On 2016 Aug 23, David Keller commented:

Results are misleadingly presented; mortality is reduced with moderate alcohol consumption

The Results section of the above abstract misleadingly states:

"The hazard ratio and 95% confidence interval in fully adjusted analyses was 1.02 (0.94-1.11) for <7 drinks/week, 1.14 (1.02-1.28) for 7 to <14 drinks/week, 1.13 (0.96-1.35) for 14 to <21 drinks/week, and 1.45 (1.16-1.81) for ≥ 21 drinks/week."

The above quote falsely implies that all amounts of alcohol consumption increased mortality, either with statistical significance, or at least by trend (depending on whether the confidence interval for a Hazard Ratio crosses 1.0).

These results are from line 5 of Table 2 of this paper, which gives the fully-adjusted results for all study participants. They are misleading, as presented, for two reasons. First, they are normalized by the Hazard Ratio of a newly-defined category called "occasional drinkers", which is a flawed and erroneously defined category of drinkers, for reasons I detail elsewhere [1]. Second, a very important data point has been omitted from these Results, namely the Hazard Ratio for non-drinkers, which is 1.19 (1.11-1.27). Why is the Hazard Ratio for non-drinkers elevated? Because it is normalized by the Hazard Ratio for "occasional drinkers", a statistical maneuver which introduces errors and obscures the true relationship of mortality with alcohol intake.

Thus informed, we see that the non-drinker can lower his Hazard Ratio for all-cause mortality from 1.19 (1.11-1.27) to 1.02 (0.94-1.11) by starting the light consumption of alcohol, drinking <7 standard alcoholic beverages per week. The confidence intervals for the Hazard Ratios of non-drinkers and light drinkers touch at 1.11, but do not overlap, so this is a significant reduction of mortality.

Again, an average non-drinker can significantly lower their risk of all-cause mortality by adding one standard 14 gram serving of ethanol per day, preferably in a dilute form such as beer (to avoid carcinogenic effects on the upper aerodigestive tracts [2]).

References

1: Keller DL, Goulden's data actually confirms that minimum mortality occurs with light-to-moderate alcohol intake, PubMed Commons, accessed on 8/22/2016 at the following URL:<br> http://www.ncbi.nlm.nih.gov/pubmed/27453387#cm27453387_26107

2: Keller DL. Dose-response relationship observed between concentration of ingested alcohol and cancer rate. Comment on PMID 26386538. In PubMed Commons [Internet]. National Library of Medicine; 2015 Sept 26 [cited 2015 Oct 12] at: http://www.ncbi.nlm.nih.gov/pubmed/26386538#cm26386538_11980 The above comment is also posted on the following Annals of Internal Medicine web page: http://annals.org/article.aspx?articleid=2456121

On 2016 Aug 30, Robert Goulden commented:

Thank you David - I think that's a fair summary of our discussion, and you usefully highlight areas where the evidence can reasonably be interpreted in different ways. I hope our exchange will be read by those interested in the article and that will enrich their understanding of this complex question.

On 2016 Aug 30, David Keller commented:

A reader disagrees with the conclusion of this study

The author's conclusion that "moderate alcohol consumption is not associated with reduced all-cause mortality" conflicts with prior studies, which have reported reduced all-cause mortality with moderate alcohol consumption, compared with non-consumption, or heavy consumption. I noted that the data in this study actually show minimum mortality for subjects who drank from 1 to 6 drinks per week, compared with non-drinkers or for heavier drinkers.

The author replied with this explanation: "You’re absolutely correct that mortality is lower in regular moderate alcohol consumers than in long-term abstainers. My paper does not contradict the existing literature on this point. The important question, however, is whether that difference is due to alcohol consumption, or due to other (confounding) factors. If it’s due to confounding factors (which I argue it is) then the comparison is not especially helpful – in fact, it’s misleading."

So, despite the title of this study, it actually confirms prior reports of significant reduction in mortality associated with moderate alcohol drinking, compared with non-drinking or heavy drinking. The author argues that this reduced mortality is not due to alcohol consumption itself, but to unidentified "confounding factors".

Occam's razor is the principle that favors the simplest explanation for an observed phenomenon, when more complicated theories exist in equipoise. For example, Occam's razor favored Newton's laws of motion until they could no longer explain all of the experimental data, at which time the more complicated Theory of Relativity was called for.

The simplest explanation for the results observed in this and prior studies is that moderate alcohol consumption is the cause of the associated reduced mortality in the populations studied. A randomized, placebo-controlled experiment is required to test the more complicated theory that some unknown confounding factor is responsible for the mortality reduction associated with moderate drinking. Meanwhile, Occam's Razor favors consideration of reduced overall mortality when discussing the potential beneficial and adverse effects of moderate alcohol consumption.

On 2016 Aug 29, Robert Goulden commented:

Thanks David. As suggested; here's those two emails:

You’re absolutely correct that mortality is lower in regular moderate alcohol consumers than in long-term abstainers. My paper does not contradict the existing literature on this point. The important question, however, is whether that difference is due to alcohol consumption, or due to other (confounding) factors. If it’s due to confounding factors (which I argue it is) then the comparison is not especially helpful – in fact, it’s misleading. In contrast, comparing moderate drinkers to occasional drinkers is much more informative in helping us figure out if alcohol reduces mortality. These groups (occasionals vs. moderates) are much more similar across a range of variables than abstainers vs. moderates, and thus any differences in mortality can more plausibly attributed to the effects of alcohol. The lack of a mortality difference between those who drink less than once per week, versus those who drink around 1 drink per day, surely calls into question a beneficial effect of alcohol? Of course, people may not be reporting their consumption levels accurately, but in that sense my study suffers the same weakness as just about every other study on this question. Hence the need for an RCT to settle this question once and for all; I sincerely hope somebody gets round to organising one sooner rather than later!

The reason I think the benefit [of moderate consumption vs. long-term abstention] survives [adjustment] is that there is significant residual confounding. This is a judgement – by definition you cannot measure unmeasured confounding – but which I think is supported by the lack of any difference between the occasional and moderate drinkers. As the Naimi et al. paper from 2005 suggests, the differences between non-drinkers and drinkers is so marked across a huge numbers of variables that it’s likely that no amount of adjustment can truly balance out the groups. Despite my extensive adjustments, in reality I’m only scratching the surface of the thousands of potential variables which affect people’s health outcomes. This gets to the heart of why observational studies, even based on pretty good quality data like the Health and Retirement Study, can only take us so far.

On 2016 Aug 29, David Keller commented:

Thank you for your outstanding replies to my questions, including your last two private emails to me, which I found very helpful and would urge you to consider posting. PubMed Commons does not permit the recipient of an email to post it, but the author can do so. I do not think anyone will object, however, if I quote your final sentence:

Hence the need for an RCT [randomized controlled trial] to settle this question once and for all; I sincerely hope somebody gets round to organising one sooner rather than later!

I concur completely, until which time we will have to agree to disagree regarding the benefits of moderate alcohol consumption.

A votre sante!

On 2016 Aug 27, Robert Goulden commented:

Hi Dr Keller,

One of the key methodological feature of my paper - in addition to extensive confounder adjustment and measurement of consumption at several different time points - is the use of occasional drinkers as the reference group. For the reasons I've outlined in the paper and in my previous comment, I believe they are a more appropriate comparator group than long-term abstainers, because the comparison is less likely to be subject to confounding. In contrast, the HR for long-term abstainers is likely subject to significant unmeasured confounding and I therefore felt should be left out of the abstract, where there is insufficient space to explain its limitations. To an extent that's a value judgement, but I think a reasonable one.

If my abstract had not mentioned that occasional drinkers were the reference group, I think that would have been grounds to publish a correction to the abstract (though not to retract the paper!). But given that I've clearly stated that occasional drinkers are the reference group, I'm not sure any change is warranted.

In a separate comment you raise a very valid point about who the 'occasional drinkers' are. You correctly suggest that someone who binges on a bottle of whisky once every 8 days would fall into that category. This reflects the structure of the questioning in the Health and Retirement Study (see question C129 onwards in the script used by interviewers), whereby those who drink 'less than once per week' are not probed for further details on how much they drink. My guess is that such drinkers are sufficiently rare to not significantly affect the results, but even if they weren't, they would presumably make the occasional drinkers group less healthy (if we can all agree a bottle of whisky every 8 days is probably bad for you!). Therefore, in so far as there are bingers among the occasional drinker group, the direction of bias would be to favour the regular light drinker group. Instead, they are almost indistinguishable (HR 1.02) from the occasional drinkers.

I hope this addresses your questions, but am happy to continue the conversation.

Regards

Rob

On 2016 Aug 27, David Keller commented:

Dear Professor Goulden,

Thanks for your helpful reply to my comments. PubMed Commons is a unique forum for authors to fully explain and defend their work, and readers to make their best effort to question results or methods.

My main point has still not been answered. Line 5 of your Table 2 indicates that the Hazard Ratio for all-cause mortality for non-drinkers was 1.19 (1.11-1.27), and for light drinkers, it was 1.02 (0.94-1.11). The associated confidence intervals touch but do not overlap. Does this not indicate a significant decrease in the risk of all-cause mortality, associated with the light consumption of alcohol? If so, then how can you conclude that "Moderate alcohol consumption is not associated with reduced all-cause mortality" in the population you studied? Your own fully adjusted data tells us that a significant decrease in the Hazard Ratio for all-cause mortality, from 1.19 to 1.02, is associated with drinking 1-6 alcoholic beverages per week. Is this not an example of alcohol consumption which is indeed associated with reduced all-cause mortality? If so, shouldn't a retraction be issued?

David L. Keller, MD, FACP

On 2016 Aug 27, Robert Goulden commented:

Hi David

Many thanks for your comments. Your key objection, in the comment here, your letter to Am J Med, and your PubMed commons comment on my response to that letter, is around the use of occasional drinkers as the reference category. You are completely right that this use of a reference category is the principal (but not sole) reason that I don’t find evidence of a benefit to moderate alcohol consumption. I had hoped that the use of this reference category was adequately explained in the paper, but it’s an important point so I’m happy to discuss it further.

Studies of the association between alcohol consumption and health are plagued with the problem of confounding. Non-drinkers and moderate drinkers differ in myriad important ways which conventional regression analyses cannot adequately adjust for (1). This leaves us with the difficult question of how to isolate the effects of alcohol on health, as opposed to the effect of all the other health-related variables which differ between non-drinkers and drinkers. One proposed solution is to use occasional drinkers as the reference category – not a novel “statistical maneuver” developed by me – but an approach used in the largest ever study of this question (2) (they were called light drinkers in that study, but the volume of alcohol consumed [0-2 g/day] meant they were occasional drinkers, as the accompanying editorial noted (3)). As I say in my paper, occasional drinkers “drink at levels for which a physiologic effect of alcohol is not plausible, but are likely to be more similar in other characteristics to moderate drinkers than long-term abstainers, thus reducing confounding”.

When this approach to addressing confounding is taken, my results are actually consistent with the wider literature, as Stockwell and Naimi note in their commentary on my paper (4). Their systematic review (5) reported “similar findings” and had my paper been included in their meta-analysis (it was published after their search window) it would have been coded as “high quality”. Other approaches which try to isolate the causal effect of alcohol and minimize confounding, such as mendelian randomization, also find no evidence of a benefit of moderate alcohol consumption (6).

Making sense of all the evidence is tricky, but my gut instinct (for what it's worth!), based on my paper and the wider literature, is that moderate alcohol consumption (up to 21 drinks per week) likely has very little effect (positive or negative) on health. My paper only finds unambiguous evidence of harm for those drinking over 21 drinks per week; of course, whether that association is driven by residual confounding is hard to say, but it’s certainly consistent with well-established links between heavy alcohol use and adverse outcomes such as liver disease, certain cancers, and trauma.

I think the claim that my abstract is ‘misleading’ isn’t warranted; I explicitly state in the abstract that occasional drinkers are the reference category, but by necessity this choice can only be fully explained in the main text.

Finally, you make some explicit causal claims about alcohol’s effects on health which I think go beyond what the observational literature can tell us. You state “the non-drinker can lower his Hazard Ratio for all-cause mortality…by starting the light consumption of alcohol” and “an average non-drinker can significantly lower their risk of call-cause mortality by adding one standard 14 gram serving of ethanol per day”. Until we have an RCT to demonstrate this, I think claims about what alcohol can and cannot do should be made more cautiously. If such a study were performed, and indeed showed benefit, I’d be the first to raise a glass to the results!

Regards,

Rob

References

1: Naimi TS, Brown DW, Brewer RD, Giles WH, Mensah G, Serdula MK, et al. Cardiovascular risk factors and confounders among nondrinking and moderate-drinking U.S. adults. Am J Prev Med. 2005 May;28(4):369–73.

2: Bergmann MM, Rehm J, Klipstein-Grobusch K, Boeing H, Schütze M, Drogan D, et al. The association of pattern of lifetime alcohol use and cause of death in the European Prospective Investigation into Cancer and Nutrition (EPIC) study. Int J Epidemiol. 2013 Dec 1;42(6):1772–90.

3: Stockwell T, Chikritzhs T. Commentary: Another serious challenge to the hypothesis that moderate drinking is good for health? Int J Epidemiol. 2013 Dec 1;42(6):1792–4.

4: Stockwell T, Naimi T. Study raises new doubts regarding the hypothesised health benefits of “moderate” alcohol use. Evid Based Med. 2016 Jul 7;ebmed-2016-110407.

5: Stockwell T, Zhao J, Panwar S, Roemer A, Naimi T, Chikritzhs T. Do “Moderate” Drinkers Have Reduced Mortality Risk? A Systematic Review and Meta-Analysis of Alcohol Consumption and All-Cause Mortality. J Stud Alcohol Drugs. 2016 Mar;77(2):185–98.

6: Holmes MV, Dale CE, Zuccolo L, Silverwood RJ, Guo Y, Ye Z, et al. Association between alcohol and cardiovascular disease: Mendelian randomisation analysis based on individual participant data. BMJ. 2014;349:g4164.

On 2016 Aug 23, David Keller commented:

Results are misleadingly presented; mortality is reduced with moderate alcohol consumption

The Results section of the above abstract misleadingly states:

"The hazard ratio and 95% confidence interval in fully adjusted analyses was 1.02 (0.94-1.11) for <7 drinks/week, 1.14 (1.02-1.28) for 7 to <14 drinks/week, 1.13 (0.96-1.35) for 14 to <21 drinks/week, and 1.45 (1.16-1.81) for ≥ 21 drinks/week."

The above quote falsely implies that all amounts of alcohol consumption increased mortality, either with statistical significance, or at least by trend (depending on whether the confidence interval for a Hazard Ratio crosses 1.0).

These results are from line 5 of Table 2 of this paper, which gives the fully-adjusted results for all study participants. They are misleading, as presented, for two reasons. First, they are normalized by the Hazard Ratio of a newly-defined category called "occasional drinkers", which is a flawed and erroneously defined category of drinkers, for reasons I detail elsewhere [1]. Second, a very important data point has been omitted from these Results, namely the Hazard Ratio for non-drinkers, which is 1.19 (1.11-1.27). Why is the Hazard Ratio for non-drinkers elevated? Because it is normalized by the Hazard Ratio for "occasional drinkers", a statistical maneuver which introduces errors and obscures the true relationship of mortality with alcohol intake.

Thus informed, we see that the non-drinker can lower his Hazard Ratio for all-cause mortality from 1.19 (1.11-1.27) to 1.02 (0.94-1.11) by starting the light consumption of alcohol, drinking <7 standard alcoholic beverages per week. The confidence intervals for the Hazard Ratios of non-drinkers and light drinkers touch at 1.11, but do not overlap, so this is a significant reduction of mortality.

Again, an average non-drinker can significantly lower their risk of all-cause mortality by adding one standard 14 gram serving of ethanol per day, preferably in a dilute form such as beer (to avoid carcinogenic effects on the upper aerodigestive tracts [2]).

References

1: Keller DL, Goulden's data actually confirms that minimum mortality occurs with light-to-moderate alcohol intake, PubMed Commons, accessed on 8/22/2016 at the following URL:<br> http://www.ncbi.nlm.nih.gov/pubmed/27453387#cm27453387_26107

2: Keller DL. Dose-response relationship observed between concentration of ingested alcohol and cancer rate. Comment on PMID 26386538. In PubMed Commons [Internet]. National Library of Medicine; 2015 Sept 26 [cited 2015 Oct 12] at: http://www.ncbi.nlm.nih.gov/pubmed/26386538#cm26386538_11980 The above comment is also posted on the following Annals of Internal Medicine web page: http://annals.org/article.aspx?articleid=2456121

On 2016 Aug 30, David Keller commented:

A reader disagrees with the conclusion of this study

The author's conclusion that "moderate alcohol consumption is not associated with reduced all-cause mortality" conflicts with prior studies, which have reported reduced all-cause mortality with moderate alcohol consumption, compared with non-consumption, or heavy consumption. I noted that the data in this study actually show minimum mortality for subjects who drank from 1 to 6 drinks per week, compared with non-drinkers or for heavier drinkers.

The author replied with this explanation: "You’re absolutely correct that mortality is lower in regular moderate alcohol consumers than in long-term abstainers. My paper does not contradict the existing literature on this point. The important question, however, is whether that difference is due to alcohol consumption, or due to other (confounding) factors. If it’s due to confounding factors (which I argue it is) then the comparison is not especially helpful – in fact, it’s misleading."

So, despite the title of this study, it actually confirms prior reports of significant reduction in mortality associated with moderate alcohol drinking, compared with non-drinking or heavy drinking. The author argues that this reduced mortality is not due to alcohol consumption itself, but to unidentified "confounding factors".

Occam's razor is the principle that favors the simplest explanation for an observed phenomenon, when more complicated theories exist in equipoise. For example, Occam's razor favored Newton's laws of motion until they could no longer explain all of the experimental data, at which time the more complicated Theory of Relativity was called for.

The simplest explanation for the results observed in this and prior studies is that moderate alcohol consumption is the cause of the associated reduced mortality in the populations studied. A randomized, placebo-controlled experiment is required to test the more complicated theory that some unknown confounding factor is responsible for the mortality reduction associated with moderate drinking. Meanwhile, Occam's Razor favors consideration of reduced overall mortality when discussing the potential beneficial and adverse effects of moderate alcohol consumption.