Author + information
- Thomas F. Heston, MD⁎ ( and )
- Richard L. Wahl, MD
- ↵⁎Johns Hopkins University, Johns Hopkins Nuclear Medicine, 600 North Caroline Street, Suite 3223, Baltimore, Maryland 21287
In a recent issue of the Journal, Tamaki et al. (1) found that in their study sample of 106 consecutive patients with stable congestive heart failure (CHF), those experiencing a sudden cardiac death (SCD) had on average a higher washout rate of iodine-123 metaiodobenzylguanidine (MIBG WR) compared with those who survived. Statistically, the cardiac MIBG WR was a powerful predictor of SCD in patients with mild-to-moderate CHF. But how can this best be applied clinically?
The mean (X1) washout rate in those with SCD was 39.9% with a standard deviation (SD1) of 15.2%. For those without SCD, the mean (X2) washout rate was 27.6% with a standard deviation (SD2) of 14.2%. Using this data, we can determine the crossover point below which a patient is more likely than not to fall into the low-risk group (no SCD) and above which a patient is more likely than not to fall into the high-risk group (SCD) (2).
The crossover point (CP) = (SD1 · X2 + SD2 · X1)/(SD1 + SD2) = 33.5%. This CP falls 0.42 SDs above X2 and 0.42 SDs below X1 (i.e., X2 + 0.42 · SD2 = X1 − 0.42 · SD1). In normally distributed data, using a z-score table, we find that, at best, 34% of the patients will be miscategorized when using the MIBG WR if a fixed threshold value is utilized. If we use a threshold of 27%, as proposed by Ogita et al. (3), then over 50% of the low-risk patients will be miscategorized. Threshold values either above or below the CP will only lead to a miscategorization rate >34%.
In clinical practice, fixed threshold values for continuous data such as the MIBG WR are not rigidly followed. Patients are frequently categorized as “borderline normal” or “borderline abnormal.” Are there better ways to make sense of the data so it can be more clinically useful? Simply reporting the means, SDs, and a threshold value does not adequately characterize the data for the clinician caring for an individual patient.
We propose that a more useful way to report continuous variables that impact patient care is to give at least 3 reference values: 1) the point where an individual patient is just as likely as not to belong to group 1 as to group 2; 2) the odds of belonging to group 1 at X1; and 3) the odds of belonging to group 2 at X2. In some situations, additional reference values may be useful. For the MIBG WR data, the CP = 33.5%. This is the point at which the odds are 50/50 in regard to whether the patient is in the high-risk or in the low-risk group. The formula to determine this point is given in the preceding text.
When a patient's MIBG WR is 39.9% or greater, the odds are at least 2.6 to 1 that the patient is in the high-risk group. This is calculated by finding the z-score of the absolute value of (X1 − X2)/SD2, then dividing 0.5 by the area under the curve to the right of this z-score. When a patient's MIBG WR is 27.6% or less, the odds are at least 2.4 to 1 that the patient is in the low-risk group. This is calculated by finding the z-score of the absolute value of (X1 − X2)/SD1, then dividing 0.5 by the area under the curve to the right of this z-score.
This type of numerical summary helps clinicians reasonably apply and explain the MIBG WR to individual patients with stable CHF. When a patient's MIBG WR is around 33%, the test does not help categorize the patient into a low- or high-risk category (a coin flip is just as accurate). However, when the MIBG WR is 27% or less, the odds are greater than 2:1 that the patient is at low risk. When the MIBG WR is 40% or higher, the odds are greater than 2:1 that the patient is at high risk. Basing medical management upon MIBG WR values between 30% and 36% is basically just guessing, and will lead to suboptimal care in a high percentage of patients.
- American College of Cardiology Foundation
- Tamaki S.,
- Hamada T.,
- Okuyama Y.,
- et al.
- Ogita H.,
- Shimonagata T.,
- Fukunami M.,
- et al.