Cepeda Opry Repeatability of Visual Acuity Measurement THOMAS W. RAASCH, OD, PhD, FAAO, IAN L. BAILEY, OD, MS, FAAO, and MARK A. BULLIMORE, MCOptom, PhD, FAAO The Ohio State University College of Optometry, Columbus, Obio (TWR, MAB), and University of California School of Optometry, Berkeley, California (LB) al acuity chart design and acuity testing scoring methods ‘Methods. Visual acuity was measured using the Sloan and British Standard letter series, and Landolt rings. Identifiability of the different letters as a function of size was acuity measurements with a va mated, and expressed in the form of frequency-of-seeing curves. These functions were then used to simulate of chart designs and scoring criter chart design parameters and acuity score, and acuity score repeatability Results. Systematic relationships exist between in particular, an important feature of a chart, that largely determines the repeatability of visual acuity measurement, is the amount of size change attributed to each letter. The methods used to score visual acuity performance also affect repeatal evaluate acuity score validity and repeatability using the stat 75:342-348) Conclusions. It is possible to cussed here. (Optom Vis Sci 1998; al principles Key Words: visual acuity, visual acuity charts, optotypes, scoring methods, measurement validity, measurement repeatability rained essentially unchanged since Snellen introduced them in 1862. Acuity chars typically consist of upper case lecers within rows of decreasing size from top to bortom. Charts of this type are very commonly used to measure acuity, primarily because the test is readily administered with relatively simple equipment, and the results can provide valuable insights inco the state of the eye's optics and the visual pathways. Clinical acuity measurement is probably the most commonly measured visual psychophysical threshold. Acuity is often considered tobe the sin- tle most important indicator of the quality of vision. For example, leccer acuity has been used as a primary outcome variable in large clinical als, as in the Early Treatment Diabetic Retinopathy Scudy (ETDRS),? and in the Macular Photocoagulation Study.” Acuity is often che only visual function used to predict visually guided performance, such as in automobile driver licensing. It is likely that acuity performance will continue to be relied upon for a wide variety of applications. The pervasive use of acuity resting emphasizes the importance of understanding the ffeces that par ticular acuity chare designs and scoring methods have on acuity, In 1980, the Committee on Vision, Working Group 39 of the National Research Council (NAS-NRC). issued a report that ad- T: basic principles of visual acuity measurement have re- “Thomas W. Raich iran AOF Exl Flow dressed several aspects of clinical visual acuity measurement.4 The report stated the need for standards in visual acuity measurement, and made several recommendations. One goal of this working gr0up was to improve che reliability of acuity esting, enabling examiners t0 obtain repeatable results that could be reasonably compared with the results of others. They identified nine issues that are amenable to standardization, and four of chese issues are addeessed in this paper. These four are: (1) comparison of opto- types tothe Landolt rng; (2) the gradation and range in optotype size; (3) the number of symbols at each size level; and (4) the method of scoring. The 1980 Commitee on Vision di nor raise the issue oF repet- ability f acuity scores. Because repeatability isa critically impor- rane characteristic ofany clinical test, we focus primarily on repeat- abiligy when considering the effess of different chart designs and scoring methods Early workin the area of visual acuity chart design was primarily concerned with the design and selection of optotypes to be used in acuity testing (for a review, see Bennet’). One ofthe principal goals was to select a family of optotypes of neary equal legibility. Sloan etal. evaluated three test arget types. She recommended a set oflecters chat could serve asa sutablesubstcuce for the Landolt sing. The recommended leter set (the so-called “Sloan” levers) is composed of 10 nonserfed, uppercase letters formed within 3 square outline, with a stoke wideh of one-fifth the letter height. Optomeny and Viton Scene, Vol75, No.5, May 1998‘The Bricish Standards Institution recommended a different leer series, The selection of these letters was based on the work of Coates® and of Woodruff” The British Standard lerters consist of| asec of 10 nonserifed, uppercase letters, with a height/width ratio of 5/4, and a stroke wideh chat is one-fifth the letter height. The members of these ewo letter sets are: CONKNORSYZ DEFHNPRUVZ ‘Sfoan letters Bricish Standard Letters: Ichas long been assumed that nearly equal legibility of letters is necessary in order to produce reliable acuity scores." -* We evaluate whether commonly used letter series are sufficiently sim- ilar in legibility so as not to inflate variability of acuity measures, ‘Most authors agree that alogatithmic size progression offers the greatest practical and theoretical advantages > Sloan! proposed that resolution thresholds ace similar to other visual thresholds in that they probably follow a Weber's law-like function. Ifresolution ‘thresholds do indeed obey Weber's law, then equal increments in acuity would be represented by equal logarithmic steps in leer size. In other words, leters that ae jut noticeably different in size will be different by a constant proportion, regardless of the letter Most existing charts which do use a logarithmic size progression use a value of 0.1 log unit (a ratio of 1.26%) berween adjacent lines on the chart !"!3 Letters on such chares approximately double in size every 3 lines, and increase 10-fold in 10 lines. An example ofthe commonly used sie progression of.1 logunits per size level is: = 2025 32.40 50 63 80 100 125 160 200 .. ‘This particular choice of size progression has proven ro be conve- nent, but is nonetheless arbitrary. Therefore, we evaluate the in- fluence that size progression has on the chart score and score re- peatabilcy ‘Acchare design feature thats closely related to size progression is ‘the number of leters per size level. The size progression and num ber of levers per line deceemine the total number of lerters chat -must appear within a chare in order o span a given size range. We demonstrate that these two Features, ie, the sie progression and the number of letters at each size level, are the features of a chart that are primarily responsible for determining the variability of acuity measurement with chat chart. “The method chosen to score acuity performance also has an influence on the score, a8 well asthe repeatability of that score. Scoring methods that assign scores in “whole-line” increments sac- rifce repeatability." "4-16 We show chat the beneficial features of this type of scoring protocol extend across a wide range of chart design parameters. Finally, several prior reports state there isan upper bound tothe repeatability of acuity measurement because of the limitations of | commonly available acuity charts.!® We shove, however, cht itis possible to improve repeatability if is possible ro simply increase the number of trials in a measurement of acuity Repeatability of Visual Acuity—Raasch etal MATERIALS AND METHODS Empirical measurement of visual acuity 343 High contrat, black-on-white visual acuity charts were photo- graphically prepared. Landolcrings, the 10 British Standard letters, and the 10 Sloan letters were used. For each oprotype, three pairs of essentially equivalent charts were produced. The setcs consisted of nine char pairs (3 letter ses by 3 spacings). The spacing berween successive rows was equal tothe berween-lerter spacing ofthe larger row. The three spacings were 0.8, 1.0, and 1.25X the letcer wideh Each row contained five letters, and there were seven rows of eters imeach chart. The size progression between successive rows was 0.1 log units. For chars in an “equivalent” pair, only the leter se- «quencing was different berween the two charts. These chart spec- fications are similar to the well-known Baiey-Lovie and ETDRS visual acuity chares?: "> 17 Ae the test distance of 4m, the lewer ‘heights spanned visual angles from 10 to 2.5 min are (6/12 t0 6/3 ‘equivalent, of 20/40 vo 20/10 equivalens) All resting was done tunder incandescent illumination, and the photopic luminance of the white background of the eters was 160 cd/m? Subjects Nineteen normally sighted subjects were tested. The subjects were volunteers from the student population of the School of Op- cometry at the University of California, Berkeley. The age range was from 20 to 33 yeats. All subjects had visual acuity of 20/25 (6775) orbbewer, were iee of any known eye disease, and each used his of her habitual correction. Rigid contact lens wearers were excluded, and soft concac lenses wearers were allowed to partici- pateifaleast 12 h had elapsed since chey had last worn theit ens. Visual acuity scoring procedure ‘The subjects were asked to read each char, starting with the largest row. Subjects were permitted to stop reading a chart when they missed three or more ofthe five lexcersin a row. The subjects? responses were recorded on facsimiles ofthe charts, noting which levers were read correctly. The experimental sessions lasted ap- proximately 20 min. Charts were presented in a random sequence. ‘A visual acuity score was obtained foreach chart by assigning a value of logMAR = 0.02 to cach leter. The top row subtended a visual angle of 10 min arc, or MAR of 10/5 = 2 min atc, so ifthe subject read this entte line and no more, he/she would be assigned an acuity of 20/40, or 0.30 logMAR. Each additional letter read ‘would improve his/her acuity score by 0.02 log units. The rationale for this procedure has been described elsewhere.!!-"* ‘Modeling of acuity measurement with simulations ‘The empirical data described above were used to model and simulate acuity performance. Probability of seeing curves for each [eur in the ewo leer series were computed. The threshold sie for cach subject on each chare was determined using the single-lrter scoting criterion described above. To construct the probability of sing curves, leer size was adjusted according to the threshold size obtained by that subject on that char. Data were accurmulared across subjects to build probability of seeing values for each of che Optomenry and Vision Science, Vol. 75, No.5, May 1998344 Repeatability of Visual Acuity—Raasch etal 10 lewers in the Sloan and British Standard letter series. These probability of seeing data were then fie with cumulative normal distribution curves using Probit analysis.” This process is similar to that used by others! 18 in describing acuity performance “These curves represent the probability that a letter will be iden- tified correctly asa Function of es siz relative to the threshold chat that subject obtained on that chare. Using these probability of seeing curves, itis possible to simulate acuity performance across a wide range of chare parameters. The effects on acuity scores and repeatability are studied by manipulating the size ratio berween lines, and che numberof eters per line, and chen simulating acuity performance, using the probability of seeing functions. Chars used in the simulations were constructed so that no letters ap- peared more chan once ata given size level, all charts contained Seven rows of leer, and the size ofthe chart was randomly se so thatthe height of the letters inthe middle lin subtended an angle somewhere between 4 and 6.3 min arc For each simulated chart condition, 1000 versions were generated and scored. Derivation of scores and score variances from simulated acuity measurement ‘The probability that a letter in a chart, ¢y will be identified correctly has a value between 0.0 and 1.0, and the expected score for cis represented by p(c)- Likewise, the probability that this lett will nor be read correctly is q(c) = 1 ~ ple)» The expected score for the entire char, inthe number of correctly read letes, i the sum of these individual letter probabilities, Bp(q).. Furthermore, che variance ofthe score, in leters* can be esti- mated. Forasingle letter, the variance ofthe expected score for that lewer is o2 = ple) X q(c). This single eter variance ranges beeween 0.0 letter® (for letters much above and much below threshold size) co 0.25 letets,? for lerters with p(c) = 0.5. The variance of the score for the whole chats the sum of the variances of the letters making up the chart, V = ia}. The SD of the chart score, 5, is the square root of cis variance, VV. Furthermore in a tesezecest experiment in which the test and retest measurements are independent, the SD of the discrepancy scoreis D = -/2 X S. Fig. 1 is an example of a chart consisting of Sloan lees, in which che lerters span a four-fold size range. Included under each, lerer is the probability of correctly idencifying the leter and the variance ofthat probability. Given these particular letters and sizes, this particular chart has an expected score of 19.975 letters, a variance of that score of 1.67 lewers* and a SD of the score of 1.292 Jeers. in tis chart each letter aecounts for 0.02 logunits, 0 this seore SD is: 1.292 leters X 0.02 log units/lecer = 0.02584 log unis. OF couse, if this chac is viewed by a differen observer, ppethaps with adiferene acuity, and from a different distance, these probabilities and variance will differ RESULTS Comparison of empirical acuities with letter optotypes and Landolt rings ‘These data indicate there are small differences among acuites ‘obtained wich the lecter oprorypes and Landolt rings (see Table 1). Simla results ate reported elsewhere.“ Acaities obtained with Sloan letters are slightly becter than hose obtained with Landole Angular feat a De Ne mien! Probably: 1.000 1.000 1.000 1.000 1.000 100 ‘Vananee: 0.000 0.000 0.000 0.000 0.000 mn ae sR oz kK oD Probably: 1.000 1.000 1.000 4.000 1.000 8.0 ‘Variance: 0.000 0.000 0.000 0.000 0.000 Mee ce ol ve ic Probabity: 1.000 1.000 0s72 1.000 0.003 6 ‘Varance: 0.000 0.000 0.027 0.000 0.007 No ov oop 09 K Probabity; 0780 0924 81704020730 5" Variance: 0.171 0070 01490280 0.103 vow c¢ No Probabity: 03080572 007 0.122 0025 4.0" Variance: 0212 0.245 | 007 107 onze Ss ovo oc oz Probabity: 0.000 0007 0039 0.000 ott az ‘Varance: 0.000 0.007 0.088 0.000 0.000 re ee Probabiy: 0.000 0.000 0.000 0.000 0000 28+ Variance: 0.000 0.000 0.000 0.000 0.000, FIGURE 1. Example ofa char composed of Sloan letters. Angular size i iste tothe right. and the acuity ofthe observers arbivaly set tobe approximately 20120. The probability and variance values ae listed under each eter. Fot this partiular chart, the expected score i the sum of the individual letter probabiles, which is 19.975 letters (0.0005 logMAR), and the vaiance ‘ofthis score isthe sum of the individual eter variances, which is 1.665 Teters? The SO is therefore 1,292 letter, or 0.026 log units. Note that smost ofthe variance in the score is contibuted by nearthreshold lets, TABLE 1. ‘Acuity and discrepancy in the three families of optotypes, averaged across subjects. ‘Acuity Relative to Test Retest Letter Family Landolt Ring, Discrepancy (SD) LogMAR Letters Lines Letters Log units Landolt Rings 250.050) British Letters -0.005 -0.25 -005 18 0.036 Sloan Letters. -0.038 -1.9 038 24 0.047 rings (betcer by 0.038 log units, or almost 2 “letters” beter with ‘Sloan eters). Sloan® obtained data thae suggested to her that che difference berween acuities obtained with Sloan levers and Landole rings is nor clinically signficane. The difference becween acuities obtained with the British Standard letters and with Landoleringsis very small, ic, 0.005 log units (or 0.25 leters)berter acuity with the British leters. Included in Table 1 is an indication of che width of the distr- bution of test-retest discrepancy scores. The disctepancy scores for theSloan lewersand the Landolt rings ae very similar, and both are larger than for the British Standard leters. These data suggest that charts comprised of British Standard letters will produce slightly more repeatable acuity measurements than charts comprised of cither Landolt rings or Sloan letters. “Table 2 lises the acuity and test-retest discrepancy values, aver- aged across subjecs for the three spacing arrangements. The aver- Optometry and Vision Scene, Vol. 75, No.5, May 1998TABLE 2. Visual acuity (relative to spacing = 1.0) and testretest discrepancy scores for the three spacing arrangements, averaged across subjects and British and Sloan letter sets. spacing Aaity "Discrepancy 80 (X letter wisth) TogMAR Snellen notation Letters Log units oax 0.009 2020. (6/6.12) 240 0.038 10x — 2020 (66) 2.05 0.041 125% -0014 2.10 0042 2019.4 (615.8) age acuity level improves with increasing spacing berween letters, and this was expected from the results of other studies. Diserep- ancy values, however, do not change consistently with spacing. Thats, although acuity evel does change systematically with spac- ing, repeatability does not Repeatability of Visual Acuity—Raasch et al. 345 Probal The experimental data were used to construct probability of correct identification curves, s shown in Fig. 2, A and B, From these curves, acuity performance was simulated under a variety of chare design formats. The 10 curves in each figure specify the probability that a leter will be identified correctly, given its size relative to threshold, Their horizontal positions reflect their differ ent threshold sizes for a given percent correct. Theit vertical posi- tions reflect thei dfficulry (or pereent correct) ata given size. The dificulty rankings listed in this figure are based on percent correct levels a chreshold size. Some curves cross others, so those rankings say sift if different size level is selected. This issu is discussed by Alexander etal.* As pointed out by Alexander et al, che spectral power of the object spatial frequencies of these levers differs, and so their relative identifiability can change, depending upon their angular size. of correct ider ication of letters Effects of size progression and letters per line Fig, 3 illustrates the results of a simulation of acuity perfor- ‘mance, in which size progression and letters per line were system- AL __atically varied. Fig, 3 shows that acuity score does depend upon size ‘ome ‘progression between adjacent lines on the chare, but does not de- | pend upon the number of letters at each size level, ie, the 10 lines ea ate superimposed, These lines havea slope near %, indicating that sz 2s sie progression increases by n log unit, the acuity score in- é creases (ie, ges worse) by n/2log units. 3 Fig. 4shows thatthe repeatability ofthe char score does depend | Penh upon both the size ratio and the number of letters at each size $ i “These plots on log-log axes have slopes near 4, suggesting a square 3 ay root relationships That asthe ne progression beeween lines ge increases by a factor of n, the SD of the score increases by a factor é of Va. Fig. 5s derived from Fig 4. Fig. 5, however, combines sie ratio o% | 02 oF 0 04 ae ttre (ut) B 015 100%. | 3 s Bon Brew. | |e | i 3 0.05 Fon 3 7 & > | a a 0.05 + i 0.05 01 015 02 025 03 035 04 | ‘Size Progression (log units) | om | enidbewernattfgmny | 2 Save to reno (Sunt) curs ‘Aaly resold 2 a funeton of size progression and numberof eters 7 per line. Sloan letters. There are 10 lines, derived from simulations of 1, 2, FIGURE 2, Probability of correct identification asa function of letter size. Lete size is expressed relative ta threshold, A: Sloan letter series, B: British Standard Teter series. 5... 1D leters per line. The 10 lines are superimposed, indicating thatthe chart score i Independent ofthe number of lees per line. The slope of (05 indicates that as size progression changes by an increment of nuns, chart score will change by Vn unis Optometry and Vision Science, Vol.75, No.5, May 1998346 Repeatability of Visual Acuity—Raasch etal [7 A. g ‘tne : ; | om : = oD ‘5 lotarctine cL tm ‘Size Progression (log units) SD ofthe acuity score a a fancton of size progression and numberof letors perline, Sloan ltrs. The bold line represents the commonly sed fve eters Per line. The slope of 0.5 indicates that as size progression changesby afactor (fm unis, char score will change by 2 factor of Vn unis o4 Tt I 2 g a 8 oni Z| | | 0.001 oot 04 1 Log unitsetter FIGURE 5. SO of the acuily score expressed asa function ofthe numberof log units accounted fr by each leter onthe chart. Sian letters. These 10 lines are Superimposed, indicating that regardless ofthe numberof letters per line, the repeatability of the score Is determined primarily by how many log unis of size change ae atribted to each lter. and lettets per line into a single variable. The size progression per line, divided by the numberof lerters per line, yields log units of size progression accounted for by each lever. For example, che common char format of 0.1 log units per line, and five leters per line, yields 0.1 log units/line FTeneaing = 002 log unic/lecer. Fig. 5 illustrates an imporcant concept: the repeatability of an acuity measurement is determined by the number of log units in sae change accounced for by each lecter. This holds true across a wide range of letters per line and size progressions, Scoring criteria Scoring criteria have been evaluated by a number of authors in recent years. 16-20 The issue is whether “whole line” or “in- dividual lewer” scoring methods yield more repeatable acuity scores. All agee that a scoring method that awards credit in ince- ments of single leters is better than a method that awards credit in ‘whole lin increments. The present findings are in agreement with these other studies. The standard error ofthe visual acuity mea- surement using whole line scoring methods tends to yield values shatare approximately 1.6 1.7 times larger than individual leer scoring methods. This difference becween individual leter and whole line scoring methods stays a this level, across. wide range of sine progressions and levers pet line. DISCUSSION Reliability of acuity scores facuityis tested ewice on essentially equivalent chars, the expected difference in score is zero Ifteted many times, there wl be 2 disti- bution of dilference scores, and this disribution reflects the underly- ing variability in he acuity measurement. These types of tesrerest cata have been analyzed in a least two ways, including use of the Inacass Correlation Coefficient" and tse-eest discrepancy values, or the coefficient of rlibily, as described by Bland and Altman? ‘We believe chaces- retest discrepancy scoresarean appropriate typeof analysis, ait isa ditect indication ofthe within-subject variability. The Intraciass Correlation Coefficient does incorporate a rem for within- subject variance, bu it is alo affected by the range of the acuiies, teited, ie, the berween-subject variance. IFesults are to be compared across experiments, the Incracass Correlation Coefficient may obscure the comparison, because berween-subjec variance may mask differ ences in within-subject variance. There is of course, a relationship beween within-subject vari- ance and test-retest discrepancy scores. The Intaclass Correlation Coefficient, R, is equal to: ote? where of isthe variance of the mean values of the n subjects i the beeween-subject variance, and a2 is the within-subject vari ance As the beeween-subject variance increases, Rwill also tend, to increas, illustrating that this statistic is affected by the range of ‘mean values. When the mean discrepancy score is 270 (i, when there is no overall erend from test to retest), the variance of the dlscrepancy sores is: 0%, = 4(@3) ‘A number of other authors in recent years have studied repeat- ability of visual acuity measurement.!! 46:29 There are small differences in the results reported by these different authors, asthe daca sets are the result of diferent experimental methods and sub- ject populations. However, the differences in repeatability Found, for visual acuity measurement are not great. Table 3 lists repeat- ability data from several studies in which Sloan letters were used, and with a size progression of 0.1 log units and five lees per size level. Most studies find that test-retest SDs are on the order of two 10 three lerters (ie, 0.04 to 0.06 log unis), Optometry and Viton Science, Vo. 75, No.5, May 1998Repeatability of Visual Acuity—Raasch et al. 347 TABLE 3. Repeatability of visual acuity testing reported in the literature. Bailey & Lovie Elliott & Sheridan Bailey et al. Arditi & Cagenello Vanden Bosch Raasch et al 976" (1988F? (19914 1993)" __& Wall (1997)'* (present study) SD of VA Score. 18 1.93 14 2.16 19 1.66 Test-Retest SD 255 273 1.98 3.06 2.69 235 95% Confidence Interval 5 5.35 3.88 6 5.27 461 All scares from use of Sloan Teters, a size progression of 0.1 log units, and five letters per size level. Units are letters (0.02 log units/eter. Important chart design principles Previous authors*®: "1? have argued in favor ofa logarithmic size progression in letter size. Likewise, it has been proposed that the task be essentially the same at each size level? This requires, among other things, thatthe same numberof letters be present at cach size level. When these ewo features are present, each letter in a chare accounts for an equal increment in acuity. “The slope ofthe function relating SD ofthe acuity score wo the size change per lever is equal o %, suggesting a square root rule. Indeed, this would be expected ifan acuity scores simply considered a sample of the subject's acuity. According to a square root rue, increasing the number oflerters used to measure the acuity bya factor ofa will reduce the standard error ofthe estimate by a factor of Vn. As Arti and ‘Cagenello!® point out, the repeatability of an acuity measurement could be improved ifthe size progression were 0.05 log units, rather than che common 0.10 log units. This is not the only way to improve repeatabilty, however. For example, the number of lerters could be doubled by doubling the number ofleers in each line, or by measus- ing the acuity «vice (with differene but otherwise equivalent chars), and averaging the results Ether ofthese improves the precision of the estimate of acuity by a factor of V/2. ‘The results ofboth the empirical data and the simulations com- paring the four differen scoring criteria indicate that acuity mea- surements are most accurate when measured in the finest size in crements available, i essentially o the fact chat letet-by-leter scoring will yield rela- tively few instances in which the disparity is as large asa whole ine, aresule that arises relatively feequenely with a whole-line eriter- ‘on.!4 The end resule is chat the distribution of test-retest discrep- ancies (and hence the score SD) is narrower when an individual letter scoring method is used. in single letter increments. This is due Is equal legibility of letters an advantage? Tehas been assumed that in order wo achieve the most reliable acuity measurements possible, the letters within a series should be equally legible 29 Te is unlikely that perfecdy equal legibility can be achieved, even with targets such as Landolt sings, where meridional differences may favor some orientations over others Recognizing tha the available leer sets are not equally recognizable, what are the consequences? Are the differences in ebilitysuficiendly great to con- tribute to variably in acuity measurement? To answer these ques- tions, 0 additional simulations were performed, In the fst sim faced acuity measurement was performed using Sloan levers arranged Jn the common chare format of5 lees per line and 0.1 log unit size progression, The second simulation was performed in the same way, afer substituting a hypotheical eter se in which all had the same legibiry. That is, the probability ofdentifation curves (asin Fig. 2) ‘was identical forall letters in this set, and was the average of chose for the Sloan leer. ‘The results ofthese simulated acuity measurements indicate that Sloan leter chart scores ate jus slightly more variable than scores forthe “equal legibility” chart. The Sloan chart SD was equal to 1.072 lecters vs. 1.026 letters for the equal legibility chart. The difference is less than 1/20ch ofa leer, and suggests that, for most purposes, the inequality of identifiability of these esters is not a significant factor. It should be noted, however, that if a chart is made up primarily of “easy” or “hard!” lerters, scores obtained with that chare would be affected. In designing a chart, i is probably sufficient to simply use all letters inthe set with about equal fre~ quency, and mote or less randomly. Itis also probably unnecessary {0 pursue an alternative set of leters or symbols that are perfectly equal in legibilcy, as the benefits of even perfect equality are quite modes. Evaluation of existing charts Existing charts can be evaluated with these principles in mind Some charts may have a format in which che size progression is as low as 0.012 log units pe letter atthe smaller letter sizes. The same chart may have a size progression at the larger lever sites of 0.3 log ‘units per lever, as when single leeers appear atthe 400 and 200 foot-lerter sizes. The score variability is exceedingly high at this level, which is noteworthy because the threshold for determination of legal blindness isin this acuity region (see Fig. 5). This is an important reason why an acuity measurement with substancal consequences, such as establishment of legal blindnes, should per- haps not be done with an acuity char chat has this weakness, ‘Clinicians have observed that some low vision patients exhibit large acuity differences from one measurement to another. Greater varabil iry in acuity performance by some low vision patients may play a part in this? bursome acuity chares may contribute addtional variability. ‘Charts used for the measurement of low visual acuity necessarily in- clude large characters, so in order to limi the overall size of these charts, relatively few lewers may be used. Consequently, such charts often use large size ratios from one line to the next (to reduce the ‘number oflines), and may have few leters on each line. Although this reduoes the ste of such charts, the effect isto make the measured acuities more variable In oder to achieve greater repeatability, the size increment accounted for by each letter should be kept as small sis practical, and preferably constane, as letter size is changed. Optomeny and Vision Science, Val. 75, No.5, May 1998348 Repeatability of Visual Acuity—Raasch etal NAS-NRC standards for visual acuity measurement The NRC Committee on Vision has offered several recommen- dacions for the measurement of visual acuity.* The Landole ri has been proposed asthe standard optotype although the commit- tee sates that other optorypes which yield similar acuities are ac- ceptable. The empicial data of the swo series of leer optorypes indicate cha auities obtained with the British Standard leers are very cose to those obtained with Landolt rings. Acuities obtained ‘with Sloan letters may be slightly differen, yielding threshold sizes, that are approximately 89 smaller chan those obtained with Landolt sings or British Standard lerers. ‘The NRC commits has recommended that a logarithmic size progression of 0.1 log units be used, and chat all 10 leters in the series be presented at each size level. Existing charts infrequently include so many symbols at each size. Size progression has two cffcrs First, change in size progression by an increment of a will change the score by an increment of Vn units (see Fig. 3) Second, and more important, multiplying che size progression by a factor of ‘will multiply the wideh ofthe distribution of discrepancy scores, by Va (sce Fig. 4 oF Fig. 5). “The NRC committee has further recommended that a scoring cicerion of7 of 10 (or 6 oF 8) letters be required to pass size evel ‘The results of ths study and of others!” “151-29 indicate that these whole-line scoring methods do a relatively poor job in cms of score repeatability. A scoring method that awards creditin single letter increments produces scores that are more repeatable. Charts inwhich each lewer accounts fora constant proportion oflerer size are conducive to scoring methods that award credit in individual lewer increments, and the use of such scoring methods is reeom- ‘mended for situations in which accurate and reliable acuity mea- surement is desited (see Fig, 5). Finally ithas been suggested that there isan upper bound onthe precision wich which acuity can be measured. This is true if there are limitations placed on the measurement procedure such as 2 shortage of time oF the number of acuity charts or letter sequences, available for use. IFivis possible to measure acuity more than once, with chars that are equivalent (eg., the same design except for letter sequencing), then the repeatability of the results ean be en- hanced by simply averaging the results. The average is beret in terms of the standard error of measurement, by a factor of ¥/2. ‘Thisisa consequence ofthe slope in Figs. 4 and 5. Testing ewice cssentially halves the number of log units accounted for by each letter. If ested with still more letters iis possible to enkance the repeatability of the measurement further. ACKNOWLEDGMENT Dr, Rasch wa supported by an EzllFellownip during the period this study twas conducted. Received August 5, 1993; revision recived January 5, 1998. REFERENCES 1. Bennett AG. Ophthalmic est eypes. Br J Physiol Ope 1965:22; 238-71 2, Early Treatment Diabetic Retinopathy Study Group. Photocoagula- tion for diabetie macular edema. Arch Ophthalmol 1985;103: 1796-806. 10. n 2, 16. 17 23, m4, 25, Macular Photocoagulation Study Group. Visual outcome afer laser photocoagulation for subfoveal choroidal neovascularizaton second ary to age-related macular degeneration. The influence of inital le- sion size and inital visual acuity. Arch Ophthalmol 1994:112.480-8, [National Research Council Committee on Vision. Recommended standatd procedures forthe clinical measurement and specification of visual acuity. Adv Ophthalmol 1980;41:103-48, Sloan LL, Rowland WM, Aleman A. Comparison of three ypes of test target for the measurement of visual acuity. Q Rev Ophehalmal 1952;8:4-16. ‘Costes WR. Visual acuity and test lentes. Transactions ofthe Insti- tute of Ophthalmic Opticians IT, 1935. /. Woodruff EW. Visual acuity andthe selection of est eters. In: Some Recent Advances in Ophthalmic Optics: Transactions. London: Hat- ton Press, 1948:59-70. Alexander KR, Wei X Derlaci DJ. Visual acity and contrast sensi- tivey for individual Sloan lewes. Vision Res 1997:37:813-7. Bailey IL, Lovie JE. New design principles for viual acuity leer chars, Am J Optom Physiol Op: 1976;53:740-5, Sloan L. The measurement of visual acuity. A rita review. Arch (Ophthalmol 1951:45:704-25 Ferris II FL, Kassoff A, Bresnick GH, Bailey I. New visual acuity chars for clinical research. Am J Ophthalmol 1982;94:91-6. ‘Westheimer G, Sealing of visual auity measurements. Arch Ophihal- ‘mol 1979597:327-30. Regan D. Low-contrst lester chares and sinewave grating test in ‘ophthalmological and neurological disorders. Clin Vision Sci. 1988; 2:235-50, Bailey IL, Bullimose MA, Raasch TW, Taylor HR. Clinical grading and che effect of scaling, Invest Ophthalmol Vis Sei 1991:32: 40-32, Audi A, Cagenello R. On che statistical reliably of lete-chart visual acuity measurements, Invest Ophthalmol Vie Sci 199334: 120-9, Vanden Bosch ME, Wall M. Visual acuity sored by the lener-by- lewer or probie methods has lower retst variably chan the line assignment method. Eye 1997;11:411-7. Strong G, Woo GC. A distance visual seuity chart incorporating Some new design features. Arch Ophthalmol 1985;103:44-6, ioe: DB, Bullimoce MA, Bailey IL Improving th reliability of che Peli-Robson contrast sensitivity char. Clin Vision Sei 1991;63471-5, Finney DJ. Probit Analysis. 3rd ed. Cambridge University Press, 1971, Eliott DB, Sheridan M. The use of accurate visual acuity measure- -eataract formulation wal, Ophthalmic Physiol (Ope 1988:8:397-401. . Fis JL. The Design and Analysis of Clinial Experiments, New York: John Wiley & Sons, 1986. 2. Bland JM, Aleman DG. Statistical methods for asesing agreement between two methods of clinical measurement. Lancet 1986 1(8476):307-10. Hedin A, Olsson K. Leer lgibilcy and the construction of new Visual acuity chare. Ophthalmologica 1984:189:147-56, Mitchell DE, Freeman RD, Millodor M, Haegersom G. Mesiional amblyopia: evidence for modification ofthe human vial sytem by early visual experience. Vision Res 1973:13:535-58. Rausch TW, Flom RE. Precision of visual acuity measurement in low vision. Invest Ophthalmol Vis Sci 1994;35(Supp):1413, ‘Thomas W. Rasch The Ohio State University College of Optomesry 338 W. 10th Avenue Columbus, Obio 43210 Oprometry and Vision Science, Vol.75, No.5, May 1998