Household budget survey nutritional data in relation to mortality from ...

3 downloads 114 Views 96KB Size Report
heart disease, colorectal cancer and cancer of the female breast. Design: Ecological ... mortality rates from chronic diseases with fairly strong nutritional roots. HBS data could ... heart disease has been critically reviewed by the (USA). National ...
European Journal of Clinical Nutrition (1999) 53, 328±332 ß 1999 Stockton Press. All rights reserved 0954±3007/99 $12.00 http://www.stockton-press.co.uk/ejcn

Household budget survey nutritional data in relation to mortality from coronary heart disease, colorectal cancer and female breast cancer in European countries P Lagiou1, A Trichopoulou1*, HK Henderickx2, C Kelleher3, IU Leonhauser4, O Moreiras5, M Nelson6, A Schmitt7, W Sekula8, K Trygg9, G Zajkas10 for the DAFNE I and II projects of the European Commission 1

Department of Nutrition and Biochemistry, National School of Public Health, 196 Alexandras Avenue, Athens, GR 115 21, Greece; and See Appendix for af®liations and list of participants of the DAFNE I and II projects

2 ± 10

Objective: We have undertaken a study to examine whether Household Budget Survey (HBS)-derived nutritional patterns are related to mortality from diseases with strong nutritional components, namely coronary heart disease, colorectal cancer and cancer of the female breast. Design: Ecological correlation study. In the context of the Data Food Networking projects of the European Union, raw data from the national HBS of 10 European countries were provided. For each of the 10 participating countries, daily food availability per capita around 1990 was calculated. Individual foods were aggregated into 12 major food groups that were linked with the diseases under consideration. Mortality data were available from a World Health Organisation database. We have used a composite score to summarise the postulated in¯uence of diet. Setting: Ten European countries circa 1990. Results: The correlation coef®cients between the composite score and the age-adjusted mortality from each of the studied diseases were: ‡ 0.51 (P~0.14) for colorectal cancer; ‡ 0.72 (P~0.02) for female breast cancer; and ‡ 0.60 (P~0.07) for coronary heart disease, after adjustment for tobacco smoking. Conclusions: We conclude that dietary information from HBS is suf®ciently reliable to reveal correlations with mortality rates from chronic diseases with fairly strong nutritional roots. HBS data could be used, with little cost, for monitoring secular trends in dietary patterns with a view to their health implications. Sponsorship: The DAFNE I and II projects were funded by the European Commission, in the context of the `Cooperation in Science and Technology with Central and Eastern European Countries' and `Agriculture and Agro-Industry, including Fisheries' programs respectively. Descriptors: Household Budget Surveys; nutrition; diet; mortality; ecological correlation

Introduction In most European countries, a large amount of information concerning diet is collected in the context of the regularly conducted Household Budget Surveys (HBS). HBS are undertaken by the National Statistical Services at intervals varying from one to several years and cover household food purchases over a period ranging from one week to a few months. The HBS samples are representative of all households in the respective countries and their size is in the order of several thousands. HBS are mainly undertaken to ascertain trends in food consumption and their economic implications, and they are also useful for the derivation of price indices. Details about HBS are provided elsewhere (Van Staveren et al, 1991; Trichopoulou & Lagiou, 1997). Efforts have recently been made to exploit the potential of HBS for nutritional purposes. The Data Food Networking (DAFNE) projects involve 10 European countries, the

*Correspondence: Dr A Trichopoulou, Department of Nutrition and Biochemistry, National School of Public Health, 196 Alexandras Avenue, Athens, GR 115 21, Greece. Received 1 September 1998; revised 11 November 1998; accepted 24 November 1998

representatives of which provided raw data from their national HBS. These data were subsequently uniformly coded to assure interpretability and comparability. We have undertaken a study to examine whether HBSderived nutritional patterns are related to the frequency of occurrence of diseases with strong nutritional components which are frequently fatal. We have chosen to study mortality from coronary heart disease (CHD), cancer of the female breast and colorectal cancer. Diet is an important component cause of several other chronic diseases, mainly diverticulosis and cancer of the stomach, but for these diseases either case fatality ratio is very low (diverticulosis) making mortality a poor frequency indicator, or information is lacking with respect to other important causes (for stomach cancer, prevalence of Helicobacter pylori infection by age). Therefore, our study was an attempt to study HBS-derived nutritional data by undertaking ecological correlations of nationwide food patterns with mortality rates from common diseases with established nutritional aetiology. Materials and methods HBS data from the following 10 European countries were used, with calendar year of the respective HBS survey in parenthesis: Belgium (1987=88), Germany (Western,

HBS nutritional data and mortality P Lagiou et al

1988), Greece (1988), Hungary (1991), Ireland (1987), Luxembourg (1993), Norway (1992=94), Poland (1988), Spain (1990=91), and the United Kingdom (1993). For each of these countries food availability per household was retrieved and food availability per capita per day was calculated through division by the product of the referent time period and the mean household size (Trichopoulou & Lagiou, 1997; Trichopoulou & Lagiou, 1999, in press). Subsequently, individual foods were aggregated into 12 major food groups that have been linked, with variable degree of certainty, with one or more of the chronic diseases under consideration: added unsaturated lipids, added saturated lipids, cereals, fruits, vegetables, legumes, red meat, dairy products, ®sh, eggs, mono- and disaccharides and products, and ethanol. Added lipids should be distinguished from those `hidden' in particular foods, for example meat or ®sh. Because tobacco products are not part of the DAFNE database, per capita per year consumption data circa 1970 were retrieved from a recent World Health Organisation publication (WHO, 1997). A general agreement has gradually been formed with respect to the role of diet and tobacco in the aetiology of coronary heart disease, cancer of the female breast and colorectal cancer. For cancer, the relevant information has been summarised in two recent publications (Willett & Trichopoulos, 1996; WCRF=AICR, 1997), whereas the information for coronary heart disease has been critically reviewed by the (USA) National Research Council (National Research Council, 1989) and Willett (Willett, 1994). Table 1 summarises this information in a semi-quantitative way. It is assumed that de®nitive associations re¯ect stronger underlying relations and vice versa. Mortality data from coronary heart disease, cancer of the female breast and colorectal cancer circa 1990, adjusted to the European population of 1991, were retrieved from a WHO database (WHO, 1991). Since bivariate correlations between particular food groups and mortality from speci®c diseases over a total of only ten countries do not allow control for mutual confounding, likely to be generated by the inter-relations of the various food groups, we have used a composite score to summarise the postulated in¯uence of Table 1 Risk implications for coronary heart disease, breast and colorectal cancer by the indicated nutritional variables and tobacco Coronary heart disease Cereals Pulses Vegetables Fruits Red meat Fish Milk and products Eggs Sugars Unsaturated= saturated added lipids Ethanol Smoking

reduce reduce? reduce reduce increase reduce increase increase increase? reduce reduce increase

Breast cancer

Colorectal cancer

reduce? reduce?

reduce reduce increase

reduce?

increase? increase? reduce?

increase

increase

`?' indicates current data are only suggestive (possible associations). Roman type indicate likely associations. Bold type indicates data are convincing (de®nitive associations). Sources: National Research Council, 1989; Willett, 1994; Willett & Trichopoulos, 1996; WCRF=AICR, 1997.

diet (including also smoking when needed). Composite scores are often used to describe the likely effects of total diet (Hjermann et al, 1981; Manousos et al 1983; Trichopoulou et al, 1995), and they are also employed in several other areas of clinical research (Alvarez M et al, 1998). For the construction of the composite score, the ten countries were ranked with respect to each of the 12 study variables (added unsaturated and saturated lipids were studied in terms of a single variable, that is, their ratio, so that there were 11 nutritional variables plus smoking). The ranking was done in a way such that number 1 was the country with the highest consumption of a bene®cial food group or the lowest consumption of a harmful food group or smoking. As indicated, the ranking was repeated for each food group, as well as for smoking, generating for every country 12 rankings for each of the three diseases. The ranks with respect to each food and each disease were subsequently weighted, with weights 1.5 for a de®nitive association, 1 for a likely association and 0.5 for a possible association. When there is no evidence for an association between a particular food group and a certain disease, the relevant ranking receives a weight of zero and is discounted. Lastly, for every particular country, the weighted ranks for each of the 12 variables were added, and this process was followed for each of the three diseases. Because the composite score is a summation of individual ranks for foods de®nitely, likely or possibly associated with a particular disease, there is bound to be a higher composite score for a disease with several such associations (notably CHD) than for a disease with fewer such associations (notably breast cancer) (Table 1). Correlation coef®cients (and, when needed, partial correlation coef®cients) were calculated using standard procedures (Armitage & Berry, 1988). Results Table 2 shows availability per capita per day of the 10 nutritional variables, the ratio of unsaturated to saturated added lipids and the average number of cigarettes smoked daily. Table 3 shows gender and age adjusted mortality from coronary heart disease, female breast cancer and colorectal cancer. In Table 4, simple correlation coef®cients between each of the eleven dietary exposures and smoking on the one hand and each of the three outcomes on the other are given. Mutual confounding among the dietary variables hinders the interpretability of these correlation coef®cients. On the basis of the procedures indicated in the materials and methods section, the data shown in Table 5 were generated. Lower values of the composite nutritional score re¯ect a more favourable dietary pattern for the respective disease. Therefore, if the nutritional patterns were indeed associated with mortality from the diseases under consideration, positive correlation coef®cients would be generated. For colorectal cancer, the correlation coef®cient is ‡ 0.51 and corresponds to one-tailed P value of 0.07 (two-tailed P~0.14). For female breast cancer, the correlation coef®cient is ‡ 0.72 and corresponds to one-tailed P value of 0.01 (two-tailed P~0.02). Finally, for coronary heart disease, the correlation coef®cient is ‡ 0.63 and corresponds to one-tailed P value of 0.03 (two-tailed P~0.06); after adjustment for tobacco smoking, the partial correlation coef®cient becomes ‡ 0.60 and corresponds to

329

HBS nutritional data and mortality P Lagiou et al

330

Table 2 Availability per capita per day of 10 nutritional variables, ratio of unsaturated to saturated added lipids, and average number of cigarettes smoked daily in the 10 DAFNE countriesa

Cereals (g) Pulses (g) Vegetables (g) Fruits (g) Red meat (g) Fish (g) Milk and products (g) Eggs (pieces) Sugars (g) Unsaturated=saturated added lipids Ethanol (g) Smoking (number of cigarettes)

Belgium

Germanyb

177 0.5 162 198 147 26 267

167 0.8 141 202 126 12 343

Greece 268 16 267 350 142 40 356

Hungary

Ireland

Luxembourg

Norway

Poland

Spain

264 6.1 201 159 138 4 309

239 1.2 130 103 107 10 527

174 2.3 180 234 155 28 274

176 1.3 102 174 116 53 465

283 3.2 202 99 153 15 408

168 18 180 308 120 75 418

UK 167 0.8 158 132 104 21 367

0.4 55 0.21

0.5 54 0.13

0.5 82 9.3

0.7 80 0.42

0.5 70 0.11

0.3 76 0.41

0.4 82 0.04

0.6 106 0.10

0.6 35 18.8

0.3 39 0.22

30 8.5

34 6.7

18 7.2

29 8.1

22 8.4

30 8.5

12 5.6

16 8.2

28 6.0

21 8.9

a

Data circa 1990, except for cigarettes (1970). The German HBS data used in the DAFNE project (German contract data base) do not necessarily correspond to the non-anonymised statistical microdata from which the contract data base was prepared.

b

Table 3 Gender and age adjusted Ð to the European population of 1991 Ð mortality from coronary heart disease, female breast cancer and colorectal cancer circa 1990 (deaths per 100 000 person-years) Coronary heart disease

Female breast cancer

Colorectal cancer

103 150 95 235 255 124 187 119 73 223

37 33 21 32 43 33 27 22 24 42

26 29 10 35 30 27 26 18 17 26

Belgium Germany Greece Hungary Ireland Luxembourg Norway Poland Spain UK

Belgium Germany Greece Hungary Ireland Luxembourg Norway Poland Spain UK

Coronary heart diseasea

Female breast cancer

Colorectal cancer

66.1 67.3 50.8 65.0 87.0 60.4 78.3 84.0 50.8 78.3

17.0 19.5 5.0 13.5 18.0 14.3 14.0 12.5 9.8 14.0

38.3 37.5 23.8 36.0 35.5 36.8 32.6 41.3 23.8 24.8

a

For coronary heart disease, tobacco smoking is also incorporated in the index.

Source: WHO, 1991. Table 4 Simple Pearson's correlation coef®cients between each of the exposure variables (per capita per day) and each of the outcome variables (deaths per 100 000 person-years)a Coronary heart Female breast disease cancer Cereals (g) Pulses (g) Vegetables (g) Fruits (g) Red meat (g) Fish (g) Milk and products (g) Eggs (pieces) Sugars (g) Unsaturated=saturated added lipids Ethanol (g) Smoking (number of cigarettes)

Table 5 Nutritional indices for the three diseases under consideration in the 10 DAFNE countries, with lower values indicating more favourable dietary patterns

0.13 7 0.53 7 0.46 7 0.70 7 0.58 7 0.57 0.37 7 0.01 0.05 7 0.58 7 0.15 0.31

(0.03) (0.08) (0.08)

(0.08)

Colorectal cancer

7 0.36 7 0.65 (0.04) 7 0.54 7 0.53 7 0.46 7 0.51 0.00 7 0.44 7 0.44 7 0.52

7 0.23 7 0.69 7 0.60 7 0.61 7 0.23 7 0.58 7 0.10 7 0.05 7 0.10 7 0.65

0.33 0.58 (0.08)

0.41 0.29

(0.03) (0.06) (0.06) (0.08)

(0.04)

a

For correlation coef®cients corresponding to two-tailed P values of less than 0.10, exact P values are shown.

one-tailed P value of 0.04 (two-tailed P~0.07). The use of one-sided P values is defensible in this context, because the composite score was constructed with the a priori assumption that it would be positively associated with mortality from each of the studied diseases.

Discussion Ecological associations are not easily interpretable, because they are susceptible to confounding by unmeasured factors, as well as to bias generated by the aggregation of data which is characteristic of this type of design (Morgenstern, 1982). Nevertheless, when the risk gradient is substantial and the exposure is wide-spread in a population, ecological correlations do re¯ect the underlying causal relations as, for instance, prevalence of hepatitis B surface antigen and mortality from hepatocellular carcinoma, or geographical latitude and incidence of skin cancer (IARC, 1990). Diet is, of course, universal and the extremes of a nutritional index could be associated with sharp risk gradients. Therefore, there is a sound foundation for the use of an ecological design for the study of the nutritional information generated from HBS. In the early days of nutritional epidemiology, ecological associations were frequently used for the generation of hypotheses concerning the dietary aetiology of various chronic diseases (Armstrong & Doll, 1975; Carroll, 1975). Today, however, nutritional epidemiology has advanced considerably and there is a wealth of information from analytical epidemiological studies (WCRF=AICR, 1997). We have not attempted multivariate analysis,

HBS nutritional data and mortality P Lagiou et al

because data from only 10 countries were available, which limits the usable number of independent variables to no more than one or two. The use of a composite score bypasses this problem, because it substitutes one composite variable for several independent ones. The age-adjusted mortality data we have used are of®cial WHO data, but this does not guarantee that they are error free. Moreover, the HBS nutritional data, which are derived from population samples designed to be country-representative, may incorporate subtle errors due to misunderstandings, mis-reporting or non-response from the randomly selected households. When the errors in the two correlated variables are unrelated, however, the correlation coef®cient between them will, if anything, be underestimated in absolute terms. In this investigation, we have used concurrent data for dietary factors, but data for 1970 with respect to smoking. Two reasons led us to this decision: there are reliable data pointing to long average latency for tobacco smoking (Doll & Peto, 1981), whereas there is little information concerning latency with respect to dietary exposures; dietary changes over time have been gradual, while tobacco consumption in several developed countries, for example the United Kingdom, has changed substantially, following the introduction of anti-smoking campaigns (WHO, 1997). In any case, mis-speci®cation of latency with respect to the dietary factors is a form of misclassi®cation, in which the associated error is unrelated to the error in the outcome variables (mortality rates). We did not attempt to apply the procedures employed in this present study to data assembled by the Food and Agriculture Organisation (FAO) and published as food balance sheets (Kelly et al, 1991). These data have their own strengths and weaknesses, but it is possible that they would generate, under similar assumptions, correlations of similar order of magnitude. Since, however, HBS and FAO data are to a very large extent independent, a combination of information from both these data sources can substantially reduce non-differential misclassi®cation and improve the association with the desirable, but not directly measured variable, which is the average individual dietary consumption (Marshall, 1989).

Conclusions We have been able to document that dietary information from HBS is associated with mortality from three diseases with fairly strong nutritional roots. Other diseases with strong nutritional aetiology have not been studied either because mortality from them is very low in comparison to their incidence, for example, diverticulosis, or because no information is available concerning other causes, for example, cancer of the stomach. Since HBS are regularly undertaken for other purposes in most European countries, their dietary data could be used, with little additional cost, for monitoring secular changes in dietary patterns with a view to their health implications. Acknowledgements ÐThe DAFNE I and II projects were funded by the European Commission, in the context of the `Cooperation in Science and Technology with Central and Eastern European Countries' and `Agriculture and Agro-Industry, including Fisheries' programs respectively.

References Alvarez M, Nava JM, Rue M & Quintana S (1998): Mortality prediction in head trauma patients: performance of Glasgow Coma Score and general severity systems. Crit. Care 26, 142 ± 148. Armitage P & Berry G (1988): Statistical Methods in Medical Research. Oxford: Blackwell Scienti®c Publications. Armstrong BK & Doll R (1975): Environmental factors and cancer incidence and mortality in different countries, with special reference to dietary practices. Int. J. Cancer 15, 617 ± 631. Carroll KK (1975): Experimental evidence of dietary factors and hormonedependent cancers. Cancer Res. 35, 3374 ± 3383. Doll R & Peto R (1981): The causes of cancer: quantitative estimates of avoidable risks of cancer in the United States today. J. Natl. Cancer Inst. 66, 1191 ± 1308. Hjermann I, Velve Byre K, Holme I & Leren P (1981): Effect of diet and smoking intervention on the incidence of coronary heart disease. Report from the Oslo study group of a randomized trial in healthy men. Lancet ii, 1303 ± 1310. International Agency for Research on Cancer (IARC) (1990) In: Cancer: Causes, Occurrence, and Control. L Tomatis (ed). IARC Scienti®c Publication no. 100. Lyon: IARC. Kelly A, Becker W & Helsing E (1991): Food balance sheets. In Food and Health Data: Their Use in Nutrition Policy-Making. WHO Regional Publications, European Series, No 34. World Health Organization. Manousos O, Day N, Trichopoulos D, Garovassilis F, Tzonou A & Polychronopoulou A (1983): Diet and colorectal cancer: a case-control study in Greece. Int. J. Cancer 32, 1 ± 5. Marshall J (1989): The use of multiple reports in epidemiological studies. Stat. Med. 8, 1041 ± 1050. Morgenstern H (1982): Uses of ecologic analysis in epidemiologic research. Am. J. Public Health 72, 1336 ± 1344. National Research Council (1989): Diet and Health: Implications for Reducing Chronic Disease Risk. Washington, DC, USA: National Academy Press. Trichopoulou A, Kouris-Blazos A, Wahlqvist ML, Gnardellis C, Lagiou P, Polychronopoulos E, Vassilakou T, Lipworth L & Trichopoulos D (1995): Diet and overall survival in elderly people. Br. Med. J. 311, 1457 ± 1460. Trichopoulou A & Lagiou P, eds (1997): Methodology for the Exploitation of HBS Food Data and Results on Food Availability in Five European Countries Ð DAFNE I. Luxembourg: Of®ce for Of®cial Publications of the European Communities. Trichopoulou A & Lagiou P, eds (1999): Methodology for the Exploitation of HBS Food Data and Results on Food Availability in Six European Countries Ð DAFNE II. Luxembourg: Of®ce for Of®cial Publications of the European Communities (in press). Van Staveren WA, Van Beem I & Helsing E (1991): Household Budget Surveys. In Food and Health Data: Their Use in Nutrition PolicyMaking. WHO Regional Publications, European Series, No 34. World Health Organisation. Willett WC (1994): Diet and health: what should we eat? Science 264, 532 ± 537. Willett WC & Trichopoulos D (1996): Summary of the evidence: Nutrition and cancer. Cancer Causes Control 7, 178 ± 180. World Cancer Research Fund in association with the American Institute for Cancer Research (1997): Food, Nutrition and the Prevention of Cancer: A Global Perspective. World Cancer Research Fund=American Institute for Cancer Research. World Health Organisation, Regional Of®ce for Europe (1991): Health and Food Indicators. Copenhagen, Denmark: WHO. World Health Organisation (1997): Tobacco or Health Ð A global Status Report. Geneva: WHO.

Appendix Participants of the DAFNE I and II projects Belgium University of Gent, Faculty of Agricultural and Applied Biological Sciences, Department of Food Technology and Nutrition 2 HK Henderickx & AM Remaut-de Winter National Institute of Statistics, Brussels H Buermans & L Merckx

331

HBS nutritional data and mortality P Lagiou et al

332

Germany JL University Giessen, rnahrungsberatung & Vebraucherverhalten, Institut fur Ernahrungswissenschaft 4 IU Leonhauser Statistiches Bundesamt, Zweigstelle Berlin J Kaiser Greece (coordinating centre) National School of Public Health, National Centre for Nutrition, Athens 1 A Trichopoulou, A Kanellou, 1P Lagiou, A Naska, P Thriskos, V Vasdekis, T Vassilakou & E Zintzaras National Statistical Of®ce, Athens G Douros & I Tsaousi Hungary National Institute of Food Hygiene and Nutrition, Budapest 10 G Zajkas & M Korom Central Statistical Of®ce, Department of Living Standards and Human Resources Statistics, Budapest P Szivos Ireland National Nutrition Surveillance Centre, Department of Health Promotion, Clinical Science Institute, University College, Galway S Friel & 3C Kelleher Central Statistics Of®ce, Dublin K McCormack Central Statistics Of®ce, Cork S MacFeely Luxembourg Interdisciplinary Nutrition Policy Group 7 A Schmitt Central Statistics Of®ce J Langers Center of Socio-economic Population Studies (CEPS) M Zanardelli

Norway Institute of Nutrition Research, University of Oslo 9 K Trygg Norwegian Board of Health, Oslo E Helsing Norwegian Food Control Authority, Oslo K Lund-Larsen Statistics Norway, Oslo E Mork Poland National Food and Nutrition Institute, Warsaw M Morawska, Z Niedzialek & 8W Sekula Central Statistical Of®ce, Department of Social and Demographic Surveys, Warsaw A Bienkunska Spain Departamento de Nutricion y Bromatologia, Facultat de Farmacia, Universidad Complutense de Madrid A Carbajal & 5O Moreiras Instituto Nacional de Estadistica, Madrid ML Boned & PS Spiegelberg United Kingdom King's College London, Department of Nutrition and Dietetics, London 6 M Nelson & S Paterakis Of®ce of National Statistics, Family Expenditure Survey, London J King Ministry of Agriculture, Fisheries and Food, National Food Survey Branch, London SM Speller.