Aim
Our aim was to evaluate the effect of age, period and birth cohort on lung cancer mortality inequalities in men and women aged 30 years and older in Andalusia, the southernmost region in Spain, from 2002 to 2016.
Design
We used the Longitudinal Database of the Andalusian Population (LDAP), which started with the population registered in the 2001 Population and Housing Census (7,357,547 individuals) residing in Andalusia on the 1 January 2002 (19, 20). This population was tracked until 31 December 2016. The LDAP merges information from the 2001 Population and Housing Census with events recorded in the Natural Population Movement (NPM) database, such as deaths, births, marriages and changes in the residential status that have occurred since 2002. The end of follow-up could be the result of (i) a death registered in the NPM, (ii) emigration outside Andalusia or (iii) censorship due to termination of the study.
Setting
This study was carried out in Andalusia, the most populated region in Spain, which had 8,403,936 inhabitants in 2016. Economic indicators there are largely are below the European average, with the region having the highest poverty rate (32.3% in 2021) among the Spanish autonomous communities. Health indicators are also well under the average, with a high mortality rate in both men and women compared to the rest of the regions.
Lung cancer is the first cause of cancer mortality (19.5% of cancer-related deaths) and the fourth most frequent cancer in Spain. Mortality rates in women continue to increase annually, while consistently trending downward in men (21). In Andalusia, an east-west mortality pattern in men has been detected for several cancer types, including lung cancer, with higher mortality rates in the west. Moreover, an association between deprivation at the small area level and lung cancer mortality in men has been shown. This association is negative for women(22). In 2017, lung cancer accounted for more than 30% of tobacco-attributable mortality in Andalusia among those 35 years and older(18).
In Spain and Andalusia, smoking was a widespread practice, with a smoking rate of more than 60% among men until 1980. This percentage has been decreasing since then, and in 2020, 26% of Spanish men considered themselves to be current smokers(23). In women, tobacco consumption was very low until 1970, increased rapidly until 1990, and has been decreasing since then. In 2020, 19% of Spanish women were smokers(23).
Population And Variables
The initial census population of 7,357,547 individuals was tracked for 15 years until December 2016, yielding 98,842,980.9 person-years of follow-up (48,415,311.3 men-years and 50,427,669.6 women-years). Individuals were living in 5,381 census tracts corresponding to 770 municipalities in 8 provinces.
Assessment of individual variables
The outcome of this study was individual lung cancer mortality as assessed from the time of the 2001 census until 31 December 2016. The locations of tumours analysed corresponded with the International Classification of Diseases (10th rev.) codes C33 and C34, malignant neoplasm of trachea, bronchus and lung.
We considered 3-year age groups (i.e. 12–14, 15–17, 18–20, 21–23, and so on) and time intervals for calculating death rates were divided into five three-year periods (2002–2004, 2005–2007, 2008–2010, 2011–2013 and 2014–2106). We classified educational level into five categories: very low (illiterate or less than one year of formal education), primary (elementary school, i.e. 6 to 8 years of formal education), secondary first circle (elementary baccalaureate or similar degree), secondary second circle (up to12 years of formal education) and university studies. Information was available at the census tract level. Therefore, data were structured in groups defined by two sex categories (men and women), 34 age categories, 5 education categories, 5 time periods and 5,381 census tracts, yielding 9,147,700 groups. Each group had its corresponding number of person-years and count of lung cancer deaths.
To assess the contextual socioeconomic status, we used a deprivation index (DI) at the census tract level(24). The index was constructed with data from the 2001 Population and Housing Census regarding (i) percentage of people with low educational level, (ii) percentage of unskilled workers and (iii) unemployment rate. We carried out a principal component analysis to calculate the DI that separated the census tracts into five levels of deprivation, according to the quintiles of the respective factorial scores. Census tracts with the lowest social deprivation were designated level 1, and those with the highest social deprivation were designated level 5.
Statistical Analyses
First, the number of lung cancer deaths and person-years of follow-up were presented for period, educational level and DI, separately for each sex. In a second step, world population age-adjusted rates(25) by period, educational level and deprivation quintile were calculated. Third, three-year group age-specific rates were estimated for each time period, and lung cancer mortality rates by age group and birth cohort were estimated for both men and women. In a fourth step, we used Poisson regression models to assess temporal trends in inequalities, using education and deprivation as socioeconomic variables.
Finally, we used an age-period-cohort analysis trying to uncover the diverse birth-cohort and period effects related to the epidemiology of lung cancer in different socioeconomic groups. Age-period-cohort modelling is a well-known quantitative method used to improve understanding of disease trends by attempting to unravel the factors influencing all ages. In addition to the evaluation of age effects as more related to biological or social factors, these models allow researchers to assess period effects, such as changes in medical practice or in public health policies that occur simultaneously, and cohort effects, which are related to circumstances that affect an entire generation, such as similar behaviours, exposures to risk or protective factors(26). In this case, in order to extend our research with a focus on social inequalities in health, we carried out an age-period-cohort analysis according to educational level.
These models suffer from an identifiability problem, since cohort = period-age. The literature reports different approaches to solve this drawback. We used Rutherford's Stata package "apcfit"(27), which is based on Carstensen's method(28), and which takes age, period and cohort as continuous variables using appropriate cubic spline functions in the framework of a generalised linear model. We used diverse parametrizations for the models, and chose the best-fitting following the Akaike information criterion. In our parametrization, age effects were expressed as rates, period effects as rate ratios relative to the reference period, and cohort effects as rate ratios constrained to be 0 on average on the log scale.
All analyses were performed using Stata software version 16.