Preventing Sudden Cardiac Death: Harnessing the Power of Decision Analysis, Bayesian Techniques, and Clinical Trials
On September 21, 2011, Gillian D. Sanders and Lurdes Y. Inoue made this presentation at the 2011 Annual Conference. Select to access the PowerPoint® presentation (1.49 MB). Plugin Software Help.
Slide 1
Preventing Sudden Cardiac Death: Harnessing the Power of Decision Analysis, Bayesian Techniques, and Clinical Trials
Gillian D Sanders Ph.D.
Associate Professor of Medicine
Duke University
Lurdes Y Inoue Ph.D.
Associate Professor of Biostatistics
University of Washington
Funded by AHRQ R01-HS018505
Slide 2
Specific Aims
- To develop a generalizable decision modeling framework for the prevention of SCD.
- To use Bayesian statistical techniques to devise a model for predicting patient and population health and economic outcomes.
- To use the framework and Bayesian model from Specific Aims 1-2, and patient level data from existing clinical trials, to explore timely clinical and policy questions.
- To develop a web-based dissemination system to allow providers and policy makers to interact with the decision modeling framework and to explore clinical and policy questions as evidence evolves.
Slide 3
Timing of ICD Clinical Trials
Image: A t-table shows the major ICD RCTs and their timing. Each trial is listed in a row, and then years are shown in the columns with the shaded area showing when the trial was ongoing.
Those in green are considered “secondary prevention trials”—that is trials in patients who have previously experienced a sudden cardiac arrest and therefore are at high risk for a recurrent event
In blue are the “primary prevention” trials—those that look at prophylactically implanting an ICD in patients who are at increased risk compared to the general population—but who haven't had a previous ventricular event
Slide 4
Timing of ICD Clinical Trials
An image showing when the findings from these different trials were published in the clinical literature from 1987-2005 is shown.
Slide 5
Clinical Characteristics of Trial Patients
Trial | Tx | Num | Age | LVEF | % Ischemic | % NYHA Class | |||||
---|---|---|---|---|---|---|---|---|---|---|---|
Mean | %≥65y | Mean | %≤30 | I | II | III | IV | ||||
AVID | Ctrl | 509 | 65.33 | 57.76 | 30.82 | 58.22 | 85.07 | 61.49 | 26.72 | 11.79 | 0 |
ICD | 507 | 64.83 | 54.83 | 32.15 | 54.17 | 85.80 | 64.89 | 28.40 | 6.71 | 0 | |
CABG | Ctrl | 454 | 64.95 | 50.00 | 27.05 | 71.15 | 100.0 | 56.95 | 18.76 | 17.88 | 6.4 |
ICD | 446 | 64.07 | 50.00 | 27.13 | 71.08 | 100.0 | 55.88 | 19.68 | 16.52 | 7.92 | |
CASH | Ctrl | 189 | 57.83 | 23.28 | 45.18 | 20.47 | 88.83 | 29.35 | 57.61 | 13.04 | 0 |
ICD | 99 | 57.46 | 27.27 | 45.89 | 24.21 | 88.89 | 24.49 | 57.14 | 18.37 | 0 | |
DEFINITE | Ctrl | 229 | 58.11 | 33.19 | 21.84 | 93.89 | 0 | 17.90 | 60.70 | 21.40 | 0 |
ICD | 229 | 58.41 | 35.37 | 20.88 | 95.63 | 0 | 25.33 | 54.15 | 20.52 | 0 | |
MADIT-I | Ctrl | 101 | 63.8 | 51.49 | 24.57 | 83.17 | 100.0 | 32.67 | 49.50 | 17.82 | 0 |
ICD | 95 | 62.12 | 44.21 | 26.66 | 69.47 | 100.0 | 37.89 | 46.32 | 15.79 | 0 | |
MADIT-II | Ctrl | 490 | 64.57 | 53.47 | 23.16 | 99.59 | 100.0 | 38.80 | 34.23 | 22.82 | 4.15 |
ICD | 742 | 64.45 | 53.50 | 23.17 | 100 | 100.0 | 34.83 | 25.34 | 25.44 | 4.49 | |
MUSTT | Ctrl | 353 | 64.87 | 54.11 | 27.65 | 64.87 | 100.0 | 36.41 | 38.46 | 25.13 | 0 |
ICD | 167 | 65.42 | 56.89 | 27.72 | 65.27 | 100.0 | 34.55 | 39.09 | 29.36 | 0 | |
SCDHEFT | Ctrl | 847 | 58.58 | 33.53 | 25.71 | 60.57 | 53.48 | 0 | 70.13 | 29.87 | 0 |
ICD | 829 | 59.41 | 35.46 | 24.96 | 61.40 | 51.99 | 0 | 68.28 | 31.72 | 0 |
Slide 6
Clinical and Policy Questions
- Controlling for EF, ischemia, age, and NYHA class—are the patients within the available trials similar?
- Is there evidence that the devices used during the different trials differ in terms of their efficacies?
- Is there evidence that the ICD is effective in patients:
- Over 65 years of age? Over 75?
- With EF > 30%.
- With non-ischemic disease.
- With NYHA class I? II? III? IV?
- Are there specific patient subgroups for which:
- The ICD is particularly ineffective or effective?
- Decision makers might benefit from additional trial data?
Slide 7
Methods
- Considered patient-level data from 8 trials (MADIT-I, -II, MUSTT, DEFINITE, SCDHeFT, AVID, CASH, CABG).
- Primary outcome is overall survival.
- Treatment: ICD versus control.
- Prognostic variables:
- Age (years).
- Ejection fraction (%).
- NYHA class (I, II, III, IV).
- Presence of ischemic disease (yes/no).
- Used Bayesian hierarchical model to explore data from trials and specific subgroups of interest.
Slide 8
Does ICD Efficacy Differ Among Trials?
An image showing the different trials are shown. Going up the y axis—with the overall findings combining data from all trials at the top in black.
The x-axis indicates the treatment effect with the vertical dashed line at 0 indicating where no effect was demonstrated from ICD therapy.
Slide 9
Is the ICD Effective?
Trial | Hazard Ratio | Probability HR ≤ 0.80 | ||
---|---|---|---|---|
2.5% | 50% | 97.5% | ||
AVID | 0.50 | 0.64 | 0.86 | 0.92 |
CABG-PATCH | 0.82 | 1.04 | 1.41 | 0.02 |
CASH | 0.56 | 0.83 | 1.25 | 0.43 |
DEFINITE | 0.39 | 0.62 | 0.95 | 0.87 |
MADIT-I | 0.26 | 0.45 | 0.71 | 0.99 |
MADIT-II | 0.48 | 0.62 | 0.82 | 0.95 |
MUSTT | 0.28 | 0.43 | 0.64 | 1.00 |
SCD-HeFT | 0.60 | 0.73 | 0.91 | 0.82 |
Overall | 0.41 | 0.65 | 1.03 | 0.82 |
Slide 10
Is the ICD Effective in Patients Over 65?
Image of a line chart which looks at the evidence of ICD efficacy in patients over age 65 and less than 75.
Slide 11
Is the ICD Effective in Patients With EF > 30%
Image of a line chart which looks at the evidence of ICD efficacy in patients with an ejection fraction greater than 30%.
Slide 12
Is the ICD Effective in NYHA II Patients?
Image of a line chart which looks at the evidence of ICD efficacy in NYHA Class II Patients.
Slide 13
Is the ICD Effective in NYHA III Patients?
Image of a line chart which looks at the evidence of ICD efficacy in NYHA Class III Patients.
Slide 14
Is the ICD Effective in NYHA IV Patients?
Image of a line chart which looks at the evidence of ICD efficacy in NYHA Class IV Patients.
Slide 15
Is the ICD Effective in NYHA IV Patients?
Trial | Hazard Ratio | Probability HR ≤ 0.80 | ||
---|---|---|---|---|
2.5% | 50% | 97.5% | ||
AVID | 0.03 | 0.84 | 18.64 | 0.48 |
CABG-PATCH | 0.34 | 1.00 | 3.07 | 0.35 |
CASH | 0.03 | 0.57 | 11.70 | 0.61 |
DEFINITE | 0.04 | 0.75 | 14.64 | 0.53 |
MADIT-I | 0.03 | 0.67 | 12.22 | 0.56 |
MADIT-II | 0.37 | 1.63 | 5.37 | 0.16 |
MUSTT | 0.04 | 0.86 | 19.27 | 0.47 |
SCD-HeFT | 0.02 | 0.77 | 13.96 | 0.51 |
Overall | 0.20 | 0.84 | 3.43 | 0.49 |
Slide 16
Is the ICD Effective in Ischemic Patients?
An image of a line graph showing the evidence supporting ICD effectiveness in patients with ischemic disease is shown.
Slide 17
Are Patients Within The Trials Similar?
An image of a line graph showing the estimated posterior baseline survival functions under each trial and over all trials is shown.
Slide 18
ICD Effect in Specific Patient Subgroups
Age Group | EF | NYHA | Ischemia | Hazard Ratio | Probability HR ≤ 0.80 | ||
---|---|---|---|---|---|---|---|
Lower | Median | Upper | |||||
< 65 | < 30% | I | No | 0.29 | 0.59 | 1.09 | 0.84 |
< 65 | < 30% | I | Yes | 0.24 | 0.57 | 1.21 | 0.80 |
< 65 | < 30% | II | No | 0.26 | 0.62 | 1.29 | 0.77 |
75+ | < 30% | I | No | 0.23 | 0.58 | 1.40 | 0.78 |
75+ | < 30% | I | Yes | 0.20 | 0.56 | 1.52 | 0.77 |
Slide 19
Clinical and Policy Research Priorities
- Are there specific patient subgroups for which policy makers might benefit from additional clinical trial data and how best can such trials be designed given the available prior information from the existing clinical trials?
- Are there clinical subgroups within the population at risk for SCD where the ICD appears to be particularly effective? Cost effective? Futile?
- How best can we use current (and novel) risk stratification techniques to either rule in "low-risk" patients who are currently ineligible for ICD therapy, or rule out current "high-risk" patients who are currently recommended ICD implantation?
- What are effects of comorbid diseases on the effectiveness and cost effectiveness of ICD use in the elderly population for primary or secondary prevention of SCD?
- Are the outcomes observed in the community predicted by the available clinical trial evidence?
- What is the cumulative survival benefit from SCD prevention therapies in the US?
- How can the clinical trial data be best used to predict the prognosis of patients within the CMS ICD registry and as longitudinal data becomes available for the registry participants—how does our modeling framework predict the patients' outcomes?
Slide 20
Next Steps
- Incorporate data from DINAMIT trial.
- Combine full patient level data sets.
- Further develop underlying Bayesian models.
- Explore main prognostic variables of: sex, age, NYHA class, LVEF, prior MI, and QRS duration.
- Look at additional endpoints of sudden cardiac death, rehospitalizations, quality of life, appropriate/ inappropriate shocks.
- Explore other potential prognostic variables including race, time from MI /CABG, renal disease, and ICD programming and types.
- Combine with decision analytic framework in a format that clinical providers and policymakers can use to explore the underlying evidence, our models, and the findings.