Aspects of tobacco attributable mortality: systematic review

The objective of the article was to assess methodologies published and applied in calculating mortality attributable to smoking. A review of the literature was made for the period 1990 to 2006, in the electronic databases MEDLINE and LILACS. A total of 186 studies were found, which measured mortality based on calculating the smoking-attributable risk. Of these, a total of 41 were selected. The studies that were carried out in the United States and Canada presented a more standard methodology and reported smoking attributable mortality to be 18%-23%, with male mortality being 25%-29% and female mortality 14%-17%. The variations can be attributed to methodological differences and to different estimates of the main tobacco-related illnesses. DESCRIPTORS: Smoking, mortality. Attributable risk. Epidemiologic studies. Review [Publication type]. Revisão Sistemática | Systematic Review 336 Mortalidade atribuível ao tabagismo Oliveira AF et al O tabaco é a droga mais utilizada e disseminada no mundo, responsável por 50% de aproximadamente cinco milhões de mortes registradas no ano 2000 nos países em desenvolvimento.19,65 Estima-se que no período de 2002/2030 as mortes atribuíveis ao tabaco irão diminuir em 9% em países desenvolvidos, mas aumentar em 100% (para 6,8 milhões) em países em desenvolvimento. Estima-se que em 2015 as mortes relacionadas ao fumo superarão em 50% aquelas causadas pela epidemia de HIV/Aids e que o tabaco será responsável por aproximadamente 10% de todas as mortes no mundo.38 Revisão sistemática de 139 estudos sobre a prevalência do tabagismo em adultos encontrou que mais de 1,1 bilhão de pessoas em todo o mundo fumam, dos quais 82% dos fumantes residiam nos países em desenvolvimento.31 Em 2000, as maiores prevalências de tabagismo no mundo foram encontradas no sexo masculino, ainda que a diferença entre os gêneros tenha diminuído nos países desenvolvidos (37% entre homens e 21% em mulheres). Na América Latina e Caribe, essa prevalência foi estimada em 32% para 2000, sendo 40% no sexo masculino e 24% no sexo feminino.32 Os prejuízos causados à saúde pelo hábito de fumar são amplamente conhecidos e seu controle é considerado pela OMS como um dos maiores desafi os da saúde pública atualmente.26 Há fortes evidências de que o tabaco faça parte da cadeia de causalidade de quase 50 diferentes doenças, destacando-se o grupo das doenças cardiovasculares, cânceres e doenças respiratórias.59,60,64 Ezzati et al20 (2005) estimaram que 11% de todas as mortes cardiovasculares ocorridas no mundo em 2000 poderiam ser atribuídas ao tabaco, embora com maior destaque para as doenças isquêmicas do coração e cerebrovasculares. Além disso, o tabaco seria ainda responsável por 21% de todas as mortes por câncer no mundo, chegando a 29% e 18% nos países desenvolvidos e em desenvolvimento, respectivamente.15 O impacto do tabagismo na sociedade é multifatorial, podendo ser medido em várias dimensões, como a carga de mortalidade. Esta pode ser medida por meio das mortes atribuídas ao tabagismo.62 A mortalidade atribuível ao tabaco (smoking attributable mortality – SAM) tem sido amplamente utilizada nos estudos e apontada como uma das estatísticassumário de maior relevância, devido à sua capacidade de mostrar os prejuízos para saúde advindos do tabagismo.63 Todavia, Tanuseputro et al54 (2005) têm apontado problemas metodológicos no cálculo dessas estimativas. A SAM tem sido utilizada nos estudos sob a forma de SAM%, ou seja, do total de óbitos gerais ou por determinada causa específica, quantos são atribuíveis ao tabagismo. O objetivo do presente estudo foi analisar as metodologias publicadas e empregadas no cálculo da mortalidade atribuível ao fumo.


INTRODUCTION
Tobacco, the most widespread and widely used drug in the world, was responsible for approximately 50% of 5 million offi cial deaths in 2000, in developing countries. 19,65It is estimated that in the period 2002/2030, tobacco-attributable deaths will decrease by 9% in developed countries, but increase by 100% (to 6.8 million) in developing countries.It is also estimated that by 2015, smoking related deaths will be 50% more than those caused by the HIV/AIDS epidemic and that tobacco will be responsible for approximately 10% of all deaths on the planet. 38systematic review of 139 studies concerned with the prevalence of tobacco smoking in adults found that more than 1.1 billion people across the world smoke, of which 82% reside in developing countries. 31In 2000, the prevalence of tobacco smoking in the world was greater among men, although the difference between the sexes has been decreasing in developed countries (prevalence is 37% among men and 21% among women).In the Latin America and Caribbean region, the prevalence was estimated to be 32% in 2000, which corresponds to 40% among men and 24% among women. 32e negative health effects caused by cigarette smoking are well documented and the control of smoking is considered by the WHO to be one of the greatest present challenges to public health. 26ere is strong evidence to suggest that tobacco forms part of the causality chain of nearly 50 different illnesses, in particular cardiovascular diseases, cancer and respiratory illness. 59,60,64zati et al 20 (2005) estimated that 11% of all cardiovascular deaths in the world in 2000 could be attributed to tobacco, in particular ischemic heart disease and cerebrovascular disease.In addition, cancer has been attributed to 21% of all cancer deaths in the world, including 29% of deaths in developed countries and 18% in developing countries. 15he various impacts that tobacco has on society can be measured in number of ways, such as the mortality burden, which represents tobacco-attributable deaths. 62oking attributable mortality (SAM) has been widely used in studies and is considered to be one of the most relevant summary statistics, due to its capacity to show the harm that tobacco causes to health. 63However, some methodological problems in the calculation of its estimates have been found (Tanuseputro et al, 2005). 54AM has been used in studies in the form SAM%, meaning that of all deaths in general or of those with a specifi c cause, the proportion that are attributable to tobacco.
The objective of the study was to analyze the methodologies used and published to calculate smoking attributable mortality.

METHODS
In May 2006, a review of the MEDLINE and LILACS electronic databases was carried out for the period 1990 to 2006.Search terms taken from the Medical Subject Headings (MeSH) were used, including "attributable risk", "mortality", "smoking" and the key words "tobacco", "smoking habit".The systematic review method was used to analyze the studies.A total of 186 articles were found, 30 of which were selected since they were concerned with SAM as a method for calculating the attributable risk in a given population.As a result of this fi rst review, a further 11 articles and abstracts were identifi ed, of which three from the 1970s and 1980s were included since they were among the most cited articles.In this way, a total of 41 articles were included in the analysis.
Articles published in Portuguese, Spanish, English, French and Italian were included; those in other languages were excluded irrespective of whether they contained a summary in English.Another criterion for their inclusion was the measurement of SAM based on the calculation of the population attributable fraction (PAF).The PAF uses parameters relating to the prevalence of smoking according to the level of exposure (smokers, ex-smokers and non-smokers) and the relative risk (RR) of death from tobacco-related illnesses.Potential causes of error that are normally ignored in the calculation of the PAF include: uncertainties about present and past exposure to smoking, the use of estimates for prevalence, mortality or relative risk by stratum, and the long latency period between exposure and occurrence of the disease.If estimates are to be more applicable and accurate, these variables need to be taken into consideration.The PAF is useful for estimating the proportion of cases of a disease that could have been prevented with reduction or elimination of the risk factor. 45,47ta extraction from the chosen articles was carried out by just one reviewer using a pre-determined method.The following information was gathered: authors, location where the study was carried out, publication year, study period, age or age range of the population under study, way of calculating the SAM, parameters used for this calculation, main fi ndings and limitations or problems identifi ed.The SAM is obtained by multiplying the number of deaths for each tobacco-related disease by the population attributable fraction, PAF.
In the 1980s, the Centre for Disease Control and Prevention (CDC) created a software called SAMMEC (Smoking-Attributable Mortality, Morbidity, and Economic Costs Software, version II), with the aim of calculating the SAM and thus estimating the impact of tobacco-related diseases.This software allows for the rapid calculation of deaths, years of life lost, direct costs of health care, indirect costs of death and costs of smoking associated incapacity. 49The SAMMEC was used as a criterion for evaluating the scientifi c articles.It uses 22 tobacco related diseases in adults, four in children (resulting from mothers who smoke), RR drawn from the Cancer Prevention Study 59 (CPS) II with a calculation method that measures prevalence according to the level of exposure to smoking in different countries and smoking attributable deaths by burning.Those articles that fulfi ll these criteria and also include deaths resulting from passive smoking scored higher points.The other articles score proportionally lower marks in relation to the criteria.

RESULTS
Table 1 shows the results of the articles, by author/year, location of publication, period of the study, age range under analysis and method of calculating the SAM.Table 2 presents the principle fi ndings and general characteristics of the studies.
The studies were organized according to the methods used for calculating the SAM.Firstly those articles that scored highest points in the criteria for evaluating the methodology were taken into consideration, followed by those that involved one or more uncertainty or discrepancy.
Deaths associated with passive smoking were not calculated in most of the studies.However, some used deaths associated with lung cancer and heart disease among non-smokers as estimates. 11,12,29,62In addition to these illnesses, other studies included cerebrovascular diseases 63 and lung cancer alone. 6,7,8,13,27Some studies also failed to mention the method used for the calculation. 28,42,51,59timates for deaths attributed to passive smoking and fi res were, in the majority of studies, drawn from national studies or used relative risk estimates taken from studies that address this question.
Seven studies received a positive evaluation for their calculation of the SAM %. 6,7,8,13,27,42,51 These studies also received a higher scoring because they included in the general calculation of the SAM deaths resulting from passive smoking (Table 1).
Three methods for calculating the PAF were identifi ed (Table 1): PAF% -the proportion of smoking attributed deaths in a population: PAF% = Pi(RRi-1)/[1+P(RRi-1)] where P is the prevalence of exposure to smoking in the population and RR is the relative risk of death (among smokers and ex-smokers), compared with non smokers.The SAM is calculated by multiplying the PAF% by the number of deaths in each disease category.Number 3 is derived from this formula.PAF% includes the incidence rate of selected causes of death in the general population, and in smokers, non-smokers and ex-smokers.The proportion of smokers to non-smokers and the relative risk of death among smokers and non-smokers are also taken into consideration.The SAM is calculated by multiplying the PAF% by the number of deaths in each disease category.

3.
among women.In developed countries, this fi gure rose to 19% and in developing countries was 9%.Peto et al 44 (1996) observed that in the 44 developed countries that were analyzed, tobacco was responsible for 24% of all deaths in men and 7% of all deaths in women.In the studies assessed here, the general SAM was between 18% and 23%.In men, the rate was between 25.4% and 29.0% and in women, between 14% and 17% (Table 2).
In the USA and Canada, values for the SAM varied between 15% and 23% and in European countries, between 13.7% and 24.0%.In some Latin American countries, such as Mexico and Porto Rico, the values ranged from 4.2% and 11.4% respectively (Table 2).
As well as the general SAM, articles that calculate the SAM for the four principle tobacco attributable illnesses were also taken into consideration (lung cancer, chronic obstructive pulmonary disease -COPD -cerebrovascular diseases and ischemic heart disease).Figure 1 shows the values and the confi dence intervals for the SAM% -lung cancer and COPD -by sex and age range in the different studies.
Studies (Figure 1) show that an important proportion of deaths by lung cancer and COPD among men are attributable to tobacco, independent of age, with more precise confi dence intervals.The same is not true for women, for whom there are signifi cant variations in the SAM% and wide confi dence intervals.
Cardiovascular and ischemic heart diseases were the only illnesses that showed differences in the age range for the calculation of the SAM% in the studies (35-64 and 65 and over) in relation to the specifi c values of the RR for these illnesses.This was not the case in some articles 14,46,48 that used other age ranges.Figure 2 makes a comparison of these two diseases by sex and age range.

DISCUSSION
A comparison between the results that Ezzati & Lopez 18 (2003) and Peto et al 44 (1996) reported for SAM% with those from studies that use a more uniform methodology (USA and Canada) shows that the general mortality rate (18% -23%; including 25% -29% amongst men and 14% -17% amongst women) was higher for the world and for developed countries in the former studies.
Those tobacco related illnesses that most contribute towards the SAM were cancer of the trachea/ bronchial/ lungs, 2,8,23,51 ischemic heart disease, 11,29,30,42,50,51 COPD 22,66 and cerebrovascular diseases. 3,22,23zati & Lopez 19 (2004) also found cardiovascular disease, COPD and lung cancer to be the three principal causes of smoking related deaths in developed and developing countries in the year 2000.
It is widely recognized that a considerable number of deaths occur among people aged 65 or over, resulting from ischemic heart and cerebrovascular diseases.Tobacco and other risk factors have been shown to be important causes for these deaths (González Enríquez et al, 22 1997).The SAM is low for these diseases in the 65 and over age group, when compared with the 35-64 age group, in which the number of deaths is fewer but the percentage of tobacco attributable deaths is high (40% -60%), mainly among men.This involves a young adult population that is economically active and dies early from a modifi able risk factor that could be reduced or even eliminated if measures to promote and prevent tobacco addiction amongst younger age groups were established as public health policies.
The differences that were observed in the SAM for the four principle diseases associated with tobacco may refl ect not only the methodological differences in the studies, but also the different prevalences of smoking that are used to calculate the FAF in the different countries.
The studies reviewed here are quite heterogeneous in many aspects: the method for calculating the attributable fraction, 46,48 the inclusion or not of certain tobacco-related diseases in adults or children, 2,14,22,40 the age range considered, 50 the inclusion of death by burning, 8,12 passive smoking 8,12,29 and the application of the current prevalence to calculate the SAM.All these factors infl uence the results of the attributable mortality in the various studies.
In addition, factors such as changes in mortality rates, reductions in smoking prevalence, differences in the methods used to calculate the FAF, omission of the consumption of other tobacco related products (cigars, pipes) in calculating the SAM may also have contributed towards the differences in the studies 8,11,12,61 and represent important limitations in the use of the SAM-MEC software.
The SAM is the result of a previous exposure to tobacco (around ten years between exposure and the development of the disease), a fact that must be taken into consideration in the studies.In a discussion on the fi ndings of Illing & Kaiserman 29 (2004), Tanuseputro et al 53 (2004) found that when adjustments are made to take into account the latency period (either two or three decades) between exposure to smoking and measurement of the associated effect (mortality), there is an increase in the estimation of the SAM by between 8% and 22%, depending on the adjustment method that is applied.Just one study calculated the SAM using a ten year latency period. 63ile cigarette smoking is associated with a series of illnesses, its infl uence does not appear to be the same for each disease.This fact justifi es the use of different relative risks of death for different illnesses.The RR must be estimated for each population in the study, taking into account different biological, cultural and socioeconomic variables.Obtaining these RR for each country is likely to be costly, since it would require specifi c studies.For this reason, the majority of studies used the RR of death of the CPS II. 59me authors discussed the criticisms that are held against using the RR from the CPS II to calculate the SAM, since it is likely to overestimate the burden of death.The most relevant criticism points to the fact that the CPS II is a national mortality survey based on a sample of approximately 1.2 million adult Americans aged 30 and over, who present different characteristics to those of the general population of the USA.The great majority of participants in the study were married, white and with high levels of schooling and income.In short, it is held that the sample is not representative of the overall population and as a result, this compromises the possibility of generalizing the results for the whole American population. 15,35e second critique that was commonly made was that the national estimates were adjusted for age, but not for other potentially confounding factors such as alcohol use, level of education, hypertension, and the prevalence of diabetes mellitus. 39,52In response to these criticisms, Thun et al 57,58 (2000) adjusted the RR obtained from the CPS II for potential confounders such as age, race, education, marital status, occupation, total daily consumption of citrus fruits and vegetables, and alcohol.The results show that adjustments for demographic and behavioral factors did not signifi cantly alter the estimates for the SAM, with a difference of no greater than 1.0%.Malarcher et al 39 (2000) and Wen et al 63 (2005) also showed that changes in the results after adjusting for confounding variables were minimal.
In order to reduce the excess risk attributed to smoking in the RR of the CPS II, Ezzati & Lopez 18 (2003) used a constant corrective factor (30.0%) to avoid overestimating mortality as a result of the repetition of risk estimates, although these were adjusted only for age and sex.The authors used as the basis of their work a method proposed by Peto et al 43 (1992), who used mortality attributed to lung cancer as an indirect marker for the accumulated risk of smoking.This method incorporates the RR of death for tobacco related diseases from the CPS II -also only adjusted for sex and age -corrected by an excess risk of 50.0%.For Sterling et al 52 (1993) and Bronnum-Hansen & Juel 4 (2000), the advantage of this method is that it does not include in the calculation the prevalence estimate of the at risk population.
These methodological variations may, in part, account for the differences found in the over-and under-estimation of the general SAM in the studies considered here and the different estimates for the principle tobaccorelated diseases.This analysis of different studies has shown the powerful impact that tobacco consumption has on the mortality of populations in different countries.It is essential that public policies take into consideration the infl uence that smoking has on mortality and incapacity levels of any population, including the Brazilian population.It is further hoped that by making available factual information and quantitative data, this can also have an impact on the policies and programs that aim to reduce tobacco related deaths.This systematization of articles highlighted the importance of tobacco as a risk factor and its impact on diseases that most affect populations.

Table 1 .
Methods used in calculating the SAM by location and date of study.
2,8,12,13,14,26.Tobacco attributable mortality (%) and confi dence intervals (CI) for cerebrovascular disease and ischemic hearth disease (IHD), by sex and age range in the different studies.Studies 1 and 44 do not provide absolute numbers for calculating the CI.Certain studies provide a very accurate number for the CI, making it impossible to be included visually in the fi gures below.2,8,12,13,14,26Referencesfollowed by letters indicate different years of a study.SAMMEC: Smoking-Attributable Mortality, Morbidity, and Economic Costs Software CPS II: Cancer Prevention Study PAF: Population attributable fraction N: no; Y: yes; N/A: information not given; A: information absent about number of tobacco-related illnesses among adults and children * Method used to calculate the PAF (%).See the Results section.

Table 2 .
Principle fi ndings and general comments in selected studies.SAM given as a rate.Used the mortality rate of the population of Indiana in 1990 and adjusted as a percentage.Deaths caused by cigarette related fi res were estimated as 50% of total deaths by fi re.Not possible to calculate the IC.Used the mortality rate of the US population in 1985 and adjusted as a percentage.Not possible to calculate the IC.RR estimated on the basis of 4 perspective studies.-20.4 8.6 SAM given as a rate.Used the mortality rate of the US population in 1984 and adjusted as a percentage.Age range: > 20 e < 1 years.Not possible to calculate the IC.RR estimated on the basis of 4 prospective studies.---SAM% only for tobacco related illnesses.Data presented in graphic form.Age range: 35 -64 years.Tobacco prevalence estimated through telephone based research, resulting in the under-representation of certain subgroups such as youth, men and the poor.SAM under-estimated.RR estimates agregated, rather than based on just one study -under-estimation of SAM. Adult illnesses also obtained from 2 other studies.Estimates for RR grouped according to the Surgeon General's Report -1989.Projections for SAM based on different scenarios: reductions in the prevalance of smoking by 0%, 2%, 4% and 10%.Did not include important smoking related illnesses: cancer of the larynx, pancreas, bladder and kidney.Deaths by arterial coronary disease and strokes in people aged 65 and above were not included.Short duration of cohort study in Taiwan in 1982 -RR with low level of signifi cance for certain smoking Age range: 1-70 years.CPS II had not yet happened