On 2016 Aug 21, David Keller commented:

Goulden's data actually confirms that minimum mortality occurs with light-to-moderate alcohol intake

Goulden's study concludes: "Moderate alcohol consumption is not associated with reduced all-cause mortality in older adults", a finding he admits is contrary to that of many prior studies. He bases his analysis on a new category he designates as "occasional drinkers", but he gives two differently worded definitions of an "occasional drinker" in different parts of his paper. Goulden states he intends "occasional drinkers" to consume less alcohol than "light drinkers", a standard category of drinkers who consume 1 to 6 standard drinks per week, each containing about 14 grams of ethanol. Unfortunately, both of his definitions of "occasional drinker" can include heavy binge alcohol abusers, clearly not what he intends. By ignoring this new and superfluous group of drinkers, we see that his remaining data confirms that the minimum risk for mortality is associated with light to moderate alcohol intake (the familiar J-shaped curve).

In his reply to my letter, Goulden wrote: "Keller raises the possibility that the 'occasional drinkers' group is, in fact, a group of light drinkers who have under-reported their level of consumption."

Or worse, as we shall see. In addition, both of the definitions he gives for "occasional drinker" do not make physiological sense, are superfluous, and confusing. This new category does not contribute to understanding the data, and it increases the possibility of erroneous classification of drinkers.

In the abstract, Goulden defines an "occasional drinker" as one who reports drinking "at least once, but never more than less than once a week [sic]". In the body of the paper, he defines an "occasional drinker" as one who reports drinking on at least 1 occasion, but always less than once per week. By failing to specify the amount of alcohol consumed on each occasion, Goulden's definitions classify both of the following as "occasional drinkers": a subject who drinks a glass champagne once a year on New Year's eve; and another who drinks an entire bottle of whiskey in one sitting every 8 days. These very different kinds of drinkers are both included in Goulden's definitions of "occasional drinker", by which I think he means those who have tried alcohol at least once, but only drink less than one drink per week. This definition rules out the heavy binge drinker, but it is easier to understand, and thus might reduce errors when classifying drinkers.

Now, for my main point: Look at these hazard ratios (from Table 2) for all-cause mortality with their confidence limits removed for improved visibility, and the "occasional drinker" column removed because of reasons cited above. We are left with 5 columns of data, in 3 rows, which all exhibit a minimum hazard ratio for mortality at <7 drinks per week, which increases when you shift even 1 column to the left or right:

Drinks/week..................zero.....<7.....7-13....14-20...>20

Fully adjusted...............1.19....1.02....1.14....1.13....1.45

Fully adjusted, men..........1.21....1.04....1.16....1.17....1.53

Fully adjusted, women........1.16....1.00....1.13....1.11....1.59

Note that the data in every row approximates a J-shaped curve with the minimum harm ratio in the column labeled <7 [drinks per week], which is light drinking. The next-lowest point is in the 7-13 drinks per week column, or about 2 drinks per day, which is moderate drinking. Although in some instances the confidence intervals overlap, we still have a set of trends which are consistent with past studies, demonstrating the typical J-shaped association between daily ethanol dose and mortality harm ratios. Such trends would likely become statistically significant if the study power were increased enough. The bottom line is that the data tend to support, rather than contradict, the often-observed phenomenon that all-cause mortality is minimized in persons who consume mild-to-moderate amounts of alcohol.

On 2016 Aug 21, David Keller commented:

Goulden's data actually confirms that minimum mortality occurs with light-to-moderate alcohol intake

Goulden's study concludes: "Moderate alcohol consumption is not associated with reduced all-cause mortality in older adults", a finding he admits is contrary to that of many prior studies. He bases his analysis on a new category he designates as "occasional drinkers", but he gives two differently worded definitions of an "occasional drinker" in different parts of his paper. Goulden states he intends "occasional drinkers" to consume less alcohol than "light drinkers", a standard category of drinkers who consume 1 to 6 standard drinks per week, each containing about 14 grams of ethanol. Unfortunately, both of his definitions of "occasional drinker" can include heavy binge alcohol abusers, clearly not what he intends. By ignoring this new and superfluous group of drinkers, we see that his remaining data confirms that the minimum risk for mortality is associated with light to moderate alcohol intake (the familiar J-shaped curve).

In his reply to my letter, Goulden wrote: "Keller raises the possibility that the 'occasional drinkers' group is, in fact, a group of light drinkers who have under-reported their level of consumption."

Or worse, as we shall see. In addition, both of the definitions he gives for "occasional drinker" do not make physiological sense, are superfluous, and confusing. This new category does not contribute to understanding the data, and it increases the possibility of erroneous classification of drinkers.

In the abstract, Goulden defines an "occasional drinker" as one who reports drinking "at least once, but never more than less than once a week [sic]". In the body of the paper, he defines an "occasional drinker" as one who reports drinking on at least 1 occasion, but always less than once per week. By failing to specify the amount of alcohol consumed on each occasion, Goulden's definitions classify both of the following as "occasional drinkers": a subject who drinks a glass champagne once a year on New Year's eve; and another who drinks an entire bottle of whiskey in one sitting every 8 days. These very different kinds of drinkers are both included in Goulden's definitions of "occasional drinker", by which I think he means those who have tried alcohol at least once, but only drink less than one drink per week. This definition rules out the heavy binge drinker, but it is easier to understand, and thus might reduce errors when classifying drinkers.

Now, for my main point: Look at these hazard ratios (from Table 2) for all-cause mortality with their confidence limits removed for improved visibility, and the "occasional drinker" column removed because of reasons cited above. We are left with 5 columns of data, in 3 rows, which all exhibit a minimum hazard ratio for mortality at <7 drinks per week, which increases when you shift even 1 column to the left or right:

Drinks/week..................zero.....<7.....7-13....14-20...>20

Fully adjusted...............1.19....1.02....1.14....1.13....1.45

Fully adjusted, men..........1.21....1.04....1.16....1.17....1.53

Fully adjusted, women........1.16....1.00....1.13....1.11....1.59

Note that the data in every row approximates a J-shaped curve with the minimum harm ratio in the column labeled <7 [drinks per week], which is light drinking. The next-lowest point is in the 7-13 drinks per week column, or about 2 drinks per day, which is moderate drinking. Although in some instances the confidence intervals overlap, we still have a set of trends which are consistent with past studies, demonstrating the typical J-shaped association between daily ethanol dose and mortality harm ratios. Such trends would likely become statistically significant if the study power were increased enough. The bottom line is that the data tend to support, rather than contradict, the often-observed phenomenon that all-cause mortality is minimized in persons who consume mild-to-moderate amounts of alcohol.