZJ, Mensah G and McAnulty J. Current burden of sudden cardiac death: multiple source surveillance versus retrospective death certificate-based review in a large U.S. community. J Am Coll Cardiol. 2004;44:1268-75.
3. Kahneman D and Tversky A. Prospect Theory: An Analysis of Decision under Risk. Econometrica. 1979;47:263-291.
4. Kannel WB and Thomas HE, Jr. Sudden coronary death: the Framingham Study. Ann N Y Acad Sci. 1982;382:3-21.
5. Moss AJ, Zareba W, Hall WJ, Klein H, Wilber DJ, Cannom DS, Daubert JP, Higgins SL, Brown MW and Andrews ML. Prophylactic implantation of a defibrillator in patients with myocardial infarction and reduced ejection fraction. N Engl J Med. 2002;346:877-83.
6. Bardy GH, Lee KL, Mark DB, Poole JE, Packer DL, Boineau R, Domanski M, Troutman C, Anderson J, Johnson G, McNulty SE, Clapp-Channing N, Davidson-Ray LD, Fraulo ES, Fishbein DP, Luceri RM and Ip JH. Amiodarone or an implantable cardioverter-defibrillator for congestive heart failure. N Engl J Med. 2005;352:225-37.
7. Buxton AE, Lee KL, Fisher JD, Josephson ME, Prystowsky EN and Hafley G. A randomized study of the prevention of sudden death in patients with coronary artery disease. Multicenter Unsustained Tachycardia Trial Investigators. N Engl J Med. 1999;341:1882-90.
8. A comparison of antiarrhythmic-drug therapy with implantable defibrillators in patients resuscitated from near-fatal ventricular arrhythmias. N Engl J Med. 1997;337:1576-83.
9. European Heart Rhythm A, Heart Rhythm S, Zipes DP, Camm AJ, Borggrefe M, Buxton AE, Chaitman B, Fromer M, Gregoratos G, Klein G, Moss AJ, Myerburg RJ, Priori SG, Quinones MA, Roden DM, Silka MJ, Tracy C, Smith SC, Jr., Jacobs AK, Adams CD, Antman EM, Anderson JL, Hunt SA, Halperin JL, Nishimura R, Ornato JP, Page RL, Riegel B, Priori SG, Blanc JJ, Budaj A, Camm AJ, Dean V, Deckers JW, Despres C, Dickstein K, Lekakis J, McGregor K, Metra M, Morais J, Osterspey A, Tamargo JL, Zamorano JL, American College of C, American Heart Association Task F and European Society of Cardiology Committee for Practice G. ACC/AHA/ESC 2006 guidelines for management of patients with ventricular arrhythmias and the prevention of sudden cardiac death: a report of the American College of Cardiology/American Heart Association Task Force and the European Society of Cardiology Committee for Practice Guidelines (Writing Committee to Develop Guidelines for Management of Patients With Ventricular Arrhythmias and the Prevention of Sudden Cardiac Death). J Am Coll Cardiol. 2006;48:e247-346.
10. Matlock DD, Nowels CT, Masoudi FA, Sauer WH, Bekelman DB, Main DS and Kutner JS. Patient and cardiologist perceptions on decision making for implantable cardioverter-defibrillators: a qualitative study. Pacing Clin Electrophysiol. 2011;34:1634-44.
11. Matlock DD, Kutner JS, Emsermann CB, Al-Khatib SM, Sanders GD, Dickinson LM, Rumsfeld JS, Davidson AJ, Crane LA and Masoudi FA. Regional variations in physicians' attitudes and recommendations surrounding implantable cardioverter-defibrillators. J Card Fail. 2011;17:318-24.
12. Lewis KB, Stacey D and Matlock DD. Making decisions about implantable cardioverter-defibrillators from implantation to end of life: an integrative review of patients' perspectives. The patient. 2014;7:243-60.
13. Green AR, Jenkins A, Masoudi FA, Magid DJ, Kutner JS, Leff B and Matlock DD. Decision-Making Experiences of Patients with Implantable Cardioverter Defibrillators. Pacing Clin Electrophysiol. 2016;39:1061-1069.
14. Goldberger Z and Lampert R. Implantable cardioverter-defibrillators: expanding indications and technologies. Jama. 2006;295:809-18.
15. Ramirez J, Monasterio V, Minchole A, Llamedo M, Lenis G, Cygankiewicz I, Bayes de Luna A, Malik M, Martinez JP, Laguna P and Pueyo E. Automatic SVM classification of sudden cardiac death and pump failure death from autonomic and repolarization ECG markers. J Electrocardiol. 2015;48:551-7.
16. Goldstein BA, Chang TI, Mitani AA, Assimes TL and Winkelmayer WC. Near-term prediction of sudden cardiac death in older hemodialysis patients using electronic health records. Clinical journal of the American Society of Nephrology : CJASN. 2014;9:82-91.
17. Rea TD, Pearce RM, Raghunathan TE, Lemaitre RN, Sotoodehnia N, Jouven X and Siscovick DS. Incidence of out-of-hospital cardiac arrest. Am J Cardiol. 2004;93:1455-60.
18. Bertoia ML, Allison MA, Manson JE, Freiberg MS, Kuller LH, Solomon AJ, Limacher MC, Johnson KC, Curb JD, Wassertheil-Smoller S and Eaton CB. Risk factors for sudden cardiac death in post-menopausal women. J Am Coll Cardiol. 2012;60:2674-82.
19. Demirovic J and Myerburg RJ. Epidemiology of sudden coronary death: an overview. Prog Cardiovasc Dis. 1994;37:39-48.
20. Hulleman M, Mes H, Blom MT and Koster RW. Conduction disorders in bradyasystolic out-of-hospital cardiac arrest. Resuscitation. 2016;106:113-9.
21. Kadish AH, Buxton AE, Waxman HL, Flores B, Josephson ME and Marchlinski FE. Usefulness of electrophysiologic study to determine the clinical tolerance of arrhythmia recurrences during amiodarone therapy. J Am Coll Cardiol. 1987;10:90-6.
22. Glikson M, Lipchenca I, Viskin S, Ballman KV, Trusty JM, Gurevitz OT, Luria DM, Eldar M, Hammill SC and Friedman PA. Long-term outcome of patients who received implantable cardioverter defibrillators for stable ventricular tachycardia. J Cardiovasc Electrophysiol. 2004;15:658-64.
23. Josephson ME. Clinical Cardiac Electrophysiology: Techniques and Interpretations. 4th Edition ed. Philadelphia, PA: Lippincott Williams and Wilkins; 2008.
24. Sipido KR, Bito V, Antoons G, Volders PG and Vos MA. Na/Ca exchange and cardiac ventricular arrhythmias. Ann N Y Acad Sci. 2007;1099:339-48.
25. Ducceschi V, Di Micco G, Sarubbi B, Russo B, Santangelo L and Iacono A. Ionic mechanisms of ischemia-related ventricular arrhythmias. Clinical cardiology. 1996;19:325-31.
26. Cabo C and Wit AL. Cellular electrophysiologic mechanisms of cardiac arrhythmias. Cardiol Clin. 1997;15:517-38.
27. Nass RD, Aiba T, Tomaselli GF and Akar FG. Mechanisms of disease: ion channel remodeling in the failing ventricle. Nature clinical practice Cardiovascular medicine. 2008;5:196-207.
28. Gaztanaga J, Paruchuri V, Elias E, Wilner J, Islam S, Sawit S, Viles-Gonzalez J, Sanz J and Garcia MJ. Prognostic Value of Late Gadolinium Enhancement in Nonischemic Cardiomyopathy. Am J Cardiol. 2016;118:1063-8.
29. Disertori M, Rigoni M, Pace N, Casolo G, Mase M, Gonzini L, Lucci D, Nollo G and Ravelli F. Myocardial Fibrosis Assessment by LGE Is a Powerful Predictor of Ventricular Tachyarrhythmias in Ischemic and Nonischemic LV Dysfunction: A Meta-Analysis. JACC Cardiovasc Imaging. 2016;9:1046-1055.
30. Moss AJ, Hall WJ, Cannom DS, Daubert JP, Higgins SL, Klein H, Levine JH, Saksena S, Waldo AL, Wilber D, Brown MW and Heo M. Improved survival with an implanted defibrillator in patients with coronary disease at high risk for ventricular arrhythmia. Multicenter Automatic Defibrillator Implantation Trial Investigators. N Engl J Med. 1996;335:1933-40.
31. Kruska M, El-Battrawy I, Behnes M, Borggrefe M and Akin I. Biomarkers in Cardiomyopathies and Prediction of Sudden Cardiac Death. Current pharmaceutical biotechnology. 2017;18:472-481.
32. Medina A, Moss AJ, McNitt S, Zareba W, Wang PJ and Goldenberg I. Brain natriuretic peptide and the risk of ventricular tachyarrhythmias in mildly symptomatic heart failure patients enrolled in MADIT-CRT. Heart Rhythm. 2016;13:852-9.
33. Ramirez J, Orini M, Minchole A, Monasterio V, Cygankiewicz I, Bayes de Luna A, Martinez JP, Laguna P and Pueyo E. Sudden cardiac death and pump failure death prediction in chronic heart failure by combining ECG and clinical markers in an integrated risk model. PLoS One. 2017;12:e0186152.
34. Haider AW, Larson MG, Benjamin EJ and Levy D. Increased left ventricular mass and hypertrophy are associated with increased risk for sudden death. J Am Coll Cardiol. 1998;32:1454-9.
35. Wachtell K, Okin PM, Olsen MH, Dahlof B, Devereux RB, Ibsen H, Kjeldsen SE, Lindholm LH, Nieminen MS and Thygesen K. Regression of electrocardiographic left ventricular hypertrophy during antihypertensive therapy and reduction in sudden cardiac death: the LIFE Study. Circulation. 2007;116:700-5.
36. Sabbag A, Suleiman M, Laish-Farkash A, Samania N, Kazatsker M, Goldenberg I, Glikson M and Beinart R. Contemporary rates of appropriate shock therapy in patients who receive implantable device therapy in a real-world setting: From the Israeli ICD Registry. Heart Rhythm. 2015;12:2426-33.
37. Providencia R, Marijon E, Lambiase PD, Bouzeman A, Defaye P, Klug D, Amet D, Perier MC, Gras D, Algalarrondo V, Deharo JC, Leclercq C, Fauchier L, Babuty D, Bordachar P, Sadoul N, Piot O and Boveda S. Primary Prevention Implantable Cardioverter Defibrillator (ICD) Therapy in Women-Data From a Multicenter French Registry. Journal of the American Heart Association. 2016;5.
38. Germano JJ, Reynolds M, Essebag V and Josephson ME. Frequency and causes of implantable cardioverter-defibrillator therapies: is device therapy proarrhythmic? Am J Cardiol. 2006;97:1255-61.
39. Kuck KH, Cappato R, Siebels J and Ruppel R. Randomized comparison of antiarrhythmic drug therapy with implantable defibrillators in patients resuscitated from cardiac arrest : the Cardiac Arrest Study Hamburg (CASH). Circulation. 2000;102:748-54.
40. Connolly SJ, Gent M, Roberts RS, Dorian P, Roy D, Sheldon RS, Mitchell LB, Green MS, Klein GJ and O'Brien B. Canadian implantable defibrillator study (CIDS) : a randomized trial of the implantable cardioverter defibrillator against amiodarone. Circulation. 2000;101:1297-302.
41. Domanski MJ, Sakseena S, Epstein AE, Hallstrom AP, Brodsky MA, Kim S, Lancaster S and Schron E. Relative effectiveness of the implantable cardioverter-defibrillator and antiarrhythmic drugs in patients with varying degrees of left ventricular dysfunction who have survived malignant ventricular arrhythmias. AVID Investigators. Antiarrhythmics Versus Implantable Defibrillators. J Am Coll Cardiol. 1999;34:1090-5.
42. Moss AJ, Schuger C, Beck CA, Brown MW, Cannom DS, Daubert JP, Estes NA, 3rd, Greenberg H, Hall WJ, Huang DT, Kautzner J, Klein H, McNitt S, Olshansky B, Shoda M, Wilber D and Zareba W. Reduction in inappropriate therapy and mortality through ICD programming. N Engl J Med. 2012;367:2275-83.
43. Hohnloser SH, Kuck KH, Dorian P, Roberts RS, Hampton JR, Hatala R, Fain E, Gent M and Connolly SJ. Prophylactic use of an implantable cardioverter-defibrillator after acute myocardial infarction. N Engl J Med. 2004;351:2481-8.
44. Steinbeck G, Andresen D, Seidl K, Brachmann J, Hoffmann E, Wojciechowski D, Kornacewicz-Jach Z, Sredniawa B, Lupkovics G, Hofgartner F, Lubinski A, Rosenqvist M, Habets A, Wegscheider K and Senges J. Defibrillator implantation early after myocardial infarction. N Engl J Med. 2009;361:1427-36.
45. Solomon SD, Zelenkofske S, McMurray JJ, Finn PV, Velazquez E, Ertl G, Harsanyi A, Rouleau JL, Maggioni A, Kober L, White H, Van de Werf F, Pieper K, Califf RM and Pfeffer MA. Sudden death in patients with myocardial infarction and left ventricular dysfunction, heart failure, or both. N Engl J Med. 2005;352:2581-8.
46. Levine YC, Matos J, Rosenberg MA, Manning WJ, Josephson ME and Buxton AE. Left ventricular sphericity independently predicts appropriate implantable cardioverter-defibrillator therapy. Heart Rhythm. 2016;13:490-7.
47. Levine YC, Rosenberg MA, Mittleman M, Samuel M, Methachittiphan N, Link M, Josephson ME and Buxton AE. B-type natriuretic peptide is a major predictor of ventricular tachyarrhythmias. Heart Rhythm. 2014;11:1109-16.
48. Huertas-Vazquez A, Nelson CP, Guo X, Reinier K, Uy-Evanado A, Teodorescu C, Ayala J, Jerger K, Chugh H, Braund PS, Deloukas P, Hall AS, Balmforth AJ, Jones M, Taylor KD, Pulit SL, Newton-Cheh C, Gunson K, Jui J, Rotter JI, Albert CM, Samani NJ and Chugh SS. Novel loci associated with increased risk of sudden cardiac death in the context of coronary artery disease. PLoS One. 2013;8:e59905.
49. Arking DE, Junttila MJ, Goyette P, Huertas-Vazquez A, Eijgelsheim M, Blom MT, Newton-Cheh C, Reinier K, Teodorescu C, Uy-Evanado A, Carter-Monroe N, Kaikkonen KS, Kortelainen ML, Boucher G, Lagace C, Moes A, Zhao X, Kolodgie F, Rivadeneira F, Hofman A, Witteman JC, Uitterlinden AG, Marsman RF, Pazoki R, Bardai A, Koster RW, Dehghan A, Hwang SJ, Bhatnagar P, Post W, Hilton G, Prineas RJ, Li M, Kottgen A, Ehret G, Boerwinkle E, Coresh J, Kao WH, Psaty BM, Tomaselli GF, Sotoodehnia N, Siscovick DS, Burke GL, Marban E, Spooner PM, Cupples LA, Jui J, Gunson K, Kesaniemi YA, Wilde AA, Tardif JC, O'Donnell CJ, Bezzina CR, Virmani R, Stricker BH, Tan HL, Albert CM, Chakravarti A, Rioux JD, Huikuri HV and Chugh SS. Identification of a sudden cardiac death susceptibility locus at 2q24.2 through genome-wide association in European ancestry individuals. PLoS Genet. 2011;7:e1002158.
50. Meyer C, Carvalho P, Brinkmeyer C, Kelm M, Couceiro R and Muhlsteff J. Wearable sensors in syncope management. Medical science monitor : international medical journal of experimental and clinical research. 2015;21:276-82.
51. Sakhi R, Theuns D, Bhagwandien RE, Michels M, Schinkel AFL, Szili-Torok T, Zijlstra F, Roos-Hesselink JW and Yap SC. Value of implantable loop recorders in patients with structural or electrical heart disease. J Interv Card Electrophysiol. 2018;52:203-208.
52. Przyzycka P, Kalowski M, Poddebska I, Kaczmarek K, Ptaszynski P, Wranicz JK and Cygankiewicz I. Self-terminating ventricular fibrillation recorded by an implantable loop recorder as a cause of syncope - A case report. J Electrocardiol. 2018;51:617-619.
53. LeCun Y, Bengio Y and Hinton G. Deep learning. Nature. 2015;521:436-44.
54. Johnson KW, Torres Soto J, Glicksberg BS, Shameer K, Miotto R, Ali M, Ashley E and Dudley JT. Artificial Intelligence in Cardiology. J Am Coll Cardiol. 2018;71:2668-2679.
55. Sanfey AG, Rilling JK, Aronson JA, Nystrom LE and Cohen JD. The neural basis of economic decision-making in the Ultimatum Game. Science. 2003;300:1755-8.
56. Güth W, Schmittberger R and B S. An experimental analysis of ultimatum bargaining. J Econ Behav Organ. 1982;3:367-388.
57. Takahashi T. Economic decision-making in the ultimatum game by smokers. Neuro endocrinology letters. 2007;28:659-61.
58. Cesarini D, Dawes CT, Fowler JH, Johannesson M, Lichtenstein P and Wallace B. Heritability of cooperative behavior in the trust game. Proc Natl Acad Sci U S A. 2008;105:3721-6.
59. Proctor D, Williamson RA, de Waal FB and Brosnan SF. Chimpanzees play the ultimatum game. Proc Natl Acad Sci U S A. 2013;110:2070-5.
60. Camerer C. Behavioral Game Theory: Experiments in Strategic Interaction. Princeton, NJ: Princeton Univ Press; 2003.
61. Kubera B, Klement J, Wagner C, Radel C, Eggeling J, Fullbrunn S, Kaczmarek MC, Levinsky R and Peters A. Differences in fairness and trust between lean and corpulent men. International journal of obesity (2005). 2016;40:1802-1808.
62. Minkoff H, Zafra K, Amrita S, Wilson TE and Homel P. Physician morality and perinatal decisions. European journal of obstetrics, gynecology, and reproductive biology. 2016;206:36-40.
63. Stiegler MP and Tung A. Cognitive processes in anesthesiology decision making. Anesthesiology. 2014;120:204-17.
64. Matlock DD, Jones J, Nowels CT, Jenkins A, Allen LA and Kutner JS. Evidence of Cognitive Bias in Decision Making Around Implantable-Cardioverter Defibrillators: A Qualitative Framework Analysis. J Card Fail. 2017;23:794-799.
65. Malhotra C, Xiang L, Ozdemir S, Kanesvaran R, Chan N and Finkelstein EA. A comparison of attitudes toward length and quality of life between community-dwelling older adults and patients with advanced cancer. Psycho-oncology. 2017;26:1611-1617.
66. Kirchler M, Andersson D, Bonn C, Johannesson M, Sorensen EO, Stefan M, Tinghog G and Vastfjall D. The effect of fast and slow decisions on risk taking. Journal of risk and uncertainty. 2017;54:37-59.
67. Lim SL and Bruce AS. Prospect theory and body mass: characterizing psychological parameters for weight-related risk attitudes and weight-gain aversion. Frontiers in psychology. 2015;6:330.
68. Johnson AK and Mehta SD. A comparison of Internet search trends and sexually transmitted infection rates using Google trends. Sexually transmitted diseases. 2014;41:61-3.
69. Dugas AF, Jalalpour M, Gel Y, Levin S, Torcaso F, Igusa T and Rothman RE. Influenza forecasting with Google Flu Trends. PLoS One. 2013;8:e56176.
70. Ma-Kellams C, Bishop B, Zhang MF and Villagrana B. Using "Big Data" Versus Alternative Measures of Aggregate Data to Predict the U.S. 2016 Presidential Election. Psychological reports. 2018;121:726-735.
71. Stephens-Davidowitz S. Everybody Lies: Big Data, New Data, and What the Internet Can Tell Us About Who We Really Are. New York, NY: HarperCollins; 2017.
72. Chae DH, Clouston S, Martz CD, Hatzenbuehler ML, Cooper HLF, Turpin R, Stephens-Davidowitz S and Kramer MR. Area racism and birth outcomes among Blacks in the United States. Social science & medicine (1982). 2018;199:49-55.
73. Kazungu JS, Barasa EW, Obadha M and Chuma J. What characteristics of provider payment mechanisms influence health care providers' behaviour? A literature review. The International journal of health planning and management. 2018.
74. Al-Khatib SM, Hellkamp A, Curtis J, Mark D, Peterson E, Sanders GD, Heidenreich PA, Hernandez AF, Curtis LH and Hammill S. Non-evidence-based ICD implantations in the United States. Jama. 2011;305:43-9.
75. Desai NR, Bourdillon PM, Parzynski CS, Brindis RG, Spatz ES, Masters C, Minges KE, Peterson P, Masoudi FA, Oetgen WJ, Buxton A, Zipes DP and Curtis JP. Association of the US Department of Justice Investigation of Implantable Cardioverter-Defibrillators and Devices Not Meeting the Medicare National Coverage Determination, 2007-2015. Jama. 2018;320:63-71.
76. Shariff N, Rahim S, Jain S, Barrington W and Saba S. Long-term outcome of defibrillator recipients included in the federal audit conducted by the Department of Justice. Am J Cardiol. 2014;114:723-6.
77. Stella BM and Alessandro Z. ICD Implantation Practice Within Europe: How To Explain The Differences Beyond Economy? Journal of atrial fibrillation. 2015;8:1262.
78. Raatikainen MJ, Arnar DO, Zeppenfeld K, Merino JL, Levya F, Hindriks G and Kuck KH. Statistics on the use of cardiac electronic devices and electrophysiological procedures in the European Society of Cardiology countries: 2014 report from the European Heart Rhythm Association. Europace. 2015;17 Suppl 1:i1-75.
79. Castellanos JM, Smith LM, Varosy PD, Dehlendorf C and Marcus GM. Referring physicians' discordance with the primary prevention implantable cardioverter-defibrillator guidelines: a national survey. Heart Rhythm. 2012;9:874-81.
80. Sadarmin PP, Wong KC, Rajappan K, Bashir Y and Betts TR. Questionnaire survey of cardiologists' knowledge, attitudes, and guideline application of implantable cardioverter defibrillator therapy. Pacing Clin Electrophysiol. 2012;35:672-80.
81. Harvard University Faculty of Medicine Policy on Conflicts of Interest and Commitment. 2016.
82. Lewin J and Arend TE, Jr. Industry and the profession of medicine: balancing appropriate relationships with the need for innovation. J Vasc Surg. 2011;54:47s-9s.
83. Fugh-Berman AJ, Scialli AR and Bell AM. Why lunch matters: assessing physicians' perceptions about industry relationships. The Journal of continuing education in the health professions. 2010;30:197-204.
84. Relations between professional medical associations and the health-care industry, concerning scientific communication and continuing medical education: a Policy Statement from the European Society of Cardiology. Eur Heart J. 2012;33:666-74.
85. Pak TR, Chacko KI, O'Donnell T, Huprikar SS, van Bakel H, Kasarskis A and Scott ER. Estimating Local Costs Associated With Clostridium difficile Infection Using Machine Learning and Electronic Medical Records. Infection control and hospital epidemiology. 2017:1-9.
86. Wang Y, Iyengar V, Hu J, Kho D, Falconer E, Docherty JP and Yuen GY. Predicting Future High-Cost Schizophrenia Patients Using High-Dimensional Administrative Data. Frontiers in psychiatry. 2017;8:114.
87. Krishnan A, Xu T, Hutfless S, Park A, Stasko T, Vidimos AT, Leshin B, Coldiron BM, Bennett RG, Marks VJ, Brandt R, Makary MA and Albertini JG. Outlier Practice Patterns in Mohs Micrographic Surgery: Defining the Problem and a Proposed Solution. JAMA dermatology. 2017;153:565-570.
88. Chan B, Anderson GM and Theriault ME. High-billing general practitioners and family physicians in Ontario: how do they do it? An analysis of practice patterns of GP/FPs with annual billings over $400,000. CMAJ : Canadian Medical Association journal = journal de l'Association medicale canadienne. 1998;158:741-6.
89. Kosy JD, Kassam AA and Hockings M. National Joint Registry data inaccuracy: a threat to proper reporting. British journal of hospital medicine (London, England : 2005). 2013;74:691-3.