Portal de Periódicos da CAPES

Model Building as an Educational Hobby

2016; Lippincott Williams & Wilkins; Volume: 9; Issue: 8 Linguagem: Inglês

10.1161/circheartfailure.116.003457

1941-3297

Lynne W. Stevenson, Roger B. Davis,

Organizational Learning and Leadership

HomeCirculation: Heart FailureVol. 9, No. 8Model Building as an Educational Hobby Free AccessEditorialPDF/EPUBAboutView PDFView EPUBSections ToolsAdd to favoritesDownload citationsTrack citationsPermissions ShareShare onFacebookTwitterLinked InMendeleyRedditDiggEmail Jump toFree AccessEditorialPDF/EPUBModel Building as an Educational Hobby Lynne Warner Stevenson and MD Roger B. DavisScD Lynne Warner StevensonLynne Warner Stevenson From the Advanced Heart Disease Section, Cardiovascular Division, Brigham and Women's Hospital, Boston, MA (L.W.S.); and Division of General Medicine and Primary Care, Beth Israel Deaconess Medical Center, Boston, MA (R.B.D.). and Roger B. DavisRoger B. Davis From the Advanced Heart Disease Section, Cardiovascular Division, Brigham and Women's Hospital, Boston, MA (L.W.S.); and Division of General Medicine and Primary Care, Beth Israel Deaconess Medical Center, Boston, MA (R.B.D.). Originally published11 Aug 2016https://doi.org/10.1161/CIRCHEARTFAILURE.116.003457Circulation: Heart Failure. 2016;9:e003457The traditional hobby of building models such as ships and airplanes requires meticulous assembly of small-scale pieces in sequence. Many models designed to commemorate historic events provide education during the assembly, and those that are completed often become objects of display on a shelf. Model building can also become a career, using past events as a basis from which to predict future events of high commercial interest. Fatigue analysis of components in engineering, for example, has reached a high state of accuracy with narrow error margins, because the conditions of the testing can be tightly controlled to approximate the conditions of future use. However, using the past to predict a future with different conditions is flawed by wide uncertainty, as reviewed by Taleb in the Black Swan.1 Errors have been notorious for predictions from the economist, defined as an expert who will know tomorrow why the things he predicted yesterday did not happen today.2 Financial experts have been demonstrated to consistently underestimate the contribution of unknown factors to the accuracy of their predictions, which although often no better than amateur predictions are reliably accompanied by greater confidence, which increases with the number of variables modeled.1See Articles by Upshaw et al and Lagu et alHigh levels of randomness apply also to both the individual human behavior and biological events related to heart failure (HF), creating even greater uncertainty at their interfaces, where outcomes of hospitalization and death occur. Two articles in the current issue of Circulation: Heart Failure provide insight into the construction and use of models for predicting HF events. Both emphasize the primacy of simple clinical factors available to the clinician. Lagu et al3 have compared models for death during hospitalization for heart failure, derived from data of various eras and tested in a large contemporary clinical population. The multistate model by Upshaw et al4 also emphasizes simple clinical factors, which influence the separate transition rates to death and to HF hospitalization but contribute less once the journey has passed through HF hospitalization.The substantial mismatches in these prediction models provide a sober reminder that all models of these outcomes will be undermined by a large random component; heart failure thus remains a more stochastic process than the failure of machine parts. HF models currently have limited use for prediction, and targeted prevention is limited additionally by a dearth of therapies specific for patients at high risk. Even with limited impact on prediction or prevention, however, consideration of relevant models by the healthcare team may encourage preparation with patients and families facing an uncertain future.Challenges of Model BuildingThe detail of data required for model assembly has often been available only from clinical trials, in which both the population and the care settings are more carefully selected and supervised than general clinical practice. Clinical data sets from retrospectively obtained chart review are flawed by lack of uniformity about the indications and timing for examination and laboratory testing. Alternatively, administrative data sets offer a large population and generalizability, but the required data elements are arbitrarily coded and often binary, neglecting key continuous variables such as renal function, natriuretic peptide levels, blood pressure, clinical congestion, and nutrition.There is a large gap between derivation and validation of HF models, as discussed by Califf and Pencina.5 The majority of proposed models are validated only by bootstrapping within the original data set. External validation sets often share similar limitations with the derivation set, particularly if from clinical trials or limited to a few selected sites. Risk prediction models that have undergone external validation studies in HF were recently reviewed by Alba et al6, who found disappointing performance. Calibration of absolute event rates was variable, overestimating mortality in some populations and underestimating in others. Similar results of HF models were reviewed by Rahimi et al.7 The C statistics (which can be calculated in at least 4 ways to reflect how often a patient with an event had a higher risk than a patient without an event) indicated poor-to-modest discrimination, usually about half-way or less between random and perfect.One of the inescapable limitations is that the contemporary HF population continues to move beyond the conditions present when the models were derived and validated. From the previous half century, prediction models could maintain their initial accuracy for a while, but progress in HF therapy has more recently gained speed.8 With chronic renin–angiotensin system inhibition, vasoconstriction during acute decompensated heart failure admissions became less common. The combination of angiotensin-converting enzyme inhibitors and β-blockers decreased sudden death before recommendation of primary prevention implantable cardioverter defibrillators.9 These devices further decreased sudden death in both early- and late-stage heart failure without altering disease progression, thus leading to an increased prevalence of secondary right heart failure and cardiorenal syndrome, which was uncommon during heart failure hospitalization in the 1980s.10 Cardiac resynchronization and β-blocker therapy as part of guideline-directed medical therapy have both increased the proportion of HF better ejection fraction, which now characterizes up to a third of outpatient HF populations.11 Patients today with low ejection fraction that has not improved on therapy represent a more refractory group than previous low ejection fraction HF.Because the patient journey is rerouted by the cumulative exposure to therapies already given, poor prediction of mortality shown in the study by Lagu et al from models such as EFFECT (Enhanced Feedback for Effective Cardiac Treatment), derived from patients between 1999 and 2001 and ADHERE (Acute Decompensated Heart Failure National Registry), derived from patients between 2001 and 2003, is not surprising. The review by Alba et al6 indicated that higher use of beta blockers and implantable cardioverter defibrillators in the validation sets was associated with worse model performance. However, it is dangerous to link on additional therapies that were not included in the models when derived. This has been done in the current multistate model with the addition of defibrillators, for example. Patients not on contemporary standard therapies differ from the control group of pivotal trials that led to approval. Primary device implantation is done with a healthy user bias, such that patients receiving them are already healthier than those patients whose clinicians felt that devices were contraindicated by competing risks.12 This leads to a larger difference in outcomes than seen in the trials, only part of which is because of the therapy itself. On the contrary, digoxin, which had a neutral impact in the largest randomized trial,13 is now generally used only after recurrent hospitalizations, leading to a sick user bias. A major inaccuracy in extrapolating benefit is the relatively short duration of the trial period. One of the most important trial lessons is from the 12-year analysis of the SOLVD trial (Studies of Left Ventricular Dysfunction), revealing that patients who began angiotensin-converting enzyme inhibitors while asymptomatic derived more survival benefit at 12 years than those in the symptomatic arm, although the latter is where the early survival benefit was shown.14For Display Only Versus A Working ModelAlthough perfect prediction is not attainable, the classic models such as the Heart Failure Survival Score, the Seattle Heart Failure Score, and the Heartmate II Risk Scores have provided useful starting points from which to track and sometimes reroute the HF journeys. As often quoted from the eminent statistician George E.P. Box, "All models are wrong but some are useful."15Learning About Patient PopulationsMuch of the value of a model comes from lessons learned while building it. Previous modeling of HF outcomes has revealed and confirmed many important physiological contributors, exemplified by neurohormonal activation,16 hyponatremia,17 right ventricular pacing,18 early renal dysfunction,19 elevated cardiac filling pressures,20 and natriuretic peptide levels, which predict across a large entire HF spectrum.21,22 Even with the progressive mechanistic insights, the traditional New York Heart Association class, which is a loose subjective integrator, has consistently retained its high rank for prediction of outcomes, as in the current multistate model by Upshaw et al.Unfortunately, education is diminishing as the degree of separation is increasing between the clinical investigators who select the data elements and the statisticians who develop the model. Considerable learning takes place during the tedious iterations of model building as predictors enter and exit the model depending on other variables, exclusions for missing data, and population subsets. It is only the hands-on modeling experience that provides insights such as the impact of how we deal with missing data, the futility of ranking variables that are tightly linked, the failures of proportional hazard assumptions, and the robustness or frailty of predictors.The current article looking at mortality during hospitalization provides useful confirmation that common factors such as demographics and comorbidities remain relevant to outcomes across different time periods and that common admission laboratory profiles improve the receiver-operating curve. The article looking at multistate transitions demonstrates that common clinical factors predict both HF hospitalization and death, with some distinctions, such as the greater contribution of ischemic pathogenesis to death than to HF hospitalization, possibly because of fatal reinfarction. Neither model includes natriuretic peptide levels or diuretic doses. A notable result from the multistate model is that most deaths in the derivation and validation data sets were sudden and most occurred without a previous HF hospitalization. It should be determined whether this proportion of sudden death with or without previous hospitalization has changed after the subsequent guideline recommendations for primary prevention implantable cardioverter defibrillators.Models have frequently been used to select populations for future trials. Matching specific population risk to therapy is clearly useful to test therapies targeting a specific physiological process. Less useful are the modeled event rates, which are notoriously overestimated. This gap is anticipated given the inexorable forward journey of the population with advancing therapy, as discussed above. The mismatch between modeled and contemporary event rates makes it particularly treacherous to assess the impact of a new therapy by comparison to hypothetical outcomes of what would have happened based on an outdated model.Models can be used for adjustment of events that have already happened, an application implemented nationally to align HF event rates in simultaneous populations from different institutions for benchmarking and efforts to measure quality of care.23 There are many unintended consequences possible in this quest for quality, including penalties for care of the underserved, impediments to necessary readmissions and pressure toward hospice care in patients who still enjoy meaningful quality of life despite multiple rehospitalizations. The penetration of electronic medical records will hopefully allow replacement of current arbitrary coding terms with more physiological and clinical variables in the Centers for Medicare and Medicaid Services models, so that these can reflect our best attempts to level the outcomes that need to be compared.Clinical Use: Prediction, Prevention, or Preparation?PredictionPredictive models are often proposed to help identify those individuals at the highest risk of future events, with the implication from both of the current modeling articles that identification of high-risk individuals can focus intervention to decrease event risk. Both articles present the traditional HF outcome of death. Although life expectancy is lower for heart failure compared with the age-matched population, death is uncommon in the short term, in the current studies 4.3% during hospitalization and 5% to 6% at 1 year. With the uncertainty around all biological estimates, the positive predictive value of a model prediction for death is usually too low to be actionable. In this issue, Lagu et al have provided not only the usual receiver-operating curve areas for model comparison but also the more practical comparison of specificity at a given sensitivity across models for hospital mortality. This allows determination of how often patients died as predicted, which is not often (Table). Interestingly, the best performing model at admission in this article in this issue was the LAPS2 model, based on admission laboratory profiles in patients with a broad range of diagnoses besides HF. Even in this model, for every one patient dying in the hospitalization during which death was predicted, 11 patients predicted to die were instead discharged alive.Table. Models Predict Hospital Death in Many Patients Who SurviveModel as Applied at AdmissionNo. of Patients Discharged After Predicted to Die in Hospital, Among 100 PatientsDead After Predicted to Die in Hospital*, Among 100 PatientsDead After Predicted to Survive to Discharge*, Among 100 PatientsRatio of Patients Actually Dead to Patients Discharged After Predicted to DieADHERE48311–16EFFECT43311–14Premier (at admission)36311–12LAPS2 (HF+other diagnoses)32311–11*Specified by arbitrary selection of sensitivity of 0.75 to compare models, mortality rate rounded to 4% for simplicity. ADHERE indicates Acute Decompensated Heart Failure National Registry; EFFECT, Enhanced Feedback for Effective Cardiac Treatment; HF, heart failure; and LAPS2, Laboratory Based Acute Physiology Score 2.This discouraging result is not the fault of the models but of the inherent stochastic nature of the events we seek to prevent, which undermines every model. Prediction may be improved using dynamic modeling to include changes over time, as discussed by Califf and Pencina.5 In the current article, adding events occurring early after admission increased the specificity from 0.63 to 0.81 for hospital mortality. However, analysis is treacherous for models that include early events as independent variables. Dynamic models are most appropriate for sequential decision making over prolonged follow-up, such as recently proposed in patients who might be considered for ventricular assist devices.24Predicting death is challenging for most terminal diseases.25 The potential for primary arrhythmic death and the wide oscillations in clinical status are invoked to explain the difficulty of predicting death with HF compared with diseases such as cancer. However, the predictability of death with cancer is itself generally overestimated. It has been demonstrated that for patients with late-stage cancer predicted to die at a certain time, ≈50% of patients will either die in less than half the predicted time or survive for more than twice as long.26PreventionThe ultimate application of accurate prediction would be to intervene to avert or delay the adverse outcomes. When outcomes are combined to increase event rates, models become more powerful at prediction, but the predictions become less actionable. For instance, translation to prevention is complicated for the common composite of death and HF hospitalization. The article by Upshaw et al importantly separates the predictions of death and HF hospitalization and also specifies whether or not hospitalization occurred at some time before death. Prevention of a given event is further complicated by the mosaic of proximal causes. Prevention of nonadherence would be a different intervention than preventing atrial fibrillation as a cause of hospitalization, although socioeconomic status and arrhythmia history may be included in the same model. Decreasing the chances of unexpected sudden death, end-stage circulatory failure, and fatal pneumonia would also require different strategies.The most practical limitation to use of our models to prevent HF events is the dearth of options for further intervention. There is no obvious strategy by which to intensify therapy for most patients, in the hospital or at home. Guideline-directed medical therapy applies to patients across a broad spectrum of HF severity.27 The therapies with a level 1 recommendation should be initiated even in patients at low risk, so what else can be added when risk is high? There is no data that higher doses of therapy ameliorate higher risks. Some early HF models were designed to trigger referral to heart transplantation. However, the excess of currently waiting candidates restricts transplantation to those with ongoing circulatory compromise or ventricular assist device dysfunction that can easily be identified clinically without modeling. The risk models derived from the recent transplant waiting list will have diminishing validity once they influence the priority for transplantation. Referral for ventricular assist device is a more complex decision, which requires simultaneous consideration of survival, adverse events, and quality of life both with and without a ventricular assist device.28 Shared decision making about ventricular assist devices in ambulatory patients will be facilitated by models comparing for these outcomes. However, modeling differences between 2 therapies for predicted outcomes will multiply the ranges of uncertainty.PreparationAlthough the translation of prediction to effective prevention is the highest working order for a model, many events cannot be prevented, whether or not they could be predicted. Preparation for events in a population at risk is a more reachable goal. Patients at risk for HF hospitalization should be coached on vigilance for symptoms that warrant attention even if it leads to rehospitalization. Individual care plans may be created in advance to streamline emergency department evaluation without unnecessary testing and to accelerate effective decongestion once admitted. These interventions are reasonable even with a wide margin of error around prediction of rehospitalization.An urgent unmet need is better preparation of patients and families for the possibilities of death with HF.28,29 In this respect, the challenge is clearly greater than that for cancer, a diagnosis that immediately summons the specter of death. Despite the grim nature of the term heart failure, many patients and families do not recognize it as a lethal disease and generally overestimate survival by several folds. An important lesson of the Upshaw et al multistate model in this issue arises from the high proportion of patients who did not have a HF hospitalization before HF-related death. If this remains true in the current era, it means that we should not wait until after a HF hospitalization to review prognosis and discuss mortality, although the risks do escalate rapidly with multiple hospitalizations.30 At the least, prognostic models may serve to define a denominator of patients for whom documenting outcomes of goals of care discussions could be a quality measure.Looking forward, construction will continue on new models in hopes of predicting and preventing more HF events in the future. These will become more sophisticated as the polynomics provide multilevel molecular maps and the big population data add more signals and more noise.10 New factors for disease progression or regression will emerge as new therapies push progress of the HF population further beyond its natural history of half a century ago. There is much to be learned from the meticulous construction and comparison of models, as shown in the 2 accompanying articles by Lagu et al3 and Upshaw et al4. However, once assembled, our models need to be labeled "For Display Only" or validated widely and repeatedly if intended to serve as working models to navigate individual patient care. We can examine and admire the ship in a bottle, but we cannot sail away on it.DisclosuresNone.FootnotesThe opinions expressed in this article are not necessarily those of the editors or of the American Heart Association.Correspondence to Lynne Warner Stevenson, MD, Cardiovascular Division, Brigham and Women's Hospital Medicine, 75 Francis St, Boston, MA 02115. E-mail [email protected]References1. Taleb NN. The Black Swan. New York, NY: Random House; 2007.Google Scholar2. Parker LJ. http://www.brainyquote.com/quotes/quotes/l/laurencej163120.html. Accessed July 14, 2016.Google Scholar3. Lagu T, Pekow PS, Shieh M-S, Stefan M, Pack QR, Kashef MA, Atreya AR, Valania G, Slawsky MT, Lindenauer PK. Validation and comparison of seven mortality prediction models for hospitalized patients with acute decompensated heart failure.Circ Heart Fail. 2016; 9:e002912. doi: 10.1161/CIRCHEARTFAILURE.115.002912.LinkGoogle Scholar4. Upshaw JN, Konstam MA, van Klaveren D, Noubary F, Hugging GS, Kent DM. Multi state model to predict heart failure hospitalizations and all-cause mortality in outpatients with heart failure with reduced ejection fraction: model derivation and external validation.Circ Heart Fail. 2016; 9:e003146. doi: 10.1161/CIRCHEARTFAILURE.116.003146LinkGoogle Scholar5. Califf RM, Pencina MJ. Predictive models in heart failure: who cares?Circ Heart Fail. 2013; 6:877–878. doi: 10.1161/CIRCHEARTFAILURE.113.000659.LinkGoogle Scholar6. Alba AC, Agoritsas T, Jankowski M, Courvoisier D, Walter SD, Guyatt GH, Ross HJ. Risk prediction models for mortality in ambulatory patients with heart failure: a systematic review.Circ Heart Fail. 2013; 6:881–889. doi: 10.1161/CIRCHEARTFAILURE.112.000043.LinkGoogle Scholar7. Rahimi K, Bennett D, Conrad N, Williams TM, Basu J, Dwight J, Woodward M, Patel A, McMurray J, MacMahon S. Risk prediction in patients with heart failure: a systematic review and analysis.JACC Heart Fail. 2014; 2:440–446. doi: 10.1016/j.jchf.2014.04.008.CrossrefMedlineGoogle Scholar8. Stevenson LW, Pande R. Witness to progress.Circ Heart Fail. 2011; 4:390–392. doi: 10.1161/CIRCHEARTFAILURE.111.963066.LinkGoogle Scholar9. Cubbon RM, Gale CP, Kearney LC, Schechter CB, Brooksby WP, Nolan J, Fox KA, Rajwani A, Baig W, Groves D, Barlow P, Fisher AC, Batin PD, Kahn MB, Zaman AG, Shah AM, Byrne JA, Lindsay SJ, Sapsford RJ, Wheatcroft SB, Witte KK, Kearney MT. Changing characteristics and mode of death associated with chronic heart failure caused by left ventricular systolic dysfunction: a study across therapeutic eras.Circ Heart Fail. 2011; 4:396–403. doi: 10.1161/CIRCHEARTFAILURE.110.959882.LinkGoogle Scholar10. Udelson JE, Stevenson LW. The future of heart failure diagnosis, therapy, and management.Circulation. 2016; 133:2671–2686. doi: 10.1161/CIRCULATIONAHA.116.023518.LinkGoogle Scholar11. Stevenson LW. Heart failure with better ejection fraction: a modern diagnosis.Circulation. 2014; 129:2364–2367. doi: 10.1161/CIRCULATIONAHA.114.010194.LinkGoogle Scholar12. Setoguchi S, Warner Stevenson L, Stewart GC, Bhatt DL, Epstein AE, Desai M, Williams LA, Chen CY. Influence of healthy candidate bias in assessing clinical effectiveness for implantable cardioverter-defibrillators: cohort study of older patients with heart failure.BMJ. 2014; 348:g2866.CrossrefMedlineGoogle Scholar13. Digitalis Investigation Group.The effect of digoxin on mortality and morbidity in patients with heart failure.N Engl J Med1997; 336:525–533.CrossrefMedlineGoogle Scholar14. Jong P, Yusuf S, Rousseau MF, Ahn SA, Bangdiwala SI. Effect of enalapril on 12-year survival and life expectancy in patients with left ventricular systolic dysfunction: a follow-up study.Lancet. 2003; 361:1843–1848. doi: 10.1016/S0140-6736(03)13501-5.CrossrefMedlineGoogle Scholar15. Box GEP. Robustness in the strategy of scientific model building., Launer RL, Wilkinson GN. In: Robustness in Statistics. New York, New York:Academic Press; 1979:201–236.Google Scholar16. Johnson G, Carson P, Francis GS, Cohn JN. Influence of prerandomization (baseline) variables on mortality and on the reduction of mortality by enalapril. Veterans Affairs Cooperative Study on Vasodilator Therapy of Heart Failure (V-HeFT II). V-HeFT VA Cooperative Studies Group.Circulation. 1993; 87(suppl 6):VI32–VI39.MedlineGoogle Scholar17. Lee WH, Packer M. Prognostic importance of serum sodium concentration and its modification by converting-enzyme inhibition in patients with severe chronic heart failure.Circulation. 1986; 73:257–267.CrossrefMedlineGoogle Scholar18. Saxon LA, Stevenson WG, Middlekauff HR, Stevenson LW. Increased risk of progressive hemodynamic deterioration in advanced heart failure patients requiring permanent pacemakers.Am Heart J. 1993; 125(5 pt 1):1306–1310.CrossrefMedlineGoogle Scholar19. Dries DL, Exner DV, Domanski MJ, Greenberg B, Stevenson LW. The prognostic implications of renal insufficiency in asymptomatic and symptomatic patients with left ventricular systolic dysfunction.J Am Coll Cardiol. 2000; 35:681–689.CrossrefMedlineGoogle Scholar20. Cooper LB, Mentz RJ, Stevens SR, Felker GM, Lombardi C, Metra M, Stevenson LW, O'Connor CM, Milano CA, Patel CB, Rogers JG. Hemodynamic predictors of heart failure morbidity and mortality: fluid or flow?J Card Fail. 2016; 22:182–189. doi: 10.1016/j.cardfail.2015.11.012.CrossrefMedlineGoogle Scholar21. Masson S, Latini R, Anand IS, Barlera S, Angelici L, Vago T, Tognoni G, Cohn JN; Val-HeFT Investigators. Prognostic value of changes in N-terminal pro-brain natriuretic peptide in Val-HeFT (Valsartan Heart Failure Trial).J Am Coll Cardiol. 2008; 52:997–1003. doi: 10.1016/j.jacc.2008.04.069.CrossrefMedlineGoogle Scholar22. Doust JA, Pietrzak E, Dobson A, Glasziou P. How well does B-type natriuretic peptide predict death and cardiac events in patients with heart failure: systematic review.BMJ. 2005; 330:625.CrossrefMedlineGoogle Scholar23. Krumholz HM, Wang Y, Mattera JA, Wang Y, Han LF, Ingber MJ, Roman S, Normand SL. An administrative claims model suitable for profiling hospital performance based on 30-day mortality rates among patients with heart failure.Circulation. 2006; 113:1693–1701. doi: 10.1161/CIRCULATIONAHA.105.611194.LinkGoogle Scholar24. Stewart GC, Mountis M, Kittleson M, Ambardekar AV, Palardy M, Forde-McLean RC, Thobodeau JT, Teuteberg JJ, Pamboukian SV, Devore AD, Cadaret L, Cantor RS, Stevenson LW. Early reassessment of ambulatory advanced heart failure detects high-risk trajectories along a dynamic baseline.J Heart Lung Trans2015; 34:S56.CrossrefGoogle Scholar25. Siontis GC, Tzoulaki I, Ioannidis JP. Predicting death: an empirical evaluation of predictive tools for mortality.Arch Intern Med. 2011; 171:1721–1726. doi: 10.1001/archinternmed.2011.334.CrossrefMedlineGoogle Scholar26. Henderson R, Keiding N. Individual survival time prediction using statistical models.J Med Ethics. 2005; 31:703–706. doi: 10.1136/jme.2005.012427.CrossrefMedlineGoogle Scholar27. Yancy CW, Jessup M, Bozkurt B, Butler J, Casey DE, Drazner MH, Fonarow GC, Geraci SA, Horwich T, Januzzi JL, Johnson MR, Kasper EK, Levy WC, Masoudi FA, McBride PE, McMurray JJ, Mitchell JE, Peterson PN, Riegel B, Sam F, Stevenson LW, Tang WH, Tsai EJ, Wilkoff BL. 2013 ACCF/AHA guideline for the management of heart failure: executive summary: a report of the American College of Cardiology Foundation/American Heart Association Task Force on practice guidelines.Circulation. 2013; 128:1810–1852. doi: 10.1161/CIR.0b013e31829e8807.LinkGoogle Scholar28. Allen LA, Stevenson LW, Grady KL, Goldstein NE, Matlock DD, Arnold RM, Cook NR, Felker GM, Francis GS, Hauptman PJ, Havranek EP, Krumholz HM, Mancini D, Riegel B, Spertus JA; American Heart Association; Council on Quality of Care and Outcomes Research; Council on Cardiovascular Nursing; Council on Clinical Cardiology; Council on Cardiovascular Radiology and Intervention; Council on Cardiovascular Surgery and Anesthesia. Decision making in advanced heart failure: a scientific statement from the American Heart Association.Circulation. 2012; 125:1928–1952. doi: 10.1161/CIR.0b013e31824f2173.LinkGoogle Scholar29. Butler J, Binney Z, Kalogeropoulos A, Owen M, Clevenger C, Gunter D, Georgiopoulou V, Quest T. Advance directives among hospitalized patients with heart failure.JACC Heart Fail. 2015; 3:112–121. doi: 10.1016/j.jchf.2014.07.016.CrossrefMedlineGoogle Scholar30. Setoguchi S, Stevenson LW, Schneeweiss S. Repeated hospitalizations predict mortality in the community population with heart failure.Am Heart J. 2007; 154:260–266. doi: 10.1016/j.ahj.2007.01.041.CrossrefMedlineGoogle Scholar Previous Back to top Next FiguresReferencesRelatedDetails August 2016Vol 9, Issue 8Article InformationMetrics Download: 447 © 2016 American Heart Association, Inc.https://doi.org/10.1161/CIRCHEARTFAILURE.116.003457PMID: 27514752 Originally publishedAugust 11, 2016 KeywordsEditorialsuncertaintycardiomyopathystatistical modelheart failurerisk assessmentPDF download SubjectsHeart Failure