British Journal of Preventive and Social Medicine, 1977, 31, 71-80

Foods and diseasesE. G. KNOX

From the Health Services Research Centre, Birmingham

summARY An examination was made of the statistical correlations between the main foodstuffand nutrient intakes and the chief causes of mortality in 20 different countries, comprising 17 inEurope, and Canada, USA, and Japan. Subsidiary examinations were made of the effects ofincluding and excluding Japan, and of the effects of various statistical standardisation procedures.Complex food patterns were identified and related both to geographical latitude and to levels ofaffluence; these, in turn, were related to complex patterns of mortality. Criteria for drawingspecial attention to specific associations were identified, based partly on statistical significance testsand also on strength-of-association yardsticks supplied by diseases with known causes. Findingssuggesting causal interpretations were: (a) alcohol intakes and cirrhosis of the liver, cancer of themouth, and cancer of the larynx; (b) total fat intakes and multiple sclerosis, cancer of the largeintestine, and cancer of the breast; and (c) beer and cancer of the rectum.

International variations in the incidence of illnessprovide striking epidemiological characteristics ofmany diseases. Examples are: the relative rarity ofanencephalus in France compared with England,the high incidence of carcinoma of the breast inNorth America compared with Japan, the tem-perate zone distribution of multiple sclerosis, thehigh incidence of ischaemic heart disease and ofcarcinoma of the large intestine in Europecompared with Africa.

Transfers of complex living patterns from onecountry to another-for example, from Europe tothe emerging middle classes of developing Africanand Asian countries-have been followed by thedisease patterns of the donor country, and studiesof migrants have in other cases demonstrated theirprogressive acquisition of the disease patterns of thenew domicile. Studies have implicated environ-mental rather than (purely) racial and geneticfactors and the dietary component of the changedculture reasonably excites suspicion.A wide range of geographical associations between

specified food and nutrient intakes and diseaseincidences continue to be reported, but the com-plexities of both disease and dietary patterns, eachwith its own set of internal correlation, and eachwith a set of associations with non-dietarycomponents of life-style, make critical interpretationa formidable task. Certainly, the arbitrary selectionof any single food/disease pair, and its examinationfor a positive or negative association across a rangeof different countries or different times, canscarcely do justice to the complexity of the

background from which the particular example waspicked. It would seem necessary, at least, toexamine a range of foodstuffs and a range ofdiseases besides the ones in question. This study isan investigation in this sense and is based upon theavailability of two sets of international data from acommon range of countries, which do not seempreviously to have been associated in a systematicway.

Materials and method

The Organisation for European Co-operation andDevelopment (OECD) produces at intervals a seriesof tabulated Food Consumption Statistics and thedata used for this study are the figures for 1960-68(Organisation for European Co-operation andDevelopment, 1970). The tables refer to 20 countriesnamely: Canada, United States of America, Japan,Austria, Belgium/Luxembourg, Denmark, Finland,France, Germany (FR), Greece, Ireland, Italy,Netherlands, Norway, Portugal, Spain, Sweden,Switzerland, United Kingdom, and Yugoslavia.A large variety of foodstuffs and foodstuff-

groupings is listed, 113 altogether, but the listing isnot uniform and 57 commodities are indexed asoccurring in only a 'small number' of countries.Examples in this class are buckwheat, molasses,tapioca, goat's meat, cottage cheese, shell-fish, andsea weeds. An abridged list was constructed, mainlyof foods not so indexed, although a few wereincluded because of possible aetiological connectionswith disease. The values used were mean grams

71

copyright. on 28 M

ay 2019 by guest. Protected by

http://jech.bmj.com

/B

r J Prev S

oc Med: first published as 10.1136/jech.31.2.71 on 1 June 1977. D

ownloaded from

E. G. Knox

consumption per caput each day for the years1960-64, or, if recording started later, the earliestavailable figure. The foods were: wheat, rye, totalbread grain, barley, oats, total coarse grain,rice, total cereals, refined sugar, honey, potatoes,pulses, vegetables, citrus fruits, fresh non-citrusfruits, canned fruit, dried fruit, nuts, fruit juices,cocoa beans, beef, veal, lamb/mutton, pig meat,horse meat, poultry, other meat, edible offals, totalmeat, eggs, cows' milk, skimmed milk, cream,condensed milk, d-ied whole milk, dried skimmedmilk, cheese, fresh and frozen fish, salt/cured/driedfish, canned fish, olive oil, other vegetable oils,total vegetable oils, marine oils, lard, margarine,butter, wine, and beer. Total calories from allsources, from animal and vegetable sourcesseparately, and within each class, from fat andprotein, were also prepared. Total grams weightof the main foodstuffs (excluding wine and beer) wasused as an index of total dietary intake. Thelatitude of the capital city of each country was alsorecorded.

The World Health Annual Statistics for 1970(World Health Organisation, 1973) was used as thesource of disease-mortality data. All the countriesin the OECD list were matched except for theBelgium/Luxembourg economic union, for whichLuxembourg was the nearest equivalent. The agespecific mortality rates per 100000 were classifiedin accordance with the 'A' list of the 8th revisionof the International Classification ofDiseases (WorldHealth Organisation, 1967) for 137 non-violent and13 violent causes of death.Once more, an abridged list was prepared.

Diseases with very low mortalities were excludedas were most of the infective and violent causesalthough respiratory tuberculosis, suicide, deathfrom road accidents and from drowning, wereincluded. The main neoplastic and circulatorydisease groups were included as were goitre,thyrotoxicosis, diabetes, psychosis, neurosis, mul-tiple sclerosis, epilepsy, chronic bronchitis, pepticulcer, appendicitis, cirrhosis, gall bladder disease,chronic nephritis, hyperplasia of prostate, toxaemiaof pregnancy, arthritis and spondylitis, spina bifida,and congenital heart disease.

In many cases the mortality within a single agegroup was used, most frequently the age group55 to 64 years, within which there were usuallysufficient numbers for statistical stability and anexpectation of reasonably accurate diagnosis. Thepurpose of age-localisation was to avoid artefactsarising from varying age distributions in differentpopulations. For some diseases such as thecongenital diseases, a different age group was moreappropriate and for some a range of age groups, or

an accumulation of rates over several age groups,was used-for example, in ischaemic heart disease,cerebrovascular disease, hypertension, epilepsy,multiple sclerosis, cancer of breast, leukaemia, andcirrhosis.

Analysis proceeded as follows. First a simplecorrelation matrix was assembled comparing eachfood with each disease. For reasons given laterthis was repeated with Japan excluded. Correlationmatrices were then produced comparing each foodwith every other food and each disease with everyother disease. Finally, the food/disease correlationswere repeated after a series of step-wise standardisa-tion procedures.

Results

FIRST CORRELATIONSA first matrix of 58 foods (or groupings) x 70disease mortality sets produced 4060 correlationcoefficients of which 715 (17-6%) differed fromzero by ± 1 * 96 standard errors. Repeated calculationwith Japan excluded produced 610 coefficientsoutside this range. The main concordant resultsare given in Table 1 including all associations with± r >0 78 on both occasions and >0 80 on atleast one. It will be noticed that all the correlations

Table 1 Mainfood disease correlations, with and withouttJapan

Food x disease Japan Japanincluded excluded

Cereals x Respiratory tuberculosis +0-80 +0 78Sugar x Haemopoietic cancer +0-84 +0-83Sugar x Breast cancer +0-84 +0-82Meat x Colonic cancer +0 90 +0-89Meat x Breast cancer +0-84 +0 78Eggs x Colonic cancer +082 +083Milk x Skin cancer +0-84 +0-82Butter x Arthritis +0-88 +0-88Wine x Laryngeal cancer +0-84 +0-84Animal sources x Haemopoietic cancer +0-84 +0-84Animal sources x Breast cancer +0-83 +0 79Animal protein x Haemopoietic cancer +0-86 +0 85Animal protein x Breast cancer +0-82 +0-78All fats x Breast cancer +0*91 +0O88

selected in this way were positive, and that cancerwas by far the commonest disease group. In theseand most subsequent calculations ± 1-96 standarderrors corresponds with r = ± 0*46. Japanesemortalities are noted for high values for cancer ofthe stomach and cerebrovascular disease and lowvalues for cancer of the breast and cardiovasculardisease. The Japanese diet is notable for a highcontent of rice, of salt fish (cured or raw), and for alow level of fat intake. In fact, the strong correlationbetween fats and cancer of the breast was notaffected by removing Japan from the analysis, andthe only associations to be affected critically by

72

copyright. on 28 M

ay 2019 by guest. Protected by

http://jech.bmj.com

/B

r J Prev S

oc Med: first published as 10.1136/jech.31.2.71 on 1 June 1977. D

ownloaded from

Foods and diseases

this exclusion were those of stomach cancer andcerebrovascular disease with salt fish. Nevertheless,because of the ectopic position of Japanesemortalities and diet, all subsequent analysesreported in this paper exclude Japan.

FOOD PATTERNS AND DISEASE PATTERNSThe striking feature of the full results was thesystematic nature of the pattern of correlations.For many foods the same pattern of positive andnegative correlations tended to recur over a range ofdiseases, while for other foods the sign wasreversed. For example, total meat showed positiveassociations with multiple sclerosis, ischaemic heartdisease, suicide, cancer of the large intestine, cancerof the breast, and leukaemia and so did most of theindividual meats. Total cereals showed negativeassociations with each of these as did most of itscomponent grains. Foods which displayed negativeassociations with this group of diseases tended toshow positive associations with epilepsy, pepticulcer, cirrhosis of the liver, chronic nephritis, andcancers of the larynx, mouth, oesophagus, andstomach-and vice versa. Refined sugar, milk,eggs, and butter resembled meat in their patterns ofassociations, while pulses, vegetables, nuts, and fishresembled the cereals. The overall pattern suggestedan interaction between complex but systematicdietary and disease spectra, patterns of life-style,and geographical latitude, The question thereforearose whether within this pattern, more specificassociations could be detected. A first approachwas as follows.The foods most strongly associated with (a)

ischaemic heart disease, cancers of the largeintestine and breast, and multiple sclerosis, or with(b) respiratory tuberculosis, peptic ulcer, cirrhosis,nephritis and carcinoma of larynx, mouth, oeso-phagus, and stomach were readily identified. Athird class of foods showed irregular associations.An arbitrary list of 20 foodstuffs covering all threeclasses was then selected. These were: wheat, rice,coarse grain, pulses, vegetables, non-citrus fruits,wine, beer, sugar, potatoes, beef, pig meat,mutton, eggs, milk, cheese, fish, lard, butter, andmargarine.A matrix of correlation coefficients was next

assembled, and for each food the sum of theabsolute coefficients against all the other foods was

assembled. Large sums were taken to indicate that a

food would be near to one or other end of a

spectrum, and small sums to indicate that it wouldbe near the middle. Sugar and rice had the highestsums, and were negatively correlated with eachother, and they were therefore used to mark theopposite ends of a scale. The other foods were then

arranged in order, to construct the list given inTable 2. The coefficients in this table indicate highlevels of association between adjacent foods, anddissociations between foods situated some distancefrom each other. In an inexact sense this rankingevidently represents a gradient of diets passingfrom the Atlantic/North-Sea/Baltic coastline to theMediterranean, from high latitudes to low, andfrom affluence to less than affluence. The last pointis confirmed in Table 2 in the form of correlationswith total calorie intakes. Those foods near thefoot of the table are mainly 'primary' in the senseof being cultivated vegetables, and foods fromanimals not consuming cultivated crops. Those inthe upper part of the table are mainly 'secondary'and therefore more expensive foods in the sense ofbeing processed vegetable products (sugar, mar-garine, beer) or obtained from animals consumingcultivated crops.

Table 2 Spectrum offoodstuffsFood Adjacent-r Opposite-r Total

calories-r

I Sugar +0-48 -0-73 +0-722 Beer +0 90 -0-68 +0-283 Pig meat +058 -054 +0O504 Eggs +0 41 -050 +0 455 Butter +0-81 -054 +0O506 Milk +0 35 -0 40 +0537 Beef -0 19 -052 +0-318 Margarine +0<37 -0 41 +0*169 Potatoes +0 17 +0-13 +0<3010 Lard +0-06 +0-06 +0-2111 Cheese +0 11 +0-06 +0.0112 Mutton -0O05 +0 13 +0-2413 Coarse grains +0-02 -0-41 -0 1014 Fish +0-24 -052 -05915 Wine +0-48 -0A40 -0 3316 Fruit (non-citrus) +0-29 -054 -0-6217 Wheat +0*45 -0*50 +0 0118 Vegetables +0*58 -0*54 -0*3319 Pulses +0*61 -0*68 -0*5220 Rice - -0 73 -0 70

'Adjacent-r' is the correlation between the intake of each food andthe next on the list. 'Opposite-r' is the correlation between each food.and the food symmetrically opposite in rank-for example, food 4and food 17. 'Total calories-r' is the correlation between each foodintake and the total intake of calories from all sources.

The attempted arrangement of diseases into anordered set proved a more difficult and lessmeaningful task. Nevertheless, disease clustersrelated to common dietary patterns were evidentin the data, and a selection of these patterns isgiven in Table 3.

STANDARDISATIONA series of standardisation procedures was under-taken in which disease mortalities were expressedas excesses or deficiencies relative to 'expected'values as determined by total calories, total fat,total protein, and others, singly and in variouscombinations. Some associations persisted through-out all of these procedures, while others disappeared

73

copyright. on 28 M

ay 2019 by guest. Protected by

http://jech.bmj.com

/B

r J Prev S

oc Med: first published as 10.1136/jech.31.2.71 on 1 June 1977. D

ownloaded from

Table 3 Food spectrum and certain diseases

Stomach Cerebro- Ischaemic Cancer Cancer ofFood lTubercullosis Scmancr Peptic ulcer vascular heart Of large breastFood Tuberculosis cancer ~~~~~~~disease disease intestine

I Sugar -0-50 -0-42 -0-42 -0-42 0-74 0-67 0-822 Beer -0-14 -0-16 -0-30 -0-43 0-11 0-66 0-703 Pig meat -0-20 -0-28 -0-23 -0-42 0-30 0-57 0-674 Eggs -0-34 -0-51 -0-31 -0-41 0-61 0-88 0-72S Butter -0-17 -0-14 -0-18 -0-05 0-57 0-27 0-416 Milk -0-25 -0-16 -0-28 -0 09 0-71 0-23 0-417 Beef -0-47 -0-61 -0.40 -0-41 0-71 0-75 0-668 Margarine -0-48 -0-24 -0-45 -0-47 0-14 -0.09 0-309 Potatoes -0-09 0-16 0-05 0-18 0-18 0-19 0-1910 Lard 0-28 -0-08 -0-17 -0-24 -0-18 -0-93 0-2211 Cheese -0-15 -0-47 -0-63 -0-55 -0-35 -0-20 0-0412 Mutton 0-12 -0-06 -0-08 0-20 -0-04 -0-05 -0-2613 Coarse grain 0-38 0-45 0-59 0-63 -0-16 -0 40 -0 5014 Fish -0-19 0-24 0-13 0-22 -0-28 -0-37 -0-4415 Wine 0-06 0-15 0-16 0-24 -0-62 0-03 -0-3416 Fruit (non-citrus) 0-20 0-39 0-35 0-30 -0-71 -0-37 -0-5517 Wheat 0-78 0-23 0-25 0-26 -0-54 -0-54 -0-6818 Vegetables 0-15 0-29 0-34 0-38 -0-64 -0-16 -0-5319 Pules 0-61 0-21 0-22 0-32 -0-56 -0-63 -0-8020 Rice 0-06 0-48 0-46 0-63 -0-43 -0-45 -0-68

or appeared selectively depending upon the criterionof standardisation. These investigations were pur-sued on an intuitive basis relating to a number offood and disease clusters. Each will be presentedseparately.

ALCOHOL, CIRRHOSIS, AND DISEASES OFTHE UPPER DIGESTIVE TRACTWine intakes were recorded in 13 Europeancountries and varied from 8 g/day in Finland to308 g/day in Italy and 327 g/day in France. Wine

2000'

intakes were strongly associated with mortalityfrom cirrhosis of the liver in adults, particularlyin the age group 15-54 years (0- 79). Theassociation is shown in Fig. 1. Since this is a knowncausal effect, the strength ofthe statistical associationprovides a valuable yardstick against which otherassociations may be assessed when consideringother causal hypotheses.

Intakes of wine also showed strong associationswith mortality from cancer of the mouth (0-70)and of the larynx (0- 84) (see Fig. 2) in the 55-66-year group. Cancer of the oesophagus (0- 51)showed a weaker association. By contrast, cancerof the lung showed a negative correlation (-0-41),and cancer of the stomach a neutral one (O-15).

0

7-

0._

o0 .

N.R. 100Wine (q/doy)

200

Fig. 1 Wine intake and death from cirrhosis of liver inthose aged between 15 and 64 years.

0

N.R. 100 200 300Wine (q/day)

Fig. 2 Wine intake and death from carcinoma of larynxin those aged between 55 and 64 years.

0

74 E. G. Kniox

copyright. on 28 M

ay 2019 by guest. Protected by

http://jech.bmj.com

/B

r J Prev S

oc Med: first published as 10.1136/jech.31.2.71 on 1 June 1977. D

ownloaded from

Foods and diseases

The associations between wine and cirrhosis,cancer of the mouth and of the larynx survived arange of standardisation procedures and theassociation with cancer of the oesophagus wasenhanced (for example, to 0 * 57 after standardisationfor animal protein).

Intakes of beer were recorded for 10 countriesand, within the limited data available, were notpositively associated with cirrhosis nor with cancerof the mouth, the larynx, or oesophagus. The mainunstandardised associations of beer were withmultiple sclerosis (0' 93), hyperplasia of the prostate(0 74), appendicitis in the age group 55-64 years(0' 85), and cancer of the rectum (0 85).

GEOGRAPHICAL LATITUDEMultiple sclerosis, suicides, arthritis, cancer of thebreast, cancer of the prostate, and haemopoieticcancers were all associated with higher latitudes.Conversely, death from cirrhosis, chronic nephritis,respiratory tuberculosis, and cancer of larynxdecreased at higher latitudes. The foodstuffs in theupper part of Tables 2 and 3 were eaten more in thenorth-particularly, sugar, potatoes, milk, mar-garine, and butter. Conversely, rice, vegetables,pulses, nuts, non-citrus fruits, and wine were moreoften eaten and drunk further south. In the presentcontext the main interest in these latitudinalchanges of mortality and diet relates to therequirement for standardisation in interpreting awide range of positive and negative associations.Two diseases, diabetes and cancer of the skin, werehowever of particular interest in that their latitudeand gradients were the reverse of those that mighthave been expected.

Diabetes, contrary to a widespread belief, did notseem to follow an affluent life-style and was morefrequent in the most southern countries. It shouldbe remembered that we are considering deathsrather than incidence, which for this disease mayreflect the quality of the medical services. This isan important point which needs to be recalled in theinterpretation of other north/south associations.The other surprising result was an increase with

northerliness of death from cancer of the skin,presumably both epithelioma and melanoma. Studieswithin individual countries have more often shownthe reverse and the current material cannot beattributed to gradients similar to those suggestedin explanation of the diabetes gradient. It is possibleto envisage a saw-tooth north/south pattern, asoutherly increase within each country runningcounter to an international gradient running in theother direction. For example, racial variations inskin pigmentation may more than compensate forthe effects of excessive exposure to sunlight, or

people accustomed to sunshine may deliberatelyprotect themselves from it. Skin cancer may berelated to episodic sunburn rather than chronicexposure. Increasing exposure to industrial car-cinogens with increasing northerliness may alsohave contributed to a complex pattern.

VASCULAR DISEASEHypertensive disease, ischaemic heart disease, andcerebrovascular disease together constitute a majorpart of the total mortality for each country listed.Despite the common rubric of 'vascular disease'they constitute a heterogeneous group from anepidemiological point of view. Death rates fromhypertensive disease and cerebrovascular diseasetended to be high in the same countries and to havecommon foodstuff correlations, but were jointlydissociated from ischaemic heart disease mortalitiesand negatively correlated with its associateddietary patterns. The foodstuff associations ofcerebrovascular disease and ischaemic heart diseaseare shown in Table 3. Ischaemic heart disease hadpositive associations with total animal protein(0 74), total animal fats (0 69), and total fats(0 57), while cerebrovascular disease had negativeassociations with these aggregates. The onlynotable exceptions to these inverse patterns wererelated to the consumption of milk, butter, cheese,and vegetable oils. Cheese and vegetable oils hadnegative associations with all three disease groups.Milk and butter had positive associations not onlywith ischaemic heart disease, but also with hyper-tension in the age groups 15-54, although this wasnot true in older age groups. These patternswithstood standardisation for total calories and fora variety of other dietary components. Because ofdiffering theories on the aetiology of ischaemicheart disease the effects were examined ofstandardising for animal fats and refined sugar.In fact, the effects were barely distinguishableexcept that standardisation for refined sugarresulted in the deletion of a long list of previouslysignificant associations between animal fats and avariety of diseases (not just vascular diseases),while standardisation for animal fat resulted in thedeletion of a similar substantial list of unstandard-ised associations with refined sugar. This is scarcelysurprising in view of the high positive correlationbetween the intakes of the two foods (0 * 895).The negative associations with vegetable oils andcheese persisted not only with respect to ischaemicheart disease but also for hypertension in the15-54-year age group.The general conclusion to be drawn is that no

specific foodstuff, no foodstuff group, and noneof the main calorie-providing nutrients can be

75

copyright. on 28 M

ay 2019 by guest. Protected by

http://jech.bmj.com

/B

r J Prev S

oc Med: first published as 10.1136/jech.31.2.71 on 1 June 1977. D

ownloaded from

76

resolved on the basis of these data as a primarydeterminant of death rates from ischaemic heartdisease, hypertension, or cerebrovascular disease.

DISEASES OF THE DIGESTIVE SYSTEMThe relationships between wine, cirrhosis, cancersof mouth, larynx, and oesophagus have beenmentioned already. A more comprehensive reviewof digestive tract diseases shows that the patternof food associations undergoes a systematicchange as the site moves from the lips to the anus.The diseases of the upper digestive tract areassociated with low calorie intakes and a highproportion of vegetable sources, and in manycountries these diseases are decreasing in frequency.Those at the lower end are associated with highcalorie and protein intakes and a high proportionof animal sources, and in many countries areincreasing in frequency. We must recall again thatin the present investigation we are concerned onlywith deaths, which for diseases such as appendi-citis and gall bladder disease may be mainlyreflections of case fatality and of the quality ofavailable medical care, while for many of thecancers the death rates are a true reflection ofincidence.The correlation between wine and cancer of the

mouth was the strongest of the associations forthat disease and additional correlations with 'edibleoffals' (0 *67), veal (0*67), and poultry (0*55)reflect the characteristic meat diets of the winedrinking countries. Wine also gave the strongestcorrelation with cancer of the larynx (0'84);additional associations with vegetables (0 74), veal(0 56), and nuts (0 48) can also be regarded assecondary associations. Cancer of oesophagus hadweaker associations with wine intakes than did thetwo cancers already mentioned. Indeed, its moststriking association was with intakes of horsemeat (0 60). This estimate was based upon 13countries for which statistics were quoted, andcan be attributed almost entirely to the high ratesand the high intakes of two of them-namelyBelgium and France. Almost certainly the inter-pretation is indirect.Cancer of stomach (and other gastrointestinal

cancers below the stomach) showed no associationswith intake of wine. Most of the food associationsfor stomach cancer can probably be interpreted ingeneral rather than in specific terms as indicators oflow levels of affluence. For example, its associationswith the various aggregate sources of calorieswere all negative and the only positive associationswere rather weak ones with wheat (0 52), rice(0'49), total cereals (0'48), and nuts (0'46). Theassociation with salt/cured/dried fish (-0 27)

E. G. Knox

was negative and not significant, when Japanwas excluded. It would be difficult from these datato support the suggestion made by Correa et al.(1975) that abrasive foods might pre-dispose.Cancer of the large intestine excluding the

rectum presented a contrasting picture. It hadstrong positive associations with several caloriesources, particularly animal protein (0*74), fat(0-72), and total calories (0-72), and a negativeassociation with vegetable protein (-0 71). It is adisease of affluent northerly flesh-eating nationsand the pattern of positive and negative associationsis compatible with the observation that intakes ofvegetable proteins were negatively correlated withintakes of animal proteins (-0 50) and with totalcalories (-0 78). Nevertheless, a number ofindividual foodstuffs and foodstuff groups yieldedeven stronger correlations than those of theaggregate sources including total meat (0 *89),eggs (0 83), and beef (0 75). Beer (0 66), sugar(0-67), and pig meat (0 57) also supplied positiveassociations. The correlation with total fat isshown in Fig. 3.

50' T-T^^J = Japon

0

0

0

0 100Fot calories per day

Fig. 3 Fat intake and death from carcinoma of largeintestine in those aged between 55 and 64 years.

Cancer of the rectum behaved much like cancerof the large intestine except that most of thecorrelations were somewhat weaker. The strongestassociations with aggregate calorie sources includedtotal fat (0 62) and animal fat (0'60) and thosewith specific foodstuffs included a number con-taining fats such as pig meat (0'72), total meat(0'60), and eggs (0'52). Sugr (0-51) and beer(0' 85) were also associated; the strength of the lastassociation was particularly notable although basedon only 10 countries.

I copyright. on 28 M

ay 2019 by guest. Protected by

http://jech.bmj.com

/B

r J Prev S

oc Med: first published as 10.1136/jech.31.2.71 on 1 June 1977. D

ownloaded from

Foods and diseases

Death from peptic ulcer followed the patternfor carcinoma of the stomach and associationswith aggregate calorie sources were similar for thetwo diseases. Among individual foodstuffs therewere only two significant positive associations withpeptic ulcer-namely, total coarse grains (0*59)(including oats, barley, and maize) and nuts (0 50).The most reasonable interpretations are in generalrather than in specific terms.

Deaths from gall bladder disease revealed onlytwo significant associations-with beer (0 63) andrye (0 60). Appendicitis deaths were significantlyrelated to the same two foods and to no others, theassociation with beer (0*85) being especiallystriking in this case. Despite the fact that gallbladder disease and appendicitis have both beenlinked with cancer of the large intestine and rectumas diseases determined by 'western' diets, neither-at least as far as deaths are concerned-wasparticularly related to intake of fat or meat.

FAT ASSOCIATED DISEASESAssociations with fat intakes have already beendescribed both for ischaemic heart disease andcancer of the large intestine. Particularly for thesecond the strength of the association, thespecificity of foodstuffs involved, and the resistanceof the correlations to deletion through standardisa-tion, all make for the reasonable formulation of acausal hypothesis related to total fats, particularlyanimal fats. The question arises whether otherdiseases could be incorporated within a morecomprehensive postulate.

Because fat is a concentrated source of caloriesit is necessary to examine the pattern of both, tosee if one set of associations represents a primaryeffect, and the other a secondary. In fact, totalcalories supplied weaker and fewer correlationsthan fats in general, or animal fats in particular.Total calories gave significant positive correlationsfor multiple sclerosis (0-61) and cancer of thebreast (0 65). Fat intakes showed striking corre-lations with cancer of the breast (088), cancer ofthe haemopoietic system (0 75), multiple sclerosis(075), and cancer of the large intestine (0 72).There were less striking associations with cancerof the rectum (0*62), ischaemic heart disease(0-57), cancer of the prostate (0 56), and cancer ofthe lung (0 56). The strength of the correlationwith cancer of the lung provides a useful yardstick;we know the main cause of this disease and, sinceit is non-dietary, we can attribute other correlationsof this strength to secondary associations.Examination was also made of division of the

diet between animal and vegetable sources ratherthan according to nutrient content. Several diseases

followed a pattern of positive associations withanimal products and negative associations withvegetable products and this must reflect the life-styles with which these diseases are associated.However, for four diseases in particular-namely,multiple sclerosis, cancer of the large intestine,cancer of the rectum, and cancer of the breast-thepositive correlations with total fats were evenstronger than the correlations with total animalproducts. The findings of chief interest are theextraordinary levels of statistical dependence uponfat of cancer of the breast (see Fig. 4), multiplesclerosis, and cancer of the large intestine (Fig. 3).

60I=aopon

50

30 * _

20-~~~~~40 /

10/

20 100 200fat colories per day

Fig. 4 Fat intake and death from carcinoina of breastin those aged between 55 and 64 years.

OTHER ASSOCIATIONS OF INTERESTA number of additional associations, not alreadylisted, raised some points of interest.

1. Nephritis and nephrosis were associated withpulses (0*78), and total cereals (0*79)including bread grain (0 69) (wheat andrye) and coarse grain (0 70) (oats, barley,and maize), and rice (0 78). In the first threerespects these results resembled those forrespiratory tuberculosis (0-83, 0 -78, 0 -61)although its associations with coarse grain(0 38) and rice (0-04) were not significant.A common interpretation relates both diseasesto the diets associated with poverty, althoughassociations between high extraction cereals

77

copyright. on 28 M

ay 2019 by guest. Protected by

http://jech.bmj.com

/B

r J Prev S

oc Med: first published as 10.1136/jech.31.2.71 on 1 June 1977. D

ownloaded from

E. G. Knox

and stones in the renal tract suggest ahypothesis that may be worth exploringfurther.

2. Death from arthritis and spondylitis showedstriking correlations with intakes of oats(080), sugar (0'70), milk (0X81), butter(0 88), but not with meat, eggs, or cheese.Oats and rye are atypical of cereal foodsin reflecting northern patterns of living andthe explanation of this association probablylies here. The extreme positions of milk andbutter within this pattern, however, raises aquestion of interest with the possibility of aspecific protein allergy.

3. Epilepsy was another disease reflecting lowlevels of affluence. Coarse grain (0 57), rice(0 70), and nuts (0-55) had strong positiveassociations, while sugar (-0 56) and eggs(-0 46) had negative ones. There were noassociations, sufficiently striking to suggestspecific dietary causes.

Discussion

The complexity of the associations between food-stuffs and diseases has been demonstrated. Varyingreporting practices, varied precision, the differentage distributions of different populations, and theconcatenation of both incidence and fatality ratewithin mortality statistics, all give rise to difficulties.A first conclusion must be that international corre-lation analyses which limit themselves to a singledisease or disease group and a small range ofarbitrarily selected foods can contribute little to theformulation and testing of causal hypotheses.Some authors such as Shennan and Bishop (1974)and Armstrong and Doll (1975) have recognisedthese difficulties but many others have adopted toodirect and naive an approach.

In these circumstances the utility of statisticalprocedures is difficult to assess. Significance testsprovide an inefficient and a largely inappropriatescreen and it may be better to use as yardsticks,estimates of correlation obtained for diseaseswhose aetiologies are well established, includingboth those where the causes are dietary and thosewhere they are not. The correlations of cirrhosisof the liver (with alcohol), of lung cancer, roadaccidents, and suicide (with fat) of respiratorytuberculosis (with cereals) and of death fromdrowning (with fish) suggest that correlationcoefficients with values less than about 0-55 arecapable of being explained indirectly (if not simplyas sampling variation), while those greater thanabout 0 70 require closer attention. The utility ofstandardisation procedures is also open to question;

in a complex situation the assignation of a 'primary'association to one variable and the relegation ofanother to 'secondary' status may depend uponquite small differences, and the outcomes mayvary for different data sets. For example, Keys(1971, 1973) carried out analyses which he interpretedas showing that the relationship between athero-sclerosis and intake of sugar was secondary to anassociation with intake of fats. This is the reverse ofYudkin's assertion (Yudkin, 1957) and differs againfrom the conclusion reached in the current reportnamely, that the data do not permit theirdifferentiation. Burbank (1972) conducted a so-phisticated geographical time analysis ofcancer mort-alities in the United States of America in which heidentified disease 'clusters' within the total pictureand attempted to relate them to identifiable andhypothetical carcinogens. This too is a moreformal approach than was attempted or thoughtdesirable in relation to the data presented here.The view taken is that sophisticated analyticalmethods applied to crude data should be regardedas legitimate means of display, awaiting subsequentinformal interpretation, but that they are not sup-portable as formal decision processes. There is noescaping the necessity to continue to seek sets ofdata of different kinds, from different sources, andat different times, if we wish to discover which of theassociations is sufficiently consistent to warrantattempted causal explanations.The necessarily tentative nature of any causal

hypothesis should now be evident but withinthese restrictions the following associations deservespecial notice.

I. The strength of the correlation of wine withcirrhosis was exceeded by the strength of itsassociation with cancer of the larynx andclosely approached by that for cancer of themouth; there was a substantial, althoughweaker, association with cancer of theoesophagus. The topographical continuityof these three cancer sites lends supportto the hypothesis that wine drinking, or theconsumption of alcoholic beverages whoseamounts are highly correlated with those ofwine, may contribute causally to thesediseases. This is not the first time that such asuggestion has been made and, particularlyfor the mouth and larynx several case/controlstudies have demonstrated associations atthe level of the individual. Kissin et al.(1973) demonstrated an excess of head andneck cancer in alcoholics compared withcontrols, while Wynder and Bross (1957),Wynder et al. (1957), and Vincent andMarchetta (1963) demonstrated excessive

78

copyright. on 28 M

ay 2019 by guest. Protected by

http://jech.bmj.com

/B

r J Prev S

oc Med: first published as 10.1136/jech.31.2.71 on 1 June 1977. D

ownloaded from

Foods and diseases

alcohol intakes in patients with cancer of thebuccal cavity, compared with controls. Thelast of these investigations demonstratedsimilar effects in relation to cancer of thepharynx and larynx, confirming a findingof Wynder et al. (1956) in patients withcancer of larynx. Keller and Terris(1965) conducted similar investigations in adetermined effort to separate the effects oftobacco and alcohol usage, which arethemselves correlated; they confirmed theconclusion of Wynder et al. that the effectsare separate and additive. The study ofShennan and Bishop (1974) did not presentor analyse information related to alcoholicdrinks. Although Vogler et al. (1962) failedto find an association between this groupof cancers and alcohol intake in the southernUnited States of America, the evidence ofthese various studies, combined with that ofthe current investigation can leave littledoubt that wine or-as Wynder suggestsspirits, are potent causes of buccal pharyn-geal and laryngeal cancers, although notexclusive causes. Clemmeson (1965) suppliedan extensive review of the literature up to1965 and Breslow and Enstrom (1974) havecarried out more recent and extensive studies.These recent studies, limited to the UnitedStates of America, appeared to show that theeffects of alcohol and tobacco were separable,and that both were important. This workalso demonstrated the associations betweenintakes of beer and cancers of the rectumand colon.

2. Intakes of total fat were sufficiently highlyand specifically correlated with three diseasesto permit consideration of causal hypotheses.They were multiple sclerosis, cancer of thelarge intestine, and cancer of the breast.Cancer of the rectum had a similar butweaker pattern of association to thatexhibited by cancer of the large intestine.The extent and specificity of these cancerassociations was found also by Shennanand Bishop (1974) in a study of 32countries of which 12 were also included inthe current investigation. Drasar and Irving(1973) noted the high correlation betweenthe occurrence of cancer of the colon andcancer of the breast and the high correlationbetween each and the intake of fat. Asso-ciations between fat intake and multiplesclerosis have been noted before; onerecent study (Agranoff and Goldberg, 1974),suggested a specific association with milk

79

intake; this was not confirmed as anassociation of special note in the currentinvestigation.

The epidemiology of multiple sclerosis drawnupon a world scale is so complex that it would befoolish to interpret the fat association in thelimited context provided by these data. The diseaseis known to demonstrate latitudinal variation inprevalence in the same direction as variation in thefat content of diet, within this range of material;the explanation might indeed be found eventuallyin a component of the diet but it could also beexplained in other terms and the association withdiet then seen to be secondary. However, withinthe range of information assembled for this studyit must be remarked that the correlation with fatintake is a strong one.Cancer of the large intestine and rectum provide

more direct scope for causal dietary interpretations.High fat diets probably contain relatively little fibreand to this extent the association is in keeping withcurrent popular hypotheses (for example, Burkitt,1971). Nevertheless, although data on fibre contentwere not available, there were no significantcorrelations between large bowel cancer andconsumption of vegetables or the total gramsweight of the diet. The same findings and the samecomment have been made by other workers, forexample, by Haenszel et al. (1973) in their study oflarge bowel cancer in immigrant Hawaiian Japanese.Several reviews of this problem (Lancet, 1974;British Medical Journal, 1975) have pointed to theweak data on which the fibre hypothesis is basedand the apparently stronger case for incriminatingfat among the aggregate nutrients, or of beef andsugar among the specific foodstuffs. Although therelationship between fat and breast cancer hasreceived less attention than that between fat andcolon cancer, the dependence of breast cancermortality upon fat in the data presented here isremarkable and a causal association between thisdisease and this nutrient, or some component orclose correlate of the nutrient, must seriously beconsidered.The fat associations of breast cancer provide an

indirect comment upon those with colon cancer.Within the context of a unitary causal hypothesisthe findings suggest that the effect must be morethan a simple mechanical process within theintestine. Unless the relationships of the twodiseases represent completely different mechanisms,there is a suggestion of absorbable cancer-producingmaterials capable of acting at a distance. It mightthen be difficult, for example, to discard as indirectthe fat associations of cancer of the haemopoieticsystem.

copyright. on 28 M

ay 2019 by guest. Protected by

http://jech.bmj.com

/B

r J Prev S

oc Med: first published as 10.1136/jech.31.2.71 on 1 June 1977. D

ownloaded from

80

Reprints from: Professor E. G. Knox, HealthServices Research Centre, Department of SocialMedicine, Medical School, Edgbaston, BirminghamB15 2TJ.

References

Agranoff, B. W., and Goldberg, D. (1974). Diet and thegeographical distribution of multiple sclerosis. Lancet,2, 1061-1066.

Armstrong, B., and Doll, R. (1975). Environmentalfactors and cancer incidence and mortality in differentcountries, with special reference to dietary practices.International Journal of Cancer, 15, 615-631.

Breslow, N. E., and Enstrom, J. E. (1974). Geographicalcorrelations between cancer mortality rates andalcohol-tobacco consumption in the US. Journal ofthe National Cancer Institute, 53, 631-639.

British Medical Journal (1975). Editorial: Faecal fibrefortunes. British Medical Journal, 2, 580-581.

Burbank, F. (1972). A sequential space-time analysis ofcancer mortality in the United States: etiologicimplications. American Journal of Epidemiology, 95,393-417.

Burkitt, D. P. (1971). Some neglected leads to cancercausation. Journal of the National Cancer Institute,47, 913-919.

Clemmeson, J. (1965). Statistical studies in the aetiologyof malignant neoplasms. I. Review and results. Actapathologica et microbiologica Scandinavica, Supplement174.

Correa, P., Haenszel, W., Cuello, C., Tannenbaum, S.,and Archer, M. (1975). A model for gastric cancerepidemiology. Lancet, 2, 58-59.

Drasar, B. S., and Irving, D. (1973). Environmentalfactors and cancer of the colon and breast. BritishJournal of Cancer, 27, 167-172.

Haenszel, W., Berg, J. W., Segi, M., Kurihara, M., andLocke, F. B. (1973). Large bowel cancer in HawaiianJapanese. Journal of the National Cancer Institute,51, 1765-1779.

E. G. Knox

Keller, A. Z., and Terris, M. (1965). The association ofalcohol and tobacco with cancer of the mouth andpharynx. American Journal of Public Health and theNation's Health, 55, 1578-1585.

Keys, A. (1971). Sucrose in the diet and coronary heartdisease. Atherosclerosis, 14,193-202.

Keys, A. (1973). Letter: Sucrose in the diet and coronaryheart disease. Atherosclerosis, 18, 352.

Kissin, B., Kaley, M. M., Su, W. H., and Lerner, R.(1973). Head and neck cancer in alcoholics. Journal ofthe American Medical Association, 224, 1174-1175.

Lancet (1974). Editorial: Beware the ox. Lancet, 1,791-792.

Organisation for European Co-operation and Develop-ment (1970). Food Consumption Statistics 1960-1968.OECD: Paris.

Shennan, D. H., and Bishop, 0. S. (1974). Diet andmortality from malignant disease in 32 countries.West Indian Medical Journal, 23, 44-53.

Vincent, R. G., and Marchetta, F. (1963). The relation-ship of the use of tobacco and alcohol to cancer of theoral cavity, pharynx, and larynx. American Journal ofSurgery, 106, 501-505.

Vogler, W. R., Lloyd, J. W., and Milmore, B. K. (1962).A retrospective study of etiological factors in cancer ofthe mouth, pharynx, and larynx. Cancer, 15, 246-258.

World Health Organisation (1967). International Classifi-cation of Diseases, 8th revision. WHO: Geneva.

World Health Organisation (1973). Vital statistics andcauses of death. In World Health Annual Statisticsfor 1970. WHO: Geneva.

Wynder, E. L., and Bross, I. J. (1957). Aetiologicalfactors in mouth cancer. British Medical Journal, 1,1137-1143.

Wynder, E. L., Bross, I. J., and Day, E. (1956). A studyof environmental factors in cancer of the larynx.Cancer, 9, 86-1 10.

Wynder, E. L., Bross, I. J., and Feldman, R. M. (1957).A study of the etiological factors in cancer of themouth. Cancer, 10, 1300-1323.

Yudkin, J. (1957). Diet and coronary thrombosis:hypothesis and fact. Lancet, 2, 155-162.

copyright. on 28 M

ay 2019 by guest. Protected by

http://jech.bmj.com

/B

r J Prev S

oc Med: first published as 10.1136/jech.31.2.71 on 1 June 1977. D

ownloaded from