CHANGES IN LUNG-CANCER MORTALITY TRENDS IN SPAINJose FRANCO

1* , Santiago PEREZ-HOYOS2 and Pedro PLAZA

3

1Servicio de Neumologıa, Hospital Clınico Universitario, Valencia, Spain2Escuela Valenciana de Estudios para la Salud, Valencia, Spain3Servicio de Neumologıa, Hospital Universitario Dr. Peset, Valencia, Spain

Several changes in smoking patterns over the past decadesin Spain can be expected to result in a shift in lung-cancermortality rates. We examined time trends in lung-cancermortality from 1973–1997 using a log-linear Poisson age-period-cohort model. The standardized lung-cancer mortal-ity rate for men almost doubled, from 31.4 per 100,000 in1973 to 58.6 in 1997, with an average annual increase of 2.7%.Mortality increased for male generations born until 1952 as aconsequence of the increasing cigarette smoking in succes-sive birth cohorts. However, the slight downward trend ob-served for the 2 youngest generations suggests a more favor-able outcome of the lung-cancer epidemic among Spanishmales in the coming years, if this trend continues. Forwomen, mortality rates were 5 to 9 times lower than thosefor men, 6.3 per 100,000 in 1973 and 6.4 in 1997. However,the increasing mortality among younger generations bornsince 1942 reflects the rise in the prevalence of smokingwomen during the last decades and can be expected tospread to older age groups as a cohort effect, indicating theearly phase of the smoking-related lung-cancer epidemicamong Spanish females. The decreasing mortality trend ob-served in women until the late 1980s could be attributed to alower exposure to environmental tobacco smoke at home asa result of a significant reduction in the prevalence of smok-ing men.© 2002 Wiley-Liss, Inc.

Key words: lung-cancer mortality; Spain; smoking habits; age-peri-od-cohort model; time trends

In Mediterranean countries and in Spain since 1990, malignanttumors are the leading cause of death, whereas heart diseases areplaced second.1 Lung cancer is the most frequent fatal canceramong men in Spain2 as well as in many other developed coun-tries.

Because of the poor survival from lung cancer, there is aclosecorrelation between incidence and mortality. The overall 5-yearsurvival rate of 10–13% has not changed over the past 2 decades,though the implementation of multimodality therapy in locallyadvanced disease has begun to modestly improve survival inpatients with more advanced stages of disease.3

Cigarette smoking plays adominant role in lung-cancer causa-tion, being responsible for up to 90% of the lung-cancer epidemic,not only directly but indirectly (passive smoking) and in associa-tion with other substances such as asbestos and radon.4 Smokingpatterns have changed markedly and in different directions inseveral countries over the past decades; therefore, time trends inlung-cancer mortality differ between countries, cohortsand sexes.5Several changes in smoking habits in Spain, such as adecline inthe prevalence of smoking among men and a rise among women,can be expected to result in ashift in lung-cancer mortality trends.

Because tobacco smoking habits are generally established at anearly age and are characteristic for a given birth cohort,5 thedevelopment of the lung-cancer epidemic can be analyzed mostaccurately by studying age-specific rates by birth cohort.6

We analyzed trends in lung-cancer mortality in Spain from1973–1997, using a Poisson log-linear age-period-cohort model.

MATERIAL AND METHODS

Population figures and data on deaths from lung cancer in Spainduring the period 1973–1997 were obtained from official publica-

tions of the Instituto Nacional de Estadıstica (INE, National Insti-tute of Statistics). The population was that estimated at 1 July ofeach year by the INE based on official censuses. Deaths from lungcancer corresponded to code162 of the International Classificationof Diseases, Eighth and Ninth Revisions (ICD-8 and ICD-9).

From these data, age-specific death rates for 5 calendar periodsof 5 years each (1973–1977 to 1993–1997) and 9 age groups of 5years each (30–34 to 70–74 years) were derived. Using thisclassification of age and time, 13 overlapping birth cohorts of 10yearseach were identified and defined according to thecentral yearof birth (1902/3 to 1962/3). Because death certification is lessreliable at elderly ages and problems arise from random variationfrom small numbersat young ages, only theagegroupsbetween 30and 74 years were considered.7

Age-standardized mortality rates were calculated for men andwomen using the direct method with the Spanish population of1980 as the reference.

The percent annual change in age-specific rates was calculatedfrom a regression equation after transforming the dependent vari-able (single calendar year rate) into natural logarithms, with theyear as the independent variable. The coefficient B (slope) in theequation with a 95% confidence interval (CI) was considered themean annual percentage of variation.

Based on the matrix of lung-cancer deaths and population bro-ken down by 5-year calendar period and age group, the effects ofage, period and cohort were calculated through a log-linear Pois-son model fitted using the S-PLUS program (Insightful Corp.,Seattle, WA) with the appropriate macro adapted from the Decarliand La Vecchia8 GLIM macro. Estimates were derived from themodel, including the 3 factors that minimize the sum of theEuclidean distances from the 3 possible 2-factor models: ageperiod (AP), age cohort (AC) and cohort period (CP). The proce-dure of minimization is based on the least squares weighted on theinverse of the log-likelihood of each 2-factor model.7

Log-likelihood ratio statistics (deviance) were used to test thegoodness of fit, while the difference between 2 models was testedby comparing the change in the deviance with the degrees offreedom. This test, however, often indicates lack of fit in popula-tion-based data even when the model appears qualitatively todescribe the data well since the number of events (deaths) is oftenlarge, so even small departures from the model are detected.9Assuming no period or cohort slope, the average of the periodvalues was fixed at unity, as was the average of the cohort values;thus, age values were of the same order of magnitude as age-specific rates.7

Becauseage, period and birth cohort variablesarearithmeticallyinterrelated, knowing 2 of them fixes the third, implying that thechosen solution is not unique,10 a problem known as nonidentifi-

*Correspondence to: Servicio de Neumologıa, Hospital Clınico Univer-sitario, Avda. Vicente Blasco Ibanez 17, 46010 Valencia, Spain.Fax: 134-96 3862657. E-mail: jfs01v@nacom.es

Received 4July 2000; Revised 17 April , 19 July 2001; Accepted 30 July2001

Int. J. Cancer: 97, 102–105 (2002)© 2002 Wiley-Liss, Inc.

Publication of the International Union Against Cancer

ability of parameters. Moreover, when the majority of age-specificrates are in the same direction, both cohort-of-birth and period-of-death patterns are similar and it is difficult to establish whether themajor underlying trend is a cohort or period effect. Consequently,the results of the model should be chiefly viewed as a guide towardsummarizing overall tendencies, and age-specific rates should beconsidered before any conclusions are drawn.7

RESULTS

The standardized lung-cancer mortality rate for men almostdoubled over the study period, from 31.4 per 100,000 in 1973 to58.6 in 1997, with an average annual increase of 2.7% (95% CI2.4–3.0). For women, mortality rates were 5 to 9 times lower thanthat for men (Fig. 1), 6.3 and 6.4 per 100,000 in 1973 and 1997,respectively; there was a slight annual decrease of –0.7% (95% CI

–1.1 to –0.3) until 1988 and then an increase of 1.5% (95% CI1.1–1.9) annually.

Age-specific trends in lung-cancer mortality rates for menshowed a statistically significant annual increase over the wholestudy period in all groups aged 35 years and over, ranging from2.1% (95% CI 1.8–2.4) annually for the age group 65 to 69 yearsto 4.5% (95% CI 3.7–5.2) for the age group 40 to 44 years. In theyoungest group (30–34 years), rates increased until 1988, with anopposite trend thereafter of –5.4% (95% CI –10.1 to –0.7) annu-ally.

For women, the patterns were quite different: age-specific mor-tality rates showed a statistically significant decrease over thewhole study period in groups aged 55 years and over, ranging from–0.5% (95% CI –0.1 to –0.9) annually among 70- to 74-year-oldsto –1.0% (95% CI –0.3 to –1.6) among 55- to 59-year-olds; thisdecline was also observed until 1984 for groups aged 35 to 54years. However, since 1984, rates have increased significantly inyounger groups (30–49 years), ranging from 3.9% (95% CI 1.6–6.1) annually in 45- to 49-year-olds to 7.9% (95% CI 3.6–12.1) in35- to 39-year-olds.

Tables I and II show the goodness of fit (scaled deviances) forthe age-period-cohort models and comparisons between models.Mortality trends for both genders were better explained by the full

TABLE I – GOODNESS OF FIT FOR AND COMPARISONS BETWEENDIFFERENT AGE, PERIOD AND COHORT POISSON REGRESSION MODELS

OF LUNG-CANCER MORTALITY FOR MALES IN SPAIN

ModelResidual Change

Deviance d.f. Deviance d.f. p

Age (A) 6,031.9 36 5,641.5 4 ,0.0011

5,594.0 12 ,0.0012

Age-period (AP) 390.4 32 320.0 11,0.0013

Age-cohort (AC) 437.9 24 367.5 3,0.0014

Age-period-cohort(APC)

70.4 21

1A vs.AP.–2A vs.AC.–3AP vs.APC.–4AC vs.APC. d.f., degrees offreedom.

TABLE II – GOODNESS OF FIT FOR AND COMPARISONS BETWEENDIFFERENT AGE, PERIOD AND COHORT POISSON REGRESSION MODELS

OF LUNG-CANCER MORTALITY FOR FEMALES IN SPAIN

ModelResidual Change

Deviance d.f. Deviance d.f. p

Age (A) 187.7 36 69.4 4,0.0011

117.1 12 ,0.0012

Age-period (AP) 118.3 32 91.6 11,0.0013

Age-cohort (AC) 70.6 24 43.9 3,0.0014

Age-period-cohort(APC)

26.7 21

1A vs.AP.–2A vs.AC.–3AP vs.APC.–4AC vs.APC. d.f., degrees offreedom.

FIGURE 1 – Age-standardized lung-cancer mortality rates for menand women in Spain, from 1973–1997.

FIGURE 2 – Age-specific lung-cancer mortality rates by 5-year ageintervals and birth cohort for Spanish males.

FIGURE 3 – Results of age, period and cohort modeling for Spanishmen. Age values are expressed as rates per 100,000 population. Peri-od-of-death and cohort-of-birth values are expressed in relative termsagainst their weighted average set to unity.

103LUNG-CANCER MORTALITY IN SPAIN

age1 period1 cohort (APC) model, though the APC model fitmortality data only for women (deviance5 26.7 on 21 degrees offreedom,p 5 0.18). For men, deviances were greater than thosefor women and the APC model did not fit, probably due to thelarge number of deaths, though overdispersion could exist. Onemust be cautious with the interpretation of the multiplicative age,period and cohort effects resulting from the modeling. Both periodand cohort effects were statistically significant for men and womenwhen the APC model was compared with the AC or AP submodel.As expected, mortality increased with age in both sexes. For men(Figs. 2, 3), the cohort effect was steadily upward to the cohortborn around 1952, with a slight reversal downward for the 2youngest generations; also, a rising period effect was observed upto 1992. For women (Figs. 4,5), cohort values decreased moder-ately until the cohort born around 1932, then leveled off andincreased for young women born since 1942; moreover, there wasevidence of a period effect downward until 1987 and upwardthereafter.

DISCUSSION

Recent changes in mortality from lung cancer are mainly relatedto changes in smoking patterns over the past decades.11 Unfortu-

nately, in Spain, question-based studies on the general public’stobacco consumption are available only since 1978. Rates ob-served in young adults should reflect recent carcinogenic exposureand can be expected to spread to older age groups in future years12

because smoking habits tend to be a characteristic of particulargenerations. Our analysis of lung-cancer mortality in Spain showsdifferent trends for men and women.

In men, mortality increased for generations born up to 1952 asa consequence of the increasing cigarette smoking in successivebirth cohorts. However, the decrease in the prevalence of smokingmen (Fig. 6) from 64% in 1978 to 45% in 199713,14and the slightdecrease in mortality for the youngest cohorts observed in ourstudy suggest a more favorable outcome of the lung-cancer epi-demic among Spanish males in the coming years, though majorefforts to discourage tobacco use should be made so that thedecreasing trend in the number of smoking men continues.

The relative risk of lung cancer decreases following smokingcessation,15 and the impact of the decline in cigarette smoking onlung-cancer mortality would be expected to show up first in theyounger groups as a cohort effect. The downturn in cigaretteconsumption since the 1960s combined with lower tar content ofcigarettes was a clear precursor to the decline in lung-cancermortality rates in the United States and the United Kingdom.6

According to data from Tabacalera, a state owned company thatmonopolized the tobacco market in Spain until 1998, the contin-uous increase in cigarette sales began to decline in the early 1990s,with a decrease of 17.1% from 2,685.5 cigarettes per adult aged 15years and older in 1991 to 2,226.2 in 1996.16,17However, sales ofblack tobacco, which has been related to a higher risk of lungcancer in case-control studies,18 decreased over the last decades,falling from 94% of total sales in 1961 to 31.3% in 1996, beingreplaced by Virginia tobacco.

The low lung-cancer mortality rate observed among Spanishwomen differs hardly at all from what might be expected if nonehad ever smoked, suggesting that few of their lung-cancer deathsare still due to smoking.19 However, the increasing mortalityamong younger generations born since 1942 reflects the rise intobacco consumption among females during the last decades andcan be expected to spread to older age groups in the coming yearsas a cohort effect. The increasing prevalence of smoking women(Fig. 6) by 58.8%, from 17% in 1978 to 27% in 1997,13,14and theupward mortality trend in young adult life indicate the early phaseof the smoking-related lung-cancer epidemic among females. Thelater onset of the epidemic in Spain compared with other devel-oped countries can be attributed to smoking having become wide-spread among women only in the last few decades as well as to the

FIGURE 4 – Age-specific lung-cancer mortality rates by 5-year ageintervals and birth cohort for Spanish females.

FIGURE 5 – Results of age, period and cohort modeling for Spanishwomen. Age values are expressed as rates per 100,000 population.Period-of-death and cohort-of-birth values are expressed in relativeterms against their weighted average set to unity.

FIGURE 6 – Prevalence of smoking among men and women aged 16years and above in Spain, from 1978–1997.

104 FRANCO ET AL.

latency period between the onset of exposure and the developmentof disease.

The small but statistically significant decrease of mortality ob-served in Spanish women until 1988 is difficult to explain.5 Al-though caution is warranted in making inferences on the basis ofstatistical modeling,20 the age-period-cohort analysis shows a co-hort effect for generations born up to 1932, aside from the decreasein mortality with the period of death up to 1987. In addition,age-specific rates decreased continuously for women aged 55 yearsand over and until the mid-1980s for younger women.

In countries with a female population that smokes relativelyrarely and a high male smoking prevalence, the risk and populationburden of lung cancer due to environmental tobacco smoke (ETS)are considered relatively important.21 Therefore, because of thereduction in the prevalence of smoking men in Spain, we canhypothesize that the decreasing mortality trend observed in womenmight be related to changes in exposure to ETS at home, thoughwe cannot exclude a role of other modifiable risk factors, such asdomestic radon, dietary factors, occupational carcinogens and aircontamination. Exposure to ETS from a spouse is a risk factor forlung cancer among nonsmoking women, and the risk increasesconsistently with increasing levels of exposure.22 In Spain, forsome decades, while smoking was rare among women, a greatnumber of nonsmoking wives would have been exposed to ETSfrom their husbands’ smoking.

Changes in the prevalence of risk factors usually alter thepattern of risk seen among birth cohorts; however, a substantialdecrease in a relatively common carcinogenic exposure couldcause a calendar-period decrease in risk after a sufficient latencyperiod. The effect of reducing tobacco carcinogen exposure on thelate stage of the carcinogenic process will be seen soon after thechange in exposure.20 The risk of lung cancer decreases with timesince cessation of ETS exposure, and there is no detectable risk

after a 15-year period.23 Therefore, once a significant number ofsmoking men quit, the reduction in risk for wives decreases as acohort effect and, by affecting all age groups simultaneously, canbe manifest also as a period effect. Thus, the decreasing lung-cancer mortality trend observed among Spanish women until thelate 1980s could be attributed to a lower exposure to ETS at homeas a result of a significant reduction in the prevalence of smokingmen.

The quality of official data on mortality can be affected indifferent ways. Changes in the international death classificationrules or inaccuracy in certification may lead to variation in mor-tality statistics. Two ICD revisions came into force in Spain duringthe study period; but code 162, assigned to malignant tumors of thetrachea, bronchus and lung, remains unchanged for both ICD-8and ICD-9 revisions. Previous studies on the quality of deathcertificates in Spain have shown reliable data on the cause of deathwith regard to malignant tumors.24,25

In summary, on the basis of the variations observed amongyounger generations, changes in the epidemiologic trends of lung-cancer mortality rates can be expected in Spain. The upward trendobserved in younger women shows the beginning of the smoking-related lung-cancer epidemic due to the continuous increase in theprevalence of smoking women. Conversely, considering the de-creasing tobacco consumption among men and the decline inmortality observed in the youngest, an opposite downward trendcan be expected if the prevalence of smoking men continues todiminish. The decrease in mortality observed for women until thelate 1980s could be related to a lower spousal ETS exposure.

ACKNOWLEDGEMENTS

We thank Dr. V. Moreno from Institut Catala` d’Oncologia forproviding theS-PLUSmacro.

REFERENCES

1. Alonso I, Regidor E, Rodrı´guez C, et al. Principales causas de muerteen Espana, 1992. Med Clin (Barc) 1996;107:441–5.

2. Fernandez E, Borras JM, Levi F, et al. Mortalidad por cancer enEspana, 1955–1994. Med Clin (Barc) 2000;114:449–51.

3. Reif MS, Socinski MA, Rivera MP. Evidence-based medicine in thetreatment of non-small-cell lung cancer. Clin Chest Med 2000;1:107–20.

4. Yesner R. Pathogenesis and pathology. Clin Chest Med 1993;1:17–30.

5. Coleman MP, Esteve J, Damiecki P, et al. Trends in cancer incidenceand mortality. IARC Sci Publ 121. Lyon: IARC, 1993.

6. Kubik AK, Parkin DM, Plesko I, et al. Patterns of cigarette sales andlung cancer mortality in some central and eastern European countries,1960–1989. Cancer 1995;75:2452–60.

7. La Vecchia C, Negri E, Levi F, et al. Cancer mortality in Europe:effects of age, cohort of birth and period of death. Eur J Cancer1998;34:118–41.

8. Decarli A, La Vecchia C. Age, period and cohort models: review ofknowledge and implementation in GLIM. Riv Stat Appl 1987;20:32–46.

9. Moolgavkar SH, Lee JAH, Stevens RG. Analysis of vital statisticsdata. In: Rothman KJ, Greenland S. Modern epidemiology. NewYork: Raven Press, 1998. 481–97.

10. Osmond C. Using age, period and cohort models to estimate futuremortality rates. Int J Epidemiol 1985;14:124–9.

11. Bailar JC, Gornik HL. Cancer undefeated. N Engl J Med 1997;336:1569–74.

12. Doll R. Progress against cancer: an epidemiologic assessment. Am JEpidemiol 1991;134:675–88.

13. Regidor E, Rodrı´guez C, Gutierrez-Fisac JL. Indicadores de Salud.Tercera evaluacion en Espana del programa regional europeo Saludpara todos. Madrid: Ministerio de Sanidad y Consumo, 1995.

14. Ministerio de Sanidad y Consumo. Encuesta Nacional de Salud deEspana 1997. Madrid: Ministerio de Sanidad y Consumo, 1999.

15. Halpern MT, Gillespie BW, Warner KE. Patterns of absolute risk oflung cancer mortality in former smokers. J Natl Cancer Inst 1993;85:457–64.

16. Tabacalera SA. Series historicas de consumo de tabaco elaborado,1957–1991. Madrid: Tabacalera SA, 1992.

17. Tabacalera SA. Memoria de actividades, 1992–1996. Madrid: Taba-calera SA, 1997.

18. Armadans-Gil I, Vaque-Rafart J, Rossello J, et al. Cigarette smokingand male lung cancer risk with special regard to type of tobacco. IntJ Epidemiol 1999;28,614–9.

19. Peto R, Lopez AD, Boreham J, et al. Mortality from tobacco indeveloped countries: indirect estimation from national vital statistics.Lancet 1992;339:1268–78.

20. Jemal A, Chu KC, Tarone RE. Recent trends in lung cancer mortalityin the United States. J Natl Cancer Inst 2001;93:277–83.

21. Lee CH, Ko YC, Coggins W, et al. Lifetime environmental exposureto tobacco smoke and primary lung cancer of non-smoking Taiwanesewomen. Int J Epidemiol 2000;29:224–31.

22. Zhong L, Goldberg MS, Parent ME, et al. Exposure to environmentaltobacco smoke and the risk of lung cancer: a meta-analysis. LungCancer 2000;27:3–18.

23. Boffetta P, Agudo A, Ahrens W, et al. Multicenter case-control studyof exposure to environmental tobacco smoke and lung cancer inEurope. J Natl Cancer Inst 1998;90:1440–50.

24. Benavides FG, Bolumar F, Peris R. Quality of death certificates inValencia, Spain. Am J Public Health 1989;79:1352–4.

25. Panella H, Borrell C, Rodrı´guez C, et al. Validacion de la causabasica de defuncion en Barcelona, 1985. Med Clin (Barc) 1989;92:129 –34.

105LUNG-CANCER MORTALITY IN SPAIN