select search filters
roundups & rapid reactions
before the headlines
Fiona fox's blog

expert reaction to study looking at fried food and heart disease and stroke risk

A study published in Heart looks at whether fried food intake is linked to risk of heart disease and stroke.


Prof Riyaz Patel, Professor of Cardiology, and Consultant Cardiologist, UCL, said:

“This is a well conducted analysis into the effects of eating fried foods on heart disease, stroke and early death.  Rather than conducting a new study, the authors have searched medical databases to find previously completed studies that looked at this question and analysed these all together to determine if any meaningful signal can be identified.

“Studying the link between fried food intake and disease is challenging, as questionnaires, usually used to capture what and how much was eaten, are dependent on participant recall and can therefore be incomplete.  Moreover, we also don’t eat foods in isolation, so it is hard to fully capture the complexity of what we eat and how, especially over many years.  Importantly, other factors that go with eating fried food could also be contributing to risk, like a tendency to drink more sugary drinks, added salt use, eating other unhealthy foods, less exercise, smoking and deprivation levels.  Much of this data may not have been captured in prior studies so cannot be fully accounted for.

“As expected then, the studies the authors found were variable and as a result the conclusions they could draw by combining them together, despite using appropriate statistical methods, need to be interpreted with caution.

“Nonetheless, the overall signal they found, that fried food was associated with an increased risk of heart disease and stroke, with each portion being associated with a greater risk, seems to fit with our understanding of biology.  We know that frying food can degrade its nutritional value, generate trans-fats which are known to be harmful, as well as increasing the calorie content of the food, all of which eventually lead to processes that can cause heart disease.

“However, as this study only shows an association, it cannot say for sure that eating fried food itself causes heart disease.  More studies are needed, perhaps with novel technologies that can record our diets more accurately over long periods of time.”


Prof George Davey Smith, Director of the MRC Integrative Epidemiology Unit, and Professor of Clinical Epidemiology, University of Bristol, said:

“One of the very few positives of the pandemic has been the absence of frequent media reports on the confusing and often contradictory findings from the latest dietary/nutritional epidemiological study.  The paper by Pei and colleagues is better than many of the studies in this field, as it acknowledges the weaknesses of the data available.

“The authors carried out a systematic review and meta-analysis of observational studies of fried-food consumption and the risk of cardiovascular disease and overall mortality.  They demonstrate that there are very considerable differences in the results obtained from studies using different methods and that were carried out in different contexts.  Whilst the overall effects in the meta-analysis suggested a positive association of fried food intake with heart disease, the variation between studies – with many showing no association – is striking.  Clearly in many contemporary contexts high degrees of fried food consumption will go together with less favourable socioeconomic circumstances and patterns of health related behaviours, such as smoking.  Conventional statistical approaches cannot deal with such confounding, as shown by the fact that epidemiological studies demonstrated that use of vitamin E supplements was associated with a substantial reduction in coronary heart disease risk, whilst randomised controlled trials revealed there was no actual protection.   Looking at findings from studies carried out in contexts where such confounding would be different is a powerful check on such observational epidemiological results.  A study – not identified by the authors – carried out 70 years ago in the US1 investigated fried food consumption at a time when it was socioeconomically favoured groups that consumed most fried food.  Amusingly, in this circumstance, it appeared that fried foods were associated with a substantially reduced risk of mortality.  15 or more servings of fried food a week was apparently associated with the lowest risk.  This is, of course, very likely to be yet another spurious finding – just like the myriad of nutritional epidemiological findings that in pre-pandemic days often received extensive media coverage.”

1 E. Cuyler Hammond, Sc.D., Smoking in Relation to Mortality and Morbidity. Findings in First Thirty-Four Months of Follow-Up in a Prospective Study Started in 1959, JNCI: Journal of the National Cancer Institute 1964;32:1161–1188


Prof Alun Hughes, Professor of Cardiovascular Physiology and Pharmacology, Head of Department of Population Science & Experimental Medicine, and Associate Director of the MRC Unit for Lifelong Health and Ageing, UCL, said:

“This article is a systematic review and meta-analysis that pools information from published observational studies looking at the correlation between self-reported intake of fried food and measures of cardiovascular disease and mortality.  Most of the studies were conducted in the USA and this may limit the generalizability of the findings.  There is also evidence of publication bias, i.e. studies showing a positive association were published more often that would be expected by chance.  This results in over-estimates of the strength of associations.  The pooled results indicate that a higher intake of fatty food was associated with a higher risk of a composite and some individual cardiovascular events –  on average a 16% higher risk of a composite cardiovascular event in the category of people with the highest intake of fatty food after accounting for publication bias.  Associations between fatty food intake and either cardiovascular deaths or all-cause mortality were weak and unconvincing (3% increase for both comparing high with low intake (unadjusted for publication bias)) and compatible with chance variation around no association.

“The major issue with this study is the potential for bias, and particularly confounding.  This is a ubiquitous problem for observational studies.  For example, high consumption of fried foods is likely to be associated with other unhealthy behaviours, consequently these other linked factors (rather than fried food intake itself) could account for any observed association.  Some of the included studies attempted to control for some of these risk factors.  Overweight/obesity (assessed as body mass index) and physical activity (presumably based on self-report) appear to explain some of the observed association.  Unfortunately, these potential confounders are unlikely to be fully controlled for, and other important risk factors, e.g. smoking, socioeconomic disadvantage, ethnicity are not accounted for at all.  This means that it is not possible to conclude that the association between fried food intake and cardiovascular events is due to a causal effect of fried food.

“The findings of this study are consistent with current guidance to limit intake of fried foods but cannot be considered as providing definitive evidence on the role of fried food consumption in cardiovascular health.”


Prof Gunter Kuhnle, Professor of Nutrition and Food Science, University of Reading, said:

“This study investigates how “fried food” affects health, and at first glance, it seems to suggest that there is a rather modest increase in risk.  This is not surprising, as people who consume a lot of fried food are often less healthy than those who don’t, although it is likely that there are a wide range of explanations well beyond the actual diet.

“The study was conducted as a meta-analysis, combining results from 19 studies that were conducted previously.  The authors do this very well, and technically, this study is very sound.  However, combining so many different studies from different parts of the world and different types of fried foods makes it impossible to interpret the results.  There is a big difference between different types of fried foods, and it is unlikely that they all have the same impact on health.  The authors combine studies on fried fish, fried snacks, fried potatoes – foods that differ considerably in food composition.  Moreover, the authors ignore that fried foods can be very different in different countries and regions – fish & chips, fish fry and pescaíto frito are quite different from each other.

“Reducing the intake of fat, especially saturated fats and trans-fats, can reduce the risk for chronic diseases, and these fats can be found in many fried foods.  But this study does not provide any new information about the impact of fried foods on health.”


Dr Duane Mellor, Registered Dietitian and Senior Teaching Fellow, Aston Medical School, Aston University, said:

“The authors undertook a review of the published research investigating the link between eating fried foods and cardiovascular disease (including heart disease and strokes) in a number of populations around the world.  Perhaps unsurprisingly the authors found a link between eating fried foods and cardiovascular disease, however they did not find a link between fried foods and with cardiovascular disease related deaths.  It is a well-designed analysis, however as is the nature with this type of research, it is only as good as the research in is analysing, in this case there is a significant amount of variation between the studies included, which can reduce the reliability of the analysis of the pooled data.  This is perhaps the result of the studies recording what is meant by fried food in different ways, some just included fried fish, others fried potatoes and one looked at food eaten outside of the home.  This means that the interpretation of the data needs to take into account this issue, that the data may not be completely reliable.

“The study does not provide any evidence why fried foods might lead to this association with risk of death from cardiovascular disease, but authors do propose that it may be due to the higher calorie content of fried foods, the process of frying foods could produce new compounds which might increase risk of cardiovascular disease and/or the use of trans-fats in the production of the foods.  It is important to note that this analysis only looked at the association between the eating of fried foods and cardiovascular disease, it did not consider the overall diet eaten by those not eating fried foods and whether they are health.  This is often a problem with this type of epidemiology, in that it only considers the association of one type of food with risk of disease and people actually typically eat a wide variety of foods.

“This study did highlight that there seemed to be an increasing linear association between eating fried foods and cardiovascular disease, with each 114g (about 4oz) eaten per week being associated with increased risk of both heart disease and heart failure.  It is worth considering that, if the relationship is causal, we can’t assume that this association is definitely down to the fat content of the foods, as many of these foods are highly processed and often contain both fat and carbohydrate together.  So, when considering this type of study, it is important to consider that although reducing fat intake is a logical part of a healthy diet, it is also important to look at what foods are eaten in its place.”


Prof Kevin McConway, Emeritus Professor of Applied Statistics, The Open University, said:

“This new research looks at something that sounds interesting and important – is eating fried food associated with cardiovascular disease (CVD: heart disease, strokes and so on), or with an increased risk of early death (from any cause)?  But there are so many ifs and buts and caveats about the findings, that in my view it really can’t tell us anything in any detail.  Generally the problems aren’t with the statistical methods that the researchers used.  Instead they are with what are usually the most difficult parts of a meta-analysis like this – the information obtained from the previous research studies that are included, and the interpretations of the pooled results.

“The first issue about the studies that are included is that they are all observational.  People in them weren’t given particular amounts of fried food to eat by the experimenters – they just ate what they would have eaten anyway, and then the researchers on the included studies recorded what the participants ate, and followed them up and recorded how many of them had various kinds of CVD, and, if they died, when the deaths happened.  The recording of deaths and of disease are likely to be pretty accurate, but the food measurements are another matter.  In most of the studies, food frequency questionnaires of some sort were used.  The participants are asked to say how often they ate certain food types, and how many servings they ate in a week or a month or some other period.  That’s fine as far as it goes, but it can involve people having to think back, possibly over as long as year, about their eating habits.  So there might be issues of inaccurate recall, as the researchers acknowledge.  Also, though, typically with one of these questionnaires, people are given guidance on how big a ‘serving’ is.  But do you know how much your lunch weighed, and so how many official ‘servings’ were involved?  Different studies used different serving sizes in some cases, and although the researchers in the new study took that into account as far as they could, I just don’t think that the data support statements about the increase in risk from eating one extra serving a week of exactly 114 grams.  (Why 114 grams?  I think it’s because that’s almost exactly four ounces, and many the studies were done in the US where the metric system is not used.  I think the researchers simply converted the round figure of 4oz to 114g.  To say that a quantity of food is about four ounces is probably relatively easy to assess, but knowing the size of a serving to the nearest gram, as the 114 gram figure indicates, is surely far beyond the accuracy of the measurement process.)

“But these measurement issues aren’t really the biggest problem with observational data.  The problem is that there will be other differences between people who eat a lot of fried food, and people who eat a little, apart from their fried food consumption.  Any of these other differences, which are often called confounders, might be the cause of differences in CVD rates, or in death rates, and not the fried food consumption at all.  In an observational study, researchers can make statistical adjustments to allow for confounders, and that type of adjustment was generally made in the studies that were included in this new research.  Such adjustments can never be perfect, so it’s never really possible to be sure what causes what from these types of observational study, and the conclusions have to be about associations and correlations, not about cause.  But it’s even more complicated when putting observational studies into a meta-analysis. Included studies will have made different adjustments.  The researchers running them may not all agree on what adjustments are advisable, and no adjustment can be made if no data are available on a particular confounder (and different studies have different data available).  So the different risk measures from different studies are not even calculated on the same basis as one another.  In fact this shows up in some of the reported results in the new study.  On comparing rates of major cardiovascular events between participants who ate the largest and the smallest amounts of fried food, there were very large differences between the findings, depending on whether the measure of risk (risk ratio) in the studies being put together was adjusted for BMI (body mass index).  It’s very well known that being obese (high BMI) is associated with higher CVD risk.  In studies adjusted for BMI, the risk of a major cardiovascular event was found to be 17% higher in people who ate the largest amount of fried food, relative to people who ate the smallest amount.  But in studies not adjusted for BMI, the risk was a whopping 46% higher in the group who ate the most fried food.  Surely quite a major part of that increased risk is likely to be due to differences in BMI rather than the fried food (though a further complication is that eating fried food may be associated with BMI).  It’s even more marked in a comparison of studies that adjusted for physical activity levels, and studies that did not adjust for it.  In those that adjusted, the risk of major cardiovascular events was 17% higher in the highest fried food consumers than the lowest consumers, but in studies not adjusted for physical activity, the risk was 80% higher in the highest fried food consumers than the lowest consumers.  A great deal of that percentage must be to do with exercise, in some way.  And, if you’re wondering how much fried food people in these highest and lowest categories actually ate, well, I can’t tell you in a simple way, because the consumption categories were defined in different ways in different studies.

“So we’re not going to be able to say what is causing any associations in this meta-analysis, because it’s putting together a lot of observational studies that can’t tell us about cause.  And it’s putting together studies that aren’t necessarily measuring the same things anyway.  Bearing that in mind, what can be said about the findings?

“In a sense, not a lot can be said.  The measures of relative risk, that the researchers calculated, come from what are called ‘random effects’ meta-analysis.  That does not mean they are just a bunch of haphazard numbers, and random effects meta-analyses are very often the right thing to do (and are definitely the right thing to do here).  But, in a random effects meta-analysis, there’s no assumption that the size of the association (say, between fried food consumption and CVD) is the same in all the studies that have been combined.  Instead what’s estimated is an average measure of association across all the studies included.  If the sizes of the associations in the individual studies aren’t too different, that’s fine, and if someone did a new study, it would be reasonably likely to have a similar size of association to the estimate from the meta-analysis.  But part of the output from each of these analyses is a measure of heterogeneity – that is, an overall measure of how different from one another the associations are in the studies that were combined.  That can be done in different ways.  These researchers use a standard measure of heterogeneity, called I-squared.  For most of their meta-analyses, the I-squared values are very large, indicating that their overall relative risk measure is the average of a set of very different results.  That indicates that a new study of the same thing, in a different context, might well produce a result that is nowhere near the result of the meta-analysis.  With this very large heterogeneity, the result depends so much on which studies happen to have been done, and included in the analysis, that any conclusion is pretty shaky.  As just one example – not even the worst, but I’ve chosen it because it’s easy to describe – the analysis looking at the association between stroke and fried food consumption has a very large I-squared value of 80.7%. It is based on four studies.  One looked at fried fish consumption in a group of elderly Americans, a second also looked at fried fish but in a younger American group, the third looked at consumption of fried potatoes in Sweden, and the fourth had collected data on consumption of deep fried snacks in Hyderabad, India.  Given these very different contexts, it’s not at all surprising that the results of the individual studies varied between essentially no association between consumption and strokes in the Swedish potato study to about double the risk of a stroke in the most frequent compared to the least frequent consumers of deep fried snacks in Hyderabad.  In a way, the point of meta-analyses is to see the overall pattern through variability, but this is really pushing it a bit too far in my view.  In the comparisons of highest and lowest consumption groups, there is high heterogeneity for every association with different measures of CVD and with overall mortality, except for the comparison of CVD mortality, but there, there was no evidence of an association with fried food consumption.

“The position is slightly clearer in the dose-response analyses that look at the extra risk of an outcome (CVD or overall mortality) for each extra serving of fried food, where two analyses show no heterogeneity: the analyses for coronary heart disease and for CVD mortality.  Again with CVD mortality, there’s no evidence of an overall association anyway.  For coronary heart disease, there is evidence of an association, but that association isn’t very strong – an increase of 2% in risk for every extra serving per week.  That doesn’t mean that, for two groups whose consumption differs by one serving a week, the risk might be 5% in the low consumption group and 5+2 = 7% in the other.  It means that the risk in the higher consumption group would be high by 2% of the lower consumption group’s risk, so 1.02 times the risk in the lower group.  To put a meaningful figure on the change in absolute risk isn’t easy, because the risk of a heart attack (or other CVD) varies hugely between the studies.  In one of them, for instance, in a reasonably young group in Spain, about 149 in every 10,000 participants got coronary heart disease in 11 years of follow-up.  That’s a pretty low risk If they all ate an extra serving of fried food each week, and if the association between fried food eating and coronary disease were known to be one of cause and effect, that would put up the number having coronary heart disease to about 153 in every 10,000 people in 11 years.  That’s not a huge difference at all.  If instead they all ate an extra serving every day, rather than just one a week, the number having coronary disease would be about 171 in 10,000, but that’s still quite a low risk and an extra serving a day is a lot of fried food.  But in a different study, of elderly Americans and their consumption of fried fish, a much greater proportion had coronary disease over the 9 years of the study – about 9.5%, or 950 in every 10,000.  Taking the research findings from the new study at face value, that would increase to 969 in every 10,000 if they all ate an extra serving of fried fish each week, or nearly 1,100 in every 10,000 if they all ate one more serving every day (but that’s a lot of fish!).  I hope this illustrates that all this variability between studies and between context mean that the results are really hard to interpret generally.  The researchers also found a 3% relative increase in the risk of a major cardiovascular event associated with one extra serving of fried food, but again in practice that’s not such large increase.

“One final point.  In several of their analyses, the researchers found evidence of publication bias.  That means that it looks as if individual studies that found no significant association with fried food were less likely to be published than studies that did find significant evidence.  (This is, unfortunately, quite a common finding.)  In some of the analyses where this was found, the authors found that attempting to fix the problem by a standard method – essentially by including some theoretical studies that didn’t actually take place – didn’t change the conclusions, so that the publication bias perhaps wasn’t a real problem.  But that didn’t apply to all of these findings.  That’s not the fault of the researchers on this new study – but they could only work with the studies they found, and this is yet another reason why those studies don’t provide anything like a clear picture.”



‘Fried-food consumption and risk of cardiovascular disease and all-cause mortality: a meta-analysis of observational studies’ by Pei Qin et al. was published in Heart at 23:30 UK time on Monday 18 January 2021.

DOI: 10.1136/heartjnl-2020-317883



Declared interests

Prof Riyaz Patel: “No conflicts to declare.”

Prof George Davey Smith: “No COIs.”

Prof Alun Hughes: “No disclosures or conflicts of interest.”

Prof Gunter Kuhnle: “No conflicts to declare.”

Dr Duane Mellor: “With respect to this, I have no declarations of interest.”

Prof Kevin McConway: “I am a Trustee of the SMC and a member of the Advisory Committee, but my quote above is in my capacity as a professional statistician.”

in this section

filter RoundUps by year

search by tag