A, An example of a sib pair included in our study. The affected (index)patient (M016-05) has the neovascular form of age-related macular degeneration(AMD) in both eyes. The left eye has a large disciform scar. The unaffected(control) sibling (M016-03) has normal fundi, including normal maculae. Thefundus photographs were obtained in subjects aged 70.4 and 80.3 years, respectively;the control sibling had fundus photographs obtained 5.1 years after the indexpatient's neovascular AMD diagnosis. B, An example of a sib pair not includedin our study. The affected (index) patient (M054-07) has the neovascular formof AMD in both eyes. Although the unaffected (control) sibling (M054-05) reportedno history of AMD, our review of fundus photographs showed pigmentary changesin the maculae, scattered drusen, and peripapillary atrophy. The fundus photographswere obtained in subjects aged 70.0 and 75.0 years, respectively; the controlsibling had fundus photographs obtained 5.0 years after the index patient'sneovascular AMD diagnosis.
DeAngelis MM, Lane AM, Shah CP, Ott J, Dryja TP, Miller JW. Extremely Discordant Sib-Pair Study Design to Determine Risk Factorsfor Neovascular Age-Related Macular Degeneration. Arch Ophthalmol. 2004;122(4):575-580. doi:10.1001/archopht.122.4.575
Copyright 2004 American Medical Association. All Rights Reserved.Applicable FARS/DFARS Restrictions Apply to Government Use.2004
To search for factors that contribute to the development of neovascularage-related macular degeneration (AMD).
In a matched-pair case-control study, we studied sib pairs in whichthe index sibling had neovascular AMD in at least 1 eye and the unaffectedsibling had normal maculae (or at most only a few small drusen) and was pastthe age at which the index case was diagnosed. Factors studied included sex,iris color, education, alcohol consumption, body mass index, vitamin use,smoking history, hypercholesterolemia, aspirin use, hypertension, other cardiovasculardisease, any autoimmune disease, and non–insulin-dependent diabetesmellitus. Conditional logistic regression was performed to identify predictorsof neovascular AMD.
On the basis of 73 sib pairs, multivariate regression analysis revealeda statistically significant 2% increase in risk of neovascular AMD with eachpack-year of smoking (odds ratio, 1.02; 95% confidence interval, 1.01-1.04; P = .007). Suggestive but nonsignificant associations werealso observed for mean lifetime alcohol consumption, adult lifetime body massindex, and hypertension in multivariate regression analyses.
Using extremely discordant sib pairs to study risk factors for AMD,a novel approach in epidemiological design, we found evidence that smokingis a risk factor for neovascular AMD.
Results of previous studies of environmental, constitutional, medical,and social factors have not been in agreement as to predictors of age-relatedmacular degeneration (AMD). Cigarette smoking appears to be the only riskfactor consistently associated with an increased risk for AMD. However, thisfactor has not been detected in some studies of the neovascular form of thedisease.1- 3 Giventhe multifactorial and heterogeneous nature of AMD, the use of epidemiologicalmethods to detect weak to moderate associations may be required to identifymodifiable risk factors.
Mathematical analyses indicate that the evaluation of sib pairs whoare extremely discordant for a quantitative multifactorial trait can be apowerful method for identifying the genes that govern the trait.4 Theprototype of an "extreme" sib pair consists of 1 member with a trait valuein the top 10% and the other member in the bottom 10% of the population distribution.4 Sib pairs composed of the top 10% and bottom 30% canbe almost as valuable.5 Such sib pairs aremore likely than pairs of siblings with intermediate trait values to differat many of the multiple genetic loci that govern a multifactorial trait. Byextension, extremely discordant sib pairs would also be expected to reportdifferent exposures to any environmental factors that influence the trait.Thus, epidemiological studies of multifactorial traits might also benefitfrom the analysis of extremely discordant sib pairs.
We studied extremely discordant sib pairs to search for epidemiologicalrisk factors associated with AMD. To do this, we needed to measure maculardegeneration as a quantitative trait. Some groups have proposed staging schemesthat in effect provide that quantification6,7 (MathewDavis, MD, unpublished data, May 2003). Regardless of the quantification scheme,it seems clear that patients with neovascular AMD or atrophic AMD involvingthe fovea (ie, geographic atrophy) are at the severe end of the spectrum ofmacular aging changes. Data from epidemiological studies of populations inthe Netherlands, Australia, and Wisconsin8 indicatethat the prevalence of neovascular AMD is about 4% and of geographic atrophyis about 2%, for a combined total of 6% in those older than 75 years.
For this study, we considered patients with these forms of AMD to bein the top (ie, most severe) decile. Even fewer younger individuals have neovascularizationor geographic atrophy, so individuals younger than 75 years with these severeforms of AMD also would be in the top decile. Patients with no aging changesin the fundus would represent the opposite end of the AMD spectrum; they wouldhave the lowest trait values in any AMD grading scheme. Drusen are presentin at least 1 eye of about 95% of those older than 43 years,7 soit would be reasonable to place individuals with no drusen or other macularaging changes in the bottom decile. Among individuals aged 75 years or olderin Beaver Dam, Wis, about 30% have no drusen greater than 63 µm in diameter,no retinal pigmentary abnormalities, no geographic atrophy, and no neovascularization.7 On the basis of these results, we considered suchindividuals to be in the bottom 30% of the spectrum of macular aging changes.
We confined this study to patients with neovascular AMD and excludedpatients with geographic atrophy because patients with neovascular AMD werereadily available for recruitment at the Massachusetts Eye and Ear Infirmary,Boston. More importantly, because some risk factors may specifically predisposepatients to this severe subtype of AMD, an analysis of only those with neovascularAMD might have more power. For epidemiological evaluation, we recruited sibpairs in which the index sibling had the neovascular form of AMD, and thematching unaffected sibling had either no macular aging changes or at mostonly a few small drusen at an age equal to or older than the age at whichthe index sibling had neovascular AMD diagnosed.
The protocol was reviewed and approved by the institutional review boardat the Massachusetts Eye and Ear Infirmary and conforms to the tenets of theDeclaration of Helsinki. Eligible patients were enrolled in this study afterthey provided informed consent either in person, on the telephone, or throughthe mail.
Recruitment for this study occurred from September 1998 to June 2003.Index patients with neovascular AMD were recruited from the Retina Serviceof the Massachusetts Eye and Ear Infirmary. Each recruited index patient wasasked if he or she had a sibling with no history of AMD and who was past theage at which the index patient first had neovascular AMD diagnosed. Afterobtaining informed consent from the index patient, we contacted those siblings,asked for their consent to enter the study, and then reviewed their ocularhistory. If a sibling was apparently eligible as an unaffected sibling, wesought to document that status by obtaining fundus photographs or by arrangingto have copies of pre-existing fundus photographs sent to us for review. Afew unaffected siblings were unwilling to come to an ophthalmologist's officefor fundus photographs, or they were too frail to do so, but they were amenableto a home visit by 1 of the investigators (T.P.D.) for a dilated fundus examination.
Blood samples from index and unaffected siblings were collected. TheDNA purified from these samples was analyzed with 3 highly polymorphic microsatellitemarkers (D2S428, D5S816, D7S796) with a heterozygosityof 94%.
All index patients were aged 50 years or older and had the neovascularform of AMD in at least 1 eye; neovascular AMD was defined as subretinal hemorrhage,fibrosis, or fluorescein angiographic presence of neovascularization documentedat the time of or before enrollment in the study. Patients whose only exudativefinding was a retinal pigment epithelium detachment were excluded becausethis finding may not represent definite neovascular AMD and, therefore, thesevere phenotype we sought. Patients with signs of pathologic myopia, presumedocular histoplasmosis syndrome, angioid streaks, or choroidal rupture werealso excluded.
The unaffected siblings had normal maculae at an age older than thatat which the index patient had neovascular AMD diagnosed. Unaffected siblingshad maculae (defined as the zone centered at the foveola and extending 2 discdiameters, or 3000 µm, in radius) fulfilling the following criteria:0 to 5 small drusen (all less than 63 µm in diameter), no pigment abnormalities,no geographic atrophy, and no neovascularization as defined previously.9,10 Disease status of every participantwas confirmed by means of fundus photography or fluorescein angiography byat least 2 investigators (T.P.D. and J.W.M.), except in 4 cases in which 1of the investigators (T.P.D.) conducted a home examination.
Interviews of participants were conducted in person or by telephone.After a sib pair was deemed eligible (Figure1), we administered a standardized questionnaire to all participantsto ascertain potential risk factors for AMD. The factors studied were sex,iris color, education, alcohol consumption, body mass index (BMI), multivitaminuse, vitamin C use, vitamin E use, smoking history, hypercholesterolemia,aspirin use, hypertension, other cardiovascular risk factors (angina, bypasssurgery, myocardial infarction, stroke, or transient ischemic attack), autoimmunedisease (thyroid disease, psoriasis or eczema, systemic lupus, multiple sclerosis,rheumatoid arthritis), and non–insulin-dependent diabetes mellitus.
Through an interview with the index patient, we established the ageat which he or she was first told by an ophthalmologist that he or she hadAMD; in most cases, AMD and neovascular AMD were diagnosed simultaneously.This age was the reference age for the index patient. The reference age forall siblings of the index patient was the reference age of the index patient.In sibships in which more than 1 sibling was affected (n = 4), the affectedsibling with the earliest age of diagnosis was used as the reference age forall siblings.
Iris color was self-reported by the patient, with brown defined as darkand nonbrown defined as blue, gray, green, or hazel. Education was self-reportedand categorized into 3 groups: college graduate and/or advanced degree (ie,master's, legal, medical, doctorate) vs high school graduate with or withoutsome college vs no high school diploma.8,11 Regularuse of alcoholic beverages was measured as grams of alcohol consumed per week,with 1 can, glass, or bottle of beer equal to 12.8 g of ethanol; 1 4-oz glassof wine equal to 11.0 g of ethanol; and 1 drink or shot of liquor equal to14.0 g of ethanol.12 The mean number of drinksper week was calculated per decade from the 20s until the decade of the referenceage for each sib pair. Weight in pounds was recorded decade by decade andthen converted to kilograms, excluding years of pregnancy or lactation, fromthe 20s until the decade of the reference age. Adult lifetime BMI was calculatedas the mean weight in kilograms across these decades divided by the squareof the height in meters at age 25 years.
Use of multivitamins, vitamin C, vitamin E, or aspirin was scored aspositive if it was regularly used (ie, taken at least twice per week for atleast 6 months before AMD diagnosis). Participants were categorized regardingtheir smoking history as "never smoked," "past smoker," or "current smoker."A past or current smoker was defined as having smoked 100 or more cigarettesin the course of his or her entire life. If the patient was a past or currentsmoker, the age when the individual started smoking was recorded. If an individualwas a past smoker, the age when the individual quit was recorded. The numberof pack-years of cigarettes smoked was estimated for each past and currentsmoker. A pack-year was defined as 1 pack of cigarettes a day for 1 year,with 20 cigarettes equaling 1 pack.
For medical conditions including hypercholesterolemia, hypertension,non–insulin-dependent diabetes, any autoimmune disease, and angina,the regular use of medication as reported by the individual for at least 6months before the reference age was necessary to define the presence of thecondition. The regular use of medication was defined as at least twice a weekfor 6 months or more.
This was a matched-pair case-control study, so conditional logisticregression with commercially available software (StataCorp, College Station,Tex) was performed to identify risk factors for neovascular AMD. Exposuresas defined earlier were evaluated initially by means of univariate analysis.Where appropriate, exposures were entered into separate models as either continuousor categorical variables, including alcohol consumption, BMI, pack-years ofsmoking, and duration of certain chronic conditions. A multivariate modelwas built by using the risk factors from the univariate model that appearedto be associated with neovascular AMD; P < .1was considered to indicate statistical significance. Nonsignificant variableswere then dropped from the multivariate model to create the most parsimoniousstatistical model predictive of neovascular AMD.
We recruited 81 extremely discordant sib pairs derived from 64 indexcases (ie, some index patients had more than 1 matching sibling). The 64 indexcases were identified after approaching 533 patients who had neovascular AMD.We obtained epidemiological data by completing standardized questionnairesin 73 (90%) of 81 extremely discordant sib pairs. The remaining 8 sib pairswere not enrolled because 1 sibling in each pair had severely deterioratinghealth or had died during the course of the study, which precluded completionof the questionnaire.
The 73 enrolled sib pairs were derived from 50 sibships. Eight of the50 sibships had more than 1 unaffected member, 2 had more than 1 affectedmember, and 2 families had more than 1 member in both the affected and unaffectedgroups; all affected members were considered index patients. The mean ageof the 56 index siblings at the time of AMD diagnosis was 69.1 years; themedian age was 70.5 years (range, 51.2-85.0 years). The mean age of the 63unaffected siblings at the time of their sibling's AMD diagnosis was 69.9years; the median age was 70.8 years (range, 44.7-86.5 years). The mean andmedian interview ages for affected siblings were 74.5 and 73.8 years (range,56.5-88.5 years), respectively, whereas the mean and median interview agesfor unaffected siblings were 75.8 and 77.0 years (range, 55.3-91.9 years),respectively. The mean difference in age at interview of the unaffected siblingand the age of AMD diagnosis of the index sibling was 8.7 years and variedbetween 2.5 months and 26.1 years. All participants were white.
To determine if there was substantial contamination of our sib pairswith sib pairs who were unknowingly half siblings or who were unrelated, weanalyzed DNA from a subset of 48 of the sib pairs with 3 unlinked microsatellitemarkers (D2S428, D5S816,D7S796) each having a heterozygosity of 94%. At each marker locus, we recordedthe number of alleles (0, 1, or 2) shared by each pair of siblings, whichis defined as the identity-by-state score. Summing all 3 markers and excludingoccasional samples that did not work for 1 or 2 markers, we found 25 instancesof an identity-by-state score of 0, 67 of an identity-by-state score of 1,and 41 of an identity-by-state score of 2. This finding was compared withthe expected identity-by-state score totals for markers with a heterozygosityof 94% (30, 66, and 37, respectively). There was no statistically significantdifference between the observed and expected distributions of identity-by-statescores (χ23 = 0.67, P >.75).This result suggests that there is no major contamination of our recruiteddiscordant sib pairs with half siblings or unrelated pairs.
In univariate analyses, associations were found at P <.1 between neovascular AMD and alcohol consumption of 105 g ormore per week, increasing pack-years of smoking, adult lifetime BMI greaterthan 20 kg/m2 and less than 25 kg/m2, and hypertension. Additionally, subjects with hypertensiondiagnosed more than 5 years before AMD diagnosis had increased risk of neovascularAMD, as compared with subjects who were normotensive or who were hypertensive5 years or less (Table 1). Testsfor trend for increasing alcohol consumption and BMI were not statisticallysignificant (data not shown). No associations were found between neovascularAMD and the use of multivitamin preparations or the use of vitamins C or Eindividually (Table 1); this lackof association was also observed when any vitamin use (vitamin C, vitaminE, or multivitamins) was considered as a group (odds ratio [OR], 1.21; 95%confidence interval [CI], 0.52-2.82; P = .66). Furthermore,no protective effect was suggested when we considered participants who usedall 3 supplements (OR, 0.98; 95% CI, 0.62-1.59; P =.92). We did not find a statistically significant association between anyother studied risk factor and neovascular AMD (Table 1).
Multivariate analyses were conducted to determine which risk factorsmight be independently associated with neovascular AMD. For the multivariateanalyses, we did not control for age because the reference age for both theunaffected and affected sibling in a pair was the same. The reference agewas the same across sibships to reduce bias in measurement of exposure levelsbetween the siblings and to ensure that the variable being studied occurredbefore the onset of AMD. Our multivariate model included adult lifetime BMI,alcohol consumption, pack-years of smoking, and duration of hypertension (orhistory of hypertension); these variables from the univariate analyses metour criteria for entry into the multivariate model (ie, significance at P <.1).
We found that alcohol consumption of 105 g or more per week, which wasa statistically significant risk factor in univariate analysis, was not associatedwith neovascular AMD in the multivariate model (OR, 3.52; 95% CI, 0.42-29.77; P = .25). Similarly, durationof hypertension for more than 5 years (OR, 1.84; 95% CI, 0.62-5.48; P = .27) and mean lifetime BMI greater than 20 kg/m2 and less than 25 kg/m2 (OR,3.58; 95% CI, 0.83-15.49; P = .09) were not significantlypredictive of neovascular AMD after controlling for other factors. A historyof hypertension was also not significantly associated with AMD risk in themultivariate model. Only pack-years of smoking was a significant predictorof neovascular AMD. Specifically, each pack-year of smoking was associatedwith a 2% increase in the risk of neovascular AMD (OR, 1.02; 95% CI, 1.01-1.04; P = .007).
Our analyses of extremely discordant sib pairs suggest that smokingincreases the risk of neovascular AMD. These results are in agreement withthose of previous studies8,11,13- 15 ofrisk factors for neovascular AMD that found that smoking increases risk. However,authors of 1 recent prospective study from Beaver Dam, Wis,3 didnot find an association between pack-years of smoking, evaluated in 10-yearincrements, and neovascular AMD. Furthermore, a retrospective study by Chaineet al2 and a prospective study by Vinding etal1 also showed no association between smokingand neovascular AMD. However, the authors of these studies did not use pack-yearsto analyze smoking but instead used smoker status (eg, past vs present). Toour knowledge, in no previous study of AMD was the extremely discordant sib-pairapproach used.
Suggestive findings in our study for BMI, alcohol consumption, and hypertensionhistory are supported by some previous research results. Body mass index waslinked to risk of neovascular AMD in the Age-Related Eye Disease Study,13 and BMI was a significant factor in AMD progressionof any type in the progression of AMD study.16 Findingsregarding associations between neovascular AMD and history of hypertensionor heavy alcohol consumption have been inconsistent. Results of 2 prior retrospectivestudies13,17 supported an association,whereas results of another retrospective study2 anda meta-analysis8 of 3 prospective studies (BeaverDam, Rotterdam, and the Blue Mountain Eye Study) showed no independent associationbetween hypertension and neovascular AMD. Klein et al3 reportedan increased risk of late-stage AMD (cases of neovascular AMD and cases ofgeographic atrophy) associated with heavy drinking. This finding is in contrastto those of Cho et al12 and Ajani et al18 who found no association of AMD with alcohol consumption.
Differences in definition and measurement of these exposures acrossstudies make it difficult to compare and interpret these findings. In addition,these factors, including smoking, are intercorrelated, and confounding ofassociations in some studies may contribute to the disparate results. In ouranalysis, ORs changed only minimally in the multivariate-adjusted analysis,but our study lacked the statistical power for a rigorous evaluation of theserelationships.
Our study was retrospective and, by design, most unaffected siblingswere older at the time of their interviews than were their affected siblings.As a result, recall bias may have been introduced, with index patients reportingexposures more accurately than did their unaffected siblings. However, webelieve this bias was minor because the difference in mean interview age betweenthe affected and unaffected siblings was less than 2 years.
It is possible that we underestimated exposure measures of unaffectedsiblings by using the youngest age of diagnosis as our reference age. Thismeasure, however, is unlikely to have affected our results because most (80%[58 of 73]) of our sib pairs involved unaffected siblings who were older thanthe affected siblings. Nevertheless, to evaluate this possibility, we performedadditional analyses by using the oldest age of diagnosis as our referenceage in families with multiple affected siblings. Results of this analysiswere similar to those of our original analysis, with smoking identified asa significant predictor of neovascular AMD.
Although analysis of extremely discordant sib pairs can be used to convincinglyidentify risk factors with relatively small numbers of cases, false-negativeresults (type II errors) may occur because of high concordance of geneticand demographic factors among siblings. For example, we found a high levelof concordance among our sib pairs for iris color, other cardiovascular factors,non–insulin-dependent diabetes mellitus, and level of education. Moreover,with only 73 sib pairs analyzed, the study may be relatively underpowered,and factors significant in univariate analysis failed to achieve significancein the multivariate models. Thus, our study provides only suggestive evidencethat other factors, including BMI, hypertension, and alcohol consumption,may have a role in the development of neovascular AMD.
Despite the limitations of our study, the fact that we found an associationbetween neovascular AMD and cigarette smoking with our admittedly modest samplesize is strong evidence in favor of the validity of this association. Theextremely discordant sib-pair approach may be a valuable tool for understandingthe role and interaction of epidemiological and genetic factors in contributingto multifactorial diseases such as AMD.
Corresponding author and reprints: Joan W. Miller, MD, Departmentof Ophthalmology, Harvard Medical School, Retina Service, Massachusetts Eyeand Ear Infirmary, 243 Charles St, Boston, MA 02114 (e-mail: email@example.com).
Submitted for publication June 9, 2003; final revision received November4, 2003; accepted December 18, 2003.
Dr DeAngelis and Ms Lane contibuted equally to this work.
This study was supported by grants from the Ruth and Milton SteinbachFund, New York, NY, and National Institutes of Health, Bethesda, Md, grantsNRSA, EY07076, and HG00008.