Healthy dietary patterns and the risk of individual chronic diseases in community-dwelling adults

It is unclear regarding associations of dietary patterns with a wide range of chronic diseases and which dietary score is more predictive of major chronic diseases. Using the UK Biobank, we examine associations of four individual healthy dietary scores with the risk of 48 individual chronic diseases. Higher Alternate Mediterranean Diet score is associated with a lower risk of 32 (all 8 cardiometabolic disorders, 3 out of 10 types of cancers, 7 out of 10 psychological/neurological disorders, 5 out of 6 digestive disorders, and 9 out of 14 other chronic diseases). Alternate Healthy Eating Index-2010 and Healthful Plant-based Diet Index are inversely associated with the risk of 29 and 23 individual chronic diseases, respectively. A higher Anti-Empirical Dietary Inflammatory Index is associated with a lower risk of 14 individual chronic diseases and a higher incidence of two diseases. Our findings support dietary guidelines for the prevention of most chronic diseases.

A strength and a limitation of the approach is the examination of a wide spectrum of diseases.The advantage is that this allows one to examine the effects of diet more broadly rather than focusing on an individual or a few diseases.The downside is that one cannot do the range of analyses to examine associations in more detail.For example, there could be detection bias for some diseases (e.g., cancers, based on screening, and diseases like cataracts based on diagnostic intensity); latencies/induction periods likely vary for many of the diseases; some diseases like dementia may require a long time lag to reliably see effects; while diet has many plausible links to diseases such as metabolic diseases, for others it is less clear; the list of confounders may be broader for some diseases.There are disease-specific nuances that cannot be examined (e.g., for prostate cancer, most evidence suggests that risk factors are different for clinically aggressive disease than for total disease).The broad view has advantages though what is sacrificed should be discussed because readers interested in a specific disease will focus on that disease.
In regards to validity, two important concerns are 1) measurement error of diet (which would tend to attenuate any true associations), and 2) residual confounding, which can impact findings in less predictable ways.Residual confounding could occur by missing some important confounders and by measurement error in the measured ones.For example, it appears that smoking was controlled as never, past, current.There could be residual confounding, esp for the strongly smoking-related outcomes like lung cancer, based on pack-years, when quit, dose in current smokers.If possible, this can be looked at.An interesting analysis would be to examine never smokers, which would remove any potential confounding by smoking.
Personally, to reduce the amount of data to look at, I would prefer two models, base model with age, sex, and energy intake, and one with everything.I view energy intake not as a confounder per se, but as an important variable to adjust for to reduce measurement error in diet.Also, in part, energy intake is determined by body size and physical activity (e.g., a large, active male would have much higher energy requirement than a small, inactive female) so energy intake (esp in balance) is not a true confounder (independent variable in another causal pathway).At least some comment would be useful in what the different models show (e.g. if HRs don't change much, that might be support for less residual confounding, etc.)As such, 3 models are shown which triples the data presented but the reason why different models are done is not specified.
Did BMI influence the associations between dietary patterns and outcomes?The paper mentions BMI in the text but none of the tables/figures include BMI.BMI could be considered a confounder or a mediator.
I might be cautious between a causal connection between diet and psychologic diseases such as depression.There is a link between mental states and poor diets.For e.g., people in a situation/circumstances or personalities prone to stress/anxiety may have poor diets and then over time, some of these people would be more likely to be diagnosed (e.g., with depression).

Minor comments:
Line 137.The scoring method for EDII as shown in Table S4 is different from what it was originally developed.The original score is a weighted sum of 18 food groups (as continuous variables) multiplied by specific weights.It seems the authors have modified EDII.I think the authors need to emphasize this in the methods and explain why they modified it.Line 50.Second sentence of the Intro is incomplete.Line 52.How much is the total number of deaths increased due to increasing size of older population?Line 184.
-Some dietary patterns include alcohol component.Inclusion of alcohol intake as a covariate might lead to overadjustment.
-The authors need to justify why they include GRS for longevity as a covariate for different diseases.In Table 1, GRS has little variation across dietary score quintiles (as would be expected).Line 299" I thought the Mediterranean diet supported higher fish intake, but here it is stated that lower fish intake contributed to the benefit?Fig1-2.Some associations appeared marginally significant (95% CI includes 1).It's not clear why the authors reported them to be statistically significant even after FDR correction.
It is not clear how many cases were included for each disease.Does total cancer include non-melanoma skin cancer?
The text mentions about average follow-up years for two diseases.I think a supplementary table of median follow-up (and IQR) for each disease will be useful.Line 328.Reference 30 seems to be incorrect.Line 342.While some items may overlap between AMED and ketogenic diet, AMED is not ketogenic.Thus, any advantage of the AMED diet is not related to ketogenesis.Some findings related to the AEDII are likely due to the fact that alcohol is in the score (e.g.association with alcohol use disorder, psychoactive substance abuse; weak effect for lung cancer because alcohol drinkers tend to smoke too).
Reviewer #2 (Remarks to the Author): The paper has well designed; analyses were appropriate.More evidence related healthy diet as a protective factor.However, these scores are based on different criteria, and it is unclear which score best predicts chronic disease risk.There is a lot of information which was well summarised, but we lost the importance of diet in each disease, including all in the same paper.Analyses were adjusted by confounding, but maybe moderator analysis by obesity, education, or other factors specific for each disease, which would be interesting too and it's lost.
Reviewer #1 (Remarks to the Author): This manuscript is well-written and generally interesting.It builds upon previous analyses on dietary patterns and extends to a comprehensive list of diseases.Even though only four dietary patterns were included, they are widely recommended and might be well-accepted by the public.
The paper also provides more evidence for the protective role of healthy diet in preventing several under-studied diseases such as chronic kidney disease, endocrine disorders etc.However, I have some concerns that need to be addressed.

Response:
We thank the Reviewer for his or her positive comments on our manuscript.We agree with the Reviewer that some concerns require attention and we have revised the manuscript accordingly.

Major comments:
Given that all these diseases were included under the umbrella term of "chronic disease", the authors need to clarify how they define "chronic diseases".The introduction includes a nice discussion about healthy ageing.However, it is not clear to me how diseases such as dyspepsia, alcohol use disorder etc. might be related to this general topic.The title suggests an investigation of co-occurrence of two or more chronic diseases, which is different from the aim of present study.
The conclusion highlights prevention of age-related chronic diseases although some diseases under investigation are not necessarily age-related, e.g., schizophrenia.

Response:
We thank the Reviewer for raising this important concern.We agree with the Reviewer that non-age-related chronic diseases such as alcohol use disorder, other psychoactive substance abuse, schizophrenia, migraine, multiple sclerosis, dyspepsia, eczema, irritable bowel syndrome, and inflammatory bowel disease were included in our analyses.

We have added the information regarding these non-age-related diseases:
1][12] Therefore, investigating significant modifiable factors for these nonage-related chronic conditions also holds considerable interest.
We agree with the Reviewer that the title was misleading, and we have revised the title to "Healthy dietary patterns and the risk of individual chronic diseases in community-dwelling adults".

Greater adherence to healthy dietary patterns especially AMED is associated with a lower risk of multiple individual chronic diseases including all CMDs, some cancers, most psychological/neurological disorders, most digestive disorders, respiratory diseases, chronic kidney disease, osteoporosis, eczema, prostate disorders, cataract, and pernicious anaemia. Our findings support dietary guidelines for the prevention of chronic diseases.
A strength and a limitation of the approach is the examination of a wide spectrum of diseases.The advantage is that this allows one to examine the effects of diet more broadly rather than focusing on an individual or a few diseases.The downside is that one cannot do the range of analyses to examine associations in more detail.For example, there could be detection bias for some diseases (e.g., cancers, based on screening, and diseases like cataracts based on diagnostic intensity); latencies/induction periods likely vary for many of the diseases; some diseases like dementia may require a long time lag to reliably see effects; while diet has many plausible links to diseases such as metabolic diseases, for others it is less clear; the list of confounders may be broader for some diseases.There are disease-specific nuances that cannot be examined (e.g., for prostate cancer, most evidence suggests that risk factors are different for clinically aggressive disease than for total disease).The broad view has advantages though what is sacrificed should be discussed because readers interested in a specific disease will focus on that disease.

Response:
We agree with the Reviewer that there are both pros and cons to investigating a broad range of diseases.We have mentioned this as a limitation:

Sixthly, investigating a broad range of chronic diseases offers certain benefits, but is also limited by narrowing the focus to a specific disease (discussion of the mechanisms).
The Reviewer is correct that there may be detection bias for some diseases.We have added this as a limitation: Fourthly, there may be detection bias for some diseases in the UK Biobank.For example,

populations may vary in their likelihood of cancer detection due to differences in screening frequency, whilst cataracts may exhibit varying degrees of severity, but the available inpatient data in the UK might have limitations in accurately distinguishing these degrees. Even though our sensitivity analysis by excluding individuals who developed dementia within the initial four years of follow-up yielded results consistent with the main findings, it is worth considering that dementia could have begun prior to the diet assessment, given that the prodromal phase of dementia can extend over one decade. 55 Evidence suggests risk factors are different for clinically aggressive prostate cancer than for non-aggressive disease. 56 The inpatient and mortality data available in the UK Biobank do not differentiate between aggressive and non-aggressive prostate cancers, potentially introducing a bias into the relationship between dietary patterns and incident prostate cancer.
We also agree with the Reviewer that the list of confounders adjusted for all individual diseases may be broader for some diseases.We have discussed this:

Fifthly, we adjusted for the same confounders including demographic and lifestyle factors, BMI, energy intake, and GRS for longevity across all health conditions (besides lung cancer), which may be broader for some diseases.
In regards to validity, two important concerns are 1) measurement error of diet (which would tend to attenuate any true associations), and 2) residual confounding, which can impact findings in less predictable ways.Residual confounding could occur by missing some important confounders and by measurement error in the measured ones.For example, it appears that smoking was controlled as never, past, current.There could be residual confounding, esp for the strongly smoking-related outcomes like lung cancer, based on pack-years, when quit, dose in current smokers.If possible, this can be looked at.An interesting analysis would be to examine never smokers, which would remove any potential confounding by smoking.

Response:
We agree with the Reviewer that there are measurement errors in diet and mentioned this as a limitation:

Firstly, while the web-based 24-hour dietary assessment tool employed in the UK Biobank study was validated against biomarkers, it is important to acknowledge the potential for measurement errors due to the self-reported nature. However, these measurement errors of diet are more likely to attenuate the true associations.
As the Reviewer suggests, we further adjusted for pack-years, the age stopping smoking, and the number of cigarettes currently smoked daily for lung cancer.In the full model, AMED and AHEI-2010 were inversely associated with the risk of lung cancer.In addition, we have conducted further analysis for the association between dietary patterns and incident lung cancer stratified by smoking (never smokers, current/former smokers).The hazard ratios (95% CI) for lung cancer associated with AMED among never smokers, former smokers, and current smokers were 0.94 (0.84-1.05), 0.81 (0.74-0.88), and 0.78 (0.70-0.88), respectively (raw P-value for interaction=0.0307).The corresponding numbers for AHEI-2010 were 0.93 (0.83-1.04), 0.89 (0.82-0.97), and 0.80 (0.71-0.90), respectively (raw P-value for interaction=0.0806).Given the multiple comparisons, the interactions were not significant after controlling false discovery rate.
Personally, to reduce the amount of data to look at, I would prefer two models, base model with age, sex, and energy intake, and one with everything.I view energy intake not as a confounder per se, but as an important variable to adjust for to reduce measurement error in diet.Also, in part, energy intake is determined by body size and physical activity (e.g., a large, active male would have much higher energy requirement than a small, inactive female) so energy intake (esp in balance) is not a true confounder (independent variable in another causal pathway).At least some comment would be useful in what the different models show (e.g. if HRs don't change much, that might be support for less residual confounding, etc.)As such, 3 models are shown which triples the data presented but the reason why different models are done is not specified.

Response:
We thank the Reviewer for raising this important concern.
We agree with the Reviewer that energy intake is determined by body size and physical activity.In our original analysis, we found the further adjustment for energy intake (Model 3) did not substantially change the associations between dietary patterns and the risk of chronic diseases Therefore, we included two models (Model 1: age, sex, and energy intake; Model 2: full covariates) in the manuscript as suggested by the Reviewer.The results have been updated in Tables S1-S12, and Figures 1-5, S5-S14 as well as the text in the manuscript.
Did BMI influence the associations between dietary patterns and outcomes?The paper mentions BMI in the text but none of the tables/figures include BMI.BMI could be considered a confounder or a mediator.

Response:
We agree with the Reviewer that BMI as an important indicator of metabolic health may influence the associations between dietary patterns and outcomes.In our original analyses, we analyzed obesity (defined by BMI) as a moderator for the association between dietary patterns and incident diseases.As the Reviewer suggests, we have included BMI as a confounder in Model 2. Some associations have been attenuated to be non-significant after the adjustment for BMI.We have updated the results in Tables /Figures.I might be cautious between a causal connection between diet and psychologic diseases such as depression.There is a link between mental states and poor diets.For e.g., people in a situation/circumstances or personalities prone to stress/anxiety may have poor diets and then over time, some of these people would be more likely to be diagnosed (e.g., with depression).

Response:
We agree with the Reviewer that individuals in a stressful situation may have poor diets, which might have biased the association between diet and psychological diseases.However, we did a sensitivity analysis by excluding individuals who developed psychologic diseases in the first four years of follow-up and found that the results for the associations between dietary patterns and incident psychologic diseases were similar to those in the main analyses.This may reduce the risk of bias that the outcome occurred before the exposure assessment.

We cannot rule out the potential reverse causation between diet and psychological diseases as people in a situation or personalities prone to stress/anxiety could potentially adopt unhealthy dietary patterns and thus were more likely to be diagnosed with psychological conditions during follow-up.
Minor comments: Line 137.The scoring method for EDII as shown in Table S4 is different from what it was originally developed.The original score is a weighted sum of 18 food groups (as continuous variables) multiplied by specific weights.It seems the authors have modified EDII.I think the authors need to emphasize this in the methods and explain why they modified it.

Response:
We modified the EDII because it is possible that the EDII score may largely depend on one or two components if individuals consumed too much of these food components.We have added explanations for this in the Methods section:

Scores for the intakes between the maximum and minimum scores were proportionately calculated.
Line 50.Second sentence of the Intro is incomplete.

Response:
We have revised the sentence accordingly: Fig1-2.Some associations appeared marginally significant (95% CI includes 1).It's not clear why the authors reported them to be statistically significant even after FDR correction.

Response:
Although these associations with the upper CIs of the HRs are 1 (actually these numbers are 0.996, or 0.997 or such data), the p-values are smaller than 0.05.It is the problem regarding decimal spaces.For those 95% CIs containing 1, we have revised the results to include values with three or more decimal places.
It is not clear how many cases were included for each disease.Does total cancer include nonmelanoma skin cancer?

Response:
We have included the number of each disease in figures 1-5, S5-S14.The total cancer does not include non-melanoma skin cancer.We have added the information in the footnotes of figures 2, 6, S1-S4, S6, and S11, and Tables S1-S12.
The text mentions about average follow-up years for two diseases.I think a supplementary table of median follow-up (and IQR) for each disease will be useful.

Response:
The incidence of dyspepsia as the most incident disease was 9.5% such that the median duration of follow-up exhibited slight variations (ranges from 8.64-8.66years).That is why we showed the average follow-up duration for the two diseases with the largest difference.For the effect size and overall disease burden, we have added the number of event cases and participants as well as incidence for each disease in Figures 1-5, and S5-S14.We have also added the description of the effect size and disease burden in the Results section.

Response:
According to the statistical convention, stratified analyses can be conducted if the interaction is significant.Given the multiple comparisons, we just displayed the results with significant interactions after controlling FDR.For those without significant interactions, the associations between dietary patterns and incident diseases were similar across subgroups.
We have added more results from subgroups analyses accordingly (new Figures S1-S4).
Line 328.Reference 30 seems to be incorrect.

Response:
We have checked the reference.
Line 342.While some items may overlap between AMED and ketogenic diet, AMED is not ketogenic.Thus, any advantage of the AMED diet is not related to ketogenesis.

Response:
We agree with the Reviewer on this point, and we have revised the explanation for the inverse association between AMED and epilepsy risk:

A diet rich in antioxidants and anti-inflammatory compounds, as found in the Mediterranean diet, may contribute to reducing inflammation and oxidative stress in the brain, 46,47 which are risk factors for epilepsy. This may partly explain why we found an inverse association between AMED and incident epilepsy.
Some findings related to the AEDII are likely due to the fact that alcohol is in the score (e.g.association with alcohol use disorder, psychoactive substance abuse; weak effect for lung cancer because alcohol drinkers tend to smoke too).

Response:
We have explained this in the Discussion section:

The positive association between AEDII and the risk of alcohol use disorder could be explained by the substantial role of alcohol consumption as a key component of AEDII.
We did not find a significant association between AEDII and the risk of lung cancer.
Reviewer #2 (Remarks to the Author): The paper has well designed; analyses were appropriate.More evidence related healthy diet as a protective factor.

Response:
We thank the Reviewer for his or her positive comments on our manuscript.
However, these scores are based on different criteria, and it is unclear which score best predicts chronic disease risk.

Response:
We agree with the Reviewer that our statement was not clear regarding the best score for the prediction of chronic diseases.AMED was linked to the largest number of chronic diseases and yielded the lowest risk for most chronic diseases.We have added the information in the first paragraph of the Discussion section: In this large cohort study, we found a higher AMED score was associated with a lower risk of 32 (all 8 CMDs, 3 out of 10 types of cancers, 7 out of 10 psychological/neurological disorders, 5 out of 6 digestive disorders, and 9 out of 14 other chronic diseases) out of 48 chronic diseases.

AHEI-2010 was inversely associated with the risk of 29 chronic diseases (7 CMDs, 4 cancers, 5 psychological/neurological disorders, 5 digestive disorders, and 8 other chronic diseases). A higher HPDI score was associated with a reduced risk of 23 chronic diseases (6 CMDs, 4 cancers, 4 psychological/neurological disorders, 5 digestive disorders, and 4 other chronic diseases). No positive associations between AMED, AHEI-2010, and HPDI and the risk of any chronic disease were observed. AEDII was inversely associated with the risk of 14 chronic diseases and positively associated with the risk of two chronic conditions (alcohol use disorder, psychoactive substance abuse). AHEI-2010 demonstrated the lowest risk for alcohol use
disorder and psychoactive substance abuse, AEDII showed the lowest risk for diabetes and thyroid disorders, while AMED yielded the lowest risk for many other chronic diseases (CVD,

cancer, COPD, CKD, chronic liver disease, psychological/neurological disorders, and digestive disorders).
There is a lot of information which was well summarised, but we lost the importance of diet in each disease, including all in the same paper.

Response:
We thank the Reviewer for raising this important concern.We agree with the Reviewer that there are both pros and cons to investigating a broad range of diseases.We have mentioned this as a limitation:

Sixthly, investigating a broad range of chronic diseases offers certain benefits, but is also limited by narrowing the focus to a specific disease (discussion of the mechanisms).
Meanwhile, we have added more statements regarding which score yielded the lowest risk for each disease.
Analyses were adjusted by confounding, but maybe moderator analysis by obesity, education, or other factors specific for each disease, which would be interesting too and it's lost.

Response:
Line 204.I would appreciate a transparent reporting of number of hypotheses being considered.Line 230-235: These numerical values are getting crowded.If you have included them in figures, you don't need to mention all of them.

Fig 6 .
Fig 6.I wonder why the authors only look at components of alternate Mediterranean diet.I can see that AMED was inversely associated with the largest number of diseases in this paper.Did the authors also consider effect size and overall disease burden when interpreting the results?

Fig 6 .
Fig 6.I wonder why the authors only look at components of alternate Mediterranean diet.I can see

Fig
Fig S1-4.I think all results from subgroup analyses should be presented.