Information Technology Reference
In-Depth Information
decrease in the blood oxygen saturation (the percentage of haemoglobin saturated
with oxygen) during sleep.
The gold standard for the diagnosis of
OSA
is an overnight in-clinic polysomnog-
raphy. The sleep study measures the frequency of apnea and hypopnea events. In
general, the diagnosis of
OSA
uses two scores: Apnea-Hypopnea Index (
AHI
)and
Apnea Index (
AI
). The apnea-hypopnea index (
AHI
), the most commonly used
score, is calculated as a number of apnea and hypopnea events per hour of sleep;
The apnea index (
AI
) is calculated as a number of apnea events per hour of sleep.
Additionally, many definitions of apnea/hypopnea events require one or both of two
factors: oxyhemoglobin desaturation of 4% or more and brief arousals from sleep.
Thus, the definition of apnea event varies.
The diagnosis of
OSA
can be based on two approaches: (1) a score of apnea/hy-
popnea events (
AHI
) or (2) a combination of
AHI
scoring and symptoms. In the
diagnosis based solely on the
AHI
index, apnea is classified as
mild
for
AHI
be-
tween 5 and 14
30.
The International Classification of Sleep Disorders (ICSD) [16] defines the severity
of
OSA
in terms of the frequency of apnea events, the degree of oxygen desaturation,
and the severity of daytime sleepiness.
The differences between the scoring and definitions of apnea have important im-
plications for the conceptual data modeling in data mining. Most published medical
research studies base the diagnosis of
OSA
on
AHI
or a combination of apnea index
(
AI
)and
AHI
obtained from overnight in-clinic PSG. For example, the authors of
two articles on craniofacial predictors [9] and [5] define respectively two criteria:
(1)
OSA
defined as
AHI
.
9,
moderate
for
AHI
between 15 and 29
.
9, and
severe
for
AHI
≥
≥
>
>
10. To il-
lustrate the difference between diagnostic criteria based on
AHI
and a combination
of
AI
and
AHI
, we applied these two criteria to 233 records from the data set ob-
tained from the authors of the first publication [3]. Figure 13.2 shows the number of
records classified into
OSA
and non-
OSA
using two diagnostic criteria:
OSA
defined
by
AHI
5and(2)
OSA
defined as
AI
5or
AHI
10 (col-
umn to the right). The second criterion is more restrictive and excludes 26 records
(26
≥
5 (column to the left) and
OSA
defined by
AH
>
5OR
AHI
>
233) from the
OSA
group and classifies them as a non-
OSA
.
The definition of
OSA
is fuzzy.
/
Different studies use different cut-off values
to indicate
OSA
, for example
AHI
15. To illustrate the
differences in prevalence of
OSA
, we applied three cut-off values to 795 records
obtained from a sleep clinic [7, 8]. Figure 13.3 shows the changing proportions
between non-
OSA
and
OSA
records for
AHI
≥
5,
AHI
≥
10,
AHI
≥
15.
OSA
is operationalized using diverse methods. Thus, the conceptual model for
data must define precisely the scoring criteria. The use of
AHI
or
AI
and three
cut-off values can result in significant differences in classification of the patients
(non-
OSA
,
OSA
), especially for patients with low
AHI
scores [13]. Furthermore,
the difficulty with the scoring of
AHI
is compounded by the natural night-to-night
variability in the severity of apnea and the differences in diagnostic equipment.
≥
5,
AHI
≥
10, and
AHI
≥