Keratoconus Screening Using Values Derived

From Auto-Keratometer Measurements: A

Multicenter Study

TAKASHI KOJIMA, TOMOYA NISHIDA, TOMOAKI NAKAMURA, AKENO TAMAOKI, ASATO HASEGAWA,
YUKI TAKAGI, HIROYUKI SATO, AND KAZUO ICHIKAWA

PURPOSE: Screening of early-stage keratoconus using keratometer–derived parameters enabled better discrimi-
auto-keratometer parameters. nation of early-stage keratoconus than the isolated

DESIGN: Evaluation of a screening approach. parameters. (Am J Ophthalmol 2020;215:127–134.

METHODS: At 5 major centers in Japan, we enrolled Ó 2020 Elsevier Inc. All rights reserved.)
123 eyes of 123 patients with Amsler-Krumeich classifi-
cation stage 1 (<50 years of age [average 26.36 ± 8.68

K
years]; 84/39 male/female) and 205 eyes of 205 healthy ERATOCONUS IS A PROGRESSIVE ECTATIC CORNEAL
subjects (average age 26.20 ± 7.34 years, 139/66 male/fe- disease that develops in both eyes. As it progresses,
male). Participants were divided 2:1 into a prediction the cornea deforms because of weakness of its
group and an application group. In the prediction group, collagen structure. If irregular astigmatism or higher-order
multivariate logistic regression analysis was performed aberrations of the cornea increase because of this deforma-
with keratoconus diagnosis as the dependent variable, tion, refractive correction with rigid gas permeable contact
and auto-keratometer parameters including average K, lenses is necessary; in severe cases, corneal transplantation
steep K, flat K, astigmatism, and astigmatic axis (no, may be necessary. Keratoconus often occurs at an early age
with-the-rule, against-the-rule, and oblique) as indepen- and, therefore, causes severe socioeconomic problems.
dent variables. The diagnostic probability determined by Generally, early keratoconus lacks characteristic slit-
regression analysis was defined as the keratometer kerato- lamp microscopy findings such as Fleischer rings or Vogt
conus index. The cutoff value was determined from the striae; thus, diagnosis with a corneal shape analyzer
receiver operating characteristic curve. This prediction including corneal topography or tomography is important.
equation was evaluated in the application group. Our pri- The protrusion of the posterior cornea and the abnormal
mary outcome measure was the accuracy of the prediction distributions of the central and peripheral corneal thick-
equation for discriminating keratoconus from normal nesses have been reported as major features of keratoconus
eyes. with corneal tomography.1 Corneal shape analyzers are

RESULTS: The selected explanatory variables were becoming increasingly sophisticated, and high levels of
steep K (partial regression coefficient [b] 1.284, odds ra- sensitivity and specificity have been reported for the diag-
tio [OR] 3.610), flat K (b L0.618, OR 0.539), and with- nosis of keratoconus using corneal tomography, such as
the-rule astigmatism (b L3.163, OR 0.042). The area Scheimpflug-based tomography and anterior segment opti-
under the receiver operating characteristic curve of kera- cal coherence tomography.2,3 Recently, it has also become
tometer keratoconus index was 0.90, which was signifi- possible to diagnose keratoconus with instruments that can
cantly better than individual parameters (P < .001). measure the corneal epithelial thickness distribution,4
The sensitivity and specificity values in the application corneal high-order aberration,5 corneal biomechanics,6–8
group were 85.0% and 86.7%, respectively. and the combination of corneal tomography with corneal

CONCLUSIONS: Although the sensitivity/specificity biomechanics.9,10
was not high, the new prediction equation using auto- Corneal cross-linking, first reported in 2003 by Wollen-
sak and associates,11 has been extensively studied, and
many reports have been published on its safety and effec-
tiveness at slowing keratoconus progression.12–16 Recent
Accepted for publication Feb 20, 2020.
From the Department of Ophthalmology (T.K.), Keio University studies have found that the number of cases of
School of Medicine, Tokyo, Japan; Nagoya Eye Clinic (T.K., T.Ni., keratoconus requiring corneal transplantation has greatly
T.Na.), Nagoya, Japan; Department of Ophthalmology (A.T., A.H., decreased since the introduction of corneal cross-
K.I.), Japan Community Health Care Organization, Chukyo Hospital,
Nagoya, Japan; Department of Ophthalmology (Y.T.), Iida Municipal linking.17,18 Moreover, according to estimates using the
Hospital, Nagano, Japan; Yuya Sato Eye Clinic (H.S.), Sendai, Japan; Markov model, the cost effectiveness of corneal cross-
and the Chukyo Eye Clinic (K.I.), Nagoya, Japan. linking has been shown to be better than that of conven-
Inquiries to Takashi Kojima, Department of Ophthalmology, Keio
University School of Medicine, Shinanomachi 35, Shinjuku-ku, Tokyo tional management. In addition, these increases improve
160-8582, Japan; e-mail: [email protected] even more with longer-lasting progress prevention.19,20

0002-9394/$36.00 © 2020 ELSEVIER INC. ALL RIGHTS RESERVED. 127

https://doi.org/10.1016/j.ajo.2020.02.017
Downloaded for Anonymous User (n/a) at Keio University - Shinanomachi Campus from ClinicalKey.jp by Elsevier on June 09, 2020.
For personal use only. No other uses without permission. Copyright ©2020. Elsevier Inc. All rights reserved.
 TABLE 1. Patient and Control Subject Demographic Information in All Participants

Keratoconus Group Control Group P Value

Eyes, n 123 205

Age (years), mean 6 SD 26.36 6 8.68 26.20 6 7.34 .2675
Sex, male:female 84:39 139:66 1.000
Auto-keratometer parameters (D), mean 6 SD
Average K 45.05 6 1.91 43.29 6 1.24 <.0001
Steep K 46.23 6 2.16 43.9 6 1.33 <.0001
Flat K 43.89 6 1.9 42.66 6 1.23 <.0001
Astigmatism 2.35 6 1.44 1.24 6 0.71 <.0001
Auto-keratometer astigmatism axis, n (%)
No astigmatism 1 (0.8) 0 (0) <.0001
WTR 73 (59.3) 203 (93.1)
ATR 17 (13.7) 5 (2.3)
Oblique 32 (25.8) 10 (4.6)
Distance-corrected visual acuity, n (%)
>
_20/20 82 (66.7) 204 (100) <.0001
>
_20/30 107 (87.0) 204 (100) <.0001

ATR ¼ against the rule; D ¼ diopter; K ¼ keratometric power; SD ¼ standard deviation; WTR ¼ with the rule.

In Japan, many universities and tertiary hospitals use by the institutional review board as an alternative for
corneal topography/tomography, but smaller clinics and informed consent.
opticians often do not offer these imaging techniques. Pa- The inclusion criteria for patients with keratoconus were
tients with undiagnosed early stage keratoconus often go as follows: 1) patients <50 years of age who visited one of
to opticians because of changes in refraction as keratoconus the facilities between January 2015 and December 2018; 2)
progresses, and many cases are not diagnosed until their patients who were diagnosed with keratoconus, Amsler-
corrected vision is poor. For these reasons, the early diag- Krumeich classification stage 1 (average K value <48 diop-
nosis of keratoconus is difficult at the primary eye care ters [D] and corneal astigmatism <5 D); 3) patients with no
level. ocular diseases other than keratoconus and refractive error;
Considering these conditions in Japan, we focused on and 4) patients with no history of ocular surgery. The kera-
auto-keratometers, which are available in most ophthalmic toconus diagnosis was verified by an ophthalmologist at one
clinics and eyeglass stores. The purpose of this study was to of the aforementioned facilities based on corneal topog-
create a prediction algorithm for keratoconus diagnosis us- raphy or tomography results and slit-lamp findings. We
ing only currently available values measured by auto- diagnosed keratoconus based on local corneal steepening
keratometers, and to evaluate the feasibility of this algo- and asymmetric astigmatism, as shown by corneal topog-
rithm for clinical screenings of early stage keratoconus. raphy. When using corneal tomography, in addition to
Our goal for the future is to integrate this prediction algo- the corneal topography criteria, keratoconus was diagnosed
rithm into auto-keratometers to allow for more detailed based on the protrusion of the posterior cornea and thin-
early corneal screenings at the primary eye care level. ning of the cornea at the area of protrusion. The Klyce/
Maeda index,21 a composite parameter, was used as a refer-
ence for diagnosis in corneal topography, and the Ectasia
Screening Index was used for the anterior segment optical
METHODS coherence tomography. The results of the corneal topog-
raphy or tomography were collated at the Nagoya Eye

PATIENTS AND STUDY DESIGN: This retrospective Clinic, and 2 corneal specialists (T.K., T.Na.) made the
multicenter study was conducted at 5 major clinical cen- final judgment on the diagnosis. Finally, 124 patients
ters: The Nagoya Eye Clinic, Gifu Red Cross Hospital, with keratoconus (124 eyes; mean age 28.26 6 10.13 years;
Iida Municipal Hospital, Satoh Yuya Eye Clinic, and Japan 85 men and 39 women) were enrolled.
Community Health Care Chukyo Hospital. The institu- Healthy subjects <50 years of age who had undergone an
tional review board/ethics committee (Chukyo Medical ophthalmic screening between January 2015 and
Ethics Committee) approved the current study (UMIN December 2018 at the Nagoya Eye Clinic or Satoh Yuya
ID 000033372). This study followed the tenets of the Eye Clinic were enrolled. During an initial prescreening,
Declaration of Helsinki. An opt-out method was approved the medical records of subjects with 1) a Klyce/Maeda

128 AMERICAN JOURNAL OF OPHTHALMOLOGY JULY 2020

Downloaded for Anonymous User (n/a) at Keio University - Shinanomachi Campus from ClinicalKey.jp by Elsevier on June 09, 2020.
For personal use only. No other uses without permission. Copyright ©2020. Elsevier Inc. All rights reserved.
 TABLE 2. Demographic Information in the Prediction Group

Keratoconus Group Control Group P Value

Eyes, n 83 123
Age (years), mean 6 SD 26.13 6 9.04 26.28 6 7.20 .5672
Sex, male:female 59:24 86:36 1.000
Auto-keratometer parameters (D), mean 6 SD
Average K 44.97 6 2.02 43.29 6 1.29 <.0001
Steep K 46.09 6 2.3 43.95 6 1.37 <.0001
Flat K 43.86 6 2.0 42.64 6 1.31 <.0001
Astigmatism 2.23 6 1.52 1.31 6 0.78 <.0001
Auto-keratometer astigmatism axis, n (%)
No astigmatism 0 (0) 0 (0) <.0001
WTR 49 (59.0) 114 (93.4)
ATR 10 (12.0) 5 (5.0)
Oblique 24 (28.9) 3 (2.5)

ATR ¼ against the rule; D ¼ diopter; K ¼ keratometric power; SD ¼ standard deviation; WTR ¼ with the rule.

index of 0% as shown by corneal topography, 2) an ectasia der axis. For the multivariate analysis, the astigmatism axis
screening score of 0% in the anterior segment with optical was classified into 4 types, depending on the axis angle: no
coherence tomography (Casia; Tomey Corp, Aichi, Japan), astigmatism, with-the-rule (WTR) astigmatism (steep axis
and 3) no abnormal findings with slit-lamp examination 608 -1198 ), oblique astigmatism (steep axis 308 -598 , 1208 -
were identified. Subsequently, corneal specialists (T.K., 1498 ), and against-the-rule (ATR) astigmatism (steep
T.Na.) examined all corneal topography/tomography re- axis 08 -298 , 1508 -1798 ).
sults and confirmed that they were normal. The The keratoconus diagnosis was set as the dependent var-
prescreened healthy subjects were enrolled using the iable, with keratoconus ¼ 1 and normal ¼ 0. A logistic
following methods while masking their keratometer values. regression analysis was performed by including all the
Since the gender ratio of the keratoconus group was parameters of the auto-keratometer (as described above)
approximately 2:1 (male:female), normal cases were as independent variables. The astigmatism axis was incor-
randomly selected from the prescreening pool so that the porated into the analysis by applying a dummy variable.
gender ratio was 2:1 for each age group. A random number The stepwise method was used for variable selection. More-
table was used as the randomization method. Finally, 205 over, the cutoff value was calculated from the receiver
eyes of 205 subjects were selected. operating characteristic (ROC) curve analysis using the
The keratoconus and control groups were randomly Youden index. The diagnosis probability obtained from
divided (2:1 ratio per group) to create the prediction equa- the prediction equation was named the keratometer kera-
tion and application groups, respectively. For the random- toconus index (KKI). In addition to the ROC analysis using
ization method, the integers 1, 2, and 3 were generated in KKI, an ROC analysis of the single parameters constituting
Excel (Microsoft Corp, Redmond, Washington, USA), the KKI was performed, and the areas under the ROC
and the prediction formula was created for cases attributed curves were compared.
to the numbers 1 and 2, whereas the application group
consisted of the cases assigned to the number 3.
APPLICATION OF THE PREDICTION EQUATION: The
Distance-corrected visual acuity (DCVA) was also KKI was calculated for the application group, and the levels
collected from the patient charts. For all patients, visual of sensitivity and specificity were determined with the pre-
acuity (VA) was measured using a decimal VA chart viously established cutoff value.
with Landolt rings, and the decimal VA was converted to
the logarithm of the minimum angle of resolution
STATISTICAL ANALYSES: x2 tests were performed to
(logMAR). analyze the differences in the age distributions and astigma-
tism axes. The Mann-Whitney U test was used for compar-

GENERATION OF THE PREDICTION EQUATION FOR THE ison of parameters between the control and keratoconus
DIAGNOSIS OF KERATOCONUS: Four parameters measured groups. The DeLong test was applied to compare the area
by the auto-keratometer were evaluated: keratometric po- under the ROC curve (AUROC) between KKI and other
wer at the steep meridian (steep K), keratometric power isolated parameters. SPSS software (v 19; IBM Corp,
at the flat meridian (flat K), amount of cylinder, and cylin- Chicago, Illinois, USA) was used for all statistical analyses,

VOL. 215 SCREENING OF EARLY-STAGE KERATOCONUS WITH AUTO-KERATOMETERS 129

Downloaded for Anonymous User (n/a) at Keio University - Shinanomachi Campus from ClinicalKey.jp by Elsevier on June 09, 2020.
For personal use only. No other uses without permission. Copyright ©2020. Elsevier Inc. All rights reserved.
 TABLE 3. Demographic Information in the Application Group

Keratoconus Group Control Group P Value

Eyes, n 40 83
Age (years), mean 6 SD 26.97 6 7.79 26.09 6 7.58 .2623
Sex, male:female 25:15 53:30 .8428
Auto-keratometer parameters (D), mean 6 SD
Average K 45.18 6 1.70 43.24 6 1.16 <0.0001
Steep K 46.48 6 1.85 43.82 6 1.28 <0.0001
Flat K 43.94 6 1.70 42.70 6 1.11 <0.0001
Astigmatism 2.54 6 1.21 1.12 6 0.57 <0.0001
Auto-keratometer astigmatism axis, n (%)
No astigmatism 1 (2.5) 0 (0) <.0001
WTR 24 (60.0) 76 (91.6)
ATR 8 (20.0) 5 (6.0)
Oblique 7 (17.5) 2 (2.4)

ATR ¼ against the rule; D ¼ diopter; K ¼ keratometric power; SD ¼ standard deviation; WTR ¼ with the rule.

including the multivariate analysis. P < .05 was considered significantly different from that of the control group
statistically significant. both in the prediction and application groups (P <
.0001).

OUTCOMES OF THE MULTIVARIATE LOGISTIC REGRES-

SION ANALYSES: In the prediction group, a multivariate
RESULTS logistic regression analysis was performed (Table 4). The
parameters steep K, flat K, and WTR astigmatism were
TABLE 1 PRESENTS THE DEMOGRAPHIC INFORMATION OF
selected as explanatory variables to predict keratoconus.
all enrolled participants. All parameters including the The regression coefficients of steep K, flat K, and WTR
average K, steep K, flat K, and astigmatism, as measured astigmatism were 1.284, 0.618, and 3.163, respectively.
by the auto-keratometer, showed significantly higher Accordingly, the odds ratios of these parameters were
values in the keratoconus group than in the control group. 3.610, 0.539, and 0.042, respectively.
Regarding the astigmatism axis, the percentages of partici- The prediction equation was determined as
pants with WTR, ATR, and oblique astigmatism were follows:logit ¼ 1.284 3 (steep K [D])0.618 3 (flat K
59.3%, 13.7%, and 25.8% in the keratoconus group, respec- [D])3.163 3 (1: WTR astigmatism; 0: no, ATR, or obli-
tively. In contrast, these values were 93.1%, 2.3%, and que astigmatism)28.662 Figure displays the ROC
4.6%, respectively, in the control group; most subjects curves for each parameter used for the diagnosis. For the pa-
were affected by WTR astigmatism. The distributions of rameters KKI, steep K, flat K, astigmatism, and absence of
the astigmatism axis were significantly different between WTR astigmatism, the AUROCs were 0.9032, 0.7846,
the 2 groups (P < .0001). A DCVA of 20/20 or better 0.6902, 0.7061, and 0.6702, respectively. The value of
was recorded in 66.7% and 100% of the eyes in the kerato- the Youden index (ie, sensitivity þ specificity  1) was
conus and control groups, respectively. Similarly, a DCVA maximized, and the cutoff value for KKI was determined
of 20/30 or better was recorded in 87.0% and 100% of the to be 0.1574. This cutoff value was converted to the prob-
eyes in the keratoconus and control groups, respectively. ability of keratoconus diagnosis (KKI) using the following
Tables 2 and 3 show the demographic information for conversion formula:Probability of keratoconus diagnosis
the prediction and application groups, respectively. The (KKI) ¼ exp(logit)/(1 þ exp[logit]) Thus, the KKI cutoff
prediction group showed no differences in age or sex value was calculated to be 0.461.
ratios between the keratoconus and control groups. In
both the prediction and application groups, all the
COMPARISON OF DIAGNOSTIC CAPABILITY OF KERATO-
measurement parameters taken by the auto-keratometer CONUS BETWEEN KKI AND OTHER ISOLATED
were significantly higher in the keratoconus group KERATOMETER-DERIVED PARAMETERS: Table 5 shows
compared with the control group. For the astigmatism the comparison of the diagnostic probability between
axis classification, 40.9% of patients in the keratoconus KKI and the other isolated parameters. According to the
group showed ATR astigmatism and oblique astigma- DeLong test, the AUROC of KKI was significantly larger
tism, and the distribution in the keratoconus group was than that of the isolated parameters (steep K, P ¼ .0049;

130 AMERICAN JOURNAL OF OPHTHALMOLOGY JULY 2020

Downloaded for Anonymous User (n/a) at Keio University - Shinanomachi Campus from ClinicalKey.jp by Elsevier on June 09, 2020.
For personal use only. No other uses without permission. Copyright ©2020. Elsevier Inc. All rights reserved.
 TABLE 4. Results of Multivariate Logistic Regression Analysis

Factors Regression Coefficient SE Wald Statistic df P Value OR (95% CI)

Constant 33.360 7.187 21.548 1 <.0001

Steep K (D) 1.824 0.303 36.321 1 <.0001 6.198 (3.425-11.218)
Flat K (D) 1.085 0.277 15.285 1 .003 0.338 (0.196-0.582)
WTR astigmatism 2.090 0.589 12.580 1 <.0001 0.124 (0.039-0.393)

CI ¼ confidence interval; D ¼ diopter; df ¼ degree of freedom; K ¼ keratometric power; OR ¼ odds ratio; SE ¼ standard error; WTR ¼ with the rule.

flat K, P < .0001; astigmatism, P < .0001; and absence of were 85.0% and 86.7%, respectively, which appeared to
WTR astigmatism, P < .0001). be better than when using the isolated parameters. Previous
keratoconus screening methods include the Klyce/Maeda

APPLICATION OF THE ACCURACY OF THE PREDICTION keratoconus prediction index24 and the Scheimpflug
EQUATION: The accuracy of the prediction equation was tomography–based Belin/Ambrósio enhanced ectasia
examined using the application group, in which the sensi- map.25 In a study discriminating keratoconus from normal
tivity and specificity levels were 85.0% (34/40) and 86.7% cornea using parameters determined by Scheimpflug-based
(72/83), respectively. tomography, the index of vertical asymmetry had a sensi-
tivity of 87.5%, a specificity of 96.3%, and an AUROC
of 95.24%.2 In comparison, the sensitivity and specificity
DISCUSSION of the prediction equation in the current study appear
low. In general, the features of the corneal shape in kerato-
TO FACILITATE AUTOMATIC RISK DETERMINATION FOR conus include an abnormal posterior elevation and
early keratoconus using only auto-keratometer values in abnormal corneal thickness distribution.1 However, the
the future, we used a multivariate logistic regression anal- auto-keratometer cannot evaluate the corneal posterior
ysis to create an equation that predicts early keratoconus. surface and the corneal thickness, which could limit the
Our initial descriptive analysis revealed that the K value diagnostic probability, especially in early stage
and the amount of astigmatism were higher in the kerato- keratoconus.
conus group than in the control group. Moreover, the per- Naderan and associates23 reported that both normal and
centages of participants with ATR astigmatism and oblique keratoconus eyes can be distinguished with a sensitivity of
astigmatism were increased in the keratoconus group. 89.5% and a specificity of 85% when an anterior corneal
These results are consistent with previous reports.22,23 astigmatism value of 1.8 D measured with Scheimpflug-
We calculated the AUROC when using KKI as 0.9032, based corneal tomography is used as the cutoff value. The
confirming that our generated prediction formula is rela- mean K value of keratoconus eyes enrolled in their study
tively accurate. Furthermore, the AUROC when using was 50.2 6 4.6 D. Our study aimed to distinguish patients
KKI was better than when using other isolated parameters, with early keratoconus from healthy subjects, and therefore
including steep K, flat K, and astigmatism. The parameters we enrolled a population with less advanced keratoconus
steep K, flat K, and WTR astigmatism were selected as (mean K, 45.05 6 1.91 D), making it difficult to make a
explanatory variables; the probability of the presence of direct comparison. In the future, it will be necessary to
keratoconus is higher if the value for steep K is higher, examine whether our prediction equation is also useful in
and if the value for flat K is lower. According to the odds more advanced cases.
ratios, the factors most related to the presence of keratoco- We found that 93.1% of individuals in the healthy con-
nus were the absence of WTR astigmatism, greater steep K trol group showed WTR astigmatism, and the mean astig-
values, and lower flat K values, in this order. In this study, matism value was 1.24 6 0.71 D. In a previous hospital-
the odds ratio for flat K was 0.539. Although selected as a based study conducted in China, 91.24% of patients in a
statistically significant explanatory variable, its clinical 20- to 29-year-old age group showed WTR astigmatism.26
impact appears to be smaller than that of steep K. Further- Moreover, the mean anterior corneal astigmatism in the
more, WTR astigmatism was selected as an explanatory 20- to 29-year-old age group was reported to be 1.26 6
variable. Given that its coefficient has a negative value, 1.04 D. Therefore, the amount of astigmatism and axis dis-
the probability of diagnosing keratoconus appears to be tribution of normal eyes in our study appears consistent
lower in eyes with WTR astigmatism compared with eyes with previous reports. However, a study conducted in
with either ATR astigmatism or oblique astigmatism. Iran reported that in patients of a similar age to those in
Our verification of the prediction equation in the appli- our study, only 71.5% of normal eyes showed WTR astig-
cation group showed that the sensitivity and specificity matism, and 22.2% showed oblique astigmatism. It is

VOL. 215 SCREENING OF EARLY-STAGE KERATOCONUS WITH AUTO-KERATOMETERS 131

Downloaded for Anonymous User (n/a) at Keio University - Shinanomachi Campus from ClinicalKey.jp by Elsevier on June 09, 2020.
For personal use only. No other uses without permission. Copyright ©2020. Elsevier Inc. All rights reserved.
FIGURE. The receiver operating characteristic curve analysis delineating the sensitivity and specificity of keratoconus screenings
using an auto-keratometer. (A) Receiver operating characteristic curves for each diagnosis parameter; the estimated area under
the curve for the keratometer keratoconus index (KKI) is 0.9032 (95% confidence interval [CI] 0.8144-0.9217). (B) The levels
of sensitivity and specificity of the representative cutoff values. Bold values indicate the cutoff value. WTR [ with-the-rule.

possible that the distribution of astigmatism may change power include age, level of education, ocular biometry
due to differences in measurement equipment or ethnicity, (eg, anterior chamber depth and axial length), and refrac-
and sufficient validation will be essential when using our tive error.37,38 For these reasons, it may be necessary to
prediction equation under different conditions. verify the prediction equation in different age and ethnicity
It is also noteworthy to mention that the remarkable ad- groups. Third, we focused on the discrimination between
vances in artificial intelligence using deep learning could early keratoconus and healthy eyes. Further studies are
make it highly useful for the screening of keratoconus27; necessary to investigate how well the established equation
such techniques could further improve the accuracy of distinguishes keratoconus in eyes with other corneal abnor-
the results of this study. However, the logistic regression malities, such as pellucid marginal degeneration, corneal
analysis used in this study to calculate the prediction equa- scarring, and irregular astigmatism after keratoplasty.
tion is arguably more accessible and can be used worldwide. Fourth, the levels of sensitivity and specificity were calcu-
This study has some limitations. First, previous studies lated based on patients with early keratoconus who had
have demonstrated that age-related corneal changes to- visited the hospital and healthy control subjects. However,
ward ATR astigmatism become detectable after 60 years in primary care settings, it is possible that earlier stage kera-
of age.26,28,29 In this study, we prioritized the early detec- toconus may exist; therefore, the sensitivity and specificity
tion of keratoconus in young patients who were susceptible values calculated in this study may not be applicable. In the
to developing this disorder; this is reflected by our inclusion future, it will be necessary to further examine how many
criteria regarding the age of the enrolled participants. In cases of early stage keratoconus can be detected in patients
future studies, it will be necessary to examine the diagnostic in primary care settings with this system.
probability using the KKI in elderly patients with keratoco- In conclusion, this study provides an equation that en-
nus. Second, the corneal refractive power has been assessed ables better discrimination of early stage of keratoconus
in many population- and hospital-based studies in many than the isolated parameters, with a clinically relevant
countries including Japan, the United States, and level of accuracy using only values generated by an auto-
Australia.30–36 The corneal refractive power has been keratometer. This new prediction equation does not
reported to have no obvious interethnic differences,37 but replace the topography/tomography-based diagnosis of
some reports recognize variations of about 1 D. In the cur- keratoconus, but rather can be used to screen for patients
rent study, the average corneal refractive power in the con- who require a more complete and complex evaluation.
trol group was 43.29 D, which seems to be at the lower limit Therefore, this risk assessment method could be useful as
of previous reports. Factors that affect corneal refractive a first-line screening tool in hospitals or clinics with no

132 AMERICAN JOURNAL OF OPHTHALMOLOGY JULY 2020

Downloaded for Anonymous User (n/a) at Keio University - Shinanomachi Campus from ClinicalKey.jp by Elsevier on June 09, 2020.
For personal use only. No other uses without permission. Copyright ©2020. Elsevier Inc. All rights reserved.
 TABLE 5. Comparison of the Diagnostic Probability Between the Keratometer Keratoconus Index and Keratometer-Derived Isolated
Parameters

Parameters Cutoff Value AUROC P Valuea Sensitivity Specificity

KKI 0.461 0.903 — 85.0% 86.7%

Steep K 44.88 D 0.785 .0049 75.0% 79.5%
Flat K 43.74 D 0.690 <.0001 60.0% 80.7%
astigmatism 1.865 D 0.70 <.0001 70.0% 86.7%
Absence of WTR astigmatism — 0.672 <.0001 37.5% 91.6%

AUROC ¼ area under the receiver operating characteristic curve; D ¼ diopter; K ¼ keratometric power; KKI ¼ keratometer keratoconus in-
dex; WTR ¼ with the rule.
A cutoff value was calculated from the ROC curve of each parameter using the application group, and then sensitivity and specificity were
analyzed from the application group.
a
P value comparing KKI with other parameters is shown.

access to corneal topography/tomography. In future large- acquisition, Investigation, Methodology, Validation.

scale population-based studies, it will be necessary to iden- Tomoaki Nakamura: Writing - review & editing,
tify how often patients who should be referred to corneal Conceptualization, Formal analysis, Project administra-
specialists can be detected using this method. tion, Resources, Software, Supervision, Validation.
Akeno Tamaoki: Writing - review & editing, Conceptu-
alization, Data curation, Formal analysis, Methodology,
Validation. Asato Hasegawa: Writing - review & editing,
CRediT AUTHORSHIP CONTRIBUTION Data curation, Formal analysis, Methodology, Validation.
STATEMENT Yuki Takagi: Writing - review & editing, Data curation,
Formal analysis, Validation. Hiroyuki Sato: Writing - re-
TAKASHI KOJIMA: WRITING - REVIEW & EDITING, CONCEP-
view & editing, Data curation, Formal analysis, Valida-
tualization, Funding acquisition, Investigation, Method-
tion. Kazuo Ichikawa: Writing - review & editing,
ology, Project administration, Validation, Visualization,
Conceptualization, Project administration, Resources,
Writing - original draft. Tomoya Nishida: Writing - re-
Software, Supervision.
view & editing, Data curation, Formal analysis, Funding

ALL AUTHORS HAVE COMPLETED AND SUBMITTED THE ICMJE FORM FOR DISCLOSURE OF POTENTIAL CONFLICTS OF INTEREST.
Funding/Support: This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. Financial Dis-
closures: Dr Kojima has received personal fees from Staar Surgical, Santen Pharmaceutical, Otsuka Pharmaceutical, Johnson & Johnson, and Alcon Japan,
outside of the submitted work. In addition, Dr Kojima has a pending patent (2019-045345) licensed to Takashi Kojima. Mr Nishida has no commercial
interests to disclose. Dr Nakamura has received personal fees from Staar Surgical, Santen Pharmaceutical, Otsuka Pharmaceutical, Carl Zeiss Meditec, and
Johnson & Johnson, outside of the submitted work. Dr Hasegawa has no commercial interests to disclose. Ms Tamaoki has received personal fees from
Tomey Corporation, Carl Zeiss Meditec, NIDEK, Santec Corporation, and Alcon Japan, outside the submitted work. Dr Takagi has no commercial in-
terests to disclose. Dr Satoh has no commercial interests to disclose. Dr Ichikawa has received personal fees from Santen Pharmaceutical, Alcon Japan,
Kowa, and Carl Zeiss Meditec, outside of the submitted work, and has 2 patents (2012-005573 and 2011-218076), both licensed to Kazuo Ichikawa.
We thank Mr Yoshiki Tanaka (Chukyo Eye Clinic, Nagoya, Japan) for his great contribution in the statistical analysis of this study. All authors attest
that they meet the current ICMJE criteria for authorship.

REFERENCES
4. Reinstein DZ, Archer TJ, Gobbe M. Corneal epithelial thick-
1. Randleman JB, Dupps WJ Jr, Santhiago MR, et al. Screening ness profile in the diagnosis of keratoconus. J Refract Surg
for keratoconus and related ectatic corneal disorders. Cornea 2009;25(7):604–610.
2015;34(8):e20–e22. 5. Colak HN, Kantarci FA, Yildirim A, et al. Comparison of
2. Hashemi H, Khabazkhoob M, Pakzad R, et al. Pentacam accu- corneal topographic measurements and high order aberra-
racy in discriminating keratoconus from normal corneas: a diag- tions in keratoconus and normal eyes. Cont Lens Anterior
nostic evaluation study. Eye Contact Lens 2019;45(1):46–50. Eye 2016;39(5):380–384.
3. Shetty R, Rao H, Khamar P, et al. Keratoconus screening 6. Vinciguerra R, Ambrósio R Jr, Elsheikh A, et al. Detection of
indices and their diagnostic ability to distinguish normal from keratoconus with a new biomechanical index. J Refract Surg
ectatic corneas. Am J Ophthalmol 2017;181:140–148. 2016;32(12):803–810.

VOL. 215 SCREENING OF EARLY-STAGE KERATOCONUS WITH AUTO-KERATOMETERS 133

Downloaded for Anonymous User (n/a) at Keio University - Shinanomachi Campus from ClinicalKey.jp by Elsevier on June 09, 2020.
For personal use only. No other uses without permission. Copyright ©2020. Elsevier Inc. All rights reserved.
 7. Sedaghat MR, Momeni-Moghaddam H, Ambrósio R Jr, et al. 22. Choi JA, Kim MS. Progression of keratoconus by longitudinal
Diagnostic ability of corneal shape and biomechanical param- assessment with corneal topography. Invest Ophthalmol Vis Sci
eters for detecting frank keratoconus. Cornea 2018;37(8): 2012;53(2):927–935.
1025–1034. 23. Naderan M, Rajabi MT, Zarrinbakhsh P. Distribution of ante-
8. Hashemi H, Beiranvand A, Yekta A, et al. Biomechanical rior and posterior corneal astigmatism in eyes with keratoco-
properties of early keratoconus: suppressed deformation signal nus. Am J Ophthalmol 2016;167:79–87.
wave. Cont Lens Anterior Eye 2017;40(2):104–108. 24. Maeda N, Klyce SD, Smolek MK, Thompson HW. Auto-
9. Ambrósio R Jr, Lopes BT, Faria-Correia F, et al. Integration of mated keratoconus screening with corneal topography anal-
Scheimpflug-based corneal tomography and biomechanical ysis. Invest Ophthalmol Vis Sci 1994;35(6):2749–2757.
assessments for enhancing ectasia detection. J Refract Surg 25. Duncan JK, Belin MW, Borgstrom M. Assessing progression
2017;33(7):434–443. of keratoconus: novel tomographic determinants. Eye Vis
10. Chan TCY, Wang YM, Yu M, Jhanji V. Comparison of (Lond) 2016;3:6.
corneal tomography and a new combined tomographic 26. Shao X, Zhou KJ, Pan AP, et al. Age-related changes in
biomechanical index in subclinical keratoconus. J Refract corneal astigmatism. J Refract Surg 2017;33(10):696–703.
Surg 2018;34(9):616–621. 27. Yousefi S, Yousefi E, Takahashi H, et al. Keratoconus severity
11. Wollensak G, Spoerl E, Seiler T. Riboflavin/ultraviolet-a- identification using unsupervised machine learning. PLoS
induced collagen crosslinking for the treatment of keratoco- One 2018;13(11):e0205998.
nus. Am J Ophthalmol 2003;135(5):620–627. 28. Naeser K, Savini G, Bregnhøj JF. Age-related changes in
12. Caporossi A, Mazzotta C, Baiocchi S, Caporossi T. Long-term with-the-rule and oblique corneal astigmatism. Acta Ophthal-
results of riboflavin ultraviolet a corneal collagen cross- mol 2018;96(6):600–606.
linking for keratoconus in Italy: the Siena eye cross study. 29. Namba H, Kawasaki R, Sugano A, et al. Cross-sectional and
Am J Ophthalmol 2010;149(4):585–593. longitudinal investigation of the power vector in astigmatism:
13. Hashemi H, Seyedian MA, Miraftab M, et al. Corneal The Yamagata Study (Funagata). Cornea 2018;37(1):53–58.
collagen cross-linking with riboflavin and ultraviolet a irradi- 30. Wong TY, Foster PJ, Ng TP, et al. Variations in ocular biom-
ation for keratoconus: long-term results. Ophthalmology 2013; etry in an adult Chinese population in Singapore: the Tanjong
120(8):1515–1520. Pagar Survey. Invest Ophthalmol Vis Sci 2001;42(1):73–80.
14. Raiskup-Wolf F, Hoyer A, Spoerl E, Pillunat LE. Collagen 31. Yuen LH, He M, Aung T, et al. Biometry of the cornea and
crosslinking with riboflavin and ultraviolet-A light in kerato- anterior chamber in Chinese eyes: an anterior segment opti-
conus: long-term results. J Cataract Refract Surg 2008;34(5): cal coherence tomography study. Invest Ophthalmol Vis Sci
796–801. 2010;51(7):3433–3440.
15. Vinciguerra R, Romano MR, Camesasca FI, et al. Corneal 32. Eysteinsson T, Jonasson F, Sasaki H, et al. Central corneal
cross-linking as a treatment for keratoconus: four-year thickness, radius of the corneal curvature and intraocular pres-
morphologic and clinical outcomes with respect to patient sure in normal subjects using non-contact techniques: Reykja-
age. Ophthalmology 2013;120(5):908–916. vik Eye Study. Acta Ophthalmol Scand 2002;80(1):11–15.
16. Hersh PS, Stulting RD, Muller D, et al. U.S. multicenter clin- 33. Ojaimi E, Rose KA, Morgan IG, et al. Distribution of ocular
ical trial of corneal collagen crosslinking for treatment of biometric parameters and refraction in a population-based
corneal ectasia after refractive surgery. Ophthalmology 2017; study of Australian children. Invest Ophthalmol Vis Sci 2005;
124(10):1475–1484. 46(8):2748–2754.
17. Godefrooij DA, Gans R, Imhof SM, Wisse RP. Nationwide 34. Shufelt C, Fraser-Bell S, Ying-Lai M, et al. Refractive error,
reduction in the number of corneal transplantations for kera- ocular biometry, and lens opalescence in an adult population:
toconus following the implementation of cross-linking. Acta the Los Angeles Latino Eye Study. Invest Ophthalmol Vis Sci
Ophthalmol 2016;94(7):675–678. 2005;46(12):4450–4460.
18. Sandvik GF, Thorsrud A, Råen M, et al. Does corneal 35. Sawada A, Tomidokoro A, Araie M, et al. Refractive errors in
collagen cross-linking reduce the need for keratoplasties an elderly Japanese population: the Tajimi study. Ophthal-
in patients with keratoconus? Cornea 2015;34(9): mology 2008;115(2):363–370.e3.
991–995. 36. Warrier S, Wu HM, Newland HS, et al. Ocular biometry and
19. Salmon HA, Chalk D, Stein K, Frost NA. Cost effectiveness determinants of refractive error in rural Myanmar: the Meik-
of collagen crosslinking for progressive keratoconus in the UK tila Eye Study. Br J Ophthalmol 2008;92(12):1591–1594.
NHS. Eye (Lond) 2015;29(11):1504–1511. 37. Jonas JB, Nangia V, Sinha A, Gupta R. Corneal refractive po-
20. Godefrooij DA, Mangen MJ, Chan E, et al. Cost-effective- wer and its associations with ocular and general parameters:
ness analysis of corneal collagen crosslinking for progressive the Central India Eye and Medical Study. Ophthalmology
keratoconus. Ophthalmology 2017;124(10):1485–1495. 2011;118(9):1805–1811.
21. Smolek MK, Klyce SD. Current keratoconus detection 38. Lee KE, Klein BE, Klein R, et al. Association of age, stature,
methods compared with a neural network approach. Invest and education with ocular dimensions in an older white pop-
Ophthalmol Vis Sci 1997;38(11):2290–2299. ulation. Arch Ophthalmol 2009;127(1):88–93.

134 AMERICAN JOURNAL OF OPHTHALMOLOGY JULY 2020

Downloaded for Anonymous User (n/a) at Keio University - Shinanomachi Campus from ClinicalKey.jp by Elsevier on June 09, 2020.
For personal use only. No other uses without permission. Copyright ©2020. Elsevier Inc. All rights reserved.