Who Participates in Population Based Studies of Visual Impairment? The Salisbury Eye Evaluation Project Experience ˜ OZ, MSC, SHEILA WEST, PHD, GARY S. RUBIN, PHD, OLIVER D. SCHEIN, MD, BEATRIZ MUN LINDA P. FRIED, MD, KAREN BANDEEN-ROCHE, PHD, AND SEE PROJECT TEAM
PURPOSE: To describe characteristics of respondents and nonrespondents to a home questionnaire and comprehensive clinical examination in a population of elderly Americans. METHODS: The SEE project is a population-based study of age-related eye diseases, visual impairment, and functional status of individuals aged 65 to 84. Potential participants were identified using the Health Care Financing Administration Medicare data base for Salisbury, MD: After sending out an introductory letter, a trained interviewer visited potential participants in their homes, obtained their consent to participate, administered a short screening interview that included questions about their general health and vision, and administered an extensive questionnaire on their diet, medical history, and difficulty performing activities related to vision. The interviewer then scheduled an appointment for the participants to visit a central site for an exam. Potential participants could fall into one of two refusal groups; refusal to take part in the study before the home questionnaire or prior to the clinic visit. RESULTS: The overall response rate for the clinic visit was 65%. Compared to individuals with complete exams, the two groups of refusals were older, less likely to have any college education, more likely to report poor health status, and more likely to need help with Independent Activities of Daily Living (IADL’s). Participants with complete home questionnaires that failed to come to the clinic were more likely to have Mini-Mental scores below 25. There were not significant differences by race, gender, or self reported vision status among the three groups. CONCLUSIONS: Population-based studies requiring an in-clinic examination may selectively undersample those with health and mental difficulties. These differential responses may introduce bias in the study results and need to be addressed when assessing the burden, type and severity of disease in the community. However, self-report of visual status was similar among refusals and participants in this study on vision. Ann Epidemiol 1999;9:53–59. 1998 Elsevier Science Inc. Population-Based, Visual Impairment, Disease Prevalence, Differential Response Rates, Visual Function, Functional Status.
KEY WORDS:
INTRODUCTION The United States population is aging; according to the 1990 census 12.6% of the population is 65 or older, with more than three million people over the age of 85 (1). In order to provide effective and timely services, it is imperative to understand the health-related issues that affect this growing sector of the population. The health status of older individuals, factors that are associated with the most common diseases in the elderly, and the implications of these different health outcomes on the ability of these individuals to func-
From the Dana Center for Preventive Ophthalmology (B.M., S.W., O.D.S.), Lions Vision Research Center (G.S.R.), Department of Biostatistics (K.B.-R.), and Departments of Medicine and Epidemiology (L.P.F.), John Hopkins University, School of Medicine, School of Hygiene and Public Health. Address reprint requests to: B. Mun˜oz, Wilmer 118, 600 North Wolfe Street, Baltimore, MD 21287–9019. Received July 10, 1997; revised March 25, 1998; accepted March 31, 1998. 1998 Elsevier Science Inc. All rights reserved. 655 Avenue of the Americas, New York, NY 10010
tion independently are priorities for gerontological research. In addition, knowing the health status and the needs of this important sector of the population will help in the rational planning of health services. Population-based studies can provide this valuable information best when the population studied is truly representative of the general population. Once a representative sample of the target population has been selected, only a proportion of the group of eligible individuals will agree to participate and be successfully enrolled in a study. To be able to generalize study findings to the target population, the characteristics of participants and nonparticipants should be similar with respect to factors associated with the outcome of interest. In general, participation rates are lower in older age groups, and reasons for refusal are more likely to be related to poor health (2). Although data on nonparticipation are available for studies on general health outcomes such as data collected for National Health and Nutrition Examination Survey (NHANES) (3) and Cardiovascular Health Study (CHS) (4), no data are 1047-2797/99/$–see front matter PII S1047-2797(98)00026-X
54
Mun˜oz et al. RESPONSE RATES IN POPULATION-BASED STUDIES OF VISION
Selected Abbreviations and Acronyms SEE 5 Salisbury Eye Evaluation MMSE 5 Mini-Mental State Exam HCFA 5 Health Care Financing Administration ADL 5 Activities of daily living IADL 5 independent activities of daily living
available on differential response rates according to factors that may be related to vision. A description of differential participation in a population-based study of vision in an elderly population is important, not only to understand the generalizability of the study results, and to measure how differential participation may affect study conclusions, but to help to identify barriers that limit participation of older people. The purpose of this study is to describe the overall recruitment experience, and the characteristics of elderly respondents and nonrespondents from a population of older Americans, to a study of visual impairment, which included a home questionnaire and a comprehensive clinical and visual examination.
METHODS The Salisbury Eye Evaluation (SEE) is a population-based study of age-related eye diseases, visual impairment, and functional status of non-institutionalized individuals aged 65–84 years. The three main objectives of the study are: first, to look at the impact of age-related eye disease on visual function; second, to identify risk factors for three of the most common eye diseases in older people: cataract, macular degeneration and dry eye; and third, to assess the impact of visual impairment on adverse health outcomes such as physical disability, falls, and nursing home admissions. To meet the objectives of the study, data were collected from a population-based sample of Salisbury residents aged 65 to 84 years, in Salisbury, Maryland. All procedures for the study were reviewed by the Institutional Review Board of Johns Hopkins Medical Institutions. Following the tenets of the Helsinki Accord, written informed consent was obtained. To be eligible for the study sampling frame, potential participants must have been between ages 65 and 84, and residing in selected zip codes from the Salisbury metropolitan area as July 1, 1993. Once contact with the potential participants was made, study eligibility was determined using the following additional criteria: participants must be non-institutionalized, able to communicate in English, able to travel to a central site for a complete exam, and score 18 or more in the Mini-Mental State Exam (MMSE) (5).
AEP Vol. 9, No. 1 January 1999: 53–59
The steps followed to select the sample are shown in Figure 1. A complete list of all individuals between the ages of 65 and 84 years old residing in the Salisbury metropolitan area, Salisbury, Maryland, was obtained from the Health Care Financing Administration (HCFA) Medicare data base, using specific zip codes. Seven thousand and four individuals, between the ages of 65 and 84 as of July 1, 1993, were identified as living in the selected zip codes. A total of 1292 (18%) were African-American. In order to achieve a final enrollment of at least 2500 participants with complete baseline information, a total of 4624 persons were randomly selected from the list of 7004 identified in the HCFA data-base. A detailed description of the sampling scheme has been reported elsewhere and is outlined here (6). To be able to recruit enough subjects from both races to fulfill the study goals, all 1292 AfricanAmericans, and an age-stratified random sample of the 5713 whites were selected. For whites, a higher proportion of individuals were selected in the older age bracket (62% of the 75 to 84 years-olds and 56% of the 65 to 74 years-olds) because of the expected lower response rates in these age groups. The demographic characteristics of the sample selected are shown in Table 1. Seventy two percent were white, 60% female, 33% were aged 65 to 69 years, 52% were aged 70 to 80 years, and 15% were aged 80 to 84 years. Contact with the potential participant was initiated with letters. An introductory letter, required by HCFA, describing the study aims and request for participation, was mailed to each person in the selected sample. Subsequently, a second letter was sent by the study team with a more detailed description of the project goals and eligibility requirements. These letters were staggered over the course of the study, so that time between receipt of letters and contact by an interviewer was minimized. Due to constraints in the number of individuals that could be examined in the clinic (up to 7 per day) a maximum of 20 months elapsed between sample selection contact with sampled individuals. The selection of persons for contact at different time points was based on random allocation. The following set of procedures was used to recruit participants; procedures common to population based studies. Within two months of the second letter, potential participants were contacted in their homes by a trained interviewer. At least six attempts to contact were made before an individual was classified as ‘unable to locate’ and these frequently involved addresses as Post Office Boxes. Once the participant was contacted, a short questionnaire (screener) with basic demographic information and ten summary questions related to the main outcomes of the study, and the MMSE was administered (5). Eligibility was assessed at this point. If eligible, a standardized home questionnaire was administered to the participant, which, included sections on self-reported vision function, medical history, food frequency questionnaire, use of alcohol, and cigarette smoking.
AEP Vol. 9, No. 1 January 1999: 53–59
Mun˜oz et al. RESPONSE RATES IN POPULATION-BASED STUDIES OF VISION
55
FIGURE 1. Sample selection and final states of the selected sample.
After completion of the interview, an appointment for a comprehensive exam at the study central site was scheduled. Transportation to and from the clinic was offered to all participants. The clinic exam had an average duration of four hours, with a maximum of five hours. The clinic visit included an ophthalmologic exam, ocular photographs, visual function testing, dry eye testing, selfreported and performance-based measures of function, risk
TABLE 1. Demographic characteristics of the sample selected Age Group 65–69 70–74 75–79 80–84 Total
Overall
% Total
% Female
% AfricanAmericans
1529 1358 1044 693 4624
33.1 29.3 22.6 15.0 100.0
54.4 59.1 61.3 68.5 59.5
29.7 28.8 25.3 26.3 27.9
factors questionnaires, questions on balance and falls, blood extraction for diabetic testing and levels of various micro nutrients, and assessment of medical conditions, other than visual problems, that could affect function. Individuals failing to fulfill the eligibility criteria were designated as “ineligible.” After any initial refusal, extensive conversion efforts were made. Only after at least six unsuccessful conversion attempts were individuals classified as refusals. These conversion attempts included contact at different hours of the day and different days of the week, by experts in refusal conversion techniques. Those who refused to both the home questionnaire and clinic exam were designated as “refusals.” Information on eligibility could not be obtained from refusals, and it is likely that some would have been ineligible. Participants who completed the home questionnaires but failed to come to the clinic visit were designated as “home questionnaire only.” Participants who completed the clinic visit were designated as “complete exams.” To characterize those who refused to participate, during the conversion effort the group of total refusals was asked
56
Mun˜oz et al. RESPONSE RATES IN POPULATION-BASED STUDIES OF VISION
TABLE 2. Screener questions asked to refusals Question description 1. What was the highest grade in school that you completed? 2. Are you now married, living together with someone, widowed, divorced, separated, or have you never married? 3. Are there any other adults living in your household? 4. At the present time, would you say your health is: excellent, very good, good, fair, or poor? 5. How would you rate your current vision with your glasses on, if you wear glasses, on a scale of 0 to 10, with 10 being excellent and 0 being blindness? 6. Have you fallen in the last 12 months? This includes falling to the ground or to another level, such as into a chair. 7. In the one occupation you held longest in life, including homemaking, were you outside in the daylight for more than 2 hours per day in the summer time? 8. Do you feel a gritty or sandy sensation in your eyes? 9. Because of any physical or health problem, do you need the help of other persons with your present care needs, such as eating, bathing, dressing, or getting around the home? 10. Because of any physical or health problem, do you need the help of other persons in handling your routine needs, such as everyday household chores, doing necessary business, shopping, or getting around for other purposes?
to answer at least the ten questions in the screener (see Table 2) that were related to the main outcomes of the study. This paper reports participation rates and describes the characteristics of ineligibles, total refusals, and those with home questionnaire only and compares their known characteristics with those of individuals who had complete exams. Figure 1 outlines the status of all the sampled individuals. Data Analysis Age at the time of the sample selection (July 1st, 1993) and answers to the 10 questions (Table 2) from the screener (for the ‘complete exams’ and ‘home questionnaire only’) and from the short interview (for the refusals) were used to characterized the three groups. Contingency tables were used to compare characteristics of the groups (ineligibles,
AEP Vol. 9, No. 1 January 1999: 53–59
total refusals, home questionnaire only and individuals with complete exams). In multi-variate analysis, logistic regression models were used to identify the independent predictors of being a refusal or a home questionnaire only when compared with the group with a complete clinic exam.
RESULTS The demographic characteristics of eligibles, questionable eligibles (total refusals and unable to locate), and ineligibles are shown in Table 3. Compared to the group of eligibles, ineligibles tended to be older, for an increase of one year in age the gender and race adjusted odds ratio (OR) was 1.10; with a 95% confidence interval (CI) of 1.08–1.11. Ineligibles were also more likely to be African American (OR, 1.54; CI, 1.29–1.85). Eligibles and questionable eligibles had similar demographic characteristics. Reasons for ineligibility are shown in Table 4. Nearly one third of the ineligibles (or 5.4% of the identified individuals) were dead before contact was made. Almost one quarter of the group (or 3.8% of the identified sample) had moved out of the study area. Seventeen percent of the ineligibles (or 2.7% of the initial sample) were institutionalized. Only 14% of the ineligibles were excluded on the basis of the mini-mental examination score. The 935 total refusals can be categorized into two groups: those who answered the short screener questions (N 5 477) and those who refused even this level of participation (N 5 458). Those who answered the short interview were similar in age, race, and gender to those who did not answer the short interview. Therefore, inferences made from the refusals with answers to the 10 questions were applied to the total refusal group. Demographic characteristics of the two groups of refusals and of individuals with complete exams are shown in Table 5. Compared with the group with complete exams, total refusals were older (OR, 1.02; CI 1.01–1.08) and more likely to be females (OR, 1.23; CI 1.06–1.44). Those who completed the home interview only were older (OR, 1.06; CI, 1.03–1.08).
TABLE 3. Demographic characteristics of the selected sample by eligibility status Eligible
Questionable Eligiblea
Ineligibles
Number of people % Female
2866 58.2
1020 62.5
718 60.2
Age group 65–69 70–74 75–79 80–84
% 35.7 30.8 21.3 12.2
% 35.3 27.9 21.7 15.1
% 19.4 25.8 29.0 25.9
% African-Americans
26.5
27.1
34.8
Characteristics
a
Refusals (N 5 935), unable to locate (N 5 85).
x2(2) 5 5.1 p , 0.001 x2(6) 5 141.3 p , 0.001 x2(2) 5 20.1 p , 0.001
AEP Vol. 9, No. 1 January 1999: 53–59
Mun˜oz et al. RESPONSE RATES IN POPULATION-BASED STUDIES OF VISION
TABLE 4. Reasons for ineligibility
57
DISCUSSION N 5 718
Description Deceased Out of the area Institutionalized Insufficient score on minimental Too ill Other
34.8% 24.2% 17.7% 14.4% 3.9% 7.1%
Answers to the short interview were compared among total refusals who answered the short interview, those with home questionnaire only, and those with complete exams. Table 6 shows the distribution of answers to the ten questions (presented in Table 2), stratified by group. Compared to the other two groups, participants were more likely to have more than twelve years of education, be married, have other adults living in the house, report better health, require less assistance with activities of daily living (ADL’s) and with independent activities of daily living (IADL’s) than the two refusal groups. However, refusals were less likely to grade their vision as poor and less likely to report dry eye symptoms. Table 7 presents the results of the multiple logistic regression model. Interactions of age with education, health status, dry eye symptoms, history of falls, reporting difficulty with ADL’s or IADL’s were explored, none were significant. In addition to the variables shown in Table 7, the final model included gender and race. Those with home questionnaire only and total refusals are compared separately with those who had complete exams. Both refusal groups were more likely to be older, less likely to have college education, more likely to report poor health status, more likely to require assistance with IADLs, and less likely to report dry eye symptoms than were those with complete exams. In addition, total refusals were more likely to rate their vision as good, and less likely to report falls in the last 12 months than were individuals with complete exams. Mini-mental scores were available for those with home questionnaire only and complete exams. Those with home questionnaire only were more likely to have scores under 25 than were those with complete exams.
Almost 16% of the selected sample was found by the interviewers to be ineligible. One third of the ineligibles (5.4% of the selected sample) were dead by the time of contact. Moving to nursing homes or out of the study area accounted for another 42% of the ineligibles (6.5% of the selected sample). A similar recruitment procedure (from the HCFA Medicare list) was used in the CHS (4). Almost 10% of the contacted individuals were ineligible and 6% were confirmed deceased. The higher proportion of ineligibles found in SEE when compared with CHS, may be due to a longer period between selection of the sample and recruitment in our study (two years versus one). As reported by others (2, 4), we found that the Medicare list was an effective and inexpensive method to identify the sample. After excluding the ineligibles, the overall response rate to the home questionnaire was 74%, comparable to rates reported in studies in populations with similar age structure (7). However, the requirement of traveling to a clinic site for a relatively long exam (4-5 hours) reduced the response to the clinic visit to 65% of the eligibles. This is better than other studies requiring a 4 to 5 hour clinic exam (3, 4). It is difficult to compare the recruitment experience of SEE with other population-based studies of vision, as other studies tend to recruit individuals 40 years and older, and response rates for individuals 65 years and older are not readily available. Moreover, the examination times are considerably shorter. The burden to the participant is an important factor in determining response rate. The main focus of other population-based studies of vision has been prevalence of eye disease, with some requiring a comprehensive eye exam only for subjects that fail some screening criteria, so that examination times are about an hour. Response rates in these settings are higher than when a 4 hour exam is necessary for study goals. The Baltimore Eye Survey reported response rate to their screening examination of 79.2% (8) with a decline by age, where 47% of the participants were over age 60. In the Beaver Dam Eye Study, 42.3% of the population was over the age of 65: refusals were more likely to be older and their overall response was 83.1%. The major characteristics of the refusals in the SEE proj-
TABLE 5. Demographic characteristics of participants and refusals Characteristics Number of people % female Age group 65–69 70–74 75–79 80–84 % African-Americans
Complete exams
Home questionn. only
Refusals short interview
Refusals no short interview
2520 57.9 % 36.8 31.3 21.0 10.9 26.4
366 60.9 % 28.1 27.0 23.5 21.3 27.0
477 61.4 % 34.4 28.5 20.3 16.8 23.7
458 65.0 % 34.3 28.2 23.6 14.0 26.4
58
Mun˜oz et al. RESPONSE RATES IN POPULATION-BASED STUDIES OF VISION
AEP Vol. 9, No. 1 January 1999: 53–59
TABLE 6. Comparison of all refusals and those with complete exams
Complete exams
Home questionn. only
Refusals short interview obtained
2520 % 5.0 66.9 28.1
366 % 9.4 73.6 17.1
477 % 6.9 75.5 17.6
57.7 30.9 8.4 3.0 56.8
50.8 39.0 6.3 3.9 50.7
55.3 34.2 7.8 2.7 52.7
10.5 25.0 35.7 25.2 3.5
7.2 15.7 37.2 28.4 11.6
9.9 19.2 32.5 26.2 12.2
2.6 14.6 43.4 39.5 20.4
1.4 18.4 43.9 36.3 24.0
0.7 12.2 43.5 43.7 17.8
41.4
45.2
39.5
27.4
22.4
9.6
Help with ADL
3.9
8.8
7.4
Help with IADL
12.3
22.5
18.7
Minimental score . 24
83.8
69.5
Not done
Characteristics Number of people Education < 5 years 6–12 . 12 Marital status Married Widowed Divorced Never married Other adults living in the house Health status Excellent Very good Good Fair Poor Vision score 0–2 (poor) 3–5 6–8 9–10 (excellent) Falls in the last 12 months > 2 hours of sunlight (summer) Eye gritty or sandy
ect were consistent with what has been found in other population based studies involving older adults (2, 4, 9–13). On average, refusals tended to be older, to have fewer years of education, and to be more likely to report poor health. Looking at the crude response rates, females were less likely to participate than were males. However, in the multivariate analyses, once reported health status and need for assistance with ADLs and IADLs were introduced in the final model, gender was no longer a predictor of refusal. The short interview obtained from refusals enable us to partially characterize nonparticipants. Some differences between participants and refusals related to some of the study outcomes were found. Estimates of these differences can be used to explore differences in inferences to “participating populations” versus “target populations.” Specifically, they provide critical information for statistical methods to correct for missingness (14).
x2(4) 5 45.5 p , 0.0001
x2(6) 5 10.9 p 5 0.09 x2(2) 5 6.7 p , 0.04
x2(8) 5 100.8 p , 0.0001
x2(6) 5 15.0 p 5 0.02 x2(2) 5 5.0 p 5 0.08 x2(2) 5 2.8 p 5 0.24 x2(2) 5 64.3 p , 0.0001 x2(2) 5 24.2 p , 0.0001 x2(2) 5 35.9 p , 0.0001 x2(1) 5 44.1 p , 0.0001
TABLE 7. Predictors of home questionnaire only and complete refusals when compared to individuals with complete examsa Characteristics Age (1 year increment) Education More than 12 years Poor health status Eye feels gritty or sandy History of falls Vision score . 5 Help with ADL Help with IADL Minimental score , 25 a
Home questionn. only
Complete refusals
OR (95% CI)
OR (95% CI)
1.04 (1.01–1.07)
1.02 (1.00–1.04)
0.64 1.61 0.66 1.10 1.21 1.39 1.44 1.78
0.57 1.34 0.29 0.73 1.62 1.68 1.56
(0.47–0.86) (1.22–2.13) (0.50–0.88) (0.83–1.45) (0.89–1.65) (0.84–2.32) (1.02–2.04) (1.34–2.35)
(0.44–0.75) (1.05–1.70) (0.20–0.40) (0.55–0.98) (1.16–2.25) (1.00–2.83) (1.09–2.21) NA
Results of a single multiple logistic regression model, adjusting for gender and race.
AEP Vol. 9, No. 1 January 1999: 53–59
Mun˜oz et al. RESPONSE RATES IN POPULATION-BASED STUDIES OF VISION
The fact that those with complete exams were slightly more likely to grade their vision as poor and to report dry eye symptoms than were refusals, may be an indication that individuals are more likely to participate in a vision project with the understanding that they could benefit from the services the study provides. A complete eye exam was offered in the clinic and referrals to their usual eye care provider were made whenever vision problems were found. Other studies in older populations have found similar trends with refusals reporting less concerns with the main study outcome. For example, in the Rural Health Promotion Project (15), refusals were more likely to report being healthier compared to participants. Refusals may have chosen not to participate because the project targeted risk factors they did not have. Although reporting less vision problems, refusals were more likely to grade their health as poor and to report difficulties with ADL’s and IADL’s, an indication that overall they were more frail than participants. Given these findings, estimates of prevalence of co-morbid conditions, impaired functional status, and of other health-related outcomes based on a recruited sample are likely to represent some underestimation of the true prevalence of these characteristics in the community. Theoretically, studies which recruit older population groups will need to account for differential response when reporting estimates of disease prevalence in the population. Practically, it may make small differences. For example, in Atherosclerosis Risk in Communities (ARIC) study, the effect of nonresponse on the prevalence of disease was examined, and with few exceptions, a 33% nonresponse rate did not seem to introduce significant bias into the prevalence estimates (13). Moreover, in the SEE study, when we adjusted the prevalence of dry eye for nonresponse bias, the estimates changes by less than 0.5% (16) from 14.6 to 15.0. Data such as ours can guide analyses of potential response biases in large, population based studies of older persons. This research was supported by The National Institute of Aging. Grant No P01 AG10184. Dr. West is a Senior Scientific Investigator for Research to Prevent Blindness.
59
REFERENCES 1. 1990 US Census Data. Database: C90STF1C. Summary Level: Nation; 1990. 2. Kelsey JL, O’Brian LA, Grisso GE, Hoffman S. Issues in carrying out epidemiologic research in the elderly. Am J Epidemiol. 1989;130: 857–866. 3. Annest JL, Mahaffey K. Blood lead levels for persons 6 months to 74 years: United States 1976–1980. Vital Health Stat [11]. National Health and Nutrition Examination Survey no 233. DHS publication no. PHS 84–1983. Hyattsville, MD: US Department of Health and Human Services; 1983. 4. Tell GS, Fried LP, Hermanson B, Manolio TA, Newman NB, Borhani NO, for the Cardiovascular Health Study Collaborative research Group. Recruitment of adults 65 years and older as participants of the cardiovascular health study. Ann Epidemiol. 1993;3(4):358–366. 5. Folstein MF, Folstein SE, McHugh PR. Mini-mental state. A practical method for grading cognitive state of patients for the clinician. J Psychiatric Res. 1975;12:189. 6. West SK, Munoz BM, Rubin GS, Fried LP, Schein OD, BandeenRoche K, et al. Function and visual impairment in a populationbased study of older adults: SEE project. Investig Ophthalmol Visual Sci. 1997;38:72–82. 7. Cornoni-Huntley JC, Foley DJ, White LR, Suzman R, Berkman LF, Evans DA, et al. Epidemiology of Disability in the Oldest Old. Methodologic Issues and Preliminary Findings. Milbank Memorial Fund Quarterly/Health and Society. 1985;63(2):350–376. 8. Tielsch JM, Sommer A, Witt K, Katz J, Royal RM, and the Baltimore Eye Survey Research Group. Blindness and Visual Impairment in an American Urban Population. Arch Ophthalmol. 1990;108:286–290. 9. Klein R, Klein BEK, Linton KLP, DeMets DL. The Beaver Dam Eye Study: Visual Acuity. Ophthalmology. 1991;98:1310–1315. 10. Hunninghake DB, Darby CA, Probstfield JL. Recruitment experience in clinical trials: Literature summary and annotated bibliography. Controlled Clin Trials. 1987;8:6S–30S. 11. Zimmer AW, Calkins E, Hadley E, Ostfeld AM, Kaye J, Kaye D. Conducting Clinical Research In Geriatric Populations. Annals of Int Med. 1985;103:276–283. 12. Vogt TM, Ireland CC, Black D, Camel G, Hughes G. Recruitment of elderly volunteers for a multicenter clinical trial: The SHEP Pilot Study. Controlled Clin Trials. 1986;7:118–133. 13. Eyal S, Aaron RF, Rodney J. The effect of non-response on prevalence estimates for a referent population: Insights from a Population-Based Cohort Study. Ann Epidemiol. 1996;6:498–506. 14. Little RJA, Rubin DB. Statistical Analysis with Missing Data. New York: John Wiley & Sons; 1987. 15. Ives DG, Traven ND, Kuller LH, Schulz R. Selection bias and nonresponse to health promotion in older adults. Epidemiology. 1994; 5(4):456–461. 16. Schein OD, Mun˜oz B, Tielsch JM, Bandeen-Roche K, West SK. Prevalence of dry eye among the elderly. Am J Ophthalmol. 1997;124: 723–728.