Biology Reference
In-Depth Information
questions is through statistical analyses, and we discuss these and other
validation questions next.
VII. DATA VALIDATING THE UTILITY OF SAMPLE
ASYMMETRY ANALYSIS OF HEART RATE VARIABILITY
1. Experimental design and methods
In this section, we describe a study involving 158 infants admitted to the
NICU at the University of Virginia that focuses on the following
question: Could SA be considered a measure for the risk of approaching
episodes of sepsis and SIRS? In this section, we follow a fundamental
statistical approach involving an experiment designed to compare
two groups of infants—those with and without episodes of sepsis and
SIRS. For each group, HRV is collected throughout the study. For the
group of infants developing sepsis, the SA is calculated for every day of
the five days preceding the medical diagnosis. We seek to corroborate
our hypothesis that SA levels increase significantly 24 hours before
sepsis.
The participants in the study were infants with high risk factors for
acquiring late-onset sepsis, including LBW, prematurity, the need for
central venous access, and NICU stays longer than 2 weeks. Fifty of
these infants had a total of 75 episodes of sepsis or SIRS during the
study, defined to be present when a physician suspected sepsis or SIRS,
obtained a blood culture, and administered antibiotic therapy for seven
or more days. These 50 infants formed our experimental group.
A control group of 50 healthy infants (i.e., infants who did not develop
sepsis and SIRS during the study) was selected from the remaining 108
consecutive admissions to the NICU to precisely match the experimental
group by birth weight and GA.
3
Table 6-1 includes the statistical
comparisons between the experimental and the control groups. A group
comparison t-test (Table 6-1, last column) indicates that there is no
evidence for rejecting the hypotheses for equality of the mean BW and
GA of the two groups.
3. To do this, we had to tackle the following problem: The entire group of
healthy infants had, on average, both a higher BW and GA, which was to be
expected because LBW is a risk factor for sepsis. Thus, our random selection of
control infants used a nonuniform random generator with a higher probability of
selecting LBW infants and lower probability of selecting higher BW infants.
Specifically, the probability of selecting a control infant ranged from 1 for VLBW
(BW
>
1451 g. These probability values were determined
by the relative distribution of BW in the experimental versus control group. Once
experimental versus control infants were matched by BW, the matching by GA
occurred naturally (Table 6-1).
656 g) to 0.16 for BW
