Table 6 Results of single studies that examined effects of different levels of alcohol (ineligible for dose-response analysis)

Hogenkamp 2014

Based on 652 men aged 70 years at point of first alcohol measure (T0: 1990).

Alcohol exposure: single assessment (T0: 1990-1994)

Outcome measures: baseline (T0), then follow-up ~7 years later (T1)

Selected outcome: executive function (Trail making test part B)

Difference in mean change from baseline from regression model with alcohol modelled as a continuous variable of grams/day.

Linear term: −0.325; p value = 0.471

Interpretation of the linear term was that the decline in executive function over time (7 years) was less as the dosage of alcohol increased per day. However, this term was not statistically significant.

Lang 2007a

Based on 13,333 men and women (57% female) aged 65 years or above at point of first alcohol measure (T0: 1998).

Alcohol exposure: single assessment at baseline (T0: 1998)

Outcome measures: single assessment ~4 years from baseline (T1: 2002)

Selected outcome: global cognitive function (sum of scores on 3 tests of memory; bottom quintile = poor cognitive function)

Odds ratios (ORs) from logistic regression with alcohol modelled as a categorical variable

Non-drinkers: OR > 1; p value < 0.05

> 0 to ≤ 1 drink/day (referent): 1.00

> 1 to ≤ 2 drinks/day: OR 0.82 (95%CI 0.64, 1.05)

> 2 drinks/day: OR < 1; p value > 0.05

No evidence of a difference in the odds of poor cognitive function in the alcohol consumption categories (> 1 to ≤ 2 drinks/day; > 2 drinks/day) compared with the referent category. Some evidence that non-drinkers had a greater odds of poor cognition compared with the referent category. The relationship was not modified by sex (specific results not reported in primary study).

Piumatti 2018

Based on 13,342 men and women (54.7% female) aged 40-73 years at point of first alcohol measure (T0: 2006).

Alcohol exposure: single assessment (T0: 2006-2010; 2nd assessment not used in prospective analysis)

Outcome measures: baseline (T0), then follow-up ~5 years later. (T1: 2011-2015)

Selected outcome: complex attention (processing speed based on reaction time task. Log reaction time. Higher score = worse cognition)

Predicted difference in log reaction time (milliseconds, ms) from a restricted cubic spline model with alcohol modelled as a continuous variable of log grams/day (outcome):

Linear effect up to 10 g/day (spline 1): −0.048 (log ms)(95%CI −0.105, −0.030); p value < 0.001

Non-linear effect (spline 2): 0.035 (log ms) (95%CI 0.007, 0.059); p value = 0.013

Interpretation of the linear effect up to 16 g/day (spline 1) was that for every 1 standard deviation unit in log grams alcohol/day, there was a predicted -0.048 standard deviation decrease in log reaction time. That is, cognitive performance improved up to 16 g/day. However, cognitive performance started to decline as alcohol consumption increased beyond 16 g/day. The study authors concluded that the relationship was modified by age for the non-linear effect, but was not modified by sex (for either of the effects).

Samieri 2013a

Based on 6174 women aged ≥ 60 at point of first alcohol measure (T0: 1992).

Alcohol exposure: single assessment (T0: 1992-1995)

Outcome measures: 3 assessments—T1 (average of 5.6 years from T0) and then T2 and T3 at ~2-year intervals (T1-T3: 1998-2004)

Total length of follow-up: ~10 years from T0.

Selected outcome: global cognitive function (average of z-scores on 5 tests: overall cognition, language and memory. Higher score=better cognition).

Mean difference (MD) from regression model with alcohol modelled as a categorical variable:

Non-drinker (referent): 0

> 0 to < 15 g/day (median 2.9): MD 0.01 (95%CI −0.03, 0.05)

≥ 15 g/day (median 25.4): MD −0.02 (95%CI −0.10, 0.05)

No evidence of a mean difference in global cognitive function between the different levels of alcohol consumption compared with the referent category of no alcohol. No information on the scale range or standard deviation of the global cognitive function outcome is provided, precluding clinical interpretation.

Solfrizzi 2007

Based on 1445 men and women (44% female) aged 65 to 84 years at point of first alcohol measure (T0: 1992).

Alcohol exposure: single assessment at baseline (T0: 1992)

Outcome measures: baseline (T0), then follow-up ~3.5 years later (T1: 1995–1996)

Selected outcome: Incidence of mild cognitive impairment (MCI; Petersen diagnostic criteria [64])

Hazard ratios (HR) from Cox proportional hazards model with alcohol modelled as a categorical variable:

Categorical model:

No-alcohol (referent): 1.00

≤ 1 drink/day: HR 0.67 (95%CI 0.37, 1.21)

> 1 to ≤ 2 drinks/day: HR 1.27 (95%CI 0.65, 2.47)

> 2 drinks/day: HR 0.85 (95%CI 0.40, 1.81)

Hazard ratios from Cox proportional hazards models with alcohol modelled as a continuous variable of (assumed by review authors) drinks/day (linear only; and linear and quadratic terms):

Linear model:

Linear term: HR 1.08 (95%CI 0.94, 1.24)

Polynomial (quadratic) model:

Linear term: HR 1.06 (95%CI 0.87, 1.28)

Quadratic term: HR 1.00 (95%CI 0.98, 1.02)

No evidence of a difference in the relative rates of MCI in any of the alcohol categories compared to no alcohol consumption, however, the confidence intervals were wide. The relationship was not modified by sex (specific results not reported in primary study).

No evidence of a linear trend between alcohol consumption and the rate of MCI (linear model). No evidence of a linear and quadratic trend between alcohol consumption and the rate of MCI (polynomial model). The relationship was not modified by sex (specific results not reported in primary study).

Wardzala 2018

Based on 486 men and women (75% female) aged ~80 years or above at point of first alcohol measure (T0: 2004 or 2008–09 depending on cohort).

Alcohol exposure: single assessment for most participants at baseline (T0)

Outcome measures: not reported, ~6–7 annual assessments based on time in study (mean 6–8 years).

Total length of follow-up: no information. Assume ~5–7 years from T0

Selected outcome: global cognitive function (mini mental state exam (MMSE). Higher score = better cognition).

Annual rate of change in MMSE from a linear mixed model with alcohol modelled as a categorical variable:

Women:

Rare/never drinkers (referent): annual rate of change < 0 (i.e. MMSE declining over time)

Moderate drinkers: annual rate of change not statistically significantly different (compared with referent category); p value > 0.05

Heavy drinkers: annual rate of change not statistically significantly different (compared with referent category); p value > 0.05

Men:

Rare/never drinkers (referent): annual rate of change < 0 (i.e. MMSE declining over time)

Moderate drinkers: annual rate of change reduced (compared referent category); p value < 0.01

Heavy drinkers: annual rate of change not statistically significantly different (compared with referent category); p value > 0.05

In women, the annual decline in global cognitive function was not found to be statistically significantly different between alcohol consumption categories and the referent category.

In men, the annual decline in global cognitive function was not found to be statistically significantly different between the heavy drinkers and rare/never drinkers; however, it was found to be statistically significantly different between the moderate drinkers and the rare/never drinkers. The rate of cognitive decline was less in moderate drinkers.

The primary study authors provided no clinical interpretation of the results beyond concluding based on statistical significance. Results are depicted in a figure, with some reporting in the text.