Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
1040-5488/100/7703-0140/0 VOL. 77, NO. 3, PP. 140–149 OPTOMETRY AND VISION SCIENCE Copyright © 2000 American Academy of Optometry ORIGINAL ARTICLE Measurement of Refractive Error in Native American Preschoolers: Validity and Reproducibility of Autorefraction ERIN M. HARVEY, MA, JOSEPH M. MILLER, MD, VELMA DOBSON, PhD, ROBERT TYSZKO, OD, and AMY L. DAVIS, OD Departments of Ophthalmology (EMH, JMM, VD, RT), Optical Sciences (JMM), and Psychology (VD), The University of Arizona, Tucson, Arizona ABSTRACT: Purpose: To examine (1) reproducibility of cycloplegic retinoscopy (C-RNS), cycloplegic autorefraction (C-Autoref), and noncycloplegic autorefraction (NC-Autoref), and (2) validity of C-Autoref and NC-Autoref compared with C-RNS in preschoolers with astigmatism. Methods: Subjects were 36 Native American preschoolers. Three measurements of right eye refractive error were obtained with each of three methods: C-RNS (by three different retinoscopists), C-Autoref, and NC-Autoref (Nikon Retinomax Kⴙ). Vector methods (vector dioptric distance, VDD) were used in the analyses. Results: Mean reproducibility was 0.41 D (SD ⴝ 0.18) for C-RNS, 0.25 D (SD ⴝ 0.17) for C-Autoref, and 0.37 D (SD ⴝ 0.21) for NC-Autoref. Mean agreement between C-Autoref and C-RNS ranged from 0.51 to 0.61 VDD (SD ⴝ 0.24 to 0.35), and ranged from 1.66 to 1.74 VDD (SD ⴝ 1.11 to 1.25) for agreement between NC-Autoref and C-RNS. Mean bias was ⴚ0.07 ⴙ0.21 ⴛ 149 and ⴚ1.33 ⴙ0.34 ⴛ 178 for C-Autoref and NC-Autoref, respectively. Conclusions: C-Autoref provided reliable and valid measurements of refractive error in young children. NC-Autoref measurements were reliable within subjects, but there was large variability in validity among subjects. (Optom Vis Sci 2000;77:140–149) Key Words: retinoscopy, autorefraction, children, astigmatism, Native Americans T he traditional method for measuring refractive errors in preschool-age children involves cycloplegic or noncycloplegic retinoscopy. Although skilled retinoscopists can provide reliable and valid measures of refractive error in children, previous research indicates that retinoscopy is subject to measurement bias and intraobserver/interobserver variation.1–7 A means of measuring refractive error that is accurate, repeatable, and free of operator bias would be desirable in clinical, research, and screening settings. From a clinical perspective, such an instrument, along with a body of research on the instrument’s measurement validity and reproducibility, could be used to detect significant change in refractive error over time, independent of any between- and within-retinoscopist bias or variability. From a research perspective, the instrument could reduce the opportunity for experimenter bias and would provide a means for standardization and comparability of measures among studies and across research laboratories. In addition, a reliable automated instrument that can be used by lay screeners and requires little cooperation from the children could prove to be an effective means of identifying children with high refractive error in a screening setting. The instrument would be especially useful for screening if normative data on the relationship between cycloplegic and noncycloplegic measurements were available and could be used to identify noncycloplegic refraction values that would provide the greatest sensitivity and specificity for detection of significant refractive error. Previous research has indicated that autorefractors can provide reliable measures of refractive error that are comparable with or better than retinoscopy.6 –10 One candidate for such an instrument is the Nikon Retinomax K⫹ autorefractor (Nikon Inc., Melville, NY), a hand-held instrument that measures refractive error. Several studies have indicated that the Retinomax K⫹ has good reproducibility of measurements, agrees well with retinoscopy measurements, and can be used with children.8 –13 However, evaluations of reproducibility of Retinomax K⫹ representative measurements compared with reproducibility of retinoscopy measurements in young children have not been reported, and there have been no Optometry and Vision Science, Vol. 77, No. 3, March 2000 Validity and Reproducibility of Autorefraction—Harvey et al. evaluations of the validity of the Retinomax K⫹ in highly astigmatic children. The aims of this study were (1) to determine if the reproducibility of cycloplegic autorefraction (C-Autoref) and noncycloplegic autorefraction (NC-Autoref) measurements using the Retinomax K⫹ compares favorably with reproducibility of cycloplegic retinoscopy (C-RNS) measurements in children and (2) to determine the validity of the Retinomax K⫹ C-Autoref and NC-Autoref measurements of a child’s refractive error, using optimal (gold standard) retinoscopy measurements as a comparison. Subjects in this study were Native American preschool children from the Tohono O’Odham Nation in Southern Arizona, a tribe with a previously documented high prevalence of astigmatism.14 –17 Data were collected as part of a larger study of “Astigmatism and Amblyopia Among Native American Children” (AANAC), which is aimed at determining the most effective screening method for the detection of high astigmatism in preschoolers, and evaluating the effectiveness of preschool screening and glasses intervention on vision once the children reach grade school.18 METHODS Subjects. Subjects were 42 preschool children (mean age, 4.6 years; range, 3.2 to 5.6 years) who were enrolled in the Tohono O’Odham Early Childhood Head Start Program. All Head Start enrollees received a complete eye examination as part of the AANAC research study. None of the children included in the study had any ocular abnormalities other than refractive error. The children included in this study were those seen on two scheduled eye exam dates (3/25/98 and 3/30/98) on which three retinoscopists were available to conduct eye examinations. The University of Arizona Institutional Review Board (IRB) approved this study. Written informed consent was obtained from parents of all children. Procedure. Details of the complete testing procedure used in the AANAC protocol have been published.18 In what follows, we provide a description of only the testing procedures that are relevant to the present study. Before dilation, the Retinomax K⫹ was used to obtain three autorefraction measurements from the right eye of each subject. Autorefraction was performed by one of the study’s trained vision screeners. Each of the three Retinomax K⫹ measurements was the “representative measurement” that the instrument calculates, based on eight “measured values.” Measurements were obtained in the instrument’s “normal” mode, rather than “quick” mode. To achieve cycloplegia, each child was first given one drop of proparacaine 0.5%, followed approximately 30 s later by a single drop of cyclopentolate 2% in each eye instilled in the inferior cul de sac. Five minutes later, each child was given a single drop of cyclopentolate 1% in each eye. Drops were administered by holding the child, with the eyelid open and the pooled drop in contact with the cornea for approximately 1 second, and then lifting the upper eyelid, allowing the pooled drop to collect under the upper eyelid. At least 40 min after the first drop of cyclopentolate was instilled, a check for residual accommodation was performed by 141 observing the retinoscopic reflex while the child was viewing a distant television monitor displaying a cartoon, and then while the child changed fixation to view a finger puppet held alongside the retinoscope. Evidence of inadequate cycloplegia was a shift in the retinoscopic reflex as viewing distance changed. If the results of this test of residual accommodation indicated that a child’s eyes had not been adequately cyclopleged, the child was asked to wait an additional 5 min. If the child’s eyes were not adequately cyclopleged 45 min after the first drop of cyclopentolate, the data from the child were eliminated from analyses. After the test for residual accommodation, each child underwent C-RNS (right eye and left eye), conducted by a pediatric ophthalmologist (JMM). Three autorefraction measurements were then obtained for the right eye of each child using the Retinomax K⫹. The autorefractions were obtained by one of the study’s trained vision screeners. Finally, two optometrists (RT and ALD) each conducted C-RNS on the right eye of each child. All three clinicians had extensive experience conducting retinoscopy on young children. None of the three retinoscopists were aware of the results of the other retinoscopists or the autorefraction measurements at the time of their examination. Data Analysis. Analysis of variability of refractive error data was conducted using refractive data (sphere, cylinder, and axis) transformed according to a method described by Long19 and modified by Harris.20 A summary of this method and calculations used in the present study is provided in the Appendix. Reproducibility of Measurements. For each subject, each of the three C-RNS measurements was converted into a value representing a point in dioptric three-dimensional space. The mean of the three values was calculated, and the vector dioptric distance (VDD) between each retinoscopy measurement and the mean of that subject’s retinoscopy measurements was determined. This yielded a dioptric error value for each retinoscopy estimate of refractive error. The mean of the dioptric errors was calculated for each subject and the grand mean dioptric error was calculated across subjects. This provided a measure of the average acrossretinoscopist VDD for three retinoscopy measurements in an individual child. The same method was used to calculate the mean VDD for the three C-Autoref measurements, and the three NC-Autoref measurements. A repeated-measures analysis of variance was used to compare mean VDDs across the three types of measurements (C-RNS, C-Autoref, and NC-Autoref). Accuracy of Autorefraction Measurements. For each subject, validity of autorefraction measurements was evaluated in comparison with that subject’s “optimal” C-RNS measurement. The optimal C-RNS measurement for each subject was the C-RNS measurement that was closest to the vector dioptric mean of the subject’s three C-RNS measurements (one from each clinician; i.e., the median of the three measurements). For each subject’s three C-Autoref measurements, the VDD from the optimal C-RNS measurement was calculated. This allowed us to determine if measurement validity tended to increase or decrease with repeated measures. The same analysis was conducted with each subject’s three NC-Autoref measurements. Optometry and Vision Science, Vol. 77, No. 3, March 2000 142 Validity and Reproducibility of Autorefraction—Harvey et al. Autorefraction Measurement Bias. VDDs are always represented in positive numbers. Therefore, they do not provide any information regarding measurement bias (e.g., consistent underor over-estimation of cylinder components). To evaluate any measurement bias present in autorefraction measurements, additional comparisons between autorefraction (C-Autoref, NC-Autoref) and optimal C-RNS were conducted. Using the vector method of Harris (see Appendix), the mean optimal C-RNS measurement across subjects was subtracted from the mean of the first C-Autoref measurements (C-Autoref measure 1 for each subject) across subjects. Results were then converted back to standard notation, for evaluation of measurement bias of the sphere and cylinder components. This procedure was repeated for each subject’s first NCAutoref measurement. RESULTS Of the 42 subjects eligible for the study, 36 completed all of the testing and are included in statistical analyses (mean age of final sample, 4.7 years; range, 3.6 to 5.6 years). Of the six subjects excluded, the eyes of one were judged to be not fully cyclopleged after the 45-min waiting time, and the other five did not complete the full testing procedure [one child was ill, one was too uncooperative (the youngest child seen), and three did not complete because of experimenter procedural errors (e.g., missing data)]. However, it should be noted that at least one autorefraction measurement and one retinoscopy measurement were obtained for each eye in 98% of subjects (41 of 42). A summary of each subject’s right eye refractive error (optimal C-RNS) is provided in Table 1. Reproducibility of Measurements. The mean VDD (average deviation from the mean of three measurements) for C-RNS was 0.41 D (SD ⫽ 0.18). The mean VDD for C-Autoref was 0.25 D (SD ⫽ 0.17), and the mean VDD for NC-Autoref was 0.37 D (SD ⫽ 0.21). A repeated-measures ANOVA yielded a significant effect of method on measurement reproducibility (mean VDD) [F2, 70 ⫽ 7.90, p ⫽ 0.001]. Post hoc comparisons indicated that the reproducibility of C-Autoref measurements was significantly better than the reproducibility for NC-Autoref and C-RNS measurements. That is, there was less variability in C-Autoref measurements than in NC-Autoref and C-RNS measurements. Results of reproducibility of measurements in standard notation (rather than vector notation) for spherical equivalent measurements are presented in Fig. 1. All mean differences are 0, because the reference was the mean of the measurements. The standard deviation, calculated across 108 measurements (three measurements per subject for 36 subjects, calculated separately for each refraction method) are 0.28 D, 0.18 D, and 0.31 D for C-RNS, C-Autoref, and NC-Autoref, respectively. Validity of Autorefraction Measurements. For each subject, the optimal C-RNS measurement (median of the three C-RNS measurements) was determined. The optimal C-RNS was the measurement from retinoscopist 1 for seven subjects (19.4%), retinoscopist 2 for 12 subjects (33.3%), and retinoscopist 3 for 17 subjects (47.2%). Statistical analysis indicated that the number of retinoscopies from each retinoscopist that contributed to the optimal did not significantly differ [22 ⫽ 4.17]. TABLE 1. Right eye refractive error data (optimal C-RNS) for each subject (plus cylinder notation). Subject ID Sphere Cylinder 440 442 443 444 445 446 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 471 472 473 474 475 476 477 481 Mean Std. Deviation Minimum Maximum 1.50 .25 1.00 1.50 .50 1.25 .00 .75 .50 .25 1.00 ⫺.25 .75 1.25 1.00 .50 ⫺3.75 1.00 .25 .75 1.00 1.00 .50 .50 .75 1.00 .75 .25 ⫺1.00 .00 .75 ⫺.25 .50 .00 .25 .25 .45 .89 ⫺3.75 1.50 1.25 .50 .25 1.00 .75 .00 1.25 2.00 .25 .75 .50 2.50 2.25 1.00 1.25 .00 4.00 .75 .25 .50 .25 3.00 .50 .75 3.75 .75 .75 .25 2.25 .25 .50 .75 .75 .75 .25 .50 1.03 1.00 .00 4.00 Axis 90 90 90 90 90 90 95 90 90 90 85 95 90 92 93 90 90 90 90 100 95 90 80 85 92 90 90 90 95 90 90 90 90 85 90.35 3.43 80.00 100.00 The mean VDDs of autorefraction measurements from the optimal C-RNS measurements were 0.57 D (SD ⫽ 0.28), 0.51 D (SD ⫽ 0.24), and 0.61 D(SD ⫽ 0.35) for the three (first, second, and third) C-Autoref measurements, respectively, and 1.71 D (SD ⫽ 1.12), 1.66 D (SD ⫽ 1.11), and 1.74 D (SD ⫽ 1.25) for the three NC-Autoref measurements, respectively. Results of validity for standard notation (rather than vector notation) spherical equivalent and cylinder measurements of each subject’s first C-Autoref measurement and each subject’s first NCAutoref measurement are presented in Fig. 2. Autorefraction is compared with optimal C-RNS. The mean difference in spherical equivalent values (autorefraction ⫺ optimal C-RNS) was 0.02 D Optometry and Vision Science, Vol. 77, No. 3, March 2000 Validity and Reproducibility of Autorefraction—Harvey et al. 143 FIGURE 1. Subjects’ deviations (three measurements per subject) from mean spherical equivalent measurements for (a) C-RNS, (b) C-Autoref, and (c) NC-Autoref. Solid reference lines indicate means; dashed lines indicate 95% confidence limits (⫾ 2 SD). FIGURE 2. Subject’s deviations from optimal retinoscopy values for C-Autoref (a and c) and for NC-Autoref (b and d). Spherical equivalent deviations are shown in a and b, and refractive cylinder deviations are shown in c and d. Solid reference lines indicate means; dashed lines indicate 95% confidence limits (⫾ 2 SD). (SD ⫽ 0.37) for C-Autoref and ⫺1.15 D (SD ⫽ 0.82) for NCAutoref. The mean difference in refractive cylinder values was ⫺0.02 D (SD ⫽ 0.37) for C-Autoref and ⫺0.21 D (SD ⫽ 0.28) for NC-Autoref. Autorefraction Measurement Bias. The mean bias for CAutoref (first C-Autoref ⫺ optimal C-RNS) was ⫺0.07 ⫹0.21 ⫻ 149. The mean bias for NC-Autoref (first NC-Autoref ⫺ optimal C-RNS) was ⫺1.33 ⫹ 0.34 ⫻ 178. DISCUSSION Autorefraction Success Rates. The results indicate that the Nikon Retinomax K⫹ can be used successfully to measure refractive error in this population of preschoolers in which there is a high prevalence of astigmatism. We were able to obtain autorefraction measurements in 98% of children eligible for this study. The results are in agreement with previous reports indicating high success Optometry and Vision Science, Vol. 77, No. 3, March 2000 144 Validity and Reproducibility of Autorefraction—Harvey et al. rates in obtaining Retinomax measurements in children.8–13, 21 For example, Cordonnier and Dramaix used the Retinomax in a screening of 897 children between the ages of 6 months and 5 years and were able to obtain a measurement on 98.5% of the children.13 Reproducibility of Measurements. The analyses of measurement reproducibility indicated that variability of NC-Autoref measurements was similar to variability of C-RNS measurements across retinoscopists. C-Autoref measurements were less variable than repeated C-RNS and NC-Autoref measurements. Reproducibility for all three methods was excellent, however, with an average deviation from the mean that was 0.41 VDD or less for each method. Table 2 summarizes the reproducibility of measurements in the present study in comparison to previous studies of reproducibility of retinoscopy and autorefraction in children. Mean difference values for sphere, spherical equivalent, and cylinder components provide an indication of measurement bias, and standard deviations and confidence intervals provide an indication of measurement variability. Studies reporting reproducibility of retinoscopy measurements (Hirsch,1 Saunders and Westall,5 and present study) indicate bias of up to 0.55 D in measurements of spherical equivalent and cylinder, with confidence intervals for the difference scores that ranged from ⫾ 0.39 D to ⫾ 1.14 D. Studies of the reproducibility of Retinomax autorefraction measurements (Harvey et al.,8 Cordonnier and Dramaix,13 and present study) indicate little bias in measurements of sphere, spherical equivalent, or cylinder either with or without cycloplegia. However, there is less variability of Retinomax autorefraction measurements with cycloplegia than without cycloplegia. In addition, variability of Retino- TABLE 2. Comparison of studies evaluating the reproducibility of refraction in children. Reproducibility (Mean difference between repeated measurements) Study Subjects Refraction Method Saunders & Westall5 10 infants (⬍2 years old) C-RNSa — Present Study 36 Native American preschoolers (3.5 to 5 years old) C-RNSc Hirsh1 72 eyes, 36 children (6 to 14-year-olds) Spherical Equivalent (SD, 95% CI) Cylinder (SD, 95% CI) VDD (SD, 95% CI) ⬵⫺0.30 D (⬵0.20, ⫾0.39) — — ⫺0.06 D (0.48, ⫾0.94) 0.05 D (0.58, ⫾1.14) 0.11 D (0.44, ⫾0.86) 0.21 D (0.43, ⫾0.84) 0.25 D (0.52, ⫾1.02) 0.04 D (0.40, ⫾0.78) 0.55 D (0.34, ⫾0.67) 0.41 D (0.32, ⫾0.63) ⫺0.14 D (0.35, ⫾0.69) 0.41 VD (0.18, ⫾0.35) Non-C-RNS — 0.28 Db (0.34, ⫾0.67) — — 10 infants (⬍2 years old) Near Retinoscopya — ⬵⫺0.48 D (⬵0.20, ⫾0.39) — — Harvey et al.8 22 children (mean age, 6.6 years) 23 children (mean age, 4.1 years) Cycloplegic Retinomaxd (measured values) ⫺0.10 D (0.20, ⫾0.39) ⫺0.04 D (0.27, ⫾0.53) ⫺0.08 D (0.16, ⫾0.31) ⫺0.03 D (0.21, ⫾0.41) 0.03 D (0.25, ⫾0.49) 0.03 D (0.25, ⫾0.49) 0.37 VD (0.52, ⫾1.02) 0.49 VD (0.40, ⫾0.78) Present Study 36 Native American preschoolers (3.5 to 5 years old) Cycloplegic Retinomax K⫹d ⫺0.04 D (0.32, ⫾0.63) ⫺0.03 D (0.22, ⫾0.43) 0.03 D (0.31, ⫾0.61) 0.25 VD (0.17, ⫾0.33) Cordonnier & Dramaix13 255 (9 to 36 months old) Non-cycloplegic Retinomax 0.09 D (1.40, ⫾2.74) — — — Present Study 36 Native American preschoolers (3.5 to 5 years old) Non-cycloplegic Retinomax K⫹d 0.00 D (0.51, ⫾1.00) ⫺0.05 D (0.50, ⫾0.98) ⫺0.10 D (0.25, ⫾0.49) 0.37 VD (0.21, ⫾0.41) Saunders & Westall5 Sphere (SD, 95% CI) SD, standard deviation; 95% CI, 95% confidence interval (1.96 ⫻ SD). Two different clinicians, values estimated from graph. b Mean of the absolute values of the differences. c Three different clinicians (three rows represent pairwise comparisons of retinoscopists). d First two measurements used for mean differences in sphere, spherical equivalent, and cylinder. The Retinomax provides up to 8 measured values and determines one representative measurement (best estimate of refractive error) based on the measured values. In the present study, we compared reproducibility and accuracy based on Retinomax representative measurements. In the study by Harvey et al., we compared reproducibility and accuracy based on Retinomax measured values. a Optometry and Vision Science, Vol. 77, No. 3, March 2000 Validity and Reproducibility of Autorefraction—Harvey et al. max autorefraction measurements with cycloplegia is equal to or less than that of both cycloplegic and noncycloplegic retinoscopy. In a recent literature review, Goss and Grosvenor22 summarized data on reliability of refraction in adults tested with a variety of techniques, including subjective refraction, retinoscopy, and autorefraction. They concluded that the 95% confidence interval for repeated clinical refractions was ⫾ 0.50 D. This value represents a standard with which to compare the data presented in Table 2 (i.e., studies of reliability of refraction in children). It is noteworthy that despite the differences in subject characteristics and refraction techniques, the 95% confidence intervals shown in Table 2 for children tested with cycloplegic Retinomax autorefraction are similar to the overall value reported by Goss and Grosvenor22 for adult refractions. In summary, the results of reproducibility of C-Autoref are comparable with findings reported in studies of clinical refraction in adults22 and are comparable with or better than reports of reproducibility of retinoscopy in children.1, 5 Validity and Bias of Cycloplegic Autorefraction Measurements. The average deviation of each of the three C-Autoref measurements from the optimal C-RNS measurement was within 0.61 VDD. Agreement between retinoscopy and C-Autoref did not seem to vary with repeated measurements, i.e., across first, second, and third autorefractions. The mean bias for the C-Autoref 145 was ⫺0.07 ⫹0.21 ⫻ 149, indicating little bias in measurements of sphere or cylinder, compared with C-RNS. In Table 3, we provide a summary of several recent studies that have examined the validity of the Retinomax in measuring cycloplegic refractive error in children. We reanalyzed our data in traditional notation (i.e., separate comparisons for sphere, cylinder, and axis components) for direct comparisons of measurement bias and variability with other studies. The results presented in Table 3 indicate little bias (i.e., mean differences close to 0) between the autorefraction and subjective refraction, or between autorefraction and retinoscopy measurements of sphere, spherical equivalent, cylinder, and axis, with the exception of the results of Weseman and Dick who reported a mean difference of 0.59 D spherical equivalent.9 Confidence intervals in the various studies indicated that 95% of Retinomax measurements of spherical equivalent were within ⫾ 0.72 D to ⫾ 1.39 D of retinoscopy, but greater variability (⫾1.65 D) was reported in a study comparing the Retinomax with subjective refinement.8 For measurements of cylinder, 95% of Retinomax measurements were within ⫾ 0.20 D to ⫾ 0.94 D of either retinoscopy or subjective refraction measurements. Kallay et al.12 reported the least variability in measurements of sphere, cylinder, and axis in their comparison between refractions obtained with the Retino- TABLE 3. Comparison of studies evaluating the accuracy of the Retinomax for use with children. All measurements were made with cycloplegia. Accuracy (Mean difference between methods) Spherical Equivalent (SD, 95% CI) Cylinder (SD, 95% CI) 0.59 D (0.44, ⫾0.86) 0.06 D (0.47 ⫾0.92) Study Subjects Comparison Sphere (SD, 95% CI) Wesemann and Dick9 25 2- to 10-yearolds (50 eyes) RetinomaxRetinoscopy — Harvey et al.8 22 children w/no occular pathology (mean age, 6.6 years) RetinomaxSubjective Refraction 0.19 D (0.90, ⫾1.76) 0.07 D ⫺0.25 D 4.18° (0.84, ⫾1.65) (0.31, ⫾0.61) (7.86, ⫾15.41) 1.03 VD (0.59, ⫾1.16) 23 children w/no ocular pathology (mean age, 4.1 years) RetinomaxRetinoscopy 0.27 D (0.75, ⫾1.47) 0.26 D (0.66, ⫾1.29) ⫺0.02 D 1.87° (0.48 ⫾0.94) (10.78, ⫾21.13) 0.82 VD (0.58, ⫾1.14) 102 5- to 72month-olds RetinomaxRetinoscopy 0.03 D (0.79, ⫾1.55) 0.09 D 0.23 D (0.71, ⫾1.39) (0.13, ⫾0.25) 132 children & young adults w/no ocular pathology (mean age, 10.9 years) RetinomaxTopcon (RMA-6000) 0.07 D (0.09, ⫾0.18) 36 Native American Preschoolers (3- to 5-year-olds) Retinomax K⫹Retinoscopyb 0.03 D (0.46, ⫾0.90) El-Defrawy et al.10 Kallay et al.12 Present Study — 0.02 D (0.10, ⫾0.20) Axis (SD, 95% CI) VDD (SD, 95% CI) — — — 0.97 VD (0.76, ⫾1.49) 6.49°a (4.71, ⫾9.23) — 0.02 D ⫺0.02 D 6.97° (0.37, ⫾0.72) (0.38, ⫾0.74) (13.87, ⫾27.18) SD, standard deviation; 95% CI, confidence interval (1.96 ⫻ SD). Only measurements with cylinder power ⱖ0.50 D included in cylinder and axis data. b Median (“optimal”) retinoscopy measurement and first cycloplegic autorefraction used in calculations. a Optometry and Vision Science, Vol. 77, No. 3, March 2000 0.57 VD (0.28, ⫾0.55) 146 Validity and Reproducibility of Autorefraction—Harvey et al. max and refractions obtained with the Topcon RM-A-6000 autorefractor. In the present study, there was slightly less variability in measurement validity (i.e., smaller standard deviations) of sphere and VDDs compared with previous studies (Table 3). It is possible that we found less variability in the present study because we compared Retinomax K⫹ measurements with the median of three retinoscopy measurements, thus eliminating any outlier retinoscopy measurements that may have been obtained, for example, in uncooperative subjects. In contrast, other studies compared Retinomax measurements with individual retinoscopy measurements. The differences in variability and measurement agreement between studies may also be attributable to age differences in the samples, and to other differences in the populations represented in each study (e.g., the high prevalence of astigmatism in our study population).14–17 Accuracy and Bias of Noncycloplegic Autorefraction: Implications for Vision Screening. The results indicate that there was good reproducibility of NC-Autoref measurements (mean VDD ⫽ 0.37 D (SD ⫽ 0.21). However, mean differences between NC-Autoref measurements and optimal C-RNS measurements ranged from 1.66 D to 1.74 D, with relatively large standard deviations (1.11 to 1.25). The large standard deviations probably reflect variations in relaxation of accommodation among subjects. The large VDDs probably indicate that accommodation is not sufficiently relaxed during NC-Autoref measurements, resulting in underestimation of hyperopia. Although VDDs do not provide an indication of the direction of measurement bias, the mean bias between NC-Autoref and optimal retinoscopy measurements support this interpretation (mean bias ⫽ ⫺1.33 ⫹0.34 ⫻ 178). This interpretation is also supported by the data in Fig. 1, which indicate a negative shift in NC-Autoref measurements of spherical equivalent (Fig. 1C), compared to C-Autoref measurements (Fig. 1B) in the same subjects. In summary, the results indicate that although NC-Autoref provides reliable measurements within subjects (i.e., good reproducibility), results of comparisons with optimal C-RNS measurements indicate that it underestimates hyperopia (overestimates myopia) and that measurement validity varies considerably among subjects. El-Defrawy and his colleagues also obtained NC-Autoref data on their subjects.10 Statistical comparisons between NC-Autoref and retinoscopy were not reported, but the authors did indicate that that NC-Autoref results were “grossly inaccurate and overestimated myopia by up to ⫺8.00 D.” In an earlier study of 77 Native American preschoolers, we reported a mean bias of ⫺1.23 D spherical equivalent, and ⫺0.08 D cylinder for NC-Autoref measurements compared with C-Autoref measurements.11 In the present study, we also found that NC-Autoref underestimated hyperopia (overestimated myopia). However, we may not have found as great an overestimation of myopia as did El-Defrawy and his colleagues (maximum difference between optimal C-RNS and NC-Autoref ⫽ 5.66 VDD in the present study), possibly because of greater cooperation in our older subjects (36 to 60 months in the present study vs. 5 to 72 months in the study by El-Defrawy et al.).10 The mean bias for NC-Autoref (⫺1.33 ⫹0.34 ⫻ 178) in the present study indicates that there was relatively little bias in measurements of cylinder. Previous studies evaluating agreement be- tween cycloplegic and noncycloplegic refractions have also found good agreement in measures of cylinder power.11, 23, 24 These data suggest that noncycloplegic Retinomax measurements might serve as a useful screening tool for detection of high astigmatism. We are currently evaluating the sensitivity and specificity of the Retinomax K⫹ in detecting astigmatism in a large sample of preschoolers with a high prevalence of astigmatism as part of the “Astigmatism and Amblyopia Among Native American Children” research study. In a preliminary analysis of data from 245 Native American preschoolers, we found that NC-Autoref measurements of refractive astigmatism had sensitivity and specificity of 91% and 86%, respectively, for detection of significant refractive astigmatism (2.00 D or more for 3-year-olds and 1.50 D for 4-year-olds, based on C-Autoref measurements).21 The Retinomax may also serve as a useful screening tool for detection of myopia and hyperopia. Additional research on large numbers of subjects with a wide range of spherical refractive errors would help determine if there is an appropriate referral value, correction constant, or regression formula that can be applied to noncycloplegic Retinomax measurements to achieve high sensitivity and specificity for detection of significant myopia or hyperopia. For example, Chan and Edwards compared the results of cycloplegic and noncycloplegic measurements of Chinese preschool children and reported that the cycloplegic measurement can be estimated by multiplying the spherical component of the noncycloplegic measurement by 1.45, and adding ⫹0.39 to the result.23 Cordonnier and Dramaix13 evaluated the Retinomax for use in screening for high hyperopia in children from 6 months to 5 years of age. They found that a threshold of ⫹1.50 D spherical refractive error in the most hyperopic meridian as measured by the Retinomax without cycloplegia provided a sensitivity of 70.2% and a specificity of 94.6% for the detection of abnormal hyperopia (defined as ⫹3.50 D by cycloplegic refraction).13 Determination of an accurate equation for estimation of refractive error from noncycloplegic measurements might make the Retinomax K⫹ extremely valuable for vision screening in young children. However, the large between-subjects variability in agreement between NCAutoref and optimal C-RNS reported in the present study make this prospect somewhat questionable. CONCLUSIONS The results of the present study indicate that cycloplegic measurements made with the Nikon Retinomax K⫹ autorefractor provide highly reliable and accurate measurements of refractive error in young children. Repeatability of the instrument’s measurements is comparable with or superior to retinoscopy in children, indicating that the Retinomax K⫹ might be useful in clinical settings, to reliably document change over time, and in research settings, to reduce measurement bias and provide standardization of measurements. The repeatability and validity of Retinomax K⫹ measurements obtained without cycloplegia indicate that although the instrument’s measurements are reliable within subjects, there is large variability in validity between subjects. Future research might be directed toward further examining this finding and evaluating the Retinomax K⫹ for use in screening of young children. Optometry and Vision Science, Vol. 77, No. 3, March 2000 Validity and Reproducibility of Autorefraction—Harvey et al. ACKNOWLEDGMENT We thank Don Everett, M.A., Jonathan Holmes, M.D., Rosemary Lopez, Maureen McGuire, Ph.D., and Karla Zadnik, O.D., Ph.D., for their support and suggestions regarding the conduct of this study. We also thank Frances Lopez, Christie Lopez, Jenniffer Funk-Weyant, Pat Broyles, and Angel Palanca-Capistrano, M.D., for their help in conducting the vision screenings and eye examinations, the parents and children who participated in the study, and the Tohono O’Odham Nation for their support of this project. Received July 12, 1999; revision received December 9, 1999. This study was supported by National Institutes of Health/National Eye Institute Grant EY11155 to JMM. APPENDIX: CALCULATION OF VECTOR DIOPTRIC DISTANCE The statistical analysis of clinical refraction data has traditionally been hampered by the fact that although sphere powers can readily be combined and subtracted from each other, the same is not true when cylinder power is present. This is because unless cylinder axes coincide with each other, simple addition or subtraction cannot combine the powers associated with each axis. In the last decade, two very clinically relevant techniques have been developed that break a spherocylindrical refracting surface into independent components that form an orthogonal set. The first method, as implemented by Harris,20 decomposes the spherocylindrical surface into two perpendicular cylindrical surfaces, and a Jackson-Cross Cylinder whose axes straddle the first two cylinders. The second method, as implemented by Thibos,25 describes the refracting surfaces as a spherical lens (whose power is equal to the spherical equivalent of the spherocylindrical lens) that is combined with two Jackson-Cross cylinder lenses, one with axis 0°, and the other with axis 45°. Either method (Harris or Thibos) greatly facilitates the comparison of refractive data, and each might be preferred for certain calculations. Each method results in the decomposition of a spherocylindrical lens into three independent (orthogonal) component parts that can be analyzed separately. In the examples that follow, sp is sphere power in clinical notation, cp is cylinder power in plus-cylinder clinical notation, and ax is axis in clinical notation in degrees. The Harris Method: Conversion from Clinical to Vector Notation 147 the oblique cross cylinder; and vcp is vertical cylinder power associated with the cylinder whose axis is at 90°. The formulas for converting sp, cp, and ax into vcp, ccp, and hcp follow (Equations 1 to 3). Several examples of sp, cp, ax, and vcp, ccp, and hcp are provided in Table 4. vcp ⫽ sp ⫹ cp ⫻ sin(ax) ⫻ sin(ax) (1) ccp ⫽ ⫺cp ⫻ sin(ax) ⫻ cos(ax) (2) hcp ⫽ sp ⫹ cp ⫻ cos(ax) ⫻ cos(ax) (3) The Harris Method: Conversion from Vector to Clinical Notation Average values of refraction, differences between refractions, and summation of refracting surfaces can be calculated by performing appropriate operations on the orthogonal components of the vcp, ccp, and hcp of the spherocylinder lens. However, clinical relevance demands that the resultant values of vcp, ccp, and hcp be converted back into clinical notation. This conversion is accomplished by the operations given in Equations 4 – 6. cp ⫽ ⫺sqrt[(vcp ⫹ hcp) ⫻ (vcp ⫹ hcp) ⫺ 4.0 ⫻ ((vcp ⫻ hcp) ⫺ (ccp ⫻ ccp))] (4) sp ⫽ (vcp ⫹ hcp ⫺ cp)/2.0 (5) ax ⫽ arctan[(sp ⫺ vcp)/cp] (6) The Thibos Method: Conversion from Clinical to Vector Notation Convert the spherocylindrical lens into a mean lens having power equal to the spherical equivalent of the spherocylinder lens, and two Jackson-Cross lenses that are oriented at 0° and 45°. In the equations that follow, M is the mean (spherical equivalent) power, J0 is the Jackson-cross cylinder power (axis 180°), and J45 is the Jackson-cross cylinder power (axis 45°). The formulas for converting sp, cp, and ax into M, J0, and J45 are given by Equations 7 to 9. Several examples of M, J0, and J45 are provided in Table 5. Convert the sphere (sp), cylinder (cp), and axis (ax) into an equivalent pair of cylinders oriented at 90 and 180°, and an obliquely oriented Jackson-Cross cylinder. In the equations that follow, hcp is horizontal cylinder power associated with the cylinder whose axis is at 180°, ccp is the cross cylinder power associated with M ⫽ sp ⫹ (cp/2) (7) J0 ⫽ ⫺(cp/2) ⫻ cos(2 ⫻ ax) (8) J45 ⫽ ⫺(cp/2) ⫻ sin(2 ⫻ ax) (9) TABLE 4. Harris method clinical to vector conversion values. sp cp ax vcp ccp hcp Comment ⫹2.00 Plano Plano ⫺1.00 0 ⫹1.00 ⫹1.00 ⫹2.00 0 090 180 135 2.00 1.00 0 0 0 0 0 1.00 2.00 0 1.00 0 Sphere Vertical cylinder Horizontal cylinder Oblique Jackson Cross Optometry and Vision Science, Vol. 77, No. 3, March 2000 148 Validity and Reproducibility of Autorefraction—Harvey et al. TABLE 5. Thibos method clinical to vector conversion values. sp cp ax M J0 J45 Comment ⫹2.00 Plano Plano ⫺1.00 0 ⫹1.00 ⫹1.00 ⫹2.00 0 090 180 135 2.00 0.50 0.50 0 0 0.50 ⫺0.50 0 0 0 0 ⫺1.00 Sphere Vertical cylinder Horizontal cylinder Oblique Jackson Cross The Thibos Method: Conversion from Vector to Clinical Notation The conversion of derived values of M, J0, and J45 back into plus-cylinder clinical notation are accomplished by the application of Equations 10 to 12. sp ⫽ M ⫺ sqrt(J0 ⫻ J0 ⫹ J45 ⫻ J45) (10) cp ⫽ 2 ⫻ sqrt(J0 ⫻ J0 ⫹ J45 ⫻ J45) (11) ax ⫽ 0.5 ⫻ arctan(J45/J0) ⫹ 90 (12) The Vector Dioptric Distance (VDD) The representation of a spherocylindrical lens as a point in three-dimensional space can be seen as a vector representation of power.25 Thus, the dioptric difference between two spherocylindrical lenses can be represented as a vector between two points in space and the length of that vector represents the magnitude of blur. The magnitude of the difference vector can be calculated using either the method of Harris20, 26 or the method of Thibos,25 and the two methods are equivalent with the introduction of an appropriate scaling factor. After a measurement in clinical notation is converted into vector notation, as described above, the resulting three values (vcp, ccp, and hcp in Harris notation; M, J0, and J45 in Thibos notation) allow component parts of different refractions to “line up” with each other and be added, subtracted, and averaged in the same way that sphere powers can be compared in the absence of cylinder. As the units of the three values (vcp, ccp, and hcp or M, J0, and J45) are diopters, and as they are “orthogonal” (meaning mutually perpendicular to each other), a given clinical measurement can be represented as a point in three-dimensional space (with the three values forming axes in the same manner as x, y, and z). In this way, the “distance” between any two points in the refraction space can be calculated, and the distance between the two points will have units of diopters. We have used the method described by Harris to allow a single number (representing the dioptric distance between two vectors) to be calculated when two refractions are compared.26 This simultaneous comparison of two pairs of sphere, cylinder, and axis is particularly helpful when measurements of repeatability are desired or when deviations from a mean value are needed to identify an outlier. To calculate the Vector Dioptric Distance, or VDD, two clinical measurements (Refractions 1 and 2, or R1 and R2) are converted to (vcp1, ccp1, hcp1) and (vcp2, ccp2, hcp2). The differences between the vcp, ccp, and hcp values for the two refractions are calculated, squared, summed, and the square root taken, after a 3-dimensional application of the Pythagorean theorem. The cross cylinder power is included twice in this calculation because it contains two refracting surfaces contributing dioptric power. The formula for VDD is given in Equation 13. VDD ⫽ sqrt[(vcp1 ⫺ vcp2)2 ⫹ 2 ⫻ (ccp1 ⫺ ccp2)2 ⫹ (hcp1 ⫺ hcp2)2] (13) If the method of Thibos is used, then the equation is slightly modified to account for a difference in scaling.25 In the Harris method, the unit vector is 1 D of cylinder power, whereas in the Thibos method, the unit vector is 1 D of sphere power. Because our use of VDD26 predates Thibos’ publication,25 we provide a formula that scales the Thibos unit vector to match the Harris unit vector. Again, refractions R1 and R2 are converted to [M1, J01, J451] and [M2, J02, J452]. The Vector Dioptric Distance is then calculated as follows: VDD ⫽ sqrt(2) ⫻ sqrt[(M1 ⫺ M2)2 ⫹ (J01 ⫺ J02)2 ⫹ (J451 ⫺ J452)2] (14) REFERENCES 1. Hirsch MJ. The variability of retinoscopic measurements when applied to large groups of children under visual screening conditions. Am J Optom Arch Am Acad Optom 1956;33:410–6. 2. Hyams L, Safir A, Philpot J. Studies in refraction. II. Bias and accuracy of retinoscopy. Arch Ophthalmol 1971;85:33–41. 3. Safir A, Hyams L, Philpot J, Jagerman LS. Studies in refraction. I. The precision of retinoscopy. Arch Ophthalmol 1970;84:49–61. 4. Hopkisson B, Arnold P, Billingham B, McGarrigle M, Shribman S. Can retinoscopy be used to screen infants for amblyopia? A longitudinal study of refraction in the first year of life. Eye 1992;6:607–9. 5. Saunders KJ, Westall CA. Comparison between near retinoscopy and cycloplegic retinoscopy in the refraction of infants and children. Optom Vis Sci 1992;69:615–22. 6. Zadnik K, Mutti DO, Adams AJ. The repeatability of measurement of the ocular components. Invest Ophthalmol Vis Sci 1992;33: 2325–33. 7. Wood MG, Mazow ML, Prager TC. Accuracy of the Nidek ARK900 objective refractor in comparison with retinoscopy in children ages 3 to 18 years. Am J Ophthalmol 1998;126:100–8. 8. Harvey EM, Miller JM, Wagner LK, Dobson V. Reproducibility and accuracy of measurements with a hand held autorefractor in children. Br J Ophthalmol 1997;81:941–8. Optometry and Vision Science, Vol. 77, No. 3, March 2000 Validity and Reproducibility of Autorefraction—Harvey et al. 9. Wesemann W, Dick B. Experiences with the hand-held autorefractometer “Retinomax” in adults and children. Klin Monatsbl Augenheilkd 1997;211:387–94. 10. El-Defrawy S, Clarke WN, Belec F, Pham B. Evaluation of a handheld autorefractor in children younger than 6. J Pediatr Ophthalmol Strabismus 1998;35:107–9. 11. Harvey EM, Miller JM, Dobson V. Effectiveness of screening for refractive errors with a hand-held autorefractor in a preschool population with a high prevalence of astigmatism. Invest Ophthalmol Vis Sci 1997;38:S977. 12. Kallay OP, Cordonnier MJ, Dramaix MM. Cycloplegic refractive errors in children: comparison of a standard and a hand-held refractor. Strabismus 1998;6:3–7. 13. Cordonnier M, Dramaix M. Screening for abnormal levels of hyperopia in children: a non-cycloplegic method with a hand held refractor. Br J Ophthalmol 1998;82:1260–4. 14. Kershner RM, Brick DC. Prevalence of high corneal astigmatism in Papago school children. Invest Ophthalmol Vis Sci 1984;25(suppl): 217. 15. Tyszko RM, Dobson V, Miller JM, Harvey EM. Characteristics of astigmatism prevalent in a preschool populaton of Native American children. Invest Ophthalmol Vis Sci 1998;39:S281. 16. Dobson V, Miller JM, Harvey EM. Corneal and refractive astigmatism in a sample of 3- to 5-year-olds with a high prevalence of astigmatism. Optom Vis Sci 1999;76:855–60. 17. Dobson V, Miller JM, Harvey EM, Sherrill DL. Prevalence of astigmatism, astigmatic anisometropia, and glasses wearing among preschool and school-age Native American children. In: Vision Science and Its Applications, OSA Technical Digest. Washington, DC: Optical Society of America, 1999:177–80. 18. Miller JM, Dobson V, Harvey EM, Sherrill DL. Astigmatism and 19. 20. 21. 22. 23. 24. 25. 26. 149 amblyopia among Native American children (AANAC): design and methods. Ophthal Epidemiol, in press. Long WF. A matrix formalism for decentration problems. Am J Optom Physiol Opt 1976;53:27–33. Harris WF. The mean and variance of samples of dioptric power: the basic calculations. Clin Exp Optom 1990;73:89–92. Miller JM, Harvey EM, Dobson V. Visual acuity screening vs. noncycloplegic autorefraction screening of Native American Preschool Children. J Pediatr Ophthalmol Strabismus 1999;3:160–5. Goss DA, Grosvenor T. Reliability of refraction—a literature review. J Am Optom Assoc 1996;67:619–30. Chan OY, Edwards M. Comparison of cycloplegic and noncycloplegic retinoscopy in Chinese preschool children. Optom Vis Sci 1994; 71:312–8. Lin LL, Shih YF, Hsiao CH, Su TC, Chen CJ, Hung PT. The cycloplegic effects of cyclopentolate and tropicamide on myopic children. J Ocul Pharmacol Ther 1998;14:331–5. Thibos LN, Wheeler W, Horner D. Power vectors: an application of Fourier analysis to the description and statistical analysis of refractive error. Optom Vis Sci 1997;74:367–75. Harvey EM, Miller JM, Dobson V. Reproducibility of corneal astigmatism measurements with a hand held keratometer in preschool children. Br J Ophthalmol 1995;79:983–90. Erin M. Harvey The University of Arizona College of Medicine Department of Ophthalmology. 655 N. Alvernon Way, Suite 108 Tucson, AZ 85711 E-mail: [email protected] Optometry and Vision Science, Vol. 77, No. 3, March 2000