WPCfJ 2kB RT Z#|x"m^+9FVV999Va+9+0VVVVVVVVVV00aaaL|rr|i_||9C|i||_|r_i||||i909PV9LVLVL9VV00V0VVVV9C0VV|VVLR"R]9+999998999999S9V0xJxJxJxJxJrrJiJiJiJiJ8.8.8.8.{SxSxSxSxS{S{S{S{SxSxJ{V|VxSxS{S`S|L|L|LrLrLrLrL|ViLiLiLiL|V|V|V|V|V|V|V|V9.909090z]C|Vi0i0i0i.i0|V|c|V|V|V|V|r9r9r9_CZA_C_Ci0i0i0|V|V|V|V|V|V||ViLlJiL|Vi0|Vr9_Ci0|V|V|V|V|VNDgFdddgF9CgggEEggzzddd~rrEgdF"m^>R{{RRR{>R>E{{{{{{{{{{RR{沥`{RER{R{nnR{EREΉ{n`R{{{na6aR>RR)RRRPRRRRRRxRExxxxxkkkkk]C]C]C]Cxxxxxx{xxx{{{nnnnnnnn{{{{{{`C`E`E`Eډ{EEE]E{{nnn`]``RRR{nknEn`R{{{NV>{R{{{{{{JQ{Txxx{TEP{{{RR{{xxxȉR{xT"m^3ETggEEEgt3E39gggggggggg99ttt[~rEP~rr~•~E9E`gE[g[g[Egg99g9ggggEP9gggg[c)coE3EE"EEECEEEEEEdEg9YYYYYY~Y~Y~Y~YC8C8C8C8ddddddddddYggdddsd[[[[[[[g~[~[~[~[ggggggggE8E9E9E9oPg~9~9~9~8~9gvggggEEErPkNrPrP~9~9~9gggggg•g~[Y~[g~9gErP~9gggggNH3]gE[ggggg9@gFdddgF%C[[gEEggzzdddwrrE[dF2&2   i"m^>Re{{RRR{>R>E{{{{{{{{{{EEn㲥R`۲鲲RERt{Rn{n{nR{{EE{E{{{{R`E{{{{nv1vR>RR)RRRPRRRRRRxR{EkkkkkۥkkkkkPCPCPCPCxxxxxxxxxxk{{xxxxnnnnnnn{nnnn{{{{{{{{RCRERERE`{EEECE{{{{{۲RRR`]``EEE{{{{{{鲲{nkn{E{R`E{{{{{NV>p{Rn{{{{{DM{Txxx{T,Pnn{RR{{xxxȉRnxT"m^3EVggEEEg3E39ggggggggggEEg~~~~E[r~~gr~~rrP9PWgEgg[g[9gg99[9ggggPP9g[[[PR9RoE3EE"EEECEEEEEEdEg9zdzdzdzdzdYzYzYzYzYC8C8C8C8dddddddddoYzdggdoYdzd~g~g~g[[[[g~[~[~[~[ggggggggE8E9E9E9d[[r9r9r9o8r9gzgggg‰~P~P~PgPdNgPgPr9r9r9ggggggr[rPoNrPgr9g~PgPr9r[r[gggNH3lgPgggggg9@gFdddgF,CrrgEEggzzdddkggErdF"m^3PrggEEEgu3E39ggggggggggEEuuugPg~~r~~~E9EugEgg[g[Egr99g9rgggPP9r[g[PH-HuE3EE"EEECEEEEEEdEg9ddddd•YYYYYN8N8N8N8oddddoooozYdggdzYdzdggg[[[[g[[[[ggggggrrP8P9P9P9kgg~9~9~9zF~9rrrgg•PPPrPoNrPrP~9~9~9rrrrrr~[~PzN~Pg~9rPrP~9~[~[ggrNH3ggPgggggg7>gCdddgC9CgggEEggzzdddsrrEgdC"m^)2>>g`)))>F)">>>>>>>>>>""FFF7rYRRYKEYY)0YKnYYEYREKYYuYYK)"):>)7>7>7)>>"">"`>>>>)0">>Y>>7;;C))){)))())))))<)>"W5W5W5W5W5nRR5L5L5L5L5(!(!(!(!YY>WK7K7K7K7Y>Y>Y>Y>Y>Y>Y>Y>)!)")")"XC0Y>K"K"K"L!K"Y>YGY>Y>Y>Y>nYR)R)R)E0@/E0E0K"K"K"Y>Y>Y>Y>Y>Y>uYY>K7N5K7Y>K"Y>R)E0K"Y>Y>Y>Y>Y>N+WW8>)7>>>>i>"&kk>*<<<>k*(77>{))xx>>II[x +++++7))&&&&,++++,,,,++,-++,"---))))-&&&&--------,!-&&&&&--$----7-)))" ""&&&------:--&'&-&-)"&-----N++5555 ><<%%.<<+$22"">9>2. ~;z"m^30****9&&)-&&--)"3)-&-&"-&3&""-)!7 %%%%%7)(%%%%(++++++++!%+-+!+%&&&))))-&&&&--------1)"""!"))%))--:)&&&"""------3)""!"-")&"""---N++ 4555 ""7<<%%.<<+ ..7"5>Times New RomanCourierCourier BoldTimes New Roman BoldTimes New Roman ItalicTimes New Roman Bold ItalicSymbolCG TimesCG Times Bold"^09MVS899V^+^+0VVVVVVVVVV00^^^Lxir{icx{8Aui{x`xoZi{xxxl889SS8JSJSJ8SS..S.SSSS>A.SSxSSJR"RNOxxSSS8JDDSSVSSS;SS;88VVS++SSfSSxSc]]8VS;"xxSxWxxqS^^^z0^S88^^^zxzzzggxxxxx8SxE]S^/zU*FxJS`vggxxxxxzzzzKxxxxggqUxx^ofx^^zxxxxzz|lr^gFiUiKiU|gY98|^v^c|YoU|^^_^egKiKvcY|^vvlKg88^ccvJlfjgz88SSSSSSSlllllllllllllllllllKKKKKKKgggggggggggggggggggg888888888888^^^^^^^ccccccccccccvvvvvvvvvvvvvvvvvvvv8`|i|9|v9v^J8"^9E]gdCEEgq3q39gggggggggg99qqq[~~wCN~sk~CCEddCYdYdYCdd88d8ddddJN8ddddYc)cN_dddCYQQddgdddFddFCChhd44ddzdddwooChdF"dhȇdqqq9"qdDDqqq||֐CdRodq8ȓf3UYds||֐ZȞ||ȇfȐqȲȧzȐqqѕq|U~f~Z~f|kEDqqwkeqqrqz|Z~ZwkqZ|DDqwwYz|CCdddddddZZZZZZZ||||||||||||||||||||DDDDDDDDDDDDqqqqqqqwwwwwwwwwwwwCs~EEqYC26` %+1"m+O6^,2EKKu222K2222KKKKKKKKKK22Cl^go^Ylo2:j^olVldQ^ollla222KK2CKCKC2KK**K*uKKKK8:*KKlKKCCKCK'KKKKK2KKKKKKKKK*lClClClClCdgC^C^C^C^C2*2*2*2*oKlKlKlKlKoKoKoKoKlKlCoKlKlKlKoKVKlClClCgCgCgCgCoK^C^C^C^ClKlKlKlKlKlKoKoK2*2*2*2*nS:jK^*^*^*^*^*oKoYoKoKlKlKld8d8d8Q:Q:Q:Q:^*^*^*oKoKoKoKoKoKllKaCaCaCoK^*oKd8Q:^*lKlKoKlKoK2KK222WxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxNlllKKK2C==KKKKKK5KKKK522NNK''KK\\rKKlKY}}SS2NK5"llKlNllK充+K22l\llllllllll2KlKSKldK2KlCKVluallllllllllEllllll}d\ul@YlllllllKlllllllCllllClllllllllllllllllllllllllllllllllllllllllll2lll2lll2lll2llllllllllllll\^S\SdC^:aCoKlC2*jCdSSoCdKlCl\VK\KK^ClCd\lSu\\\\:K**CCC\C\\\\n22KKKKKKK\\\\\\\\\\\\\\\\\\\:::::::KKKKKKKKKKKKKKKKKKKK************CCCCCCCCCCCCCCCCCCC\\\\\\\\\\\\\\\\\\\\2Vl^o2ll\2lKKC2xɇEإiiTZWxffQQ~~~c{`xllifurllBlTNN-BuurrrrxxxrrrrE6]r"m+O6^,2EKK}222K2222KKKKKKKKKK22Kldlld\uu:Hudlu\ulSdlllld222KK2KSCSC5KS*2S*}SKSSC:2SKlKKCKKKKKKKKK2KKKKKKKKS*lKlKlKlKlKllCdCdCdCdC:*:*:*:*lSuKuKuKuKlSlSlSlSlKlKlSuKuKlKlK\SlKlKlKlClClClClSdCdCdCdCuKuKuKuKuKuKuSuS:*:*:*:*VHuSd*d*d*d:d*lSlwlSlSuKuKolClClCS:S:S:S:d2d2d2lSlSlSlSlSlSllKdCdCdClSd*lSlCS:d2lKlKlSuKlS2KK222WxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxNlllKKK2K@@KKKKKK5KKKK522VVK''KK\\rKKlK^}}SS2VK5"llKlVllK充+K22l\llllllllll2KlKSKldK2KlCKVlugllllllllllHlllllld\ul@YlllllllKlllllllKllllKlllllllllllllllllllllllllllllllllllllllllll2lll2lll2lll2llllllllllllll\dS\SdCd:dCuClC:*uCdSSlCdKuCl\\K\KKdClCd\lSu\\\\:C**CCC\C\\\\n22KKKKKKK\\\\\\\\\\\\\\\\\\\:::::::CCCCCCCCCCCCCCCCCCCC************CCCCCCCCCCCCCCCCCCC\\\\\\\\\\\\\\\\\\\\2\ldu:ul\:lKKC2xɇEإiiTZWxffQQ~~~c{`xllifurllBlTNN-BuurrrrxxxrrrrE6]r"m+O6^*--PF-P----------ZPZ(PA9>C95AC#?9RCA4A<09CAUAA:--(-(-(---F----!#--A--((-(Z---]------------A(A(A(A(A(P<>(9(9(9(9(C-A-A-A-A-C-C-C-C-A-A(C-A-A-A-C-4-A(A(A(>(>(>(>(C-9(9(9(9(A-A-A-A-A-A-C-C-B2#?-99999C-C5C-C-A-A-PAZZAAAAAAAAAAZZZZZ+ZAAAAZZZZZZZZZZZZZZZZZZZAAZZZZZZN<7FZAZZZ&5AAAAZZZZZZAAA-AAZAAZZZAAA-AAAA-AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA7<272<(<#<(F(A(#F(<2U2A(<-F(A77-7--<(A(<7A2F7777#((((7(7777B-------7777777777777777777#######(((((((((((((((((((((((((((((((((((((((777777777777777777777AA.SSxSSJJSJS+SSSSS8SSSSSSSSS.xJxJxJxJxJorJiJiJiJiJ8.8.8.8.{SxSxSxSxS{S{S{S{SxSxJ{SxSxSxS{S`SxJxJxJrJrJrJrJ{SiJiJiJiJxSxSxSxSxSxS{S{S8.8.8.8.z]AuSi.i.i.i.i.{S{c{S{SxSxSxo>o>o>ZAZAZAZAi.i.i.{S{S{S{S{S{SxxSlJlJlJ{Si.{So>ZAi.xSxS{SxS{S8SS888WxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxNxxxSSS8JDDSSSSSS;SSSS;88VVS++SSffSSxSc]]8VS;"xxSxWxxS唔0S88xfxxxxxxxxxx8SxS]SxoS8SxJS`xlxxxxxxxxxxMxxxxxxofxGcxxxxxxxSxxxxxxxJxxxxJxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx8xxx8xxx8xxx8xxxxxxxxxxxxxxfi]f]oJiAlJ{SxJ8.uJo]]{JoSxJxf`SfSSiJxJofx]ffffAS..JJJfJffffz88SSSSSSSfffffffffffffffffffAAAAAAASSSSSSSSSSSSSSSSSSSS............JJJJJJJJJJJJJJJJJJJffffffffffffffffffff8`xi{8xxf8xSSJ8xߖMuu]daqqZZnkxxuqxxIx]WW2IςMy.X8+,wX\  P6P ?xxx,2x6X@`7X@D?xxx,)x `7X ?7nC3,%4Xn4  pX @AP>,%U`4  p A7iC3,ƒXi\  P6XP  A~P>,B-~\  P6PZ B5hC3,:Xh*f9 xr XX C6jC3,Q3Xj9 xOX =W!?(,,h?\  P6hPZ #{,W8+,:.UW*f9 xr X H5!,x,5\  P6,P Z $I4!,:F,4*f9 xr ,X !+ ,Ӵ \  P6PZ ",,:/*f9 xr X y.V80,m!V\  PAP:l*Y2,,t!1Y PE37P9m)\2,,}\_ pi7@6,t/6 PE37P;A7,}7_ pi77x/c81,tApc PE37P6y.f81,}f_ pi7 7gC9,mXg\  PAXPQsXRa=X?F~u O%Lu%Lu AF^1>Pu?rX?. u*>PWF\?s'ArORtQt>P>X>ORE b  6&6&StandardHPLAS4.PRSX\ 6&6&StandardHPLAS4.PRSX\  #Xn4  p4X#@IX - @       CHAPTER IX #Xi\  P6ƒXP#  b ASSESSING THE HEALTH EFFECTS  X #Xi\  P6ƒXP#INTRODUCTION 55 A fundamental principle underlying public health action has been the need to determine the causes of illhealth (or more specifically, the causes of epidemics), and to reliably document and monitor the evolution of disease due to various exposures. This central role of epidemiology in promoting public health is as important for chronic diseases as it is for infectious diseases. Indeed, chronic diseases and their causes need to be monitored with the same degree of vigilance as the infectious disease epidemics. 55 The use of tobacco products has brought about a pandemic which has already killed millions of individuals in developed countries, and is threatening to kill many millions more in developing countries. In 1964, the U.S. Surgeon General's first official report on smoking and health concluded that "cigarette smoking is a health hazard of sufficient importance in the United States to warrant appropriate remedial action". A few years later, in 1971, the Royal College of Physicians of the United Kingdom declared that smoking was "as important a cause of death as were the great epidemic diseases that affected previous generations in this country". Subsequent reports have documented the progression of the epidemic. 55 The epidemic of tobaccorelated deaths that had begun in a handful of developed countries, has spread to all developed countries. The general pattern is for deaths to first rise for men, and subsequently for women, reflecting sex differences in smoking behaviour. By the end of the 20th century, cigarette smoking will have killed about 62 million people in developed countries (52 million men, 10 million women) )-++Ԍ In the early 1990s, one in four male deaths, and, more importantly, one in three male deaths in middle age (3569 years) were caused by smoking. 55 Smokingattributable death rates are currently lower among women than men, but are rising rapidly. For example, the proportion of female deaths in middle age due to smoking has increased sixfold since 1955, rising from 2% to 13% by 1995. WHO estimates that about half of all tobaccorelated deaths occur at ages 3569 years, making tobacco much the most important cause of premature death in developed countries. 55 This chapter presents some basic epidemiological concepts useful for understanding the relationship between tobacco and disease. Data sources for mortality and to a lesser extent, morbidity, are discussed and methods are presented for estimating or projecting tobaccoattributable mortality. Several procedures are suggested to assist countries in determining what methods are most appropriate for them.  X-  COLLECTING AND INTERPRETING EPIDEMIOLOGICAL DATA ON THE  X TOBACCO EPIDEMIC 55 Most countries have some system for collecting and analysing health information. The coverage and reliability of these data vary widely, depending on the country's level of health development. Developed countries have national vital registration data on causes of death which are both complete and reliable. All deaths in these countries are registered and all, or virtually all deaths, are medically certified as to the underlying cause. On the other hand, morbidity data are much less complete, even in these statisticallyadvanced countries. For example, only a handful of developed countries have national cancer incidence registries (although many more have registries that cover part of the population). 55 In developing countries, the availability of data on mortality and morbidity is much less uniform, and it is important to realize that there are great differences amongC*-++ developing countries in their capacity to monitor tobaccorelated diseases. In some countries of Latin America and East Asia, for example, reliable national mortality statistics are available, at least for adults, which can be used to assess the local importance of these diseases. In many other parts of the developing world, data that are available are often unrepresentative, unreliable and outofdate. The challenge is thus to decide how best to use whatever data are available so as to maximize their utility in support of local action to control tobacco use. Where reliable data to monitor the health hazards of tobacco use are not available, efforts should be made to collect them using one or more of the approaches described in this chapter.  X  Basic Data and Definitions 55 Information about the size, structure and geographical distribution of a population is essential if data on health status (i.e. mortality and morbidity) are to be correctly interpreted. When assessing and monitoring the extent of a population's tobacco  Xs epidemic, it is essential to have reliable data on population size by  age and sex  so that rates of disease can be calculated from data on cases of disease or deaths in the same population. In many instances, the epidemiologist will have information on the number of cases or deaths from lung cancer or other smokingrelated diseases, but without information concerning the size and composition of the specific population at risk of incurring these diseases, the value of the data for epidemiological surveillance is considerably reduced. 55 A key requirement, therefore, is that all data on the number of cases of disease or of deaths be related to an identifiable population at risk so that rates of disease occurrence or death can be calculated. Disease incidence is strongly age and sexdependent, with rates for the major tobaccorelated diseases being higher for males and increasing sharply with age. Therefore, it is strongly recommended that population data be sought and analyzed separately for males and females and on the basis of age. For most purposes, the 10year age groups 1524 years, 2534, 3544, 4554, 5564, 6574, 75 and over are desirable. However, broader agegroups such as 1544, 4564 or 69, and 65 (or 70) and over are still useful analytical categories if these are the only data available.E*-++Ԍ X ԙ55 The crude mortality rate (M) in a specific population is defined as the number of  X] deaths (D) in that population in a given period of time (usually one year) divided by the  X population (P) in which the deaths have occurred. Usually the population at the middle of the year is used. Rates are usually expressed as the number of deaths per 1000 (or some multiple of 1000) per year.  J ddx!ddx J      X     D  X i.e.    M =  x 1000  X     P   55 Mortality rates always need to be calculated for specific agegroups, and separately for men and women. Examples are given in the table on the following page.  X 55 A crude incidence rate for a disease is similarly defined and calculated, but instead  X$ of deaths, we count the number of new cases of disease which are reported for a population in a year. Age and sexspecific incidence rates are also defined in a similar way as for mortality. 55 Agespecific disease or death rates are not always easy to interpret, particularly as the number of age categories increases. Summary indices of disease rates over all ages, or over broad agegroups, greatly assist the interpretation of overall disease levels. Moreover, they provide more reliable information on disease patterns and trends. For example, the crude mortality rate from lung cancer might be increasing in a population simply because the population is ageing, without any underlying increase in lung cancer risk. 55 Two simple indices are widely used in epidemiology to summarize agespecific rates: i)55 cumulative incidence (or mortality) rate (CIR or CMR), defined as the sum of the agespecific rates between two ages. If, as is generally the case, agespecific rates)-++ are available by fiveyear age groups (e.g. 1519, 2024, etc), then the cumulative incidence (or mortality) rate is defined as 5 x (the sum of the fiveyear agespecific death rates). That is,%"5 @!ddx Al ddx @    |`'x = b  X  CMR = 5  Mx `'x = a  55 NN  X# 55 where a is the starting age of the ageinterval, b is the upper limit of the age  X 5%"%" %"%"55 interval, and Mx is the agespecific death rate for each fiveyear interval  X between ages a and b .%"5  X  Example : Suppose the following data on lung cancer deaths were available from vital registration or some similar source.  Agegroup& & Number of deaths``)Population at9Agespecific 55 NN& &  !``)risk (midyear9lung cancer 55 NN& &  !``)population)%9death rate%" 55 NN& &  !``)19(per 1000 population)!!H 55 NN& &  !``)19ss@$"$"H$"$"H$"$"H$"$"H 3539NN& &   140!``)1 650 000190.08  Xl 55 NN& &  !``)19(= 140/1 650 000 x 1000) 4044NN& &   480!``)1 850 000190.26 4549NN& &  1070 !``)1 840 000190.58 5054NN& &  2090!``)1 670 000191.25 5559NN& &  3000!``)1 340 000192.24 6064NN& &  3200!``)1 100 000192.91 6569NN& &  2900!``) 900 000193.22 70+55 NN& &  3600!``)1 400 000 92.57 ss@$"$"H$"$"H$"$"H$"$"H$"$"H$"$"H$"$"H$"$"H$"$"H$"$"H$"$"H$"$"H   X) Then the cumulative mortality rate from lung cancer between ages 35 and 69 would be: 5 (0.08 + 0.26 + 0.58 + ... + 3.22) = 52.7 per 1000.n*-++Ԍii)55 agestandardized incidence (or mortality) rate (ASR) is calculated to facilitate comparisons of disease rates among populations which have different agestructures (overall rates of disease are very much influenced by population agestructure since chronic disease rates rise steeply with age). It is essential to have agespecific rates for this purpose. The ASR is calculated by multiplying agespecific rates by certain arbitrary weights and summing them over all age categories. The weights are in fact the number of people in each age group in a hypothetical "standard" population.%"5 55 Thus, the ASR is the rate which would be expected in a hypothetical population with a "standard" age structure. In practice, any population agestructure could be chosen as the `standard', although it is preferable to choose a standard which is not too dissimilar to that of the actual populations under study. Two widely used population age standards are the "World" and "European" standards. The agecomposition of these hypothetical populations is given in Appendix IX.1. Changing the agestandard will affect the level of the agestandardized rates being compared, but their comparative rank order should be largely preserved.  X- 55  Example :& & To calculate the agestandardized death rate from lung cancer at ages 3569 from the data given in the previous example, we proceed as follows:%"& 55 Step 1:& & Calculate agespecific death rates (already done in the example).%"& 55 Step 2:& & Multiply (or weight) each of these death rates by the corresponding population in the `standard'. This gives the expected number of deaths, by age, in the standard population.%"& 55 Step 3:& & Sum up these expected number of deaths over the age groups of interest, and divide by the total population size in the `standard' population to obtain the ASR (this last step is facilitated by choosing the total standard population size to be some multiple of 1000 since death rates are usually expressed per 1000).%"& E*-++ԌThis procedure yields an agestandardized death rate at ages 3569 of 1.30 per 1000 population, calculated as follows (choosing, arbitrarily, the `European' standard):  55 Agegroup& &  Agespecific``)`Standard'19Expected number of 55 (years)& &  lung cancer``)population9lung cancer deaths 55 NN& &  death rate!``)size19in the 'standard' 55 NN& &  (per 1000)``)19population$"$"H$"$"H$"$"H$"$"H$"$"H$"$"H$"$"H$"$"H$"$"H$"$"H  55 3539& &  0.08!``) 700019 ss@0.56 55 4044& &  0.26!``) 700019 ss@1.82 55 4549& &  0.58!``) 700019 ss@4.06 55 5054& &  1.25!``) 700019 ss@8.75 55 5559& &  2.24!``) 600019ss@13.44 55 6064& &  2.91!``) 500019ss@14.55 55 6569& &  3.22!``) 400019ss@12.88  55 3569& &  !``)43 0001 9ss@56.06   X Then the ASR at ages 3569 = 56.06/43,000 x 1000 = 1.30 per 1000 population.   XF  MONITORING TOBACCORELATED MORBIDITY  X 55 The prevalence of smoking has already been defined in Chapter VIII. Prevalence  X]  of a disease is defined in a similar way as the proportion of the population who have the disease at a given point in time. @Al ddx 2addxt# @     Number of existing cases of the disease (in a year) Disease Prevalence (DP) = 20 x 1000 Midyear population size  t# 55 The two words "incidence" and "prevalence" are sometimes used as synonymous.  X* This is incorrect. Incidence is a measure of the  risk  of disease, whereas prevalence is a*-++  X measure of the existing number of cases  of disease in the population at a point in time. They are closely interrelated, since prevalence depends upon current and past incidence, the duration of the disease, and mortality. 55 Morbidity indicators of relevance for assessing the tobacco epidemic are, in effect, measures of incidence or prevalence. Although incidence is unquestionably the most appropriate measure of the evolution of tobaccorelated diseases, it is intrinsically more difficult to obtain than mortality. Few reliable, comprehensive data collection systems for monitoring the incidence of major chronic diseases exist. The most notable exception is cancer, for which several countries have established some system of data collection which is representative of at least part of the population. However, even in developed countries, incidence data are rarely collected for the entire population, but rather for defined geographical areas. The International Agency for Research on Cancer (IARC) in Lyon routinely monitors cancer incidence and periodically publishes data from cancer registries around the world in their publication Cancer Incidence in Five Continents. IARC has also published guidelines for the establishment and maintenance of cancer registers. 55 Other sources of morbidity data include hospital records or data from other health establishments, including general practices (GPs). Although the diagnostic information on these records can give a useful (though general) idea of the occurrence of tobaccorelated diseases in the population served by these facilities, the data tend to be biased in that they reflect disease occurrence among those who have access to health services. They do not reflect patterns of disease in poorer populations who may not have access to these services, but who nonetheless incur tobaccorelated illnesses. Though these records may yield indirect evidence about tobaccorelated disease trends in a population, they do  Xr#  not yield reliable measures of disease incidence and are frequently based on small numbers. Thus, the observation that a particular hospital recorded 62 new cases of lung cancer in a year is not, by itself, useful for disease monitoring. Even a comparison with records from earlier years, which may show far fewer cases of lung cancer, is at best suggestive that lung cancer is rising, but is not a reliable monitoring tool for assessing trends in the disease since it may reflect changes in population size, better diagnosisE*-++ and/or greater access to health care facilities. 55 Reliable data on chronic disease incidence for diseases other than cancer are even less accessible. In many developed countries, good communitybased registration  Xt systems for cardiovascular diseases exist as a result of the WHO MONICA (Monitoring of Trends and Determinants of Cardiovascular Diseases) initiative which was established in the mid1980s to monitor trends in cardiovascular disease incidence and in related riskfactor exposures, including smoking. About 40 centres are collaborating in this project. Due to the strict criteria for diagnosis and the standardization of data collection and registration procedures, it is a very valuable source of data on cardiovascular morbidity for these populations. However, establishing and maintaining cardiovascular disease registries is very costly, and may not be appropriate or feasible for national surveillance, but rather more for research and development purposes. 55 Several countries are working to develop hospital discharge reporting systems to obtain relevant information on incidence trends for major chronic diseases (e.g. acute myocardial infarction). This requires that a degree of standardization of diagnostic criteria has been achieved across hospital record systems. These data should be linked with mortality registration to provide the overall incidence (fatal and nonfatal) of cases of disease. Provided that all hospitals treating patients in a given geographical area participate, and the population of that area is known, this procedure will yield useful estimates of disease incidence.  55 Another source of data on morbidity from chronic diseases are periodic health surveys. Although these methods may provide adequate information about levels of illness, they are inadequate for assessing chronic disease patterns. Many surveys ask broad questions about symptoms from which specific medical diagnoses must be drawn. However, since many chronic diseases have similar symptoms, diagnosis is often difficult. This is compounded by the fact that symptoms are selfreported by respondents, who may lack the ability to identify specific symptoms, or who may often underreport disease or symptoms. Thus, health surveys are generally of limited value inC* -++ assessing specific tobaccorelated disease incidence and prevalence. Some surveys ask whether the respondent has been told by a physician that he or she has a specific disease. Data obtained in this way can provide a broad indication of disease levels in representative samples of the population. 55 When reliable and comprehensive morbidity data are available, they have the potential for providing a more complete assessment of the health effects of tobacco use than mortality data alone. This is particularly relevant for those smokingrelated diseases which are rarely fatal (i.e. peripheral vascular disease), or for diseases in which a long period of illness precedes death (i.e. chronic bronchitis or emphysema). A decision must be made whether the additional morbidity information for a representative group of the population warrants the resources required to reliably obtain it.  X  MONITORING TOBACCORELATED MORTALITY  Xs   55 A major advantage of using mortality data to assess the evolution of tobaccorelated diseases is that they are generally widely available and unambiguous: death is a singular event for which most societies have instituted some form of legal registration. This is not the case for morbidity which can be defined in several ways. 55 The most common source of mortality data is vital registration systems which are operative in most countries. In this system, a civil registration office is notified within a short time interval following each death, and a certificate which contains basic demographic details of the deceased (e.g. age, sex, place of residence) is issued. The certificate also states the underlying cause of death, chosen according to the rules and procedures of the "International Statistical Classification of Diseases, Injuries and Causes  Xr# of Death" (ICD). For all developed countries, and for some developing countries, all, or virtually all deaths are registered and medically certified as to the cause of death. For these countries, reliable data on the age and sex structure of the population are also generally available for the calculation of mortality rates. For these 5060 countries or territories, representing about 30% of the world's population, mortality from chronic diseases can be monitored with reasonable confidence, although even for theseC* -++ populations, the validity of diagnosis could be further improved, for example, by greater use of autopsy. (See the latest WHO World Health Statistics Annual for a compilation of international mortality data.) 55 In some countries, sentinel surveillance points have been established which serve as vital registration areas for specific population groups. In this case, all vital events, including deaths, are registered for the population under surveillance (usually a defined administrative area such as a county, province, or city). All deaths should be medically certified, and estimates of the population size and age and sex structure for the `catchment' area should be available. If such `sentinel sites' are reasonably randomly distributed throughout a country or area, they can, collectively, provide a very good basis for monitoring disease incidence, particularly mortality. For example, the Chinese Disease Surveillance Point (DSP) system is a statisticallyrepresentative collection of about 135 sentinel sites, covering over 50,000 deaths (in 1993). 55 Other procedures in use for the collection of causespecific mortality data on chronic diseases include `lay reporting' of causes of death, retrospective surveys, or the analysis of hospital records. Some of these approaches suffer from the same diagnostic problems or biases mentioned earlier for morbidity data. `Lay reporting' is a practical, generally inexpensive means to obtain data at the community level. Using this approach, a nonmedical team visits the household of the deceased to enquire about symptoms and related information on the deceased prior to death. From this information, a diagnosis is made about the cause of death, using standardized algorithms. This diagnosis should be validated on a sample basis by a physician visiting the household. Such a system is operative in India, based on the populations served by about 1,400 primary health care centres throughout the country. Evaluations and use of these data suggest that while they are helpful in delineating major disease or injury levels, they are not sufficiently accurate for monitoring specific chronic diseases related to tobacco use. 55 In summary, only vital registration or sentinel surveillance sites of known reliability and completeness are likely to be useful for monitoring mortality from the major chronic diseases caused by tobacco. Rather than attempt to monitor mortality rates for allC* -++ diseases associated with tobacco use, it is recommended that attention be restricted to those diseases most strongly caused by tobacco and/or which are major causes of adult mortality. These are (ICD9/ICD10 numbers in parentheses): 55 lung cancer (162/C33, C34) 55 cancer of the mouth, oesophagus, larynx and pharynx 55 (could be grouped) (140150,161/C00C15, C32) 55 ischaemic heart disease (410414/I20I25) 55 cerebrovascular disease (430438/I60I69) 55 chronic obstructive pulmonary disease (COPD) (4902, 4946/J40J44, J47) 55 (this is primarily chronic bronchitis and emphysema) 55 55 In addition, it is useful to monitor death rates from all cancers (140209/C00C97) and all causes of death combined.  Xs  55 Focusing on these conditions will greatly reduce the analytical problems which would otherwise occur from analysing vast data sets on 3050 causes; very little additional information would be gained from examining more detailed causes at considerably greater effort. 55 The degree of agedetail for the collection and analysis of mortality statistics should also be considered for epidemiological surveillance. Death from the major chronic diseases related to tobacco use is relatively rare prior to age 35 or so. It is recommended that mortality statistics be compiled, wherever possible and feasible, according to the 5year age groups 3539, 4044, ..., 8084, 85 and over. For presentation of data, including graphs of trends in death rates for specific diseases, ten year age intervals 3544, 4554, ..., 6574, 75 and over may be more appropriate. 55 Trend analyses should not only focus on timetrends in disease, e.g. changes over a decade, or over a specific period such as 19701995, but equally, analyses of cohort patterns should be undertaken. Figure IX.1, for example, shows how lung cancer mortality for men at ages 5559 years in selected countries has changed for successiveC* -++ birth cohorts, beginning with men born at the turn of the century. This type of presentation illustrates very clearly the so called `cohort effect' whereby the progressive increase (or decrease) in smoking among successive birth cohorts is reflected, decades later, in progressively rising (or falling) lung cancer rates.  X  RISKS AND CAUSALITY IN EPIDEMIOLOGY 55 Public health specialists responsible for assessing the health effects of tobacco must be aware that all diseases caused by tobacco use are multifactorial in nature. Smoking in developed countries causes most lung cancer, but not all. Cigarettes cause some, but not all, heart and other vascular disease. Smoking and smokeless tobacco use cause many, but not all cancers of the mouth, pharynx, larynx and oesophagus. Therefore, knowing levels of diseases caused by tobacco use and their trends is not sufficient to fully understand the progression of the tobacco epidemic. Effective tobacco control measures are much better served if more precise information is available about the burden of various diseases attributable to tobacco use, and how this is changing. 55 Since epidemiology is an empirical science (it would be unethical to conduct intervention studies to actually produce disease), the "proof" that tobacco causes disease is based on the range and weight of evidence that has accumulated about the relationship between tobacco use and the diseases associated with it. That is, proof is established in the legal sense, meaning "beyond reasonable doubt". Over the years, the criteria for deciding causality have evolved from those originally laid down for infectious diseases (Koch's postulates). In chronic disease epidemiology more extensive criteria have been defined to establish causality. The main ones are listed below, with brief details about how these criteria have been applied to demonstrate the causal association between smoking and lung cancer. C* -++Ԍ X ԙ Criteria for causality and their application in the case of smoking and lung cancer  E2addxt# 2ddxF= E  Z t#  X  Criterion Strength of the association between smoking and lung  X cancer Z   F  doseresponse temporal (time) sequence strength of the association effect of cessation concordance with other evidence consistency of findings among studies'  the longer a person smokes, the higher the risk; the more cigarettes smoked per day, the higher the risk smoking precedes the onset of disease by 2 to 3 decades the risk of getting lung cancer is 2025 times higher among smokers (of long duration) compared with never smokers stopping smoking reduces the risk; the benefits increase with longer duration of cessation the carcinogenicity of tobacco has been demonstrated in laboratory studies on animals and chemical analyses of tobacco smoke thousands of independent studies in different parts of the world, using different methods, have confirmed the association    55 The measures of disease occurrence most familiar to public health workers are the death rate and the incidence rate. Both of these measures, and various summary statistics derived from them, such as the agestandardized death rate and the cumulative  X incidence rate, are indicators of absolute risk . Absolute risk indicates the likelihood of an individual incurring a disease or dying from it in a given year. Among different population groups, absolute risk can vary substantially, depending on a variety of factors, particularly exposure to a hazardous substance such as tobacco. A hypothetical example of different absolute risks for two population groups is given in Figure IX.2. 55 Data on absolute risks of death from various diseases, and on the incidence of cancers, are available for many populations. By themselves, these data to not indicate the importance of a particular exposure (e.g. cigarette smoking) since in most countries,*-++ mortality or incidence rates are only available for the entire population, not subgroups such as smokers and nonsmokers. However, in some countries, specific studies of population groups consisting of smokers and nonsmokers have been undertaken, thereby allowing comparison of mortality among smokers and nonsmokers, for each major disease caused by smoking. Dividing the death rate among smokers by that among neversmokers indicates how many times higher the mortality from that disease is among  X. smokers compared with never smokers. This measure is known as the relative risk (see  X Figure IX.2) since it summarizes the absolute risk of death of smokers relative to that of never smokers, and is the basis of the statements often made about smokers having a death rate from a specific disease which can be up to 30 times higher than that of a never smoker. 55 An even more powerful piece of information for defining the need for tobacco  X control strategies is the population attributable risk (PAR) . This measure provides an estimate of the proportion or percentage of mortality (or morbidity) from a given disease in a population which can be attributed to tobacco use. When lung cancer or heart disease rates are high, the case for urgent tobacco control action is further strengthened if it is possible to demonstrate the proportionate impact (frequently substantial) of tobaccouse alone on those rates. Figure IX.2 gives the formula for calculating population  X attributable risk based on the prevalence of smoking (P) and the relative risk (RR) of death or incidence (comparing smokers with never smokers) for a given disease. The derivation of this formula is demonstrated in Appendix IX.2.  X^  QUANTIFYING THE RELATIONSHIP BETWEEN TOBACCO USE AND DISEASE 55 The effect of tobacco use on health was first quantified in the early 20th century when an increased risk of oral cancer in smokeless tobacco users was reported from India. Further research into the health effects of tobacco use, particularly cigarette smoking, was stimulated by the observation that lung cancer, which was extremely rare had risen sharply among men in countries such as the USA and the UK during the 1940s. Early epidemiological studies to investigate the causes of this new epidemicF*-++ examined several hypotheses about the reasons for the increase, including air pollution, tarring or roads, etc. With their now famous "casecontrol" study carried out in the late 1940s, in which patients with and without the disease were asked about their exposure to a variety of factors, Doll and Hill concluded that "smoking is an important factor in the production of carcinoma of the lung". Similar conclusions were reached in a study carried out by Wynder and Graham, and others, around the same time in the USA. 55 In 1951, to confirm their findings, Doll and Hill questioned 60 000 doctors in the United Kingdom about their smoking habits and other exposures which might account for higher death rates in smokers. They followed up the survival of the 40 000 doctors who had replied to their questionnaire, and within a few years, the early results from this cohort study confirmed the higher lung cancer rates among smokers. 55 Doll and Hill's study demonstrated for the first time that myocardial infarction (heart attack), chronic lung disease (especially bronchitis and emphysema) and various other diseases were also related to smoking. Several other prospective studies in America, Europe and Japan further confirmed the causal relationship between smoking and atherosclerotic cardiovascular diseases (coronary heart disease, stroke, peripheral vascular disease). Even though the relative risk from smoking for cardiovascular  X diseases is generally lower than for cancers caused by smoking, the number of cardiovascular deaths due to smoking is larger in most developed countries since cardiovascular diseases are much more common. These studies also demonstrated the multifactorial nature of cardiovascular diseases. 55 So far, about 25 causes of death that are significantly associated with tobacco use have been identified. Some of these are minor conditions which are comparatively uncommon, and hence the excess risk among smokers does not lead to a great many deaths. Others, however, are major public health problems in a number of countries (e.g. ischaemic heart disease), and thus the excess risk of smokers from such diseases contributes significantly to overall levels of mortality. 55 Table IX.1 provides a summary of the major diseases caused by smoking, andC*-++ relative risks for smokers versus neversmokers (based on American data). Relative risks are shown separately for men and women and for two different time periods. The relative risks calculated from the American Cancer Society's first Cancer Prevention Study (CPSI) refer to the period 195965. The relative risks calculated from the second Cancer Prevention Study (CPSII) refer to 198286, about 25 years later. Almost without exception, the relative risk in CPSII increased substantially essentially due to a longer duration of smoking, although some of the increase may have been due to other factors. For example, women in CPSI had been smoking for about 1020 years whereas women of comparable age in CPSII had been smoking for 30 years or more. In the early 1960s, the relative risk of lung cancer among male smokers was 11.4 times that of lifelong male nonsmokers. Twentyfive years later, this rate ratio had doubled to 22.4 for males, and rose from 2.7 to 11.9 for women. 55 This increase in relative risks for smokingrelated diseases with longer duration of smoking has been found in other populations as well. For example, Figure IX.3 shows the steady increase in relative risks from lung cancer among male smokers in Japan. The fact that these RRs for lung cancer are still substantially below that of the USA merely reflects the fact that smoking in Japan has been a much more recent phenomenon than in countries such as the USA and the UK. With longer exposure, relative risks from smoking for lung cancer and other smokingrelated diseases can be expected to rise in Japan in much the same manner as they have done elsewhere. 55 Two types of epidemiological study can be used to estimate relative risks, either  X[ prospective (cohort) studies or casecontrol studies. The principal features of each of these study designs are outlined below; more details about the conduct and analysis of these studies are given in standard epidemiological texts listed in the reading list for this chapter. In addition to the diseases monitored above, these studies have also been used to assess other health risks due to tobacco use including the effects on reproductive health and the effects on nonsmokers.  X( 55 Prospective Studies: The basic concept of these studies is to interview a group of people about their smoking practices, exposure to other risk factors (e.g. alcohol, airF*-++ pollution), personal characteristics (sex, age, height, weight, blood pressure) and medical history, and then to monitor their morbidity and mortality from various diseases over several years. The success of a prospective study rests on being able to identify the deaths, and especially the cause of death, of persons participating in the survey (i.e. the cohort). Provided that mortality followup is feasible, the death rates of tobacco users (especially smokers) and lifelong nonusers can be compared, and relative risks calculated. 55 The great advantage of cohort studies is that they enable scientists to monitor the change in disease specific relative risks over time, by comparing death rates of smokers and nonsmokers in sequential periods of followup (i.e. every five years). Moreover, prospective studies yield information on relative risks of death from a variety of diseases. 55 There are major problems in carrying out prospective studies, especially in developing countries. These studies are expensive, time consuming and very dependent on the quality and availability of cause of death data. The linkage of baseline data with mortality data is extremely difficult, if not impossible, without adequate personal identification data. Followup can be further complicated by high migration rates. 55 The magnitude of health hazards from smoking and other tobacco use in developing countries needs to be urgently studied and monitored as the epidemic of tobaccorelated diseases in developing countries evolves. To encourage and facilitate this research, WHO, in collaboration with the IARC and the University of Oxford in the UK (a WHO Collaborating Centre) has established a network of prospective studies on tobaccorelated mortality in several developing countries including Argentina, Brazil, China, Cuba, Egypt and India. This research will yield local evidence on how the risks of tobacco use are changing which will be useful both for local action to control tobacco use and for global monitoring of the epidemic.  X' 55 Casecontrol studies: In this study design, individuals with a given disease ( cases )  X( are compared with persons without the disease ( controls ) drawn from the population from which the cases arose. Controls are selected in such a way so as to be comparableG*-++ to cases in terms of age, sex, and other sociodemographic characteristics. For example, in a casecontrol study of lung cancer, "controls" should preferably be healthy people, but might also be patient controls (including cancer controls) drawn from patients with  X diseases not causallyrelated to smoking. In a typical casecontrol study, information would be collected from all subjects on a variety of exposures including tobacco use, occupation, history of disease, indoor air pollution, etc. By comparing the proportion of smokers among the cases and controls, an "odds ratio" (or approximate relative risk) can be calculated. 55 The advantage of casecontrol studies over cohort studies is that they are relatively inexpensive, can be done rapidly on a relatively small number of individuals (typically a few hundred), and give information on current risks very soon after the study has been completed. Their main limitation is that they are restricted to the one disease under study, and do not show how the epidemic of tobaccorelated diseases is changing unless they are periodically repeated. Both types of study designs are useful in order to estimate relative risks, and either or both could be carried out in developing countries depending on the availability of resources and epidemiological skill, and the purposes of the investigation whether for monitoring the progress of the epidemic or assessing current risks.  XG  Calculating the number of deaths caused by tobacco 55 For stimulating tobaccocontrol action in a community, reliable information on mortality caused by tobacco use is extremely helpful. One means of calculating this  X death toll is to compute the population attributable risk (PAR) for each major disease caused by smoking (as described in Appendix IX.2). Multiplying this population attributable risk by the number of deaths observed in a population from each of these causes yields the estimated number of deaths caused by tobacco. This should be done  X/& separately for men and women in each age group. Thus,  tobaccoattributable mortality  X(  from disease i , TAM(i), can be expressed as: E2ddxF= 2 ddxH* E    0aTAM (i) = PAR(i) x D(i)*-++ H* X ԙwhere PAR(i) is the tobaccoattributable fraction for disease i , and D(i) is the number of  X] deaths from disease i . Smokingattributable mortality (SAM) is defined in a similar  X way, with the PAR in this case referring only to smoking risks.  Xv 55 When calculating the PAR , it is preferable that both prevalence of smoking and relative risks come from the same population at approximately the same period. Differences in smoking patterns, consumption, type of cigarette smoked, etc., mean that the tobacco exposure of contemporary populations (as measured by prevalence) and earlier cohorts will not be identical, and may not even approximate each other in many cases. This is particularly true in situations where tobacco consumption, and particularly cigarette smoking, has only recently become widespread as, for example, in developing  X countries. As demonstrated in Table IX.1 and Figure IX.3,  relative risks  rise with increasing duration of smoking. It is inappropriate to use relative risks from populations such as the USA or UK with long smoking histories to estimate smokingattributable mortality in developing countries. This approach is likely to exaggerate estimates of smokingattributable deaths in these countries. 55 The direct method for calculating smokingattributable deaths will yield reliable estimates for countries where smoking histories are comparable to those of populations used to derive the relative risks. Thus, for the USA, the Centres for Disease Control and Prevention (CDC) routinely calculate smokingattributable deaths using the attributable risk formula. The CDC have developed a computerbased package for calculating  X SmokingAttributable Mortality, Morbidity and Economic Costs (SAMMEC) which facilitates the monitoring of the epidemic in developed countries with smoking histories comparable to the United States. Countryspecific relative risks should preferably be utilised with such software programmes.  X$  Indirect methods for estimating smokingattributable deaths 55 Another way to estimate smokingattributable mortality is by assuming that the current lung cancer rate in a given country reflects adequately the entire smoking history of that country in terms of prevalence, duration, intensity and relative risk for lungJ*-++ cancer. Implicit in this approach is the basic assumption that lung cancer is essentially unicausal (i.e. smoking) and that other cofactors have a negligible impact. This also implies that lung cancer mortality among nonsmokers is similar in countries to which this method is applied. There are other causes of lung cancer, some of which have been identified, such as asbestos, while others are still unknown and can result in comparatively high lung cancer rates in nonsmoking women. This has been observed in  X. Chinese women, for example. Hence, this approach should only be applied to countries where the assumptions about lung cancer are valid, and where reliable cause of death data are readily available.  55 A more detailed description of the method is given in Appendix IX.3. The results of the application of the method to all developed countries have been published by WHO and Oxford University Press (see reading list at the end of this chapter). Variations on this approach have been reported which have yielded essentially similar results. The method described in Appendix IX.4, for example, was applied to data for Latin America countries. 55 Both of these indirect approaches yield reasonably reliable estimates of smokingattributable mortality in populations which have reliable vital statistics on causes of death and in which lung cancer death rates among nonsmokers are likely to be similar to those observed in the USA. The World Health Organization may be contacted for assistance in their application.  X]  Projecting SmokingAttributable Mortality 55 None of the methods described above for estimating current smokingattributable mortality are likely to apply for the majority of developing countries where reliable cause of death data, and information on lung cancer rates among nonsmokers, are generally not available. However, many of these countries do have reliable smoking prevalence data and the question arises as to how to use this information to support the implementation or strengthening of tobacco control policies and strategies. E*-++Ԍ55 A priority for prevention is to reduce the future mortality which will occur from  X] current smoking patterns. It is recommended, therefore, to use current prevalence to  X estimate  future deaths from tobacco and to use these projections to draw attention to the  X need for  current action to control tobacco use. 55 To make the projections, a valid estimate of current smoking prevalence in young adults (for example, ages 2024) is required. Since most regular smokers are already persistent smokers by these ages, this prevalence figure will give a good indication of those likely to be exposed to risk of a tobaccorelated death in the future. 55 Prospective studies in populations such as the USA and the UK, with long smoking histories, suggest that death rates for smokers are between 2 and 3 times those of nonsmokers at all ages. If smokers have at least a twofold excess mortality at all ages, then at least half of them will eventually die from smoking, either in middle age, or in old age. We do not know what future smoking patterns will be and how smoking will interact with other factors which could increase disease risks. Consequently future health hazards of smoking may be substantially different. For example, a reduction in prevalence could be expected to lead to lower risks in the future. If, however, these estimates do apply, then to estimate the future deaths from current smoking patterns, a useful reference group is children and youth aged 019. -++Ԓ @2 ddxH* 2ddx @   X  Suppose N is the number of children and youth alive today at ages 019 in any given population. Further, suppose the prevalence of smoking in the population  Xb (preferably at ages 2024) is P . Then N x P is the estimated number of future smokers among those alive today, aged 019, and 1/2 x N x P is the projected number of smokingattributable deaths among those alive at ages 019. For example, suppose for a developing country that we know: N = number of children alive at ages 019 = 2 000 000 P = smoking prevalence at ages 2024 = 30% Then the future number of smokers is estimated to be 600 000 (0.3 x 2 000 000), of which 300 000 are predicted to die from smoking, half (150 000) in middle age, half in old age. When done at a global level, this procedure leads to a prediction of  Xa at least 200 300 million deaths from smoking among those currently alive today at ages 019. Most of these deaths will occur around the middle of next century when annual mortality from tobacco is projected to be 1015 million.  55 Evidently these projected numbers of deaths will be lower if there is substantial quitting among those young people who currently smoke and/or if the health hazards of tobacco use in developing countries are different to those observed elsewhere. 55 This type of information, or the more detailed current estimates of smokingattributable deaths now available for developed countries, can and should be used by tobaccocontrol advocates to highlight the need for urgent tobacco control action to combat this global epidemic. It is also clear that a potential epidemic of this magnitude needs to be monitored with much greater confidence and reliability. Prospective studies which will do so are urgently required in developing countries to assess current hazards, and to monitor the much greater epidemic which, based on current smoking patterns, is yet to occur. )-++Ԍ X  CONCLUSION: KEY ISSUES IN ASSESSING THE HEALTH EFFECTS *55 Denominators (i.e. population data) are critical for calculating rates of disease and should always be used; these should be age and sex specific.%"5 *55 Care should be taken to ensure that denominators (population data) are those from which numerators (cases or deaths of tobaccorelated disease) are derived. Too often, cases are drawn from restricted samples, or are otherwise underreported, yet are compared to the total population of a region or country, thus grossly underestimating disease levels.%"5 *55 There are many sources of population data, but the most reliable will usually be population censuses (typically carried out every five or ten years), combined with intercensal estimates of the population by age and sex. Data from population censuses are collected and compiled for administrative regions of a country (as are intercensal estimates) which should facilitate the calculation of disease rates for population subgroups within a country. In some countries, continuous population registers are available, and may be used to yield population data.%"5 *55 Several sources of morbidity data might be available, including hospital records, other health services data, disease registries, and health surveys. Data from health services are likely to be biased and cannot generally be related to a populationatrisk to calculate disease rates, and thus should be used with caution.%"5 *55 Good data on cancer incidence (and on cardiovascular events) are available for certain populations, and may be used to determine rates of disease. It is strongly recommended that this source of data be encouraged, strengthened and utilized for monitoring chronic disease illness levels and trends. These registers should also be used to help develop and validate incidence monitoring systems based on hospital discharge records.%"5 C*-++Ԍ*55 Health surveys have been widely used to collect data on selfreported illnesses. Those that are only symptombased are generally of limited utility for reliably monitoring disease trends. Better data can be obtained from surveys that ask about physiciandiagnosed conditions.%"5 *55 Methods are available to estimate the number of deaths due to tobacco use. These should only be used when the assumptions of these methods have been met. In  X particular, it is  not  appropriate to use diseasespecific relative risks derived from studies in developed countries to estimate deaths from smoking in developing countries.%"5 *55 For policy purposes, a useful estimate is the projected number of deaths among young people which might be expected on current smoking patterns. %"5 *55 Reliable data on the health effects of tobacco use are useful for monitoring the epidemic and for guiding policy action; however, enough is known about the extremely hazardous nature of tobacco to take action NOW to reduce consumption.%"5 -++  X  Appendix IX.1 55 Commonly used standard populations ("World" and "European")  55 NNAge group!``)World19European  55 NN& & 0 !``) 2 400191 600 55 NN& & 14 !``) 9 600196 400 55 NN& & 59 !``)10 000197 000 55 NN& & 1014!``) 9 000197 000 55 NN& & 1519!``) 9 000197 000 55 NN& & 2024!``) 8 000197 000 55 NN& & 2529! ``)8 000197 000 55 NN& & 3034! ``)6 000197 000 55 NN& & 3539! ``)6 000197 000 55 NN& & 4044! ``)6 000197 000 55 NN& & 4549! ``)6 000197 000 55 NN& & 5054! ``)5 000197 000 55 NN& & 5559! ``)4 000196 000 55 NN& & 6064! ``)4 000195 000 55 NN& & 6569! ``)3 000194 000 55 NN& & 7074! ``)2 000193 000 55 NN& & 7579! ``)1 000192 000 55 NN& & 8084! ``) 500191 000 55 NN& & 85+ !``) 500191 000  55 NN& & Total! 100 0001 100 000  ,&---  X  Appendix IX.2  X  DERIVATION OF FORMULA FOR CALCULATING POPULATION  X ATTRIBUTABLE RISK FROM SMOKING 55 Suppose the proportion (or percentage) of a given population who smoke (i.e.  X population smoking prevalence) is p. Furthermore, let us define the incidence (or death) rate from a specific smokingrelated disease (e.g. lung cancer) as follows:  X 55 It = rate of disease (death or incidence) in the total population.  XJ 55 Ie = rate of disease among smokers (the `exposed' group).  X 55 Iu = rate of disease among non smokers (`unexposed' group) 55 Then the overall rate of disease in the total population can be expressed as the sum of the rates in the exposed and unexposed groups, weighted by the size (relative) of the two groups. That is:  @2ddx 2: ddxaN @   X  It = p. Ie + (I p). Iu00-_____ (1) a 55 Following the description given in Figure IX.2, the relative risk (RR) (or rate ratio) is defined as: 55 @2: ddxaN 2 ddx @ a  X@ RR = Ie/Iu ______ (2)   X+ 55 To compute the population attributable risk (PAR) from smoking for a specific disease, we firstly compute the excess risk (absolute) of smokers versus nonsmokers by subtracting the risk among nonsmokers (i.e. the `background' risk) from that among smokers. What remains is the absolute risk attributable to smoking. We then apply this risk to the fraction of the population who are exposed (i.e. smokers) to obtain the expected amount of disease contributed by smoking. This can be expressed algebraically as: @2 ddx 2! ddx) @   X^* PAR = p(Ie Iu)/It+_________ (3) )`,---ԌЙ55 It is then a matter of simple algebra to demonstrate that equation (3), for the population attributable risk, can be written as: 55 NN p (RR 1) 55 PAR =  55 NN p (RR 1) + 1  as given in Figure IX.2.  ---  X  Appendix IX.3& & A method for calculating smokingattributable mortality based on  X] the observed lung cancer rate. %"& 55 The method proposed here uses the absolute lung cancer rate observed in a population to estimate the proportionate mortality from other diseases attributable to smoking. A high lung cancer rate, such as that for Hungarian males, thus implies that smoking is also a major cause of death from other diseases, whereas in a population where the lung cancer death rate is still low, such as Spanish females, by implication relatively few deaths from other diseases can yet be due to smoking. To determine lung cancer deaths attributable to smoking in country "x", use the following formula: $!{dddddyddw` SAM = { f } OVER { 1+f }D  SAMh Mf:1F::fD$ In this formula, D is the number of lung cancer deaths in country "X" and: _Adddddddw` 1f = { L SUB x L SUB usns } OVER { L SUB usns }  fJ +dLdd4/xbddLdd(/usnsL_Ldd*usns_ where: 55 x = country "x" %"5 55 L = the lung cancer death rate per 100 000 population%"5 55 usns =& & U.S. nonsmoker lung cancer death rate as determined by ACS%"& 55 NN& & CPSII (from Table 1)%"& 55 uss =NN& & U.S. smoker lung cancer death rate as determined by ACS%"& 55 NN& & CPSII (from Table 1)%"& 55 This calculation should be done separately for males and females and for each age group for which data are available. 55 To estimate smokingattributable mortality from diseases other than lung cancer, a+--- more complicated procedure is required since it cannot be assumed that the absolute rates among nonsmokers will be comparable in different populations as was done for lung cancer. There may well be important differences in other major risk factors for vascular disease (e.g. hypertension, blood lipid levels), or upper aerodigestive cancers (alcohol) among nonsmokers in different countries. 55 Smokers may also be expected to be generally less healthconscious than nonsmokers, and more likely to adopt other deleterious health habits (e.g. poor diet, excessive alcohol consumption) with either independently or synergistically interact with smoking to increase their risk of death. Thus, in the CPSII cohort, part of the excess mortality of smokers from diseases other than lung cancer may well be due to factors other than smoking. In an attempt to control for this confounding, thereby ensuring that  X\ the risks of tobacco are not exaggerated, the estimated excess mortality of smokers is  X halved before calculating the attributable mortality due to smoking. 55 55 That is, for all diseases "i", other than lung cancer, smoking attributable mortality (SAM) is calculated as: adddddddddw`$; SAM SUB i = f SUB i OVER { 2 + f SUB i } D SUB i  SAMddki )mdfdd/i_2f__fdd*i1Dddi߅  XJ where Di is the total number of deaths due to disease "i" and mD!dddddbddw` f SUB i = (RR SUB i1)(SIR) _fddF*ic__(_RRdd*i_V_1_)_(_SIR_)m  Xa where RRi is the known relative risk of smoking for disease "i", as determined by the American Cancer Society Cancer Prevention Study II (ACS CPSII) (from Table 2) and  X" SIR is the Smoking Impact Ratio as determined by the following formula: u)dddddOddw` :DSIR SUB x = { L SUB x L SUB usns } OVER {L SUB uss L SUB usns }  SIRddxC dLddX/xddLddL/usns_Ldd-*uss__Lddv*usns:߷ 55 This calculation should also be repeated for males and females separately, and for each agegroup. +---ԌЙ55 SAM is then summed over all agesexcause groups for which it was calculated to obtain an estimate of total smokingattributable mortality in country "x". 55 This procedure was developed by Peto and colleagues in the early 1990s. Further details of the method, and its application to the developed countries, can be found in the reference to the study provided in the reading list for this chapter.  X  TABLE 1:& & Lung cancer: annual rates per 100 000 in 5year and 10year age groups among current cigarette smokers and lifelong nonsmokers in the ACS CPSII prospective survey%"& #Y P7!1P#)a h2! ddx) Addx ||||||||| h &    )& \ MALES\ :FEMALES0       0  AGE & SMOKERS   (crude)  (uss) "_NONSMOKERS "/(smoothed) $(usns) 3SMOKERS 3(crude) 4(uss) @7NONSMOKERS @(smoothed) B(usns)(     \( 3539  *  12  2  2  *  10  2  2 (     ( 4044  *   3   *   3 (     ( 4549r  35r  91r  5r  6r  49r  63r  4 r  6(     ( 5054   114     7     71     7  (     r( 5559  227  296   10  12  136  164  10  12(      ( 6064  375   14   195    14 (     ( 6569d  599d  708d  20d  23d  310d  321d  19d  22 (     ( 7074C!  899C! C!  27C! C!  339 C! C!  26C! (     ( 7579"# 1168"#  1174"#  35"#  46"#  429 "#  420"#  34"#  39(     !( 80+$ 1191$ $  46$ $  400$ $  44$ 0      "#0  mJD% RR** s vs. ns' 26' <13   $ )a   *55 Inadequate data to estimate reliably at ages 3544%"5 **55 MantelHaenszel analysis of all ten age groups, without smoothing. For relative risk with smoothing, see Table 2.%"5 +---Ԍ mJ  TABLE 2:NNCigarette smokers versus "nonsmokers" (never smoked regularly): selected risk ratios* from years 3 to 6 inclusive (approximately 198488) of ACS CPSII prospective study of one million U.S. adults%"N r Addx ||||||||| addx$ E r   $ X"ICD9 L?Male !!EFemale  $ Lung cancer (ICD 162)x @?24.22x ?!!F12.50   Upper aerodigestive cancermouth, pharynx, larynx or oesophagus (ICD 140150 and 161) M? 7.87 L!!F 6.95  x Other cancer (rest of ICD 140209) M? 1.69 L!!F 1.20   Chronic obstructive pulmonary disease (ICD 4902, 4926)F @?13.82F ?!!F14.21   mJ Other medical causes (rest of ICD 000799)**   F  Age3559z M? 3.05z L!!F 2.69    6064 M? 2.31 L!!F 2.68  z  6569 M? 2.09 L!!F 2.52   7074H M? 2.00H L!!F 2.00    75+ M? 1.54 L!!F 1.44   H *55 Risk ratios are standardized by the method of Mantel and Haenszel for whichever are relevant of the fiveyear age groups from 3034 to 7579, and 80+. Female risk ratios may rise in future years in the U.S., at least in the older age groups as women who have smoked for only part of their lives are replaced by lifelong smokers.%"5 **55 Except in extreme old age, the chief other medical causes were vascular disease, particularly coronary heart disease and stroke. For vascular diseases alone, the agespecific risk ratios for the same age groups shown in the table, respectively, were 3.45, 2.33, 2.01, 1.87 and 1.53 for males, and 2.96, 2.89, 2.53, 2.09 and 1.43 for females.%"5 ! --- #Xi\  P6ƒXP# Appendix IX.4 Method used for calculating smokingattributable mortality in the  X Americas  Xt  1. Estimate overall mortality For each country, evaluate vital registration and use the portion of the data that provides an accurate populationbased mortality estimate. For the 10 jurisdictions without mortality data, use United Nations population schedules and apply mortality rates from countries with similar sociodemographic configurations. Do not correct for underreporting. Exclude and do not correct for illdefined causes.  X  2. Estimate causespecific mortality Identify the major smokingassociated disease groups (coronary heart disease; cerebrovascular disease; lung cancer; oral, laryngeal, and oesophageal cancer; bladder cancer; and chronic obstructive pulmonary disease [COPD]). Use causespecific mortality data for countries for which such data are available. For the 10 jurisdictions without such data, use data from four countries representative of the demographic and socioeconomic spectrum of the Americas (Guatemala, Colombia, Argentina, and the United States).  X,&  3. Estimate relative risk and attributable risk Use US estimates for relative risk since countyspecific relative risk is generally not available. +!---ԌDetermine the smokingattributable fraction (SAF) for the US by using the attributablerisk calculation.  X  4. Adjust estimates by using an index related to lung cancer Use an index of the maturity of the epidemic that relates the lung cancer rate for each country to that of the United States. For each country, determine an adjusted SAF for each disease by multiplying the index by the US SAF for each disease. For each country, multiply the adjusted SAF for each disease by the number of deaths from the disease to obtain smokingattributable mortality (SAM) (approximately 375 000).  Xs  5. Adjust the estimate further Calculate SAM for the US alone by using this method and compare the result with the official value reported for 1985 (US Department of Health and Human Services 1989). For each cause, calculate the difference between the result from this method and from the official method. Apply these upward adjustments to the causespecific SAMs: increase COPD by 230%, increase cancers by 10.4% (using the difference in lung cancer estimates), and increase other diseases and causes by 16.4%. Calculate the adjusted estimate of SAM in the Americas (526 000). ,&"---  #c P7ApP#