If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Geriatric Research, Education and Clinical Center (GRECC),VA Boston Healthcare System, Boston, MADivision of Aging, Department of Medicine, Brigham and Women's Hospital, Harvard Medical School, Boston, MA
GJ is a 79-year-old woman with hypertension, diabetes, osteoporosis, depression, and New York Heart Association class II heart failure with a left ventricular ejection fraction of 30%. She is a potential candidate for an implantable cardioverter-defibrillator (ICD), and you would like to discuss this with her using evidence from a clinical trial. Which of the following statistics would be most helpful in explaining the possible survival benefit of an ICD?
The P value comparing mortality of ICD and placebo groups was 0.007.
The hazard ratio (HR) for mortality was 0.77.
The absolute risk reduction was 7%, from 36% to 29%, over 5 years.
The number-needed-to-treat (NNT) was 15 over 5 years.
An ICD will prolong life from 49.1 to 51.4 months, an average of 2.3 months, over 5 years.
Shared decision making has become a central aspect of providing goal-concordant care. Recently, the Centers for Medicare and Medicaid Services issued directives requiring physicians and patients to engage in shared decision making conversations, particularly for procedures such as ICD therapy for primary prevention.
One challenge to successfully implementing such conversations is the ability of clinicians to accurately interpret statistical conclusions from clinical trials and present the data in a way that is meaningful to patients. Conventional statistics reported in clinical trials, such as P values, relative risk reduction (eg, HR), absolute risk reduction, and NNT require statistical sophistication and are often difficult for patients to understand.
Translational statistics aim to bridge the divide between complex statistical analyses and clinically actionable information. A translational statistic is an empirical summary derived from the data in a clinical study that is readily comprehended by practitioners and patients, thereby facilitating shared decision making conversations. Unfortunately, most clinical studies do not provide translational statistics. For instance, a statistically significant P value at the arbitrary threshold of 0.05 does not necessarily indicate clinically meaningful treatment benefit. Equally, a nonsignificant P (≥0.05) cannot exclude the possibility of such a benefit. This was the key point in a recent statement from the American Statistical Association
and is 1 reason the P value is not a translational statistic.
The effect of a treatment is usually quantified by contrasting the risk of clinical events between treatment groups. As strongly emphasized in the recent International Conference on Harmonisation of Technical Requirements for Registration of Pharmaceuticals for Human Use guidelines for industry and regulatory agencies,
represents a ratio of hazards between groups. Since the Cox model only estimates the ratio without providing actual hazard rates, the clinical meaning of HR cannot be easily determined. Furthermore, if the ratio of 2 hazards is not constant over time (ie, the proportional hazards assumption is not met), the resulting HR may misrepresent the treatment effect.
The absolute risk reduction and NNT are often reported in conjunction with HRs. Because these measures can minimize overinterpretation of treatment effect, they are commonly used in decision aids (eg, the Colorado Program for Patient Centered Decisions
). However, the absolute risk reduction and NNT, which are calculated at a specific time point (eg, study end), do not distinguish between early and late events during the study. If treatment only delays the occurrence of a clinical event by some amount of time, then that benefit is not captured by the absolute risk reduction or NNT, even though it may be meaningful to patients.
Returning to the question posed in the vignette, evidence for the benefit of ICD therapy is available from the Sudden Cardiac Death in Heart Failure Trial.
This trial evaluated the benefit of ICD vs amiodarone or placebo for prevention of death in 2521 patients with mild to moderate heart failure who were followed for a median of 45.5 months. The 5-year risks of death were 36.1% for placebo, 34.0% for amiodarone, and 28.9% for ICD. The authors reported a statistically significant HR of 0.77 (reconstructed 95% confidence interval [CI], 0.62-0.91; P = 0.007) in favor of ICD placement vs placebo. This result means that ICD therapy reduced the hazard of death by 23% over the trial duration. A clinician might be left wondering whether such a statistically significant reduction in mortality is in fact clinically meaningful. Describing this hazard reduction to a patient is challenging. The absolute risk reduction for ICD vs placebo at 5 years is 7.1% (95% CI, 1.2%-13%; P = .020), which corresponds to an NNT of 14.2 (95% CI, 7.7-86). This means that 15 patients would need to be treated with an ICD to prevent 1 death in 5 years.
An alternative measure for estimating the treatment effect is based on the restricted mean survival time (RMST), or the mean survival time during a prespecified time interval, such as 5 years. This measure represents the average “event-free survival time” over the specified follow-up period. Graphically, the RMST represents the area under the survival curve up to the designated time.
Treatment effect can be quantified using the difference of 2 RMSTs, which can be intuitively interpreted as a gain (or loss) in the event-free survival time. In addition, this measure does not require the proportional hazards assumption. To illustrate this, we analyzed reconstructed data from the Sudden Cardiac Death in Heart Failure Trial. We plotted the Kaplan-Meier curves (Figure, A) and estimated the RMSTs up to 5 years as the areas under Kaplan-Meier curves (Figure, B). RMSTs were 49.1 and 51.4 months in the placebo and ICD arms, respectively. The difference was 2.29 months (95% CI, 0.59-3.98; P = 0.01). In other words, among patients with mild to moderate heart failure on conventional therapy, ICD therapy prolongs survival by an average of about 2.3 months over 5 years. The 95% CI suggests that ICD therapy may increase survival by between 0.6 and 4.0 months. These quantifications are more informative for assessing the clinical benefit of ICD therapy than a HR of 0.77 (23% relative reduction), ranging from a pronounced benefit of 0.62 (38% reduction), as suggested by the lower bound of the 95% CI, to a modest benefit of 0.91 (9% reduction), as suggested by the upper bound.
In the case of GJ, choice E, An ICD will prolong life from 49.1 to 51.4 months, an average of 2.3 months, over 5 years, is most likely to be intuitively understood by the patient. Compared with the P value, HR, absolute risk reduction, and NNT, a precise estimate of the gain in survival time is more concrete. Whether a survival gain of 2.3 months over 5 years is meaningful or not can then be placed in the context of the patient's own values, which is the foundation of shared decision making. As exemplified by this case, we believe that translational and interpretable statistics, such as RMST, should be routinely presented in addition to traditional statistics in reports of clinical studies. These summary measures can facilitate informed shared decision making between physicians and patients, particularly in light of risk-benefit and cost considerations. Importantly, RMST can be easily calculated using standard statistical software, including the R procedures surv2sampleComp and survRM2 (CRAN), and equivalent programs for StataCorp (College Station, Tex) and SAS (Cary, NC).
Merchant FM, Dickert NW, Howard DH. Mandatory shared decision making by the Centers for Medicare & Medicaid Services for cardiovascular procedures and other tests. JAMA320 (7), 641–642.