Author + information
- Received January 9, 2019
- Revision received March 29, 2019
- Accepted March 31, 2019
- Published online July 29, 2019.
- Søren Lund Kristensen, MD, PhDa,b,∗ (, )
- Wayne C. Levy, MDc,
- Ramin Shadman, MDd,
- Jens C. Nielsen, MD, DMSce,
- Jens Haarbo, MD, DMScf,
- Lars Videbæk, MD, PhDg,
- Niels E. Bruun, MD, DMSch,i,
- Hans Eiskjær, MD, DMSce,
- Henrik Wiggers, MDe,
- Axel Brandes, MDg,
- Anna Margrethe Thøgersen, MD, PhDj,
- Christian Hassager, MD, DMSca,k,
- Jesper H. Svendsen, MD, DMSca,k,
- Dan E. Høfsten, MD, PhDa,
- Christian Torp-Pedersen, MD, DMScj,
- Steen Pehrson, MD, DMSca,
- James Signorovitch, PhDl,
- Lars Køber, MD, DMSca,k and
- Jens Jakob Thune, MD, PhDb,k
- aDepartment of Cardiology, Rigshospitalet, Copenhagen University Hospital, Copenhagen, Denmark
- bDepartment of Cardiology, Bispebjerg University Hospital, Copenhagen, Denmark
- cUniversity of Washington, Seattle, Washington
- dSouthern California Permanente Medical Group, Los Angeles, California
- eDepartment of Cardiology, Aarhus University Hospital, Aarhus, Denmark
- fDepartment of Cardiology, Gentofte University Hospital, Copenhagen, Denmark
- gDepartment of Cardiology, Odense University Hospital, Odense, Denmark
- hDepartment of Cardiology, Zealand University Hospital, Roskilde, Denmark
- iClinical Institute, Copenhagen and Aalborg Universities, Denmark
- jDepartment of Cardiology, Aalborg University Hospital, Aalborg, Denmark
- kDepartment of Clinical Medicine, Faculty of Health and Medical Sciences, University of Copenhagen, Denmark
- lAnalysis Group, Boston, Massachusetts
- ↵∗Address for correspondence:
Dr. Søren Lund Kristensen, Department of Cardiology, Bispebjerg Hospital, University of Copenhagen, Copenhagen 2100, Denmark.
Objectives This study aims to identify patients with nonischemic heart failure who are more likely to benefit from implantable cardioverter-defibrillator (ICD) implantation by use of established risk prediction models.
Background It has been debated whether an ICD for primary prevention reduces mortality in patients with nonischemic heart failure.
Methods The Seattle Heart Failure Model (SHFM) predicts all-cause mortality whereas the Seattle Proportional Risk Model (SPRM) predicts the proportion of sudden cardiac death (SCD) versus nonsudden death, with a higher score indicating a greater proportion of SCD. We report the effect of ICD implantation on all-cause mortality and SCD, according to median SPRM and SHFM scores in all 1,116 patients enrolled in the DANISH (Danish study to Assess the Efficacy of ICDs in Patients with Non-ischemic Systolic Heart Failure on mortality) trial.
Results Among patients with an SPRM score above the median (n = 558), ICD implantation reduced all-cause mortality (hazard ratio [HR]: 0.63; 95% confidence interval [CI]: 0.43 to 0.94), whereas patients with lower SPRM scores (n = 558) had no effect (HR: 1.08; 95% CI: 0.78 to 1.49, p for interaction = 0.04). The corresponding numbers for SHFM score above and below the median were HR: 0.84; 95% CI: 0.62 to 1.13 and HR: 0.82; 95% CI: 0.53 to 1.28, respectively (p for interaction = 0.980). In 177 patients with upper SPRM/upper SHFM, ICD implantation reduced all-cause mortality (HR: 0.45; 95% CI: 0.25 to 0.80) when compared to 381 patients with lower SPRM/upper SHFM (HR: 1.09; 95% CI: 0.76 to 1.55) (p for interaction <0.001).
Conclusions Nonischemic heart failure patients with high predicted relative likelihood of SCD, as estimated by higher SPRM score, seemed to benefit from ICD implantation. (DANISH [Danish ICD Study in Patients With Ditaled Cardiomyopathy]; NCT00542945)
The use of an implantable cardioverter-defibrillator (ICD) for primary prevention is recommended in patients with heart failure (HF) and reduced ejection fraction irrespective of etiology (1–3). In patients with ischemic HF, the absolute risks of all-cause and sudden cardiac death (SCD) are higher, and the evidence for ICD benefit stronger, than in patients with nonischemic HF, even if the relative risk reduction is similar (4–9). As ICDs mainly work by terminating ventricular arrythmia, the impact of ICD implantation is related to the patient’s risk of SCD. Recently, the DANISH (Danish Study to Assess the Efficacy of ICDs in Patients with Non-ischemic Systolic Heart Failure on Mortality) study found no benefit on all-cause mortality (10). However, a subgroup analysis of DANISH indicated mortality benefit from ICD implantation in the younger patients, and in a post hoc analysis this corresponded to a lower proportion of SCDs relative to nonsudden deaths with increasing age (11).
In the present study we applied combinations of previously validated HF risk prediction models to identify patients with nonischemic HF with a greater risk of SCD, both relatively and absolutely, who would be more likely to benefit from an ICD. We used the Seattle Proportional Risk Model (SPRM) which estimates the proportional risk of sudden versus nonsudden death, and the Seattle Heart Failure Model (SHFM) which estimates overall mortality, in patients enrolled in the DANISH trial (Central Illustration) (12,13).
DANISH was a randomized trial studying the effect of ICD implantation in patients with nonischemic HF on mortality. The design and main findings of DANISH are published (10,14). Briefly, patients were eligible if they had a nonischemic cause of HF, were in New York Heart Association (NYHA) functional class II to III or NYHA functional class IV if cardiac resynchronization therapy (CRT) was planned, had a documented left ventricular ejection fraction (LVEF) ≤35% and N-terminal pro-B-type natriuretic peptide (NT-proBNP) level >200 pg/ml (23.6 pmol/l). The trial was approved by regional ethics committees, and all patients provided written informed consent.
Risk stratification and outcomes
We applied the SPRM and SHFM multivariable risk models on all 1,116 patients from DANISH. The SPRM score predicts the likelihood of SCD versus nonsudden death with the following parameters increasing relative likelihood of SCD: younger age, male sex, low LVEF, NYHA functional class II versus NYHA functional class III/IV, higher body mass index, and use of digoxin. Conversely, diabetes, hyper/hypotension, renal dysfunction, and hyponatremia reduce relative SCD likelihood as estimated by SPRM (13). SHFM predicts all-cause mortality with these parameters associated with an increased risk: older age, male sex, higher NYHA functional class, reduced LVEF, ischemic etiology, hyponatremia, uric acid, and low/high hemoglobin, whereas high systolic blood pressure, high cholesterol or lymphocytes, use of statins, beta blockers, mineralocorticoid receptor antagonists, and having an ICD or CRT are associated with lower risk (12). We did not include information on ICD in our estimation of SHFM score as our goal was to estimate the ICD benefit adjusted for the risk models.
In sensitivity analyses, we repeated the above-mentioned analyses in patients with and without CRT separately (Online Tables 1 and 2), and we assessed ICD implantation effect for quartiles of SPRM and SHFM scores separately and irrespective of other factors (Online Tables 3 and 4).
Baseline differences between patients according to median SPRM and SHFM scores were tested by use of chi-square test for categorical variables and analysis of variance or Kruskal-Wallis test for continuous variables. Event rates were reported per 100 person-years. We derived Kaplan-Meier curves for all-cause mortality, and cumulative incidence curves for SCD with death from nonsudden causes as a competing risk by use of the Aalen-Johansen method (15). Log rank test and Gray’s test were used to analyze unadjusted differences, respectively. Cox proportional hazard models were used to assess the ICD implantation effects within risk stratification groups, and to test for interactions between risk groups, ICD implantation, and outcomes. Analyses were performed by use of Stata version 14 (StataCorp, College Station, Texas).
Baseline characteristics according to median SPRM and SHFM scores are presented in Table 1. Patients with SPRM above versus below median were less likely to be scheduled for CRT (49% vs. 67%), had better kidney function (median estimated glomerular filtration rate 79 ml/min/1.73 m2 vs. 59 ml/min/1.73 m2), and lower level of NT-proBNP (948 pg/ml vs. 2,015 pg/ml). Patients with SHFM score above the median versus below were more likely to have a history of atrial fibrillation (50% vs. 31%) and be scheduled for CRT (69% vs. 46%).
Event rates in patients with and without ICD according to risk stratification are depicted in Table 2 and Figure 1. Among 558 patients with SPRM score above the median, event rates of all-cause mortality per 100 patient-years were 2.9 (95% CI: 1.7 to 5.0) in the ICD group versus 4.4 (95% CI: 3.4 to 5.7) among controls (hazard ratio [HR]: 0.63; and 95% confidence interval [CI]: 0.43 to 0.94). In 558 patients with an SPRM score below the median, event rates were 6.1 (95% CI: 4.9 to 7.6) and 5.6 (95% CI: 4.4 to 7.1) per 100 patient-years, respectively, with an HR of 1.08; 95% CI: 0.78 to 1.49 for ICD versus no ICD. The p value for interaction between SPRM and randomization to ICD was 0.04. Patients with SHFM scores above versus below median did not display any difference in effect of ICD implantation (Table 2) (p for interaction = 0.980). Among patients with an SHFM score below median, a higher SPRM score was not associated with a significantly different ICD implantation effect when compared to a lower SPRM score (p for interaction = 0.552) (Figure 2A), whereas among patients with an SHFM score above the median, there was a significant difference in ICD implantation effect according to SPRM: higher SPRM, event rates ICD group 3.8 (95% CI: 2.4 to 5.9) versus controls 8.2 (95% CI: 5.7 to 11.8) per 100 patient-years (HR: 0.45, 95% CI: 0.25 to 0.80). In patients with upper SHFM and lower SPRM event rates were 7.8 (95% CI: 6.1 to 9.9) in ICD groups versus 7.1 (95% CI: 5.5 to 9.2) in controls per 100 patient-years (HR: 1.09, 95% CI: 0.76 to 1.55) (p for interaction = 0.011). We found similar patterns and significant interactions when repeating analyses restricted to the 645 patients with CRT (data not shown).
Application of the SPRM score in the DANISH study predicted a mean proportion of SCD versus nonsudden death of 54% (in medically treated patients), which was somewhat higher than the 35% we observed among the control group, and 20% among those who received an ICD (Online Table 3). In Figure 2 the cumulative incidence of SCD and nonsudden death is shown in patients with SPRM scores below and above the median. After risk stratification according to SPRM and SHFM, we analyzed ICD implantation effects on SCD across groups but found no significant interactions. The interpretability of these analyses was somewhat limited due to the relatively low overall number of SCD events (n = 70). We found a similar pattern of ICD benefit when we repeated analyses in patients with (n = 645) and without CRT (n = 471), although interaction analyses did not reach statistical significance (Online Tables 1 and 2).
Absolute mortality reduction
The absolute survival at 7-year follow-up according to treatment arm (ICD or control) is listed in Table 3. In the full study population, difference in 7-year survival between the ICD versus the control group was 2.4% (95% CI: −5.3% to 9.1%). Among patients with SPRM above the median, mortality benefit was 10.0% (95% CI: 1.3% to 18.7%), and among patients 70 years of age or younger it was 8.4% (95% CI: −1.8% to 18.4%). The strongest benefit was seen among patients with upper SHFM/upper SPRM with 19.3% absolute mortality reduction (95% CI: 2.8% to 35.8%).
In the present study we found that for patients with nonischemic HF at a predicted high relative likelihood of SCD (SPRM score above the median), ICD implantation was associated with a 37% relative reduction in all-cause mortality, and the reduction was even larger (55%) among patients who in addition had a predicted high absolute mortality. In contrast, for patients with a low SPRM, ICD implantation was not associated with reduced mortality or any signal of harm, irrespective of absolute mortality risk.
We did not see any difference in absolute rates of SCD according to median SPRM, indicating that high SPRM did not identify patients at higher absolute SCD risk, but rather patients at lower risk of competing nonsudden causes of death. This seemingly paradoxical finding is due to the nature of the SPRM model which was derived from differences between HF patients who died suddenly versus nonsudden death, and did not take their absolute mortality risk into consideration (13).
There were no statistically significant differences in effects of ICD implantation on SCD according to SPRM and SHFM subgroups. Although this finding may not be surprising, 1 reason could be the low number of SCD events (n = 70). Finally, we assessed ICD benefit estimates separately in patients with and without CRT and similarly to the main analysis; we observed a pattern of benefit in those with SPRM above the median, particularly in presence of high SHFM irrespective of CRT status (non-CRT HR: 0.37 [95% CI: 0.15 to 0.90] and CRT HR: 0.50 [95% CI: 0.23 to 1.08]).
The selection of patients for primary prevention ICDs relates to the relative and absolute risk of SCD. Suggestions for ICD indication in patients with nonischemic HF range from no indication at all to the guideline-recommended approach with similar criteria as for patients with ischemic cardiomyopathy (11). Some researchers have even suggested re-evaluating the ICD indication in all HF patients irrespective of etiology on the basis of the DANISH study and an observed trend of declining rates of SCD over time across HF trials potentially due to the improvement in drug therapy (16,17). Our analysis and the prior DANISH-age publication suggest that risk stratification may identify nonischemic HF patients who derive a meaningful benefit from an ICD (11).
Differences in absolute mortality at 7-year follow-up for the ICD group versus controls were 2.4% in the overall population but rose to 8% among patients 70 years of age or younger, to 10% in those with SPRM above the median, and to 20% in the subgroup of patients who had both a high SPRM and SHFM scores, although with wide confidence intervals.
In an analysis of the primary prevention SCD-HeFT (Sudden Cardiac Death in Heart Failure Trial) study, dividing patients by SHFM score quintiles displayed ICD benefit in all but those with highest predicted overall mortality (18). Varying ICD benefit with the SPRM was observed in the HF-ACTION (Heart Failure and A Controlled Trial Investigating Outcomes of Exercise Training) study and the National Cardiovascular Data Registry ICD registry. In the National Cardiovascular Data Registry ICD registry, patients with SHFM <4.75% and SPRM <47.5% had no ICD benefit (ICD HR: 0.92; 95% CI: 0.78 to 1.08), similar to the DANISH results (19–21). In DANISH, we found no significant interaction between SHFM quartile and treatment effect although we found no effect of ICD implantation in those with SHFM score in the highest quartile, similar to the findings in SCD-HeFT (Online Table 2). However, there were very few patients in DANISH (5%) with SHFM ≥16.5% annual mortality (10th decile), where the SHFM identified no ICD benefit within SCD-HeFT. Overall mortality and likelihood of SCD was somewhat overestimated by SPRM and SHFM but nonetheless worked fairly well as risk stratification tools (Online Table 5).
Most importantly, the present analyses were post hoc and were not pre-specified when DANISH was designed. The risk prediction is based on clinical variables evaluated at baseline, and we cannot surely extrapolate our findings beyond the mean follow-up of 67 months in DANISH. Particularly, hospitalizations or other signs of worsening HF during follow-up may merit re-evaluation and reclassification of SCD risk. Furthermore, additional caution is warranted when interpreting results of subgroup analyses in studies where the overall result is neutral (as is the case in DANISH). For these reasons, the findings from the current study should be considered hypothesis-generating and would need further validation.
Among patients with nonischemic systolic HF who had a high predicted relative likelihood of dying suddenly, fewer deaths occurred in the group that was randomized to an ICD. Risk prediction models of absolute and relative risk of SCD could potentially aid in identifying patients more likely to benefit from an ICD.
COMPETENCY IN MEDICAL KNOWLEDGE: The benefit of implanting an ICD for primary prevention of SCD in patients with nonischemic systolic HF has been questioned due to improved pharmacologic treatment. Our exploratory analyses of the DANISH trial suggest that risk prediction models may add guidance in identifying patients more likely to benefit from an ICD.
TRANSLATIONAL OUTLOOK: Implantation of an ICD reduces the risk of SCD by approximately 50% in patients with nonischemic systolic HF, irrespective of predicted risk. More knowledge is needed in terms of understanding the pathophysiologic mechanisms leading to refractory malignant arrhythmia.
The DANISH trial was supported by unrestricted grants from Medtronic, St. Jude Medical, TrygFonden, and the Danish Heart Foundation. The University of Washington Comotion owns the copyrights to the SHFM and SPRM and has received licensing fees for the SHFM from various companies. Dr. Levy is on Steering Committee for GE Healthcare; is on the Clinical Endpoint Committee for CardioMems (Abbott) and EBR Systems, Inc.; and is a consultant to Novartis and Impulse Dynamics. Dr. Nielsen has received a research grant from Novo Nordisk Foundation. Dr. Videbæk has received fees for serving on advisory boards from Novartis, AstraZeneca, and Boehringer Ingelheim; and has received lecture fees from Novartis. Dr. Bruun has received lecture fees from Biotronik. Dr. Brandes has received lecture fees from Bayer, Boehringer Ingelheim, MSD, and Pfizer. Dr. Svendsen has received fees for serving on an advisory board from Medtronic; has received lecture fees from Medtronic, Biotronik, AstraZeneca, and Boehringer Ingelheim; and has received grant support from Medtronic, Biotronik, and Gilead Sciences. Dr. Torp-Pedersen has received lecture fees and grant support from Bayer. Dr. Pehrson has received lecture fees from Bayer, Boehringer Ingelheim, Bristol-Myers Squibb, and Servier. Dr. Køber has received lecture fees from Sanofi and Novartis. Dr. Thune has received lecture fees from Bristol-Myers Squibb; and has received personal fees and travel support from Novartis. All other authors have reported that they have no relationships relevant to the contents of this paper to disclose.
- Abbreviations and Acronyms
- cardiac resynchronization therapy
- implantable cardioverter-defibrillator
- sudden cardiac death
- Seattle Heart Failure Model
- Seattle Proportional Risk Model
- Received January 9, 2019.
- Revision received March 29, 2019.
- Accepted March 31, 2019.
- 2019 American College of Cardiology Foundation
- Ponikowski P.,
- Voors A.A.,
- Anker S.D.,
- et al.
- Yancy C.W.,
- Jessup M.,
- Bozkurt B.,
- et al.
- Al-Khatib S.M.,
- Stevenson W.G.,
- Ackerman M.J.,
- et al.
- Bansch D.,
- Antz M.,
- Boczor S.,
- et al.
- Strickberger S.A.,
- Hummel J.D.,
- Bartlett T.G.,
- et al.
- Elming M.B.,
- Nielsen J.C.,
- Haarbo J.,
- et al.
- Levy W.C.,
- Mozaffarian D.,
- Linker D.T.,
- et al.
- Shadman R.,
- Poole J.E.,
- Dardas T.F.,
- et al.
- Thune J.J.,
- Pehrson S.,
- Nielsen J.C.,
- et al.
- Aalen O.O.,
- Johansen S.
- Petrie M.C.,
- Connelly D.T.,
- Gardner R.S.
- Shen L.,
- Jhund P.S.,
- McMurray J.J.V.
- Levy W.C.,
- Lee K.L.,
- Hellkamp A.S.,
- et al.
- Levy W.C.,
- Li Y.,
- Reed S.D.,
- et al.
- Bilchick K.C.,
- Wang Y.,
- Cheng A.,
- et al.
- Levy W.C.,
- Hellkamp A.S.,
- Mark D.B.,
- et al.