Cancer Therapy Vol 4, 153-162, 2006

Nutritional patterns and lung cancer risk in Uruguayan men

Research Article

Eduardo De Stefani¹, Alvaro L. Ronco², Paolo Boffetta³, Hugo Deneo-Pellegrini¹, Pelayo Correa⁴, Giselle Acosta¹, Luis Piñeyro Gutiérrez⁵ and María Mendilaharsu¹

¹Grupo de Epidemiología, Departamento de Anatomía Patológica, Hospital de Clínicas, Facultad de Medicina, Montevideo, Uruguay

²División de Epidemiología, Instituto de Radiología, Hospital Pereira Rossell, Montevideo, Uruguay

³International Agency for Research on Cancer, Lyon, France

⁴Department of Pathology, Louisiana State University Health Sciences, New Orleans, Louisiana, USA

⁵Instituto de Neumología, Facultad de Medicina, Montevideo, Uruguay

__________________________________________________________________________________

*Correspondence: Dr. Eduardo De Stefani, Avenida Brasil 3080 dep. 402, Montevideo, Uruguay; Tel.: (598) 2 708 23 14; Fax: (598) 2 402 08 10; E-Mail: estefani@adinet.com.uy

Key words: lung cancer, factor analysis, principal components, dietary patterns

Abbreviations: food-frequency questionnaire (FFQ); heterocyclic amines, (HCA); odds ratios, (ORs)

Supported by a grant N° ECE/98/17 from International Agency for Research on Cancer, Lyon, France

All authors contributed to the statistical analysis, interpretation of the data, and preparation of the manuscripts.

Received: 26 January 2006; Revised: 20 February 2006

Accepted: 10 April 2006; electronically published: April 2006

Summary

Lung cancer is the more frequent malignancy among Uruguayan men. The age-adjusted incidence rate is 76.5 per 100,000 inhabitants of Montevideo. In comparisons between American registries, the rate of Uruguay is only second, following the rate observed among Black men in United States. Doubtless the main etiologic factor of lung cancer in Uruguay is tobacco smoking. Uruguayan men display a high prevalence rate of smoking and this population is also characterized by an elevated use of black tobacco cigarettes and by the use of hand-rolled cigarettes. Both types of cigarettes have been considered as particularly risky. Also occupation and dietary factors have been considered as risk factors for this malignancy. Previous studies on diet and lung cancer suggested that vegetables, fruits, carotenoids and fat were associated with lung cancer risk. A case-control study which included 846 newly diagnosed and microscopically confirmed male cases with lung cancer and 846 male hospitalized patients with non-neoplastic diseases was conducted in Montevideo, Uruguay. Both series of participants were directly interviewed by two trained social workers in the four major public hospitals in Montevideo. Cases and controls were frequency matched on age and residence. The patients were analyzed by factor analysis (principal components) with an eigenvalue of 1.25 as a limit for retained principal components. The analysis allowed to retain four factors which were labelled as “healthy”, “western”, “fatty” and “cheese and rice”. The “cheese and rice” and “healthy” patterns were protective (OR for the higher score of the “cheese and rice” factor 0.55, 95 % CI 0.37-0.80), whereas the “western” and the “fatty” patterns were directly associated with a significant increase in risk of lung cancer (OR for the “fatty” pattern 2.33, 95 % CI 1.65-3.28). We concluded that factor analysis is a valuable statistical method which allows to reduce complex sets of data in a smaller number of factors. These, in turn, are possibly more explicative than the traditional logistic analysis of individual foods or food groups. Aside from the recognized and well-known role of tobacco smoke in the etiology of lung cancer, the role of diet should be taken into account as an important non-tobacco risk factor for this malignancy.

I. Introduction

Lung cancer is the most frequent malignancy among Uruguayan men, with an age-adjusted incidence rate of 76.5 per 100,000 inhabitants of Montevideo (Parkin et al, 2002). In fact, in comparisons between American registries, the rate of Uruguay is only second, following the rate observed among Black men in United States (Parkin et al, 2002).

Undoubtly the main etiologic factor of lung cancer in Uruguay is tobacco smoking (Carámbula et al, 1995). Uruguayan men display a high prevalence rate of smoking and this population is also characterized by an elevated use of black tobacco cigarettes and by the use of hand-rolled cigarettes (Carámbula et al, 1995; Engeland et al, 1996; De Stefani et al, 1997a). Both types of cigarettes have been considered as particularly risky (Engeland et al, 1996; De Stefani et al, 1997a). Also occupation and dietary factors have been considered as risk factors for this malignancy (Hueper, 1966; Ziegler et al, 1996; World Cancer Research Fund, 1997).

Traditionally, research on the role of diet and lung cancer risk have been conducted, both in prospective and case-control studies, by individual analysis of foods and/or nutrients (Ziegler et al, 1996; World Cancer Research Fund, 1997). Rather recently, factor analysis, that is a statistical method for simplifying and reducing complete sets of data into a rather small number of factors, has begun to be employed in the field of dietary epidemiology (Randall et al, 1992; Slattery et al, 1998; De Stefani et al, 1999; McCann et al, 2001; Palli et al, 2001; Terry et al, 2001a, b; Handa and Kreiger, 2002; Markaki et al, 2003; Masaki et al, 2003; Sieri et al, 2004; De Stefani et al, 2005; Fung et al, 2005). Initially used in the field of psychology (Pearson, 1901; Spearman, 1904), this method has been employed in the field of social sciences, economy and other sciences (Harman, 1976; Kim and Mueller, 1978; Kline, 2002).

The role of dietary patterns in the etiology of lung cancer was studied in two previous reports (Tsai et al, 2003; Balder et al, 2005). The first study (Tsai et al, 2003) used cluster analysis and reported that a “healthy” pattern had a protective effect against lung cancer. However, after controlling for smoking this inverse association was no longer significant. The second study used factor analysis and reported a protective effect for a so-called “salad” pattern whereas the “pork, processed meat and potatoes” pattern was directly associated with lung cancer risk (Balder et al, 2005).

We considered that a research on foods and lung cancer risk, using exploratory factor analysis, could be worthwhile in order to replicate previous studies in the field (Tsai et al, 2003; Balder et al, 2005). This is the objective of the present research.

II. Methods

In the time period 1995-2001, a case-control study on environmental risk factors and lung cancer was conducted in Montevideo, Uruguay.

A. Selection of cases

All newly diagnosed and microscopically confirmed cases of lung carcinoma were considered eligible for the present study. Thus, 870 patients with lung carcinoma were identified in the four major hospitals in Montevideo, Uruguay. Twenty one patients presented extensive disease with brain metastases and four patients refused the interview. These patients were excluded from the study, leaving a final number of 846 participants (response rate 97.2 %). The cases were distributed by cell type as follows: squamous cell (308 patients, 36.4 %), small cell (105, 12.4 %), adenocarcinoma (212, 25.1 %), large cell (37, 4.4 %) and unclassified carcinoma (184, 21.7 %) using the World Health Organization (WHO) classification for tumors of the lung and pleura (Travis et al, 1999). The relatively large number of unclassified cancers was, in part, to the fact that 87 % of these tumors were of peripheral location and most of them (64 %) were microscopically diagnosed by cytology.

B. Selection of controls

In the same time period and in the same hospitals, all male patients hospitalized for conditions not related with tobacco smoking, alcohol drinking and without recent changes in their diets were considered as eligible for the study. One thousand and six hundred (1600) patients were identified through the log book of admissions. Forty two patients refused the interview, leaving a final number of 1158 potential controls (response rate 96.5 %). Among them 846 controls were frequency matched to cases on age (in ten-years intervals) and residence (Montevideo, other counties) and retained in the current analysis. Controls presented the following diseases: eye disorders (178 patients, 21.0 %), abdominal hernia (170, 20.1 %), fractures (130, 15.4 %), injuries (77, 9.1 %), skin diseases (67, 7.9 %), acute appendicitis (66, 7.8 %), varicose veins (42, 5.0 %), hydatid cyst (40, 4.7 %), urinary stones (25, 3.0 %), blood disorders (21, 2.5 %), prostate hypertrophy (18, 2.1 %) and osteoarticular diseases (12, 1.4 %).

C. Questionnaire

Cases and controls were interviewed shortly after admittance in the hospitals. All the interviews were performed face-to-face by two trained social workers. No proxy interviews were accepted. The questionnaire administered to the participants included the following sections: (1) sociodemographics, (2) a complete occupational history based in job titles and its duration, (3) family history of lung cancer or other cancers in first degree relatives, (4) self-reported height and weight five years before the date of the interview, (5) a complete history on tobacco smoking, including age at start, age of quit, average number of cigarettes smoked per day, type of tobacco used (based in brands), type of cigarette, filter use, degree of inhalation (no inhalation, mouth and chest), (6) a complete history of alcohol drinking including age at start, age of quit, number of glasses drunked per week or day, type of alcoholic beverage, (7) a maté drinking section including age at start, age at quit, liters or fractions of liter ingested per day (or week), temperature of the beverage and (8) a food-frequency questionnaire (FFQ) on 64 food items. This FFQ is considered as representative of the usual diet of Uruguayans and allowed the estimation of total energy intake. Furthermore, the FFQ was not validated but it was tested for reproducibility with good results (see Appendix I). Queries about food intake concerned to consumption five years before the date of the interview, both for cases and controls. We have already reported results on selected nutritional aspects and lung cancer which are partially based on this population (Deneo-Pellegrini et al, 1996; De Stefani et al, 1999).

D. Foods included in the study

The following food and food groups were included in the factor analysis: red meat (beef, lamb), white meat (poultry, fish), processed meat (bacon, sausage, mortadella, salami, saucisson, hot dog, ham, salted meat), cheese, butter, whole milk, total eggs (boiled eggs, fried eggs), desserts (milk with sugar, rice pudding, custard, marmalade, ice cream, cake), white rice, fresh vegetables (carrot, tomato, lettuce, onion), cooked vegetables (garlic, swiss chard, spinach, potato, sweet potato, winter squash, beetroot, zucchini, cabbage, cauliflower, red pepper), citrus fruits (orange, tangerine), other fruits (apple, pear, grape, peach, plum, banana, fig, fruit cocktail), coffee with milk, maté, and total alcohol (beer, wine, hard liquor).

E. Statistical analysis

Factor analysis (principal components) was conducted to derive food patterns based on the 16 foods and food groups. This analysis was conducted using the principal components factor procedure in STATA (1999). The factors were rotated by the varimax function to achieve a simpler structure with greater interpretability. In determining the number of factors to retain we used eigenvalues greater than 1.25 and the Scree test. Positive loadings indicates that the foods are directly correlated with the factor, and negative loadings indicate that the foods are inversely correlated with the factor. Labeling of the food patterns was entirely subjective and was based on our interpretation of the loadings observed in the correlation matrix. Scores for the retained factors were obtained by the score command in STATA (1999). Correlations between selected variables and scores were estimated by the procedures of Pearson and Spearman (STATA, 1999).

Odds ratios of lung cancer for estimated scores were estimated by multiple unconditional logistic regression (Breslow and Day, 1980). The basic model included the following terms: age (continuous), residence, urban/rural status, education (categorical), family history of lung cancer among first degree relatives, body mass index (continuous), total energy intake (continuous), and the following tobacco variables: smoking status, years since quit among former smokers (categorical), cigarettes per day among current smokers (categorical), age at start smoking (continuous). Since scores are conditional on each other, the four scores were included simultaneous in the models (Balder et al, 2005). Scores was introduced in this basic model as categorical variables. The categories were defined in quartiles, following the distribution of the controls. Departure from the multiplicative model was determined by assessing the likelihood ratio test statistic. An a level of 0.05 was used as the indicator of statistical significance and, accordingly, 95 % CIs were reported. All p-values were derived from two-sided statistical tests. Each factor was entered as continuous into the model in order to estimate the p-value for trend.

Comparisons between cell types of lung cancer for food patterns were estimated by polytomous (multinomial) regression (Rothman and Greenland, 1998). All the calculations were performed with the STATA software programme (STATA, 1999).

III. Results

Distribution of cases and controls by sociodemographic variables and selected risk factors are shown in Table 1. As a result of the matched design, cases

Table 1. Distribution of cases and controls by sociodemographics and selected risk factors

		Cases	Controls
Variable	Category	Nº %	Nº %
Age (in years)	30-39	14 1.6	14 1.6
	40-49	93 11.0	93 11.0
	50-59	186 22.0	186 22.0
	60-69	334 39.5	334 39.5
	70-79	200 23.6	200 23.6
	80-89	19 2.3	19 2.3
Residence	Montevideo	410 48.5	410 48.5
	Other counties	436 51.5	436 51.5
Urban/rural status	Urban	645 76.4	664 78.5
	Rural	201 23.8	182 21.5
Education (years)	0-2	224 26.5	204 24.1
	3-5	333 39.4	311 36.8
	6+	289 34.1	331 39.1
Income (US dollars)	≤ 144	326 38.5	322 38.1
	145+	326 38.5	318 37.6
	Unknown	194 22.9	206 24.3
Family history of lung cancer	No	765 90.4	811 95.9
	Yes	81 9.6	35 4.1
Body mass index	≤ 22.8	282 33.3	218 25.8
	22.9-24.8	216 25.5	205 24.2
	24.9-27.0	165 19.5	214 25.3
	27.1+	183 21.6	209 24.7
Total energy intake	≤ 1722	171 20.2	211 24.9
	1723-2071	181 21.4	212 25.1
	2072-2466	211 24.9	212 25.1
	2467+	283 33.5	211 24.9
Tobacco smoking	Never smokers	19 2.3	146 17.3
	Former 20+yrs	35 4.1	86 10.2
	Former 10-19 yrs	51 6.0	88 10.4
	Former 1-9 yrs	154 18.2	124 14.7
	Current 1-9 cig/day	19 2.2	84 9.9
	Current 10-19 c/day	84 9.9	133 15.7
	Current 20-29 c/day	183 21.6	96 11.3
	Current 30+ cig/day	301 35.6	89 10.5
Nºpatients		846 100.0	846 100.0

and controls were similar concerning age and residence. Also the distribution of controls and cases were rather similar concerning urban/rural status and monthly income. Cases were slightly less educated than controls and were significantly leaner when compared with controls. On the other hand total energy intake was significantly higher among cases than controls and family history of lung cancer among first-degree relatives displayed an increased risk of the malignancy (OR 2.4, 95 % CI 1.6-3.7). Finally, tobacco smoking, as expected, was strongly and directly associated with lung cancer risk (OR for current heavy smokers 26.0, 95 % CI 13.3-59.9, p-value for linear trend <0.0001).

Four major food patterns were retained (Table 2). Factor or pattern 1 was labeled “cheese and rice” since it reflected the correlated intake of some dairy foods, white rice and processed meat, explaining 14.2 % of the variance. Factor 2 was labeled as “Western” since it presented high loadings of red meat and alcohol. This pattern explained 9.5 % of the variance. Pattern 3 was labeled as “fatty” and was rich in whole milk and coffee with milk. This factor explained the 8.5 % of the variance. Factor 4 was labeled as “healthy” and presented high correlations of white meat, fresh vegetables, cooked vegetables, citrus fruits and non-citrus fruits and explained 8.0 % of the variance.

Correlations between patterns and selected variables for controls are shown in Table 3. The “cheese and rice” pattern was directly associated with total energy inatake and with lactose. The “western” factor was composed by patients who were less educated. Also this pattern was significantly correlated with smoking intensity, total energy intake, saturated fat, monounsaturated fat, polyunsaturated fat and with cholesterol. The “western” pattern was negatively correlated with total carbohydrates, calcium and lactose. The “fatty” pattern displayed elder patients and there was a significant

Table 2. Factor-loading matrix among controls¹

Foods	Factor 1 (Cheese & Rice)	Factor 2 (Western)	Factor 3 (Fatty)	Factor 4 (Healthy)
Red meat	0.04	0.67	0.05	0.03
White meat	0.30	-0.27	-0.08	0.46
Processed meat	0.47	0.40	0.09	-0.02
Cheese	0.67	-0.07	-0.04	0.14
Butter	0.68	0.05	0.01	-0.04
Whole milk	0.01	-0.02	0.77	0.09
Eggs	0.14	0.38	0.08	0.08
Desserts	0.47	0.08	0.27	0.22
White rice	0.46	0.10	-0.09	-0.01
Fresh vegetables	-0.04	0.10	0.01	0.66
Cooked vegetables	0.15	0.24	0.17	0.43
Citrus fruits	-0.11	0.18	0.08	0.57
Other fruits	0.13	-0.12	0.03	0.72
Coffee w/milk	0.00	0.01	0.78	-0.03
Maté	0.07	0.55	-0.24	0.04
Alcohol	-0.14	0.45	-0.01	-0.11
Variance (%)	14.2	9.5	8.5	8.0
Nº of zeros	5	5	10	8
Nºof high loadings	5	4	2	5

Table 3. Energy-adjusted Pearson correlation coefficients between food patterns and selected variables.

Variable	Cheese & rice	Western	Fatty	Healthy
Age	-0.05	-0.21	0.13	0.09
Education	0.06	-0.06	-0.06	0.07
Body mass index	0.01	-0.07	-0.04	0.06
Smoking	-0.05	0.23	-0.00	-0.03
Total energy	0.46	0.51	0.48	0.28
Saturated fat	-0.04	0.30	0.13	-0.16
MUFA	-0.11	0.43	-0.10	-0.18
PUFA	0.04	0.27	-0.10	-0.09
Cholesterol	-0.03	0.40	0.01	0.00
Carbohydrates	0.06	-0.43	-0.05	0.16
Vitamin C	-0.12	0.07	0.05	0.64
Vitamin A	0.00	-0.00	0.03	0.25
b-carotene	-0.01	0.01	-0.01	0.23
Vitamin E	0.05	-0.10	0.03	0.66
Folate	-0.08	0.06	0.19	0.59
Calcium	0.11	-0.30	0.57	0.14
Lactose	0.40	-0.20	0.28	-0.06

positive correlation with total energy intake (r=0.48) and with saturated fat, folate, lactose and calcium. The “healthy” pattern showed a strong inverse association with saturated and monounsaturated fat. This pattern showed a positive correlation with total energy intake, vitamin C, vitamin A, b-carotene, vitamin E and folate.

Odds ratios of lung cancer (all histologies) for food patterns are shown in Table 4. The “cheese and rice”

Table 4. Odds ratios of lung cancer (all histologies) for food patterns¹

Food Pattern	Cases/Controls	OR^1 95%CI	OR^2 95%CI
Cheese and rice	210/211	1.0	1.0
	261/212	1.28 0.97-1.69	1.17 0.86-1.61
	225/212	1.09 0.83-1.45	0.93 0.67-1.30
	150/211	0.69 0.51-0.93	0.55 0.37-0.80
	p-value for trend	0.008	0.001
Western	98/211	1.0	1.0
	170/212	1.70 1.23-2.34	1.38 0.96-1.99
	240/212	2.58 1.89-3.53	1.85 1.28-2.67
	338/211	3.83 2.81-5.23	2.05 1.37-3.06
	p-value for trend	<0.0001	<0.0001
Fatty foods	165/211	1.0	1.0
	159/212	1.02 0.76-1.39	1.00 0.71-1.41
	191/212	1.30 0.96-1.74	1.31 0.94-1.84
	331/211	2.20 1.66-2.92	2.33 1.65-3.28
	p-value for trend	<0.0001	<0.0001
Healthy	260/211	1.0	1.0
	203/211	0.79 0.60-1.04	0.77 0.56-1.05
	198/213	0.69 0.52-0.91	0.75 0.55-1.04
	185/211	0.64 0.48-0.85	0.74 0.53-1.04
	p-value for trend	0.001	0.06

¹Age-adjusted.

²Adjusted for age, residence, urban/rural status, education, family history of lung cancer among first-degree relative, body mass index, smoking status, years since cessation, number of cigarettes smoked per day among current smokers, age at start smoking and total energy intake.

Table 5. Odds ratios of lung cancer for food patterns by smoking status¹

Pattern	Former smokers		Current smokers
Cheese and rice	Cases/Controls	OR 95%CI	Cases/Controls	OR 95%CI
Low score	70/116	1.0 reference	140/95	1.0 reference
2	63/117	0.83 0.50-1.36	198/95	1.40 0.92-2.13
3	78/108	0.96 0.57-1.59	147/104	0.86 0.55-1.35
High score	48/103	0.41 0.22-0.76	102/108	0.59 0.35-0.98
Heterogeneity 0.10	p-value for trend	0.02		0.008
Western	Cases/Controls	OR 95%CI	Cases/Controls	OR 95%CI
Low score	40/144	1.0 reference	58/67	1.0 reference
2	58/115	1.49 0.87-2.54	112/97	1.20 0.71-2.04
3	76/105	1.98 1.14-3.44	164/107	1.60 0.96-2.67
High score	85/80	2.19 1.17-4.08	253/131	1.83 1.05-3.17
Heterogeneity 0.52	p-value for trend	0.009		0.01
Fatty	Cases/Controls	OR 95%CI	Cases/Controls	OR 95%CI
Low score	43/97	1.0 reference	122/114	1.0 reference
2	45/106	0.95 0.53-1.69	114/106	1.00 0.65-1.54
3	59/114	1.01 0.58-1.75	132/98	1.46 0.94-2.26
High score	112/127	1.36 0.79-2.34	219/84	3.36 2.12-5.31
Heterogeneity 0.51	p-value for trend	0.18		<0.0001
Healthy	Cases/Controls	OR 95%CI	Cases/Controls	OR 95%CI
Low score	62/83	1.0 reference	198/128	1.0 reference
2	62/110	0.58 0.34-0.99	141/101	0.87 0.58-1.29
3	65/119	0.57 0.33-0.97	133/94	0.86 0.57-1.31
High score	70/132	0.48 0.28-0.83	115/79	0.91 0.58-1.42
Heterogeneity 0.06	p-value for trend	0.02		0.59

¹Adjusted for age, residence, urban/rural status, education, family history of lung cancer among first-degree relatives, body mass index, cigarettes per day, years since quit and total energy intake.

pattern was inversely associated with lung cancer risk. (OR 0.55, 95 % CI 0.37-0.80, p-value for trend=0.001). The “western” pattern was strongly and directly associated with lung cancer risk. The OR of the higher category was associated with a strong increase in risk (OR 2.05, 95 % CI 1.37-3.06, p-value for trend<0.0001). Also the “fatty” factor was directly associated with lung cancer risk (OR 2.33, 95 % CI 1.65-3.28, p-value for trend<0.0001). The “healthy“ pattern was inversely associated with lung cancer risk (OR for the high score 0.74, 95 % CI 0.53-1.04, p-value for trend=0.06).

Odds ratios of lung cancer for food patterns by smoking status are shown in Table 5. The “cheese and rice” pattern displayed a significant reduction in risk among former smokers and current smokers (OR for former smokers 0.41, 95 % CI 0.22-0.76, p-value for trend=0.02). Also the “western” pattern was positively associated with lung cancer risk in both strata of smokers, although the effect was more impressive among former smokers (OR 2.19, 1.17-4.08, p-value for trend=0.009). The “fatty” pattern was much more directly associated among current smokers (OR 3.36, 95 % CI 2.12-5.31, p-value for trend<0.0001). Finally, the “healthy” pattern showed a strong reduction in risk among former smokers, but not among current users of cigarettes (OR for former smokers 0.48, 95 % CI 0.28-0.83, p-value for trend=0.02). The p-value for heterogeneity was close to significance (p-value=0.06).

The effect of the four patterns in different cell types is shown in Table 6. The “cheese and rice” pattern displayed an inverse association with squamous cell carcinoma (p-value for trend=0.001), whereas small cell carcinoma and adenocarcinoma of the lung were not associated with this factor. Large cell carcinoma and other lung cancers were moderately protective (p-value for trend=0.03). On the other hand, the “western” pattern displayed a significant increase in risk for squamous cell carcinoma and for adenocarcinoma of the lung. The “fatty” pattern was positively associated with all histologies. The highest risky effect was observed among squamous cell and adenocarcinoma of the lung. The “healthy” pattern was significantly protective for large cell carcinoma and othe types of lung cancer, whereas the remaining cell types were not associated with risk.

IV. Discussion

According to our results, principal components analysis retained four patterns which were labeled as follows: “cheese and rice”, “western”, “fatty” and “healthy”. Whereas the “western” and the “fatty” factors

Table 6. Odds ratios of different histologies of lung cancer for food patterns¹

Factor 1 (Cheese and rice)
	Score 2	Score 3	Score 4
Cell type	OR 95%CI	OR 95%CI	OR 95%CI	p-value trend
Squamous cell	1.06 0.71-1.57	0.87 0.57-1.33	0.41 0.25-0.69	0.001
Small cell	1.50 0.82-2.78	0.78 0.39-1.56	0.80 0.38-1.68	0.20
Adenocarcinoma	1.06 0.67-1.68	1.13 0.71-1.81	0.73 0.42-1.27	0.32
Other types	1.44 0.92-2.24	0.99 0.61-1.60	0.57 0.32-1.02	0.03
Factor 2 (Western)
	Score 2	Score 3	Score 4
Cell type	OR 95%CI	OR 95%CI	OR 95%CI	p-value trend
Squamous cell	1.73 1.04-2.88	2.31 1.39-3.84	2.35 1.36-4.08	0.003
Small cell	1.09 0.48-2.47	1.70 0.78-3.73	1.45 0.62-3.37	0.28
Adenocarcinoma	1.46 0.83-2.58	1.68 0.94-2.98	2.78 1.53-5.05	<0.0001
Other types	1.04 0.60-1.80	1.60 0.94-2.72	1.49 0.82-2.71	0.10
Factor 3 (Fatty)
	Score 2	Score 3	Score 4
Cell type	OR 95%CI	OR 95%CI	OR 95%CI	p-value trend
Squamous cell	0.93 0.59-1.48	1.21 0.77-1.90	2.26 1.45-3.53	<0.0001
Small cell	0.69 0.34-1.40	1.22 0.64-2.33	1.81 0.95-3.44	0.02
Adenocarcinoma	1.14 0.68-1.92	1.59 0.96-2.63	2.60 1.58-4.28	<0.0001
Other types	1.14 0.70-1.85	1.33 0.81-2.18	2.17 1.33-3.55	0.001
Factor 4 (Healthy)
	Score 2	Score 3	Score 4
Cell type	OR 95%CI	OR 95%CI	OR 95%CI	p-value trend
Squamous cell	1.01 0.67-1.51	1.07 0.71-1.61	0.82 0.53-1.29	0.43
Small cell	0.74 0.40-1.36	0.76 0.41-1.40	0.89 0.47-1.68	0.68
Adenocarcinoma	0.82 0.52-1.28	0.77 0.49-1.21	0.90 0.56-1.44	0.82
Other types	0.63 0.41-0.97	0.54 0.34-0.85	0.54 0.33-0.88	0.003

¹Adjusted for age, residence, urban/rural status, education, family history of lung cancer among first-degree relative, body mass index, smoking status, years since cessation, number of cigarretes smoked per day among current smokers, age at start smoking and total energy intake.

²Pattern 1: squamous cell vs adenocarcinoma=0.05/

³Pattern 4: squamous cell vs other types=0.04

⁴Pattern 4:adenocarcinoma vs other types=0.04

were positively associated with lung cancer risk, both the “cheese and risk” and “healthy” patterns were somehow protective. The “cheese and rice” pattern showed high loadings for white meat, processed meat, cheese, butter, desserts and rice. Taking into account that six different food groups contributed in an important loading, we considered the possibility of labeling this pattern as a “mixed” pattern. Finally, since both cheese and rice were protective in lung cancer we decided to label this pattern as “cheese and rice” factor. This pattern was not identified in previous studies on cancer and factor analysis. The “cheese and rice” pattern showed high correlations with processed meat, cheese, butter, desserts and rice. All these variables have been traditionally directly associated with risk of lung cancer (Ziegler et al, 1996; World Cancer Research Fund, 1997; Slattery et al, 1998; Terry et al, 2001a). Nevertheless, some studies reported a protective effect of cheese in non-smokers (Brennan et al, 2000). In our study, the “cheese and rice” food pattern has been inversely associated with lung cancer risk. In the Dutch prospective study the “sweet” pattern has some common characteristics (Balder et al, 2005). Further studies are needed in order to clarify the possible mechanisms by which these foods could be protective in lung cancer.

On the contrary, the “western” pattern, which has high loadings for red meat and alcohol, replicated similar patterns in other studies which employed factor analysis (Slattery et al, 1998; Terry et al, 2001a). Red meat, processed meat, maté drinking and alcohol drinking have been considered as risk factors for the lung mucosa or bronchial lining (Harris et al, 1996; Ferguson and Harris, 1998; Slattery et al, 1998; Tsai et al, 2003). Our “western” pattern is close to the “pork, processed meat, potatoes” pattern of Balder et al, (2005). Red meat consumption has been considered as risk factors for two possible mechanisms: 1) through its high content of saturated fat and/or 2) through its high amount of HCA’s resulting from the cooking method of this meat (Slattery et al, 1998; Tsai et al, 2003). Since both chemicals have been suggested as carcinogens, its activity together with the carcinogenic chemicals present in tobacco smoke (Sinha et al, 1998), suggest that heavy smokers and high consumers of well-done red meat are, possibly, at high risk of developing lung cancer. To our knowledge, our study on maté drinking and lung cancer is the only report which suggested a possible direct effect of this herbal tea in lung cancer (De Stefani et al, 1997b). For this reason, it is impossible to be sure that the results are not due to residual confounding from tobacco or due to other confounders (IARC, 1980). Concerning alcohol consumption, a recent meta-analysis suggested that alcohol could be a risk factor only on at very large doses (Korte et al, 2002). Therefore, although our findings regarding the “western” pattern are somehow reassuring in replicating previous findings, it is necessary to be cautious since exploratory factor analysis is essentially subjective both in the selection of the variables and in its interpretation (De Stefani et al, 2005; Fung et al, 2005).

The “fatty” factor was associated with a significantly increase in risk of lung cancer and this applies to all the cell types. The only variables with high loadings for this pattern were whole milk and coffee with milk. It is noteworthy that Uruguayan population consumed high amounts of whole milk (rich in fat) and coffee with milk, being pure coffee an unfrequent habit. Therefore, this pattern is rich in foods which contains high amounts of fat, mainly saturated fat. Thus, our findings support the role of saturated fat in lung carcinogenesis, replicating previous findings (Terry et al, 2001a; Balder et al, 2005).

The “healthy” factor displayed high loadings of white meat, fresh vegetables, cooked vegetables, citrus fruits and non-citrus fruits. rich in carotenoids, winter squash, potato, and white rice. These foods are considered as protective. The “healthy” pattern was particulary protective among former smokers and among patients with large cell carcinoma. Vegetables which contained b-carotene, like carrot and sweet potato, were inversely associated with lung cancer risk. A recent study reported that sweet potato is a rich source of cis-b-Cryptoxanthin, b-carotene, and 9-cis b-Carotene (Benamotz and Fishler, 1998). These stereoisomeric forms of carotenoids deserve more attention in the field of protection of lung cancer (Benamotz and Fishler, 1998). Potatoes have been considered as inversely associated with some cancers (Cambie and Ferguson, 2003). It is possible that cell walls from potatoes adsorb heterocyclic amines (HCA) in the intestinal lumen (Harris et al, 1996; Ferguson and Harris, 1998). It is well-known that HCA could be carcinogens for lung tissue (Sinha et al, 1998; Weisburger, 2002). As it was suggested by Weisburger, (2002), HCA could act as initiators and fats as promoters in lung carcinogenesis. Our “healthy” pattern was different with the patterns identified by Balder et al, (2005), although the “cooked vegetables” pattern presented some charcteristics in common. Whereas the “salad vegetables” of the Netherlands Cohort Study was essentially protective, the “healthy” pattern of Tsai et al. was no longer significant after controlling for tobacco smoking (Tsai et al, 2003; Balder et al, 2005). In traditional studies (both prospective and case-control), the role of vegetables and fruits were inversely associated with lung cancer risk (Ziegler et al, 1996; World Cancer Research Fund, 1997; IARC, 2003). However, the study of Feskanich et al, (2000) reported results close to the null for fruit and vegetables intake. Perhaps residual confounding from smoking could attenuate the inverse association with plant foods. In this sense we adjusted the estimates for fruit and vegetables consumption using a index which included smoking status, number of cigarettes per day among current smokers, years since quit and age at start smoking.

The use of factor analysis raises some concerns. The first problem is related with the construction and analysis of the FFQ. In other words, the FFQ could be not adequate for the purposes of the study. We devoted a considerable time in order to constructing our FFQ and we consider that it is representative of the usual diet among Uruguayans. An important decision in factor analysis is the choice of the number of factors to be retained (McCann et al, 2001; Terry et al, 2001a; Handa and Kreiger, 2002). In order to retain the number of factors which explains more of the variance than a single variable, we set an eigenvalue of 1.25. In spite of some controversial findings, the factors retained by exploratory factor analysis, have biological sense. More precisely, the “western” pattern, the “healthy” pattern and the “fatty” factor are plausible, since all three are in accordance with the Uruguayan diet. Also they could be replicable in other datasets (Harman, 1976; Pearson, 1901; Kline, 2002). The retained factors explained 45 % of the total variance of the model. This percentage is rather good when compared with other studies which employed factor analysis in cancer epidemiology (De Stefani et al, 1999; McCann et al, 2001; Palli et al, 2001; Terry et al, 2001b; Handa and Kreiger, 2002; Masaki et al, 2003; Sieri et al, 2004; De Stefani et al, 2005; Fung et al, 2005). Also the communalities were reasonably high.

As other case-control studies, our study has limitations. Aside from the usual problems of selection and classification biases, our study could suffer from recall bias. At difference with prospective studies, case-control studies has this common problem. Although the interviews were conducted blindly by the two social workers, it is impossible to discard some degree of faulty recall. It is true that our participants are mostly unawere of the role of diet in lung cancer or of other diseases, but it is impossible to discard that some patients were more healthy than others. We have excluded proxy interviews, but sometimes proxies are more exact then the participants. Our study has strengths. We excluded other hospitals which could be sources of patients with lung cancer since they not were a source of controls. Although we do not match by hospital, the proportion of cases and controls were rather similar. The statistical power of the study allowed to detect as significant an OR of 1.3. Another strength is the high response rate both for cases and controls.

V. Conclusions

In summary, we conducted a case-control on foods and risk of lung cancer. To do so, we used exploratory factor analysis (principal components) and we retained four factors with an eigenvalue of 1.25. These patterns explained 40.2 % of the total variance. Whereas the “cheese and rice” and “healthy” patterns were significantly protective, the “western” and the “fatty” patterns were directly associated with lung cancer risk. As important variables, white meat, tomatoes, green leafy vegetables, onions, carotenoid vegetables, winter squash, potatoes, citrus fruits and white rice deserve attention in further studies. Also, some foods rich in fat could be also protective, like cheese. On the other hand, red meat, processed meat, maté consumption, alcohol drinking, whole milk and coffee with milk appears to be associated with lung cancer risk. It is our opinion that is too soon to suggest preventive messages. Perhaps, the only preventive message is to quit smoking.

Acknowledgements

This research was supported by the International Agency for Research on Cancer.

Appendix I

Results of the reproducibility test

Table 1. Means and Pearson correlation coefficients for food groups

Means	Correlations^2
Food groups^1	First interview	Second interview	rho
Red meat	355.4	369.0	0.77
White meat	86.2	93.7	0.60
Processed meat	212.1	258.6	0.55
Total meat	660.2	735.7	0.67
Dairy foods	606.1	597.8	0.64
Eggs	122.8	130.9	0.45
Desserts	165.7	177.8	0.50
Grains	1035.9	1072.5	0.69
Raw vegetables	254.4	271.6	0.64
Cooked vegetables	603.6	662.9	0.39
Total vegetables	858.1	934.5	0.46
Citrus fruits	132.6	111.9	0.46
Other fruits	323.1	309.3	0.51
Total fruits	455.8	421.2	0.54
All tubers	337.4	332.5	0.66
Legumes	37.9	38.1	0.70
Total vegetables & fruits	1313.8	1355.7	0.59
Coffee^3	124.7	110.2	0.33
Tea^3	65.9	83.8	0.51
Alcohol^4	80.0	81.1	0.70

¹Servings/year

²Energy-adjusted Pearson correlations

³Mililiters/day

⁴Mililiters/ethanol day

Table 2. Means and Pearson correlation coefficients for nutrients

Means	Correlations
Nutrient	First interview	Second interview	Crude rho	Energy-adjusted rho
Energy	2015	2098	0.78	-
Protein	102.8	108.4	0.61	0.43
Carbohydrates	263.6	264.0	0.75	0.69
Total fat	116.2	123.3	0.63	0.62
Saturated fat	46.4	48.7	0.67	0.69
MUFA^1	45.4	48.4	0.66	0.67
PUFA^2	11.9	12.9	0.55	0.49
Linoleic acid	10.4	11.3	0.54	0.48
Linolenic acid	1.3	1.3	0.65	0.59
Cholesterol	533	655	0.53	0.47
Vitamin A	11918	13082	0.49	0.50
b-Carotene	5596	5782	0.50	0.50
Total carotenoids	10356	10641	0.45	0.49
Vitamin C	123	132	0.51	0.46
Vitamin E	3.8	4.0	0.50	0.39
Vitamin B6	1.5	1.6	0.60	0.44
Vitamin B12	6.5	6.8	0.68	0.71
Thiamine	1.4	1.6	0.67	0.45
Riboflavin	1.9	1.9	0.71	0.61
Folate	194.3	211.3	0.56	0.49
Fiber	22.1	22.6	0.66	0.61
Calcium	646	662	0.56	0.58
Iron	16.7	17.7	0.61	0.46
Sodium	954	1042	0.57	0.51

¹Monounsaturated fat

²Polyunsaturated fat

References

Balder HF, Goldbohm RA, van den Brandt PA (2005) Dietary patterns associated with male lung cancer risk in the Netherlands Cohort Study. Cancer Epidemiol Biomarkers Prev 14, 483-90.

Benamotz A, Fishler R (1998) Analysis of carotenoids with emphasis on 9-cis b-carotene in vegetables and fruits commonly consumed in Israel. Food Chem 62, 515-520.

Brennan P, Fortes C, Butler J, Agudo A, Benhamou S, Darby S, Gerken M, Jokel KH, Kreuzer M, Mallone S, Nyberg F, Pohlabeln H, Ferro G and Boffetta P (2000) A multicenter case-control study of diet and lung cancer among non-smokers. Cancer Causes Control 11, 49-58.

Breslow NE and Day NE (1980) Statistical methods in cancer research. The analysis of case-control analysis. Volume 1IARC scientific publication N° 32 IARC Lyon.

Cambie RC, Ferguson LR (2003) Potential functional foods in the traditional Maori diet. Mutat Res 523-524,109-17.

Carámbula JA, Ronco AL, Niedworok NT (1995) Programa Nacional de Prevención del Cáncer de Pulmón: Investigación sociolaboral del tabaquismo en el Uruguay urbano. Dirección General de Salud, Ministerio de Salud Pública, Montevideo, Uruguay.

De Stefani E, Boffetta P, Deneo-Pellegrini H, Mendilaharsu M, Carzoglio JC, Ronco A and Olivera L (1999) Dietary antioxidants and lung cancer risk:a case-control study in Uruguay. Nutr Cancer 34, 100-110.

De Stefani E, Boffetta P, Ronco AL, Correa P, Oreggia F, Deneo-Pellegrini H, Mendilaharsu M and Leiva J (2005) Dietary patterns and risk of cancer of the oral cavity and pharynx in Uruguay. Nutr Cancer 51, 132-139.

De Stefani E, Correa P, Boffetta P, Deneo-Pellegrini H, Ronco AL and Mendilaharsu M (2004) Dietary patterns and risk of gastric cancer: a case-control study in Uruguay. Gastric Cancer 7, 211-220.

De Stefani E, Deneo-Pellegrini H, Mendilaharsu M, Carzoglio JC and Ronco A (1997a) Dietary fat and lung cancer: a case-control study in Uruguay. Cancer Causes Control 8, 913-921.

De Stefani E, Fontham ETH, Chen V, Correa P, Deneo-Pellegrini H, Ronco A and Mendilaharsu M (1997b) Fatty foods and the risk of lung cancer:A case-control study from Uruguay. Int J Cancer 71, 760-766.

Deneo-Pellegrini H, De Stefani E, Mendilaharsu M, Ronco A and Carzoglio J (1996) Meat consumption and lung cancer risk. Lung Cancer 14, 195-205.

Engeland A, Haldorsen T, Andersen A, Tretli S (1996) The impact of smoking habits on lung cancer risk: 28 years' observation of 26,000 Norwegian men and women. Cancer Causes Control 7, 366-76.

Ferguson LR, Harris PJ (1998) Suberized plant cell walls suppress formation of heterocyclic amine-induced aberrant crypts in a rat model. Chem Biol Interact 114, 191-209.

Feskanich D, Ziegler RG, Michaud DS, Giovannucci EL, Speizer FE, Willett WC and Colditz GA (2000) Prospective study of fruit and vegetable consumption and risk of lung cancer among men and women. J Natl Cancer Inst 92, 1812-1823.

Fung TT, Hu FB, Holmes MD, Rosner BA, Hunter DJ, Colditz GA and Willett WC (2005) Dietary patterns and the risk of postmenopausal breast cancer. Int J Cancer 116, 116-121.

Handa K, Kreiger N (2002) Diet patterns and the risk of renal cell carcinoma. Public Health Nutr 5, 757-67.

Harman HH (1976) Modern factor analysis.Third Edition Revised. The University of Chicago Press Ltd., London.

Harris PJ, Triggs CM, Roberton AM, Watson ME, Ferguson LR (1996) The adsorption of heterocyclic aromatic amines by model dietary fibres with contrasting compositions. Chem Biol Interact 100, 13-25.

Hueper WC (1966) Occupational and environmental cancers of the respiratory system. Springer-Verlag, New York.

IARC (2003) Handbooks of Cancer Prevention. Fruits and Vegetables. IARC Press Lyon. Volume pp. 81-84.

IARC (1980) Monographs on the evaluation of the carcinogenic risk of chemicals to humans. Tobacco smoke and involuntary smoking. Volume 83.

Kim JO and Mueller CW (1978) Factor analysis. Statistical methods and practical issues. Sage University Paper. California. USA.

Kline P (2002) An easy guide to factor analysis. Routledge, New York.

Korte JE, Brennan P, Henley SJ, Boffetta P (2002) Dose-specific meta-analysis and sensitivity analysis of the relation between alcohol consumption and lung cancer risk. Am J Epidemiol 155, 496-506.

Markaki I, Linos D, Linos A (2003) The influence of dietary patterns on the development of thyroid cancer. Eur J Cancer 39, 1912-9

Masaki M, Sugimori H, Nakamura K, Tadera M (2003) Dietary patterns and stomach cancer among middle-aged male workers in Tokyo. Asian Pac J Cancer Prev 4, 61-6.

McCann SE, Marshall JR, Brasure JR, Graham S, Freudenheim JL (2001) Analysis of patterns of food intake in nutritional epidemiology: food classification in principal components analysis and the subsequent impact on estimates for endometrial cancer. Public Health Nutr 4, 989-97.

Palli D, Russo A and Decarli A (2001) Dietary patterns, nutrient intake and gastric cancer in a high-risk area of Italy. Cancer Causes Control 12, 163-172.

Parkin DM, Whelan SL, Ferlay J, Teppo L and Thomas DB (2002) (eds): Cancer Incidence in Five Continents. Vol VIII. IARC Scientific Publications n° 155, Lyon, IARC.

Pearson K (1901) On lines and planes of closest fit to systems of points in space. Phil Mag 6, 559-572.

Potter JD, Sellers TA, Folsom AR, McGovern PG (1992) Alcohol, beer, and lung cancer in postmenopausal women. The Iowa Women's Health Study. Ann Epidemiol 2, 587-95.

Randall E, Marshall JR, Brasure J and Graham S (1992) Dietary patterns and colon cancer in western New York. Nutr Cancer 18, 265-276.

Rothman KJ and Greenland S (1998) Modern Epidemiology.Second Edition. Lippincott-Raven Pub., Philadelphia.

Sieri S, Krogh V, Pala V, Muti P, Micheli A, Evangelista A, Tagliablue G and Berrino F (2004) Dietary patterns and risk of breast cancer in the ORDET cohort. Cancer Epidemiol Biomarkers Prev 13, 567-572.

Sinha R, Kulldorff M,Curtin J, Brown CC, Alavanja MCR and Swanson CA (1998) Fried, well-done red meat and risk of lung cancer in women (United States). Cancer Causes Control 9, 621-630.

Slattery ML, Boucher KM, Caan BJ, Potter JD and Ma Y (1998) Eating patterns and risk of colon cancer. Am J Epidemiol 148, 4-16.

Spearman C (1904) General intelligence, objectively determined and measured. Am J Psychol 15, 201-293.

STATA Stata reference guide (1999) Release Stata Press. College Station Texas.

Terry P, Hu FB, Hansen H and Wolk A (2001a) Prospective study of major dietary patterns and colorectal cancer risk in women. Am J Epidemiol 154, 1143-1149.

Terry P, Suzuki R, Hu FB and Wolk A (2001b) A prospective study of major dietary patterns of the risk of breast cancer. Cancer Epidemiol Biomarkers Prev 10, 1281-1285.

Travis WD, Colby TV, Corrin B and Shimosato Y (1999) Histological typing of lung and pleural tumors. Berlin, Springer.

Tsai YY, McGlynn KA, Hu Y, Cassidy AB, Arnold J, Engstrom PF, Buetow KH (2003) Genetic susceptibility and dietary patterns in lung cancer. Lung Cancer 41, 269-81. Erratum in 2004, 44, 271.

Weisburger JH (2002) Comments on the history and importance of aromatic and heterocyclic amines in public health. Mutat Res 506-507, 9-20.

World Cancer Research Fund (1997) Food, nutrition and the prevention of cancer: a global perspective. American Institute for Cancer Research .Washington DC.

Ziegler RG, Mayne ST and Swanson CA (1996) Nutrition and lung cancer. Cancer Causes Control 7, 157-177.