Journal of Refractive Surgery

Biomechanics 

Analysis of Waveform-Derived ORA Parameters in Early Forms of Keratoconus and Normal Corneas

Bruna V. Ventura, MD; Aydano P. Machado, PhD; Renato Ambrósio Jr, MD, PhD; Guilherme Ribeiro, MD; Luana N. Araújo, MS; Allan Luz, MD; João Marcelo Lyra, MD, PhD

Abstract

PURPOSE:

To evaluate the Ocular Response Analyzer (ORA; Reichert Ophthalmic Instruments, Depew, NY) performance in differentiating grades I and II keratoconus from normal corneas using 41 parameters individually and to assess the effect of analyzing all parameters together.

METHODS:

This study compared the mean value of 41 ORA parameters in grades I and II keratoconus with healthy age-matched control eyes. Only eyes with a central corneal thickness between 500 and 600 μm were included. The area under the receiver operating characteristic curve was calculated for each of the 41 parameters independently and for all of the parameters together.

RESULTS:

This study included 136 eyes with normal corneas and 68 eyes with grades I and II keratoconus. When analyzed individually, four ORA parameters (p1area, p1area1, p2area, and p2area1) had an area under the curve greater than 0.900 for discriminating between both groups. The p2area was the parameter that achieved the largest area under the curve individually (0.931). The area under the curve increased to 0.978 when analyzing all parameters together.

CONCLUSION:

Alternative ORA parameters are better for differentiating grades I and II keratoconus from normal corneas than the four parameters originally available for ophthalmologists (corneal hysteresis, Goldmann-correlated intraocular pressure, corneal-compensated intraocular pressure, and corneal resistance factor). Although the ORA did not achieve 100% accuracy, the discrimination between these two groups was optimized by combining all parameters.

[J Refract Surg. 2013;29(9):637–643.]

From the Brazilian Study Group of Artificial Intelligence and Corneal Analysis, Maceió, Rio de Janeiro, and Recife, Brazil (BVV, APM, RA, GR, LNA, AL, JML); Altino Ventura Foundation, Recife, Brazil (BVV); Federal University of Alagoas, Maceió, Brazil (BVV, APM, LNA); Instituto de Olhos Renato Ambrósio, Rio de Janeiro, Brazil (RA, AL); Federal University of São Paulo, São Paulo, Brazil (RA, AL); and Universidade de Ciências da Saúde de Alagoas, Maceió, Brazil (GR, JML).

Dr. Ambrósio is a consultant for Oculus Optikgeräte GmbH and Reichert Ophthalmic Instruments. The remaining authors have no financial or proprietary interest in the materials presented herein.

AUTHOR CONTRIBUTIONS

Study concept and design (BVV, APM, RA, GR, LNA, AL, JML); data collection (RA); analysis and interpretation of data (BVV, APM, GR, LNA, AL, JML); drafting of the manuscript (BVV); critical revision of the manuscript (APM, RA, GR, LNA, AL, JML); statistical expertise (BVV); obtained funding (LNA); administrative, technical, or material support (BVV, APM, RA, GR, LNA, AL, JML); supervision (APM, RA, AL, JML)

Correspondence: Bruna V. Ventura, MD. Rua da Soledade, 170, Boa Vista, Recife, Pernambuco, Brazil 50070-020. E-mail: brunaventuramd@gmail.com

Received: February 06, 2012
Accepted: May 13, 2013

Abstract

PURPOSE:

To evaluate the Ocular Response Analyzer (ORA; Reichert Ophthalmic Instruments, Depew, NY) performance in differentiating grades I and II keratoconus from normal corneas using 41 parameters individually and to assess the effect of analyzing all parameters together.

METHODS:

This study compared the mean value of 41 ORA parameters in grades I and II keratoconus with healthy age-matched control eyes. Only eyes with a central corneal thickness between 500 and 600 μm were included. The area under the receiver operating characteristic curve was calculated for each of the 41 parameters independently and for all of the parameters together.

RESULTS:

This study included 136 eyes with normal corneas and 68 eyes with grades I and II keratoconus. When analyzed individually, four ORA parameters (p1area, p1area1, p2area, and p2area1) had an area under the curve greater than 0.900 for discriminating between both groups. The p2area was the parameter that achieved the largest area under the curve individually (0.931). The area under the curve increased to 0.978 when analyzing all parameters together.

CONCLUSION:

Alternative ORA parameters are better for differentiating grades I and II keratoconus from normal corneas than the four parameters originally available for ophthalmologists (corneal hysteresis, Goldmann-correlated intraocular pressure, corneal-compensated intraocular pressure, and corneal resistance factor). Although the ORA did not achieve 100% accuracy, the discrimination between these two groups was optimized by combining all parameters.

[J Refract Surg. 2013;29(9):637–643.]

From the Brazilian Study Group of Artificial Intelligence and Corneal Analysis, Maceió, Rio de Janeiro, and Recife, Brazil (BVV, APM, RA, GR, LNA, AL, JML); Altino Ventura Foundation, Recife, Brazil (BVV); Federal University of Alagoas, Maceió, Brazil (BVV, APM, LNA); Instituto de Olhos Renato Ambrósio, Rio de Janeiro, Brazil (RA, AL); Federal University of São Paulo, São Paulo, Brazil (RA, AL); and Universidade de Ciências da Saúde de Alagoas, Maceió, Brazil (GR, JML).

Dr. Ambrósio is a consultant for Oculus Optikgeräte GmbH and Reichert Ophthalmic Instruments. The remaining authors have no financial or proprietary interest in the materials presented herein.

AUTHOR CONTRIBUTIONS

Study concept and design (BVV, APM, RA, GR, LNA, AL, JML); data collection (RA); analysis and interpretation of data (BVV, APM, GR, LNA, AL, JML); drafting of the manuscript (BVV); critical revision of the manuscript (APM, RA, GR, LNA, AL, JML); statistical expertise (BVV); obtained funding (LNA); administrative, technical, or material support (BVV, APM, RA, GR, LNA, AL, JML); supervision (APM, RA, AL, JML)

Correspondence: Bruna V. Ventura, MD. Rua da Soledade, 170, Boa Vista, Recife, Pernambuco, Brazil 50070-020. E-mail: brunaventuramd@gmail.com

Received: February 06, 2012
Accepted: May 13, 2013

The cornea is a tissue with viscoelastic properties. The alterations in its stromal structure are intimately related to its biomechanical behavior.1 The commercially available Ocular Response Analyzer (ORA; Reichert Ophthalmic Instruments, Depew, NY) uses a dynamic bidirectional applanation process to quantify corneal biomechanical properties in vivo and to determine intraocular pressure (IOP).2 The four ORA parameters that were originally introduced in ophthalmology are corneal hysteresis (CH), corneal resistance factor (CRF), Goldmann-correlated IOP (IOPg), and corneal-compensated IOP (IOPcc). An additional 37 parameters are generated from the ORA waveform. They are the mathematical representation of this waveform, not corresponding to specific corneal biomechanical properties.3

The investigation of some ORA parameters independently in keratoconic corneas has shown that even though they possess a statistically lower mean value when compared to normal corneas, the large measurement overlap between both groups of patients results in poor test performance for differentiating normal and abnormal corneas.4–6 However, a parameter that is completely useless by itself may significantly improve the test’s performance when combined with other parameters. Thus, the purpose of the current study was to evaluate and compare each ORA parameter mean value in patients with early keratoconus to healthy control individuals. In addition, we assessed the ORA’s ability to differentiate grades I and II keratoconus from normal corneas using each of the 41 parameters individually and the effect of analyzing ORA parameters together.

Patients and Methods

Patients

This comparative case series study was approved by the Ethics Committee of the Altino Ventura Foundation and followed the tenets of the Declaration of Helsinki. All study participants were told the purpose of the study and gave informed consent before inclusion.

All patients who submitted to ORA (version 3.01) in the private practice of one surgeon (RA) from 2007 to 2009 were eligible to participate in this study. Exclusion criteria included patients younger than 18 years, previous ocular surgery, eye disease other than keratoconus, chronic use of topical medications, corneal scars or opacities, and grades III and IV keratoconus according to the Krumeich severity classification (central keratometry greater than 53.0 diopters [D]).7 Contact lenses were removed at least 72 hours prior to the ORA examination.

Patients underwent a comprehensive ophthalmologic examination, including review of their medical history, corrected visual acuity, slit-lamp microscopy, and funduscopic examination. In addition, patients submitted to testing with the Placido disk topography (Humphrey ATLAS; Carl Zeiss Meditec, Inc., Dublin, CA), Pentacam tomography (Oculus Optikgeräte GmbH, Wetzlar, Germany), and ORA. All measurements were taken during the same visit between 8 AM and 6 PM. Only high-quality ORA readings (defined by the manufacturer as fairly symmetrical height force-in and force-out applanation signal peaks on the ORA waveform) were accepted for the study.

The same experienced corneal specialist (RA) identified eyes with normal corneas and eyes with keratoconus. Diagnosis of keratoconus was made by topographic evaluation (increased area of corneal power surrounded by concentric areas of decreasing power, inferior-superior power asymmetry, and skewing of the steepest radial axes above and below the horizontal meridian).8 Only patients with grades I and II keratoconus (central keratometry 53.0 D or less)7 and central corneal thickness (CCT) between 500 and 600 μm were included in the study group. Pentacam CCT measurements were used as opposed to the thickness at the thinnest point of the cornea because the air-puff system of the ORA is aligned with the corneal center and not with the thinnest point. Keratoconic eyes were matched with healthy controls according to age (±2 years). The control eyes also had a CCT between 500 and 600 μm.

Ocular Response Analyzer

The ORA has an infrared electro-optical system that monitors corneal deformations. It delivers a precisely metered collimated air pulse to the eye. The cornea suffers an inward movement, passing a first applanation state before assuming a concave shape. The air pressure progressively declines after this first applanation and the cornea passes through a second applanation state while returning to its normal convex curvature. The examination generates a waveform that contains two peaks, corresponding to the inward and outward applanation events (Figure 1).2 Forty-one parameters originate from this examination (Table A, available in the online version of this article).2,3,9 In addition, the Keratoconus Match Index (KMI) is generated. The KMI is a composite value that derives from seven ORA waveform parameters using a proprietary algorithm. It represents the similarity of the waveform from an eye to the average waveform characteristics of various eyes with keratoconus.

The Ocular Response Analyzer (Reichert Ophthalmic Instruments, Depew, NY) generates a waveform from which parameters originate. Several parameters are derived from the upper 75% of the applanation peaks and others describe characteristics of the upper 50%.

Figure 1. The Ocular Response Analyzer (Reichert Ophthalmic Instruments, Depew, NY) generates a waveform from which parameters originate. Several parameters are derived from the upper 75% of the applanation peaks and others describe characteristics of the upper 50%.

Data Analysis

Statistical analysis was performed using SPSS (version 12.0; SPSS, Inc., Chicago, IL) for Windows (Microsoft Corporation, Redmond, WA). Qualitative characteristics were expressed by their absolute and relative frequencies, and compared between the groups using the chi-square test. Quantitative characteristics were expressed by their minimum and maximum value, mean, and standard error of the mean. They were compared between the keratoconus and normal cornea groups using Student’s t test. To evaluate the effect of the thinnest corneal thickness in each ORA parameter and in the KMI, we further subdivided the keratoconus and normal cornea groups based on the thinnest corneal thickness (greater than or less than 500 μm). We compared the groups by analysis of variance. Bonferroni correction for multiple comparisons was used when analyzing pairwise differences. A P value less than .05 was used to reject the null hypothesis.

To better evaluate the ORA performance in differentiating normal corneas from those with early stages of keratoconus, we calculated the receiver operating characteristic curve for each of the 41 ORA parameters individually using Medcalc software (version 12.0.1; MedCalc Software bvba, Mariakerke, Belgium). The curves were quantified using the area under the curve (AUC). An AUC greater than 0.900 was considered to reflect a good ability to distinguish keratoconic from normal corneas, whereas 0.700 was the cutoff point that defined poor parameter performance.3 In addition, we analyzed all 41 ORA parameters together to investigate the possible interdependence between parameters for keratoconus identification optimization using ORA. We used RapidMiner software (version 5.1; Rapid-I GmbH, Dortmund, Germany)10 to apply the data of all of the parameters to the radial basis function neural network and we used Medcalc software to calculate this ROC curve.

To validate the radial basis function neural network, a ten-fold cross-validation technique was used. Briefly, in this technique, the cases are divided into 10 mutually exclusive and equal-sized subsets. These subsets are randomly constructed, maintaining the class distribution of the whole dataset. When testing each subset, the algorithm is first trained on the union of all other subsets.11 We also used the radial basis function neural network and the Medcalc software to calculate the AUC of the composite of the ORA parameters that achieved an AUC greater than 0.900 when analyzed independently. Additionally, we calculated the AUC of the KMI. Because differentiating grade I keratoconus from normal corneas and differentiating keratoconus with thinnest corneal thickness greater than 500 μm from normal corneas can be clinically challenging, we analyzed the AUC attained by each ORA parameter when differentiating these two sets of patients. The AUC achieved by the KMI and by the composite of all 41 ORA parameters when separating these patients was also assessed.

Results

A total of 106 patients (204 eyes) were enrolled in the study: 54 (50.9%) men and 52 (49.1%) women. One hundred thirty-six (66.7%) eyes were classified as normal and 68 (33.3%) eyes were diagnosed as having keratoconus. According to the Krumeich severity classification,7 49 (72.1%) eyes had grade I keratoconus and 19 (27.9%) eyes had grade II. In the keratoconus group, the mean patient age was 31.1 ± 11.6 years (range: 18.0 to 63.0 years) and the male-to-female ratio was 23:15. In the normal cornea group, the mean patient age was 30.9 ± 11.3 years (range: 18.0 to 63.2 years) and the male-to-female ratio was 31:37. Both groups of patients had similar ages (P = .899, Student’s t test) and gender distribution (P = .203, chi-square test). The mean CCT was 510.8 ± 1.2 μm (range: 500 to 569 μm) in the keratoconus group and 543.6 ± 2.2 μm (range: 500 to 599 μm) in the normal cornea group (P < .0001, Student’s t test). The mean thinnest corneal thickness was 472.2 ± 2.9 μm (range: 412 to 569 μm) in the keratoconus group and 543.4 ± 2.3 μm (range: 501 to 599 μm) in the normal cornea group (P < .0001, Student’s t test).

The differences in the mean values of most ORA parameters were statistically significant (P < .0001, Student’s t test) between the keratoconus group and the control group. In all of these cases, the mean value in the keratoconus group was lower than in the control group, except for aplhf, path1, path11, path2, and path21. However, there was a large overlap in the parameters’ values between the groups (Table B, available in the online version of this article).

When we subdivided the keratoconic and normal corneas based on the thinnest corneal thickness, 52 keratoconic corneas had a thinnest corneal thickness less than 500 μm, 16 keratoconic corneas had a thinnest corneal thickness greater than 500 μm, and all 136 normal corneas had a thinnest corneal thickness greater than 500 μm. The mean thinnest corneal thickness was 457.9 ± 3.2 μm (range: 412 to 499 μm) in the keratoconus subgroup with thinnest corneal thickness less than 500 μm and 518.5 ± 5.0 μm (range: 500 to 569 μm) in the keratoconus subgroup with thinnest corneal thickness greater than 500 μm (P < .0001, Student’s t test). Both keratoconus subgroups had similar distribution of grades I and II keratoconus (subgroup with thinnest corneal thickness less than 500 μm: 15 [28.8%] eyes with grade I keratoconus; subgroup with thinnest corneal thickness greater than 500 μm: 4 [25.0%] eyes with grade I keratoconus; P = .518, chi-square test). The values of the 41 ORA parameters and of KMI in each subgroup and the comparison between the subgroups are shown in Table C (available in the online version of this article). The three parameters (dslope2, dslope21, and IOPcc) that had similar mean values when comparing keratoconic and normal corneas in the previous analysis continued having similar mean values when comparing the groups subdivided based on thinnest corneal thickness. Twenty-three parameters that had a significantly different mean value when comparing the keratoconic and normal corneas in the previous analysis had similar mean values between both subgroups of keratoconic corneas and maintained a different mean value when compared to the normal corneas (parameters shown in bold in Table C). However, although not statistically significant, five of these parameters (dive1, h1, h11, KMI, and mslew1) tended to have an increase in parameter value related to an increase in thinnest corneal thickness. Ten parameters (aindex, aplhf, aspect2, aspect21, bindex, dslope11, path11, us-lope11, w11, and w2) that had a significantly different mean value when comparing the keratoconic and normal corneas in the previous analysis had similar mean values between both subgroups of keratoconic corneas and maintained a different mean value when comparing one of the keratoconus subgroups with the normal corneas. However, this was not seen when comparing the other keratoconus subgroup and the normal corneas.

Five parameters (aspect1, aspect11, dslope1, slew1, and uslope1) that had a significantly different mean value when comparing the keratoconic and normal corneas in the previous analysis had a statistically lower mean value in the keratoconus group with thinnest corneal thickness less than 500 μm when compared to the other two groups. For these parameters, the keratoconus group with thinnest corneal thickness greater than 500 μm had a lower mean value than the normal cornea group, but this difference did not reach statistical significance. One parameter (w1) that had a significantly different mean value when comparing the keratoconic and normal corneas in the previous analysis had a statistically lower mean value in the keratoconus group with thinnest corneal thickness greater than 500 μm when compared to the other two groups. The keratoconus group with thinnest corneal thickness less than 500 μm had a lower mean value than the normal cornea group, but this difference did not reach statistical significance.

When analyzed independently, four ORA parameters had a good ability to distinguish mild stages of keratoconus from normal corneas: p1area, p2area, p1area1, and p2area1. Of these, the biggest AUC was achieved with p2area (0.931). Fourteen parameters had poor performance: aindex, aspect2, aspect21, bindex, dslope11, dslope2, dslope21, IOPcc, path11, path21, w1, w2, w11, and w21. The lowest AUC was attained with IOPcc (0.579). The KMI achieved an AUC of 0.909. The mean KMI was 0.4 ± 0.05 (range: −0.5 to 1.0) in the keratoconic corneas and 0.9 ± 0.03 (range: −0.2 to 1.8) in the normal corneas (P < .0001, Student’s t test). The composite of p1area, p2area, p1area1, and p2area1 reached an AUC of 0.941. When using all ORA parameters together to differentiate keratoconic and normal corneas, the AUC increased to 0.978 (Figure 2).

Receiver operating characteristic (ROC) curves for distinguishing between mildly forms of keratoconic and normal corneas. The ROC curves express the performance achieved by p2area, Keratoconus Match Index (KMI), and the composite of Ocular Response Analyzer (ORA; Reichert Ophthalmic Instruments, Depew, NY) parameters that achieved an area under the curve (AUC) greater than 0.9 when analyzed independently and the composite of all 41 ORA parameters.

Figure 2. Receiver operating characteristic (ROC) curves for distinguishing between mildly forms of keratoconic and normal corneas. The ROC curves express the performance achieved by p2area, Keratoconus Match Index (KMI), and the composite of Ocular Response Analyzer (ORA; Reichert Ophthalmic Instruments, Depew, NY) parameters that achieved an area under the curve (AUC) greater than 0.9 when analyzed independently and the composite of all 41 ORA parameters.

In Table 1 we report the AUC achieved by each ORA parameter when differentiating grade I keratoconus from normal corneas and when differentiating keratoconus with thinnest corneal thickness greater than 500 μm from normal corneas. The KMI achieved an AUC of 0.887 and 0.857 when differentiating these two sets of patients, respectively. The former AUC increased to 0.964 and the latter increased to 0.916 when using the composite of all 41 ORA parameters.

Area Under ROC Curve Representing Each ORA Parameter’s Ability to Differentiate Between Normal Corneas and Grade I Keratoconus, and Between Normal Corneas and Keratoconic Corneas With a Thinnest Corneal Thickness of > 500 μm

Table 1: Area Under ROC Curve Representing Each ORA Parameter’s Ability to Differentiate Between Normal Corneas and Grade I Keratoconus, and Between Normal Corneas and Keratoconic Corneas With a Thinnest Corneal Thickness of > 500 μm

Discussion

Corneal biomechanical characteristics depend on many factors, including CCT and corneal collagen fiber density, orientation, and distribution.5,12–14 The alterations observed in keratoconic corneas are intrinsically related to these factors and, consequently, alter the tissue’s biomechanics.1 Similarly to another study,6 we only included eyes with a CCT between 500 and 600 μm to compare our groups, although the mean CCT and minimum corneal thickness in the keratoconus group were significantly lower. Previous studies13,15 have also suggested that variations in biomechanical measures occur with age. Therefore, our keratoconic eyes were age matched with healthy controls.

Most ORA parameters had a significantly lower mean value in the keratoconus group when compared to the control group, in agreement with previous studies4–6 that investigated CH, CRF, IOPg, IOPcc, and h1 in keratoconic and normal corneas. This suggests that keratoconic corneas applanate slightly earlier and in response to a slightly lower rate of air pressure, which has also been reported in forme fruste keratoconus.14,16 Similar to Touboul et al.,6 the mean IOPg was lower in keratoconic corneas. This parameter appeared to be independent of corneal thickness because the same result was found after subdividing the keratoconus group into thin and thick corneas. Touboul et al.6 hypothesized that there might be an artifact in the IOP assessment of keratoconic corneas using IOPg because of the lower elastic modulus, which would cause an underestimation of the measured pressure. Our finding that the mean IOPcc was similar between both groups and even between all subgroups supports the fact that it is the pressure measurement least dependent on corneal characteristics.2

The parameters related to the absolute value of the path lengths around the peaks (path1, path11, path2, and path21) and the parameter related to the irregularity of the waveform region between the peaks (aplhf) were significantly lower in our control eyes. A previous case report on unilateral corneal ectasia after bilateral LASIK described multiple oscillations in the waveform from the ectatic eye.16 This supports our finding of higher mean path lengths around the peaks and a higher mean aplhf in keratoconic eyes.

Previous studies by Fontes et al.4,5 have shown a significant difference in mean CH and CRF between keratoconic and normal corneas associated with a large overlap in these parameter values. They postulated that the other parameters derived from the waveform could be more sensitive for discriminating abnormal corneas, which was not verified in our study. Although there was a statistically significant difference between the groups with respect to the mean value of the majority of the parameters, there was also a large scatter of the parameter values, which limits the use of each parameter as a sole criterion for differentiating mild keratoconus from normal corneas.

The analysis subdividing the groups based on the thinnest corneal thickness has to be analyzed cautiously due to the number of corneas in each of the three subgroups. Also, it is difficult to isolate the influence of pachymetry on the ORA data because eyes with thicker corneas might also have less advanced keratoconus, which could possibly affect the ORA parameters’ values. Twenty-six parameters had the same results when comparing the subgroups versus when comparing all keratoconic corneas to normal corneas. However, although not statistically significant, five of these parameters (dive1, h1, h11, KMI, and mslew1) tended to have an increase in parameter value related to an increase in thinnest corneal thickness. This finding may possibly become statistically significant with a larger sample size. Ten parameters that had a significantly different mean value when comparing all keratoconic corneas to normal corneas had similar mean values between both subgroups of keratoconic corneas and maintained a different mean value when comparing one of the keratoconus subgroups with the normal corneas, whereas this was not observed when comparing the other keratoconus subgroup with the normal corneas. Although not statistically significant, the values of all of these parameters, except for aspect21, tended to be related to corneal thickness. This finding can possibly become statistically significant with a larger sample size. A larger sample may also result in a significant difference between the aspect21 mean values of both keratoconus subgroups when compared to the normal group.

Our results suggest that aspect1, aspect11, dslope1, slew1, uslope1 and w1 are possibly influenced by the thinnest corneal thickness. Five of these parameters had a statistically lower mean value in the keratoconus subgroup with thinnest corneal thickness less than 500 μm when compared to the other two groups. The keratoconus subgroup with thinnest corneal thickness greater than 500 μm had a lower mean value than the normal cornea group, but this difference did not reach statistical significance. A larger sample may confirm that these five ORA parameters are influenced by the thinnest corneal thickness, with an increase in parameter value being related to an increase in thinnest corneal thickness. In addition, one parameter (w1) had a statistically lower mean value in the keratoconus group with thinnest corneal thickness greater than 500 μm when compared to the other two groups. The keratoconus group with thinnest corneal thickness less than 500 μm had a lower mean value than the normal cornea group, but this difference did not reach statistical significance. This possibly suggests that other variables in addition to the thinnest corneal thickness and the presence of keratoconus biomechanical modifications also influence w1, which is supported by its poor performance in differentiating keratoconic and normal corneas. By dividing the groups based on corneal thinnest point, we observed that although corneal thickness might influence the ORA data to a degree, affecting some parameters more than others, it alone cannot explain the differences observed between the normal and keratoconus groups. Thus, future studies are necessary to confirm and further investigate these findings and assess their clinical significance.

When analyzed independently, four of the 41 ORA parameters achieved the best performance in distinguishing grades I and II keratoconus from normal corneas: p1area, p1area1, p2area, and p2area1. Although the AUCs decreased when assessing only grade I keratoconus and normal corneas, these four parameters were still the best. In addition, despite the lower AUCs reached when differentiating keratoconic corneas with a thinnest corneal thickness greater than 500 μm from normal corneas, p1area, p1area1, and p2area achieved the highest AUCs.

A previous study3 also highlighted p1area, p1area1, p2area, and p2area1 for their good performance in identifying grades I and II keratoconus. However, that study also reported an AUC higher than 0.900 for CRF, CH, dive2, h1, h11, h2, h21, and time1 (a temporal parameter created by the authors). In our analysis, although the latter parameters (except time1, which was not tested) did not achieve an AUC of 0.900 when differentiating grades I and II keratoconus from normal corneas, they all had an AUC greater than 0.850. This disparity in the identified parameters is possibly due to the different patient sample included in both studies, one from Brazil and the other from Europe, because previous studies17,18 have shown variations in ORA parameter values between races.

Two of the four ORA parameters that achieved the best performance in identifying keratoconus in the current study are related to the first applanation peak (p1area and p1area1) and comprise information pertaining to the 75% and 50% area under the peak, respectively. The two other parameters (p2area and p2area1) express this same information regarding the second applanation peak. In accordance with the literature,3–5,19 our study suggests that the parameters related to the area under the peaks are the best ORA parameters to distinguish early keratoconus from normal corneas and that the four original parameters (CH, IOPg, IOPcc, and CRF) are not ideal for discriminating these groups. In our study, the CRF and CH reached an AUC of 0.881 and 0.856, respectively, when discriminating grades I and II keratoconus from normal corneas. Although a previous study3 assessing the differentiation between normal and mild keratoconic eyes found an AUC of 0.968 for CRF and 0.900 for CH, another study6 described an AUC as low as 0.790 for CRF and 0.680 for CH when distinguishing these two groups. Thus, this latter study and our results support previous findings5,19 that discourage the use of CRF and CH as the only criteria for discriminating between keratoconic and normal corneas.

In our analysis assessing grades I and II keratoconus and normal corneas, the greatest AUC obtained by the ORA parameters individually was 0.931 (p2area), which was larger than the AUC achieved by the KMI. The AUC of the composite of the four parameters that reached an AUC greater than 0.900 when analyzed independently was slightly larger than when using p2area alone. However, the analysis of all ORA parameters together increased the AUC to 0.978. An increase in AUC when using the composite of the 41 ORA parameters was also seen when analyzing only grade I keratoconus and normal corneas, and only keratoconic corneas with a thinnest corneal thickness greater than 500 μm and normal corneas. These findings suggest that among the 41 ORA parameters there are some parameters that do not have a good performance in differentiating mildly keratoconic from normal corneas when used independently, but that significantly improve test performance when combined with other parameters. KMI was developed based on this principle. However, it is not the best performing composite parameter to distinguish between keratoconic and normal corneas. Further research is warranted to select the specific parameters that improve ORA performance in distinguishing between normal and mildly keratoconic corneas when applied together, excluding those parameters that are redundant or that add noise. Moreover, despite our results, corneal topography still distinguishes these two groups better than the ORA, because the diagnosis of keratoconus was known in these cases by topography.20 Thus, future studies integrating ORA parameters and those of other keratoconus screening methods, such as corneal topography and tomography, may enhance the identification of early keratoconus even more.

Most ORA parameters were statistically lower in eyes with grades I and II keratoconus when compared to healthy control eyes. However, the large overlap between the groups limits the use of each parameter individually. Our study suggests that the four ORA parameters related to the area under the peaks are better for discriminating between early keratoconus and normal corneas than the four original parameters used independently, and that corneal thickness possibly does not have an important influence in these former parameters. In addition, KMI was not the best composite parameter to distinguish the groups. Although the ORA did not achieve 100% accuracy, the test performance was enhanced using the combined analysis of all 41 ORA parameters.

References

  1. Dupps WJ Jr, Wilson SE. Biomechanics and wound healing in the cornea. Exp Eye Res. 2006;83:709–720. doi:10.1016/j.exer.2006.03.015 [CrossRef]
  2. Luce DA. Determining in vivo biomechanical properties of the cornea with an ocular response analyzer. J Cataract Refract Surg. 2005;31:156–162. doi:10.1016/j.jcrs.2004.10.044 [CrossRef]
  3. Mikielewicz M, Kotliar K, Barraquer RI, Michael R. Air-pulse corneal applanation signal curve parameters for the characterisation of keratoconus. Br J Ophthalmol. 2011;95:793–798. doi:10.1136/bjo.2010.188300 [CrossRef]
  4. Fontes BM, Ambrósio R Jr, Velarde GC, Nosé W. Ocular response analyzer measurements in keratoconus with normal central corneal thickness compared with matched normal control eyes. J Refract Surg. 2011;27:209–215.
  5. Fontes BM, Ambrósio R Jr, Jardim D, Velarde GC, Nosé W. Ability of corneal biomechanical metrics and anterior segment data in the differentiation of keratoconus and healthy corneas. Arq Bras Oftalmol. 2010;73:333–337. doi:10.1590/S0004-27492010000400006 [CrossRef]
  6. Touboul D, Bénard A, Mahmoud AM, Gallois A, Colin J, Roberts CJ. Early biomechanical keratoconus pattern measured with an ocular response analyzer: curve analysis. J Cataract Refract Surg. 2011;37:2144–2150. doi:10.1016/j.jcrs.2011.06.029 [CrossRef]
  7. Krumeich JH, Daniel J, Knülle A. Live-epikeratophakia for keratoconus. J Cataract Refract Surg. 1998;24:456–463. doi:10.1016/S0886-3350(98)80284-8 [CrossRef]
  8. Wang M, ed. Topographic presentation of common corneal diseases. In: Zhang Z. Corneal topography in the waveform era: A guide for clinical application. Thorofare, NJ: SLACK Incorporated; 2006:85–96.
  9. Ehrlich JR, Haseltine S, Shimmyo M, Radcliffe NM. Evaluation of agreement between intraocular pressure measurements using Goldmann applanation tonometry and Goldmann correlated intraocular pressure by Reichert’s ocular response analyzer. Eye (Lond). 2010;24:1555–1560. doi:10.1038/eye.2010.83 [CrossRef]
  10. Mierswa I, Wurst M, Klinkenberg R, Scholz M, Euler T. Yale: rapid prototyping for complex data mining tasks. Presented at the Twelfth International Conference of the Association for Computing Machinery Special Interest Group on Knowledge Discovery and Data Mining. ; August 20–23, 2006. ; Philadelphia, PA. .
  11. Chen S, Zhou S, Yin FF, Marks LB, Das SK. Investigation of the support vector machine algorithm to predict lung radiation-induced pneumonitis. Med Phys. 2007;34:3808–3814. doi:10.1118/1.2776669 [CrossRef]
  12. Torres RM, Merayo-Lloves J, Jaramillo MA, Galvis V. Corneal biomechanics [article in Spanish]. Arch Soc Esp Oftalmol. 2005;80:215–223.
  13. Zarei-Ghanavati S, Ramirez-Miranda A, Yu F, Hamilton DR. Corneal deformation signal waveform analysis in keratoconic versus post-femtosecond laser in situ keratomileusis eyes after statistical correction for potentially confounding factors. J Cataract Refract Surg. 2012;38:607–614. doi:10.1016/j.jcrs.2011.11.033 [CrossRef]
  14. Schweitzer C, Roberts CJ, Mahmoud AM, Colin J, Maurice-Tison S, Kerautret J. Screening of forme fruste keratoconus with the ocular response analyzer. Invest Ophthalmol Vis Sci. 2010;51:2403–2410. doi:10.1167/iovs.09-3689 [CrossRef]
  15. Touboul D, Roberts C, Kérautret J, et al. Correlations between corneal hysteresis, intraocular pressure, and corneal central pachymetry. J Cataract Refract Surg. 2008;34:616–622. doi:10.1016/j.jcrs.2007.11.051 [CrossRef]
  16. Kerautret J, Colin J, Touboul D, Roberts C. Biomechanical characteristics of the ectatic cornea. J Cataract Refract Surgery. 2008;34:510–513. doi:10.1016/j.jcrs.2007.11.018 [CrossRef]
  17. Song Y, Congdon N, Li L, et al. Corneal hysteresis and axial length among Chinese secondary school children: the Xichang Pediatric Refractive Error Study (X-PRES) report no. 4. Am J Ophthalmol. 2008;145:819–826. doi:10.1016/j.ajo.2007.12.034 [CrossRef]
  18. Kirwan C, O’Keefe M, Lanigan B. Corneal hysteresis and intraocular pressure measurement in children using the reichert ocular response analyzer. Am J Ophthalmol. 2006;142:990–992. doi:10.1016/j.ajo.2006.07.058 [CrossRef]
  19. Fontes BM, Ambrósio R Jr, Jardim D, Velarde GC, Nosé W. Corneal biomechanical metrics and anterior segment parameters in mild keratoconus. Ophthalmology. 2010;117:673–679. doi:10.1016/j.ophtha.2009.09.023 [CrossRef]
  20. Rabinowitz YS. Keratoconus. Surv Ophthalmol. 1998;42:297–319. doi:10.1016/S0039-6257(97)00119-7 [CrossRef]



Area Under ROC Curve Representing Each ORA Parameter’s Ability to Differentiate Between Normal Corneas and Grade I Keratoconus, and Between Normal Corneas and Keratoconic Corneas With a Thinnest Corneal Thickness of > 500 μm

Parameter Normal vs Grade I Keratoconus Normal vs Keratoconic Corneas With a Thinnest Corneal Thickness > 500 μm
aindex 0.635 0.674
aplhf 0.749 0.720
aspect1 0.727 0.767
aspect11 0.693 0.699
aspect2 0.654 0.688
aspect21 0.662 0.732
bindex 0.556 0.674
CH 0.860 0.718
CRF 0.867 0.722
dive1 0.772 0.780
dive2 0.851 0.779
dslope1 0.695 0.740
dslope11 0.650 0.565
dslope2 0.605 0.636
dslope21 0.612 0.656
h1 0.858 0.849
h11 0.858 0.849
h2 0.840 0.857
h21 0.840 0.857
IOPcc 0.543 0.513
IOPg 0.789 0.628
mslew1 0.746 0.766
mslew2 0.725 0.719
path1 0.694 0.590
path11 0.636 0.533
path2 0.709 0.672
path21 0.606 0.602
p1area 0.904 0.882
p1area1 0.883 0.861
p2area 0.908 0.867
p2area1 0.880 0.821
slew1 0.711 0.752
slew2 0.729 0.764
uslope1 0.712 0.759
uslope11 0.708 0.758
uslope2 0.734 0.764
uslope21 0.730 0.749
w1 0.686 0.544
w11 0.610 0.573
w2 0.649 0.567
w21 0.660 0.540

10.3928/1081597X-20130819-05

Sign up to receive

Journal E-contents