Model Reporting Area for Blindness Statistics,

Transcription

1 EPIDEMIOLOGIC REVIEWS Copyright 1983 by The Johns Hopkins University School of Hygiene and Public Health All rights reserved Vol. 5, 1983 Printed in U.SA. METHODOLOGICAL PROBLEMS IN EYE DISEASE EPIDEMIOLOGY 1 FRED EDERER BACKGROUND Until recently, research activity in eye disease epidemiology was sparse. The sparseness was accounted for by a lack of ophthalmologists with training in epidemiology and of epidemiologists with interest in eye disorders. An indication of recent expansion of activity is the increase in the number of eye epidemiology papers published in the American Journal of Epidemiology from four in to 15 in Furthermore, in June 1982, the first international symposium on this subject was held in Bethesda, Maryland, drawing more than 100 scientists from the United States and abroad. Thirteen papers from this symposium appear in the August 1983 issue of the American Journal of Epidemiology, together with an introductory article discussing the recent increase of activity in the field and factors contributing to it (1). These factors include increases in the number of ophthalmologists with epidemiologic training and of epidemiologists interested in eye diseases. The present paper is an informal survey of selected methodological problems in the epidemiology of eye disease. It is largely limited to the author's encounter with epidemiologic research in the four chronic diseases that are the major causes of blindness in the United States: senile cat- Abbreviation: HANES, Health and Nutrition Examination Survey. 1 From the Office of Biometry and Epidemiology, National Eye Institute, National Institutes of Health, Bethesda, MD (Reprint requests to Mr. Ederer.) The author thanks those who read and commented on the essay, and is especially grateful to Drs. Robert D. Sperduto, Dean E. Krueger, and Frederick L. Ferris, HI for their critical comments. aract, senile macular degeneration, diabetic retinopathy, and glaucoma. Although the coverage of methodological problems and eye diseases is limited, the intended coverage of types of epidemiologic studies is broad: case registers, population surveys, cohort studies, case-control studies, clinical trials, and natural history studies. As an introduction to the discussion of specific methodological problems, it will be useful to review critically the methodological features of and problems met by five large studies carried out in this country during the past two decades: the Model Reporting Area for Blindness Statistics, the Health and Nutrition Examination Survey, the Framingham Eye Study, the Diabetic Retinopathy Study, and the pilot study for the Visual Acuity Impairment Survey. The Office of Biometry and Epidemiology, National Eye Institute, was involved in each of these studies. Model Reporting Area for Blindness Statistics In the 1960s, the National Institute of Neurological Diseases and Blindness organized the Model Reporting Area for Blindness Statistics, a consortium of state blindness registers which pooled their information to produce population-based data on the incidence and prevalence of blindness by cause. This study attempted to improve the uniformity and reliability of the state register data by establishing uniform definitions and procedures (2). The Model Reporting Area understated the extent of blindness in the population, perhaps by a factor of as much as 2 or 3 (3). Its greatest epidemiologic weaknesses 51

2 52 EDERER were that neither the extent of underregistration nor the variability of underregistration according to subgroups was adequately evaluated. It was taken for granted that registration of the middle and upper socioeconomic classes was less complete than of poor people, and the generally higher registration rates for blacks than for whites were taken as presumptive evidence for the social class difference in registration (4). Despite these problems, the Model Reporting Area emerged with certain findings for which may be accepted as real: incidence and prevalence of registered blindness remained stable over the nine-year period ; senile macular degeneration, diabetic retinopathy, glaucoma, and senile cataract were the major causes of blindness (4); black women had a higher prevalence of blindness from diabetic retinopathy than black men, white women, or white men (5); and blacks had excess prevalence of blindness from glaucoma (6). Nevertheless, blindness register data will continue to be of limited value in epidemiologic research unless the aforementioned deficiencies can be corrected. Health and Nutrition Examination Survey In , the newly established National Eye Institute participated in the National Center for Health Statistics Health and Nutrition Examination Survey (HANES) by contracting with local departments of ophthalmology throughout the United States to provide ophthalmologists to conduct a "complete" eye examination of persons aged 1-74 years in a probability sample of the United States population (7). The timetable for the start of the survey allowed little time to prepare a detailed eye examination protocol or to standardize examination procedures, and the time that was available for training the examiners in the protocol procedures was also severely limited. Although more than 30,000 persons were surveyed by HANES in , the eye examinations were discontinued in 1972 after some 10,000 subjects had been examined by 91 examiners in 35 locations, because the National Eye Institute encountered too much difficulty in recruiting examiners. The task of completing numerous detailed examinations day after day of eyes that were largely disease-free may have been an inadequate challenge to the ophthalmologists. Cooperation by subjects was fair: of 14,147 persons aged 1-74 years in the sample, 10,126 (71.6 per cent) came in for the eye examination (8). A limitation of the HANES eye examination was that it excluded a visual field test, in part because that test is too time-consuming, leaving the assessment of glaucoma incomplete. Again, as in the case of the Model Reporting Area for Blindness Statistics, despite limitations, a number of useful descriptive (8-11) and analytic (12-17) findings have emerged from HANES. The analytic findings emerged because HANES included various eye examinations as well as other kinds of clinical examinations, histories, laboratory tests, and a nutrition interview. Nevertheless, future HANES, or similar, efforts to examine the eyes of a large population sample will not be wholly successful unless a) the problem of getting the eye examinations done can be solved, and b) adequate time is allowed for protocol preparation, diagnostic standardization, and examiner training. Possible solutions for the eye examiner problem are considered under Needed Technologic Advances. Framingham Eye Study In , the National Eye Institute again participated in a large eye survey (2675 examinees), this time of an elderly population (aged years). The Framingham Eye Study was limited to the four chronic eye diseases that are the

3 METHODOLOGICAL PROBLEMS IN EYE DISEASE EPIDEMIOLOGY 53 major causes of blindness in the United States. Although not as large as HANES, because the chronic eye diseases that cause blindness mainly affect older people, the per capita yield of cases in the Framingham Eye Study was much greater than in HANES and the work of the examiners, ophthalmologists from Boston University, was more challenging. Cooperation by ophthalmologists and subjects was excellent. Eighty-four per cent of the subjects residing in the Framingham, Massachusetts, area were examined. They were the survivors of the Framingham Heart Study cohort who had been under continual surveillance since 1948 and had cooperated well with that study. The Framingham Eye Study emphasized quality assurance: there was ample time for protocol preparation, diagnostic standardization, and examiner training. The study was mainly limited to the four major chronic eye diseases to permit the examiners and other investigators to concentrate on high quality data collection. To assure quality, replicate examinations were performed on a sample of subjects, interexaminer variability was monitored throughout the study, and retraining seminars were held from time to time. Despite these efforts at diagnostic standardization and quality assurance, considerable examiner variability was found for some observations (18). Detailed data from the Framingham Eye Study, including information on data quality, have been published in a monograph (19). Numerous additional epidemiologic papers based on this study have been published (13, 20-29). Whereas HANES omitted the visual field test entirely, the Framingham Eye Study included this test, but only for a) those who met the glaucoma history, intraocular pressure, or cup-disc ratio criteria for "glaucoma suspect" during the screening examination, and b) those in a 10 per cent sample of the entire study population. Data from the 10 per cent sample indicate that an appreciable number of persons who did not meet the criteria for glaucoma suspect also had glaucoma (19, 24); therefore, the assessment of glaucoma, as in HANES, was incomplete. Because pupils had been dilated during the screening examination to identify suspects, the visual field test for glaucoma suspects was scheduled for another date. The delay was epidemiologically costly: 18 per cent of suspects failed to appear for that visit (19). Diabetic Retinopathy Study The Diabetic Retinopathy Study was the first in a series of randomized clinical trials in diabetic retinopathy sponsored by the National Eye Institute. The study investigated the value of photocoagulation treatment in preventing severe visual loss from proliferative diabetic retinopathy. More than 1700 patients, enrolled in 15 clinical centers in the United States from 1972 to 1975, had one eye selected randomly for treatment, while the fellow eye served as untreated control. The main finding was that photocoagulation reduced the incidence of severe visual loss by 60 per cent (30). Patient follow-up terminated in The control (untreated) eyes provided natural history data on proliferative diabetic retinopathy, a major complication of diabetes, and the quality control procedures used in the Diabetic Retinopathy Study have application to other epidemiologic studies. The study made an extensive effort to ensure that the data collected were of good quality. Diagnostic and therapeutic procedures were standardized, training and certification requirements for clinical personnel were established, and a Clinic Monitoring Committee was appointed to develop and carry out an ongoing quality assurance program (31, 32). The quality assurance procedures included training and certification of ophthalmologists, visual acuity examiners, and photographers; monthly tele-

4 54 EDERER phone calls from the Clinic Monitoring Committee to each clinic to review various procedures, personnel changes, training and certification activities, and problems encountered; at least one visit a year from the committee to each clinic to review procedures and problems and to replicate visual acuity tests; internal surveillance of the Fundus Photograph Reading Center; and internal and external surveillance of the Coordinating Center. The Diabetic Retinopathy Study Research Group has published a series of papers (30, 31, 33-38). Visual Acuity Impairment Survey {pilot study) In , a pilot study was carried out in three large cities of the United States to determine whether it is feasible to conduct a national Visual Acuity Impairment Survey, a survey of unilateral and bilateral visual acuity impairment in a probability sample of the adult population of large cities (39). The main objectives of a national Visual Acuity Impairment Survey would be to measure the prevalence of visual acuity impairment by cause of impairment and to carry out casecontrol studies of possible etiologic factors of visual impairment from specified causes. Visual impairment by cause was determined in two stages. The first stage was a household visual acuity screen of more than 2000 persons conducted by a US Census Bureau interviewer, and the second stage an ophthalmologic examination in an eye clinic. A report on the pilot study is in preparation (40). DISEASE DEFINITIONS AND CLASSIFICATIONS Disease definitions used in clinical practice tend to be broad, and this allows clinicians to use their experience and judgment in selecting correct diagnoses for individual patients. But broad definitions are unsatisfactory in epidemiology, where national and international comparisons of disease frequency or of studies of disease etiology depend on standardized definitions. The absence of standardized disease definitions has been a handicap to epidemiologic research on eye diseases. The problem of arriving at a standardized epidemiologic definition for open angle glaucoma is particularly difficult because pathogenesis is unknown and ophthalmologists have a conceptual disagreement on the disease's constituent elements. The three elements that make up the diagnostic triad of open angle glaucoma are elevated intraocular pressure, cupping of the optic disc, and characteristic nerve fiber bundle defects of the visual field. As Leske (41) points out, "When all these signs are present, the diagnosis of glaucoma [is]... not questioned, but if the diagnostic triad is incomplete, there is no general agreement regarding diagnosis..." Cataract also lacks a standardized definition. A cataract is present when the loss of normal transparency of the lens results in interference with the passage of light through it (42). Because early senile lens changes, often insufficient to cause visual impairment, are present in more than half of persons aged 65 or more, the usual diagnostic criteria for cataract are the presence not only of lens opacity but also of visual acuity deficit. The extent of the deficit and the causal relationship of the deficit to the opacity are not standardized. HANES, for example, defined cataract as lens change consistent with visual acuities of 20/25 or worse (43). The Framingham Eye Study, on the other hand, defined cataract as lens change in the presence of visual acuity of 20/30 or worse (20). If the difference between these definitions were limited to the extent of the visual acuity deficit, then it would merely be a severity difference. The difference

5 METHODOLOGICAL PROBLEMS IN EYE DISEASE EPIDEMIOLOGY 55 between "consistent with" and "in the presence of however, is conceptual. The phrase "in the presence of" used by the Framingham Eye Study has the advantage of being operationally simple and of not requiring a subjective judgment. The expression "consistent with" used by HANES requires the examiner to make a subjective judgment, in some instances a difficult judgment. The disadvantage of the definition given by the Framingham Eye Study is that lens opacities not severe enough to cause the requisite visual acuity deficit are nevertheless classified as cataract when the deficit is due to another cause, such as senile macular degeneration. Undoubtedly, the published cataract data of the Framingham Eye Study (20) include cases of visual acuity deficit attributable to causes other than cataract. Similarly, because senile macular disease in the Framingham Eye Study had a visual acuity criterion similar to cataract, cases undoubtedly include those with visual acuity deficit from other causes. This problem is discussed by Ferris (44). For each disease or condition present in an eye with a visual acuity deficit of 20/ 50 or more, the examiners in the pilot study of the Visual Acuity Impairment Survey were required to estimate the effect on visual acuity "if this were the only abnormality present" (39). This design provides for possible tabulations of frequency of visual acuity impairment by single cause (the disease or condition effecting the greatest deficit) or multiple cause (each condition effecting a deficit of 20/50 or more). The clinical examiners made this estimate knowing the visual acuity. Because knowledge of visual acuity might bias the estimates, additional estimates were made, in a sample of cases, by clinical examiners who did not know the visual acuity. Further estimates of this effect were independently made by the readers of eye photographs and visual fields who did not know the visual acuity. An evaluation of the outcome of these three methods should be helpful in assessing their value. A first constructive step toward the development of disease definitions would be the development of disease classifications (taxonomies). This has been done for diabetic retinopathy with the Airlie House classification (45), later modified (37). In this approach, the presence and severity of lesions seen on stereo fundus photographs of seven fields are classified on a three- to six-step scale according to specific written definitions and by comparing the lesions with standard photographs. The Airlie House classification permits not only various categorizations of the disease but also various definitions. Fundus photograph classifications for senile macular degeneration have also been proposed and used by the Framingham Eye Study (19) and by Hyman et al. (46). As in the case of diabetic retinopathy, these classifications can also serve as a basis for disease definition. Ferris (44) has made some general recommendations regarding the definition of senile macular degeneration. If visual acuity is a part of the definition, he proposes that an attempt be made "to estimate whether the level of decreased vision can be explained by the degree of senile macular degeneration present. Masked experiments to test the reproducibility of the evaluation should be included." As noted earlier, this attempt was made by the Visual Acuity Impairment Survey (39). OPHTHALMIC OBSERVATIONS AND MEASUREMENTS Reproducibility of observations "Many ophthalmic observations are inherently subjective. A particular item of information may be influenced by the training and experience of the ophthalmologist, the conditions under which the

6 56 EDERER observation is made, and the cooperativeness of the patient" (19). The Framingham Eye Study made intensive efforts to standardize procedures with written protocols of examination procedures, standard fundus photographs for comparison with clinical findings, repeated training sessions for examiners, replication of examinations, and monitoring of interexaminer differences in recorded frequency and magnitude of observations. Despite these efforts, during the study Kahn et al. (18) discovered interexaminer differences in estimates of cupdisc ratio and diagnoses of macular pigment disturbance, mild nuclear sclerosis, arcus senilis, and myopic cup that were too large to be explained by chance variation. They consequently revised the protocol so as to improve and clarify it, held additional training sessions, and, where appropriate, adopted standard reference photographs. These efforts seemed helpful in reducing certain interobserver differences, but excessive variation persisted for cup-disc ratio and macular pigment disturbance. From this experience, they recommended, as measures to reduce observer variation, more tightly written protocols, increased examiner training, and more reliance on objective measurements. Optic disc cupping, as has been mentioned, is one of the elements in the diagnostic triad of open angle glaucoma, a leading cause of blindness in the United States. Kahn et al.'s (18) findings of large variation in cup-disc ratio measurement are supported by those of Lichter (47) and Schwartz (48). Whereas Kahn studied reproducibility among residents in ophthalmology, Lichter studied it among glaucoma experts. These results are particularly discouraging in the light of Kahn's failure to better the situation after protocol improvements, additional training, and use of standard reference photos. It is clear that a better method of measuring cupping of the disc is needed for both clinical and epidemiologic use. Brilliant et al. (49), as part of a blindness survey in Nepal, included 529 replicate examinations in order to study observer reliability, assessing the latter with the kappa statistic (50). In their evaluation of four variables related to visual acuity, seven to cataract, 16 to trachoma, and seven to other diseases and conditions, they found that five of the 16 trachoma variables and three of the seven variables related to other diseases and conditions had poor interobserver agreement (kappa less than 0.40). Their report includes a literature review of observer variation in ophthalmic research. In summary, studies of reproducibility have found good observer agreement on a large number of ophthalmic measures, but have identified enough problem areas to caution epidemiologists not to take good reproducibility for granted. Subjectivity of observations appears to be a major source of difficulty, suggesting a need for developing objective measures. More on this follows under Needed Technologic Advances. Visual acuity and visual field tests The two most important measures of visual function are those of visual acuity and visual field. Measurement of (central) visual acuity is necessary for epidemiologic studies of the many diseases that cause visual impairment. The test is commonly performed by having the subject read letters on a chart from a distance of 10 to 20 feet. The visual field test, which maps out that portion of space in which objects are visible to the steadily fixating eye, is needed for the diagnosis not only of glaucoma but also of a number of other eye diseases that cause visual field impairment. The usual visual field examination determines the outer limits of the visual field and detects areas of defective vision within that field. Both of these visual function tests are subjective and de-

7 METHODOLOGICAL PROBLEMS IN EYE DISEASE EPIDEMIOLOGY 57 pend on the subject's cooperation, alertness, and ability to follow instructions. (To some extent, the visual acuity test may be a test of intelligence in that intelligent subjects may more easily be able to guess correctly letters they cannot clearly discern.) Both the examiner and the subject can bias the results of these vision tests. This possibility becomes important, for example, in clinical trials in which visual function is the outcome variable. Both biases can be eliminated, when feasible, by "double masking" (51). Examiner masking, although cumbersome, should be feasible in most studies, even in surgical trials, but subject masking is feasible only in trials in which the study treatments are indistinguishable to the subject. In surgical trials, possible patient malingering, although not entirely preventable, can be dissuaded by having the examiner urge, cajole, and encourage the patient to a maximum effort to read or guess, and to continue testing until the patient makes, say, two or more mistakes in a line of five letters (52). In addition to possible personal biases, the visual acuity measurement is influenced by physical variables: 1) location of light for the eye chart front, rear (through glass), or projected image; 2) amount of chart illumination; 3) amount of room illumination; 4) distance of patient from chart; 5) degree of black-white contrast between letters and background; 6) degree of reading difficulty of letters (some letters are more difficult than others); and 7) printing style of letters (32). Because variation in any of these factors can influence test outcome, both equipment and procedures should be standardized, not only in multicenter studies, where the need is obvious, but also in single-center studies to obtain comparability between examiners within a study and between outcomes of similar studies in different settings. Although standards exist for visual acuity testing (53), they are not generally followed. Two recent papers have proposed detailed methods of standardizing the visual acuity test which are particularly suitable for epidemiologic research. Ferris et al. (54) have designed visual acuity charts, available commercially, which overcome several deficiencies of currently widely used charts and have certain optimal properties for quantitative visual acuity measurement: the number of letters per line is constant (five), the lines are of equal difficulty, and letter size progresses geometrically from line to line, so that, independent of initial visual acuity, a three-line worsening amounts to a doubling of the visual angle. Three charts are available, one for refraction and two (one for each eye) for visual acuity measurement. Ferris and Sperduto (55) have reported on the design of two visual acuity boxes, one front-lighted and the other retroilluminated, to achieve evenness of illumination and contrast across the face of the chart and to maintain constant illumination over time. NEEDED TECHNOLOGIC ADVANCES This review has identified a number of problems that point to the need for technologic advances through which direct examinations by ophthalmologists would be replaced either by examinations performed by nonophthalmologists or by observations made by instruments, such as cameras, automated perimeters, automated refractors, or electrophysiologic recorders. In this section, some specific technologic advances are discussed which may be feasible in the current state of knowledge. The objectives of these substitute examinations are one or more of the following: 1) reduce the ophthalmologist's workload; 2) reduce inconvenience (e.g., time lost from work) to the subject; 3) reduce subjectivity (increase objectivity) of the examination; 4) facilitate replicate examinations; and 5) facilitate develop-

8 58 EDERER ment of standard definitions and classifications. The ophthalmologist's workload can be reduced either by training nonophthalmologists (e.g., technicians) to perform examinations (which is feasible for some examinations) or by using an instrument to make the observations. Even if the ophthalmologist eventually has to assess an observation recorded by an instrument, such as a photograph, he or she may be able to do that in less time or at a more convenient time. Subject convenience can be increased by reducing the length of the examination. Moreover, recorded observations such as photographs, electric tracings, or digital computer output, as opposed to direct observations, may facilitate a) objectivity of the examination, b) the performance of replicate examinations, and c) reassessment at a later time according to different definitions or standards. Recorded observations can be a) examined at leisure, b) compared with standard records, and c) circulated among several examiners in the mail. It may also be easier to develop standardized definitions and classifications from recorded than from direct observations; an example is the Airlie House classification of diabetic retinopathy (37, 45), which relies on standard photographs. Technologic problems Methodological problems identified in this review which might be remedied by suitable technologic advances are the following: Inability to obtain ophthalmologists as examiners. The inability to get ophthalmologists as examiners to conduct the HANES eye examinations may be related to the length of the examination and to the inadequate challenge to the ophthalmologists of examining a large, mostly disease-free population. HANES, a US Public Health Service effort, is a potentially invaluable resource to eye epidemiology in that it combines various examinations on a large (some 30,000 individuals in ) representative sample of the US population. The examination, for example, included, among other components, a physical examination pertinent to certain chronic diseases, a nutrition interview, a dermatologic examination, and laboratory tests. Thus, the ophthalmologic data collected in this survey provide not only descriptive epidemiologic information, but information for studying associations between eye diseases and findings from the other examinations. The HANES eye data have in fact been used in this way (12-17). It should be noted that these HANES studies, mainly of the more common eye conditions, were possible although the number of subjects whose eyes were examined was limited to one third of the total population examined. How many more studies would be possible if the eyes of the entire sample had been examined? With suitable technologic advances, and this includes examinations by nonophthalmologists, a future HANES eye survey may be feasible. Lengthy visual field test. The Goldmann visual field test usually takes minutes to complete for both eyes. Its length can lead to fatigue, which may influence performance, particularly in elderly people, and can also be an imposition on a subject's time. As a result, neither HANES nor the Framingham Eye Study included this test for all subjects, and to that extent the glaucoma examination was incomplete in both studies. Visual fields were not tested in HANES. In the Framingham Eye Study, visual field defects were probably missed in a substantial proportion of the cases, in part because the sensitivity of the screening criteria was well below 100 per cent (24) and in part because appreciable numbers of suspects failed to attend the visual field examination (19). The visual field examination was scheduled for a later date because the subjects' pupils had been dilated

9 METHODOLOGICAL PROBLEMS IN EYE DISEASE EPIDEMIOLOGY 59 during the screening examination. The experiences of both these studies point to the need for the development of a rapid visual field screening test suitable for epidemiologic surveys of glaucoma, a test that can be performed on all subjects whether or not their pupils have been dilated. The recently developed automated perimeters (56), discussed below, may have the required properties. Absence of standardized disease definitions and classifications. Here is an area where appropriate instrumentation can be helpful. We have cited fundus photography in the classification of diabetic retinopathy and senile macular degeneration as examples. Some recent work in densitometric measurement of lens photographs (57), discussed below, may pave the way toward a suitable photographicdensitometric classification of cataract. Objectivity and reproducibility of observations. Observations recorded by instruments (possibly computer-assisted), such as cameras, automated refractors, or densitometers, may be more objective and therefore more reproducible than direct observations by a human observer. There is an acute need for the development of instruments suitable for measurement of cupping of the disc and of density of lens opacities. Possible technologic advances Automated perimeter. We have mentioned that because the standard visual field test is time-consuming, it has not been used routinely in epidemiologic surveys. Various visual field screening devices have been developed which take less time. Among these, the automated perimeter holds out greater promise for epidemiologic application than a manual perimeter (56) because it reduces the possibility of examiner bias and because of its potential for a) high validity and reliability, b) precise standardization of test procedures, and c) minimal training of technicians to operate it. A screening visual field test on the automated perimeter usually takes less than 10 minutes. Before the automated perimeter is used in epidemiologic studies (its potential value may be greatest for prevalence surveys and case-findings surveys), it should be field-tested for validity and reliability. Densitometric measurement of lens photographs. Hockwin et al. (57) have described a system for densitometric measurement of lens photographs which may provide a useful solution to the problem of objective measurement, reproducibility, classification, and definition of cataract. The camera in this system has greater depth of focus, permitting more accurate and detailed assessment of cataract than conventional slit-lamp photography. As in the case of automated perimetry, the validity and reliability of this method for epidemiologic application need to be tested. Nonophthalmologists as examiners. In recent years, the use of technicians for measuring visual acuity and visual fields has increased. A decade ago, there was some reluctance to accept technicians as visual acuity examiners in the multicenter Diabetic Retinopathy Study, but their proficiency and conscientious execution of a complex examination protocol was soon recognized, and this task was gradually transferred from ophthalmologists to technicians. Technicians tended to be more thorough in adhering to the protocol, and their results more similar to those obtained by an ophthalmologist site visitor than to those of the ophthalmologists (32). Technicians have been used successfully to grade fundus photographs in diabetic retinopathy (58). A recent report describes the use of a nurse practitioner in screening patients for diabetic retinopathy (59). Not enough use has been made of nonophthalmologists as examiners in epidemiologic studies of eye disease. Concerted efforts should be made to explore

10 60 EDERER the use of optometrists, nurse practitioners, physicians' assistants, and technicians in this work. STUDY DESIGN Case-control studies As yet, little information is available about risk factors in the etiology of eye diseases, and this handicaps case-control studies in that investigators may easily fail to collect necessary information on possibly confounding factors. Patients as controls. A convenient source of both cases and controls for case-control studies is a clinic or office practice. Drawing the controls from the same source as the cases provides convenience as well as a possible theoretic benefit: the cases and controls can often be presumed drawn from the same segment of the population and, therefore, matched on social, economic, ethnic, and environmental factors. Using patients as controls can, however, lead to bias (confounding) when the prevalence of the "exposure" in the control group differs from that in the segment of the general population from which the cases are drawn. For example, we can envision a negative association with diabetes emerging from a case-control study of senile macular disease if a large proportion of the controls have conditions associated with diabetes, such as diabetic retinopathy, cataract, or glaucoma. Although this source of possible bias is not unique to studies of eye diseases, it has recently come up in several case-control studies of eye disease. Two case-control studies of senile macular degeneration, both drawing cases and controls from a single office practice, have reported an association between the disease and hyperopia (60, 61). Neither report discussed the possibility that excessive myopia among the controls might have contributed to the reported association. Myopia is a common reason for seeking eye care. In a case-control study of 284 cases of open angle glaucoma and 673 controls at the Oxford Eye Hospital in England, Steinmann (62) took steps to assess possible bias from the use of patients as controls. First, when an exposure under study was known to be associated with an ophthalmic condition of the controls, the controls with that condition were excluded and the analysis was repeated (e.g., controls with diabetic retinopathy were excluded when the association with diabetes was assessed). Second, having identified, among the controls, 13 diagnostic entities with at least 10 cases each, the investigators repeated the analysis for association between glaucoma and all exposures 13 times, each time excluding one of the 13 diagnostic entities. Finding no essential changes in results from this analysis provided some assurance of absence of bias. This method is not devoid of difficulties. For example, when controls with diabetic retinopathy are excluded, cases with diabetic retinopathy would also have to be excluded, and these exclusions lessen the power to detect an association with diabetes. In summary, drawing controls from the same practice as the cases has the advantages of convenience and of a certain kind of matching on socioeconomic, ethnic, and environmental factors. At the same time, it has the disadvantage of incurring possible bias, which, when corrected for, may reduce the power to detect an association. Finding enough controls. We have mentioned that lens opacities are present in more than 50 per cent of persons aged 65 years or more, and this presents a problem in finding enough controls for case-control studies of cataract. In persons aged 75 or more, this percentage is as high as 92 (13). If the case-control investigators stratify on age (as they should), they may have to screen as many as 11 persons aged 75 or more to find one disease-free control in that age range. The definition of cataract usually includes a visual acuity criterion. Adding that criterion substantially reduces the

11 METHODOLOGICAL PROBLEMS IN EYE DISEASE EPIDEMIOLOGY 61 prevalence of cataract (e.g., 32 per cent in persons aged 75 or more with acuity of 20/ 30 or worse (13)). Therefore, if controls are defined as persons without cataract (but not necessarily without lens opacity), the problem of finding enough controls is less severe. However, it should be noted that these controls are not free of the disease. Nevertheless, given the difficulty of the problem, this way of defining controls may be reasonable because lens opacities without visual acuity impairment represent a less severe form of the disease. The reasonableness of this approach derives some support from the finding that in two populations under age 65 the relative risk of diabetics having senile lens changes was much higher for cataractous than for precataractous lens changes (13). The problem of finding disease-free controls also confronts case-control investigations of senile macular degeneration. The Framingham Eye Study found senile macular changes present in 40 per cent of persons aged years and 46 per cent of persons aged 75 or more (19). Although the problem is not as severe as for cataract, it does mean that, as in case-control studies of cataract, investigators should examine the controls in detail to determine their disease status. Hyman et al. (63), in their case-control study, photographed the maculae of all "probable" cases and controls and had the photographs independently graded, according to standard photographs, by two masked readers. They found that many of the probable controls had some drusen and that many of the probable cases had too few drusen to be classified as definite cases; using information from the graded photographs, they excluded 66 of 228 probable cases and 62 of 237 probable controls from their main analysis (46, 63). Cohort studies There have not been any cohort studies of eye disease. Although cohort studies have the same objective as case-control studies, namely to study the association between a disease and specified exposures, it is usually easier in cohort than in case-control studies 1) to determine whether the exposure anteceded the disease, and 2) to quantify the exposure. Furthermore, while case-control studies are usually limited to the study of a single disease, cohort studies readily lend themselves to the study of several diseases simultaneously. Cohort studies usually require large study populations and take years to complete particularly when, as is true for most eye diseases, disease incidence is low. However, opportunities appear from time to time to "piggy-back" eye studies onto cohort studies of other diseases. For example, the subjects of the Framingham Heart Study, a cohort study, were surveyed for eye disease in (19-21), establishing the foundation for a future cohort study of this population with a wealth of ocular, medical, and laboratory information for the study of associations. Opportunities for cohort studies, such as was provided by the Framingham Heart Study, should be sought out and exploited. Clinical trials Clinical trials of bilateral eye disease which compare two local (e.g., surgical) treatments, or a local treatment with no treatment, have the unusual opportunity to incorporate a highly efficient design. When both eyes are eligible for study, one eye can be randomly chosen for one treatment, while the other eye can serve as control. The availability for study of two eyes of each patient reduces the number of patients needed by a factor of two. Sample size can then usually be further reduced because of positive correlation between paired eyes (64, 65). This design is not only efficient but also advantageous to the patient. In the case of a treatment that impairs vision, for example, the design "gives each patient the best chance of maintaining vision in at least one eye" (66). If the new treatment is of value, each

12 62 EDERER patient has the opportunity to benefit from it; if it is harmful, damage will be limited to one eye. DATA ANALYSIS: SHALL WE COUNT SUBJECTS OR EYES? When data are collected on both eyes, data analysis presents a problem not usually encountered in epidemiologic studies. The problem stems from the fact that the paired observations made on a person's two eyes are generally not independent, but positively, often highly, correlated. The commonly used methods of statistical inference apply only to independent (uncorrelated) observations. In observational studies (e.g., surveys, case-control studies, cohort studies, natural history studies), in which the data analysis generally concerns the sum or average of paired measurements, positively correlated measurements contribute less information than uncorrelated measurements. In experimental studies (controlled clinical trials) of two local (e.g., surgical) treatments, in which the statistic of interest is the difference between paired measurements, positively correlated measurements provide more information than uncorrelated measurements (64). In the extreme case of perfect positive correlation (r = 1), the second eye adds no information to an observational study. For observational studies, Rosner (67) has presented methods of adjusting the common statistical tests for correlation between paired measurements. These methods can be applied either to studies in which all persons studied contribute information on both eyes or to studies in which some persons contribute information on both eyes and some on only one eye. The need for adjustment can be obviated by limiting the results to only one set of eyes (e.g., left eyes, right eyes, worse eyes, better eyes) or by presenting results separately for the two sets of eyes. When the correlation is high, this obviation loses little information (64). Milton and Ederer (68) have presented a method of adjusting the variance of the difference between paired, correlated measurements in experimental studies. Their presentation also provides for studies in which some observations are paired and some unpaired ("partially paired designs"). Seigel and Podgor (69), addressing the problem of testing paired experimental data for survival differences, have described a sign test for truncated data. The test, a special case of Mantel's "log-rank" method (70), makes no parametric assumptions about the shapes of the survival curves. In addition to a significance test, a nonparametric method for constructing a confidence interval for paired experimental survival data is needed. SUMMARY AND COMMENTS The recent burst of activity in eye disease epidemiology, occasioned by increases in the number of ophthalmologists with training in epidemiology and of epidemiologists interested in eye diseases,' makes this an appropriate time for taking stock of methodological problems in this field. In this review, based largely on the author's personal encounter with five large epidemiologic studies (2, 7, 19, 30, 39) of the four major chronic eye diseases (senile cataract, senile macular degeneration, diabetic retinopathy, and glaucoma), the following methodological areas are discussed: quality assurance; disease definitions and classifications; reproducibility of observations; visual acuity and visual field tests; needed technologic advances; study design of case-control studies and clinical trials; and data analysis (shall we count subjects or eyes?). The chief weakness of blindness registers has been inadequate evaluation of the

13 METHODOLOGICAL PROBLEMS IN EYE DISEASE EPIDEMIOLOGY 63 extent of underregistration and of the variability of underregistration according to socioeconomic and racial subgroups. The Health and Nutrition Examination Survey (HANES), conducted by the United States Public Health Service, is a potentially valuable source of descriptive and analytic epidemiologic information on eye diseases. It is an examination of a large probability sample of the US population and includes various histories, clinical and laboratory examinations, and a nutrition interview. In the early 1970s, HANES included eye examinations, but these were curtailed because of inability to get ophthalmologists as examiners. In future HANES studies, efforts should be made to minimize the involvement of ophthalmologists in the eye examinations and to increase the use of nonophthalmologists (e.g., optometrists, technicians) and instruments (e.g., cameras, densitometers, automated perimeters). The validity and reliability of any new procedures will need to be determined. Studies may need to be limited to certain major eye diseases and conditions. Population glaucoma surveys have tended to omit partially or entirely the time-consuming visual field test. Yet, without this test, the diagnostic examination for glaucoma is incomplete. The automated perimeter, with which visual fields can be more quickly measured, promises to be an important screening tool for glaucoma prevalence and case-finding surveys. It should be field-tested for epidemiologic use. The absence of standardized disease definitions, without which comparisons between studies are uncertain, has been a handicap to epidemiologic research in eye disease. A first constructive step toward the development of disease definitions would be the development of disease classifications. This has been done, with standard photographs, for diabetic retinopathy and senile macular degeneration. A recently developed densitometric measurement of lens photographs could lead to a reproducible cataract classification. A standardized classification and definition is needed for open angle glaucoma. A number of studies have found that certain ophthalmic observations or measurements are poorly reproducible, and among these the cup-disc ratio, an element in the diagnosis of open angle glaucoma, is prominent. The subjectivity of many ophthalmic observations no doubt contributes to this problem and emphasizes the need for standardized procedures, written examination protocols, standard photographs, thorough examiner training, and the development of objective measures. Ophthalmic observations made by instruments rather than ophthalmologists could reduce the ophthalmologist's workload and inconvenience to the subject. Recorded observations made by instruments (e.g., photographs, digital computer output) may facilitate objectivity, the performance of replicate examinations, and the development of standardized classifications and definitions. The most important measures of visual function are those of visual acuity and visual field. Both tests are subjective and therefore liable to subject and examiner bias. In clinical trials, prevention of one or both biases may be particularly important and also feasible through single or double masking or by having the examiner encourage the patient to a maximum effort. The visual acuity measurement is influenced by a number of physical variables (e.g., room and eye chart illumination, black-white contrast of letters on eye chart, reading difficulty of letters). Standardization of this measurement has been facilitated by recently developed eye charts and visual acuity boxes.

14 64 EDERER In case-control studies of eye disease, paucity of information about risk factors is a handicap in that investigators may fail to collect necessary information on possibly confounding factors. Use of patients as controls can lead to confounding, and special preventive steps are recommended. Because of the high prevalence in elderly people of lens opacities and drusen, finding enough controls is likely to be a problem for case-control studies of cataract and senile macular degeneration; possible solutions are discussed. Clinical trials of bilateral eye disease which compare two local treatments, or a local treatment with no treatment, have the opportunity to use a highly efficient design: one eye can be randomly chosen for treatment and the other can serve as control. When data are collected on both eyes in epidemiologic studies, the fact that the paired observations on a person's two eyes are generally (often highly) positively correlated presents a special problem in statistical inference. Methods of handling this problem are discussed. REFERENCES 1. Ederer F. The National Eye Institute Symposium on the Epidemiology of Eye Diseases and Visual Disorders: Introduction. Am J Epidemiol 1983;118:130-l. 2. Goldstein H, Goldberg ID. The Model Reporting Area for Blindness Statistics. Sight Sav Rev 1962;32: Ederer F. The Framingham Eye Study. In: Blindness , Vol 13. Washington, DC: American Association of Workers for the Blind, 1979: Kahn HA, Moorhead HB. Statistics on Blindness in the Model Reporting Area (DHEW publication no. (NIH) ). Washington, DC: US GPO, Kahn HA, Hiller R. Blindness caused by diabetic retinopathy. Am J Ophthalmol 1974;78: Hiller R, Kahn HA. Blindness from glaucoma. Am J Ophthalmol 1975;80: Miller HW. Plan and operation of the Health and Nutrition Examination Survey, United States, National Center for Health Statistics. (Vital and Health Statistics, Series 1, no. 10a) (DHEW publication no. (HRA) ). Washington, DC: US GPO, Roberts J, Ludford J. Monocular visual acuity of persons 4-74 years, United States, National Center for Health Statistics. (Vital and Health Statistics, Series 11, no. 201) (DHEW publication no. (HRA) ). Washington, DC: US GPO, March Roberts J, Ludford J. Eye examination findings among youths years, United States, National Center for Health Statistics. (Vital and Health Statistics, Series 11, no. 155) (DHEW publication no. (HRA) ). Washington, DC: US GPO, November Roberts J, Rowland M. Refraction status and motility defects of persons 4-74 years, United States, National Center for Health Statistics. (Vital and Health Statistics, Series 11, no. 206) (DHEW publication no. (HRA) ). Washington, DC: US GPO, August Ganley JP, Roberts J. Eye conditions and related need for medical care among persons 1-74 years, United States National Center for Health Statistics. (Vital and Health Statistics, Series 11, no. 228) (USPHS publication no ). Washington, DC: US GPO, Hiller R, Giacometti L, Yuen K. Sunlight and cataract: an epidemiologic investigation. Am J Epidemiol 1977;105: Ederer F, Hiller R, Taylor HR. Senile lens changes and diabetes in two population studies. Am J Ophthalmol 1981;91: Klein BE, Klein R. Intraocular pressure and cardiovascular risk variables. Arch Ophthalmol 1981;99: Hiller R, Sperduto RD, Krueger DE. Race, iris pigmentation, and intraocular pressure. Am J Epidemiol 1982;115: Sperduto RD, Seigel D, Roberts J, et al. The prevalence of myopia in the United States. Arch Ophthalmol 1983;101: Hiller R, Sperduto RD, Ederer F. Epidemiologic associations with cataract in the National Health and Nutrition Examination Survey. Am J Epidemiol 1983;118: Kahn HA, Leibowitz H, Ganley JP, et al. Standardizing diagnostic procedures. Am J Ophthalmol 1975;79: Leibowitz HM, Krueger DE, Maunder LR, et al. The Framingham Eye Study Monograph. Surv Ophthalmol 1980;24: Kahn HA, Leibowitz HM, Ganley JP, et al. The Framingham Eye Study. I. Outline and major prevalence findings. Am J Epidemiol 1977; 106: Kahn HA, Leibowitz HM, Ganley JP, et al. The Framingham Eye Study. II. Association of ophthalmic pathology with single variables previously measured in the Framingham Heart Study. Am J Epidemiol 1977;106: Krueger DE, Milton RC, Maunder LR. The Framingham Eye Study: introduction to the monograph. Surv Ophthalmol 1980;24: Kahn HA, Milton RC. Revised Framingham Eye Study prevalence of glaucoma and diabetic retinopathy. Am J Epidemiol 1980;lll: Kahn HA, Milton RC. Alternative definitions of open-angle glaucoma. Effect on prevalence and

15 METHODOLOGICAL PROBLEMS IN EYE DISEASE EPIDEMIOLOGY 65 associations in the Framingham Eye Study. Arch Ophthalmol 1980;98: Sperduto RD, Seigel D. Senile lens and senile macular changes in a population-based sample. Am J Ophthalmol 1980;90: Sperduto RD, Hiller R, Seigel D. Lens opacities and senile maculopathy. Arch Ophthalmol 1981;99: Hiller R, Sperduto RD, Krueger DE. Pseudoexfoliation, intraocular pressure and senile lens changes in a population-based survey. Arch Ophthalmol 1982;100: Seigel D, Sperduto RD, Ferris FL. Aspirin and cataracts. Ophthalmology 1982;89:47A-49A. 29. Leske MC, Podgor MJ. Intraocular pressure, cardiovascular risk variables, and visual field defects. Am J Epidemiol 1983;118: The Diabetic Retinopathy Study Research Group. Preliminary report on effects of photocoagulation therapy. Am J Ophthalmol 1976;81: The Diabetic Retinopathy Study Research Group. Diabetic Retinopathy Study. Design, methods, and baseline results. Report no. 6. Invest Ophthalmol Vis Sci 1981;21(1, pt 2): Ferris FL, Ederer F. External monitoring in multiclinic trials. Applications from ophthalmologic studies. Clin Pharmacol Ther 1979; 25: The Diabetic Retinopathy Study Research Group. Photocoagulation treatment of proliferative diabetic retinopathy. The second report of Diabetic Retinopathy Study findings. Ophthalmology 1978;85: The Diabetic Retinopathy Study Research Group. Four risk factors for severe visual loss in diabetic retinopathy. The third report from the Diabetic Retinopathy Study. Arch Ophthalmol 1979;97: The Diabetic Retinopathy Study Research Group. Photocoagulation treatment of proliferative diabetic retinopathy. A short report of long range results. Diabetic Retinopathy Study (DRS). Report no. 4. In: Diabetes Proceedings of the 10th Congress of the International Diabetes Federation. Amsterdam: Excerpta Medica 1980; The Diabetic Retinopathy Study Research Group. Photocoagulation treatment of proliferative diabetic retinopathy. Relationship of adverse treatment effects to retinopathy severity. Diabetic Retinopathy Study (DRS). Report no. 5. In: Modern problems in ophthalmology Proceedings of the meeting of Club Jules Gonin. Dev Ophthalmol 1981;2: The Diabetic Retinopathy Study Research Group. A modification of the Airlie House classification of diabetic retinopathy. Report no. 7. Invest Ophthalmol Vis Sci 1981;21(1, pt 2): The Diabetic Retinopathy Study Research Group. Photocoagulation treatment of proliferative diabetic retinopathy. Clinical application of Diabetic Retinopathy Study (DRS) findings. Report no. 8. Ophthalmology 1981;88: VAIS Manual of Procedures. Office of Biometry and Epidemiology, National Eye Institute, National Institutes of Health, Bethesda, MD, Final report of the VAIS pilot study. Office of Biometry and Epidemiology, National Eye Institute, National Institutes of Health, Bethesda, MD (manuscript). 41. Leske MC. The epidemiology of open-angle glaucoma: a review. Am J Epidemiol 1983; 118: Leske MC, Sperduto RD. The epidemiology of senile cataracts: a review. Am J Epidemiol 1983;118: National Center for Health Statistics. HANES Examination Staff Procedures Manual for the Health and Nutrition Examination Survey, National Center for Health Statistics. US DHEW, Public Health Service, Health Services and Mental Health Administration. Washington, DC: US GPO, June Ferris FL III. Senile macular degeneration: review of epidemiologic features. Am J Epidemiol 1983;118: Davis MD, Norton EWD, Myers FL. The Airlie House classification of diabetic retinopathy. In: Symposium on the Treatment of Diabetic Retinopathy. Goldberg MF, Fine SL, eds. (USPHS publication no. 1890:7-22). Washington, DC: US GPO, Hyman LG, Ferris FL, Fine SL, et al. Senile macular degeneration: an epidemiologic investigation (manuscript). 47. Lichter PR. Variability of expert observers in evaluating the optic disc. Trans Am Ophthalmol Soc 1976;74: Schwartz JT. Methodologic differences and measurement of cup-disc ratio. Arch Ophthalmol 1976;94: Brilliant LB, Lepkowski JM, Musch DC. Reliability of ophthalmic diagnoses in an epidemiologic survey. Am J Epidemiol 1983;118: Fleiss J. Statistical methods of rates and proportions. New York: John Wiley & Sons, Ederer F. Patient bias, investigator bias and the double-masked procedure in clinical trials. Am J Med 1975;58: Aiello LM, Berrocal J, Davis MD, et al. The Diabetic Retinopathy Study (Letter). Arch Ophthalmol 1974;92: NAS-NRC Committee on Vision. Recommended standard procedures for the clinical measurement and specification of visual acuity. Adv Ophthalmol 1980;41: Ferris FL, Kassoff A, Bresnick GH, et al. New visual acuity charts for clinical research. Am J Ophthalmol 1982;94: Ferris FL, Sperduto RD. Standardized illumination for visual acuity testing in clinical research. Am J Ophthalmol 1982;94: Keltner JL, Johnson CA, Balestrery MS. Suprathreshold static perimetry: initial clinical trials with the Fieldmaster automated perimeter. Arch Ophthalmol 1971;97: Hockwin O, Dragomirescu V, Koch HR. Photographic documentation of disturbances of the lens transparency during ageing with a Scheimpflug camera system. Ophthalmic Res

16 66 EDERER 1979;ll:405-ll. 58. Milton RC, Ganley JP, Lynk RH. Variability in grading diabetic retinopathy from stereo fundus photographs: comparison of physician and lay readers. Br J Ophthalmol 1977;61: Sackett CS, Ferris FL. Screening for diabetic retinopathy in a diabetic management clinic. J Ophthalmol Nurs Technol 1982;1: Maltzman BA, Mulvihill MN, Greenbaum A. Senile macular degeneration and risk factors: a case-control study. Ann Ophthalmol 1979; 11: Delaney WV, Oates RP. Senile macular degeneration: a preliminary study. Ann Ophthalmol 1982;14: Steinmann WC. Case-control study of the risk factors for open angle glaucoma (manuscript). 63. Hyman LG, Lilienfeld AM, Ferris FL in, et al. Senile macular degeneration: a case-control study. Am J Epidemiol 1983;118: Ederer F. Shall we count number of subjects or number of eyes? Editorial. Arch Ophthalmol 1973;89:l Kahn HA. Letter. Invest Ophthalmol 1974; 13: Davis MD. Application of the principles of clinical trials. Am J Ophthalmol 1975;79: Rosner B. Statistical methods in ophthalmology: an adjustment for the intraclass correlation between eyes. Biometrics 1982;38: Milton RC, Ederer F. Partially paired designs. Biometrics note no. 5. Office of Biometry and Epidemiology, National Eye Institute, National Institutes of Health, Bethesda, MD, May Seigel D, Podgor M. A sign test for survivorship curves. Controlled Clin Trials 1982;3: Mantel N. Evaluation of survival data and two new rank order statistics arising in its consideration. Cancer Chemother Rep 1966;50(3):