Dear JULES,
How much fruit & veg is optimal?
Executive Summary
* Five-a-day was invented in 1991 at a meeting of the National Cancer Institute (NCI) and the Produce for Better Health Foundation in California. It was a marketing slogan to help companies that benefited from increased fruit and veg consumption.
* A study was published on March 1st, 2021 in Circulation. It came from the Harvard nutritional epidemiology ‘factory’. This one sought to investigate the optimal intake of fruit and vegetables for mortality (all-cause, cardiovascular (CVD), cancer and respiratory disease mortality.)
* The study included 66,719 women from the Nurses’ Health Study (NHS) and 42,016 men from the Health Professionals Follow-up Study (HPFS).
* The study found that the optimal intake of fruit and vegetables was five portions. More than this was not better.
* The usual limitations of epidemiology applied: 1) This was association, not causation. 2) The absolute differences were tiny and 3) There was a healthy person confounder. (Those consuming the most fruit and veg were less likely to smoke/drink and more likely to be active and taking nutrient supplements etc.).
* The raw data showed that the incident rate was very small and differed little (for nurses) and favoured lower intake (for health professionals) when those consuming 2 portions a day were compared with those consuming 7-8.
* After adjustment, the numbers flipped the other way round, which doesn’t make sense knowing the healthy person confounder. I’ve written to the authors for clarification.
Introduction
I am known for challenging a number of dietary beliefs – dietary fat is bad… saturated fat is worse… if you create a deficit of 3,500 calories you will lose 1lb of fat… fibre is essential… and, the most famous dietary belief of all, five-a-day. I’ve done a few media articles and interviews about five-a-day (Ref 1) – including a live interview on Sky News on April Fools Day in 2014 – the day a paper was published, which generated headlines that “seven a day would save lives” (Ref 2).
I first looked at the evidence for the five-a-day slogan for my obesity book, which was published in 2009. The research I did at the time found that five-a-day was first coined at a meeting of the National Cancer Institute (NCI) and the Produce for Better Health Foundation in California in 1991 (Ref 3). The Produce for Better Health Foundation is a trade body of companies that benefit from increased fruit and veg consumption. It did include companies as diverse as Driscoll’s (berries), General Mills and McDonalds, when five-a-day was invented in 1991. The latest list of members includes Del Monte, Dole, Campbell Soup, Kellogg’s, and Subway, to name just a few of the current 160 members (Ref 4). Five-a-day was created as a marketing slogan. The NCI trademarked the term and has defended the trademark since (Ref 5).
Why five? Why not! It’s the number of digits on one hand. It would seem like a good target to encourage – not too low to miss out on sales and not too high to seem unachievable.
My key issue with five-a-day is that it is not evidence based. For something to be evidence based, the evidence must come first. This critical fact seems to be beyond the grasp of public health officials. Next week I will share the exchange I had with Public Health England, back in 2013, when I challenged them on their five-a-day advice.
This week, we’re taking another look at five-a-day. A study was published on March 1st, 2021 in Circulation entitled “Fruit and Vegetable Intake and Mortality: Results From 2 Prospective Cohort Studies of US Men and Women and a Meta-Analysis of 26 Cohort Studies” (Ref 6). It came from some of the usual Harvard epidemiology authors. Being a Harvard publication, it used the usual two favourite studies of theirs – the Nurses’ Health Study (1984-2014) and the Health Professionals Follow-up Study (1986–2014).
The study sought to investigate the optimal intake of fruit and vegetables for mortality outcomes. The mortality outcomes reviewed were all-cause, cardiovascular (CVD), cancer and respiratory disease mortality.
The study
The study included 66,719 women from the Nurses’ Health Study (NHS) and 42,016 men from the Health Professionals Follow-up Study (HPFS) who were free from cardiovascular disease (CVD), cancer, and diabetes at baseline. Follow-up was conducted for up to 30 years.
Diet, in this study, was assessed using a Semi-quantitative Food Frequency Questionnaire (FFQ) (Ref 7). The questionnaire was completed at baseline and updated every 2 to 4 years (Ref 8). The fact that the questionnaire was repeated is a positive, as many studies take the baseline questionnaire and then never repeat it. Interestingly, the average intakes of fruit and vegetables, were stable during the follow-up years in the NHS and HPFS. This shows that public health campaigns have had little effect, which the Caerphilly study also found (Ref 9).
Participants reported their usual fruit and vegetable intake over the preceding year (from never to ≥6 times per day) of a standard portion size of each item (e.g., half a cup of strawberries, 1 banana, or half a cup of cooked spinach etc.). Frequencies and portions of each individual fruit and vegetable item were converted to average daily intake for each participant.
Table 1 in the paper, as usual, was the characteristics table. This tells us the characteristics of the people in the lowest fruit and vegetable intake groups, through to the highest. Another positive of the study was that people were split into quintiles (five equal groups) of person-years. This means that the same number of person-years (approximately 364,000) were in each group. This gave us equal sized groups, which is the fairest way of comparing people. This avoids the small group comparator issue where the same change has a much bigger impact on a small group than a large group.
The characteristics table painted the classic healthy person issue. The women in the NHS and the men in the HPFS were reported on separately. In both cases, compared with the highest intake group, the lowest intake group were more than twice as likely to be current smokers. They drank more alcohol. They were far less likely to do moderate/vigorous physical activity. They were less likely to take multivitamins. They were more likely to have hypertension. Age and some family histories of conditions favoured the lower intake group, but, overall, the higher intake group represented the healthy person. There was no adjustment for income or education as it was assumed that nurses and health professionals would have similar education and income. This is not unreasonable.
Results
During 30 years of follow-up in the NHS (1,822,058 person-years), 18,793 deaths were documented. During 28 years of follow-up in the HPFS (1,033,007 person-years), 15,105 deaths were documented.
There were two major findings:
1) Compared to the lowest intake (2 servings a day), intake of 5 portions of fruit and vegetables a day, or two servings of fruit and 3 of vegetables was optimal.
2) Higher intake than this wasn’t better.
The first finding alone made me smile. We had no evidence for five-a-day back in 1991. Yet here we are in 2021, 30 years later, and it turns out we got the right number way back then. What chance!
The numbers given for mortality were – in comparison with 2 servings a day, daily intake of 5 servings of fruit and vegetables was associated with a 13% reduced risk of total mortality, a 12% reduced risk for CVD mortality, a 10% reduced risk for cancer mortality and a 35% reduced risk for respiratory disease mortality (Ref 10).
More not better
Table 2 showed that the fourth highest intake group was the closest to five-a-day. This group (quintile 4) averaged 5.3 portions a day for the nurses and 5.4 portions a day for the health professionals. The highest intake group (quintile 5) averaged 7.3 portions a day for the nurses and 7.6 portions a day for the health professionals. Table 2 confirmed that the fourth quintile was best for mortality and that quintile five (the highest intake) wasn’t better.
This finding was best illustrated in Figure 1, which presented the association between fruit and vegetable intake and mortality as intake increased in portions per day. Figure 1 also had diagrams for CVD, cancer, and respiratory disease mortality.
As you can see from Figure 1 above, many of the curves started to turn upwards beyond the optimal intake (the optimal intake is at the lowest point of any curve). In most cases, the tip of the turn up was still lower than the reference line of 1.0. This means that, for example, 6 portions were still better than 2 but not as good as 5. However, in some cases – e.g., vegetable intake and cancer mortality or fruit intake and CVD mortality – the tip of the turn up was back to, or beyond, the reference line meaning that this higher intake was as bad as, or worse, than the lower portion intake.
However…
I spotted an issue with Table 2 (given what we learned in Table 1). The following numbers in black are extracted from Table 2. They show the extreme groups – Q1(lowest intake) to Q5 (highest intake) – before any adjustment has taken place. The two red rows show the relative risk after adjustment, using the low intake group as the reference.
The black numbers show that for the NHS, the incident rate was virtually the same for the lowest and highest intake groups. That’s the raw data for deaths in person-years. For the HPFS, the incident rate was higher for the highest intake group – quite markedly so. The people in Q5 are the healthier people remember (less likely to smoke/drink, more active, taking nutrient supplements etc). Hence, when we adjust for these health advantages, we would expect the incident rate to go up in the Q5 group (relative to Q1). But it didn’t. It went down.
The low intake people were slightly younger and had some family history advantages, but at best these should have counterbalanced the smoking/drinking/activity/hypertension etc. At worst, the adjustment should have favoured the low intake group and made this look even better relative to the high intake group. Adjustment in epidemiological papers is always a black hole. This one raised an alarm for me. I have written to the corresponding author about this (Ref 11).
The three issues with epidemiology
We have the usual three issues, with which we are familiar:
1) This is association, not causation.
Notwithstanding the concern about adjustment, we cannot say that fruit and vegetable intake reduces mortality. We can only observe that fruit and vegetable intake is associated with reduced mortality. Does fruit and veg make people healthy or do healthy people eat fruit and veg?
2) This is relative and not absolute risk.
Notwithstanding the concern about adjustment, even the relative risks – of 10-13% – are small and in no way suggest causation. The respiratory mortality differential was the largest. The discussion part of the paper offered antioxidant properties of fruit and vegetables as a possible explanation for the difference in respiratory mortality. The supplemental file reported that there were far fewer respiratory deaths than other deaths and so the respiratory number will be less precise (the wider confidence intervals in the diagrams above confirm this).
The absolute risk difference – before adjustment – was minute. In the NHS, Q1 people averaged 106 incidents in 10,000 person-years while Q5 people averaged 107 incidents in 10,000 person-years. That’s a difference of 1 in 10,000 person-years – and it was in favour of the lower intake. The HPFS reported unadjusted differences of 28 in 10,000 person-years – again in favour of the lower intake.
3) The healthy person confounder is significant.
This has been addressed throughout this note. Indeed, it was confirmation of the usual healthy person confounder that led to the discovery that the incident rate looked wrong.
Different types of fruit & veg
This paper also looked at different types of fruit and vegetables and found that not all were associated with lower mortality. Intake of starchy vegetables (including corn and peas), cruciferous vegetables (broccoli, cabbage, cauliflower, sprouts, kale), potatoes and fruit juice were not associated with lower mortality. Intake of green leafy vegetables, non-starchy vegetables, citrus fruits and vitamin C or beta carotene rich fruits and vegetables were associated with lower mortality.
The detail for this was in the supplemental file. It lost credibility when kale was included in both cruciferous vegetables (no association) and green leafy vegetables (an association) and when broccoli, cabbage, cauliflower, sprouts and kale were included in cruciferous vegetables (no association) and the same five were included in non-starchy vegetables (an association).
I’ll let you know if I hear back from the authors. Until then, we have no more evidence for five-a-day than we did when it was invented.
Until the next time
All the best – Zoë
References
Ref 1: https://www.mirror.co.uk/news/uk-news/five-fruit-vegetables-day-fairy-3340324
https://www.walesonline.co.uk/news/health/five-a-day-rule-fact-fiction-1838420
https://www.independent.co.uk/life-style/health-and-families/features/five-day-enough-2248856.html
Ref 2:Oyebode et al. Fruit and vegetable consumption and all-cause, cancer and CVD mortality: analysis of Health Survey for England data. BMJ Epidemiology & Community Health. 2013. https://jech.bmj.com/content/68/9/856.full
Ref 3: https://www.cdc.gov/nccdphp/dnpa/nutrition/health_professionals/programs
/5aday_works.pdf (see Preface)
Ref 4: https://fruitsandveggies.org/contributors/?contributor-type=members
Ref 5: Schultz. Food Fight: NCI Aims for Consumer Clarity With Complaint Against Dairy Group. Journal of the National Cancer Institute. 2003. https://academic.oup.com/jnci/article/95/13/934/2520300
Ref 6: Wang et al. Fruit and Vegetable Intake and Mortality: Results From 2 Prospective Cohort Studies of US Men and Women and a Meta-Analysis of 26 Cohort Studies. Circulation. March 2021. https://www.ahajournals.org/doi/10.1161/CIRCULATIONAHA.120.048996
Ref 7: A Food Frequency Questionnaire (FFQ) is a checklist of foods and drinks with a frequency response section for subjects to report how often each item was consumed over a specified period of time. Semi-quantitative FFQs collect portion size information as standardized portions or as a choice of portion sizes.
Ref 8: Willett W. Food frequency methods and Reproducibility and validity of food-frequency questionnaires. In: Nutritional Epidemiology. 3rd ed. Oxford University Press; 2013.
Ref 9: Elwood et al. Healthy Lifestyles Reduce the Incidence of Chronic Diseases and Dementia: Evidence from the Caerphilly Cohort Study. PLoS ONE. 2013. https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0081877
Ref 10: The abstract of the paper reported the results as. “In comparison with the reference level (2 servings/d), daily intake of 5 servings of fruit and vegetables was associated with hazard ratios (95% CI) of 0.87 (0.85–0.90) for total mortality, 0.88 (0.83–0.94) for CVD mortality, 0.90 (0.86–0.95) for cancer mortality, and 0.65 (0.59–0.72) for respiratory disease mortality.”
Ref 11: Dear Dr Wang
I am enjoying reading your recent paper in Circulation (https://www.ahajournals.org/doi/10.1161/CIRCULATIONAHA.120.048996).
I have a query about incidence and adjustment please…
Table 1 tells us that (as usually happens) the lowest intake group (Q1) is the less healthy group (more than twice as likely to be current smokers, less likely to be active or taking supplements, drinking more, less likely to be consuming an overall healthy diet etc). Q1 were younger and had some advantages in terms of family history of conditions, but one would expect adjustment of raw data to reflect that Q5 people were healthier than Q1.
Table 2 tells us that the incident rate before adjustment between Q1 and Q5 was virtually identical (106 vs 107 per 10,000 person-years) for NHS and in favour of Q1 (136 vs 164 per 10,000) for HPFS. I would thus expect adjustment to either maintain these differentials (if age counter balanced all the other factors) or to extend these differentials – in favour of Q1. The adjustment lines in Table 2 report the opposite – after adjustment, Q5 has significantly lower incidence than Q1.
Please can you help me to understand this?
Thank you
Kind regards – Zoe
