Age, Period, and Cohort Analysis of Smoking Intensity Among Current Smokers in Malaysia, 1996–2015

Abstract

Introduction

Tobacco use is one of the major preventable risk factors for premature death and disability worldwide. Understanding the trend of tobacco use over time is important for informed policy making.

Aims and Methods

The present study aimed to examine the changes in mean daily cigarette consumption among random samples of the Malaysian current smoker population over 20 years using an age-period-cohort (APC) approach. We conducted APC analysis using a multilevel hierarchical age-period-cohort model and data from four nationally representative, repeated cross-sectional surveys (National Health and Morbidity Survey) conducted in 1996, 2006, 2011, and 2015 among individuals aged 18 to 80 years. Analyses were also stratified by gender and ethnicity.

Results

Overall, mean daily cigarette consumption (smoking intensity) among current smokers increased with age until 60, after which a drop was observed. There were increases in daily cigarette consumption across birth cohorts. Age and cohort trends did not vary by gender but by ethnicity. The decreasing cigarette consumption after age 60 among the current smoker population was consistent with those observed among the Chinese and Indians, a trend that was not observed in Malays and other aborigines. In contrast, the increasing cohort trend was consistent with those observed among the Malays and other bumiputras.

Conclusions

The present study highlighted important ethnic-specific trends for mean daily cigarette consumption among the current smoker population in Malaysia. These findings are essential in guiding the formulation of interventional strategies or implementation of national tobacco control policies and help achieve the Ministry of Health Malaysia’s 2025 and 2045 targets for smoking prevalence.

Implications

This is the first APC study on smoking intensity among current smokers in a multiracial, middle-income nation. Very few studies had performed gender- and ethnic-stratified APC analyses. The ethnic-stratified APC analyses provide useful insights into the overall age and cohort trends observed among the current smoker population in Malaysia. Therefore, the present study could add evidence to the existing literature on the APC trends of smoking intensity. The APC trends are also important in guiding the government to develop, implement, and evaluate antismoking strategies.

Introduction

Smoking-related morbidity and mortality remain a significant public health issue worldwide, with low- and middle-income countries facing a more critical challenge since they constitute more than 80% of the world’s smoker population.¹ In Malaysia, the prevalence of current smoking is still plateauing at about 25% since 1996 despite continuous concerted antismoking efforts. As advocated in the WHO Framework Convention Tobacco Control’s MPOWER approach, “Monitoring tobacco use and prevention policies,” accurate analysis of the smoking trend among the Malaysian population is imperative to understand temporal changes in smoking and its long-term policy effects on guiding the implementation of more effective public health strategies.

Malaysia, a middle-income developing country, is experiencing an aging population (age effect), rapid socioeconomic development (period effect), and demographic transition that leads to changes in age structure as the younger cohort replaces the older ones (cohort effect). These secular changes can influence smoking patterns and behaviors. Therefore, decomposing the independent effects of age, period, and cohort from the temporal trend of smoking among Malaysian adults is critically important.

Several local cross-sectional studies have shown that smoking prevalence decreased with advancing age (age effect).² Because of the different working definitions for the current smoker in the national health surveys, the National Health and Morbidity Survey (NHMS), the trend for smoking has not been reported.³ However, a direct comparison of smoking prevalence among Malaysian adults across NHMSs showed that smoking prevalence has been plateauing (period effect). Furthermore, the prevalence of current smoking among the younger generation (adolescents aged 15–24 years) had increased by 2.3% in 2015 compared to that in 2011 (cohort effect).⁴ However, these smoking-related estimates were generated from either single cross-sectional data or comparing the estimates across cross-sectional data. This limits time trend analysis as data are collected at a particular time point, and analysis was performed without taking into account the effect of other time components (age and cohort effect). As such, trend analysis using the age-period-cohort (APC) model is most suited to study the independent effect of age on smoking behavior while taking into account the influences of period and cohort. To date, there has been no APC study of smoking in Malaysia.

Therefore, we combined data from four nationally representative health surveys, the NHMS in 1996, 2006, 2011, and 2015, among current smokers aged 18 to 80 years, and used the Bell’s Hierarchical Age-Period-Cohort (HAPC)⁵ model to disentangle the APC trend on daily cigarette consumption among Malaysian adults aged 18 to 80 years. In addition, since previous studies had demonstrated that males, Malays, and other aborigines^3,4 had higher smoking prevalence, we had also stratified the APC analysis by gender and ethnicity to better elucidate the smoking trends among these subpopulations.

Methods

Study Sample

The present study is a repeated cross-sectional study where a total of 94 537 non-institutionalized respondents aged 18–80 years that were selected via a two-stage stratified sampling method were pooled from four cross-sectional NHMSs conducted in 1996 (n = 22 631), 2006 (n = 34,184), 2011 (n = 18 017), and 2015 (n = 19 705). Respondents aged household members who resided in the selected living quarters, the secondary sampling units (Enumeration Blocks were the primary sampling units), were asked to complete the smoking module during face-to-face interviews. These NHMSs had an overall response rate ranging from 93.0% to 96.9%.^6–9

Measure

The primary outcome of the present study is daily cigarette consumption among current smokers. Current smoker is defined as those who smoked in the past 30 days prior to the interview (in NHMS 1996, 2006, and 2011) or those who smoked daily or less than daily (NHMS 2015). Respondents who reported themselves as current smokers were asked the next question, “On the average, how many cigarettes do you smoke in a day?” in NHMS 1996; “In the last one month, what was the average number of cigarette you smoked in a day?” in NHMS 2006; “Usually, how many cigarettes, cigars, or many times do you smoke pipes, shisha, etc. in a day?” in NHMS 2011; “On average, how many cigarettes, kretek, cigarillos, pipe, bidis, etc do you currently smoke each day?” in NHMS 2015. The inclusion of cigars, pipes, shisha, kretek, cigarillos, bidis, and other tobacco products in more recent surveys would not make a difference in the trend of smoking intensity since cigarette is still the most common tobacco product in Malaysia.¹⁰ We examined the daily cigarette consumption trend among current smokers instead of smoking prevalence. The rationale for this is that different working definitions for current smoking were used in the NHMSs, whilst a similar question for the average number of cigarettes smoked per day was used in all four NHMSs. The use of a standardized question across NHMSs ensured the generation of reliable smoking trends.

Statistical Analysis

Univariate analysis of current smokers was reported as weighted prevalence among the general populations, whilst daily cigarette consumption and other covariates were reported as mean ± SD or weighted prevalence among the current smokers. All univariate analyses were stratified by survey periods.

To decompose the APC effects, we adopted Bell’s HAPC model,⁵ a cross-classified random effects model extended from Yang et al.¹¹ for repeated cross-sectional data. The HAPC model which allows the incorporation of random variations (or random effect) in the model, in addition to the usual fixed effects in ordinary regression, can address the intra-period and intra-cohort correlation issues by specifying individual covariates such as age within each period-by-cohort group at level-1 of the model, while the crossed (non-nested) random variations for period and cohort were specified at level-2 and 3. This specification gives rise to a three-level model (multilevel model). For the present study, because of the sampling nature of these NHMSs, an additional geographical identifier (a combination of state and urban or rural locality) was added to the model, thus extending it to a four-level cross-classified model.

Despite this conceptually appealing APC framework, the well-known non-identifiability of APC or the inherent mathematical dependency between APC (cohort = period − age) made the simultaneous inclusion of APC into the ordinary regression model, or even the fixed part of a multilevel model impossible since there are an infinite number of APC coefficient sets with equal fits for the data.^12,13 As such, to disentangle the APC, one has to impose constraints on either one of these three. Two different conceptualizations of cohort effect arise to deal with this identification problem: Sociologic and epidemiologic. The sociologic definition of cohort effect posits that cohort-specific exogenous factors may uniquely shape people’s health behaviors and outcomes across their life course. Whilst the epidemiologic definition of cohort effect conceptualizes a cohort effect as a period effect (eg, populational-level environmental causes) that are differentially experienced by people of different age (ie, age as an effect modifier).¹⁴ The former can be translated into statistical modelings, such as constrained-based regression and HAPC, and the latter into models that focus on the second-order effect produced by the interaction between period and age effects.

We adopted the sociologic cohort effect definition for the present study and are contented to assume no linear period trend because of few justifications. First, we believe that people born in a similar time range (birth cohorts) shared similar historical, social, and environmental experiences in their formative years, which may bring about distinctive health behaviors across life trajectories compared to their younger or older counterparts. Second, there are rarely continuous period-specific ubiquitous factors,⁵ such as the coronavirus disease 2019 pandemic, in a population that can lead to a shift in behaviors and attitudes across all ages and cohorts. Therefore, having made these assumptions (and assuming these assumptions are valid), we included only the effects of age and cohort but excluded periods in the fixed part of the HAPC model, whilst the variations between periods were captured by the random period residuals in the random part. In order to complement the APC trends of daily cigarette consumption among current smokers, the period trend was plotted in Figure 1A but was not interpreted in further detail since our model explicitly assumed no linear period trend.

Figure 1.

Adjusted association between age, period, birth cohort, and number of cigarettes smoked per day among current smokers. From left to right, top to bottom, the graph reports the age, cohort, and period trend of the number of cigarettes smoked/day in the general population (A); age trend of the number of cigarettes smoked/day by birth cohorts (B); by gender (C); and ethnicity (D); and cohort trend by gender (E); and ethnicity (F).

Respondent’s age was self-reported and counter checked by enumerators using the date of birth documented in the respondent’s identity card and the interview date. Period denotes the year during which the NHMS survey was conducted. Year of birth (birth cohort) was generated by subtracting the survey year (period) by age. Age and birth cohort were mean-centered at 40 and 1960, respectively, to reduce correlations between the linear and higher-order terms. Covariates such as age, cohort, gender, ethnicities, and the interaction terms between gender and age, gender and cohort, ethnicity and age, ethnicity and cohort, and higher-order terms for age and cohort were added, one at a time, into the fixed part (level-1) of the model based on the significance of the likelihood ratio tests. These interaction terms were included to assess the gender and ethnic differences in age and cohort trends of daily cigarette consumption. In addition, 5-year cohort groups, survey years, and state-by-locality were added, one at a time, into the random part (levels 2, 3, and 4) of the model to account for random variations. The significance of the random effects was determined via likelihood ratio tests. Besides, due to the heterogeneous mix of races and nationalities of the “Others” category under ethnicity, though they were included in all analyses, their results were not presented in the APC graphs since they were not sufficiently represented in the NHMSs to obtain robust estimates. Natural log-transformed values of mean daily cigarette use were used in the APC analysis as the data were skewed. The final model is as specified below:

$${\displaystyle \begin{array}{ll}{\displaystyle lnCi{g}_{i(jkm)}}& ={\beta }_o+{\beta }_{1}Ag{e}_{i(jkm)}+{\beta }_{2}Ag{e}_{i(jkm)}^2\\ & +{\beta }_{3}Ag{e}_{i(jkm)}^3+{\beta }_{4}Cohor{t}_{i(jkm)}+{\beta }_{5}Cohor{t}_{i(jkm)}^2\\ & +{\beta }_{6}Gende{r}_{i(jkm)}+{\beta }_{7}Ethnicit{y}_{i(jkm)}\\ & +{\beta }_8{(EthnicxAge)}_{i(jkm)}+{\beta }_9{(EhtnicityxAg{e}^2)}_{i\left(jkm\right)}\\ & +{\beta }_{10}{(EthnicityxAg{e}^3)}_{i\left(jkm\right)}+{\beta }_{11}{(EhtnicityxCohort)}_{(i(jkm)}\\ & +{\beta }_{12}{(EhtnicityxCohor{t}^2)}_{i(jkm)}+{\mu }_{1j}^{}+{\mu }_{2k}^{}+{\mu }_{3m}^{}+e_{i(jkm)}\end{array}}$$

, where In$Ci{g}_{i(jkm)}$ is the natural log-transformed values of mean daily cigarette use of individual i that is born in cohort j, being surveyed in period k, and resided in state-by-locality m, with fixed coefficients ${\beta }_1$to${\beta }_{12}$_,${\beta }_o$is a non-varying constant, ${\mu }_{1j}^{}$is the random variation of the 5-year-interval cohort group j, ${\mu }_{2k}^{}$is the random variation of period k,${\mu }_{3im}^{}$is the random variation for state-by-locality m, and $e_{i(jkm)}$is the residual error term due to variations not explained by the model. This model was fitted using the mixed program in STATA version 14 (StataCorp., College Station, TX, USA). Model estimates were presented in Appendix I.

Sensitivity Analysis

Sensitivity analysis was performed based on an alternative assumption of no linear cohort effects (age-period model). The resulting age, period, and cohort trends (refer to Appendix II) were compared with those from the abovementioned model (age-cohort model). Parameter estimates for the age-period model were presented in Appendix III.

Results

The prevalence of current smoking has been plateauing from 1996 to 2015, except that a decrease was observed in 2006. The median daily cigarette consumption has increased from 12.0 in 1996 to 16.0 in 2015. Both genders, all ethnic groups (except the Chinese), age groups, and birth cohort groups (except those born in 1940–1949) showed an increase in median daily cigarette use (Table 1).

Table 1.

Prevalence of Current Smoking and Distribution of Smoking Intensity by Sociodemography Among Current Smokers Aged 18––80 Years in Malaysia, NHMS 1996, 2006, 2011, and 2015.

	NHMS
	1996		2006		2011		2015		Total		n
Current smoker	%	95% CI	%	95% CI	%	95% CI	%	95% CI	%	95% CI
Overall	25.4	[24.6 to 26.1]	23.5	[23.0 to 24.0]	25.1	[24.1 to 26.1]	24.1	[23.1 to 25.2]	24.4	[23.9 to 24.9]	22,160
Daily cigarette use	median	IQR	median	IQR	median	IQR	median	IQR	median	IQR	n
Overall	12.0	13.0	10.0	10.0	10.0	14.0	16.0	22.0	10.0	14.0	20,732
Sociodemography
Gender
Female	5.5	9.0	5.0	7.0	5.0	4.0	11.5	22.0	5.0	7.0	1101
Male	14.0	12.0	10.0	9.0	10.0	14.0	16.0	22.0	10.0	13.0	19555
Ethnicity
Malay	12.0	13.0	10.0	9.0	10.0	14.0	16.0	22.0	10.0	13.0	12128
Chinese	20.0	10.0	10.0	13.0	10.0	14.0	17.0	20.5	15.0	11.0	3198
Indian	10.0	15.0	10.0	10.0	8.0	16.0	13.5	15.0	10.0	15.0	1038
Other Bumiputra	10.0	15.0	10.0	10.0	10.0	11.0	16.0	21.0	10.0	15.0	2678
Others
Age (10-year interval)
18–29	7.0	17.0	10.0	9.0	10.0	10.0	14.0	24.5	10.0	10.0	4678
30–39	12.0	12.0	10.0	9.0	10.0	13.0	15.5	22.0	12.0	13.0	5662
40–49	14.0	10.0	12.0	13.0	10.0	13.0	20.0	20.0	14.0	12.0	4528
50–59	12.0	14.0	10.0	14.0	10.0	13.0	20.0	20.0	12.0	13.0	3260
60–69	10.0	14.0	10.0	10.0	10.0	14.0	15.0	17.5	10.0	15.0	1838
70–79	8.0	11.0	8.0	6.0	7.0	10.0	12.0	15.0	10.0	11.0	742
80 & above	6.0	15.5	11.5	20.0	10.0	17.0	15.0	20.0	8.0	16.5	24
Year of birth (10-year interval)
1929 and earlier	10.0	16.0	9.0	7.0	0.0	—	0.0	—	10.0	16.0	465
1930–1939	10.0	14.0	8.5	8.0	10.0	11.0	12.0	15.0	10.0	15.0	1261
1940–1949	14.0	13.0	10.0	11.0	10.0	15.0	10.0	14.0	10.0	14.0	2291
1950–1959	14.0	10.0	12.0	13.0	10.0	13.0	20.0	22.0	12.0	12.0	3794
1960–1969	12.0	13.0	12.0	13.0	10.0	13.0	20.0	20.0	12.0	13.0	4585
1970–1979	0.0	—	10.0	9.0	10.0	13.0	17.0	20.0	10.0	13.0	3481
1980–1989	0.0	—	10.0	9.0	10.0	10.0	15.0	22.0	10.0	12.0	3860
1990–1997	0.0	—	—	—	8.0	7.0	14.0	23.0	10.0	15.0	995
Locality
Rural	10.0	14.0	10.0	10.0	10.0	14.0	19.0	21.0	10.0	14.0	10089
Urban	14.0	12.0	10.0	10.0	10.0	14.0	15.0	23.0	11.0	14.0	10643

IQR = interquartile range, n = study sample of current smokers, NHMS = National Health and Morbidity Survey.

% - weighted prevalence.

Figure 1A shows that daily cigarette consumption among current smokers increases as age increases from 18 to 60 years old, after which a drop in cigarette use was observed. There was a gradual increase in daily cigarette consumption over birth cohorts.

Figure 1B further examined the age effect by eight 10-year birth cohort groups. A more profound increasing trend for cigarette use was observed among the three most recent cohorts (ie, 1990–1997, 1980–1989, and 1970–1979), whilst the older cohorts have a decreasing age trend. In addition, more recent birth cohorts reported higher daily cigarette consumption than those of the same age but from earlier cohorts.

Figure 1C shows that the age trend of daily cigarette consumption did not vary by gender (p > .05). Despite the insignificant gender modification effect, males had a higher level of cigarette use than females across life trajectories. In contrast, the age trajectory of daily cigarette consumption differed by ethnicity (Figure 1D, p < .05). A decrease in daily cigarette use at age 60 was only observed among the Chinese and Indian current smokers.

When age was held constant at 40 years old, no gender difference in cohort trend was observed (Figure 1E, p > .05). In contrast, the cohort trend differed significantly by ethnicity (Figure 1F, p < .001). An increasing cohort trend was observed among the Malays and other aborigines, whilst the mean daily cigarette consumption remained at about 12 to 13 cigarettes per day among the Chinese across birth cohorts.

Sensitivity analysis revealed that the age-period model (that assumed zero cohort trend) has similar age, period, and cohort trends (Appendix II) compared to those observed in the model of interest (that assumed zero period trend), except that it has an earlier decreased age trend and a less profound increased cohort trend.

Discussion

Using an APC approach, the present study pooled four nationally representative health surveys to examine the 20-year trend of gender- and ethnic-specific daily cigarette consumption among Malaysian adults. Generally, smoking intensity among Malaysian current smokers increased with age (except a drop in smoking intensity was observed during elderhood) and cohort recency. However, the present findings must be cautiously interpreted as we explicitly assumed the absence of a linear period trend, and this assumption may not be applicable to other studies.

APC studies on smoking prevalence are primarily conducted among adults in developed countries^15–19 and some among adolescents,^20–22 with one exception of APC study among the Asian adult population.²³ Compared to those observed among the Western nations, where smoking prevalence peaked at a younger age of around 20,^17,19 the mean daily cigarette consumption among the Malaysian population peaked at age 60, after which a decreasing age trend was observed. Differences in socioeconomic and development status, tobacco control policies (GBD 2015 Tobacco Collaborators, 2017), and the use of smoking prevalence instead of smoking intensity as the dependent outcome can be plausible reasons for the heterogeneous findings between countries.

The increasing smoking intensity (daily cigarette consumption) over life trajectories (until age 60 years) among current smokers can be attributable to increased nicotine addiction because of early smoking initiation. This explanation is corroborated by previous findings, where nearly 80% of Malaysian adolescents who had ever smoked tried their first cigarette before the age of 14, and three in five of them had nicotine dependence.²⁴ Furthermore, individuals who smoked at a younger age were more likely to smoke more cigarettes per day,²⁵ continue their smoking behavior into adulthood,²⁶ and were less likely to cease smoking due to nicotine addiction.²⁷ All of these factors can contribute to the increased smoking intensity as young smokers age into adulthood. However, further research is needed to elucidate the underlying factors for the age-related increase in daily cigarette consumption among current smokers. On the other hand, the decreasing age trend after year 60 can be due to a survival bias since current smokers with greater daily use of cigarettes are more likely to suffer from smoking-related diseases and may die prematurely. On the contrary, current smokers that survive into later life are more likely to reduce daily cigarette consumption due to increased awareness of smoking-related health hazards.

The positive cohort effect implies that smoking intensity increased with cohort recency. In addition, by further stratifying the age trend by birth cohorts, we were able to demonstrate that, comparing individuals of the same age, for instance, a 40-year-old adult that was born in 1970 had higher daily cigarette consumption than the 40-year-old counterparts that were born in 1960 and 1950. These findings were in line with those reported in previous studies. In cross-sectional studies, a higher smoking prevalence was observed among young Malaysian adults⁴ and Taiwanese adults born between 1980 and 1986.²⁸ Using a two-factor APC model, a positive cohort effect on smoking initiation was observed among Europeans born after 1980.¹⁸ Few reasons are postulated for the observed cohort effect. First, substantial growth of employment and occupational shift from agricultural to manufacturing due to the rapidly growing economy after the postwar era had led to a successively more affluent environment in this country.²⁹ As a result, those that were born in more recent cohorts may have increased affordability for cigarettes than the preceding cohorts, thus causing a rise in smoking intensity in more recent cohorts. Second, people in modern life are encountering higher levels of stress than their counterparts in the preceding cohorts,^30,31 and this may lead to increased smoking intensity as a maladaptive stress-coping strategy. A local study among 343 school-going boys revealed that stress was the most common reason for continuing smoking.³² Third, social media influence and the growing number of cigarette retailers in this modern era could have increased the ease of cigarette access, thus causing an increase in cigarette consumption among the young generations. Despite the prohibition of cigarette sales to minors under 18 years, more than half of the young adolescents reported that they were able to purchase cigarettes from commercial stores such as grocery stores, supermarkets, and roadside stalls.^24,33,34 Fourth, although Malaysia has banned all forms of direct tobacco advertisements since 1982, tobacco companies were the top advertisers during the 1980s and 1990s via indirect advertisement and trademark diversification to circumvent the ban.³⁵ This can, in turn, lead to earlier smoking initiation and subsequently increased smoking intensity since numerous longitudinal studies have demonstrated the strong association between tobacco advertising and adolescent smoking,³⁶ and those who smoke at a younger age tend to smoke more because of nicotine addiction.²⁷

As the tobacco industry is now targeting the previously untapped market of adolescents, if no effective tobacco control initiatives were in place, the observed cohort effect is expected to increase even more among the coming generations, Gen Z and Gen Alpha. Furthermore, the late Gen-Xers (born between 1970–1979) and Millennials (born between 1980–1989 and 1990–1997) are now the parents of Generation Z and Generation Alpha, and a local study had showed that adolescents were more likely to adopt smoking behavior of their parents.³³ The recent proposal to ban smoking and the possession of tobacco products, including e-cigarettes, for people born after 2005 by the Minister of Health Malaysia is hoped as a “generational end game” for smoking in Malaysia.³⁷ If the legislation is passed, the mid-Gen Zers and Gen Alpha-ers will never be able to buy cigarettes in this country ever again.

The present study did not find significant gender differences in the age and cohort trend for daily cigarette consumption. This finding is corroborated by previous studies among regular smokers, where women had been reported to have similar levels of plasma nicotine and cotinine compared to men, although they smoked fewer cigarettes per day on average than men.³⁸ This may imply that nicotine addiction did not vary by gender and therefore lend support to the present findings of similar age and cohort trends of daily cigarette consumption among male and female current smokers. Nonetheless, studies conducted elsewhere reported otherwise for smoking prevalence, where a less profound decrease^17,39 or an increase²³ across age and birth cohorts³⁹ was observed among females than males. Apart from the use of different outcome measures (smoking prevalence vs. daily cigarette consumption) and APC methods, variations in social norms for female smoking⁴⁰ and socioeconomic developments across countries could have caused inter-country heterogeneities. The underreporting of cigarette use among Malaysian women because of prevailing sanctions against female smoking in the comparatively conservative Malaysian society could have potentially obscured the actual cigarette use trend over age trajectories and birth cohorts and caused the insignificant gender modification effect on time trends of smoking intensity.

It is not surprising to observe an ethnic modification effect on the age and cohort trend of daily cigarette consumption in the present study as Malaysia is a multiethnic and multicultural country, and therefore health perception and behavior may vary. The decreasing overall age trend at age 60 is most likely due to the decrease in consumption among the Chinese and Indian elderly current smokers, a trend that was not seen among the Malays and other aborigines. On the other hand, the increasing overall cohort trend among Malaysian current smokers is most plausibly because of the increasing cohort trends among the Malays and other aborigines, since cigarette use among the Chinese remained almost constant at 12 to 13 cigarettes per day, whilst the Indians had the lowest mean cigarette use across birth cohorts and showed a decreasing trend in more recent cohorts. These findings suggest the need to further strengthen the National Strategic Plan on Tobacco Control 2015–2020 among older adults and young generations of Malay or other aborigine descent. In addition, the plateauing trend of cigarette use among the Chinese across birth cohorts also warrants attention as it may imply a widespread insensitivity to tobacco control strategies across all birth cohorts in this subpopulation.

Several social, cultural, environmental, genetic, and individual factors can contribute to the ethnic differences in daily cigarette consumption over age and cohort trajectories. Over the decades, many studies have concluded the important role of genetics in smoking behavior,⁴¹ nicotine metabolism,⁴² and dependency.⁴³ Studies have reported that single nucleotide polymorphisms of nicotine acetylcholine receptors (ie, Cholinergic Receptor Nicotinic Alpha 4 Subunit [CHRNA4] gene) were associated with nicotine dependence in different populations.^44,45 These ethnic-specific associations between CHRNA4 gene variants and nicotine dependence may suggest the presence of ethnogenetic heterogeneity in nicotine dependence and hence the ethnic-specific age and cohort trends of daily cigarette consumption in the present study. Furthermore, the possible generational effect on smoking could explain the positive associations between smoking intensity and cohort recency among the Malays and other aborigines. Since the Malay and other aborigine adults were more likely to smoke than the Chinese,⁴ their progenies are therefore more likely to expose to secondhand smoke and thus have increased smoking intensity. This explanation is substantiated by previous findings, where the likelihood of secondhand smoke exposure was significantly higher among adolescents with at least one smoking parent compared to their counterparts with nonsmoking parents,⁴⁶ and secondhand smoke exposure was significantly associated with increased susceptibility to nicotine addiction among smokers.⁴⁷ However, properly designed research to identify the underlying predictors for the increased cigarette consumption among the older and younger generation Malays and other aborigines and for the plateauing trend among the Chinese current smokers is greatly necessitated.

Conclusion

The present study highlighted important ethnic-specific trends for mean daily cigarette consumption among the current smoker population in Malaysia. Targeted interventional strategies or implementation of national tobacco control policies should be focused on the older adults and young generations of Malay or other aborigine descents in order to reduce daily cigarette use and thus curb the increasing burden of smoking-related diseases and the associated healthcare cost. The main strength of the study is the use of four nationally representative health surveys and APC analysis to disentangle the effects of age, period, and cohort from the overall trend of daily cigarette consumption among current smokers in the Malaysian population. The high overall response rate and standardized survey methodology in all NHMSs ensure the representativeness, comparability, and validity of the results. Besides, although we assumed no linear period trend, we recognized that discrete period effects because of the occurrence of specific events in a given time period (such as the recent coronavirus disease 2019 pandemic) might exist. Therefore, it was implicitly taken into account in the HAPC model as a random effect, and this ensured validity of the predicted age and cohort trends. In addition, gender- and ethnic-stratified APC analyses also helped to inform policy makers about targeted tobacco controls. Nonetheless, the present study is not immune to limitations like any other study. First, cigarette use was self-reported, and there may be an underestimation of female smoking and reporting bias of daily cigarette use such as heaping of number of cigarettes smoked. Second, pack-years—a better measure of smoking intensity over a period of time was not used in the present study and should be investigated in future APC studies. Third, the present findings should be cautiously interpreted as we assumed the absence of period trend in the extended HAPC model to deal with the APC identification problem. Sensitivity analyses revealed an earlier decrease and a less profound increase in smoking intensity across age and cohorts, respectively when we assumed a zero cohort trend instead of period. Therefore, researchers that aim to investigate the APC trends need to understand that there is no standard APC model to disentangle the APC effect, and the model that we specified here may not be applicable to other studies/data since the model specification and assumptions are subjected to the research question, subject area, and expert opinion on the contextual effect of APC.

Supplementary Material

A Contributorship Form detailing each author’s specific involvement with this content, as well as any supplementary data, are available online at https://academic.oup.com/ntr.

Funding

The NHMSs were supported by the Ministry of Health (MOH) Research Fund, Malaysia (P42-251-170000-00500033, NMRR-10-757-6837, and NMRR-14-1064-21877).

Ethical Approval

The present study was registered under the National Medical Research Registry (NMRR-18-3790-44039), and ethical approval was granted by the Medical Research and Ethics Committee (MREC).

Declaration of Interest

The authors declared no conflict of interests or competing interests.

Data Availability

Analysis codes and a dataset sample were deposited in GitHub and accessible at https://github.com/chienhueymoh/hapc-smoking.git and https://github.com/chienhueymoh/hapc-smoking-data.git, respectively. However, the complete dataset is only available upon request and is subject to approval by the Director-General of Health Malaysia.