To derive limits of metric keratoconus indices for classification into keratoconus stages.
Validity and reliability analysis of diagnostic tools.
A total of 126 patients from the keratoconus center of Homburg/Saar were evaluated with respect to Amsler criteria, using Pentacam (Keratoconus Index [KI], Topographic Keratoconus Classification [TKC]), Topographic Modeling System (Smolek/Klyce, Klyce/Maeda), and Ocular Response Analyzer (Keratoconus Match Probability [KMP], Keratoconus Match Index [KMI]). Mean value, standard deviation, 90% confidence interval, and the Youden J index for definition of the thresholds were evaluated.
For separation of keratoconus stages 0/1/2/3/4 we derived the following optimum thresholds: for KI 1.05/1.15/1.31/1.49 and for KMI 0.77/0.32/-0.08/-0.3. For Smolek/Klyce and Klyce/Maeda high standard deviations and overlapping confidence intervals were found; therefore no discrete thresholds could be defined. Nevertheless, for them we still found a good sensitivity and specificity in discriminating between healthy (stage 0) and keratoconus (stages 2–4) eyes in comparison with the other indices.
We derived thresholds for the metric keratoconus indices KI and KMI, which allow classification of keratoconus stages. These now need to be validated in clinical use. Smolek/Klyce and Klyce/Maeda were not sufficiently sensitive to allow classification into individual stages, but these indices did show a good specificity and sensitivity in discriminating between keratoconus and healthy eyes.
Keratoconus is an ectatic noninflammatory corneal disorder in which the cornea turns to a conical shape owing to thinning of the corneal stroma. This corneal thinning typically induces irregular astigmatism, myopia, and protrusion, leading to a mild to marked deterioration of visual acuity. In most cases the disease is bilateral and mostly diagnosed in the second or third decade of life, with an incidence of 55:100 000. Different diagnostic devices are available to assist the examiner in confirming the diagnosis and in classifying the keratoconus into stages for monitoring the disease. In clinical routine, tomography or topography systems, aberrometers, or the Ocular Response Analyzer (ORA) are used. Each device offers the examiner a series of indices, which may be interpreted as stand-alone values or in combination.
Keratoconus indices in general can be split into categorical values , which allow a classification into grade 0 (healthy), 1 (suspect), 2 (mild), 3 (moderate), or 4 (severe), and metric values . Metric values typically do not allow classification, but based on a threshold a binary decision is possible, indicating whether a finding is normal or pathologic.
The purpose of the current study is to investigate the impact of metric keratoconus indices in classifying the keratoconus stages normal, suspect, mild, moderate, and severe, and to derive proper limits for classification.
Patients and Methods
This validity and reliability study of diagnostic tools was conducted at the Department of Ophthalmology, Saarland University Medical Center in Homburg/Saar (UKS), Germany. The study and data acquisition were carried out with approval from the ethics committee of the Saarland Medical Association (Ethik-Kommission der Ärztekammer des Saarlandes, Nr. 157/10). The study adhered to the tenets of the Declaration of Helsinki. Each subject provided informed written consent to participate in this research study.
Patients were recruited from the database of the Homburger Keratoconus Center (HKC). In the HKC, patients from our outpatient service with unilateral or bilateral keratoconus were included, as well as patients without corneal abnormality but with thyroid diseases. Additional patients from the outpatient service of the endocrinology department of the UKS, Internal Medicine II, were also recruited. All patients underwent a complete ophthalmologic examination, including visual acuity test and refractometry and slit-lamp biomicroscopy to estimate possible keratoconus stage based on Amsler criteria, and also topography, tomography, and biomechanical examination.
We used Topographic Modeling System 5 (TMS; Tomey Corporation, Nagoya, Japan) for generating topography. A trans-illuminated cone with concentric ring patterns (Placido disc) is reflected off the cornea and imaged onto an on-axis charge-coupled device camera. The distortion of the Placido mires with respect to the reference is transferred to a map of corneal slope, curvature, or height. Maeda and Klyce created different indices, such as the Keratoconus Index (KCI), which relates to the probability of the measured topography being diagnosed as keratoconus or normal. The KCI uses a binary decision-making tree with the input from 5 other indices (Keratoconus Prediction Index, Differential Sector Index, Opposite Sector Index, and Center/Surround Index). The Keratoconus Severity Index (KSI) is a combination of a neural network and a binary decision-making tree.
For tomography we used the Pentacam (Oculus Optikgeräte GmbH, Heidelberg Germany). A rotating camera captures the diffuse volume scattering of a monochromatic slit-light source projected onto the cornea and the anterior eye segment. Scheimpflug imaging creates a series of keratoconus-specific indices. We selected the parameters Keratoconus Index (KI) and Topographic Keratoconus Classification (TKC) for keratoconus classification. TKC allows a classification into 5 grades: 0 (normal) to 4 (severe keratoconus). In some cases intermediate grades (eg, 1–2) are displayed. In this case we documented the lower of both values.
With the ORA (Reichert Ophthalmic Instruments, Depew, New York, USA) biomechanical parameters can be extracted alongside intraocular pressure. ORA applanates the cornea by means of an air puff and records the intensity of the light reflected off the corneal vertex. In software version 3.01 or higher a normative database has been developed and a keratoconus screening tool implemented. The mathematical characterizations of the waveform signal such as height, slope, and width are compared with the reference dataset in the database. The Keratoconus Match Index (KMI), which is a neural network calculation of 7 parameters, represents the similarity of the waveform of the patient’s eye compared with the waveform in the database. The Keratoconus Match Probability (KMP) classifies the measurement as potentially normal, suspect, mild, moderate, or severe keratoconus.
An overview on the thresholds of TMS, Pentacam, and ORA is given in Table 1 .
|KI||1.04–1.07 normal |
1.07–1.15 grade 1
1.10–1.25 grade 2
1.15–1.45 grade 3
>1.5 grade 4
|TMS||KCI||0–5% keratoconus suspect |
>5% manifest keratoconus
|KSI||0–15% normal |
15%–30% keratoconus suspect
>30% manifest keratoconus
|ORA||KMP||Normal, suspect, mild, moderate, severe||Categorical|
|KMI||Typically between 0 and 1||Metric|
Statistical analysis was performed using SPSS software (SPSS version 19.0; International Business Machines Corporation, Armonk, New York, USA). Descriptive evaluation of data was performed using mean, standard deviation, and 90% confidence interval. Mean, standard deviation, and 90% confidence interval of the metric indices were allocated to the stages of the categorical values. To compare the different measurement systems Spearman R 2 was calculated.
To derive thresholds of the metric keratoconus indices for classifying into keratoconus grades the Youden J statistic was used. This was calculated as the sum of sensitivity and specificity minus 1. J lies within a range from −1 to +1. A diagnostic test is considered to yield reasonable results for positive values of J. Higher values indicate better performance of the diagnostic test in terms of discrimination between stages. We analyzed receiver operating characteristic (ROC) curves in order to determine the value giving the best separation between stages based on the Youden index and this was defined as the threshold for discrimination between adjacent stages.
In cases where stages could not be separated, we evaluated the diagnostic capacity for binary decision—to discriminate between normal (stage 0) vs keratoconus (stage 2–4) by calculating cross tables, sensitivity, specificity, and P values. Cases classified as suspect (stage 1) were excluded from the binary decision as they do not strictly refer to either normal or keratoconus cases.
One hundred and twenty-six eyes of 126 patients were enrolled in this study. The mean age was 37 years (11–75 years), and 26 of 126 (33%) were female.
Mean values, standard deviations, and 90% confidence intervals were assessed for the metric values and allocated to the different stages of categorical values ( Supplemental Table , available at AJO.com ). Table 2 shows the comparison of the different measurement systems. Spearman coefficient of determintation R 2 is given for each pair. All P are <.01.
|Spearman R 2||KI||KMI||KCI||KSI|
With the KCI and KSI values, the standard deviations and confidence intervals are large and mainly overlapping. Therefore, finding thresholds for discrimination between stages was not possible. We decided to determine the thresholds to discriminate between the stages using KI and KMI. Table 3 shows the thresholds of each stage of KI and KMI referring resampling of classification according to TKC, KMP, and Amsler. In Table 4 a clinical recommendation for the thresholds based on our datasets is provided for a classification of keratoconus into grades 0–4.
|Threshold of KI (Pentacam)||Youden Index|
|TKC (Pentacam)||KMP (ORA)||Amsler||TKC (Pentacam)||KMP (ORA)||Amsler|
|Threshold of KMI (ORA)||Youden Index|
|TKC (Pentacam)||KMP (ORA)||Amsler||TKC (Pentacam)||KMP (ORA)||Amsler|
|Limits||KI (Pentacam)||KMI (ORA)|
|2||1.15–1.30||0.32 to −0.07|
|3||1.31–1.48||−0.08 to −0.29|
Cross tables, sensitivity, and specificity for the evaluation of a binary separation of keratoconus and normal eyes are shown in Table 5 . A high sensitivity and specificity could be reached with all systems.