Skip to main content

Table 2 Characteristics of included papers (n = 16)

From: Are there non-linear relationships between alcohol consumption and long-term health?: a systematic review of observational studies employing approaches to improve causal inference

First author (year) Health outcome/s and interval to follow-up Study design Participant characteristics Cohort/s name Sample size Statistical methodology
(reference group bolded)
Explicit testing for non-linearity? Main results (statistically significant findings bolded) Conclusion
Cancer
Dickerman (2016) [43] Prostate cancer (PC) and prostate cancer mortality
-median 30 yrs. of follow-up
Twin Male twins; mean age 40.1 at baseline Older Finnish Twin Cohort 11,372;
225 and 43 discordant twin pairs for PC and PC-mortality respectively
-Co-twin (discordant for both alcohol consumption level and either time to diagnosis among outcome-concordant pairs concordant or time to event vs death/end of follow-up among outcome-discordant pairs) and pooled cohort Cox analyses to examine risk for PC and PC-mortality
-Alcohol consumption measured twice (6 years apart) and averaged; categories: abstainers, light (0.01–3 drinks/wk), moderate (> 3–14 drinks/wk), heavy (> 14 drinks/wk); 1 drink considered = 12 g of alcohol
ϰ -PC-risk:
HR (95% CI) Cohort MZ twins DZ twins All twins
Abstainers 1.27 (.94,1.71) 2.85 (.67,12.1) 3.80 (1.36,20.6) 2.98 (1.35,6.60)
Moderate 1.2 (.99,1.46) 1.28 (.6,2.74) 1.54
(0.92,2.57) 1.36 (.91,2.04)
Heavy 1.46 (1.12,1.91) 2.00(.62,6.45) 1.71 (.87,3.39) 1.63 (.92,2.88)
-PC-mortality:
HR (95% CI) Cohort MZ twins DZ twins All twins
Abstainers 1.90 (1.04,3.47) 2.31 (.19, 27.4) 1.83 (.42,8.03) 1.37 (.44,4.28)
Moderate 1.22 (.76,1.97) 9.13 (.70,119) 1.43 (.37,5.62) 2.44 (.79, 7.52)
Heavy 1.32 (.66,2.62) - 2.39 (.33,17.3) 7.31 (1.3, 41)
Reverse J-shaped relationships, with light drinking at the nadir, were evident in all twin analyses of PC risk. J-shaped curves were evident in twin analyses of PC mortality. The evidence is consistent with a causal protective effect of light drinking, but twin analyses in general lacked power.
Diabetes
Carlsson (2003) [44] Type 2 diabetes (T2D)
−20 yrs. of follow-up
Twin Same-sex twins without diabetes at baseline; mean age 34.3 (M) and 35.4 (F) at baseline Finnish Twin Cohort 22,788; 27 discordant twin pairs analysed -Co-twin (discordant for alcohol category) and pooled cohort Cox analyses to examine risk for T2D
-Categories for pooled cohort (based on 3 questionnaires over 15 years): abstainers, < 5 g/day, 5–19.9 g/day (women)/5–29.9 g/day (men), ≥ 20 g/day (women)/ ≥ 30 g/day (men); for twin analyses (based on baseline questionnaire only): low (< 5 g/day), moderate (15–29.9 g/day), high (≥ 30 g/day)
RR (95% CI) Cohort Men Cohort Women OR (95% CI) Twin
Abstainers 1.1 (.7,1.5) 1.1 (.9,1.5) Moderate .5 (.2,1.3)
5–29.9 g/day (m); 5–19.9 g/day (f) .8 (.6,1.1) .7 (.4,1.1) High 1.2 (.4,3.9)
≥ 30 g/day (m); ≥ 20 g/day (f) .9 (.6,1.4) 1.6 (.8,3.5)
Twin analyses were critically underpowered, preventing causal inference.
Peng (2019) [45]
*See also cardiovascular outcomes
Diabetes-related biomarkers (FBG, P2hBG, HbA1c, HOMA-IR, HOMA-beta)
-cross-sectional
MR Chinese adults living in the Yi-Ling district of Yichang; mean age 55 (SD 0.1) at baseline One community from the Risk Evaluation of cAncers in Chinese diabeTic Individuals: a LONgitudinal (REACTION) study 4536 −1-sample MR using a single variant and standard IV analysis (2SLS); local average treatment effects (LATEs) computed for subgroups of observed alcohol consumption (non-zero LATE slopes indicate non-linearity)
-Only standard, linear IV analysis was performed for categorical diabetes risk, not allowing for detection of non-linearity
-Analysis performed in women as a negative control due to their lack of alcohol consumption
-No conventional analyses conducted for comparison
-Non-linear analyses: the LATE slopes for all diabetes-related biomarkers were not sig. Different to zero, indicating no non-linear relationships
Unstandardizedβ (95% CI) Using log-transformed alcohol intake
FBG (log-transformed) -.01 (−.04,.01)
P2hBG (log-transformed) .01 (−.04,.06)
HbA1c (log-transformed) -.01 (−.03,.00)
HOMA-IR (log-transformed) -.00 (−.09,.08)
HOMA-beta (log-transformed) .03 (−.04, .11)
-Standard IV analyses:
Unstandardizedβ (95% CI)Per 1-unit increase in log-transformed genetically-predicted alcohol consumption
FBG (log-transformed).04 (.02,.05)
P2hBG(log-transformed).07 (.04,.11)
HbA1c (log-transformed).00 (−.01,.01)
HOMA-IR(log-transformed).10 (.04, .17)
HOMA-beta (log-transformed).00 (−.05,.06)
No evidence of non-linear relationships between genetically predicted alcohol consumption and diabetes-related biomarkers. Alcohol appears to raise FBG, P2hBG and HOMA-IR in a linear fashion in males, although these are small effects.
Dementia
Handing (2015) [46] Dementia
−43 yrs. of follow-up
Twin Twins without dementia prior to baseline; <=65 at baseline and > =60 at study end; mean age 54.2 (SD 5.9) at baseline; reported drinking <=100 g/day of alcohol Swedish Twin Registry 12,362;
576 dementia discordant pairs (177 MZ); 396 concordant pairs (160 MZ)
-Co-twin (discordant for dementia) logistic regressions and pooled cohort Cox models used to examine risk for dementia; co-twin (concordant for dementia) mixed-effects analyses used to examine age at onset
-Categories for pooled cohort + twin age of onset analyses: abstainers, light (> 0 and ≤ 5 g/d), moderate (> 5 and ≤ 12 g/d), heavy (> 12 and ≤ 24 g/d), and very heavy (> 24 g/d). For twin dementia risk analyses: abstainers, light (> 0 and ≤ 5 g/d), and moderate-to-very heavy (> 5 g/d)
-Dementia risk:
HR (95% CI) Cohort OR (95% CI) MZ twins All twins
Abstainers 1.05 (1.00,1.11) Abstainers 1.37 (.6,3.16) 1.39 (.89,2.16)
Moderate .98 (.92–1.04) Moderate-to-very heavy 3.07 (1.37,6.86) 1.57 (1.04,2.37)
Heavy 1.1 (1.01,1.19)
Very heavy 1.18 (1.01,1.36)
-Dementia age at onset:
Mean difference in years between twin diagnosed first vs twin diagnosed later (p value) MZ twins All twins
Abstainers − 5.37 (.83) -5.49 (.68)
Light −6.28 (NA – reference group) -6.79 (NA – reference group)
Moderate − 5.41 (.13) -7.00 (.98)
Heavy − 6.33 (.99) -6.73 (.32)
Very heavy − 12.67 (.09) -10.67 (.02)
Results are consistent with a J-shaped curve. Twin analyses do support a causal role for increased risk of dementia for moderate-to-very heavy drinking, and faster dementia onset for very heavy drinkers.
Mental health
Gemes (2019) [47] Depression
− 7-9 yrs. of follow-up
MSM General population aged 20–64 at recruitment; mean age 43.3 (SD 12.2) at baseline Psykisk Hälsa–Arbete–Relationer (PART) cohort (Sweden) 5087 -MSM (weighted logistic regression) + standard logistic regression for comparison
-Alcohol consumption measured pre-baseline (for use as time-variant confounder) and baseline; categories: no consumption, light (> 0- ≤ 7 drinks/wk), moderate (> 7- ≤ 14 drinks/wk), and excessive (> 14 drinks/wk); 1 drink = 12 g
-Adjusting for baseline MDI score:
RR (95% CI) Non-MSM MSM
Abstainers 1.10 (.69,1.74) 1.60 (1.27,2.01)
Moderate .54 (.28,1.04) 1.05 (.83,1.32)
Excessive .61 (.21,3.07) 1.77 (1.13,2.78)
-Excluding those with baseline depression (MDI > 26):
RR (95% CI) Non-MSM MSM
Abstainers 1.24 (.72,2.14) 1.46 (1.10,1.95)
Moderate .63 (.29,1.41) .75 (.55,1.03)
Excessive 2.72 (.61,12.19) 2.83 (1.80,4.44)
MSM results support a non-linear (U/J-shaped) relationship between alcohol consumption and depression.
Samuelsson 2013 [48] a Disability pension (DP) due to mental health diagnoses (MHD)
-median 10 yrs. of follow-up from prior study (baseline)
Twin Twins with data from a prior study, and at time of that study were living in Sweden, < 65, and without DP/old age pension; mean age 52.9 (SD 5.6) Swedish Twin Study of Disability Pension and Sickness Absence (STODS), drawn from the Swedish Twin
Registry
28,613; 229 DP-discordant twin pairs (95 MZ pairs) -Co-twin (discordant for dementia) Cox regressions + pooled cohort Cox regressions performed
-Differentiated between frequent and infrequent drinkers (have/not consumed alcohol in previous two months); categories for pooled cohort + twin analyses: abstainers, light frequent (12-84 g/wk), moderate frequent (M: 85-168 g/wk.; F: 85–108 g/wk) and ≤ 12 g/d), heavy frequent (M: > 168 g/wk.; F: > 108 g/wk), light infrequent (12 g/occasion), moderate infrequent (M:13-48 g/occasion; F: 13-36 g/occasion), heavy infrequent (M: > 48 g/occasion; F:> 36 g/occasion)
HR (95% CI) Cohort MZ twins DZ twins All twins
Abstainers 1.99 (1.57,2.54) 1.93 (.54,6.96) 2.63 (1.05,6.62) 2.17 (1.06,4.45)
Moderate frequent 1.07 (.78,1.49) .46 (.11,1.87) 2.70 (.81–9.02) 1.24 (.53,2.91)
Heavy frequent .98 (.61,1.54) - .46 (.09,2.36) .48 (.12,1.90)
Light infrequent 1.09 (.69,1.73) 2.80 (.24,32.6) 3.98 (.71,22.4) 3.67 (.91,14.8)
Moderate infrequent 1.18 (.91,1.54) .52 (.18,1.49) 1.71 (.75,3.91) 1.03 (.55,1.93)
Heavy infrequent 1.20 (.92,1.57) 1.09 (.33,3.53) 3.61 (1.28,10.2) 2.10 (.99,4.46)
Results support a non-linear relationship between alcohol consumption and DP due to MHD, such that abstainers are at increased risk compared to light frequent drinkers. Light and heavy infrequent drinkers are also at increased risk, but CIs are very wide for the former, and effect disappears for the latter in MZ-only twins.
Cardiovascular events/diagnoses
Ilomaki (2011) [49] Myocardial infarction (MI)
− 12-14 yrs. of follow-up
MSM General population males; mean age 52 (SD 6.7) at initial exam (4 yrs. before ‘baseline’) Kuopio Ischaemic Heart Disease Risk Factor Study (KIHD); Finland 1030 −5 discrete-time hazard models run and compared: M1 (baseline alcohol consumption with no covariate adjustment or inclusion of t-4 consumption), M2 (baseline alcohol consumption adjusted for covariates and t-4 consumption), M3 (baseline and t + 7 alcohol consumption with no covariate adjustment or inclusion of t-4 consumption), M4 (baseline and t + 7 alcohol consumption adjusted for covariates and t-4 consumption), M5 (MSM with baseline and t + 7 alcohol consumption with stabilized IP weights at t and t + 7)
-Baseline alcohol categories: < 12 g/wk., 12-83 g/wk, 84-167 g/wk., ≥168 g/wk.; additional alcohol measurements at t-4 and t + 7 included in various models
RR (95% CI) M1 M2 M3 M4 M5 (MSM)
< 12 g/wk. 1.20 (.86,1.67) 1.01 (.70,1.45) 1.55 (1.08,2.21) 1.27 (.88,1.81) 1.27 (.88,1.83)
84-167 g/wk. 1.05 (.71,1.56) 1.13 (.75,1.72) 1.20 (.78,1.84) 1.27 (.81,2.00) 1.18 (.75,1.87)
≥168 g/wk. .98 (.60,1.58) 1.20 (.68,2.12) 1.40 (.91,2.18) 1.71 (1.03,2.85) 1.59 (.93,2.72)
Results are generally consistent with increased risk for MI for those drinking less than weekly and those drinking heavily. However, effect sizes varied with model specifications.
Kadlecova (2015) [50] Stroke and transient ischaemic attack (TIA) events
− 43 yrs. of follow-up
Twin Same-sex twins ≤60 with no history of stroke at baseline and with ≥5 yrs. of follow-up; mean age 50.5 (SD 5.29) at baseline Swedish Twin Registry 11,644; 370 stroke/TIA discordant pairs (all MZ); 167 stroke/TIA concordant pairs (all MZ) -Co-twin (discordant for stroke) logistic regressions and pooled cohort Cox models used to examine risk for stroke/TIA; co-twin (concordant for stroke) mixed-effects analyses used to examine time to stroke/TIA
-Categories for pooled cohort + twin analyses: abstainers, very light (> 0-5 g/day), light (> 5 − 12 g/day), moderate (> 12-24 g/day), heavy (> 24 g/day)
-Stroke/TIA risk:
HR (95% CI) Cohort OR (p value) Twins
Abstainers 1.11 (.98,1.23) Abstainers 2.22 (.058)
Light .98 (.85,1.15) Light 1.56 (.17)
Moderate .99 (.85,1.15) Moderate 1.59 (.26)
Heavy 1.34 (1.04,1.70) Heavy 1.23 (.78)
-Time to stroke/TIA:
Mean difference in yrs. to stroke/TIA (p value) Twins
Abstainers 2.07 (.11)
Light .77 (.54)
Moderate 1.15 (.47)
Heavy − 5.68 (.029)
Twin analyses are consistent with an increased risk of stroke/TIA for abstainers compared to very light drinkers, with somewhat increased risks for heavier drinking groups as well (reverse J-shape). Results support a causal role of heavy alcohol consumption in hastening time to event among those who experience stroke/TIA.
Millwood 2019 [51] Ischaemic stroke, intracerebral haemorrhage (ICH), total stroke, acute myocardial infarction (AMI), total coronary heart disease (CHD)
-roughly 10 yrs. of follow-up
MR Permanent residents from 10 Chinese regions aged roughly 35–74 and without major disabilities, (those with a history of CVD were excluded from analyses of disease incidence); mean age 52 (SD 11) at baseline China Kadoorie Biobank 512,715; 161,498 of which had genotype data (male and female combined) -1-sample MR using Cox models
-Instrument composed of 9 combinations of 2 SNPs (ALDH2 rs671 and ADH1B rs1229984) within each of 10 geographic areas, producing 90 combinations overall; then, based on mean ‘usual’ alcohol consumption within each of those combinations (incorporating repeat measurements to account for measurement error), six final categories were produced, aligned with increasing genetically-predicted alcohol consumption: Category1: 0-10 g/wk, Category2: 10-25 g/wk., Category 3: 25-50 g/wk., Category 4: 50-100 g/wk., Category 5100-150 g/wk., Category 6 > 150 g/wk.
-Conventional analyses using observed alcohol consumption conducted for comparison using Cox models, with consumption categories (for men): ex-drinker, non-drinker, occasional drinker (less than weekly), < 140 g/wk, 140-279 g/wk., 280-419 g/wk., ≥420 g/wk
-Ischaemic stroke:
Log RR (95% CI) MR Conventional
C2 1.00 (.91,1.10) Ex-drinker 1.39 (1.32,1.46)
C3 1.03 (.96,1.11) Non-drinker 1.21 (1.16,1.25)
C4 1.11 (1.02,1.20) Occasional 1.00 (.97,1.03)
C5 1.23 (1.15,1.33) 140-279 g/wk. 1.13 (1.07,1.19)
C6 1.23 (1.12,1.35) 280-419 g/wk. 1.23 (1.15, 1.32)
≥420 g/wk. 1.31 (1.21,1.41)
Per 280 g/wk. (assuming linearity)
1.27 (1.13,1.43) 1.28 (1.19,1.38)
-ICH:
Log RR (95% CI) MR Conventional
C2 1.01 (.88,1.14) Ex-drinker 1.52 (1.38, 1.68)
C3 1.02 (.88,1.14) Non-drinker 1.33 (1.25,1.43)
C4 1.08 (.96,1.22) Occasional 1.02 (.96,1.09)
C5 1.29 (1.15,1.44) 140-279 g/wk. 1.35 (1.21,1.52)
C6 1.54 (1.36,1.76) 280-419 g/wk. 1.52 (1.32,1.74)
≥420 g/wk. 1.73 (1.52,1.97)
Per 280 g/wk. (assuming linearity)
1.58 (1.36,1.84) 1.59 (1.37,1.85)
-Total stroke:
Log RR (95% CI) MR Conventional
C2 1.02 (.95,1.10) Ex-drinker 1.40 (1.34,1.46)
C3 1.05 (.95,1.10) Non-drinker 1.23 (1.19,1.27)
C4 1.13 (1.06,2.10) Occasional 1.00 (.98, 1.03)
C5 1.27 (1.19,1.34) 140-279 g/wk. 1.16 (1.10,1.21)
C6 1.35 (1.26,1.45) 280-419 g/wk. 1.28 (1.20,1.36)
≥420 g/wk. 1.39 (1.31,1.48)
Per 280 g/wk. (assuming linearity)
1.38 (1.26,1.51) 1.35 (1.27,1.44)
-AMI:
Log RR (95% CI) MR Conventional
C2 1.02 (.87,1.19) Ex-drinker 1.66 (1.48,1.85)
C3 1.05 (.93,1.19) Non-drinker 1.63 (1.51,1.76)
C4 .93 (.81,1.07) Occasional 1.23 (1.16,1.31)
C5 .94 9.81,1.09) 140-279 g/wk 1.11 (.97,1.26)
C6 .97 (.83,1.15) 280-419 g/wk 1.19 (1.01,1.41)
≥420 g/wk 1.14 (.95,1.37)
-Total CHD:
Log RR (95% CI) MR Conventional
C2 1.03 (.93,1.13) Ex-drinker 1.45 (1.38,1.52)
C3 1.08 (1.01,1.15) Non-drinker 1.31 (1.26,1.36)
C4 .94 (.87,1.02) Occasional 1.07 (1.04,1.11)
C5 1.04 (.97,1.11) 140-279 g/wk 1.03 (.97,1.09)
C6 1.06 (.98,1.15) 280-419 g/wk 1.11 (1.03,1.19)
≥420 g/wk 1.13 (1.04,1.22)
In contrast to conventional analyses, MR results are consistent with a monotonically increasing relationship between alcohol and stroke events. The results do not support a causal relationship between alcohol consumption and AMI or total CHD.
Ropponen 2014 [52]
*See also musculoskeletal health
DP due to circulatory system diagnoses
−5 − 10 years of follow-up
Twin Twins with data from a prior study, and at time of that study were living in Sweden, < 65, working and without DP/old age pension; mean age at baseline 53.7 (SD 5.7) a Swedish Twin Registry 31,206; 216 DP due to circulatory system diagnoses-discordant pairs (of which 95 are MZ) -Co-twin (discordant for DP due to MSD) Cox models and pooled cohort Cox models used to examine risk (stratified for sex in pooled model)
-Categories for pooled cohort + twin analyses: abstainers, light (≤3 drinks/wk), moderate (F: > 3- ≤ 7 drinks/wk.; M: > 3- ≤ 14 drinks/wk), heavy (F: >7drinks/wk.; M: > 14 drinks/wk)
HR (95% CI) Cohort MZ twins DZ twins All twins
Abstainers 1.22 (.91,1.64) .97 (.31,3.01) 1.55 (.70,3.44) 1.3 (.69,2.45)
Moderate .85 (.63,1.15) - .86 (.39,1.91) 1.54 (.78,3.04)
Heavy .79 (.63,.98) .91 (.49,1.68) .89 (.52,1.51) .86 (.57,1.28)
Results do not offer clear support for causal relationships between alcohol consumption and later receipt of DP due to circulatory system diagnoses.
Cardiovascular disease biomarkers
Peng (2019)
28
*See also diabetic outcomes
Lipids: HDL-C, non-HDL-C, triglycerides (TG), total cholesterol (TC); blood pressure: systolic blood pressure (SBP), diastolic blood pressure (DBP); obesity anthropometric measures: BMI, waist circumference (WC), hip circumference (HC), waist-to-hip ratio (WHR)
-cross-sectional
MR Chinese adults living in the Yi-Ling district of Yichang; mean age 55 (SD 0.1) at baseline One community selected from the Risk Evaluation of cAncers in Chinese diabeTic Individuals: a LONgitudinal (REACTION) study 4536 −1-sample MR using ALDH2 to instrument for alcohol consumption with standard IV analysis (2SLS); local average treatment effects (LATEs) computed for subgroups of observed alcohol consumption (non-zero LATE slopes indicate non-linearity)
-Analysis performed in women as a negative control due to their lack of alcohol consumption
-No conventional analyses conducted for comparison
-Non-linear analyses: the LATE slopes for all lipids, blood pressure and obesity parameters in both sexes were not sig. Different to zero, indicating no presence of non-linear relationships
Unstandardizedβ (95% CI) Using log-transformed alcohol intake
HDL-C −.01 (−.06,.04)
Non-HDL-C −.01 (−.13,.10)
TG (log-transformed) .03 (−.05,.10)
TC −.03 (−.16,.08)
SBP −.85 (− 3.15,1.3)
DBP −.70 (− 2.32,.71)
BMI −.15 (−.51,.21)
WC −.02 (− 1.09,1.09)
HC .34 (−.37,1.07)
WHR −.00 (−.01,.00)
-Standard IV analyses:
Unstandardizedβ (95% CI) Per 1-unit increase in log-transformed genetically-predicted alcohol consumption
HDL-C .04 (−.00,.08)
Non-HDL-C .11 (.02,.20)
TG (log-transformed) .11 (.05,.16)
TC .15 (.06,.25)
SBP 2.91 (1.06,4.76)
DBP 3.03 (1.87,4.19)
BMI .57 (.28,.87)
WC 2.37 (1.47,3.27)
HC 1.01 (.39,1.62)
WHR .02 (.00,.02)
LATE results offer no evidence of non-linear relationships between genetically predicted alcohol consumption and lipid markers, blood pressure or obesity parameters. Alcohol appears to raise BMI, WC, HC, non-HDL-C, TG, TC, SBP and DBP in a linear fashion in males, with little effect on HDL-C or WHR.
Silverwood (2014) [53] Lipids: non-HDL-C, HDL-C, TG; blood pressure: SBP; obesity anthropometric measures: BMI, WC; inflammatory markers: CRP, interleukin 6 (IL-6)
-cross-sectional
MR Individuals of European descent from Europe and North America; mean age 56.75 (calculated from Holmes et al. [cite]) 22 individual studies (18 cohorts, 2 nested case-control, 1 RCT, 1 case-control) from the Alcohol-ADH1B consortium 80,057 individuals total; 78,172 for SBP, 60,140 for non-HDL-C, 60,227 for HDL-C, 79,454 for BMI, 57,172 for WC, 63,367 for CRP, 23,535 for IL-6, 63,667 for TG -1-sample MR using the rs1229984 polymorphism in ADH1B to instrument for alcohol consumption and standard IV analysis (2SLS); local average treatment effects (LATEs) computed for subgroups of observed alcohol consumption (non-zero LATE slopes indicate non-linearity)
-Where non-linearity is present, the difference in outcome between no alcohol and median observed consumption in low (> 0–7 units/wk), moderate (7–21 units/wk), heavy (21–70 units/wk) and very heavy (> 70
units/wk) groups is predicted, as well as curve nadir, difference in outcome between nadir and abstinence, and level of consumption matching outcome for abstinence
-No conventional analyses conducted for comparison
-Non-linear analyses: the LATE slopes for SBP, non-HDL-C, BMI, WC and CRP were all sig. Different to zero, indicating non-linearity
Unstandardizedβ (95% CI) Using log-transformed alcohol intake
HDL-C .00 (−.06,.06)
Non-HDL-C .37 (.19,.55)
TG (log-transformed) -.02 (−.10,.06)
SBP 3.30 (1.0,5.5)
BMI .90 (.3,1.4)
WC 2.00 (.6,3.6)
CRP (log-transformed) .26 (.10,.43)
IL-6 (log-transformed) -.13 (−.34,.29)
-Predicted differences in outcome between category medians and abstinence (unstandardized):
3.04 units/wk. 12.15 units/wk. 31.90 units/wk. 84.52 units/wk.
Non-HDL-C −.39 (−.79,.06) -.15 (−.72,.47) .40 (−.28,1.10) 1.30 (.45,2.16)
SBP .1 (− 5.5,6.1) 5.2 (− 2.6,13.9) 12.4 (3.4,22.1) 22.8 (12.2,34.6)
BMI −.6 (− 2.2,.8) .2 (− 2.0,2.1) 1.6 (−.8,3.8) 3.9 (1.2,6.3)
WC −.6 (− 4.7,3.5) 1.9 (− 3.9,7.8) 5.7(−.6,12.5) 11.5 (4.5,12.5)
CRP (log-transformed) -.29 (−.68,.15) -.15 (−.68,.5) .22 (−.37,.95) .83 (.15,1.69)
-Predicted curve features for those outcomes with evidence of non-linearity (unstandardized):
Nadir (units/wk.; 95% CI) Difference in outcome at nadir Units/wk. (95% CI) with outcome equivalent to abstinence
Non-HDL-C 3.2 (.7,6.0) −.39 (−.85,-.03) 16.9 (2.1, 48.2)
SBP 1.00 (.0,3.6) −.7(− 5.4,.0) 2.8 (.0,19.6)
BMI 2.3 (.0,6.0) −.6 (− 2.3,.0) 10.1 (.0,48.4)
WC 1.5 (.0,5.4) −.8 (− 4.9,.0) 5.3 (.0,37.4)
CRP 3.5 (.0,7.2) −.30 (−.75,.00) 19.4 (.0,66.0)
-Standard IV analyses for those outcomes with no evidence of non-linearity:
β (95% CI) Per 1-unit increase in log-transformed genetically-predicted alcohol consumption
HDL-C −.02 (−.07,.03)
TG (log-transformed) −.01 (−.06,.07)
IL-6 (log-transformed) .30 (.16,.45)
Results support non-linear relationships between alcohol consumption and SBP, non-HDL-C, BMI, WC and CRP, with nadirs for these outcomes falling in the low drinking range. These relationships are best characterised by J-shapes with shallow nadirs, followed by gentle, elongated inclines. Results are consistent with a positive linear relationship between alcohol and IL-6, and do not support any causal relationships between alcohol and HDL-C or TG.
Vu (2016) [54] Lipids: TG, total cholesterol, HDL-C, HDL2-C, HDL3-C, LDL-C, sdLDL-C, apoB, Lp(a)
-cross-sectional
MR European Americans; mean age 54.3 (SD 5.7) at baseline Atherosclerosis Risk in Communities (ARIC); USA 10,893 individuals total; 9911 for TG, 9751 for total cholesterol and LDL-C, 10,132 for HDL-C, 10,120 for HDL2-C and HDL3-C, 8102 for sdLDL-C, 7663 for apoB, 9924 for Lp(a) − 1-sample MR using a genetic risk score composed of 5 SNPs (rs2066702, rs1693457, rs1789891, rs698, and
rs1126671) and standard IV analysis (2SLS); model fitted separately for quartiles of genetically-predicted alcohol consumption (q1 = 1.49–3.63 g/wk, q2 = 3.63–4.66 g/wk., q3 = 4.66–10.57 g/wk., q4 = 10.57–19.54 g/wk) to determine presence of non-linearity
-Conventional analyses regressing outcomes on observed alcohol consumption were also conducted, using consumption categories: never drinkers, former/infrequent drinkers (< 1 drink/wk), low-to-moderate current drinkers (M: ≤210 g/wk., F: ≤105 g/wk), heavy current drinkers (M: > 210 g/wk., F: > 105 g/wk)
-TG (log-transformed):
Unstandardized β (95% CI) MR Using log-transformed alcohol intake Conventional
Q2–.06 (−.09,-.02) Former/ infrequent −.08 (−.11,-.05)
Q3–.13 (−.20,-.07) Low-to-moderate −.16 (−.19,-.13)
Q4–.08 (−.17,.00) Heavy −.13 (−.17,-.09)
-TC:
Unstandardized β (95% CI) MR Using log-transformed alcohol intake Conventional
Q2–5.54 (− 8.23,-2.85) Former/ infrequent − 2.75 (− 4.99,-.51)
Q3–7.71 (− 13.26,-2.15) Low-to-moderate −.73 (− 3.03,1.57)
Q4–4.56 (− 11.36,2.25) Heavy 4.05 (.73,7.37)
-HDL-C (log-transformed):
Unstandardized β (95% CI) MR Using log-transformed alcohol intake Conventional
Q2 .01 (−.01,.03) Former/ infrequent .03 (.01,.04)
Q3 .04 (.00,.07) Low-to-moderate .14 (.12,.15)
Q4 .03 (−.02,.07) Heavy .26 (.23,.28)
-HDL2-C (log-transformed):
Unstandardized β (95% CI) MR Using log-transformed alcohol intake Conventional
Q2 .04 (.00,.07) Former/ infrequent .07 (.04,.10)
Q3 .10 (.03,.17) Low-to-moderate .17 (.14,.20)
Q4 .06 (−.03,.15) Heavy .30 (.26,.35)
-HDL3-C:
Unstandardized β (95% CI) MR Conventional
Q2–.19(−.87,.49) Former/ infrequent .40 (−.14,.94)
Q3 .08 (− 1.23,1.38) Low-to-moderate 4.16 (3.60,4.73)
Q4 .11 (− 1.51,1.74) Heavy 8.70 (7.84,9.55)
-LDL-C:
Unstandardized β (95% CI) MR Using log-transformed alcohol intake Conventional
Q2 −4.60 (−7.18,-2.03) Former/ infrequent −2.59 (− 4.69,-.49)
Q3 −6.87 (− 12.24,-1.50) Low-to-moderate − 4.38 (− 6.53,-2.23)
Q4 − 4.57 (− 11.11,1.96) Heavy −7.48 (− 10.70,-4.25)
-sdLDL-C (log-transformed):
Unstandardized β (95% CI) MR Using log-transformed alcohol intake Conventional
Q2 −.04 (−.08,-.01) Former/ infrequent −.04 (−.07,-.01)
Q3 −.08 (−.15,-.01) Low-to-moderate −.05 (−.08,-.01)
Q4 −.08 (−.17,.01) Heavy .01 (−.04,.06)
-apoB (log-transformed):
Unstandardized β (95% CI) MR Using log-transformed alcohol intake Conventional
Q2 −.03 (−.04,−.01) Former/ infrequent -.01 (−.02,.01)
Q3 −.04 (−.07,.00) Low-to-moderate −.01 (−.03,.01)
Q4 −.04 (−.08,.01) Heavy −.02 (−.04,.01)
-Lp(a) (log-transformed):
Unstandardized β (95% CI) MR Using log-transformed alcohol intake Conventional
Q2 −.02 (−.09,.06) Former/ infrequent −.03 (−.10,.03)
Q3 −.05 (−.20,.10) Low-to-moderate .01 (−.06,.08)
Q4 −.01 (−.20,.18) Heavy −.06 (−.16,.03)
Results support causal relationships between alcohol and TG, TC, HDL2-C, LDL-C, sdlDL-C and apoB, such that some level of consumption leads to more favorable levels than abstinence/very low consumption does. Analyses suggest non-linearity, with benefits peaking for most outcomes at roughly .5–.1 drinks per week, although effect sizes vary. Results do not support causal relationships between alcohol and HDL-C, HDL3-C and Lp(a).
Mortality
Sipila 2016 [55] All-cause mortality
-median 30.2 yrs. of follow-up
Twin Same-sex twins aged 18–54 at baseline and free of chronic disease 6 years post-baseline; mean age 35.9 at baseline Older Finnish Twin Cohort 14,787; 3389 drinking-discordant pairs (of which 926 pairs are MZ) a -Co-twin (discordant for alcohol consumption) Cox models and pooled cohort Cox models used to examine risk for mortality
-Categories for pooled cohort + twin analyses based on average of two measurements 6 yrs. apart: abstainers,
1-69 g/mnth, 70-139 g/mnth, 140-209 g/mnth, 210-419 g/mnth, 420-839 g/mnth, 840-1199 g/mnth, ≥1200 g/mnth
ϰ HR (95% CI) Cohort MZ twins All twins
0 g/mnth 1.02 (.85,1.22) .43 (.17,1.11) .96 (.63,1.45)
70–139 g/mnth .95 (.81,1.10) .64 (.29,1.40) .96 (.68,1.36)
140–209 g/mnth 1.08 (.91,1.29) 1.12 (.47,2.65) .85 (.57,1.26)
210–419 g/mth 1.29 (1.09,1.53) 1.55 (.65,3.71) 1.40 (.94,2.09)
420–839 g/mnth 1.56 (1.31,1.85) 1.70 (.69,4.22) 1.60 (1.06,2.43)
840–1199 g/mnth 2.17 (1.74,2.70) 1.65 (.54,5.08) 2.65 (1.50,4.69)
≥1200 g/mnth 2.81 (2.26,3.50) 3.18 (.81,12.43) 2.99 (1.60,5.59)
Results are consistent with a monotonically increasing risk function. Twin analyses support a causal role for increased risk of all-cause mortality for moderate-to-very heavy drinking.
HIV seroconversion
Sander (2013) [56] HIV seroconversion
-median 10.5 yrs. of follow-up
MSM Men who have sex with men and who were sexually active and HIV-seronegative at baseline; median age 33.4 at baseline Multicenter AIDS Cohort Study (MACS); USA 3752 -MSM (Cox models) + standard Cox models for comparison
-Categories for analyses based on average of two measurements 1 yr apart: abstainers, moderate (1–14 drinks/wk), heavy (> 14 drinks/wk); 1 drink considered = 14 mL
ϰ RR (95% CI) Non-MSM MSM
Moderate .91 (.65,1.27) 1.10 (.78,1.54)
Heavy 1.19 (.83,1.70) 1.61 (1.12,2.29)
Results are consistent with a monotonically increasing risk function. Abstainers and moderate drinkers appear to have similar risk for HIV seroconversion, while heavy drinkers are at increased risk.
Musculoskeletal health
Pietikainen 2011 [57] DP due to low back disorders (LBD)
− 29 years of follow-up
Twin Same-sex twins aged 18–64 and not receiving pension at baseline; mean age 33.2 (SD 12) at baseline Finnish Twin Cohort 24,043; 504 (284 M) pairs discordant for DP due to LBD -Co-twin (discordant for DP due to LBD) Cox models and pooled cohort Cox models used to examine risk for mortality
-Categories for pooled cohort + twin analyses: abstainers, light (≤3 drinks/wk), moderate (F: > 3- ≤ 7 drinks/wk.; M: > 3- ≤ 14 drinks/wk), heavy (F: >7drinks/wk.; M: > 14 drinks/wk)
ϰ HR (95% CI) Cohort All twins
Abstainer .85 (.61,1.20) .79 (.49,1.27)
Moderate 1.07 (.79,1.43) .94 (.62,1.42)
Heavy 1.08 (.80,1.46) 1.07 (.69,1.66)
Results suggest a roughly monotonically increasing relationship between alcohol and later receipt of DP due to LBP, with reduced risk for abstainers the clearest feature.
Ropponen 2014 [52]
*see also CVD section
DP due to musculoskeletal diagnoses (MSD)
− 5-10 years of follow-up
Twin Twins with data from a prior study, and at time of that study were living in Sweden, < 65, working and without DP/old age pension; mean age at baseline 53.7 (SD 5.7) b Swedish Twin Registry 31,206; 922 DP due to MSD-discordant pairs (of which 357 are MZ) -Co-twin (discordant for DP due to MSD) Cox models and pooled cohort Cox models used to examine risk (stratified for sex in pooled model)
-Categories for pooled cohort + twin analyses: abstainers, light (≤3 drinks/wk), moderate (F: > 3- ≤ 7 drinks/wk.; M: > 3- ≤ 14 drinks/wk), heavy (F: >7drinks/wk.; M: > 14 drinks/wk)
HR (95% CI) Cohort MZ twins DZ twins All twins
Abstainers .93 (.80,1.07) 1.49 (.98,2.26) .8 (.54,1.19) 1.07 (.81,1.42)
Moderate .8 (.69,.93) 2.33 (1.39,3.91) .67 (.48,.95) 1.02 (.78,1.34)
Heavy .73 (.67,.81) 1.11 (.82,1.51) .77 (.60,.98) .88 (.73,1.07)
Results do not support a clear relationship between alcohol and later receipt of DP due to MSD, particularly given the discrepancy in direction of effects between MZ and DZ twins.
Ropponen 2011 [58] DP due to musculoskeletal disorder (MSD) and osteoarthritis specifically (OA)
−29 years of follow-up
Twin MZ and same-sex DZ twins aged ≥18 and working at baseline; mean age 33.2 (SD 12) at baseline Finnish Twin Cohort 24,043; 1317 pairs discordant for DP due to MSD, 461 pairs discordant for DP due to OA -Co-twin (discordant for DP due to MSD/OA) Cox models and pooled cohort Cox models used to examine risk for MSD and OA in men and women separately
-Categories for pooled cohort + twin analyses: abstainers, light (≤3 drinks/wk), moderate (F: > 3- ≤ 7 drinks/wk.; M: > 3- ≤ 14 drinks/wk), heavy (F: >7drinks/wk.; M: > 14 drinks/wk)
ϰ -For DP due to MSD:
HR (95% CI) Cohort (M) Cohort (F) All twins (M) All twins (F)
Light .91 (.59,1.40) 1.15 (.93,1.42) 2.04 (1.09,3.82) 1.07 (.81,1.41)
Moderate .95 (.70,1.30) 1.23 (1.00,1.51) 1.50 (.94,2.38) .97 (.74,1.27)
Heavy 1.01 (.73,1.39) 1.08 (.84,1.39) 1.58 (.99,2.53) 1.37 (.98,1.91)
-For DP due to OA specifically:
HR (95% CI) Cohort (M) Cohort (F) All twins (M) All twins (F)
Light .99 (.50,1.96) 1.19 (.86,1.63) 4.07 (1.15,14.36) 1.33 (.82,2.14)
Moderate .78 (.48,1.28) .96 (.69, 1.32) 2.01 (.82,4.93) 1.12 (.70,1.82)
Heavy 1.04 (.64,1.71) .86 (.60,1.25) 2.39 (.97,5.89) 2.32 (1.21,4.47)
Results do not offer enough support for a clear functional form, but are consistent with abstinence representing the lowest risk for DP due to MSD and OA specifically. Heavy drinkers were also at increased risk for both outcomes and sexes.
  1. a Ropponen et al. (2014) also used this cohort to examine the relationship between alcohol and disability pension due to mental health diagnoses – not reported on here as Samuelsson et al. used the more informative exposure categorization
  2. b From correspondence with authors
  3. Note: Where multiple models are reported in the original paper, the most adjusted analyses are reflected in the above table if (excluding for MR studies)