Diagnostic accuracy of self-reported age-related macular degeneration in the ASPREE Longitudinal Study of Older Persons

Background The validity of findings from epidemiological studies using self-report of ophthalmic conditions depends on several factors. We assessed the diagnostic accuracy of self-reported age-related macular degeneration (AMD) among older Australians enroled in a primary prevention clinical trial and compared diagnostic accuracy between demographic subgroups. Methods At baseline (2010–2015), Australian sub-study participants of the ASPirin in Reducing Events in the Elderly (ASPREE) trial, underwent bilateral two-field, 45° non-mydriatic colour retinal photography. Beckman classification of any-stage AMD was used as the reference standard diagnosis. Participants were asked whether a doctor had ever diagnosed them with “macular degeneration” (the index test) via a paper-based questionnaire as part of the ASPREE Longitudinal Study of Older Persons (ALSOP) within the first year of enrolment. Results In total, 4193 participants were included (aged 70–92 years, 50.8% female). Of those, 262 (6.3%) reported having AMD and 92 (2.2%) were unsure. Retinal grading detected 2592 (61.8%) with no AMD, 867 (20.7%) with early, 686 (16.4%) with intermediate and 48 (1.1%) with late AMD (n = 1601 with any-stage AMD, 38.2%). Self-reported AMD had 11.4% sensitivity (95% CI 9.9–13.1) and 96.9% specificity (95% CI 96.2–97.6) for any-stage AMD, with 69.8% and 63.9% positive and negative predictive values. Sensitivity was higher among participants with late-stage AMD (87.5%), older participants (26.8%), and those with poorer vision (41.0%). Conclusions Although most participants with late-stage AMD were aware of having AMD, the majority with early and intermediate AMD were not. Therefore, findings from studies that rely on disease self-report should be interpreted with caution.


INTRODUCTION
Large community-based studies can play an important role in capturing the level of ocular health in society and in identifying sectors of the population with additional eyecare needs.However, ocular diagnostic testing is not always feasible in studies primarily designed to investigate non-ocular conditions and participantreport of eye disease may be the most practical way to obtain that information [1][2][3][4].Given findings from these studies can have implications for policy development, allocation of resources, and generation of hypotheses for future interventional research, it is important that the reliability of self-report be quantified for each condition and population of interest [5,6].
Accurate self-report of ocular conditions is dependent on participants having the motivation and means to undergo a comprehensive eye exam, appropriate communication of findings by the eyecare provider, and correct recall of those details.Therefore, the diagnostic accuracy of self-report is likely to vary and may be influenced by factors such as access to eyecare, visual function, level of health literacy, and the presence of comorbid conditions [7].
Self-report has previously been shown to be unreliable for AMD with errors increasing with age, time since last eye exam, and poorer vision [7][8][9].However, the extent to which diagnostic accuracy of self-reported AMD differs according to population characteristics has not been investigated in detail.Therefore, we aimed to investigate the diagnostic accuracy of questionnairebased self-report of AMD among generally healthy older Australians using expert-graded bilateral two-field 45° colour fundus photography as the reference standard in this cross-sectional study with prospective and standardised data collection.We compared diagnostic accuracy between demographic and clinical subgroups to identify differences in the ability to self-report.

MATERIALS AND METHODS Study design, eligibility, and recruitment
Australians aged ≥70 years were recruited into the ASPirin in Reducing Events in the Elderly (ASPREE) randomised placebo-controlled trial of 100 mg daily aspirin through general practices across five states/territories between 2010 and 2014 (clinicaltrials.govregistration number: NCT01038583) [10,11].Participants in the USA arm of the ASPREE trial were not invited to participate in the ASPREE-AMD or ASPREE Longitudinal Study of Older Persons (ALSOP) studies and therefore have not been included in this analysis.Exclusion criteria precluded enrolment of people with known cardiovascular disease, independence-limiting physical disability, anaemia, high risk of bleeding, dementia, uncontrolled high blood pressure, or ongoing use of antiplatelet/anticoagulant medication [10].People who needed assistance to complete basic activities of daily living (eating, dressing, walking across a room, bathing, toileting and transferring) were excluded.Participants were required to be able to read and sign a consent form as part of the eligibility criteria and the use of visual aids was permitted for these tasks.Potential participants were also informed that they would be required to see well enough to complete selected written questionnaires confidentially without assistance from other individuals.
Retinal photography was conducted between March 2010 and January 2015 ranging from six months before to three months after the randomisation date as part of three ASPREE sub-studies: the ASPREE-AMD sub-study (Australian New Zealand Clinical Trial Registry: ACTRN12613000755730) [12], the Study of Neurocognitive Outcomes, Radiological and Retinal Effects of Aspirin in Sleep Apnoea (SNORE-ASA) sub-study (ACTRN12612000891820) [13], and the Aspirin for the prevention of cognitive decline in the Elderly: a Neuro-Vascular Imaging Study (ENVIS-ion, ACTRN12609000613202) [14].Some ASPREE participants were enroled prior to the sub-studies commencing in their region and therefore were not invited to participate.The sample size of the ASPREE-AMD study was chosen to detect a difference in the rate of AMD progression between randomisation groups [12].
A paper-based medical questionnaire was mailed to Australian participants who were still active in the ASPREE study within the first year of enrolment (the majority between 3-6 months post randomisation) as part of ALSOP [3] [15].

Demographic data and medical history
Age at randomisation, gender, race, primary language, country of birth, living situation, years of education, and area of residence were collected by study staff at the baseline visits of the ASPREE study.Area of residence was used to derive remoteness area (major city vs not major city) and the Index of Relative Socio-economic Advantage and Disadvantage (IRSAD) decile [16].
As part of the ALSOP baseline medical questionnaire, participants were asked if a doctor had ever diagnosed "Macular degeneration" and the response options were Yes, No, and Don't know.They were also asked, "At the present time, how would you rate your eyesight?(with glasses or contact lenses, if you wear them)."Response options were Excellent, Good, Fair, Poor, Very poor, and Completely blind.

Retinal photography and grading
Bilateral digital 45° macular-and disc-centred colour fundus photographs were captured using non-mydriatic fundus cameras (Canon Inc., Tokyo) with Digital Health Care software (UK) [12].Digital images were viewed immediately and repeated if necessary.All images were graded by two senior graders from the Centre for Eye Research Australia while masked to participant characteristics and questionnaire responses.Images were viewed on high resolution monitors using FastStone Image Viewer software v6.5 (FastStone Corporation).
Per-person AMD status was assigned as the stage of disease in the worse affected eye according to the Beckman classification system [17].Participants were classified as having no apparent ageing in the absence of drusen and retinal pigment epithelium abnormalities.Those with drusen <63 μm and no retinal pigmentary abnormalities were classed as having normal ageing changes.Early AMD was classified as medium drusen (≥63-<125 μm) with no pigmentary abnormalities, while intermediate AMD was defined as medium drusen with retinal pigmentary abnormalities, or large drusen (≥125 μm) with or without AMD pigmentary abnormalities.Late AMD was defined as either neovascular AMD (nAMD), or geographic atrophy (GA).All cases of late AMD were adjudicated by a retinal-specialist ophthalmologist [18].

Statistical methods
Complete-case analyses were conducted, i.e., only participants with nonmissing data on self-reported AMD and gradable retinal images in at least one eye were included in the analyses.Demographic variables were compared according to self-reported AMD status via Pearson's chisquared test.
Photograph-graded AMD status was used as the reference-standard diagnosis.For the primary analysis, AMD was diagnosed as the detection of any-stage AMD, i.e., early, intermediate, or late AMD (versus no AMD or normal ageing changes only).To compare with previous studies that have used self-report of vision-affecting AMD [8], the use of late AMD as a reference-standard (compared to those with no AMD/normal ageing, early AMD, or intermediate AMD) was also examined, as was intermediate AMD or worse (compared to those with no AMD/normal ageing or early AMD).
Self-report of AMD, the index test, was dichotomised as "aware" versus "not aware" of having AMD.Participants who responded "Don't know" (an inconclusive index test) were treated as not being aware of the condition (i.e., best-case scenario) [19].Sensitivity analyses were conducted under the worst-case scenario (inconclusive index tests treated as positive responses) and after exclusion of participants with an inconclusive index test.
Sensitivity, specificity, positive and negative predictive values, positive and negative likelihood ratios, diagnostic odds ratio, and area under the receiver operating characteristic curve were estimated (see Supplementary Table 1 for formulae and interpretation).Interaction terms were included in mixed-effects logistic regression models to compare sensitivity and specificity between demographic subgroups.

Participant characteristics
Among 16,703 Australian ASPREE participants, 5422 (32.5%) underwent retinal imaging.Of those, 879 (16.2%) did not have any data on self-reported AMD status and 350 (6.5%) did not have a gradable retinal photograph for at least one eye, leaving 4193 participants (77.3%, see Fig. 1).Included participants were aged 70-92 years (median 73 years, IQR 71-76) and 2129 (50.8%) were female (see Table 1).Australian ASPREE participants who were excluded from the current analysis were more likely to be older, female, have fewer years of education and live in an area with higher levels of disadvantage (see Supplementary Table 2).

Self-reported and photograph-graded AMD
In response to the questionnaire, 262 (6.2%) participants reported having AMD diagnosed by a doctor and 92 (2.2%) did not know.People who reported having AMD were more likely to be older and more likely to be female than those who reported never being diagnosed (see Table 1).People who reported that they did not know if they had AMD, were more likely to have been born overseas, have a primary language other than English, have fewer years of formal education, and live in an area with less advantage/ more disadvantage (see Table 1).AMD was detected on photographs of 1601 (38.2%) participants (early n = 867, 20.7%; intermediate n = 686, 16.4%; and late AMD n = 48, 1.1%, see Table 2).

Diagnostic accuracy
The numbers of true and false positive and negative responses under each scenario are presented in Supplementary Table 3.
The questionnaire item had poor sensitivity for any-stage AMD (11.4%, see Table 3): positive responses were recorded from only 4.4%, 15.0%, and 87.5% of those with early, intermediate and late AMD, respectively, indicating that people with late AMD were more likely to be aware of having AMD than those with earlier stages of disease.Sensitivity for any-stage AMD increased with increasing age and among those with poorer self-rated eyesight, meaning that participants with AMD in these groups were more aware of having the disease than people with AMD who were either younger or had better vision (see Table 4).
Among those without any AMD, a high proportion (96.9%) reported no diagnosis of AMD or unsure (i.e., excellent specificity, see Table 3).This value decreased slightly with increasing age and among those with poorer self-rated eyesight, meaning the participants without AMD in these categories were slightly more likely to falsely report having AMD than younger people with better vision (see Table 4).Similar patterns were observed under the alternative reference standard diagnoses (intermediate AMD or worse Supplementary Table 4; late AMD Supplementary Table 5).
The questionnaire was only slightly better than chance at distinguishing between participants with and without any-stage AMD (area under the receiver operating characteristic curve 0.54, see Table 3).Of the 262 participants who reported having AMD, evidence of any AMD was detected among 183 (positive predictive value 69.8%), including 38 (20.8%), 103 (56.3%), and 42 (23.0%)with early, intermediate and late AMD, respectively.A negative or inconclusive response (n = 3931) was strongly associated with the absence of late AMD (excellent negative predictive value for late AMD).However, early and intermediate AMD were detected in 829 (21.1%) and 583 (14.8%), respectively, of those with negative/inconclusive responses, and the suboptimal negative likelihood ratio (0.91) indicates a negative/inconclusive response is associated with only a minimal decrease in the probability of having any-stage AMD compared to a positive response (i.e., the questionnaire item is not very helpful for ruling out AMD).Similar results were found under the worst-case scenario (inconclusive index test "Don't know" treated as a positive response) and after inconclusive cases were excluded (see Supplementary Table 6).That is, excellent specificity but poor sensitivity for any-stage AMD.

DISCUSSION
In this large prospective study, we found that asking participants whether a doctor had ever diagnosed them with "macular degeneration" severely underestimated the prevalence of photograph-graded AMD.Although most people with visionthreatening late-stage AMD were aware of their condition, the overwhelming majority of people with early and intermediate AMD were not.While this suggests that many healthy older Australians are not undergoing regular eye examinations, it is also possible that clinicians are not passing on information about early-stage disease when identified, or that individuals have not remembered or understood the information they have been given.
Upon diagnosis with the earlier stages of AMD, patients may be encouraged to take steps to slow disease progression such as quitting cigarettes, becoming more active, and taking dietary supplements [20].People with intermediate AMD, in particular, should be aware of their condition so they can closely monitor their vision for signs of progression and seek timely intervention if needed to prevent severe visual impairment.Thus, it is important for those affected to be aware of having AMD, even in its early stages.We found that the proportion of people correctly identifying themselves as having AMD was lowest among younger people and those with better self-rated eyesight.These characteristics should be considered as potential sources of bias in studies that use self-reported AMD status to estimate population prevalence or to assess associations between AMD Australian residents aged 65 years and over have access to government funding for annual comprehensive ocular exams and around 97% of optometry services were fully covered by government funding during the study period [21].While it is estimated that over 80% of older Australian adults would have had an eye examination within the previous two years, access to eyecare services can vary according to location [22].Although we did not find strong evidence that diagnostic accuracy of selfreport differed according to socio-economic factors in the current study, lower sensitivity in particular demographic groups could attenuate the estimated magnitude of effect between those demographic risk factors and AMD, such as those reported in the UK Biobank studies [23].

Comparison to previous research
Poor diagnostic accuracy for self-report of eye conditions has been acknowledged for many years, with a sensitivity (i.e., proportion of positive reports among those with the condition) for any-stage AMD of 18% in the Beaver Dam Eye Study and 5% in the Los Angeles Latino Eye Study [7,9].Despite the development of therapeutic agents for nAMD and improved access to eyecare services since those studies were conducted, awareness of having AMD remains low.Like the Beaver Dam Eye Study, we found the proportion of people without AMD who correctly report their status to decrease with increasing age [7].However, unlike previous findings, the proportion of people with AMD who were aware of the condition was greater among our older participants and those with poorer self-rated vision, i.e., among those who are more likely to have had a comprehensive eye exam.
Comparable findings were reported among adults with visual acuity worse than 6/12 within a similar population to this study [8].However, interpretation of the estimates from that study is challenging given the incongruity between the questionnaire wording ("Age-related macular degeneration [loss of your central vision]") and the reference standard diagnosis which included the earlier, non-vision threatening stages of AMD [8].

Strengths and limitations
Strengths of this study include the large sample size, the prospective and standardised collection of data, photo documentation of the central retinal status, and inclusion of participants from both metropolitan and regional areas.Experienced retinal image graders provided the reference standard diagnosis based on the Beckman clinical classification for AMD [17].We investigated the diagnostic accuracy of a questionnaire item for detecting both late-stage and any-stage AMD, unlike previous studies which have only used any-stage AMD as their reference standard [7][8][9].ASPREE participants were required to see well enough to independently complete a number of written tasks, potentially Table 3. Diagnostic accuracy parameters for self-reported age-related macular degeneration with 95% confidence intervals (n = 4193).

Intermediate/late Late
Prevalence of reference standard (%) 38.deterring recruitment of people with very poor vision due to ocular pathology.However, the prevalence of AMD detected in this study is consistent with that expected in this age group, indicating that the estimates of positive and negative predictive values may be reasonably extrapolated to the wider community [24].Although the distribution of age in this study broadly reflects that of the community, people with serious health issues were excluded from enroling in the ASPREE study and the proportion of current smokers was lower in APSREE than the wider community [3].In addition, less than 1% of participants in the current study were Indigenous Australians.Thus, we are unable to comment on the accuracy of self-report of eye conditions among these important groups.It is possible that additional cases of AMD would have been detected using multimodal retinal imaging, including more cases of late AMD which are more difficult to diagnose on colour photography alone [25].However, the number of undetected AMD cases is likely to be low given the diagnostic accuracy expected from expert graders of colour fundus photographs [26].
Retinal imaging was performed prior to completing the ALSOP questionnaire for the majority of participants.Some participants with healthy eyes at the time of retinal photography may have developed signs of AMD in the interim.However, this number is expected to be small given the slowly progressing nature of the condition.Other participants may have become aware of existing macular pathology prior to completing the questionnaire due to their involvement in the ASPREE-AMD study.The ALSOP questionnaire was completed at home, allowing participants to refer to records from past ocular exams if available.

Implications for future research
We have shown that the questionnaire item, as it is currently worded, captures some people with the earlier stages of AMD.Identification of these participants may be of interest to researchers who are investigating the underlying risk factors for AMD and its sequalae [27].However, if self-report is being used to estimate the prevalence, causes, or burden of vision impairment due to nAMD and/or GA, then revision of the wording of the questionnaire item could be beneficial given many people with these types of AMD are aware of their diagnosis.For people who respond that a doctor has diagnosed them with AMD, the inclusion of an additional item may assist in differentiating between people who have been informed of early-stage disease and those who have experienced visual disturbance due to late AMD.Consultation with patients, clinicians, and researchers would be needed to develop and validate any new questionnaire items.
Given the suboptimal positive predictive value of self-reported AMD (i.e., the probability of actually having late AMD given positive self-report), confirmatory steps such as retinal imaging or medical record review following positive self-report may be of benefit in future epidemiological studies that lack the resources required to capture retinal images for all participants [28,29].However, this approach would not improve the sensitivity of the questionnaire item for detecting AMD among people who are not aware of having the condition.Linkage with medical records and administrative claims databases may be useful in determining which participants have nAMD via exploration of treatment codes [6,30].Diagnostic codes are not routinely collected for outpatient visits in Australia, meaning that linkage with administrative claims data would not assist in the identification of earlier-stage AMD and GA cases.Likewise, the use of patient-accessible health and pharmacy records could improve participants' ability to accurately report medical history [31], but these approaches will only be of benefit if participants undergo regular eye examinations with accurately-recorded findings.Therefore, retinal imaging of all participants is strongly recommended for future studies that aim to rigorously investigate AMD status.

CONCLUSIONS
AMD is asymptomatic in the early stages and many Australians are unaware of these early changes.This study highlights the need for regular eye exams among older adults to detect common eye diseases, allowing treatment to be initiated in a timely manner to avoid vision loss if appropriate.Results from studies that rely on self-report should be interpreted with caution while AMD remains underdiagnosed in the community.

Table 1 .
Participant characteristics according to self-reported age-related macular degeneration (AMD) status.
a p-values from Pearson's chi-squared test.

Table 2 .
Agreement between photograph-graded and self-reported age-related macular degeneration.

standard photograph-graded AMD grade Index test "Has a doctor ever told you that you have macular degeneration?", n (%) a
a Row percentages given.

Table 4 .
Sensitivity and specificity for any-stage age-related macular degeneration within subgroups of demographic variables.
IRSAD index of relative socio-economic advantage and disadvantage, T true, F false, P positive, N negative.a Missing values for IRSAD (n =