Introduction: Controversies about the long-term health consequences of low-carbohydrate diets (LCD) persist. Studies have shown that the associations between LCD, all-cause mortality, and type-2 diabetes, a key risk factor for gastrointestinal (GI) cancers, depend on macronutrient quality, with lower risks for healthy LCD and higher risks for unhealthy LCD. To understand the impact of carbohydrate restriction on GI cancer risk, we leveraged three prospective cohorts with high-quality dietary assessment followed for clinically adjudicated cancer cases.
Methods: We included 222,745 participants from the Health Professional Follow-up Study (1988-2016), Nurses’ Health Study (NHS, 1982-2018), and NHSII (1990-2018) without a prior history of cancer or heart disease who returned validated food frequency questionnaires every four years. We calculated total LCD scores based on 11 equally sized ranks of protein and fat in ascending order and carbohydrates in descending order. Similarly, animal- and plant-based LCD scores were created according to animal and plant proteins and fats, respectively. Additionally, we created unhealthy LCD scores based on animal protein, saturated fat, and inverse-ordered high-quality carbohydrates (e.g., whole grains, fruits) and healthy LCD scores based on plant protein, unsaturated fat, and inverse-ordered low-quality carbohydrates (e.g., refined grains, added sugar). Using multivariable Cox models, we estimated hazard ratios (HRs) and 95% confidence intervals (CIs) for incident GI cancers.
Results: During 2,878,573 person-years of follow-up across three cohorts, we documented 7,035 incident GI cancers (i.e., oral and pharyngeal, esophageal, gastric, small intestinal, colon, rectal, pancreatic, gallbladder, and liver cancers). Compared with the lowest quintile, the HR for GI cancers was 1.11 (95% CI=1.02-1.20) for the highest quintile of both unhealthy and animal-based LCD scores (Figure 1). Similar adverse associations were observed for colorectal and pancreatic cancers, while these LCD scores were not associated with other types of GI cancers. The risks for all GI cancers were not materially different across quintiles of total, healthy, and plant-based LCD scores. When assessing individual carbohydrate sources (Figure 2), we found that carbohydrates from intact whole grains were associated with reduced GI and colon cancer risks, and carbohydrates from legumes were associated with reduced pancreatic and liver cancer risks. In contrast, carbohydrates from other starchy vegetables (e.g., sweet potato) were associated with increased GI and pancreatic cancer risks.
Conclusion: Our findings suggest that the associations between LCD and GI cancer risk may depend on the quality and sources of macronutrients, especially carbohydrates. The mechanisms underlying these associations require further investigation.

<b>Fig 1. The association between low-carbohydrate diet score (LCDS) and the risk of overall gastrointestinal (GI) cancer is dependent on the quality and sources of macronutrients. </b>Cox proportional hazards regression was adjusted for age, calendar year, cohort, race, BMI, family history of cancer, personal history of diabetes, physical examination, smoking status, pack-years of smoking, physical activity, current multivitamin use, regular aspirin use, regular non-aspirin non-steroid anti-inflammatory drug use, menopausal hormone use, and total calorie intake, with a 2-year lag period between exposures and outcomes. The median value for each quintile of LCDS was modeled as a continuous variable to test for linear trend.

Fig 1. The association between low-carbohydrate diet score (LCDS) and the risk of overall gastrointestinal (GI) cancer is dependent on the quality and sources of macronutrients. Cox proportional hazards regression was adjusted for age, calendar year, cohort, race, BMI, family history of cancer, personal history of diabetes, physical examination, smoking status, pack-years of smoking, physical activity, current multivitamin use, regular aspirin use, regular non-aspirin non-steroid anti-inflammatory drug use, menopausal hormone use, and total calorie intake, with a 2-year lag period between exposures and outcomes. The median value for each quintile of LCDS was modeled as a continuous variable to test for linear trend.

<b>Fig 2. The association between each standard deviation of carbohydrate intake (%energy) from various sources and the risk of all and major gastrointestinal (GI) cancers. </b>Cox proportional hazards regression was adjusted for age, calendar year, cohort, race, BMI, family history of cancer, personal history of diabetes, physical examination, smoking status, pack-years of smoking, physical activity, current multivitamin use, regular aspirin use, regular non-aspirin non-steroid anti-inflammatory drug use, menopausal hormone use, and total calorie intake, with a 2-year lag period between exposures and outcomes.

Fig 2. The association between each standard deviation of carbohydrate intake (%energy) from various sources and the risk of all and major gastrointestinal (GI) cancers. Cox proportional hazards regression was adjusted for age, calendar year, cohort, race, BMI, family history of cancer, personal history of diabetes, physical examination, smoking status, pack-years of smoking, physical activity, current multivitamin use, regular aspirin use, regular non-aspirin non-steroid anti-inflammatory drug use, menopausal hormone use, and total calorie intake, with a 2-year lag period between exposures and outcomes.

Mingyang Song

Appearances