Low-carbohydrate, high-protein diet score and risk of incident cancer; a prospective cohort study

In recent years, low-carbohydrate diets have emerged as a controversial and popular means of achieving weight loss and controlling diabetes. In Sweden, extensive positive media support for dietary carbohydrate restriction has occurred over the past 5–7 years [1]. During the same time period, in northern Sweden, a complete reversal of previous reductions in fat intake and cholesterol levels has been reported in the general population [2, 3]. Discerning the potential long-term health effects of carbohydrate restriction, not only of stringent low-carbohydrate diets but also of more modest carbohydrate reduction, is thus an important challenge in nutrition research today.

For weight loss, low-carbohydrate diets, both extremely or more modestly reduced in carbohydrate (e.g. E% carbohydrates/protein/fat = 9/28/63 [4], and 44/18/38 [5], respectively) have been found to be at least as effective as traditional low-calorie/low-fat diets over a period of up to two years [5‐7]. The results of randomized trials have also tended to support improved metabolic parameters and blood lipids [8‐11], but elevated markers of stress and inflammation [11‐13] in subjects following a low-carbohydrate diet. These alterations might influence the risk of major chronic diseases such as cardiovascular disease and cancer [11, 14]. However, from a long-term perspective, the effects of carbohydrate reduction of varying degrees, and consequent increased consumption of various types of protein and/or fat, for health outcomes, and cancer in specific, are largely unknown.

The results of previous epidemiological studies in general populations have suggested positive or null associations between low-carbohydrate diet scores, particularly scores representing diets higher in foods of animal origin, and all-cause, cardiovascular, and cancer mortality [15‐19]. Prospective studies of cardiovascular disease incidence have reported either an increased risk [20], or reduced risk for plant-based [21], carbohydrate-restricted diets. The only previous prospective study to address overall cancer incidence found null associations for both animal- and plant-based low-carbohydrate diets [22]. An increased risk of incident breast cancer has been observed for a dietary pattern characterized by a low intake of bread and fruit juice and a high intake of processed meat, fish, butter, other animal fats, and margarine [23]. In contrast, a plant-based, low-carbohydrate diet has been related to a reduced risk of estrogen-receptor-negative breast cancer [24].

Given the high rates of overweight and obesity worldwide, and the widespread popularity of low-carbohydrate diets, evaluation of the long-term safety of carbohydrate restriction of varying degrees is crucial. The aim of the present study was to investigate macronutrient distribution, in particular a previously established low-carbohydrate, high-protein (LCHP) score [16‐20], in relation to the risk of incident cancer and specific types of cancer in a large, population-based cohort in northern Sweden.

Follow-up times ranged from 1 day to 17.9 years, median 9.7 years. Macronutrient intakes for the lowest to highest LCHP scores (2–20 points) ranged from median 60.9 to 38.9% of total energy intake for carbohydrate, 11.3 to 18.9% for protein, and 26.7 to 41.5% for fat. Relationships between baseline characteristics and LCHP score are presented in Table 1. High LCHP scores were associated with younger age (not apparent in medians due to sampling at 10-year age intervals) and higher BMI, prevalence of current smokers, sedentary lifestyle (women only) and alcohol intake. Lack of postsecondary education was more common in men with low LCHP scores and in women with high scores. LCHP scores were positively related to intake of animal protein, but negatively related to plant protein. For carbohydrate and fat, associations were consistent in sucrose and whole grain and saturated and unsaturated fat, respectively. Spearman correlation coefficients for LCHP score and energy-adjusted fat, saturated fat and unsaturated fat intakes were 0.51, 0.45, and 0.46, respectively.

There were no statistically significant associations between LCHP score and any cancer, with the exception of an increased risk of respiratory tract cancer for medium LCHP scores in men (multivariate HR for medium versus low LCHP scores 1.84; 95% CI 1.05-3.23; p-trend = 0.38) (Table 2). HR for high versus low LCHP scores for respiratory tract cancer were above one in both men and women, but not statistically significant.

Subgroup analyses based on metabolic risk profile, saturated fat intake, and energy reporting [19], had no material effects on results (Table 3). The only statistically significant finding was an inverse association between LCHP score and colorectal cancer risk in women with high saturated fat intakes (multivariate HR for a 1-point increase in LCHP score 0.92; 95% CI 0.87-0.98; p = 0.013; p-heterogeneity = 0.003). Constructing LCHP scores in which whole grain or sucrose replaced total carbohydrates, and in which vegetable or animal protein replaced total protein intake (data not shown), also did not differ from the main findings, except a statistically significant increased risk of colorectal cancer in men with higher LCHP scores based on vegetable protein (multivariate HR for a 1-point increase in LCHP score 1.07; 95% CI 1.01-1.14; p = 0.029; p-heterogeneity = 0.016).

In analyses restricted to the time period up to and including December 31, 2002, there was a tendency towards a positive association between high LCHP scores and overall cancer risk in both men (multivariate HR for high versus low LCHP scores 1.25; 95% CI 0.86-1.80; p-trend = 0.093) and women, (multivariate HR for high versus low LCHP scores 1.39; 95% CI 0.98-1.96; p-trend = 0.020) (Table 4). For prostate, breast and colorectal cancers no significant associations were found.

In this large population-based cohort study with a follow-up period of up to 17.9 years, a diet moderately low in carbohydrates and moderately high in protein was largely unrelated to overall and site-specific cancer incidence, regardless of the quantity and quality of fat intake.

The one previous prospective study to report results for overall cancer incidence, from the Iowa Women’s Health Study, reported inverse risk relationships for isocaloric substitution of either animal or vegetable protein for carbohydrates [22]. However, the results were attenuated to null in multivariate analyses. Associations reported for overall cancer mortality have also been null, non-statistically significant, or unstable [15, 17, 19, 22]. Taken together, the evidence to date does not support a role for moderate carbohydrate reduction in determining the overall risk of cancer.

Increasing LCHP scores were associated with an elevated risk of respiratory tract cancer in both men and women in the present study, but the relationship was not dose dependent and was only statistically significant for medium LCHP scores in men. Although these observations may be due to chance or reflect residual confounding due to smoking, they are consistent with a previous finding for lung cancer mortality [15]. Further study is therefore warranted.

There are several mechanisms by which a carbohydrate-reduced diet could influence carcinogenesis, through specific food items or components, such as red and processed meat for example [35], or through effects on energy metabolism and body composition [36‐39]. In analyses considering macronutrient quality and metabolic profile at baseline, two statistically significant results were observed, an inverse association between LCHP score and colorectal cancer risk in women with high saturated fat intakes, and an increased risk of colorectal cancer in men with higher LCHP scores based on vegetable protein. These findings do not support the hypothesis that high animal protein intake increases the risk for these cancer types. Previously, we have reported a null association between LCHP score and colorectal cancer mortality [19], and a positive association has been noted for an animal-based, low-carbohydrate score and colorectal cancer mortality [15]. The latter finding is more consistent with the current understanding of the role of diet in colorectal cancer. For example, there is convincing evidence that a high consumption of protein sources such as red and processed meat is associated with increased colorectal cancer risk [35]. Furthermore, in a controlled trial, a LCHP weight-loss diet has been observed to reduce fecal cancer-protective metabolites and increase hazardous metabolites, which could increase the risk of colon cancer [40].

The limited variability in macronutrient distribution in the study population may have prevented the detection of associations with cancer risk. In particular, the role of stricter carbohydrate restriction could not be assessed. This is an issue common to both the present and previous studies [15, 17, 19, 22]. Interindividual differences, such as gene-nutrient interactions and epigenetics, both emerging research fields [36], may also complicate the relationship between macronutrient distribution and cancer risk. Furthermore, carbohydrate restriction might have different roles in different stages of tumorigenesis, making it difficult to detect overall effects on incidence. For example, putative mechanisms for a role for carbohydrate in the progression from premalignant lesion to cancer diagnosis include metabolic reprogramming of cancer cells resulting in increased glycolysis and glucose requirements, the so-called Warburg effect, as well as the stimulatory effect of insulin-like growth factor on proliferating cells [37, 41].

In northern Sweden, a rapid decline in fat intake and hypercholesterolemia occurred between the years 1986–1992 [42], attributed in part to the cardiovascular disease prevention activities of the VIP [42]. Today fat intake has reached the peak levels of the 1980’s, and carbohydrate intake is decreasing [2]. LCHP scores have increased in VIP participants with repeated samples 10 years apart [19]. Furthermore, blood cholesterol concentrations in the northern Swedish population are increasing, despite increasing use of cholesterol-lowering drugs [3]. These temporal changes may have attenuated our results, as indicated by the positive association between a high LCHP score and overall cancer incidence in the sub-analysis restricted to the time period 1990–2002, when LCHP score was relatively stable in the VIP population. In the present study, roughly equal amounts of saturated and unsaturated fat replaced most of the carbohydrate reduction in subjects with high LCHP scores, and the excess protein consumed by subjects with high LCHP scores was primarily of animal origin. In Sweden, extensive positive media focus for carbohydrate-restricted diets in recent years has largely promoted fat, often animal fat, rather than protein, as the substitute for carbohydrates [1, 43]. The general enthusiasm for carbohydrate reduction, and the apparent preference for animal-based replacement foods in Sweden, thus underscores the importance of evaluating potential long-term implications for health.

The main strengths of this study were the large, population-based cohort, the extensive data available, such as an oral glucose tolerance test and BMI measured by health professionals, and the long, essentially complete, follow-up. In addition, the prospective study design reduced the risk of recall bias and reverse causation. We used an established LCHP score, which has been employed in previous studies [16‐18]. The LCHP score does not include fat intake. However, unlike macronutrient scores that incorporate fat, the LCHP score is independent of total energy intake. It is also simple, both to calculate and interpret, and it allowed separate consideration of the amount and quality of fat consumed. Food frequency questionnaires like the one employed is this study have inherent weaknesses, such as being a relatively inexact tool for the measurement of total nutrient and energy intake, but they are generally adequate for ranking and are an accepted and practical tool for large-scale epidemiological studies. Although several potential confounders were accounted for, residual confounding due to factors not measured (such as food items not included in the FFQ) or not adequately estimated (such as history of tobacco use) was likely present. Since we consider this study to be exploratory, the results were not corrected for multiple testing. The risk of chance findings should therefore be acknowledged. Numbers of cases were also low for some cancer types and in some subgroup analyses.

In conclusion, the results of this population-based, cohort study do not support an important role for a diet moderately low in carbohydrates and moderately high in protein, regardless of the quantity and quality of fat consumed, in determining the overall, long-term risk of cancer, although a possible increased risk of respiratory cancer was observed and a tendency of an increased general cancer risk over shorter time. Given the current widespread popularity of carbohydrate-restricted diets, and the limited data concerning potential long-term health effects of carbohydrate reduction and consequent increases in protein and/or fat intakes, these findings are important. In order to evaluate the role of more stringent carbohydrate restriction, investigations encompassing a wider range of macronutrient intakes, such as multicenter studies, will be needed.