QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Abstract Full Text (PDF) Purchase Article View Shopping Cart Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by Ambrosini, G. L. Articles by Musk, A. W. Search for Related Content PubMed PubMed Citation Articles by Ambrosini, G. L. Articles by Musk, A. W.

---

Nutritional Epidemiology

The Reliability of Ten-Year Dietary Recall: Implications for Cancer Research

Gina L. Ambrosini^*,2, Sofie A. H. van Roosbroeck^*, Dorothy Mackerras, Lin Fritschi^*, Nicholas H. de Klerk^* and A. William Musk^*,**

^* School of Population Health, University of Western Australia, Crawley 6009, Western Australia; Menzies School of Health Research, Casuarina 0811, Northern Territory, Australia; and ^** Department of Respiratory Medicine, Sir Charles Gairdner Hospital, Nedlands 6009, Western Australia.

²To whom correspondence should be addressed. E-mail: ginaa{at}sph.uwa.edu.au.

	ABSTRACT

TOP ABSTRACT SUBJECTS AND METHODS RESULTS DISCUSSION LITERATURE CITED

 
Remote dietary intakes may be more important than recent diet in the etiology of cancer because of the long latency in cancer development. We examined the reliability of remote dietary recall over 10 y. Subjects were 56 adults participating in a cancer prevention trial in Western Australia. All subjects completed a 28-d diet record (DR) in 1991. A food-frequency questionnaire (FFQ) modified to ask respondents about their diet 10 y earlier was sent to each subject for completion in 2001. Remote intakes recalled from 10 y earlier using the FFQ were compared with the DR using the limits of agreement (LOA) method and Pearson correlation coefficients. Mean intakes of most nutrients did not differ between dietary methods. The LOA indicated that the FFQ could under- or overestimate DR estimates by 50%. For many nutrients, agreement between methods depended on the magnitude of intake. Pearson’s correlation coefficients ranged from 0.02 for retinol to 0.66 for alcohol. These findings are similar to those of other studies that examined the reliability of recent and remote dietary intakes. They also show that using this FFQ, remote diet recalled from 10 y earlier may be as reliable as recent dietary recall.

KEY WORDS: • remote dietary recall • reliability • cancer

When studying the nutritional determinants of cancer and other chronic illness, remote dietary intake (from many years in the past) may be more important than recent diet because of the long latency of these diseases. It is possible that weak associations were observed in many studies of diet and cancer because most depended on estimates of recent diet, and possibly underestimated the true effect of diet because of misclassification of intake. Recent or current dietary intake is not thought to be a reliable predictor of past intake (1), and dietary habits change significantly with advancing age (2). This is pertinent to cancer epidemiology because estimates of recent diet from older people, whom the majority of cancers affect, are therefore unlikely to reflect the diet consumed over most of their adult life. Similarly, people diagnosed with cancer may be prone to change their diet; thus, their estimates of recent diet may also be less likely to reflect past diet. Therefore, the reliability of remote dietary recall is of considerable interest in cancer epidemiology. To investigate this, we compared dietary intakes recalled from 10 y earlier using a modified food-frequency questionnaire (FFQ) in 2001, with intakes from a diet record (DR) collected in 1991. Although this type of study is commonly referred to as a validation study, it is more correctly described as an intermethod reliability study (3).

	SUBJECTS AND METHODS

TOP ABSTRACT SUBJECTS AND METHODS RESULTS DISCUSSION LITERATURE CITED

 
Study population.

Subjects were recruited from a cancer prevention trial examining the efficacy of ß-carotene and retinol supplements in reducing the risk of mesothelioma after asbestos exposure (4). Trial participants were 3240 healthy people aged 8–87 y (80% men) recruited from two large cohorts of former workers and residents of a blue asbestos mining and milling town in Western Australia (4). All participants gave their informed consent, and the trial was approved by the Human Research Ethics Committee of the University of Western Australia and the Clinical Drug Trials Committee of the Sir Charles Gairdner Hospital, Nedlands, Western Australia. The reliability study was approved by the Human Research Ethics Committee of the University of Western Australia.

Diet records.

All subjects joining the trial during December 1990 were asked during their enrollment interview if they would provide four 7-d DR over the following year, except subjects with apparent language or literacy difficulties. Those who agreed were shown how to record everything consumed for seven consecutive days using household units (not weights). One 7-d DR was completed in December 1990 and the others in April, July and October of 1991. Participants were followed up by telephone 2 wk after each commencement date to set a date for person-to-person checking of their DR. All DR were entered into a database by the same nutritionist (G.L.A.) using Diet1 nutrient analysis software (5) and linked with Australian Food Composition Tables (6) to calculate daily nutrient intakes.

Food-frequency questionnaire.

The semiquantitative FFQ was developed by the Anti-Cancer Council of Victoria (ACCV) Australia for use in the ethnically diverse Australian population (7). It includes 74 food items and 3 questions on alcohol intake, with 10 frequency options for each, ranging from "never" to "3 or more times daily." Respondents are asked to identify usual portion sizes from a series of photographs in the FFQ. Information is also collected on types of milk, bread, spread and cheese consumed, the number of pieces of fresh fruit and how many different vegetables are usually consumed daily.

We modified the FFQ to ask respondents to recall their usual dietary intake from 10 y earlier, rather than over the past year. In July of 2001, the modified FFQ was mailed to all surviving subjects who had provided four 7-d DR during 1990 and 1991. A covering letter explained how to complete the questionnaire. Respondents were encouraged to think about where they lived and their job 10 y ago to assist dietary recall. Any missing values were followed up with the participant by telephone. Completed questionnaires were sent to the ACCV for computerized scanning of responses and the calculation of daily nutrient intakes using Australian Food Composition Tables (8).

Internal validity.

Energy intakes from both dietary methods were checked for extreme values. We utilized limits used by the ACCV (personal communication, A. Hodge, ACCV) and others (9), in which energy intakes < 3000 kJ or > 21,000 kJ were considered implausible.

Subjects who completed the FFQ in 2001 were compared with those who failed to complete the FFQ and other trial participants who did not participate in the reliability study, i.e., those invited but who declined. Reliability study subjects who died after completing their DR and before July 2001 were not included in this comparison. Gender, smoking status and use of vitamin A supplements in addition to the trial supplement at entry to the trial were compared using ² and Fisher’s exact tests. Mean age, BMI and alcohol intake at entry to the trial were compared using t tests.

Limits of agreement.

Although correlation coefficients are often used to compare two different methods, they can be misleading because the correlation coefficient expresses the strength of a linear relationship between two variables, but not the agreement between them (10,11). It has been shown that highly correlated data, including dietary intakes, may not actually agree (12,13).

To examine agreement in this study, we used the limits of agreement (LOA) method recommended by Bland and Altman (12,14). All nutrient intakes were first transformed to their natural logarithms. For each subject, the difference between nutrient intakes from each method was then calculated, e.g., FFQ(energy) - DR(energy). The mean difference and its SD (SD_diff) were calculated for each nutrient. The mean difference represents mean agreement between methods for the study population. The 95% LOA provide an interval within which 95% of all individual differences between methods are expected to lie. These were estimated for each nutrient by calculating: mean agreement ± (1.96 x SD_diff). For each nutrient we plotted the differences between methods against the mean of the two to detect any variations in level of agreement. This was formally tested by fitting the regression line of differences. A significant slope in the regression line (H₀: ß = 0, = 0.05) indicated that agreement varied according to the magnitude of intake.

Because dietary intakes were log transformed, antilogging was necessary to interpret agreement. Antilogging mean agreement, the LOA, and their 95% confidence limits result in ratios i.e., a multiple of the FFQ relative to the DR (14). All were expressed as a percentage, with 100% indicating exact agreement. For example, a mean agreement of 120% indicates that on average, FFQ estimates for that nutrient were 1.2 times the DR estimate. LOA of 50–200% indicate that 95% of all subjects’ FFQ estimates were between one half and two times their DR estimate.

Using the LOA method, levels of acceptable agreement are arbitrary, but should reflect the required sensitivity of a measure or method (14). We classified agreement between the FFQ and DR as being acceptable when 1) the LOA were between 50 and 200% and 2) there was a lack of dependency between agreement and the magnitude of intake, i.e., the slope in the regression line of differences was not significantly different from zero. Although we (13) and others (10,11) have argued against its use, we also calculated Pearson’s correlation coefficients between methods to enable comparisons with other studies that have not used the LOA method.

	RESULTS

TOP ABSTRACT SUBJECTS AND METHODS RESULTS DISCUSSION LITERATURE CITED

 
During December of 1990, 569 people enrolled in the Vitamin A Program. Of these, 118 (21%) agreed to provide four 7-d DR over the following year; 83 (70%) completed all four DR. Of these, 56 (68%) then completed the FFQ in July 2001 (19 women and 37 men). One person was excluded because his FFQ intake was >21,000 kJ/d. Twelve subjects died after completing their DR and before the FFQ could be sent to them. Four subjects refused to complete the FFQ, eight were noncompliant, two reported that they were unable to recall past diet and one could not be traced.

After adjusting for gender, no differences were found in mean age, BMI, alcohol intake or supplement use between reliability and trial subjects (Table 1). Statistical comparisons of smoking habits were hampered by the small numbers in each category. The proportion of subjects who had never smoked was higher in reliability subjects for men (P = 0.03) and probably women, also (P = 0.16) (Table 1). The proportion of previous smokers tended to be lower among reliability subjects (P = 0.07 and 0.22 for men and women, respectively, Table 1). The proportion of current smokers was similar, but female reliability subjects tended to smoke more cigarettes per day than female trial subjects (Table 1).

Even after log transformation of the data, agreement depended on the magnitude of intake for 11 of the 29 nutrients assessed, i.e., niacin, niacin equivalents, phosphorous, potassium, protein, retinol, retinol equivalents, saturated fat, sodium, vitamin C and zinc (P < 0.05). For all of these except retinol and retinol equivalents, the slope of the regression line of differences was positive, meaning that the FFQ increasingly overestimated intakes as overall intake increased (Table 3). The negative slopes observed for retinol (Fig. 1) and retinol equivalents indicate that the FFQ significantly underestimated retinol intakes when greater intakes were reported. For all other nutrients, agreement was independent of intake, as in the case of fiber (Fig. 2).

View larger version (18K): [in this window] [in a new window]   FIGURE 1 Mean agreement and upper and lower limits of agreement (LOA) between retinol intakes estimated by the diet record (DR) and 10 y later using the food-frequency questionnaire (FFQ). The trend in differences is equivalent to the regression line of differences (P = 0.02).

View larger version (17K): [in this window] [in a new window]   FIGURE 2 Mean agreement and upper and lower limits of agreement (LOA) between fiber intakes estimated by the diet record (DR) and 10 y later using the food-frequency questionnaire (FFQ). The trend in differences is equivalent to the regression line of differences (P = 0.86).

	DISCUSSION

TOP ABSTRACT SUBJECTS AND METHODS RESULTS DISCUSSION LITERATURE CITED

 
Most studies rely on assessments of recent or current diet when investigating diet and chronic disease relationships, probably because of a perceived level of difficulty in obtaining compliant respondents or reliable information. Yet, given the long latency of most chronic diseases, especially cancer, it is likely that remote dietary intake is etiologically more important than recent intake.

We found that for most nutrients, the FFQ could under- or overestimate DR intakes by 50%. According to our criteria, adequate agreement was observed for carbohydrate, energy, fiber, iron, magnesium, the percentage of energy from carbohydrates and the percentage of energy from fats. Wide LOA were observed especially for alcohol, carotene, retinol and retinol equivalents. The LOA may become narrower when using more subjects, however; estimates of carotene, retinol and retinol equivalent intakes showed poor agreement in many other reliability studies of current diet (13,15–17). Retinol and carotene intakes (vitamin A) can be especially difficult to estimate because these nutrients are highly concentrated in some foods. Sporadically high intakes of vitamin A in the DR are not likely to agree with more general intakes estimated by a FFQ. In the case of alcohol, extremes in day-to-day variation are common (18).

Agreement decreased significantly with increasing intakes for many nutrients. Sporadically high intakes of vitamin A may also explain the negative slope observed between agreement and overall intake for retinol and retinol equivalents for which the FFQ underestimated very high intakes reported in the DR (Fig. 2). For the other nine nutrients for which a positive slope was observed between agreement and overall intake, the FFQ tended to overestimate the DR, which is not uncommon for a FFQ (13,19).

Very few published reliability studies have used the LOA method, and we could find none that had used this method to assess remote dietary recall. One reliability study that compared a 10-d DR with an FFQ designed to estimate recent diet reported LOA for ascorbic acid, energy and fiber only (16). These showed that the FFQ could underestimate the DR by as much as 30%, or overestimate it by up to 90%.

Most studies examining retrospective dietary intake have repeated a single dietary method 3–5 y later (20). Generally these studies have concluded that although remote dietary recall is often influenced by current diet (20–23), remote dietary recall correlates better with past diet than with current diet (22,24,25). It would be useful to examine whether the reliability of remote dietary recall varies according to whether the respondent’s current diet is different from their remote diet. We administered an annual questionnaire during the cancer prevention trial (still running in 2003) that included a question asking all subjects if they had made any major changes to their diet. Among the reliability study subjects, 64% reported making major dietary changes during the 10-y follow-up, i.e., after the DR was completed. Their small number (n = 55) prevented comparisons of reliability according to whether dietary changes were or were not reported. This does suggest, however, that the results of this study reflect a study sample in which the majority of respondents’ current diets are likely to be different from their diet of 10 y earlier.

In one study that compared multiple diet records to an FFQ completed 10–15 y later, the reported correlation coefficients ranged from 0.07 (vitamin A) to 0.46 (saturated fatty acids) (26), which are similar to our findings. The correlation coefficients observed in this study are also very similar to those seen in reliability studies examining recent dietary intake (27–30). However, we and others have identified the limitations of using correlation coefficients to assess reliability (10,11,13).

We recently applied the LOA method in another reliability study in the same study sample, to evaluate an FFQ developed by the Commonwealth Scientific Industrial Research Organization (CSIRO) of Australia to estimate recent diet (31). In comparing the CSIRO FFQ administered at the end of 1991 to the DR used in this study, we similarly found that for most nutrients, the FFQ could under- or overestimate DR intakes by 50%. The results for men and women combined showed a greater number of nutrients for which agreement was dependent on the magnitude of intake, in comparison to the ACCV FFQ. This dependency shows that FFQ errors vary across the range of intakes for that nutrient, which can lead to differential misclassification of intakes. Mean agreement was also better for the ACCV FFQ. The main difference was that the upper LOA for the ACCV FFQ were generally higher, which might be acceptable, given that intakes were recalled from 10 y earlier. Although the FFQ was different, we expected agreement for the ACCV FFQ to be substantially worse than that found for the CSIRO FFQ because it asked respondents to recall their diet from 10 y earlier. However, we believe this not to be the case.

We acknowledge the small sample size as the main drawback of this study. The recruitment rate for this study was also quite low (21%); for this reason, a convenience sample of 100 subjects was sought during the recruitment phase. A low response rate is not unusual when a labor-intensive method like the DR is required. The majority of subjects were asked to volunteer for the reliability study immediately after they had completed their induction interview for the cancer prevention trial. This required the completion of several other questionnaires and tests, which might explain the low response rate. The completion rates in this reliability study were, however, relatively good (70 and 67% for the DR and FFQ, respectively).

In comparing the baseline characteristics of study subjects with those who did not take part (but were invited), mean age, BMI, alcohol intake and supplement use were similar. Although not all differences were significant, there were more nonsmokers and fewer previous smokers among the reliability subjects. Because these are the only obvious differences between our reliability and trial subjects, we do not believe the two groups to be sufficiently different to reduce the generalizability of this study.

This study and the few others using the LOA method have shown that it is difficult to gain high levels of agreement between an FFQ and DR. Given that they are different tools in that the FFQ provides a summary of intake whereas the DR provides several detailed "snapshots," and given the natural within-person variation in diet, it is unreasonable to expect perfect agreement between the two. Furthermore, reliability studies examining dietary methods are hampered by the lack of a "gold standard." The diet record method is influenced by random and systematic errors and as such, is not an error-free method.

Our results show that whether the correlation coefficient or the LOA method is used to assess agreement, the FFQ used in this study to collect dietary information from 10 y earlier may provide estimates as reliable as other FFQ designed to collect information about recent diet. Based on these findings, there is no obvious reason to collect information about recent diet as a surrogate when remote dietary intake is the factor of interest.

	ACKNOWLEDGMENTS

 
The authors are grateful to Barbara Telfer, Lynne Watts, Nola Olsen, Caroline Hickling, Irene Foundas and Fiona Smith for their assistance with data collection for this study.

	FOOTNOTES

 
¹ Supported by the Western Australian Workers’ Compensation and Rehabilitation Commission, the Western Australian Department of Health and the Anti-Cancer Council of Victoria, Australia.

³ Abbreviations used: DR, diet record; FFQ, food-frequency questionnaire; LOA, limits of agreement; SD_diff, difference between standard deviations from each dietary method.

Manuscript received 22 January 2003. Initial review completed 7 March 2003. Revision accepted 21 May 2003.

	LITERATURE CITED

TOP ABSTRACT SUBJECTS AND METHODS RESULTS DISCUSSION LITERATURE CITED

 

1. Dwyer, J. T., Gardner, J., Halvorsen, K., Krall, E. A., Cohen, A. & Valadian, I. (1989) Memory of food intake in the distant past. Am. J. Epidemiol. 130:1033-1046.[Abstract/Free Full Text]

2. Drewnowski, A. & Shultz, J. M. (2001) Impact of aging on eating behaviors, food choices, nutrition, and health status. J. Nutr. Health Aging. 5:75-79.[Medline]

3. Armstrong, B. K., White, E. & Saracci, R. (1992) Principles of Exposure Measurement in Epidemiology 1992 Oxford University Press New York.

4. Musk, A. W., De Klerk, N. H., Ambrosini, G. L., Eccles, J. L., Hansen, J., Olsen, N. J., Watts, V. L., Lund, H. G., Pang, S. C., Beilby, J. & Hobbs, M.S.T. (1998) Vitamin A and cancer prevention I: observations in workers previously exposed to asbestos at Wittenoom, Western Australia. Int. J. Cancer. 75:335-361.[Medline]

5. Goodsell, C. (1991) Diet/1 Nutrient Analysis Software, 3.10 ed 1991 Xyris Software Brisbane, Australia .

6. National Food Authority (1992) NUTTAB91–92 (1992) 1992 Commonwealth Government of Australia Canberra, Australia .

7. Ireland, P., Jolley, D. & Giles, G. G. (1994) Development of the Melbourne FFQ: a food frequency questionnaire for use in an Australian prospective study involving an ethnically diverse cohort. Asia Pac. J. Clin. Nutr. 3:19-31.

8. National Food Authority (1995) NUTTAB95 1995 Commonwealth Government of Australia Canberra, Australia.

9. Kristal, A. R., Cohen, J. H., Qu, P. & Stanford, J. L. (2002) Associations of energy, fat, calcium, and vitamin D with prostate cancer risk. Cancer Epidemiol. Biomark. Prev. 11:719-725.[Abstract/Free Full Text]

10. Bland, J. M. & Altman, D. G. (1990) A note on the use of the intraclass correlation coefficient in the evaluation of agreement between two methods of measurement. Comp. Biol. Med. 20:337-340.[Medline]

11. Chinn, S. (1990) The assessment of methods of measurement. Stat. Med. 9:351-362.[Medline]

12. Bland, J. M. & Altman, D. G. (1999) Measuring agreement in method comparison studies. Stat. Methods Med. Res. 8:135-160.[Abstract/Free Full Text]

13. Ambrosini, G. L., De Klerk, N. H., Musk, A. W. & Mackerras, D. (2001) Agreement between a brief food frequency questionnaire and diet records using two statistical methods. Public Health Nutr. 4:255-264.[Medline]

14. Bland, M. J. & Altman, D. G. (1986) Statistical methods for assessing agreement between two methods of clinical assessment. Lancet. 1:307-310.[Medline]

15. Harlan, L. & Block, G. (1990) Use of adjustment factors with a brief food frequency questionnaire to obtain nutrient values. Epidemiology 1:224-231.[Medline]

16. Thompson, R. L. & Margetts, B. M. (1993) Comparison of a food frequency questionnaire with a 10-day weighed record in cigarette smokers. Int. J. Epidemiol. 22:824-833.[Abstract/Free Full Text]

17. Tjonneland, A., Overvad, K. I. M., Haraldsdoltir, J., Bang, S., Ewertz, M. & Jensen, O. M. (1991) Validation of a semi quantitative food frequency questionnaire developed in Denmark. Int. J. Epidemiol. 20:906-912.[Abstract/Free Full Text]

18. Leigh, B. C. (2000) Using daily reports to measure drinking and drinking patterns. J. Subst. Abuse. 12:51-65.[Medline]

19. Amanatidis, S., Mackerras, D. & Simpson, J. M. (2001) Comparison of two frequency questionnaires for quantifying fruit and vegetable intake. Public Health Nutr. 4:233-239.[Medline]

20. Friedenreich, C. M., Slimani, N. & Riboli, E. (1992) Measurement of past diet: review of previous and proposed methods. Epidemiol. Rev. 14:177-196.[Free Full Text]

21. Bakkum, A., Bloemberg, B., Van Staveren, W. A., Verschuren, M. & West, C. E. (1988) The relative validity of a retrospective estimate of food consumption based on a current dietary history and a food frequency list. Nutr. Cancer. 11:41-53.[Medline]

22. Byers, T. E., Rosenthal, R. I., Marshall, J. R., Rzepka, T. F., Cummings, K. M. & Graham, S. (1983) Dietary history from the distant past: a methodological study. Nutr. Cancer. 5:69-77.[Medline]

23. Thompson, F. E., Metzner, H. L., Lamphiear, D. E. & Hawthorne, V. M. (1990) Characteristics of individuals and long term reproducibility of dietary reports: the Tecumseh Diet Methodology Study. J. Clin. Epidemiol. 43:1169-1178.[Medline]

24. Wilkens, L. R., Hankin, J. H., Yoshizawa, C. N., Kolonel, L. N. & Lee, J. (1992) Comparison of long-term dietary recall between cancer cases and noncases. Am. J. Epidemiol. 136:825-835.[Abstract/Free Full Text]

25. Byers, T., Marshall, J., Anthony, E., Fiedler, R. & Zielezny, M. (1987) The reliability of dietary history from the distant past. Am. J. Epidemiol. 125:999-1011.[Abstract/Free Full Text]

26. Sobell, J., Block, G., Koslowe, P., Tobin, J. & Andres, R. (1989) Validation of a retrospective questionnaire assessing diet 10–15 years ago. Am. J. Epidemiol. 130:173-187.[Abstract/Free Full Text]

27. Willett, W. C., Reynolds, R. D., Cottrell-Hoehner, S., Sampson, L. & Browne, M. L. (1987) Validation of a semi-quantitative food frequency questionnaire: comparison with a 1-year diet record. Am. Diet. Assoc. J. 87:43-47.

28. Goldbohm, R. A., Van Den Brandt, P. A., Brants, H.A.M., Van’t Veer, P., Sturmans, M.A.F. & Hermus, R.J.J. (1994) Validation of a dietary questionnaire used in a large-scale prospective cohort study on diet and cancer. Eur. J. Clin. Nutr. 48:253-265.[Medline]

29. Kim, J., Chan, M. M. & Shore, R. E. (2002) Development and validation of a food frequency questionnaire for Korean Americans. Int. J. Food Sci. Nutr. 53:129-142.[Medline]

30. Brunner, E., Stallone, D., Juneja, M., Bingham, S. & Marmot, M. (2001) Dietary assessment in Whitehall II: comparison of 7 d diet diary and food-frequency questionnaire and validity against biomarkers. Br. J. Nutr. 86:405-414.[Medline]

31. Ambrosini, G. L., Mackerras, D., De Klerk, N. H. & Musk, A. W. (2003) Comparison of an Australian food frequency questionnaire to diet records: implications for nutrition surveillance. Public Health Nutr. 6:415-422.[Medline]

This article has been cited by other articles:

				A. A Thorp, P. R. Howe, T. A Mori, A. M Coates, J. D Buckley, J. Hodgson, J. Mansour, and B. J Meyer Soy food consumption does not lower LDL cholesterol in either equol or nonequol producers Am. J. Clinical Nutrition, August 1, 2008; 88(2): 298 - 304. [Abstract] [Full Text] [PDF]

				C M. Schooling, S. Y. Ho, G. M Leung, G N. Thomas, S. M McGhee, K. H. Mak, and T. H. Lam Diet synergies and mortality--a population-based case-control study of 32 462 Hong Kong Chinese older adults Int. J. Epidemiol., April 1, 2006; 35(2): 418 - 426. [Abstract] [Full Text] [PDF]

This Article Abstract Full Text (PDF) Purchase Article View Shopping Cart Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by Ambrosini, G. L. Articles by Musk, A. W. Search for Related Content PubMed PubMed Citation Articles by Ambrosini, G. L. Articles by Musk, A. W.