ReviewThe temporal relationship between per capita alcohol consumption and harm: A systematic review of time lag specifications in aggregate time series analyses☆
Introduction
Average levels of alcohol consumption in the population are widely recognised as a relevant public health indicator. Such aggregate levels of consumption have been associated with the incidence and prevalence of a range of alcohol-related harms including morbidity and mortality from various health conditions and also rates of crime, unemployment and workplace absences (Purshouse et al., 2009). These associations can be seen in cross-sectional studies; however, stronger evidence comes from time series analyses showing that changes in per capita alcohol consumption are temporally linked to changes in rates of alcohol-related harms. Although methodological approaches for such analyses have been suggested (Norström and Skog, 2001, Rehm and Gmel, 2001) debates on aspects of applying these continue. One such aspect is the time lag problem.
Given that much alcohol-related harm is the accumulated result of years of harmful individual drinking behaviours, the full effect of changes in consumption may not be immediately apparent in harm data. Instead, the effect of changes in aggregate consumption may be delayed and distributed over a number of years. In response to this problem, it is commonplace to incorporate a lag structure into time series analyses to ensure the full long-term effect is captured.
Studies which have explored time lags in depth have tended to focus on liver cirrhosis and have found that, despite the anticipated long-term effect, much of the impact on cirrhosis mortality rates occurs in the first year following a change in consumption (Kerr et al., 2000, Skog, 1984). This somewhat paradoxical finding of immediate effects at the aggregate level on a harm which develops after many years of heavy drinking at the individual level can be clearly observed in the sharp falls in cirrhosis deaths following alcohol rationing in Paris during World War Two (Norström, 1987).
The notion of critical thresholds has typically been used to explain this paradox (Norström, 1987, Norström, 1989, Skog, 1980). It is postulated that, at any given time, there are a group of people with advanced liver cirrhosis for whom a change in alcohol consumption could prompt or prevent liver failure. It is changes in the mortality rate within this group that are often used to explain the rapid effects of changes in aggregate alcohol consumption (Norström and Skog, 2001). Simultaneous changes in alcohol consumption amongst those who are not at this critical threshold also need to be accounted for and Skog (1984) and Norström (1987) have obtained consistent results modelling lag structures specifying both short- and long-term effects for the UK and Sweden respectively. However, other work has suggested time series models accounting only for a short-term effect adequately fit the data (Kerr et al., 2000, Roizen et al., 1999).
The primary concern of this paper is to provide the first systematic review of aggregate time series lag structures which have been applied to different alcohol-related harms with a focus on three pieces of information: the time to first effect, the duration to full effect and the functional form of the accumulation of effect. Referred to hereafter as the lag specifications, these pieces of information are illustrated in Fig. 1.
Section snippets
Methods
The search was conducted across the following databases between December 2010 and February 2011: ASSIA, Campbell Collaboration, CINAHL, Econlit, IBSS, Embase, Medline, PsychINFO, Scopus, Social Care Online, Sociological Abstracts, Web of Knowledge and World Political Science Abstracts.
The full search terms can be seen in the online supplementary material. The search was conducted in three stages using a set of alcohol terms combined in turn with three sets of terms relating to lags. These three
Results
The results of the literature search are shown in Fig. 2. Of 3342 studies initially identified, 18 were included in the narrative synthesis with a further 18 studies identified through reference and citation checking of those studies. Reporting of the results is structured as follows: a summary of the different methods used to derive lag specifications and their implications are presented first, followed by the lag specifications used for liver cirrhosis, heart disease and suicide in turn.
Discussion
The above review presents current knowledge on the different ways of deriving time lag specifications when modelling the effects of aggregate alcohol consumption on rates of various alcohol-related harms. Although consistency across studies is limited, some recommendations can be made about valid lag specifications for different harms. These are presented at the start of the discussion followed by some general comments on deriving, modelling and reporting lag specifications.
The results
Limitations
The main limitation of the review was the difficulty in identifying relevant studies as lag specifications were commonly not referred to in abstracts. Reference and citation checking and professional networks were invaluable in limiting this; however, it is possible studies may have been omitted due to non-identification.
Conclusions
Time lag specifications for aggregate time series analyses of the impacts of changes in alcohol consumption on alcohol-related harms are underdeveloped for most harms. For liver cirrhosis, heart disease and suicide some recommendations can be made including the expectation of immediate, front-loaded effects. In general however, greater research attention needs to be given to the rationale for choosing or applying particular lag specifications and the inherent complexity of the processes which
Role of funding source
This work was funded by a grant from the Medical Research Council (G000043). The funder played no part in any stage of the conduct of this research, the preparation of the manuscript or the decision to publish.
Contributors
John Holmes: Lead author, conducted search, led the review of evidence and analysis of findings.
Petra Meier: Supervised and guided all stages of the research process, contributed to the analysis of findings and the redrafting of the manuscript.
Andrew Booth: Contributed to and guided the creation and refinement of the search strategy.
Yelan Guo: Contributed to the review of evidence and the design of the search strategy.
Alan Brennan: Contributed to the analysis of findings and the redrafting of
Conflict of interest
No conflict declared.
Acknowledgements
The authors would like to acknowledge the valuable contributions to this paper made by those who attended a discussion session on time lags at the 2011 KBS conference and the members of their project's scientific steering committee. We also thank Corinna Slany for her contribution to the literature search and our clinician colleagues Ravi Maheswaran and Andrew Lee for their guidance in selecting lag specifications for harms where no evidence was available.
References (49)
Explaining differential rates of mortality decline for Swedish men and women: a time-series analysis, 1945–1992
Soc. Sci. Med.
(1999)- et al.
Relationship of alcohol-consumption to atherosclerotic heart-disease
Prev. Med.
(1980) - et al.
Time Series Analysis: Forecasting and Control
(1976) Liver cirrhosis mortality trends in Eastern Europe, 1970–1989. Analyses of age, period and cohort effects and of latency with alcohol consumption
Addict. Biol.
(1998)- et al.
Are the recent trends in liver cirrhosis mortality affected by the changes in alcohol consumption? Analysis of latency period in European countries
J. Stud. Alcohol
(1997) - et al.
Alcohol-consumption and ischemic-heart-disease mortality – are time-series correlations meaningful
Am. J. Epidemiol.
(1983) Per capita alcohol consumption and ischaemic heart disease mortality
Addiction
(2001)- et al.
The effects of a large reduction in alcohol prices on hospitalizations related to alcohol: a population-based natural experiment
Addiction
(2011) - et al.
On measurement of avoidable and unavoidable cost of alcohol: an application of method for estimating costs due to prior consumption
Int. J. Environ. Res. Public Health
(2010) Econometric Methods
(1984)
Beverage-specific alcohol consumption and cirrhosis mortality in a group of English-speaking beer-drinking countries
Addiction
Per capita alcohol consumption and ischemic heart disease mortality in a panel of US states from 1950 to 2002
Addiction
Population-level relationships between alcohol consumption measures and ischemic heart disease mortality in US time-series
Alcohol. Clin. Exp. Res.
Alcohol consumption and suicide mortality by age among Finnish men, 1950–1991
Addiction
The impact of per-capita consumption on Swedish cirrhosis mortality
Br. J. Addict.
The use of aggregate data in alcohol epidemiology
Br. J. Addict.
Per capita alcohol consumption and sickness absence
Addiction
Alcohol and mortality: methodological and analytical issues in aggregate analyses
Addiction
A time-series analysis of the impact of heavy drinking on homicide and suicide mortality in Russia, 1956–2002
Addiction
Alcohol consumption and liver cirrhosis mortality with and without mention of alcohol – the case of Canada
Addiction
Alcohol and pancreatitis mortality at the population level: experiences from 14 western countries
Addiction
Population drinking and liver cirrhosis mortality: is there a link in eastern Europe?
Addiction
Homicide and alcohol intoxication in Russia, 1956–2005
Alcoholism (Zagreb)
Cited by (0)
- ☆
Supplementary material can be found by accessing the online version of this paper at http://dx.doi.org. Please see Appendix A.