Association of butter and plant-based oils with mortality: Further clarifying the butter

SUMMARY

This prospective cohort study was based on data from three large cohorts in the United States–the Nurses’ Health Study (NHS), the Nurses’ Health Study II (NHS-II), and the Health Professionals Follow-up Study (HPFS). It had 221 054 adults who were free of diabetes, cardiovascular disease (CVD), and cancer at baseline. They were followed up for 33 years.¹ The intake of butter and plant-based oils was recorded at baseline and every 4 years thereafter. Daily butter intake was measured as the sum of responses to three questions about intake of butter and margarine blend, spreadable butter added to food, and butter used in baking and frying. The intake of plant-based oil was recorded based on oil brand (safflower, corn, soybean, canola, and olive oil), type of fat, and cooking methods. The participants were divided into four levels (1-lowest to 4-highest) based on the quartile of intake. The outcomes of interest were total mortality, CVD-related mortality, and cancer-related mortality, as assessed through review of death certificates, medical records, or autopsy reports, by a physician blinded to food consumption data. Cox proportional hazards regression analysis was used to compare outcomes for higher intake levels versus the lowest intake levels. Adjustments were made for several confounders, including age, total energy intake, menopausal status, alcohol, smoking, body mass index, aspirin use, hyper-tension, hypercholesterolaemia, and family history of myocardial infarction, cancer, and diabetes. Further, the risk of mortality when replacing 10 g/day of butter intake with an equivalent amount of plant-based oils was estimated.

The study showed that participants with the highest level (level 4) of butter intake had a 15% higher mortality than those with the lowest level (hazard ratio [HR] 1.15; 95% CI 1.08– 1.22). By comparison, the highest level of plant-based oil intake was associated with a 16% lower mortality than the lowest intake (HR 0.84; CI 0.79–0.90). This was also true for intake of plant oils other than olive oil (HR 0.92 CI; 0.86–0.98).

On sub-group analysis, the highest level of butter added to food or bread had a positive association with total mortality (HR 1.09; CI 1.04–1.14), whereas butter for baking or frying did not.

For plant-based oils, higher intakes of canola (HR 0.85), soybean (HR 0.94), and olive (HR 0.92) oils also showed a significant negative association with mortality, but this was not observed for safflower and corn oils.

Higher butter intake was associated with 12% higher cancer mortality (HR 1.12 per 10 g/day; CI 1.04–1.20) and higher plant-based oil intake with 11% lower cancer mortality (HR 0.89; CI 0.85–0.94) and 6% lower CVD mortality (HR 0.94; CI 0.89–0.99). Surprisingly, butter intake was not associated with CVD-mortality.

Substitution of 10 g/day of butter with an equivalent amount of plant-based oil was estimated to be associated with a 17% lower risk of total as well as cancer-related mortality.

COMMENT

This study used data from three well-known long-term cohorts, i.e., the NHS, NHS-II, and the HPFS, to assess the effect of intake of various forms of dietary fat on mortality.

The study has several strengths, including a very large sample size of >220 000 individuals, coverage of a wide variety of age and sex groups (NHS: age mean [standard deviation]=56.1 [7.1] years, 100% females; NHS-II: 36.1 [4.7] years, 100% females; HPFS: 56.3 [9.3] years, 100% males), a long follow-up period, repeated measurement of dietary exposures at several time points and reliable determination of outcomes.

However, the design and results raise certain questions. The study had an observational design with inherent limitations, i.e. risk of various forms of bias and of confounding. First, the study cohorts consisted of highly educated, high socioeconomic status, predominantly Caucasian nurses and health professionals, whose health behaviour and outcomes are likely to differ from those of the general population. In view of this selection bias, the results of this study may not be generalizable to other population groups. Second, dietary studies also have an inherent limitation of recall bias. However, randomized controlled trials to evaluate dietary interventions are inconvenient and fraught with ethical concerns, and hence, we have to rely on observational studies while exercising appropriate caution.

Further, the participants with higher butter intake differed from those with lower butter intake, being more frequent smokers, exercising less often, and having higher total energy intake and body mass index (BMI), covariates well known to be associated with the outcomes of interest, thereby creating a risk of confounding. To the authors’ credit, they tried to correct this by adjusting for several covariates; however, such an adjustment is itself beset with several challenges, including residual confounding, failure to account for interaction between covariates, and model overfitting. This mandates caution in interpreting and applying the study’s results. Further, a closer look at the results shows that these adjusted analyses included fewer subjects than the total size of the three cohorts, further increasing the risk of selection bias. Butter, being a costlier dietary fat, is likely to be associated with a higher socio-economic status, which itself influences the risk of several diseases; however, the analysis did not adjust for this potential confounder.

The study methods state that intake of butter-margarine blends was recorded as butter.¹ As the authors themselves allude, this may have led to a false association between butter intake and mortality, due to the high quantity of trans-fatty acids (TFA) in margarine, which is made from plant oils, a form of measurement bias.

Furthermore, somewhat surprisingly, though the intake of butter was associated with total mortality, it showed no association with CVD-related mortality; it is noteworthy that the intake of butter, which is rich in saturated fatty acids (SFA), has previously been found to be linked to risk of CVD.^2,3

Though the paper’s methods refer to the comparison of four quartiles of butter intake, the groups compared were very different in size. Further, the excess risk with higher butter intake was observed only in the group with the highest or level 4 intake (n=2504) versus the least or level 1 intake (n=155 310) and not for level 2 (n=18 633) or 3 intake (n=5946). With the association of mortality risk with butter intake being based on data from only 2504 subjects, or around 1.1% of the total cohort, we would urge caution in reading too much into this finding. Further, the presence of such risk in only a few members of the cohorts means that the population impact of any intervention to reduce butter intake would be quite small.

How do the results of this study compare with the previous information? A study of a larger cohort of participants from 18 countries (largely low and middle-income countries including India) showed no association of SFA intake with total mortality.⁴ This difference may be related to the relatively higher proportion of carbohydrates and a lower proportion of fat in the diet in low and middle-income countries. Hence, in these latter settings, notwithstanding the results of the current study, restricting SFA-rich foods, e.g., butter, may not be prudent.⁵

The authors attribute the beneficial effect of plant-based oil to their antioxidant content, particularly Vitamin E. The content of antioxidants, or of other protective constituents, if any, in plant-based oils may vary with the type of extraction process used. For instance, cold-pressed oils (referred to in India as kacchi ghani) may be healthier than refined oils, since the heating step during refining is associated with some loss of antioxidants and generation of TFAs.⁶ Moreover, cooking oils may be subjected to reuse or reheating, which is known to increase their TFA content, processes that are considered hazardous.⁶ The study questionnaire did not consider these factors.

The observational studies on diet and health receive a lot of media attention, despite having several limitations, mutually contradictory results, and reporting findings that are not confirmed in subsequent randomized trials.⁷ Hence, their results should be taken with a pinch of salt (pun intended!).

This study question is of particular interest for the Indian population which is more prone to cardiovascular morbidity and mortality.⁸ How well do its results apply to us? First, as discussed above, the applicability of the study’s results to a population with a different dietary composition is unclear. Second, in India, the use of butter may not be exchangeable with that of liquid vegetable oils, since Indians use butter, a solid fat, primarily as a spread for breads and parathas and at the time of consumption rather than during cooking, and liquid vegetable oils for frying and during cooking.

Third, the nature of fats consumed in India differs greatly from those used in the study populations and even across the country. In our population, the solid fats (i.e. fats with high melting points) that rule the roost are desi ghee and vanaspati, and not butter. Desi ghee is animal-derived, clarified butter made by heating butter to remove other residual milk solids; it is rich in SFA and is largely similar to butter in fat composition. Vanaspati, on the other hand, is made by partial hydrogenation of vegetable oils, currently primarily palm oil, to generate a ghee-like consistency; it is rich in TFA.

Also, in India, the vegetable oils consumed differ from those used in the United States, with Indians consuming much more palm oil.⁹ Not all plant-based oils are alike. Palm oil, in its natural form, has been associated with CVD due to its high SFA content.⁶ It is also used to make vanaspati, which again has a high TFA content and is associated with CVD.¹⁰ Within India too, there are geographic variations in the use of vegetable oils, with mustard oil being used more often in northern India, and coconut oil in southern parts, though admittedly these distinctions are getting blurred with population migration and urbanization. Mustard oil has higher monounsaturated fatty acid (MUFA) content, similar to that of olive and canola oils; however, unlike those, it has a high concentration of erucic acid, a MUFA that has been linked to cardiac lipidosis in animal models.⁶ Although several human studies have attested to the safety of this oil, its use in the USA and Europe remains restricted.¹¹ By comparison, coconut oil is solid and has a high medium-chain SFA content. Its intake has been considered to contribute to an increase in low-density lipoprotein cholesterol and to be a risk factor for CVD.¹²

Further, the genetic differences from the Caucasian population may also limit the applicability of this study’s findings to the Indian population.¹³

In summary, this study is important since it strongly suggests that the choice of dietary oils impacts health outcomes, though its findings do have several important limitations. There is a clear need to undertake similar studies in India to explore the relationship of various oils that are commonly used here, with mortality.