 |
|
 |
|
 |
 |
|
 |
 |
 |
 |
 |
 |
 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Douglas E. Lake, PhD
J. Randall Moorman, MD
Internal Medicine and the Cardiovascular Research Center
University of Virginia Health System
Charlottesville, VA 22908
We thank Dr El-Khatib for his insights into our analysis of abnormal heart rate characteristics (HRC) before clinical signs of neonatal sepsis. We have, indeed, found that an entropy measure alone has predictive information in this important clinical setting, and we refer him to our earlier work.1 We stop short, however, of agreeing that entropy alone provides either an adequate or even a simplified estimate of risk or that entropy in this setting is a measure of regularity.
First, our multivariate models uniformly have higher predictive accuracy than models that use only the components of SD, sample entropy, or sample asymmetry. In our view, the importance of early diagnosis of neonatal sepsis easily justifies the use of more than one measure.
Second, using only an entropy measure does not really simplify the calculations very much; it is by far the most complex and time consuming computationally and itself requires calculation of another HRC component, the SD. The HRC index adds only sample asymmetry, which merely requires sorting, and a regression expression. The value of the regression is that it relates the calculated parameters to actual clinical cases of illness, making the HRC index a real-world test.
Finally, sample entropy is helpful in diagnosing sepsis in the NICU because it detects the heart rate records with reduced variability and transient decelerations that indicate imminent illness. Surprisingly, this finding is not related to the conventional interpretation of entropy as a measure of regularity, order, or complexity but is an inevitable feature of all data sets with flat baselines punctuated by spikes.1 We note that sample entropy is a better measure than approximate entropy because it has less bias, especially in short data sets, and is more likely to maintain accurate hierarchies among data sets of differing order.2,3
The complexity of the pathophysiology of sepsis defies simple diagnostic approaches. The predictive model that we described optimizes contributions from different HRC measures and gives us our best chance to date at the early diagnosis of late-onset neonatal sepsis.
FOOTNOTES
Financial Disclosure: Medical Predictive Science Corporation of Charlottesville, Virginia, has a license to market technology related to heart rate characteristics monitoring of newborn infants and supplied partial funding for this study. Drs Griffin and Moorman have an equity share in this company.
REFERENCES
Related articles in Pediatrics:
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
 |
||
|
||
Home | My Pediatrics | Journal Information | Current Issue | Past Issues | Subscriptions & Services | Contact Us Click here for faster international access © 2006 American Academy of Pediatrics. All rights reserved. |
||
|
||
