Title: Logistic regression model for distinguishing keratoconus eyes based on analysis of Orbscan parameter
1- Logistic regression model for distinguishing
keratoconus eyes based on analysis of Orbscan
parameters - Yuri Oleynikov MD,PhD
- Fellow Cornea Refractive
- Surgery
- Baris Sonmez, M.D.
- International Fellow Cornea Refractive
- Surgery
- D. Rex Hamilton, M.D., M.S.
- Director, UCLA Laser Refractive CenterAssistant
Professor of Ophthalmology
Presenters have no financial interest in the
technology presented
2Introduction
- Analysis of corneal morphology is essential to
identify corneas at risk for post-LASIK ectasia - Obvious keratoconus is detectable at the slit
lamp - Forme fruste keratoconus (FFKCN) can be subtle
and easily missed on topography - Quantitative analysis may assist the refractive
surgeon in identifying subtle cases of FFKCN
3Purpose
- To quantitatively evaluate differences in corneal
shape of normal and KCN eyes using Orbscan IIz
slit-beam topography system. - To formulate a reliable model distinguishing
normal from KCN corneal morphology. - Retrospective study of 207 normal eyes of 108
patients who presented for refractive surgery
evaluation and 42 eyes of 24 patients with KCN - Keratoconus was defined based on the
Collaborative Longitudinal Evaluation of KCN
clinical findings
4Methods
- Multiple parameters were recorded
- amount and axis of astigmatism
- central corneal power
- irregularity indices at 3 and 5mm zones
- location and magnitude of maximal posterior
elevation - magnitude of maximal central anterior elevation
- location and magnitude of thinnest optical
pachymetry - anterior and posterior best-fit sphere
- keratometry values at 3 and 5mm zones at 30, 60,
90, 120, 150, 210, 240, 270, 300, and 330 degrees
on both keratometric and tangential topographic
maps - Skewing of radial axis (SRAX) at 3 and 5mm zones.
5Methods
- anterior elevation / best fit sphere radius ratio
- I-S (inferior-superior) difference, defined as
the average superior (30, 60, 90, 120, and 150
degrees) subtracted from the average inferior
(210, 240, 270, 300, and 330 degrees) keratometry
values for both the keratometric and tangential
topographic maps
6Patient characteristics
Chi-square test. T-test. Age was available
for 106 normal patients. Spherical equivalent
was available for 38 keratoconic eyes.
7Orbscan IIz quantitative indices between normal
and keratoconic eyes
T-tests
8Multiple logistic regression model
The multiple logistic regression model for
distinguishing KCN eyes from normal controls
(MPE (p0.030), AER (p0.050), IS-K 3mm
(p0.015), and the area under ROC curve
0.99) Pr(KCN) 1 / (1 exp(21.4177 -
0.1474MPE - 25.4821AER - 5.1761IS-K
3mm)) where Pr(KCN) is the predicted probability
of KCN, MPE Maximum posterior elevation,
measured as the highest elevation of posterior
float over best-fit sphere. AER Ratio of
highest anterior elevation to the anterior best
fit sphere in diopters IS-K 3mm is
inferior-superior K difference ratio. Area under
ROC is 99 - high sensitivity and specificity.
Group separation results based on several cut-off
points of Pr(KCN)
Pr(KCN) Predicted probability of being KCN
using logistic regression models shown above SE
Sensitivity of KCN eyes with positive
results / of KCN eyes SP Specificity of
normal control eyes with negative results / of
normal control eyes AC Total accuracy ( of
KCN eyes with positive results of normal
control eyes with negative results) / all eyes.
Positive means Pr(KCN) greater than the cut-off
value.
9Normal cornea
AER0.289 MPE19 um IS-K0.09 PrKCN 0.021
10Keratoconic cornea
AER1.58 MPE142um IS-K6.21 PrKCN 100
11Analysis of atypical topography
AER0.36 MPE28 IS-K0.87 PrKCN 2.5506
12Conclusion
- A logistical regression model using Orbscan
parameters may be useful in detecting abnormal
corneal morphology - Our cut-off value of 0.02 was highly sensitive
and specific for distinguishing normal from
abnormal morphology - Further training of the model using larger data
sets and forme fruste keratoconus eyes may allow
for improved detection of subtle corneal
morphologic abnormalities