ORIGINAL CONTRIBUTIONS

Transcription

1 American Journal of Epidemiology Copyright O 1999 by The Johns Hopkins University School of Hygiene and Public Health All rights reserved Vol. 150, No. 4 Printed In USA. ORIGINAL CONTRIBUTIONS Recent Alcohol Intake as Estimated by the Health Habits and History Questionnaire, the Harvard Semiquantitative Food Frequency Questionnaire, and a More Detailed Alcohol Intake Questionnaire Susan E. McCann, 1 James R. Marshall, 2 Maurizio Trevisan, 1 Marcia Russell, 3 Paola Muti, 1 Nina Markovic, 4 Arthur W. K. Chan, 3 and Jo L. Freudenheim 1 Epidemiologic studies often rely on food frequency questionnaires (FFQs) to collect information on alcoholic beverage intake. However, estimation of alcohol intake using FFQs may be of some concern because of limited questions concerning alcohol intake. The authors compared estimates of alcohol intake during the months prior to interview obtained from the Health Habits and History Questionnaire and the Harvard Semiquantitative Food Frequency Questionnaire with those from a more extensive alcohol questionnaire, the Drinking Pattern Questionnaire, among 133 healthy subjects (75 men, 58 women) aged years, residents of western New York State. Data were collected in 1995 during two separate interviewer-administered computerassisted interviews conducted approximately 2 weeks apart. For each questionnaire, average daily ounces (1 oz = 30 ml) of alcohol intake from alcoholic beverages were calculated as the product of the reported beveragespecific drink size (ounces) and the average daily frequency of intake multiplied by a factor representing the percentage of alcohol provided by each beverage. Estimates of total alcohol and liquor intake, but not of beer and wine intake, tended to be higher for the Drinking Pattern Questionnaire compared with the FFQs. Spearman's correlation coefficients ranged from 0.69 to. These results suggest that although the Drinking Pattern Questionnaire produced higher estimates than either FFQ, both FFQs provide a reasonable ranking of participants' alcohol intake. Am J Epidemiol 1999; 150: alcohol drinking; alcoholic beverages; diet surveys; questionnaires; reproducibility of results While there is considerable interest in evaluating the possible effects of alcohol on the risk of chronic disease, estimation of associations between alcohol and disease outcomes necessarily depends on adequate quantification of alcohol intake. Epidemiologic studies often rely on the use of food frequency instruments to collect dietary information, including intake of alcoholic beverages (1-6). Food frequency questionnaires (FFQs) consist of a defined list of foods and beverages, and a participant is asked how often over a spec- Received for publication September 1, 1998, and accepted for publication February 8, Abbreviations: CLDH, Cognitive Lifetime Drinking History; DPQ, Drinking Pattern Questionnaire; FFQ, food frequency questionnaire; HFFQ, Harvard Semiquantitative Food Frequency Questionnaire; HHHQ, Health Habits and History Questionnaire. 1 Department of Social and Preventive Medicine, State University of New York at Buffalo, Buffalo, NY. 2 Arizona Cancer Center, Tucson, AZ. 3 Research Institute on Addictions, Buffalo, NY. 4 Department of Epidemiology, Graduate School of Public Health, University of Pittsburgh, Pittsburgh, PA. Reprint requests to Dr. Susan E. McCann, Department of Social and Preventive Medicine, 270 Farber Hall, SUNY Buffalo, Buffalo, NY ified time period each item was consumed. Two food frequency instruments have been used frequently in a wide variety of settings. The first, developed by researchers at the National Cancer Institute under the leadership of Dr. Gladys Block, is referred to as the Health Habits and History Questionnaire (HHHQ) (7). The other, developed as part of the Nurses' Health Study and the Male Health Professionals Study, is referred to as the Harvard Semiquantitative Food Frequency Questionnaire (HFFQ) (8). Both instruments include questions regarding intake of alcoholic beverages. Although both the HHHQ and HFFQ have been shown to estimate nutrient intake reasonably well when they are compared with food records in a wide variety of settings (9-19), estimation of alcohol intake by these methods may pose particular concerns. On both instruments, questions concerning alcohol intake are limited to the overall amount and frequency of intake of the three major beverage types: beer, wine, and liquor. The generality of the questions requires the participant to average the amount and frequency of all beverages consumed across the time period of interest. 334

2 Measurement of Alcohol Intake 335 No opportunity exists for a description of patterns of intake or periods during which alcohol may be consumed in larger amounts than usual. Persons who drink small amounts on a regular basis are not distinguished from those who drink larger amounts less frequently. For some disease processes, the pattern of intake may be more important than average intake. Furthermore, intake among some drinkers may be underestimated by the global approach generally taken in FFQs (20, 21). More detailed questionnaires that measure drinking patterns exist (20-22). The Lifetime Drinking History developed by Skinner and Sheu (22) assessed drinking patterns by repeating questions on quantity and frequency for each beverage each time a change in pattern was reported. Researchers at the State University of New York at Buffalo and the Research Institute on Addictions in Buffalo, New York, have developed a detailed questionnaire to measure lifetime alcohol intake, the Cognitive Lifetime Drinking History (CLDH), which integrates the Lifetime Drinking History with cognitive interviewing techniques (23). During the CLDH interview, a participant is oriented to usual activities during which drinking occurs, then asked detailed questions about such drinking (23). Allowing participants to report variability in drinking patterns relieves them of the responsibility of averaging intakes and provides a description of individual patterns of drinking. When compared with food records, FFQs have been shown to produce reasonably valid estimates of alcohol intake, with correlations of about (24-27). Although several of these studies have compared estimates of alcohol intake obtained from FFQs with those calculated from food records, to our knowledge none has compared FFQs with more detailed alcoholspecific questionnaires. The purpose of this study was to compare the average daily volume of alcohol consumed during the months prior to interview as calculated from two different FFQs, the Health Habits and History Questionnaire (HHHQ) and the Harvard Semiquantitative Food Frequency Questionnaire (HFFQ), with that calculated from a more detailed alcohol questionnaire. As the CLDH was designed to measure lifetime alcohol intake, we used for comparison the introductory portion of the CLDH, the Drinking Pattern Questionnaire (DPQ), which queries alcohol intake during the months preceding the interview but maintains the structure and detail of the lifetime questionnaire. MATERIALS AND METHODS Data were collected for a pilot study conducted as part of a large study of lifetime alcohol intake, lung cancer, and myocardial infarction in western New York State. The pilot study included both myocardial infarction cases and controls. Only the data regarding healthy controls are included here. During a computer-assisted personal interview completed during two separate visits approximately 2 weeks apart, all participants were asked questions regarding basic demographic information and health-related habits such as cigarette smoking and intake of vitamin supplements. Each interview consisted of the DPQ and either the HHHQ or the HFFQ. Participants completed both the HHHQ and HFFQ during the two interviews; assignment of the two FFQs to the two interviews was random. Subjects Controls were persons who were aged years and were residents of Erie and Niagara Counties. They were selected from the general population by using driver's license bureau lists for residents less than age 65 years and Health Care Financing Administration lists for those aged 65 years or older. Participants included only those persons who had consumed at least 12 drinks in at least 1 year during their lifetimes. During recruitment, 690 controls were contacted for participation. Of these, 147 (21 percent) agreed to participate in this fairly extensive study. A small percentage of data (n = 4) was not usable, and 10 participants withdrew after the first interview. The present analyses were conducted by including the 75 men and 58 women participants whose data were complete for all three questionnaires. Measures Three separate instruments were used to gather data concerning intake of alcoholic beverages during the months prior to the interview: the introductory section of the CLDH (the DPQ), the HHHQ, and the HFFQ. For the DPQ, participants were asked to recall their usual activities on specific days of the week during the past year and to think about whether they drank alcohol during those activities. They were then asked specific questions about their frequency of intake of four major types of alcoholic beverages: beer, wine, wine cooler, and liquor. For each beverage, participants defined their usual drink size in ounces (1 oz = 30 ml) in relation to sample glasses with drink sizes marked on the sides. Furthermore, they were asked the number of drinks they consumed per drinking occasion. Participants reporting intake of more than one type of beverage were asked what proportion of all beverages they consumed was accounted for by each reported beverage. Separate questions were used for occasional (less than weekly) versus regular (at least weekly) consumers of alcohol as well as for occasions

3 336 McCann et al. on which alcohol was consumed in greater than usual amounts. Regular consumers of alcohol were asked detailed questions about their frequency of intake and the number of drinks per drinking occasion for each beverage type on Fridays, Saturdays, Sundays, and weekdays for all reported beverages. Additionally, detailed questions concerning frequency of intake and number of drinks per drinking occasion were asked about periods during which alcohol was consumed in "more than usual" amounts. Both the HHHQ and HFFQ queried participants about their frequency of intake of beer, wine, and liquor during the past months. However, the two instruments differed slightly concerning how these questions were asked. Beverage size and frequency questions were separate on the HHHQ. Beverage size was defined by stating a medium portion size, and participants were asked whether, in comparison, their portion was small (one-half the stated size), medium (the stated size), or large (one and one-half the stated size). On the HHHQ, the nine frequency categories ranged from never, or less than once per month, to six or more times per day. On the HFFQ, beverage size and frequency questions were combined. Participants were asked how often they consumed a specified amount of each beverage type. The nine frequency categories were the same as those used in the HHHQ. However, the HFFQ included separate questions about white and red wines. Neither FFQ had an item for wine coolers. Calculation of alcohol Intake For the FFQs, ounces of alcohol consumed of each beverage type were calculated as the product of the reported (HHHQ) or specified (HFFQ) beveragespecific drink size in ounces and the frequency of intake multiplied by a factor representing the percentage of alcohol provided by each beverage. The factors used were 0.048, 0.12, 0.04, and 0.40, for beer, wine, wine cooler, and liquor, respectively. For the DPQ, ounces of alcohol consumed of each beverage type were calculated as the product of the reported beverage-specific drink size in ounces and the number of drinks per drinking occasion and the frequency of intake multiplied by the alcohol factors. Ounces of alcohol consumed from alcoholic beverages were expressed as average daily intake. Statistical analyses Variables examined were average daily ounces of alcohol consumed from beer, wine, and liquor and average daily total ounces of alcohol consumed as calculated from each of the questionnaires. Analyses excluded alcohol from wine cooler, as only the DPQ included this beverage and few participants reported drinking it. Because participants received the DPQ twice but the HHHQ and HFFQ only once, analyses were conducted by pairing the FFQ with the DPQ given during the same interview. and Spearman's correlation coefficients were calculated. Because reported intakes were skewed toward the lower end of the distribution, a natural logarithmic transformation was performed before the correlation coefficients were calculated. Mean differences in intake according to the three questionnaires were calculated with paired t tests. The possible effects of age, race, education, marital status, and gender as confounding factors were tested by using multivariate analysis of variance for repeated measures. Models were tested separately for the and DPQ- HFFQ pairs, with each possible confounder entered individually as a covariate. Further analyses were stratified by factors found to be significant in the multivariate analysis of variance with repeated measures analyses. RESULTS Participants ranged in age from 35 to 73 years (mean age, 60.6 years). They were generally well educated (mean number of years of education, 14.3), and 95 percent were White. With the exception of age, the demographic characteristics were comparable for men and women; men were slightly older than women (mean age 62 (standard deviation, 7.0) and 59 (standard deviation, 7.5) years, respectively). For visit one and visit two, the test-retest reliability of the DPQ estimates of alcohol intake during the months prior to interview is shown in table 1. Repeatability for the two administrations of this section of the DPQ was for total alcohol intake. Exclusion of those participants who reported at both visits no alcohol intake during the referent time period TABLE 1. Test-retest reliability of the DPQ* estimates of alcohol intake during the months prior to Interview, visit one and visit two, western New York State, 1995 Total alcohol Beer Wine Liquor All controls.133) (" = con-elation coefficient % confidence Interval, , ,, 0.89 Excluding nondrfnkeret (n = 86) correlation coefficient % confidence Interval 0.76, , ,, 0.90 * DPQ, Drinking Pattern Questionnaire, t Nondrinkers, participants who during both interviews reported no alcohol intake during the past months.

4 Measurement of Alcohol Intake 337 had minimal effects on the reliability of the DPQ between visit one and visit two. In multivariate analyses of variance examining several possible confounders (gender, age, education, race, and marital status), gender was the only factor found to be statistically significant in the relation between the DPQ and either the HHHQ or the HFFQ (t = 3.5, p = and t = 3.3, p = 0.001, respectively). All further analyses were stratified by gender. Data on average daily total and beverage-specific ounces of alcohol consumed during the months prior to interview, as estimated by the three pairs of questionnaires, are shown in table 2. The DPQ consistently produced somewhat higher estimates than either the HHHQ or the HFFQ of total and beverage-specific average daily intake of alcohol. The discrepancies between methods were generally greatest between the DPQ and the HFFQ, with the exception of wine. Average differences between the questionnaire pairs regarding total and beverage-specific average daily ounces of alcohol were greater among men than women, especially concerning liquor intake. When the HHHQ and the HFFQ were compared, the HHHQ tended to produce higher estimates of total and beverage-specific average daily ounces of alcohol consumed. Correlations between average daily ounces of alcohol consumed, as measured by the DPQ, HHHQ, and HFFQ, are shown in table 3. Because exclusion of nondrinkers from the analyses had essentially no effect on the correlations, results are shown for the entire sample. For men and women together, good agreement was demonstrated for average daily total as well as beverage-specific ounces of alcohol intake for all three pairs of questionnaires, with r values between 0.71 and The comparative ability of the three instruments to rank participants on alcohol intake was assessed with Spearman's correlation coefficients. As shown in table 3, the three pairs of questionnaires performed comparably in ranking participants on total as well as beverage-specific average daily ounces of alcohol consumed, with Spearman's correlations on the order of DISCUSSION Measurement of alcohol intake presents unique challenges when traditional dietary assessment methods are used. The brevity and close-ended nature of food frequency instruments permit efficient collection and processing of data from a large number of subjects, a desirable characteristic for a data collection instrument used in epidemiologic studies. However, FFQs assume that average exposure to a dietary component over a time period adequately approximates actual exposure as it might relate to subsequent development of a chronic disease such as heart disease or cancer. In fact, for an exposure such as alcohol, a constant, moderate amount consumed over time might produce a different physiologic response than would larger amounts consumed sporadically. To our knowledge, no comparisons have been made between FFQs and highly detailed questionnaires designed specifically to collect data on alcohol intake. Although the HHHQ and HFFQ performed comparably to the DPQ in ranking participants on total as well as beverage-specific ounces of alcohol consumed per day during the months prior to interview, the more detailed DPQ consistently produced higher estimates of intake than did either of the FFQs. This finding is consistent with the fact that FFQs do not assess variability in the number of drinks consumed per occasion and that a substantial number of participants (74 percent) reported consuming more than one drink per occasion when given that option on the DPQ. Furthermore, the current study was conducted in a relatively older population, and it has been well established that the number of drinks consumed per occasion tends to decrease with age (28, 29). Therefore, the discrepancy between total alcohol intake based on the DPQ versus the FFQs may be larger in studies of alcohol intake conducted in younger populations or in younger persons in the same population. The largest discrepancies in alcohol intake between the DPQ and either the HHHQ or the HFFQ occurred for liquor intake among men. On both FFQs, questions concerning liquor intake were the least detailed of the beverage-specific questions. Because the DPQ asks the same sequence of detailed questions for each beverage type, it seems likely that estimates of the intake of liquor would show the largest differences. That FFQs may underestimate alcohol intake needs to be considered when attempts are made to develop public health recommendations based on findings from studies of alcohol and chronic disease that use these instruments. On the other hand, given these data, it was not possible to assess the validity of the measurement of alcohol intake using these three instruments, as we had no direct measure of intake. It is possible that the estimation differences we observed for pairs of the three instruments could be explained partly by overestimation by the DPQ as well as by underestimation by the FFQs. We found that the differences in estimated alcohol intake between the DPQ and either the HHHQ or the HFFQ were, in general, smaller for women than for men. As shown in table 2, women reported drinking smaller amounts than men did. Webb et al. (26) compared two methods of measuring alcohol intake and

5 u TABLE 2. Mean intake and mean difference In intake of daily ounces* of total and beverage-specific alcohol, as estimated by the DPQ.t HHHQ.t and HFFQ.t western New York State, 1995 DPQ MOD moan uiiana \ouumaiu uaviauun; HHHQ HFFQ Mean 95% Clt Mean difference Mean 95% Cl Mean 95% Cl 3 1 Total alcohol Beer Wine Liquor 0.42(1.0) 0.64(1.3) 0.15 (0.28) 0.20 (0.56) 0.33 (0.70) 0.04 (0.14) 0.08 (0.20) 0.07(0.21) 0.08(0.19) 0.15() 0.24(1.0) 0.03 (0.09) 0.28 (0.57) 0.41 (0.72) 0.11 (0.17) 0.19 (0.54) 0.31 (0.69) 0.03 (0.08) 0.06(0.12) 0.05(0.13) 0.07(0.12) 0.04(0.12) 0.06(0.16) (0.05) 0.23 (0.40) 0.33 (0.49) 0.11 (0.15) 0.13(0.35) 0.21 (0.44) (0.04) 0.08 (0.20) 0.09 (0.25) 0.07(0.12) (0.06) 0.03 (0.07) (0.05) ,0.28 -, , , ,0.17 -, , , , -, , , ounce = 30 ml. t DPQ, Drinking Pattern Questionnaire; HHHQ, Health Habits and History Questionnaire; HFFQ, Harvard Semiquantitative Food Frequency Questionnaire; Cl, confidence interval , 0.34, , , , , -, , -, , , , , , , ,0.13 -, , -0.05, , - -, 0.00, 0.03, , 8 CD CD CO

6 Measurement of Alcohol Intake 339 TABLE 3. and Spearman's correlation coefficients for mean dally total and beverage-specific ounces* of alcoholt Intake In the months prior to Interview by the DPQ4 HHHQ.t and HFFQ.t western New York State, 1995 Spearman's Spearman's Spearman's Total alcohol Beer (n = 133) (n = 75) (n = 58) Wine Liquor ounce = 30 ml. t Total and beverage-specific ounces of alcohol transformed with natural logarithmic transformation before calculation of correlation coefficients; all correlations were statistically significant at p < t- DPQ, Drinking Pattern Questionnaire; HHHQ, Health Habits and History Questionnaire; HFFQ, Harvard Semiquantjtative Food Frequency Questionnaire. found that reliability was lower among heavy drinkers. Although few of the participants in this study could be classified as heavy drinkers, the differences in intake between men and women may be partly explained by this effect. It may also be that women are better able to estimate intake. In our sample, the test-retest reliability of the DPQ was high between visit one and visit two, with correlations between 0.80 and 0.90, even when we excluded participants who reported no alcohol intake. The high level of repeatability of the DPQ lends credence to the ability of that instrument to describe drinking patterns. However, the interval between visits was fairly short (an average of 2 weeks). Responses to the second interview may have been influenced by the recency of the responses to the first. Furthermore, participants were interviewed during each visit regarding their lifetime alcohol intake and therefore were well prompted to remember their drinking habits. This format may have increased the comparability of all three methods. Finally, repetition of response errors related to the specific types of questions used in the DPQ could have contributed to the high repeatability observed in these data. We were unable to assess the test-retest reliability of either the HHHQ or the HFFQ because each of these FFQs was administered only once. However, both instruments have been assessed for their validity and reliability in a wide variety of settings and have been shown to be reasonably reliable in estimating nutrient intake (8-19). Although, to our knowledge, the reliability of the measurement of alcohol intake has not been assessed for the HHHQ, the HFFQ has been reported to measure alcohol intake reliably, with correlations of about 0.90 (27). A potential limitation of our study was the fairly small sample size that resulted from the low rate of participation among eligible controls. While poor participation could affect the generalizability of these results, we obtained between-instrument correlation coefficients that were similar to those published previously for the reliability of measurement of alcohol intake using FFQs versus food records (24-27). In addition, our study included a sample size comparable to those in other studies of reliability and validity. Sample sizes larger than 200 have not been shown to substantially improve precision in studies such as these (30). Participant burden is an important factor in data collection and a partial determinant of response rate in studies of diet and disease. If a limited number of questions can be used to obtain information comparable to that obtained by using a detailed instrument, participant burden may be reduced, increasing the likelihood that a subject will participate. Our results suggest that both the HHHQ and the HFFQ perform comparably with a more detailed questionnaire, the DPQ, in ranking participants on alcohol intake during the months prior to interview. However, the DPQ consistently produced higher estimates of the quantity of intake than either of the FFQs. If, in fact, FFQs underestimate alcohol intake, then public health recommendations resulting from studies of alcohol and chronic disease using these instruments might need to be adjusted accordingly. ACKNOWLEDGMENTS This work was supported in part by National Institute on Alcohol Abuse and Alcoholism grant 5 P50 AA09802.

7 340 McCann et al. REFERENCES 1. Smith-Warner SA, Spiegelman D, Yaun SS, et al Alcohol and breast cancer in women: a pooled analysis of cohort studies. JAMA 1998;279: Bandera EV, Freudenheim JL, Marshall JR, et al. Diet and alcohol consumption and lung cancer risk in the New York State Cohort. Cancer Causes Control 1997;8: Ascherio A, Hennekens C, Willett WC, et al. Prospective study of nutritional factors, blood pressure, and hypertension among US women. Hypertension 1996;27: Giovannucci E, Rimm EB, Ascherio A, et al. Alcohol, lowmethonine-low-folate diets, and risk of colon cancer in men. J Nad Cancer Inst 1995;87: Freudenheim JL, Graham S, Byers TE, et al. Diet, smoking, and alcohol in cancer of the larynx: a case-control study. Nutr Cancer 1992;17: Freudenheim JL, Graham S, Marshall JR, et al. A case-control study of diet and rectal cancer in western New York. Am J Epidemiol 1990;131: Block G, Hartman AM, Dresser CM, et al. A data-based approach to diet questionnaire design and testing. Am J Epidemiol 1986;124: Willett WC, Reynolds RD, Cottrell-Hoehner S, et al. Validation of a semi-quantitative food frequency questionnaire: comparison with a 1-year diet record. J Am Diet Assoc 1987;87: Coates RJ, Eley JW, Block G, et al. An evaluation of a food frequency questionnaire for assessing dietary intake of specific carotenoids and vitamin E among low-income black women. Am J Epidemiol 1991;134: Stein AD, Shea S, Basch CE, et al. Consistency of the Willett semiquantitative food frequency questionnaire and 24-hour dietary recalls in estimating nutrient intakes of preschool children. Am J Epidemiol 1992;135: Rimm EB, Giovannucci EL, Stampfer MJ, et al. Reproducibility and validity of an expanded self-administered semiquantitative food frequency questionnaire among male health professionals. Am J Epidemiol 1992;135: Sobell J, Block G, Koslowe P, et al. Validation of a retrospective questionnaire assessing diet years ago. Am J Epidemiol 1989;130: Willett WC, Sampson L. Stampfer MJ, et al. Reproducibility and validity of a semiquantitative food frequency questionnaire. Am J Epidemiol 1985; 122: Salvini S, Hunter DJ, Sampson L, et al. Food-based validation of a dietary questionnaire: the effects of week-to-week variation in food consumption. Int J Epidemiol 1989; 18: Suitor CJ, Gardner J, Willett WC. A comparison of food frequency and diet recall methods in studies of nutrient intake of low-income pregnant women. J Am Diet Assoc 1989;89: Block G, Thompson FE, Hartman AM, et al. Comparison of two dietary questionnaires validated against multiple dietary records collected during a 1-year period. J Am Diet Assoc 1992;92: Feskanich D, Rimm EB, Giovannucci EL, et al. Reproducibility and validity of food intake measurements from a semiquantitative food frequency questionnaire. J Am Diet Assoc 1993;93: Block G, Woods M, Potosky A, et al. Validation of a selfadministered diet history questionnaire using multiple diet records. J Clin Epidemiol 1990;43: Mares-Periman JA, Klein BE, Klein R, et al. A diet history questionnaire ranks nutrient intakes in middle-aged and older men and women similarly to multiple food records. J Nutr 1993; 123: Clark WB, Hilton ME, eds. Alcohol in America: drinking practices and problems. Albany, NY: State University of New York Press, Room R. Measuring alcohol consumption in the U.S.: methods and rationales. In: Clark WB, Hilton ME, eds. Alcohol in America: drinking practices and problems. Albany, NY: State University of New York Press, Skinner HA, Sheu WJ. Reliability of alcohol use indices: the lifetime drinking history and the MAST. J Stud Alcohol 1982; 43: Russell M, Marshall J, Trevisan M, et al. Test-retest reliability of the Cognitive Lifetime Drinking History. Am J Epidemiol 1997;146: Ferraroni M, Decarli A, Franceschi S, et al. Validity and reproducibility of alcohol consumption in Italy. Int J Epidemiol 1996;25: Fidanza F, Gentile MG, Porrini M. A self-administered semiquantitative food frequency questionnaire with optical reading and its concurrent validation. Eur J Epidemiol 1995;11: Webb GR, Redman S, Gibberd RW, et al. The reliability and stability of a quantity-frequency method and a diary method of measuring alcohol consumption. Drug Alcohol Depend 1991; 27: Giovannucci E, Colditz G, Stampfer MJ, et al. The assessment of alcohol consumption by a simple self-administered questionnaire. Am J Epidemiol 1991;133: Stall R. Change and stability in quantity and frequency of alcohol use among aging males: a 19-year follow-up study. Br J Addict 1986;81:537^ Fillmore KM, Golding JM, Leino EV, et al. Patterns and trends in women's and men's drinking. In: Wilsnack RW, Wilsnack SC, eds. Gender and alcohol: individual and social perspectives. New Brunswick, NJ: Rutgers Center of Alcohol Studies, Willett W. Nutritional epidemiology. New York, NY: Oxford University Press, 1990.