Dietary fat consumption and primary open-angle glaucoma 1 3

Transcription

1 Dietary fat consumption and primary open-angle glaucoma 1 3 Jae H Kang, Louis R Pasquale, Walter C Willett, Bernard A Rosner, Kathleen M Egan, Nicholaus Faberowski, and Susan E Hankinson ABSTRACT Background: Prostaglandin F 2 analogues are effective intraocular-pressure-lowering drugs. Dietary fatty acids affect endogenous prostaglandin F 2 concentrations and may thus influence intraocular pressure. Objective: We prospectively examined dietary fat consumption in relation to primary open-angle glaucoma (POAG). Design: (n in the Nurses Health Study) and men (n in the Health Professionals Follow-Up Study) free of POAG in 1980 and 1986, respectively, were followed until 1996 if they were 40 y old and reported receiving eye exams during follow-up. Potential confounders were assessed on biennial questionnaires, and energy-adjusted cumulative averaged fat intakes were measured by using validated food-frequency questionnaires. We analyzed 474 self-reported POAG cases confirmed by medical chart review. Cohort-specific multivariate rate ratios (RRs) were obtained by using proportional hazards models and were then pooled. Results: Major fats and fat subtypes were not independently associated with POAG risk. multivariate RRs (95% CI) for POAG comparing the highest with the lowest quintile of fat intake were as follows: 0.90 (0.67, 1.21) for total fat, 1.03 (0.77, 1.38) for saturated fat, 0.76 (0.56, 1.03) for monounsaturated fat, and 0.87 (0.66, 1.16) for polyunsaturated fat, none of which were statistically significant. We found a suggestive positive association between a higher ratio of n 3 ton 6 polyunsaturated fat and risk of POAG [RR 1.49 (1.11, 2.01); P for trend 0.10], which was stronger for high-tension POAG [RR 1.68 (1.18, 2.39); P for trend 0.009]. Conclusion: A high ratio of n 3ton 6 polyunsaturated fat appears to increase the risk of POAG, particularly high-tension POAG. Further studies are needed. Am J Clin Nutr 2004;79: KEY WORDS Primary open-angle glaucoma, dietary fats, unsaturated fatty acids, food-frequency questionnaire, Nurses Health Study, Health Professionals Follow-Up Study, prospective studies INTRODUCTION Glaucoma is a major cause of blindness in the United States, especially in African Americans and Hispanic Americans (1), in whom it is the leading cause (2, 3). Elevated intraocular pressure (IOP) is an established risk factor for primary open-angle glaucoma (POAG; 4, 5), and the treatments currently available for managing glaucoma focus on lowering patients IOP (6). Latanoprost (trade name Xalatan; Pharmacia & Upjohn Company, Kalamazoo, MI) is a potent IOP-lowering drug that is widespread in its use for managing glaucoma. It is a prostaglandin F 2 analogue engineered for its bioavailability and effectiveness in lowering IOP while minimizing ocular and systematic side effects (7). A review of the initial clinical trials of latanoprost administration reported that IOP reductions of 25 35% have been observed both in healthy volunteers and in patients with ocular hypertension or glaucoma (8). Latanoprost is believed to exert its IOP-lowering effects mainly by increasing uveoscleral flow through the iris root and ciliary body, which comprises 5 15% of the total outflow (9, 10). Newer IOP-lowering drugs that are also prostaglandin-type agonists include bimatoprost and travoprost. Also, other prostaglandins have been effective in lowering IOP in animals through other mechanisms and are under investigation for therapeutic use (11, 12). Endogenous eicosanoids, including prostaglandins, are local hormones that have diverse and important effects in all human tissues, including the eye (13). Eicosanoids can be divided broadly into 2 types depending on the precursor fatty acids: the n 6orn 3 eicosanoids. The n 6 eicosanoids are metabolites of arachidonic acid, a long-chain unsaturated n 6 fat derivative. Prostaglandins such as prostaglandin F 2 are of this type. The n 3 eicosanoids are formed from eicosapentaenoic acid, the n 3 homologue of arachidonic acid, and these often have weaker physiologic effects on tissues (14). Eicosapentaenoic acid can inhibit arachidonic acid metabolism because these 2 progenitors compete as substrates for the eicosanoidsynthesizing enzymes, resulting in a suppression of n 6 eicosanoid production. The relative amount of arachidonic acid versus eicosapentaenoic acid in tissues, in turn, is determined by the composition of dietary polyunsaturated fat. Arachidonic acid is formed mainly from the essential fatty acid linoleic acid (which is abundant in corn, sunflower, and safflower oils) or is ingested in small amounts in meats. Eicosapentaenoic acid is derived from 1 From the Departments of Epidemiology (JHK, WCW, KME, and SEH), Nutrition (WCW), and Biostatistics (BAR), Harvard School of Public Health, Boston; the Channing Laboratory, Department of Medicine, Harvard Medical School and Brigham & s Hospital, Boston (JHK, WCW, BAR, and SEH); the Division of Ophthalmology, Brigham & s Hospital, Boston (LRP); the Glaucoma (LRP) and Retina (KME) Services, Massachusetts Eye and Ear Infirmary, Boston; and the Department of Ophthalmology, Boston University School of Medicine (NF). 2 Supported by grants CA87969, CA55075, EY09611, and HL35464 from the National Institutes of Health and by a grant from the Glaucoma Research Foundation. 3 Address reprint requests to JH Kang, Channing Laboratory, 181 Longwood Avenue, Boston, MA nhjhk@channing.harvard.edu. Received November 15, Accepted for publication October 10, Am J Clin Nutr 2004;79: Printed in USA American Society for Clinical Nutrition 755

2 756 KANG ET AL the essential fatty acid -linolenic acid (which is found in flaxseed, canola, and soy oils and green leafy vegetables) or is ingested in fish or fish oil (13, 15). We hypothesized that the dietary intake of fats may modulate the availability of endogenous n 6 prostaglandins, and that differences in physiologic concentrations of these agents may influence IOP. Because dietary fat consists of many subtypes that occur together in nature, we also assessed the independent associations of various types of dietary fat with POAG. We report here the results of a prospective study of 2 large cohorts of men and women, each followed for 10 y, in whom diet was assessed multiple times to examine the relation between types of dietary fat and risk of POAG. SUBJECTS AND METHODS Populations studied The Nurses Health Study (NHS) was started in 1976 with US female registered nurses aged y who responded to a mailed questionnaire on health information and medical history. The Health Professionals Follow-Up Study (HPFS) began in 1986 with the enrollment of US male health professionals (dentists, veterinarians, pharmacists, optometrists, osteopaths, and podiatrists) aged y who also responded to a mailed questionnaire. Participants complete biennial questionnaires on risk factors and newly diagnosed illnesses, such as glaucoma. The present study was approved by the Brigham & s Human Research Committee and the Harvard School of Public Health. Because the first dietary assessments occurred in 1980 for the NHS and in 1986 for the HPFS, the study period was restricted to in the NHS and in the HPFS. Persontime was accrued from the return date of the questionnaire completed when a participant first became eligible until the earliest occurrence of a report of glaucoma, cancer, or death; loss to follow-up; or 1996 while permitting only the person-time during which a participant was aged 40 y (because glaucoma risk increases after the age of 40 y) and reported having had an eye exam (to minimize possible detection bias). Person-years were accrued in approximate 2-y units and were categorized on the basis of responses to the most recent biennial questionnaire. Of the original cohort members, participants were excluded for the following reasons as of 1980 in the NHS and as of 1986 in the HPFS: 1) women did not return the first semiquantitative food-frequency questionnaire (FFQ) in 1980; 2) 1596 men and 5994 women had inadequate diet information on the first semiquantitative FFQ (adequate information was defined as follows: for men, 70 of 131 items left blank on the FFQ and a total caloric intake range of kcal/d; for women, 10 of 61 items left blank and a total caloric intake range of kcal/d); 3) 1927 men and 3625 women had prevalent cancers, excluding nonmelanoma skin cancer; 4) 787 men and 767 women had a prevalent diagnosis of glaucoma or suspected glaucoma; 5) 1029 men and 795 women were later lost to follow-up; 6) 18 men and 4 women had missing age information; and 7) 5507 men and 9686 women did not report having an eye exam during follow-up. After these exclusions, men and women remained. However, at each specific 2-y period, participants who were aged 40 y or who did not report having an eye exam were also considered ineligible. For example, the number of participants who contributed person-time for the first 2 y ( in the HPFS, in the NHS) was men and women. This was after excluding men and women considered temporarily ineligible because they were aged 40 y (221 men and women) or did not report receiving an eye exam when first asked ( men and women). At later periods, these ineligible participants were included if they reached 40 y of age and reported receiving eye exams. Hence, by 1996, a total of men and women contributed person-time. Follow-up rates were high ( 95% of the total possible person-time). We determined eligibility for the eye exam criterion by selecting those who responded positively to the question of whether an eye exam was received in the previous 2 y. For example, if a subject answered positively only in 1994 and 1996, then she contributed person-time only during and Because this question was first asked in 1990 in both cohorts, we determined eye exam eligibility in this way from 1988 on. However, for the initial periods of in the HPFS and in the NHS, eye exam eligibility was based on responses to the 1990 question (eg, an NHS participant eligible in 1980 contributed 8 y from if the response to the 1990 eye exam question was positive and 0 y if negative). Measurement of fat intake from foods and supplements Semiquantitative FFQs were administered repeatedly during the study period: in 1980, 1984, 1986, and 1990 for the NHS and in 1986, 1990, and 1994 for the HPFS. The NHS 1980 FFQ contained 61 foods found to maximally discriminate intakes of specific fats, fiber, and 12 other nutrients. The 1984 NHS FFQ was expanded to 116 foods to include more fruit, vegetable, and fish items, and similar versions of it were used from 1986 on in both the NHS (126 foods) and the HPFS (131 foods). For each food item in these FFQs, a specific serving size was given and the respondents were asked to estimate their average frequency of consuming that serving of food in the past year from 9 response selections ranging from almost never to 6 or more times a day. For each participant, we calculated the total intakes of various fats by summing the contributions from foods and oil supplements (when asked in 1990 and 1994). Food contributions were computed by multiplying the frequency of consumption of a food item by the nutrient content of specified portions. Information on the nutrient contents of various foods was obtained from US Department of Agriculture sources (16), taking into account types of margarine and fats used in cooking and baking. The validity of the FFQs has been evaluated in both cohorts, and the FFQs have been shown to measure dietary intakes of fat reasonably well (17, 18). Energy-adjusted fat intakes from the FFQ were compared with the intakes based on the average of two 1-wk diet records. The Pearson correlation coefficients corrected for attenuation due to random error in diet records ranged from 0.48 to 0.73 (0.57 for total fat and 0.48 for polyunsaturated fat) in women and from 0.37 to 0.76 (0.67 for total fat and 0.37 for polyunsaturated fat) in men. For evaluating the performance of the FFQ for measuring specific fatty acids, the intakes as a percentage of total fat from the FFQs were compared with the fatty acid composition of subcutaneous fat aspirates (19, 20). The Spearman correlation coefficients were r 0.37 for women and 0.43 for men for polyunsaturated fat, r 0.51 and 0.34 for trans polyunsaturated fat, r 0.35 and 0.37 for linoleic acid, and r 0.48 and 0.49 for long-chain n 3 fatty acids. The fat intake distribution of the participants was similar to

3 FAT INTAKE AND GLAUCOMA 757 that reported in the third National Health and Nutrition Examination Survey (21). For example, in the national survey, the mean percentages of total energy from fat for women aged 40 y ranged from 31.3% to 34.9% and those for men ranged from 33.3% to 33.9%; the corresponding mean ( SD) percentages from our participants were % and %. The mean percentages of total energy from saturated fat in the national survey ranged from 10.8% to 11.8% for women and from 11.3% to 11.8% for men; the corresponding percentages in the present study were % for women and % for men. Measurements of covariates During follow-up, we asked the participants to report information on potential confounders such as age; body mass index; diagnosis of hypertension, diabetes, cataract, or macular degeneration; history of physician exams; history of eye exams; and physical activity. Alcohol use was assessed in the FFQs. Ethnicity was asked in 1992 for women and in 1986 for men. Case definition and ascertainment We asked participants about any diagnosis of glaucoma and dates of diagnosis from 1986 onward. From the participants who self-reported glaucoma, we requested permission to review their medical records and information on which ophthalmologist or optometrist made the initial diagnosis. We requested that eye care providers send copies of ocular records or complete a brief questionnaire about IOP, the presence of optic disc cupping, any glaucomatous visual field (VF) loss, dates of diagnosis, and whether the glaucoma was primary or secondary. We selected participants whose eye doctors indicated POAG with glaucomatous VF loss and obtained complete ocular records, including VF records, from the original date of diagnosis to the most recent. These records were examined independently by 2 ophthalmologists (LRP and NF) who were masked to the participants fat intake. Cases for analysis were those with a diagnosis that the reviewers agreed was either definite or probable POAG. The standardized criteria for these designations are described below. For the date of diagnosis, we used the date of onset of the earliest glaucomatous sign in either eye. Definite POAG cases were required to meet the following criteria: 1) angles were open and not occludable in both eyes on the basis of gonioscopy, 2) there were no indications of secondary causes of glaucoma on the basis of a slit lamp exam, and 3) VF defects were present and consistent with glaucoma in the most recently available reliable VF tests (eg, nasal step, nasal depression, paracentral scotoma, arcuate defect, blind spot enlargement, or temporal wedge), were seen at the same locus on at least one prior reliable VF test, and were not due to other ocular conditions or optic disc pathology. Although there was no requirement for the type of perimetry performed, static automated perimeters had to have an age-matched normal database and VF tests had to be reliable for the affected eye or eyes. For static threshold or suprathreshold testing, we considered a field reliable if the fixation loss was 33%, the false-positive rate was 20%, and the false-negative rate was 20%. For kinetic VFs, we considered a field reliable unless there was notation by the examiner that the VF was unreliable. Probable POAG cases met the above criteria except for gonioscopy; in lieu of this, they had to have documentation that dilations were without any subsequent adverse events. Differences in reviewer assessments occurred for 8.9% of the cases included in the analyses. The 2 main reasons for differences were errors in calculating VF test reliability parameters or omissions in reading a few VF results in the charts. The 2 reviewers resolved their differences after an open discussion of the available information. During follow-up, 1274 men and 2897 women self-reported a diagnosis of glaucoma. These self-reports were confirmed by eye doctors for 61% of the men and 70% of the women as follows: 317 men and 693 women had POAG with VF loss, 265 men and 730 women had only elevated IOP or optic disc cupping, and 116 men and 519 women had other types of glaucoma or suspected glaucoma. The remaining 39% of self-reports by the men and 30% of self-reports by the women could not be confirmed because the participants (11% of the men, 6% of the women) or their eye doctors (4% of the men, 2% of the women) could not be contacted, the participants did not give permission to review their records (11% of the men, 10% of the women), the participants indicated the initial report was in error (13% of the men, 10% of the women), or the participants eye doctors disconfirmed the diagnosis of POAG (2% of the men, 2% of the women). Of the 317 men and 693 women confirmed to have POAG with VF loss by their doctors, 188 men (59%) and 320 women (47%) met the criteria for definite or probable POAG on the basis of medical records review. The remaining patients were excluded for having only one VF showing an abnormality (8% of the men, 9% of the women), only elevated IOP (12% of the men, 18% of the women), other glaucoma (11% of the men, 14% of the women), no signs of glaucoma (0.3% of the men, 3% of the women), or insufficient documentation (9% of the men, 10% of the women). After we further excluded those who were diagnosed after the end of follow-up, those without complete diet information, and those with a history of cancer, 173 men and 301 women were included as cases in the analysis. Statistical analysis We evaluated total fat, cholesterol, fat according to source (animal or vegetable fat), major fat subtypes (saturated, polyunsaturated, monounsaturated, and trans polyunsaturated), and subtypes of polyunsaturated fats. We calculated nutrient densities of fats by dividing the energy contribution of each fat by total energy (22). We examined the cumulatively updated intakes by using averages of the intakes from all the available dietary assessments up to the start of each 2-y period at risk. Because glaucoma is a slowly progressing chronic disease and because cumulatively averaged intakes represent long-term diet with minimized within-person variation in dietary intake (23), we present these intakes as the primary analysis. In all analyses, we categorized fat nutrient densities into quintiles and compared risks by using the lowest quintile as the comparison group. In the initial steps to examine the relation between a potential risk factor for POAG and total fat intake, we modeled quintile of total fat intake as the outcome by using ordinal logistic regression, with age and other factors entered into the models. For each type of fat analyzed, we examined 2 dietary substitution multivariate models. For example, for saturated fat, the first model had quintiles of saturated fat and of total energy. The coefficients for saturated fat in this model can be interpreted as the effect of substituting a specific percentage of energy from saturated fat for the same percentage from all sources of energy except for saturated fat. In the second model, quintiles of all other sources of energy, except for carbohydrates, were added to the previous

4 758 KANG ET AL model (ie, protein, monounsaturated fat, trans polyunsaturated fat, and polyunsaturated fat). The coefficients in this second model can be interpreted as the effect of substituting a specific percentage of energy from saturated fat for the same percentage from carbohydrates. The second model allowed a direct comparison of the magnitude of the associations with POAG risk between different fat types. We derived incidence rates of POAG by dividing the number of cases by the number of person-years in each quintile. We adjusted the rates for age by using 5-y categories and calculated Mantel-Haenszel age-adjusted incidence rate ratios (RR) and their 95% CIs. We controlled for the possible confounding effects of known and potential risk factors for glaucoma by including them simultaneously in Cox proportional hazards analysis stratified by age in months and the 2-y period at risk (from one questionnaire to the next biennial questionnaire administration) (24). We performed tests for trend by including the median values within each category in multivariate models and evaluating the statistical significance ( 0.05). Variables considered for inclusion were African American heritage (yes or no), body mass index (in kg/m 2 ), alcohol intake (g/d), physical activity (quartiles of activity intensity/d), report of a physician exam, and selfreported history (yes or no) of a diagnosis of hypertension, diabetes, cataract, or age-related macular degeneration. We first analyzed the data from each cohort separately and performed tests for heterogeneity of the cohort-specific results to check for the appropriateness of pooling the results. Then, we pooled the results by using meta-analytic methods incorporating random effects (25). For certain variables of interest, restricted cubic splines (26) were fitted to the proportional hazards regression models to examine the possibility of nonlinear relations between the variable and POAG risk; tests for nonlinearity were performed. In addition, for specific variables, we were interested in testing any differences in variable estimates between normal and high-tension glaucoma. Whereas IOP-independent factors may contribute to glaucomatous optic neuropathy even for those with IOP well above the population statistical norm, IOP-dependent factors likely predominate (27). We hypothesized that a dietary fat composition that may lead to lower IOP would be more effective in patients with a tendency to develop high IOP and who also had a propensity to IOP-dependent optic nerve damage. Thus, we distinguished high-tension POAG from normal-tension POAG by determining those with maximum IOP or 21 mm Hg before VF loss. To maximize our power for detecting differences between the 2 outcomes, we pooled the raw data from the 2 cohorts. We first approximated the time-varying Cox models with pooled logistic regressions models, given the assumption that the incidence of POAG is rare in any given biennial follow-up period (28); we then used multivariate polychotomous logistic regression (29) with the 3 endpoints of normal-tension POAG, hightension POAG, and no POAG; and, finally, we calculated Lagrange multiplier statistics (29) testing differences. In all analyses, P values were two-sided with RESULTS During person-years of follow-up, we identified 474 cases of POAG. The distribution of potential risk factors for glaucoma by quintiles of total fat intake is presented in Table 1. The highest consumers of fat tended to have higher body mass indexes, a history of diabetes (particularly in men), greater smoking history, and greater caloric intake, whereas they were less likely to be of African American heritage, have a history of hypertension, be regular alcohol drinkers, be physically active, and have recently received an exam by a physician. We attempted to control for these differences by accounting for them as covariates in the multivariate analyses. In all primary analyses, we did not observe heterogeneity in the general trends with increased intake of fats between studies; thus, cohort-specific results were pooled and are presented as summaries of the data. In both age-adjusted analyses and all multivariate analyses, the risk of POAG was not significantly related to intakes of total fat, animal fat, vegetable fat, or cholesterol (Table 2). Among the different subtypes of fat, saturated and trans polyunsaturated fat intake were not important risk factors for POAG (Table 3). Monounsaturated fat and polyunsaturated fats tended to be weakly inversely related (for quartile 5 of monounsaturated fat intake versus quartile 1: RR 0.76; 95% CI: 0.56, 1.03; for quartile 5 of polyunsaturated fat intake versus quartile 1: RR 0.87; 95% CI: 0.66, 1.16), but neither association was statistically significant. When we examined the relation with substituting a given percentage of energy from these fats for the same quantity from carbohydrates, the statistical power was reduced but the rate ratios changed minimally. We observed only weak associations when we examined intakes of n 3 orn 6 fatty acids individually, but we found the ratio of n 3ton 6 fatty acids to be positively associated with risk (Table 4). Comparing the highest with the lowest quintile of this ratio, the RR was 1.49 (95% CI: 1.11, 2.01; P for linear trend 0.10). When total fat was added to the model such that the estimates represent the effect of increasing the ratio of n 3 to n 6 fat by substitution for other fats, the pooled estimate was unchanged. We observed that this association was more prominent for the risk of high-tension POAG (quartile 5 versus quartile 1: RR 1.68; 95% CI: 1.18, 2.39; P for linear trend 0.009) than for the risk of normal-tension POAG (Lagrange multiplier testing differences in estimates for high versus normal tension POAG: P 0.005). On the basis of the fitted restricted cubic splines and the tests of nonlinearity of this relation (P 0.52 in the NHS; P 0.37 in the HPFS), a simple linear relation appeared to be the best fit for these data. For all types of glaucoma, the pooled RR associated with a 0.1-unit increase in the ratio was 1.26 (95% CI: 0.94, 1.69; P for linear trend 0.12) and for high-tension glaucoma it was 1.40 (95% CI: 1.10, 1.77; P for linear trend 0.006). Because significant associations in ratio measures reflect an apparent effect modification between 2 variables, in exploratory analyses we assessed the cross-classification of total n 3 and n 6 fat intakes in relation to high-tension POAG. Because we were limited in case numbers for detailed cross-classifications, we created simple combinations of n 3 and n 6 fat intake by creating 8 approximately equal numbered categories: quartiles of one fat by above and below the median of the other fat. Results were qualitatively similar with quintiles. Among those with n 3 fat intake lower than the median, the risk of high-tension glaucoma for those in the highest compared with the lowest quartile of n 6 fat intake was lower by 57%, although the P for trend was not significant (RR 0.43; 95% CI: 0.21, 0.90; P for linear trend 0.10). Among those with high n 3 fat intake, the corresponding relative risk was 0.94 (95% CI: 0.63, 1.40; P for linear trend 0.85; P for interaction 0.17). Conversely, when n 3

5 TABLE 1 Age and age-standardized characteristics at baseline according to total dietary fat intake 1 FAT INTAKE AND GLAUCOMA 759 Quintile of total fat as a percentage of total energy 2 Characteristic Total fat (% of energy) Age (y) African American heritage (%) BMI (kg/m 2 ) History of hypertension (%) History of diabetes (%) Pack-years smoked (y) Alcohol intake 15 g/d (%) Energy intake (kcal) Highest quartile of physical activity (%) Self-reported cataract diagnosis (%) Self-reported macular degeneration diagnosis (%) History of physician exam in previous 2 y (%) Mean eye exams reported Data were age-standardized by using 5-y intervals of age at baseline (1980 for the Nurses Health study, 1986 for the Health Professionals Follow-Up Study). are participants of the Nurses Health Study, and men are participants of the Health Professionals Follow-Up Study. 2 All P values from tests for age-standardized differences across fat categories were 0.05, except for the following: history of diabetes (P 0.08 for women), self-reported cataract diagnosis (P 0.90 for women; P 0.21 for men), and self-reported macular degeneration diagnosis (P 0.71 for women; P 0.65 for men). 3 Values are median SD. 4 Values are x SD. 5 During follow-up, questions on eye exams were asked 4 times in both cohorts. fat intake according to lower or higher than median n 6 fat intake was examined, we observed no meaningful trends. For food analyses, we examined foods such as peanut butter and nuts that are high in n 6 fats and low in n 3 fats in both cohorts. We observed statistically significant inverse associations with 2 of 4 foods examined: potato or corn chips and French fried potatoes. The pooled estimates of the RR (95% C.I; P for linear trend) were 0.68 (0.36, 1.29; P trend 0.12) in those consuming nuts (1 oz, or 31 g) 1 time/d compared with 2 times/wk; 0.50 (0.16, 1.59; P trend 0.49) for 2 servings/d of peanut butter (1 tablespoon, or 16 g) compared with 5 times/ wk; 0.64 (0.45, 0.91; P trend 0.04) for 2 servings/wk of

6 760 KANG ET AL TABLE 2 Cohort-specific and pooled analyses of quintiles of fat intake in relation to risk of primary open-angle glaucoma 1 Quintile of fat intake Nutrient P trend 2 P het 3 Total fat Median (% of energy) Multivariate RR 1.00 (ref) 0.92 (0.65, 1.30) 1.13 (0.80, 1.59) 0.99 (0.69, 1.42) 0.96 (0.66, 1.40) 0.92 Median (% of energy) Multivariate RR 1.00 (ref) 0.82 (0.52, 1.31) 0.66 (0.39, 1.10) 1.04 (0.66, 1.65) 0.80 (0.49, 1.29) 0.56 Multivariate RR 1.00 (ref) 0.88 (0.67, 1.17) 0.89 (0.53, 1.51) 1.01 (0.76, 1.34) 0.90 (0.67, 1.21) Multivariate protein 1.00 (ref) 0.89 (0.67, 1.17) 0.90 (0.53, 1.54) 1.02 (0.77, 1.36) 0.90 (0.67, 1.21) Animal fat Median (% of energy) Multivariate RR 1.00 (ref) 1.12 (0.80, 1.58) 0.89 (0.61, 1.29) 1.21 (0.85, 1.71) 1.00 (0.69, 1.45) 0.64 Median (% of energy) Multivariate RR 1.00 (ref) 1.10 (0.69, 1.76) 1.16 (0.72, 1.87) 0.84 (0.50, 1.41) 1.13 (0.70, 1.83) 0.93 Multivariate RR 1.00 (ref) 1.12 (0.84, 1.47) 0.98 (0.73, 1.32) 1.06 (0.76, 1.49) 1.05 (0.78, 1.40) Multivariate protein 1.00 (ref) 1.08 (0.81, 1.44) 0.96 (0.69, 1.33) 1.03 (0.75, 1.41) 0.97 (0.69, 1.37) vegetable trans unsaturated fats Vegetable fat Median (% of energy) Multivariate RR 1.00 (ref) 0.82 (0.58, 1.16) 0.77 (0.54, 1.11) 1.04 (0.74, 1.46) 0.69 (0.47, 1.01) 0.20 Median (% of energy) Multivariate RR 1.00 (ref) 0.92 (0.57, 1.49) 0.99 (0.61, 1.58) 0.96 (0.59, 1.55) 0.84 (0.51, 1.37) 0.49 Multivariate RR 1.00 (ref) 0.85 (0.65, 1.13) 0.84 (0.63, 1.13) 1.01 (0.77, 1.34) 0.75 (0.55, 1.01) Multivariate protein 1.00 (ref) 0.83 (0.62, 1.11) 0.84 (0.60, 1.19) 1.03 (0.75, 1.41) 0.80 (0.56, 1.14) animal trans unsaturated fats Cholesterol Median (% of energy) Multivariate RR 1.00 (ref) 1.07 (0.75, 1.53) 1.02 (0.71, 1.47) 1.31 (0.92, 1.85) 0.84 (0.57, 1.25) 0.82 Median (% of energy) Multivariate RR 1.00 (ref) 0.78 (0.45, 1.34) 1.32 (0.81, 2.14) 1.19 (0.74, 1.93) 1.11 (0.67, 1.82) 0.38 Multivariate 1.00 (ref) 0.97 (0.72, 1.31) 1.12 (0.83, 1.50) 1.27 (0.96, 1.68) 0.94 (0.69, 1.28) Multivariate protein vegetable fat 1.00 (ref) 0.97 (0.71, 1.31) 1.11 (0.78, 1.58) 1.23 (0.91, 1.67) 0.91 (0.57, 1.44) All values (except P values) are relative risks (RRs); 95% CIs in parentheses. Intake values represent cumulatively updated averages. Multivariate RR analysis was based on Cox proportionals hazards controlled for African heritage (yes or no), diagnosis of diabetes in the current or previous cycle (yes or no), history of hypertension (yes or no), BMI (kg/m 2 ; indicators for 22, 22 23, 24 25, 26 27, 28), indicators for quartiles of daily physical activity, alcohol intake (indicators for 0, 1 4, 5 14, 15 29, 30 g/d), pack-years of smoking (indicators for 0, 1 9, 10 19, 20 29, 30 y), and quintiles of total energy intake. Data were pooled by using the Dersimonian and Laird meta-analytic methods with random effects. ref, referent. 2 The statistical significance of the pooled estimate for the medians of nutrients as continuous variables. 3 The statistical significance of the test for between-study heterogeneity in the continuous median variables used for testing trends. potato or corn chips [1 small bag (1 oz, or 31 g)] compared with 1 time/mo; 0.45 (0.25, 0.81; P trend 0.01) for 2 servings/wk of French fried potatoes (4 oz, or 124 g) compared with 1 time/mo. We found no material differences in secondary analyses that controlled for other predictors of more frequent eye exams (cataract diagnoses, macular degeneration diagnoses, report of physician exam, or history of eye exam reports). This was also true in analyses limited to those receiving eye exams for screening, in analyses only among those aged 65 y, and in analyses restricted to those consistently reporting eye exams during follow-up (data not shown).

7 FAT INTAKE AND GLAUCOMA 761 TABLE 3 Cohort-specific and pooled analyses of quintiles of intake of subtypes of fat in relation to risk of primary open-angle glaucoma 1 Quintile of fat intake Nutrient P trend 2 P het 3 Saturated fat Median (% of energy) Multivariate RR(A) 1.00 (ref) 1.05 (0.75, 1.49) 1.08 (0.76, 1.54) 1.02 (0.71, 1.47) 1.02 (0.70, 1.49) 0.76 Median (% of energy) Multivariate RR(A) 1.00 (ref) 0.93 (0.59, 1.47) 0.67 (0.41, 1.12) 0.81 (0.49, 1.32) 1.05 (0.66, 1.66) 0.97 Multivariate RR(A) 1.00 (ref) 1.01 (0.77, 1.33) 0.89 (0.56, 1.40) 0.94 (0.70, 1.26) 1.03 (0.77, 1.38) Multivariate RR(B) 1.00 (ref) 1.02 (0.75, 1.39) 0.94 (0.63, 1.40) 1.01 (0.68, 1.50) 1.25 (0.81, 1.95) Monounsaturated fat Median (% of energy) Multivariate RR(A) 1.00 (ref) 0.97 (0.69, 1.37) 1.05 (0.74, 1.48) 1.03 (0.72, 1.47) 0.80 (0.54, 1.17) 0.50 Median (% of energy) Multivariate RR(A) 1.00 (ref) 1.09 (0.69, 1.71) 0.49 (0.27, 0.87) 1.43 (0.91, 2.23) 0.69 (0.41, 1.16) 0.38 Multivariate RR(A) 1.00 (ref) 1.01 (0.77, 1.33) 0.74 (0.35, 1.56) 1.18 (0.86, 1.61) 0.76 (0.56, 1.03) Multivariate RR(B) 1.00 (ref) 0.97 (0.70, 1.34) 0.74 (0.40, 1.36) 1.08 (0.70, 1.67) 0.65 (0.38, 1.11) Polyunsaturated fat Median (% of energy) Multivariate RR(A) 1.00 (ref) 0.80 (0.57, 1.12) 0.89 (0.63, 1.25) 0.71 (0.50, 1.03) 0.83 (0.58, 1.19) 0.25 Median (% of energy) Multivariate RR(A) 1.00 (ref) 0.83 (0.51, 1.34) 0.79 (0.48, 1.30) 0.89 (0.56, 1.43) 0.95 (0.60, 1.51) 0.94 Multivariate RR(A) 1.00 (ref) 0.81 (0.61, 1.07) 0.85 (0.64, 1.13) 0.78 (0.58, 1.04) 0.87 (0.66, 1.16) Multivariate RR(B) 1.00 (ref) 0.80 (0.60, 1.07) 0.86 (0.64, 1.17) 0.82 (0.59, 1.12) 1.00 (0.71, 1.41) trans Unsaturated fat Median (% of energy) Multivariate RR(A) 1.00 (ref) 1.29 (0.91, 1.81) 1.20 (0.84, 1.71) 1.17 (0.82, 1.69) 0.89 (0.60, 1.32) 0.53 Median (% of energy) Multivariate RR(A) 1.00 (ref) 1.04 (0.65, 1.67) 1.18 (0.74, 1.87) 0.92 (0.56, 1.50) 0.77 (0.46, 1.29) 0.25 Multivariate RR(A) 1.00 (ref) 1.19 (0.90, 1.58) 1.19 (0.90, 1.58) 1.07 (0.80, 1.44) 0.84 (0.61, 1.15) Multivariate RR(B) 1.00 (ref) 1.28 (0.95, 1.72) 1.32 (0.96, 1.82) 1.23 (0.87, 1.75) 1.02 (0.68, 1.52) All values (except P values) are relative risks (RRs); 95% CIs in parentheses. Intake values represent cumulatively updated averages. Multivariate RR(A) analysis was based on Cox proportionals hazards controlled for African heritage (yes or no), diagnosis of diabetes in the current or previous cycle (yes or no), history of hypertension (yes or no), BMI (kg/m 2 ; indicators for 22, 22 23, 24 25, 26 27, 28), indicators for quartiles of daily physical activity, alcohol intake (indicators for 0, 1 4, 5 14, 15 29, 30 g/d), pack-years of smoking (indicators for 0, 1 9, 10 19, 20 29, 30 y), and quintiles of total energy intake. Multivariate RR(B) analysis was additionally controlled for protein and other fats in the table. Data were pooled by using the Dersimonian and Laird meta-analytic methods with random effects. 2 The statistical significance of the pooled estimate for the medians of nutrients as continuous variables. 3 The statistical significance of the test for between-study heterogeneity in the continuous median variables used for testing trends. DISCUSSION In this large prospective study, although we did not find that intakes of specific types of fat were independently associated with the development of POAG, we found that the ratio of n 3 to n 6 fats was related to the risk of POAG, and more strongly so for high-tension glaucoma. The stratified analyses suggested that the association between the ratio and high-tension glaucoma arose mainly from the strong inverse association with greater n 6 fat intake in the presence of low n 3 fat intake, and not with greater n 3 fat intake in the presence of low n 6 fat intake. The effect of the relative intakes of the 2 types of polyunsaturated fat on the relative tissue concentrations of the progenitors of the n 3 and n 6 eicosanoids has been well elucidated. An empirical formula has been derived that predicts the relative proportions of the 2 progenitors by using only dietary fat composition (15). The ratio of n 3ton 6 fatty acids is a component

8 762 KANG ET AL TABLE 4 Cohort-specific and pooled analyses of quintiles of polyunsaturated fat intake in relation to risk of primary open-angle glaucoma (POAG) 1 Quintile of fat intake Nutrient P trend 2 P het 3 Total n 3 4 Median (% of energy) Multivariate RR(A) 1.00 (ref) 0.96 (0.66, 1.40) 1.11 (0.77, 1.60) 1.30 (0.91, 1.85) 1.01 (0.69, 1.48) 0.56 Median (% of energy) Multivariate RR(A) 1.00 (ref) 1.38 (0.85, 2.24) 0.71 (0.41, 1.24) 1.57 (0.98, 2.52) 0.99 (0.59, 1.65) 0.98 Multivariate RR(A) 1.00 (ref) 1.11 (0.79, 1.57) 0.94 (0.61, 1.44) 1.39 (1.04, 1.84) 1.00 (0.74, 1.36) Multivariate RR(B) 1.00 (ref) 1.15 (0.85, 1.56) 0.98 (0.56, 1.71) 1.52 (1.11, 2.07) 1.10 (0.78, 1.56) Total n 6 5 Median (% of energy) Multivariate RR(A) 1.00 (ref) 0.92 (0.66, 1.29) 0.86 (0.60, 1.22) 1.02 (0.72, 1.43) 0.74 (0.50, 1.08) 0.22 Median (% of energy) Multivariate RR(A) 1.00 (ref) 0.69 (0.42, 1.12) 0.82 (0.51, 1.31) 0.95 (0.61, 1.49) 0.75 (0.47, 1.21) 0.50 Multivariate RR(A) 1.00 (ref) 0.84 (0.63, 1.11) 0.84 (0.63, 1.12) 0.99 (0.76, 1.31) 0.74 (0.55, 1.00) Multivariate RR(B) 1.00 (ref) 0.83 (0.63, 1.11) 0.84 (0.62, 1.14) 1.02 (0.75, 1.40) 0.79 (0.55, 1.14) n 3:n 6 Median Multivariate RR(A) 1.00 (ref) 1.30 (0.87, 1.95) 1.23 (0.82, 1.85) 1.26 (0.84, 1.87) 1.67 (1.15, 2.44) 0.01 Multivariate RR (C) 1.00 (ref) 1.31 (0.87, 1.96) 1.24 (0.83, 1.87) 1.27 (0.85, 1.91) 1.73 (1.17, 2.54) Median Multivariate RR(A) 1.00 (ref) 1.13 (0.68, 1.86) 1.26 (0.77, 2.06) 0.83 (0.49, 1.41) 1.24 (0.77, 2.00) 0.58 Multivariate RR(C) 1.00 (ref) 1.13 (0.68, 1.86) 1.25 (0.76, 2.05) 0.81 (0.47, 1.40) 1.18 (0.70, 1.98) 0.54 Multivariate RR(A) 1.00 (ref) 1.23 (0.90, 1.68) 1.24 (0.91, 1.70) 1.06 (0.71, 1.58) 1.49 (1.11, 2.01) Multivariate RR (C) 1.00 (ref) 1.23 (0.90, 1.69) 1.25 (0.91, 1.71) 1.06 (0.69, 1.63) 1.49 (1.03, 2.14) High-tension POAG (ref) 1.33 (0.93, 1.91) 1.22 (0.84, 1.76) 1.03 (0.61, 1.76) 1.68 (1.18, 2.40) Normal-tension POAG (ref) 0.97 (0.50, 1.85) 1.32 (0.71, 2.44) 1.13 (0.60, 2.13) 0.82 (0.20, 3.47) All values (except P values) are relative risks (RRs); 95% CIs in parentheses. Intake values represent cumulatively updated averages. Multivariate RR(A) analysis was based on Cox proportionals hazards controlled for African heritage (yes or no), diagnosis of diabetes in the current or previous cycle (yes or no), history of hypertension (yes or no), BMI (kg/m 2 ; indicators for 22, 22 23, 24 25, 26 27, 28), indicators for quartiles of daily physical activity, alcohol intake (indicators for 0, 1 4, 5 14, 15 29, 30 g/d), pack-years of smoking (indicators for 0, 1 9, 10 19, 20 29, 30 y), and quintiles of total energy intake. Multivariate RR(B) analysis was additionally controlled for protein, saturated fat, monounsaturated fat, trans unsaturated fat, and other fats in this table. Multivariate RR(C) Multivariate RR(A) total fat. Data were pooled by using the Dersimonian and Laird meta-analytic methods with random effects. 2 The statistical significance of the pooled estimate for the medians of nutrients as continuous variables. 3 The statistical significance of the test for between-study heterogeneity in the continuous median variables used for testing trends. 4 The sum of 22:5n 3, -linolenic acid, eicosapentaenoic acid, and docosahexaenoic acid. 5 The sum of linoleic acid and arachidonic acid. 6 Separate multivariate RR(C) models for high-tension glaucoma (maximum intraocular pressure 21 mm Hg; n 221 women, 137 men) and normal-tension glaucoma (maximum intraocular pressure 21 mm Hg; n 80 women, 36 men). of the formulas, and there is a high inverse correlation with the relative tissue concentrations of highly unsaturated n 6 fatty acids such as arachidonic acid (15). Our study findings are thus consistent with the hypothesis that greater dietary n 6 fat intake leads to greater availability of the n 6 prostaglandins (such as prostaglandin F 2 ), which may help to maintain IOP at levels that are less harmful to the optic nerve and thereby reduce the occurrence of POAG. Whereas prostaglandins have been valued mostly as therapeutic agents for glaucoma patients, these results suggest that alterations in the endogenous production or variations in the physiologic concentrations of ocular prostaglandins in healthy populations may be related to the likelihood of developing glaucoma. Also, the observation that the association was stronger with high-tension glaucoma than with normal-tension glaucoma is consistent with the hypothesis that pressuredependent mechanisms would predominate for high-tension glaucoma. Thus, any IOP-lowering effects of dietary fat intake would appear stronger in relation to this group. However, because this was the first study of fat intake and POAG risk and several individual fats were examined, this result could have been a chance finding. Future studies are necessary to confirm this association.

9 FAT INTAKE AND GLAUCOMA 763 To our knowledge, the association between fat intake and risk of developing POAG has not been examined previous to our study. However, there have been 2 studies of the effect of fat-free diets on IOP (30, 31) and one of supplementation with a combination of docosahexaenoic acid, vitamin E, and B complex vitamins compared with supplementation with B complex vitamins only (32). In both studies of fat-free diets [one of a fat-free rice diet (30) and the other of fat-free parenteral nutrition (31)], there were significant average reductions in IOP in the range of 4 7 mm Hg. However, these findings are hard to interpret and may have limited applicability because both studies were based on patients with severe morbidities [patients with uncontrolled systemic hypertension secondary to end-stage renal failure (30) and patients who needed to be fed parenterally (31)]. In the study of glaucoma patients (32), improvements in the perimetric indexes were observed only in the group with docosahexaenoic acid and vitamin E, without meaningful changes in IOP. Although it is a n 3 fatty acid, docosahexaenoic acid might have beneficial effects, because it is abundant in the brain (especially in membranes with synaptic function) (33) and it deserves more study in relation to glaucomatous neuropathy. The strengths of our study include its prospective nature and the large number of cases. The prospective design makes recall or selection biases less likely, and the high follow-up rates also minimize biases. Multiple dietary assessments reduced the possibility of measurement error and allowed us to examine the effects of nutrients over time. Finally, we were able to control for established POAG risk factors, such as age and African heritage. Specific limitations of our study include the possibility of confounding by unmeasured variables, such as family history of POAG. Although we did not assess family history of POAG, it is unlikely that dietary fat intake varies substantially by one s family history, because the hypothesis examined in this study has been little studied We could not conduct repeated eye exams in these large cohorts; thus, our method of case ascertainment may have had low sensitivity. However, low sensitivity in the case definition does not bias the rate ratio estimates if the ascertainment method is not related to exposure and the case definition is highly specific (ie, few false-positive results) (34). We believe our case definition would minimize false-positive results among identified cases because we required confirmation of the original diagnosis on 2 reliable VF tests by 2 independent reviewers. Some residual detection bias issues may have occurred if the eye exams reported by those with the lowest intakes were not as comprehensive and thus less likely to detect glaucoma, if present. However, the results of the various sensitivity analyses with greater restrictions on the allowable person-time did not change substantially from the primary results. Also, because our participants are health professionals, their levels of screening would be similar to or higher than what would be observed in a general population sample. Thus, any such bias would likely be modest. Another consideration is that the participants of both the NHS and the HPFS are 90% white. Thus, our results may not be widely generalizable to populations with higher percentages of minorities, particularly those of African or Caribbean heritage who are at greater risk of POAG (2, 3). In addition, because of the insidious nature of the disease, the heterogeneous nature of eye exams for detecting POAG, and our requirement of reproducible VF defects, our protocol may have resulted in a greater percentage of cases with moderate to severe disease than would have been detected in a direct standardized ophthalmologic survey of this cohort. Thus, our results do not apply directly to the risk of developing ocular hypertension or very early glaucoma. In conclusion, in our analysis of data from 2 large prospective studies, we found a diet high in n 6 and low in n 3 polyunsaturated fats to be associated with a reduced occurrence of POAG, particularly high-tension POAG. Because this is the first examination of this relation, further studies are needed to corroborate these findings. We extend our gratitude to Maureen Ireland, Kerry Pillsworth, Karen Corsano, and Steven Stuart for their unfailing technical assistance and to Frank Speizer, the original principal investigator, and Graham Colditz, the current principal investigator, of the Nurses Health Study for their input. Each author substantially contributed to this manuscript. JHK participated in collecting and analyzing the data and writing the manuscript; WCW, KME, and SEH conceived and planned the glaucoma studies for the Health Professionals Follow-Up Study and the Nurses Health Study; LRP and NF formulated and executed the case confirmation protocol in both cohorts; and BAR and WCW provided expert guidance in the complex statistical analysis. All authors provided critical input to data interpretation and manuscript preparation. None of the authors had a financial or personal interest in any company or organization sponsoring the research, including advisory board affiliations. REFERENCES 1. Rodriguez J, Sanchez R, Munoz B, et al. Causes of blindness and visual impairment in a population-based sample of U. S. Hispanics. Ophthalmology 2002;109: Leske MC, Connell AMS, Wu S-Y, et al. Incidence of open-angle glaucoma: the Barbados Eye Studies. Arch Ophthalmol 2001;119: Sommer A, Tielsch JM, Katz J, et al. Racial differences in the causespecific prevalence of blindness in east Baltimore. N Engl J Med 1991; 325: Collaborative Normal-Tension Glaucoma Study Group. The effectiveness of intraocular pressure reduction in the treatment of normal-tension glaucoma. Am J Ophthalmol 1998;126: Kass MA, Heuer DK, Higginbotham EJ, et al. The Ocular Hypertension Treatment Study: a randomized trial determines that topical ocular hypotensive medication delays or prevents the onset of primary open-angle glaucoma. Arch Ophthalmol 2002;120: Alward WLM. Medical management of glaucoma. N Engl J Med 1998; 339: Stjernschantz JW. From PGF 2 -isopropyl ester to latanoprost: a review of the development of Xalatan. Invest Ophthalmol Vis Sci 2001;42: Camras CB. Initial clinical studies with prostaglandins and their analogues. Surv Ophthalmol 1997;41(suppl):S Schachtschabel U, Lindsey JD, Weinreb RN. The mechanism of action of prostaglandins on uveoscleral outflow. Curr Opin Ophthalmol 2000; 11: Gabelt BT. The effect of prostaglandin F 2 on trabecular outflow facility in cynomolgus monkeys. Exp Eye Res 1990;51: Dijkstra BG, Schneeman A, Hoyng PF. Flow after prostaglandin E 1 is mediated by receptor-coupled adenylyl cyclase in human anterior segments. Invest Ophthalmol Vis Sci 1999;40: Wang R-F, Lee P-Y, Mittag T, Podos S, Serle J, Becker B. Effect of 8-iso prostaglandin E 2 on aqueous humor dynamics in monkeys. Arch Ophthalmol 1998;116: James MJ, Gibson RA, Cleland LG. Dietary polyunsaturated fatty acids and inflammatory mediator production. Am J Clin Nutr 2000;71(suppl): 343S 8S. 14. Innis SM. Essential dietary lipids. In: Ziegler EE, Filer LJ, eds. Present knowledge in nutrition. 7th ed. Washington, DC: ILSI Press, 1996: Lands WEM, Libelt B, Morris A, et al. Maintenance of lower proportions of (n 6) eicosanoid precursors in phospholipids of human plasma in response to added dietary (n 3) fatty acids. Biochim Biophys Acta 1992;1180: