alcohol consumption and liver cirrhosis mortality with and without mention of alcohol—the case of...

RESEARCH REPORT

© 2003 Society for the Study of Addiction to Alcohol and Other Drugs

Addiction,

98

, 1267–1276

Blackwell Science, Ltd

Oxford, UK

ADDAddiction

1360-0443© 2003 Society for the Study of Addiction to Alcohol and Other Drugs

98Original Article

Mats RamstedtAlcohol consumption and liver cirrhosis mortality

Correspondence to:

Mats RamstedtCentre for Social Research on Alcohol and Drugs (SoRAD)Stockholm UniversitySveaplanSE-106 91 StockholmSwedenE-mail: [email protected]

Submitted 22 January 2003;initial review completed 3 April 2003;

final version accepted 21 April 2003

RESEARCH REPORT

Alcohol consumption and liver cirrhosis mortality with and without mention of alcohol—the case of Canada

Mats Ramstedt

Centre for Social Research on Alcohol and Drugs (SoRAD), Stockholm University, Stockholm, Sweden

ABSTRACT

Aims

To analyse post-war variations in per capita alcohol consumption in rela-tion to gender-specific liver cirrhosis mortality in Canadian provinces and toassess the extent to which alcohol bears a different relation to cirrhosis deathswith mention of alcohol (alcoholic cirrhosis) compared to cirrhosis deaths with-out mention of alcohol (non-alcoholic cirrhosis).

Data and method

Annual liver cirrhosis mortality rates by 5-year age groupswere converted into gender-specific and age-adjusted mortality rates. Outcomemeasures included total cirrhosis—the conventional measure of liver cirrho-sis—alcoholic cirrhosis and non-alcoholic cirrhosis. Per capita alcohol con-sumption was measured by alcohol sales and weighted with a 10-yeardistributed lag model. A graphical analysis was used to examine the regionalrelationship and the Box–Jenkins technique for time-series analysis was used toestimate the temporal relationship.

Findings

Geographical variations in alcohol consumption corresponded tovariations in total liver cirrhosis and particularly alcoholic cirrhosis, whereasnon-alcoholic cirrhosis rates were not associated geographically with alcoholconsumption. In general, for all provinces, time-series analyses revealed posi-tive and statistically significant effects of changes in alcohol consumption oncirrhosis mortality. In Canada at large, a 1-litre increase in per capita consump-tion was associated with a 17% increase in male total cirrhosis rates and a 13%increase in female total cirrhosis rates. Alcohol consumption had a strongerimpact on alcoholic cirrhosis, which increased by fully 30% per litre increase inalcohol per capita for men and women. Although the effect on the non-alco-holic cirrhosis rate was weaker (12% for men and 7% for women) it was never-theless statistically significant and suggests that a large proportion of thesedeaths may actually be alcohol-related.

Conclusions

Some well-established findings in alcohol research were confirmedby the Canadian experience: per capita alcohol consumption is related closely todeath rates from liver cirrhosis and alcohol-related deaths tend to be under-reported in mortality statistics.

KEYWORDS

Canada, liver cirrhosis mortality, per capita alcohol

consumption.

INTRODUCTION

The mortality rate from liver cirrhosis is seen as an impor-tant indicator of alcohol-related harm in society (WHO2000). Consequently, variations in this measure haveattracted much attention in the alcohol research field

where the interplay with per capita alcohol consumptionin particular has been in focus. The Canadian experiencehas played an important role in this research tradition. Inthe 1970s, the Canadian case was drawn upon as onepiece of evidence behind the idea that per capita alcoholconsumption is a key determinant of alcohol-related


Addiction,

98

, 1267–1276

1268

Mats Ramstedt

harm in society (Bruun

et al

. 1975; see also Schmidt1977). Cirrhosis rates and per capita consumption inCanada were found to be strongly and positively corre-lated during the period 1920–60, both within Canadianprovinces over time and across provinces (Popham1970). In contrast, the Canadian experience during the1970s and 1980s was a case when the development ofthese phenomena did not match as expected. Forinstance, only a slight decrease in per capita consumptionin Ontario between 1975 and 1983 was followed by 43%decline in cirrhosis mortality (Smart & Mann 1987). Atleast implicitly, the idea of overall drinking as a key deter-minant of alcohol-related harm was now questioned onthe basis of the Canadian experience.

However, not even strong temporal covariationbetween series exhibiting long-term trends (as do theseseries) is strong evidence for drawing conclusions aboutcausal relationships (Norström & Skog 2001). Even ifalcohol is regarded as the leading cause of liver cirrhosis,variations in cirrhosis rates are affected by other factors,e.g. hepatitis rates, possibly with similar long-term trendsas alcohol consumption. Moreover, as not only presentbut also past alcohol intake impinges on the risk for livercirrhosis, this should be taken into account when theassociation between these variables is examined. Becausethese methodological issues have not been considered inprevious studies, existing knowledge of the Canadianexperience is insufficient. One aim of this paper is to fillthis gap by using a statistical technique usually referredto as autoregressive integrated moving average (ARIMA)time-series analysis (described more in detail below). TheARIMA approach has been used in time-series analysesof alcohol and cirrhosis in many other countries, whichthus provide a basis for comparison. It would, forinstance, be interesting to see how Canada relates towestern Europe, where 14 countries have been examinedrecently with this approach (Ramstedt 2001).

While previous Canadian studies have considered livercirrhosis rates in the whole country or in the largest prov-ince Ontario, with men and women mainly together, thepresent work will carry out a somewhat more detailedanalysis. First, men and women will be analysed sepa-rately in order to test if there is a stronger impact of percapita consumption on male cirrhosis rate as mostly wasthe case in western Europe (Ramstedt 2001). Secondly,the Canadian constitution with several independentprovinces allows for conducting a comparative analysiswithin the same country, which is seldom practicable inother countries. It is thus feasible to test if overall drinkingis equally important for changes in liver cirrhosis in dif-ferent parts of Canada.

Thirdly, the analyses will be carried out for differentsubclassifications of cirrhosis. Although the conventionhas been to analyse all cirrhosis deaths (referred to here-

after as total cirrhosis) in relation to per capita alcoholconsumption, this measure is composed of deaths codedwith and without mention of alcohol (hereafter referredto as alcoholic and non-alcoholic cirrhosis). This strategyto use total cirrhosis has been justified by findings reveal-ing that a substantial proportion of deaths coded as non-alcoholic cirrhosis should actually have been recorded asalcohol-related (e.g. Blake

et al

. 1988). If this is the case,we would expect that non-alcoholic cirrhosis rates alsorespond to changes in alcohol consumption.

DATA AND METHOD

Liver cirrhosis deaths (underlying cause) by province formen and women and by 5-year age groups were compiledfrom Statistics Canada (CANSIM). The basic observationperiod was 1950–98, but in some provinces a few yearswere missing in the early period. The classification codeand title for the liver cirrhosis mortality category variedwith the revisions of the International Classification ofDiseases (ICD). In ICD-6 and ICD-7 (1950–68) the codewas 581, with 581.1 used for the alcoholic cases and581.0 for non-alcoholic. In ICD-8 (1968–78) and ICD-9(1979–98) the code changed to 571, with 571.0 repre-senting alcoholic cirrhosis and 571.8-9 the non-alco-holic. In ICD-9 the title ‘liver cirrhosis’ was replaced by‘chronic liver disease and cirrhosis’, with 571.0–571.3involving alcoholic cases and 571.4–571.9 includingnon-alcoholic cases. Because the label ‘liver cirrhosis’was used during the major part of the study period andalso has been applied in most previous studies, this termwill be employed here even though several non-cirrhosisdeaths are included.

Per capita alcohol consumption measures (expressedin litres 100% alcohol per inhabitant 15 years and older)from 1950 and onwards, were derived from CANSIM. Thedata were collected by financial year, i.e. from 1 April to31 March, while mortality data refer to calendar year. Itwas decided therefore that the consumption variable waslagged 1 year so that 1 April to 31 December belonged tothe same year as the mortality data, whereas the period 1January to 31 March thus had to be imputed from thecoming year. The smallest of the Atlantic provinces,Prince Edward Island, was not included in the analysis asconsumption data were not available until 1970, and theremaining fairly small Atlantic provinces Nova Scotia,New Brunswick and Newfoundland were combined asone region, denoted the Maritimes. The two territories,Yukon and North-west Territories, had to be excludedfrom the study because of small populations.

The functional form of the basic model is semilogarith-mic in order to consider that the risk for liver cirrhosis is aconvex function of alcohol intake (Pequignot

et al

. 1978;

Alcohol consumption and liver cirrhosis mortality

1269


Addiction,

98

, 1267–1276

Corrao

et al

. 1999). Because both mortality and con-sumption series exhibited strong trends in all provinces(see Fig. 4), the analyses were performed on yearlychanges rather than on raw data in order to reduce therisk of obtaining a spurious relationship between the twovariables (Norström & Skog 2001).

A lag structure was also incorporated into the alcoholmeasure in order to consider that not only present but alsopast drinking affects the cirrhosis risk (Lelbach 1974;Skog 1980; Norström & Skog 2001). A distributed lagmodel employed by Norström (1987) and Skog (1984)was used where the total impact of consumption on cir-rhosis mortality is distributed over several years. Adummy variable (D) for changes in ICD-classification wasinitially tested during the estimation process and includedif significant. Overall, the shift to ICD-9 was associatedwith a significant decrease in total and alcoholic cirrhosisrates among men whereas a significant increase in alco-holic cirrhosis among women was found with the shift toICD-7. However, these dummy variables had no influenceon the alcohol effect parameter in any analysis.

Hence, the basic model to be estimated can be writtenas:

—

LnM

t

=

a

+

b*

—

Aw

t

+

c*

—

D

t

+

—

LnN

t

,

where

—

is the differencing operator, LnM is the naturallogarithm of age-standardized cirrhosis mortality (15

+

)per 100 000, a denotes average annual changes in M dueto other causes, the effect parameter b indicates the pro-portional change in M (proportional because the naturallogarithm of mortality is used) associated with a 1-litrechange in weighted per capita alcohol consumption(Aw). The effect of changes in ICD versions is indicated bythe parameter c, and N is the noise term including othercausal factors. The temporal structure of the noise term isconsidered by estimating AR (auto-regressive) or MA(moving average) parameters in the models.

The effect for Canada at large is calculated both as apopulation-based average and as an average effect basedon the estimates for the individual provinces. In the lattercase the provinces were weighted according to popula-tion (p

i

), and the standard error of the consequential aver-age was estimated according to the formula:

where denotes the standard errors for each region.The pooling technique can be seen as a way to diminishthe large standard errors associated with differencing thedata (Norström & Skog 2001) and should yield a similarestimate as with the population-based average.

Before model estimation, scatterplots between alcoholand cirrhosis (differenced data) were inspected in orderto detect possible outliers, i.e. extreme values with an

SE p SE bpool i i i= Â ◊ ( )[ ]22ˆ

SE biˆ( )

exorbitant influence on the parameter estimate. Oneclear case was found, namely Alberta in 1980, whichwas traced to a strike during the summer. In this case adummy was included. Other strikes have been takenplace in Canada with an observable change in sales sta-tistics such as the beer strike in Manitoba of 1978(Harper

et al

. 1981) and Ontario 1958 and 1968 (Single1979). However, these strikes affected mainly beer salesand not total alcohol sales, and did not give rise to anoutlier-problem.

RESULTS

Differences in overall drinking and cirrhosis rates in Canadian provinces

Table 1 presents data for Canada and the provinces interms of some average figures for the study period. First,we can see that per capita alcohol consumption is fairlysimilar across the Canadian provinces, ranging betweenfully 7 l per capita to 10 l for the period 1950–98. TheMaritimes have a lower per capita consumption than par-ticularly British Columbia, but also lower than the Prai-ries, Ontario and Quebec. According to a Canadianalcohol survey from 1989, abstainers tend to be morecommon in the Maritimes and least common in BritishColumbia and Alberta (Health and Welfare Canada1992). Thus, per capita consumption per drinker mightbe even more homogeneous across the provinces. Livercirrhosis mortality rates vary across the provinces andbetween men and women. For the whole of Canada weobserve 17 male cirrhosis deaths per 100 000 and year,ranging from 11 in the Maritimes to around 20 in thethree largest provinces. For women, the average rate iseight cirrhosis deaths per 100 000, again with the lowestrates in the Maritimes (five per 100 000) to 10 in BritishColumbia.

Almost half of total cirrhosis rates among men arecoded as alcoholic and the corresponding figure forwomen is one-third, which in a western European con-text is the average (Ramstedt 2002). These proportionsvary to some extent between provinces, with the lowestfraction found in Newfoundland and the highest in Brit-ish Columbia.

The geographical relationship between per capita consumption and cirrhosis mortality

Both theoretical considerations and empirical findingssuggest that the higher the per capita consumption in apopulation, the higher the rate of alcohol-related mortal-ity (Edwards

et al

. 1994). We should therefore expect thatprovincial variations in cirrhosis mortality correspond to


Addiction,

98

, 1267–1276

1270

Mats Ramstedt

variations in alcohol consumption. Figure 1 elucidatesthe relationship between per capita consumption and (a)total cirrhosis (b), alcoholic cirrhosis and (c) non-alco-holic cirrhosis among men. The variables are trans-formed (as described earlier) in order to fulfil theunderlying assumption of an exponential risk functionand a lagged association between alcohol and cirrhosis.As expected, a higher per capita consumption is generallyassociated with more deaths from total cirrhosis amongmen although a few deviations can be noticed; Quebecscores somewhat higher than expected, whereas New-foundland have lower rates than predicted.

It is reasonable to assume a closer correspondencebetween alcoholic cirrhosis and per capita consump-tion. Furthermore, as the fraction of alcoholic casesshould be lower for cirrhosis deaths without mention ofalcohol, we assume a weaker relationship in this case.Both assumptions gain support in Fig. 1. Variations inalcoholic cirrhosis match variations in per capita con-sumption better than total cirrhosis (with

R

2

increas-ing from 0.37 to 0.60) and provincial differences innon-alcoholic cirrhosis show no correspondence withvariations in per capita alcohol consumption. The samepattern is found for women; a consistent positive rela-tionship between per capita consumption and total cir-rhosis, an even more consistent association withalcoholic cirrhosis and a zero correlation with non-alcoholic cirrhosis (Fig. 2).

These analyses thus suggest that the alcoholic cirrho-sis rate is the best measure of differences in alcohol-related harm across Canadian provinces. The validity ofthis measure is also substantiated by the fact that menhad higher fractions of alcoholic cirrhosis than womenhave (Table 1) and that provinces with high consumptiontend to have higher proportions of alcoholic cirrhosisamong both men and women (Fig. 3).

The temporal association between per capita alcohol consumption and liver cirrhosis rates

Figure 4 presents post-war trends in per capita con-sumption, total cirrhosis and non-alcoholic cirrhosis inCanada and seven provinces among men and women.The typical pattern involves increasing alcohol con-sumption and total cirrhosis rates from 1950 up untilaround 1975 and then a subsequent decline, which ishalted in the mid-1990s. The most notable exceptionsare Alberta and British Columbia, with a peak some-what later, around 1980, in both consumption and cir-rhosis. However, seen over the whole period, there seemsto be a fairly close match between these trends. Alsotrends in non-alcoholic cirrhosis rates correspond to thispost-war pattern, as is seen most clearly for Canada as awhole. The only notable exceptions in this respect areBritish Columbia (men and women) and Saskatchewan(women).

In order to assess if this could be related to a causalprocess, these time-series are analysed with ARIMAtime-series models. For Canada as a whole, per capitaconsumption is associated significantly with all three cat-egories of mortality. For total cirrhosis, the effect param-eter is 0.16, whereas the corresponding outcome foralcoholic cirrhosis is 0.27 and 0.12 for non-alcoholic cir-rhosis deaths (Table 2). This means that a 1-litre increasein per capita consumption on average was followed by a17% increase [calculated as exp(0.16)] in male cirrhosis,a 31% increase in alcoholic cirrhosis and a 13% increasein non-alcoholic cirrhosis.

The alcohol effect on male total cirrhosis variesbetween 0.17 and 0.20 in most provinces but is some-what weaker in Ontario (0.13). These small differences inalcohol effects are most probably a matter of chance,which is suggested by the

F

-test for heterogeneity. More-

Alcoholconsumption

Abstainers(%)

Men Women

Totalcirrhosis

Alcoholic(%)

Totalcirrhosis

Alcoholic

(%)

Canada 8.7 22.3 17.4 46 7.9 36Provinces

British Columbia 10.0 17.1 18.6 61 10.0 53Alberta 9.4 18.1 14.8 51 7.5 45Manitoba 8.5 20.7 14.1 43 7.6 34Saskatchewan 7.6 21.6 12.0 42 5.4 31Ontario 9.2 22.4 18.7 46 8.4 37Quebec 8.0 23.6 20.1 39 7.9 24

Maritimes 7.3 31.0 10.9 33 4.9 20Nova Scotia 7.4 28.8 11.3 34 5.1 22New Brunswick 6.7 32.0 12.1 34 5.1 18Newfoundland 7.4 32.4 8.4 26 4.2 16

Table 1

Per capita alcohol consumption(litres alcohol 100%, 15

+

), abstainers (esti-mates from 1989) and liver cirrhosis rates(age-adjusted per 100 000) in Canada.Average for 1950–98.


1271


Addiction,

98

, 1267–1276

over, the pooled estimate of 0.17 is similar to the estimatefor whole Canada.

The association between per capita consumption andalcoholic cirrhosis is positive and statistically significantin all provinces except Manitoba. The estimates are stron-ger than for total cirrhosis, ranging from 0.21 (Mari-times) to 0.35 (British Columbia). Again, these regionaldifferences are not large enough to exclude randomness(

F

-value

=

0.93). By pooling the regional effects weobtain a parameter estimate of 0.27, thus the same as forwhole Canada. Interestingly, and in contrast to the geo-graphical relationship, also male non-alcoholic cirrhosisrates are generally associated significantly with alcoholconsumption. Statistical significance is reached in all

provinces but the Maritimes and the size of the estimatesis lower than for the two other mortality categories exceptfrom in Manitoba and Quebec. However, the test for het-erogeneity is not significant and the pooled alcohol effectis almost identical to the one obtained for Canada in total,i.e. 0.12.

This outcome thus suggests that alcohol actuallycould be involved in many deaths recorded as not beingrelated to alcohol, although it does not reveal to whatextent this is the case. However, one way of obtaining atleast a rough estimate is to calculate an attributable frac-tion, i.e. an estimation of the fraction of cases that can beattributed to alcohol (see, e.g. Norström 1989). The for-mula for the attributable fraction in this case is: AF

=

1 – exp(– bX) where X is alcohol consumption (periodaverage) and b is the estimated effect parameter. By apply-

Figure 1

Relationship between alcohol consumption [litres 100%per capita 15

+

and (a) male liver cirrhosis mortality, (b) male alco-holic liver cirrhosis mortality and (c) male non-alcoholic liver cirrho-sis mortality]. Average figures for each province for 1950–98 (ALB:Alberta; BC: British Columbia; MAN: Manitoba; NB: New Brun-swick; NF: Newfoundland; NS: Nova Scotia; ONT: Ontario; QB:Quebec)

(a)

Per capita alcohol consumption (weighted)

1211109876

Mal

e ci

rrho

sis

mor

talit

y (l

n)

3.5

3.0

2.5

2.0

1.5

1.0

0.5Rsq=0.37

NF

BCALB

SASMAN

ONTQB

NB NS

(c)


1211109876

Mal

e no

n-al

coho

lic

cirr

hosi

s (l

n)

3.5

3.0

2.5

2.0

1.5

1.0

0.5Rsq=0.03

NF BCALBSAS

MANONT

QB

NB NS

(b)


1211109876

Mal

e al

coho

lic c

irrh

osis

mor

talit

y (l

n)

3.5

3.0

2.5

2.0

1.5

1.0

0.5Rsq=0.60

NF

BC

ALB

SASMAN

ONTQB

NBNS

Figure 2

Relationship between alcohol consumption (litres 100%per capita 15

+

) and (a) female liver cirrhosis mortality, (b) femalealcoholic liver cirrhosis mortality and (c) female non-alcoholic livercirrhosis mortality. Average figures for each province for 1950–98

(a)


1211109876

Fem

ale

cirr

hosi

s m

orta

lity

(ln)

3.0

2.5

2.0

1.5

1.0

0.5

0.0Rsq=0.72

NF

BC

ALB

SAS

MANONTQB

NB NS


1211109876

Fem

ale

alco

holic

cir

rhos

is m

orta

lity

(ln)

3.0

2.5

2.0

1.5

1.0

0.5

0.0Rsq=0.88NF

BC

ALB

SAS

MANONT

QB

NB NS


1211109876

Fem

ale

non-

alco

holic

cir

rhos

is (

ln)

3.0

2.5

2.0

1.5

1.0

0.5

0.0Rsq=0.05

NFBCALB

SAS

MANONTQB

NB NS

(b)

(c)


Addiction,

98

, 1267–1276

1272

Mats Ramstedt

ing this technique, it is estimated that 64% of male cir-rhosis deaths recorded as non-alcoholic actually has aconnection with alcohol. As expected, the correspondingestimates for total cirrhosis and alcoholic cirrhosis arelarger, 76 and 92%, respectively.

A 1-litre change in per capita alcohol consumptionwas also significantly related to all three categories of cir-rhosis mortality for women in Canada as a whole(Table 3).

The estimates for women’s total and non-alcoholiccirrhosis are weaker than for men (0.12 and 0.06,respectively), but the effect for alcoholic cirrhosis is some-what larger (0.30). The alcohol effect parameters for totalcirrhosis are significant in all provinces and fall withinthe range of 0.09–0.17 but no significant regional vari-ation was indicated by the

F

-test. Moreover, the pooledestimate and the effect for the whole of Canada are prac-tically identical.

A somewhat stronger and generally statistically sig-nificant alcohol effect was found for female alcoholic cir-rhosis in the provinces. Although some provincialvariations can be seen they are due most probably to

chance (

F

-value

=

0.271). In addition, female non-alco-holic cirrhosis rates turned out to be related to overallconsumption of alcohol in most provinces. In fact, thelink was positive in all provinces except Alberta andreached statistical significance in five provinces.

DISCUSSION

The present analyses support two well-established find-ings in alcohol research. To begin with, the Canadiancase verifies that per capita alcohol consumption isclosely related to death rates from liver cirrhosis. In par-ticular, the magnitude of this association was found tobe similar across the Canadian provinces and both maleand female cirrhosis increased with a rise in the overalllevel of drinking. Since practically identical time-seriesanalyses have been carried out in 14 western Europeancountries (Ramstedt 2001), it is possible to put theresults in a comparative perspective. We then see thatthe increase of 17% in total male liver cirrhosis mortal-ity per litre alcohol in Canada is higher than in south-ern Europe (10%) and central Europe (8%), whereas itis weaker than in northern Europe (27%). Similar dif-ferences were observed for women. Whether this out-come is related to variations in drinking patterns is onlya matter of speculation. However, recent research sug-gests that not only average amount of drinking but alsothe frequency of episodic heavy drinking contribute tothe risk of developing cirrhosis (Hsiao-ye & Dufour2002).

In accordance with findings in western Europeancountries, the effect of a change in overall consumptionwas generally stronger for male cirrhosis rates; this is rea-sonable as a result of men’s higher share of overall con-sumption. If the underlying risk function implies that aconsumption increase of a given volume entails a certainpercentage increase in mortality risk, we should thusexpect that men’s mortality respond stronger to changesin per capita consumption than does the mortality ofwomen. However, for alcoholic cirrhosis this expectationwas not fulfilled because the alcohol effect was about thesame for men and women. Considering the uncertaintyof women’s fraction of overall alcohol consumption andthe relatively low female rate of alcoholic cirrhosis, onemust be careful in the interpretation of this outcome.However, it corresponds with the idea of a higher femalesusceptibility to liver cirrhosis suggested by several stud-ies (e.g. Bellentani

et al

. 1988).Another finding that has gained support in the

present study is that alcohol-related deaths tend to beunder-reported in official mortality statistics. Thishypothesis, often based on prospective analyses of medi-cal records, was supported here by the positive and sta-

Figure 3

Relationship between alcohol consumption (litres 100%per capita 15

+

) and (1) the fraction (%) of liver cirrhosis deathsspecified as alcoholic; men and women. Average figures for eachprovince for 1950–98


1110987

Fra

ctio

n of

mal

e al

coho

lic c

irrh

osis

(%

)

70

60

50

40

30

20Rsq=0.77

NF

BC

ALB

SAS MAN

ONT

QB

NB NS

Per capita alcohol consumption (weighted) 1110987

Frac

tion

of

fem

ale

alco

holic

cir

rhos

is (

%)

60

50

40

30

20

10 Rsq=0.87 NF

BC

ALB

SAS

MAN

ONT

QB

NB

NS


1273


Addiction,

98

, 1267–1276

Figure 4

Trends in alcohol consumption per inhabitant 15

+

and gender-specific mortality from total cirrhosis and non-alcoholic cirrhosisfor Canada and provinces between 1950 and 1998

95908580757065605550

30

20

10

0

Alcohol

Total cirm

Non-alc cirm

Total cirf

Non-alc cirf95908580757065605550

40

30

20

10

0

Alcohol

Total cirm

Non-alc cirm

Total cirf

Non-alc cirf

95908580757065605550

30

20

10

0

Alcohol

Total cirm

Non-alc cirm

Total cirf

Non-alc cirf

95908580757065605550

30

20

10

0

Alcohol

Total cirm

Non-alc cirm

Total cirf

Non-alc cirf95908580757065605550

40

30

20

10

0

Alcohol

Total cirm

Non-alc cirm

Total cirf

Non-alc cirf

95908580757065605550

40

30

20

10

0

Alcohol

Total cirm

Non-alc cirm

Total cirf

Non-alc cirf95908580757065605550

30

20

10

0

Alcohol

Total cirm

Non-alc cirm

Total cirf

Non-alc cirf

Manitoba

British ColumbiaCanada

Alberta

Saskatchewan

Quebec The Maritimes

Ontario

Alcohol

Total cirm

Non-alc cirm

Total cirf

Non-alc cirf95908580757065605550

30

20

10

0


Addiction,

98

, 1267–1276

1274

Mats Ramstedt

tistically significant temporal association found betweenalcohol consumption and cirrhosis mortality withoutmention of alcohol. In fact, when the estimated increasein male non-alcoholic cirrhosis with 13% from a 1-litreincrease in per capita alcohol consumption wasexpressed as an attributable fraction, it turned out thatnearly 65% in this category could be related to alcohol.This is a somewhat higher proportion than was assumedin a meta-study assessing the role of alcohol in mortalityin Canada, where 50% of non-alcoholic cirrhosis deathswere held to be alcohol-related (Single

et al

. 1999).The reason for this difference can, of course, be related

to the different methodological approaches. However, theimportance of virus liver diseases for liver cirrhosis mor-

tality is another possibility and is also a conceivable con-founder more generally. There are now several studiesdemonstrating an additive or multiplicative effect of hep-atitis C infection and alcohol on liver cirrhosis (e.g. Cor-rao & Arico 1998; Peters & Terrault 2002), and this couldhave produced a somewhat exaggerated alcohol effect inthis analysis (and a corresponding overstated AF).Because hepatitis C and B infections have increased inrecent decades in Canada (elSaadany

et al

. 2002; Pohani2001), one rough test of the implications of this develop-ment is to examine whether the obtained results are alsovalid for an earlier part of the study period. However, byconducting the same analyses for 1950–89 no change inthe gender-specific alcohol effects was found, neither for

Table 3

Estimated effect of (weighted) total alcohol consumption (litres of pure alcohol per inhabitants 15 years and above) on female totalliver cirrhosis mortality, alcoholic cirrhosis and cirrhosis without mention of alcohol in Canada (15 years and above, age-standardized). Semi-logarithmic models estimated on differenced data.

Total cirrhosis Alcoholic cirrhosis Non-alcoholic cirrhosis

Estimate SE Model Estimate SE Model Estimate SE Model

Canada 0.123*** 0.032 0,1,1 0.302** 0.092 0,1,1 0.065** 0.019 0,1,1British Colombia 0.174* 0.066 0,1,1 0.227 0.146 0,1,0

−

0.013 0.137 1,1,0Alberta 0.154** 0.044 2,1,0 0.299*** 0.088 2,1,0 0.091 0.077 2,1,0Manitoba 0.141*** 0.023 0,1,1 0.201*** 0.004 0,1,1 0.114*** 0.031 0,1,1Saskatchewan 0.095** 0.023 0,1,1 0.208 0.190 2,1,0 0.103*** 0.035 0,1,1Ontario 0.107** 0.035 2,1,0 0.360** 0.119 2,1,0 0.072 0.056 0,1,1Quebec 0.123*** 0.026 0,1,1 0.227* 0.098 2,1,0 0.197* 0.079 2,1,0Maritimes 0.096*** 0.024 0,1,1 0.209*** 0.054 0,1,1 0.076* 0.031 0,1,1

F

-test for 0.657 NS 0.271 NS 0.757 NSheterogenityPooled estimate 0.123*** 0.017 0.278*** 0.055 0.089*** 0.024

*

P

< 0.05,

**

P

< 0.01

***

P

< 0.001.

Q

-tests for residual autocorrelation are satisfactory in all models.

Table 2

Estimated effect of (weighted) total alcohol consumption (litres of pure alcohol per inhabitants 15 years and above) on male totalliver cirrhosis mortality, alcoholic cirrhosis and cirrhosis without mention of alcohol in Canada (15 years and above, age-standardized). Semi-logarithmic models estimated on differenced data.

Total cirrhosis Alcoholic cirrhosis Non-alcoholic cirrhosis

Estimate SE Model Estimate SE Model Estimate SE Model

Canada 0.164*** 0.043 0,1,0 0.274*** 0.074 0,1,0 0.122* 0.056 0,1,0British Colombia 0.206** 0.063 0,1,1 0.352** 0.123 0,1,1

-

0.080 0.224 2,1,0Alberta 0.192** 0.059 2,1,0 0.317** 0.097 2,1,0 0.114*** 0.033 0,1,1Manitoba 0.198*** 0.035 0,1,1 0.065 0.116 0,1,1 0.214*** 0.033 0,1,1Saskatchewan 0.170*** 0.021 0,1,1 0.271*** 0.043 0,1,1 0.147*** 0.025 0,1,1Ontario 0.136* 0.060 0,1,0 0.279*** 0.071 0,1,1 0.114* 0.074 0,1,1Quebec 0.203*** 0.038 0,1,1 0.246* 0.098 0,1,1 0.206** 0.067 0,1,1Maritimes 0.167* 0.073 2,1,0 0.213* 0.101 2,1,0 0.071 0.159 2,1,0

F

-test for 0.231 NS· 0.934 NS· 0.780 NSheterogenityPooled estimate 0.173*** 0.026 0.266*** 0.119** 0.04

*

P

<

0.05,

**

P

<

0.01

***

P

<

0.001.

Q

-tests for residual autocorrelation are satisfactory in all models.


1275


Addiction,

98

, 1267–1276

non-alcoholic cirrhosis nor total and alcoholic cirrhosis(results not shown). Thus, the hypothesis that increasinghepatitis infections rates have been influencing the alco-hol effect parameters obtained here did not gain support.

As an inference of causality is always at risk by omit-ted causal factors in this type of analysis it is worth not-ing that a coherent picture was obtained from bothtemporal and geographical analyses with regard to therelation between alcohol and total cirrhosis and particu-larly alcoholic cirrhosis. Moreover, differencing and sta-tistical modelling of the noise term in the time-seriesanalyses must be considered as strong precautionarymeasures against omitted variable bias. In fact, sinceyearly changes were analysed any confounding factorhas to covary with yearly changes in both alcohol andcirrhosis, in order to produce biased alcohol effectparameters. The different outcome for the temporal andgeographical analyses analysis of non-alcoholic cirrhosisillustrates the well-known difficulty in relying uponcross-regional covariation in the assessment of causalrelationships (see, e.g. Norström & Skog 2001). Becauseother factors besides alcohol ought to be more influentialfor the non-alcoholic cirrhosis rate and these factors maybe of different importance in the provinces, the alcoholeffect was only possible to identify with a careful statisti-cal analysis.

There are time periods when trends in consumptionand cirrhosis do not match as closely as expected, whichwas the case in Ontario between 1975 and 1983 (Smart& Mann 1987). However, according to the present study,such examples of a mismatch are rare and do not chal-lenge the general conclusion; per capita alcohol con-sumption and liver cirrhosis mortality are closely relatedphenomena in Canada.

ACKNOWLEDGEMENTS

The paper is written within the framework of the projectCanadian Alcohol Experiences and Nordic Perspectives,which is supported financially by the Swedish Ministry ofHealth and Social Affairs. Kari Poikolainen, William Kerr,Sturla Nordlund, Ole-Jørgen Skog and Robin Room havegiven valuable comments on earlier versions.

REFERENCES

Bellentani, S., Pozzato, G., Saccoccio, G., Crovatto, M., Croce,L. S., Mazzoran, L., Masutti, F., Cristianini, G. & Tiribelli, C.(1988) Clinical course and risk factors of hepatitis C virusrelated liver disease in the general population: report from theDionysos study.

Gut

,

44

, 874–880.Blake, J. E., Compton, K. V., Schmidt, W. & Orrego, H. (1988)

Accuracy of death certificates in the diagnosis of alcoholic

liver cirrhosis.

Alcoholism: Clinical and Experimental Research

,

12

, 168–172.Bruun, K., Edwards, G., Lumio, M., Mäkleä, K., Pan, L., Popham,

R. E., Room, R., Schmidt, W., Skog, O. J., Sulkunen, P. & Öster-berg, E. (1975)

Alcohol Control Policies in Public Health Perspec-tive.

Helsinki: Finnish Foundation for Alcohol Studies.Corrao, G. & Arico, S. (1998) Independent and combined action

of hepatitis C virus infection and alcohol consumption on therisk of symptomatic liver cirrhosis.

Hepatology

,

27

, 914–919.Corrao, G., Bagnardi, V., Zambon, A. & Arico, S. (1999) Explor-

ing the dose–response relationship between alcohol consump-tion and the risk of several alcohol-related conditions: a meta-analysis.

Addiction

,

94

, 1551–1573.Edwards, G., Anderson, P., Babor, T. F., Casswell, S., Ferrence,

R., Giesbrecht, N., Godfrey, C., Holder, H. D., Lemmens, P.,Mäkelä, K., Midanik, L. T., Norström, T., Österberg, E.,Romelsjö, A., Room, R., Simpura, J. & Skog, O.-J. (1994)

Alcohol Policy and the Public Good.

Oxford: Oxford UniversityPress.

Harper, D. W., MacRae, L. & Lange, D. (1981) ‘Substitution’,‘restraint’ and reduction during the Manitoba beer and liquorstrikes of 1978.

Journal of Studies on Alcohol

,

42

, 132–135.Health and Welfare Canada (1992)

Alcohol and other drug use byCanadians: a national alcohol and other drugs survey.

Ontario:Health and Welfare Canada.

Hsiao-ye, Y. & Dufour, M. C. (2002)

Average daily alcohol con-sumption, episodic heavy drinking, and liver diseases.

Poster pre-sented at the 25th Annual Scientific Meeting of the ResearchSociety on Alcoholism, 28 June

-

3 July 2002, San Francisco,CA.

Lelbach, W. K. (1974) Organic pathology related to Volume andpattern of alcohol use. In: Gibbins, R. J., Israel, Y., Kalant, H.,Popham, R. E., Schmidt, W. & Smart, R. G., eds.

ResearchAdvances in Alcohol and Drug Problems

, vol. 1, pp. 93–198. NewYork: John Wiley & Sons.

Norström, T. (1987) The impact of per capita consumption onSwedish cirrhosis mortality.

British Journal of Addiction

,

82

,67–75.

Norström, T. (1989) The use of aggregate data in alcohol epide-miology.

British Journal of Addiction

,

84

, 969–977.Norström, T. & Skog, O.-J. (2001) Alcohol and mortality: meth-

odological and analytical issues in aggregate analyses.

Addic-tion

,

96

, S5–S18.Pequignot, G., Tuyns, A. & Berta, J. L. (1978) Ascitic cirrhosis in

relation to alcohol consumption.

International Journal of Epide-miology

,

7

, 113–120.Peters, M. G. & Terrault, N. A. (2002) Alcohol use and hepatitis

C.

Hepatology

,

36

, 220–225.Pohani, G., Zou, S. & Tepper, M. (2001) Trends of hepatitis B and

hepatitis C mortality in Canada, 1979–97.

Canadian Journal ofPublic Health

,

92

, 250–254.Popham, R. E. (1970) Indirect methods of alcoholism prevalence

estimation: a critical evaluation. In: Popham, R. E., ed.

Alcoholand Alcoholism

, pp. 678–685. Toronto: University of TorontoPress.

Ramstedt, M. (2001) Per capita alcohol consumption and livercirrhosis mortality in 14 European countries.

Addiction

,

96

,S19–S34.

Ramstedt, M. (2002) Alcohol-related mortality in 15 Europeancountries in the post-war period.

European Journal of Popula-tion

,

18

, 307–323.el-Saadany, S., Tepper, M., Mao, Y., Semenciw, R. & Giulivi, A.

(2002) An epidemiological study of hepatocellular carcinomain Canada.

Canadian Journal of Public Health

,

93

, 443–446.


Addiction,

98

, 1267–1276

1276 Mats Ramstedt

Schmidt, W. (1977) Cirrhosis and alcohol consumption: an epi-demiological perspective. In: Edwards, G. & Grant, M., eds.Alcoholism: New Knowledge and New Responses, pp. 15–47.London: Croom Helm.

Single, E. (1979) The ‘substitution hypothesis’ reconsidered; aresearch note concerning the Ontario beer strikes in 1958and 1968. Journal of Studies on Alcohol, 40, 485–491.

Single, E., Robson, L., Rehm, J. & Xie, X. (1999) Morbidityand mortality attributable to alcohol, tobacco and illicitdrug use in Canada. American Journal of Public Health, 89,385–390.

Skog, O.-J. (1980) Liver cirrhosis epidemiology: some meth-

odological problems. British Journal of Addiction, 5, 227–243.

Skog, O.-J. (1984) The risk function for liver cirrhosis from life-time alcohol consumption. Journal of Studies on Alcohol, 45,199–208.

Smart, R. & Mann, R. E. (1987) Large decreases in alcohol-related problems following a slight reduction in alcohol con-sumption in Ontario 1975–83. British Journal of Addiction, 82,285–291.

World Health Organization (2000) International Guide for Moni-toring Alcohol Consumption and Related Harm. Geneva: WorldHealth Organization.

alcohol consumption and liver cirrhosis mortality with and without mention of alcohol—the case of...

Documents