Substitution of Dietary Protein for Carbohydrate: Associations of Disease and Mortality in a Prospective Study of Postmenopausal Women LE Kelemen, LH Kushi, DR Jacobs Jr., JR Cerhan Mayo Clinic College of Medicine, Rochester, MN University of Minnesota, Minneapolis, MN Kaiser Permanente, Oakland, CA

Background Popular high protein (HP) diets extol benefits for weight loss –Often do not discriminate among protein types Effect of protein & protein type on long term health outcomes not widely studied

Objectives Using multivariable nutrient density models: 1)To estimate the effect of an isoenergetic substitution of total protein for total carbohydrate with cancer incidence and mortality from cancer, CHD and all causes in the IWHS 2)To estimate the effect of an isoenergetic substitution of one type of protein for another type of protein with these outcomes

In 1986: 99,826 Iowa women aged yrs (now 71-86) Identified from randomly selected driver’s licenses Mailed questionnaire –diet (FFQ) –self-reported lifestyle, medical & reproductive history 41,836 (41.9%) enrolled Study Population

Dietary Assessment Semi-quantitative Harvard FFQ Validation study, 1988, 44 Iowa women –average of five 24-hour recalls over 2 months: »r = 0.16 ( protein) »r = 0.45 (carbohydrate) »r = (fats) Reproducibility (2.5 yrs) »r = 0.59 (protein) »r = 0.53 (carbohydrate) »r = (fats)

Follow-Up Questionnaires mailed in 1987, 1989, 1992 and 1997 Incident cancers identified by linkage to Iowa SEER cancer registry Deceased non-respondents & cause of death identified by linkage to National Death Index 15 years follow-up

Excluded –Premenopausal women (n=569) –Prior history of cancer (n=3,881) –Known heart disease (n=5,116) –Known diabetes (n=2,675) –Diet  30 blanks on FFQ total energy (kcal/d)  600 or  5,000 (n=3,096) 29, 017 eligible women Eligibility Criteria

Data Analysis Dietary exposures –Macronutrients expressed as nutrient densities (i.e. % of energy from protein, carbohydrate and fats) –Micronutrient covariates were energy-adjusted (Willett & Stampfer 1986) –Categorized into quintiles RR (95% CI) estimated using Cox proportional hazards with lowest intake category as referent; age as time metric

Multivariable-adjusted nutrient density models (Willett 2nd ed 1998 p 295; Hu AJE 1999; Willett AJCN 1997) –Estimate associations from an increase in the % energy from protein intake –By forcing total energy and other intake (i.e., dietary fats) to be constant, and by excluding carbohydrate from the model, modeling the effects of an increase in protein intake, by definition, statistically results in a decrease in carbohydrate intake –Thus, the effect estimates of protein assume a substitution interpretation –The % of energy from protein that is “substituted” for carbohydrate is the difference between the median intake in the highest and lowest quintiles –Models also adjusted for other risk factors Data Analysis

Covariates Known/suspected confounders & risk factors: –Total energy –Fats (saturated, poly-, mono- & trans) (all quintiles & expressed as % of energy) –Total fiber, dietary cholesterol, dietary methionine (all quintiles & energy-adjusted) –Alcohol (≤14 vs > 14 g/d) –Smoking (never, former, current) –Activity level (active vs not active) –BMI (5 levels) –History of HTN –PM hormone use –Education (≤ high school vs > high school) –Family history of cancer –Multivitamin use –Vitamin E supplement use

Results 475,755 person-years Outcomes: –4,843 incident cancers –739 CHD deaths –1,676 cancer deaths –3,978 deaths from all causes

Table 1 Distribution of baseline characteristics by quintiles of total protein among 29,017 Iowa women, 1986 Quintiles of total protein (% of total energy) Subject characteristics1 (14.1) 2 (16.3) 3 (17.8) 4 (19.4) 5 (22.0) Age, y76 75 Education > high school, % Physical activity, active % Current smokers, %1914 Alcohol >14 g/d, % Body mass index, kg/m Vitamin E supplement use, %

Table 1 cont. Nutrient intakes1 (14.1) 2 (16.3) 3 (17.8) 4 (19.4) 5 (22.0) Carbohydrates, % energy Total fat, % energy Saturated fat, % energy Polyunsat. fat, % energy Monounsat. fat, % energy trans fat, % of energy Cholesterol, mg/d Total fiber, g/d Methionine, g/d Quintiles of total protein (% of total energy)

Table 1 cont. Food intakes, servings / 1,000 kcals 1 (14.1) 2 (16.3) 3 (17.8) 4 (19.4) 5 (22.0) Processed & red meat A Chicken & poultry Fish & seafood Dairy products B Eggs Nuts, tofu and legumes Whole grains C Refined grains D Sweets and desserts Fruits and vegetables Quintiles of total protein (% of total energy) A composite of beef, pork, processed meat B composite of milk, cream, ice-cream, yogurt, cheese C composite of dark bread, brown rice, oatmeal, whole grain cereal, bran, wheat germ & other grains (bulgar, kasha, couscous) D composite of rice, pasta, potatoes, refined cold breakfast cereal, muffins, snack foods, sweetened sodas, pizza, chocolate, cakes, cookies

Table 2 RR (95% CI) for CHD mortality by quintiles of total protein intake (% energy) substituted for isoenergetic amount of carbohydrate, IWHS 1986 to 2001 Quintiles of intakeP (95% CI)trend Total protein Median (% energy) Multivariable RR* (0.39, 1.79)0.62 Animal protein Median (% energy) Multivariable RR (0.42, 1.86)0.29 Vegetable protein Median (% energy) Multivariable RR (0.49, 0.99)0.02 *adjusted for dietary fats, total energy plus other covariates

Table 3 RR (95% CI) for cancer incidence by quintiles of total protein intake (% energy) substituted for isoenergetic amount of carbohydrate, IWHS Quintiles of intakeP (95% CI)trend Total protein Median (% energy) Multivariable RR* (0.92, 1.67)0.18 Animal protein Median (% energy) Multivariable RR (0.76, 1.37)0.95 Vegetable protein Median (% energy) Multivariable RR (0.87, 1.14)0.99 *adjusted for dietary fats, total energy plus other covariates

Table 4 RR (95% CI) for cancer mortality by quintiles of total protein intake (% energy) substituted for isoenergetic amount of carbohydrate, IWHS Quintiles of intakeP (95% CI)trend Total protein Median (% energy) Multivariable RR* (0.64, 1.79)0.81 Animal protein Median (% energy) Multivariable RR (0.47, 1.27)0.31 Vegetable protein Median (% energy) Multivariable RR (0.83, 1.32)0.43 *adjusted for dietary fats, total energy plus other covariates

Table 5 RR (95% CI) for all cause mortality by quintiles of total protein intake (% energy) substituted for isoenergetic amount of carbohydrate, IWHS Quintiles of intakeP (95% CI)trend Total protein Median (% energy) Multivariable RR* (0.71, 1.38)0.67 Animal protein Median (% energy) Multivariable RR (0.59, 1.13)0.24 Vegetable protein Median (% energy) Multivariable RR (0.82, 1.10)0.74 *adjusted for dietary fats, total energy plus other covariates

Table 6 RR (95% CI) of vegetable protein intake (% of energy) substituted for isoenergetic amount of animal protein for different outcomes Quintiles of vegetable protein intakeP (95% CI)trend Median (% energy) CHD mortality Multivariable RR* (0.51, 0.98)0.02 Cancer incidence Multivariable RR (0.87, 1.13)0.92 Cancer mortality Multivariable RR (0.89, 1.38)0.29 All cause mortality Multivariable RR (0.86, 1.14)0.82 *adjusted for carbohydrate, dietary fats, total energy, plus other covariates

Table 7 Multivariable RR* for protein foods substituted for an isoenergetic amount of carbohydrate foods (svgs/1000 kcals) for different outcomes Quintiles of intakePΔ Svg/1,000 kcals Q5:Q1 Svgs/1,000 kcals12345 (95% CI)trend CHD mortality Legumes (0.65, 1.07) Dairy (1.07, 1.87) Red meats (1.06, 1.94) Cancer mortality Legumes (1.04, 1.46) Dairy (0.80, 1.17) Red meats (0.85, 1.27) All cause mortality Legumes (0.99, 1.23) Dairy (0.97, 1.24) Red meats (1.02, 1.32) *adjusted for dietary fats, total energy, other covariates & quintiles of svgs/1000kcals: fruits & veg, eggs, poultry, fish, legumes, dairy, red meats

Summary Similar ↓ in risk of CHD mortality when vegetable protein substituted for carbohydrate or animal protein –suggests animal protein & carbohydrate may have similar potentially adverse effects on CHD mortality Animal protein not associated with any outcome –↑ risk of CHD mortality for red/processed meat servings (RR=1.44) and dairy servings (RR=1.41) when substituted for carbohydrate foods –Modest risk of red/processed meat servings with all cause mortality (RR=1.16) Modest risk of legume servings with cancer mortality (RR=1.23) but not with cancer incidence No associations with cancer incidence

Strengths & Limitations Strengths –Prospective –Large # of events –Adjust for large # of covariates Limitations –Baseline diet only –No blood samples –Food substitution analyses: measuring non- protein components? Red meat & CHD – consistent with others’ findings (Snowdon 1984, Hu 1999, Liu 2004)

Conclusions Dietary protein from animal and vegetable sources appear to be differentially associated with mortality from CHD & all causes when substituted in the diet Long-term adherence to popular HP diets, without discrimination toward protein source, may have potentially adverse health consequences

Appendix - Protein Food Groupings Legumes/nuts/tofu –composite of tofu, dried beans, nuts and peanut butter Dairy –composite of milk, cream, ice-cream, yogurt and cheese Eggs Red meats –composite of beef, pork and processed meat Poultry –composite of chicken and turkey Fish –composite of fresh fish, canned fish and seafood Fruits & Vegetables –Including juices, excluding potatoes Carbohydrate foods = referent –composite of refined carbohydrates (rice, pasta, potatoes, refined cold breakfast cereal, muffins, snack foods, sweetened sodas, pizza, chocolate, cakes, cookies), and –whole grain carbohydrates (dark bread, brown rice, oatmeal, whole grain breakfast cereal, bran, wheat germ and other grains such as bulgar, kasha and couscous)