Age-period-cohort Analysis of Change Trend in Drinking Behaviors among Adults in the Republic of Korea

Abstract

One of the major functions of public health should be monitoring levels of alcohol consumption and the changes over time. However, little attention has been paid to the changing trends in alcohol consumption in the social context. The study aim was to describe alcohol consumption levels for the period 1998–2021 in terms of the monthly prevalence of alcohol use (MPA) and the prevalence of high-risk alcohol use (PHA). Additionally, change trends were broken down by age, period, and cohort effect. The study used data from Korea’s National Health and Nutrition Examination Survey from 1998 to 2021. The data showed a differential pattern of change in both MPA and PHA between gender, with men tending to exhibit a decrease or remaining at a plateau. However, women exhibited a discernible increase in both MPA and PHA. A convergence theory of drinking behavior between gender has been confirmed among Korean adults. Age-period-cohort models identify the effects of age, with both MPA and PHA reaching the highest rates at age 45 years and then declining among both men and women. During the period 2009–2011 and around 2015, the relative risk (RR) of the likelihood of MPA and PHA was higher than the RR in any other year studied, which indicates a period effect. In addition, birth cohort effects were observed in MPA and PHA among adults, with more discernible patterns in women. Policy implications were discussed with the aim of reducing harm caused by alcohol use, with a high policy priority placed on the reduction of alcohol use among women.

Key messages

① What is known previously?

Previous monitoring of alcohol consumption among adults revealed a difference in the change trend between men and women. While levels of alcohol consumption are generally declined prevalence of high-risk alcohol use (PHA), particularly among men, tended to decrease or remain at a plateau over decades. However, PHA among women consistently increased. The observed changes in woman drinking behavior can be attributed to age, period, and cohort factors.

② What new information is presented?

Study findings indicated that changes in the prevalence of alcohol use in the previous month had the same trend pattern as PHA. Women who were born in the recent birth cohort were more likely to consume alcohol than those born in the older birth cohort.

③ What are implications?

High policy priority should be given to the prevention of alcohol use by women. Policy measures should include increasing alcohol health literacy and developing effective prevention policy options at the population level.

Introduction

The essential functions that government agencies responsible for a country’s public health should undertake are assessment, policy development, and assurance. Among these, assessment is the most crucial function to implement first [1]. It involves systematically collecting information on the population’s health status, health behaviors, and determinants, as well as analyzing and providing various kinds of information by integrating different datasets. Assurance entails monitoring how developed policies and programs are implemented and their effectiveness. In this context, the National Health and Nutrition Examination Survey (NHANES) can be regarded as performing the roles of assessment and assurance. To fulfill these functions, it is necessary to collect data and engage in monitoring activities such as analyzing the collected information and identifying the effectiveness and significance of policies or programs. However, compared to activities that collect information, there is a relative lack of activities related to analyzing the collected information and deriving policy meanings. In this regard, deriving meaning through in-depth analysis of the data collected through the NHANES is a core public health activity.

Alcohol is known to damage various organs and the mental health of the human body and is responsible for more than 200 diseases and unhealthy conditions [2]. Therefore, in the Republic of Korea (ROK), where not only the proportion of drinkers but also the rate of harmful drinking is high, alcohol-related harm is inevitable. In 2022, 78.9% of adults in the ROK reported having consumed alcohol at least once in the past year, with 14.2% categorized as high-risk drinkers (21.3% for men and 7.0% for women). In 2022, there were 5,033 deaths attributed to alcohol-induced diseases, which can be considered typical alcohol-related harm [3]. However, it should be noted that this statistic only included deaths directly attributable to diseases caused by alcohol consumption; if indirect causes were included, the number of deaths would likely increase. According to the data estimated by the Institute for Health Metrics and Evaluation in the United States, the number of alcohol-attributable deaths in the ROK was 13,200 in 2016 [4]. The institute also reported that alcohol consumption was the second factor contributing to a decrease in healthy life expectancy for Koreans, resulting in a reduction of 11.1 months [5]. In addition, alcohol consumption as a single factor was estimated to contribute to 4.8% of alcohol-attributable cancers reported [6]. Such health risks are not exclusive to drinkers alone. As seen in fetal alcohol syndrome, drunk driving accidents, and workplace safety incidents, drinkers can harm the health of others around them. As a result, alcohol consumption is a major cause of health insurance financial losses (more significant than smoking or obesity) [7]. It can also lead to social problems such as crime, domestic violence, and loss of corporate productivity [8].

Alcohol-related harm is not an inevitable social problem experienced only by some alcoholics; it is a health and social issue that can be addressed [9]. There is ample scientific evidence that alcohol-caused damage can be prevented with the implementation of appropriate alcohol-related policies [8]. Nonetheless, national efforts and societal responses to reduce the prevention and reduction of alcohol harm in the ROK are quite lackluster [10]. One reason for this is the lack of information about harms caused by alcohol use. A lack of accurate and detailed information based on monitoring and analysis of the size and type of harmful effects related to drinking behavior has led to a low understanding of the magnitude and severity of the problem. As a result, policy priority and social acceptance of alcohol harm prevention projects are low [11].

This study aimed to conduct an in-depth analysis of the information on drinking behavior from the NHANES, which has been collected for nearly a quarter of a century. We aimed to identify the factors contributing to the long-term changes in harmful drinking behaviors and derive policy implications by examining the long-term trends in monthly alcohol consumption rates and high-risk drinking rates among drinkers annually. We further sought to describe these trends through age-period-cohort (APC) analysis.

Methods

1. Study Subject

This study analyzed the secondary data from the Korea NHANES (KNHANES). The study used raw data from 1998 to 2021. The final analysis included a total of 79,024 drinkers who responded that they had consumed alcohol at least once in the past year among adult respondents aged 19 and above.

2. Variable Selection

At the national level, one of the critical elements that should be included when monitoring alcohol consumption is to figure out alcohol consumption [12]. As one of ways to investigate alcohol consumption, surveys can measure the prevalence of drinking and drinking patterns. The criteria for defining a drinker vary from country to country, but in the ROK, a drinker is defined as a person who consumes more than one drink in a month [13]. Therefore, the monthly drinking rate was included in the analysis of this study, defined as the proportion of people who drank alcohol at least once a month in the past year and calculated using the drinking frequency item related to the section of the Health Behavior of the KNHANES. This included the proportion of drinkers who responded to the question “How often do you drink alcohol?” with “about once a month,” “2–4 times a month,” “2–3 times a week,” or “4 or more times a week.”

The harm caused by alcohol consumption is determined by the total volume of alcohol consumed, which poses a complex estimation challenge due to considering the type of alcoholic beverages, their alcohol content, and the units of consumption [12]. An easy way to measure only the frequency and the average amount of units of alcohol consumed on a typical drinking occasion is to measure high-risk drinking [14]. In the ROK, high-risk drinking is defined as consuming an average of more than seven drinks per occasion for men and more than five drinks per occasion for women in the past year and doing so on two or more occasions per week. This variable is derived from the amount and frequency of drinking among the items related to the health behavior of the KNHANES. The question regarding the amount of alcohol consumption is “How much alcohol do you typically consume on one occasion in the past year?” The response options are “1–2 drinks,” “3–4 drinks,” “5–6 drinks,” “7–9 drinks,” and ten or more drinks.

3. Data Analysis

The monthly drinking rate and high-risk drinking were analyzed using a complex sampling design that considers the weighting of the KNHANES. Trend analysis was also conducted to test the significance of the trend changes in these two variables. Since drinking behavior is a social behavior [15] changing trends of it over time must be understood with respect to social contextual factors [16]. One way to do this is to perform an APC analysis [17-19]. Longitudinally repeated cross-sectional surveys, such as the KNHANES data, can be considered one of the representative cases where APC analysis can be applied [17]. Through this analysis, it is possible to understand whether the changes in monthly alcohol consumption rates and high-risk drinking rates are attributable to the characteristics of specific cohorts, various environmental changes related to alcohol consumption, or age effects reflecting biological maturation or social processes.

In the APC analysis of this study, a logarithmic linear model was applied, assuming that the number of monthly and high-risk drinking behaviors followed the Poisson distribution. As an APC analysis assumes that “cohort=period–age,” such an analysis based on multiple regression models has an identification problem in that it is not linear and independent. To overcome this problem, non-linear models and constrained generalized linear models can be proposed alternatively, and in this study, a constrained generalized linear model was used to solve the problem of de-identification, which constrains the mean value and slope of the period effect to zero.

In this study, APC analysis was performed based on gender and weighted frequency by 10-year age unit of the KNHANES data. Accordingly, for the APC analysis, 19–29 years old was treated as 25 years old, 30–39 years old as 35 years old, 40–49 years old as 45 years old, 50–59 years old as 55 years old, 60–69 years old as 65 years old, and 70 years and older as 75 years old. The reference period was set as 1998 and the reference cohort as 1923. The age effect was calculated as the prevalence (%) and the period effect and cohort effect as the relative risk (RR). The statistical analysis used the “apc,fit.” function of the R Epi package. Through a goodness-of-fit comparison between the APC submodels, the model with the lowest Akaike information criterion (AIC) value was selected as the final model.

4. Research Ethics

This study was conducted after receiving approval for exemption by the Institutional Review Board of Inje University (INJE 2023-06-024).

Results

1. Trends in Monthly Drinking Rates and High-risk Drinking Rates by Year

1) Trends in monthly drinking rates among annual drinkers

The monthly drinking rate, which is the fraction of annual drinkers who regularly drink at least once a month, is shown in Table 1 and Figure 1. Overall, it fluctuated within the range of 74–92%, but except for 1998, there was no significant difference in the pattern of change, and the fluctuation was not statistically significant (p>0.1); thus, it can be considered that the distribution is approximately stable at around 75%. When analyzed by gender, the trend differed. Men showed a slight decline overall but no statistical significance (p>0.1). However, in the case of women, there was a fluctuating pattern. After excluding the years 1998 (85.5%) and 2001 (35.2%), which were considered to be outliers, there was an overall increasing trend, which was statistically significant (p=0.022).¹⁾ However, the trend lines, including measurements from 1998 and 2001, were not statistically significant (p>0.1). The gender ratio (man/woman) of monthly drinking rate increased slightly from 1.1 in 1998 to 1.3 in 2021, but showed a clear tendency to decrease except for outlier measures (p<0.05).

Table 1. Trends in the prevalence of alcohol use among adults who had alcohol in the past 12 months, 1998 to 2021.

	1998	2001	2005	2007	2008	2009	2010	2011	2012	2013	2014	2015	2016	2017	2018	2019	2020	2021	p for trend
Monthly prevalencea)
Total	92.4	56.7	69.1	75.9	77.5	76.8	76.1	76.1	74.8	77.5	76.4	76.9	77.7	77.0	75.6	76.6	74.8	74.2	0.924
Man	97.1	75.0	83.8	86.3	87.3	87.9	88.3	88.1	85.8	87.4	86.4	86.9	86.5	85.9	82.0	84.9	82.6	82.2	0.274
Woman	85.5	35.2	50.9	61.6	64.4	62.1	60.5	60.8	60.9	64.7	64.2	64.4	66.4	65.9	67.4	65.8	64.7	63.5	0.471
Gender ratioc)	1.1	2.1	1.7	1.4	1.4	1.4	1.5	1.5	1.4	1.4	1.4	1.4	1.3	1.3	1.2	1.3	1.3	1.3	0.046
High-risk prevalenceb)
Total	19.1	19.4	14.9	16.1	19.8	17.1	17.2	17.3	17.5	15.7	17.0	16.6	17.0	17.3	18.0	15.8	17.7	17.1	0.013
Man	26.8	31.0	23.2	24.5	28.4	24.6	24.9	26.0	25.1	22.5	23.7	23.9	24.1	24.1	24.0	21.5	25.3	23.6	0.006
Woman	8.0	6.1	4.6	5.0	8.4	7.4	7.4	6.4	7.9	7.2	8.8	7.7	8.1	9.1	10.5	8.6	8.1	8.9	0.008
Gender ratioc)	3.4	5.0	5.0	4.9	3.4	3.3	3.4	4.1	3.2	3.1	2.7	3.1	3.0	2.7	2.3	2.5	3.1	2.7	0.001

Unit: %. ^a)Percentage of adults who had alcohol in previous month. ^b)Percentage of adults who had 7+ drinks for men or 5+ for women on a typical occasion at least twice per week. ^c)Gender ratio=man/woman.

Figure 1. Trends in the prevalence of alcohol use among adults who had alcohol in the past 12 months, 1998 to 2021.

(A) Monthly prevalence. (B) High-risk prevalence.

2) Trends in high-risk drinking rates among annual drinkers

The high-risk drinking rate among annual drinkers, that is, the proportion of individuals engaging in high-risk drinking behavior during the analysis period, fluctuated within the range of 14–19% for both gender combined, as shown in Table 1 and Figure 1, and was statistically significant (p<0.05). When examining these changes by gender, different patterns emerged. Except for 1998 and 2001, when the monthly drinking rate was an outlier, for men, there was a fluctuation in the range of 21–29%, showing a decreasing trend overall (p<0.05). For women, despite slight fluctuations (from 8.0% in 1998 to 8.9% in 2021), there was a continuously increasing trend, as depicted in Figure 1, with the trend line also being statistically significant (p<0.05). The gender ratio for high-risk drinking rates steadily decreased from 3.4 in 1998 to 2.7 in 2021 (p<0.05).

2. APC Analysis

1) Results of the APC analysis of monthly drinking rates among annual drinkers

Since 1998, APC analysis was conducted separately by gender to distinguish whether the change in monthly drinking rate during the analysis period was due to the age, cohort, the period effect caused by the trend of the times, or environmental changes. The model fit statistics for the APC analysis (Table 2) showed that the model fit, including all three components, was the best. The AIC and residual deviance were the smallest, and both were statistically significant. This indicates that the predictions of the model considering all three components (age, period, cohort) are the closest to the observed values compared to the predictions of other intrinsic models. Therefore, it can be interpreted that the model, including age, period, cohorts as independent variables, best explains the trend of changes in monthly drinking rates compared to other intrinsic models. This pattern was the same for both gender.

Table 2. Goodness of fit of the APC model .

Model	Man				Woman
	AIC	Deviance (df)	p-value		AIC	Deviance (df)	p-value
Model for monthly prevalence
Age	436,711.8	434,984 (103)			2,354,313.7	2,352,657 (104)
Age-drift	436,460.8	434,731 (102)	<0.0001		1,412,946.0	1,411,288 (103)	<0.0001
Age-cohort	359,988.4	358,253 (99)	<0.0001		1,302,095.5	1,300,431 (100)	<0.0001
Age-period-cohort	160,649.4	158,908 (96)	<0.0001		648,602.1	646,932 (97)	<0.0001
Age-period	210,309.3	208,573 (99)	<0.0001		695,389.0	693,725 (100)	<0.0001
Model for high-risk prevalence
Age	1,318,919.8	1,317,339 (103)			1,300,178.1	1,298,797 (104)
Age-drift	1,221,222.8	1,219,640 (102)	<0.0001		927,302.9	925,920 (103)	<0.0001
Age-cohort	1,128,631.3	1,127,043 (99)	<0.0001		900,233.3	898,844 (100)	<0.0001
Age-period-cohort	1,082,677.7	1,081,083 (96)	<0.0001		852,915.2	851,520 (97)	<0.0001
Age-period	1,188,467.5	1,186,879 (99)	<0.0001		873,118.3	871,729 (100)	<0.0001

APC=age-period-cohort; AIC=Akaike information criterion; df=degree of freedom.

As observed in Figure 2, the age effect on monthly drinking rates shows relatively more significant fluctuations among women compared to men. For men, the monthly drinking rate, which was 63.7% at age 25, slightly increased to 65.5% at age 35, 66.8% at 45, 66.9% at 55, 67.2% at 65, and 67.6% at 75. This indicates a slight increase with age up to 45 years, beyond which there was little to no change. For women, the monthly drinking rate showed a more significant variation than men, with percentages as follows: 16.6% at age 25, 18.7% at 35, 21.9% at 45, 23.9% at 55, 24.6% at 65, and 25.2% at 75. There was an increasing trend in the monthly drinking rate among women of different ages. Regarding the period effect, men and women showed a continuous increase in the RR of monthly drinking compared to the reference year of 1998 until peaking in 2010 (man RR=1.20, woman RR=1.58). After that, there was a consistent decrease in the risk of monthly drinking, with the RR for men at 1.12 and women at 1.28 in 2021, which is still higher compared to the reference year 1998. In the cohort effect, men and women showed different patterns. The RR of monthly drinking among men remained within the range of not exceeding 1.14 for cohorts born after 1923. However, for the cohort born in 1991, the RR was 1.09; for the cohort born in 1996, it was 1.06. By contrast, for women, the RR of monthly drinking compared to the cohort born in 1923 consistently increased for the cohorts born after that year. In the 1969 cohort, the RR of monthly drinking compared to the 1923 cohort was 2.0, which increased to 3.0 for the 1990 cohort and continued to rise. By the 1996 cohort, the RR of monthly drinking was 3.2 times higher than that of the 1923 cohort. It was observed that in women, the more recent the cohort, the more linearly the monthly drinking rate increased.

Figure 2. Age-period-cohort effect in the prevalence of alcohol use among adults who had alcohol in the past 12 months, 1998 to 2021.

(A) Monthly prevalence. (B) High-risk prevalence.

2) Results of the APC analysis of high-risk drinking rates among annual drinkers

Table 2 presents statistics of the model fit tests for the APC analysis of high-risk drinking rates. In the model that included age, period, and cohort, the AIC and residual deviation values were the lowest, resulting in the best fit. This was the same for both men and women (p<0.001).

APC analysis on the high-risk drinking rates among annual drinkers showed significant differences in age, period, and cohort effects (Table 2, Figure 2). However, the differences in high-risk drinking risk by period were smaller than those by age and cohort. In terms of age effects, both men and women showed a continuous increase up to the age of 45 (men 30.3%, women 2.8%), followed by a sustained decrease, with men exhibiting greater variation by age compared to women. Regarding the period effect, men showed a relatively stable distribution with little change throughout the observation period, while women showed a pattern of fluctuation with repeated increases and decreases. Specifically, for men, the RR decreased from 0.96 in 2001 to 0.88 in 2009 and remained the same until 2017, after which it increased slightly to 0.89 in 2018, 0.91 in 2019, and 0.94 in 2021. By contrast, women started at 1.03 in 2001 compared to 1998 and gradually increased their RR of high-risk drinking until it reached 1.13 in 2009. After maintaining this level for three years (2009–2011), the RR decreased to 1.11 in 2015, and then followed a gradual increase: 1.14 in 2016, 1.18 in 2017, and 1.20 in 2018. However, it decreased again to 1.18 in 2019, continuing to decline in the subsequent years (2019, 1.11; 2020, 1.04; 2021, 0.97). In terms of cohort effects, for men, the risk of high-risk drinking tended to remain stable or slightly increase with increasing birth years, but began to decrease from those born in the 1960s (RR=1.25). From 1985 onwards, it was consistently lower (RR=0.99) compared to the reference cohort born in 1923, showing a continuous decline. Conversely, for women, the trend showed that the risk of high-risk drinking increased with more recent birth cohorts. Compared to those born in 1923, the risk of high-risk drinking doubled, tripled, quadrupled, and quintupled for those born in 1958, 1974, 1981, and 1987, respectively. This indicates that there are gender-specific patterns in drinking behavior between the Baby Boomer generation and their children’s generation.

Discussion

This study aimed to decompose the long-term trends in drinking behavior into age, period, or birth cohort effects. To achieve this, APC analysis was conducted using data from the KNHANES from 1998 to 2021. When studying the trend changes observed in alcohol consumption behaviors from the NHANES, trend analysis tended to carry out with age adjustment. This is because it is a relative indicator compared to standard year population and not representing actual measurements of drinking behaviors. This means there’s potential for distortion in measuring alcohol consumption rates across age groups [17]. One of the ways to overcome this is APC analysis, which can consider both cohort and period effects. APC analysis allows us to understand the contextual (historical, social, cultural, and political) influences on trends in health behaviors such as alcohol consumption. In particular, cohort effects arise from two main factors: early-life exposure and ongoing accumulation of exposure. This perspective enables us to elucidate contextual exposure factors spanning different life stages [20].

In APC analysis, the age effect refers to the accumulation of exposures or physiological changes that can occur with increasing biological age and the effects of events experienced during different life stages (e.g., education, marriage). By contrast, the period effect refers to specific events or environmental changes that simultaneously affect all members of a population group, regardless of age. Events such as the COVID 19 pandemic or the International Monetary Fund crisis in 1997 are representative examples. The cohort effect can be interpreted as a reflection of experiences formed by the intersection of an individual’s life history and macro-socioeconomic influences. As seen in “the 58ers” or “the fateful class of '94,” cohort groups born in specific years can be interpreted as experiencing accumulated experiences at key stages of life cycles that may impact their health status. In this context, the cohort effect in APC analysis can be seen as a social structural aspect that determines trends and fluctuations in health levels and behaviors [17]. Unlike cohort studies that track groups exposed to specific factors, in this study, the term “cohort” refers to birth cohorts.

In the APC analysis of alcohol consumption behavior in this study, the observed age effect exhibited a similar pattern to the general trajectory of alcohol consumption: it increases up to a certain age, stabilizes or decreases after middle age, and then decreases as a typical developmental process of drinking behavior [21]. As individuals progress through the developmental stages of life, the proportion of high-risk drinking tends to steadily increase until the age of 45 for both men and women, followed by a decrease. Similarly, in terms of monthly alcohol consumption rates, men’s consumption increased until the age of 45, with little change after that. This age effect may be influenced by biological maturity or social processes (e.g., marriage, employment, acquisition of parental status). Health-risk behaviors, such as high-risk drinking, generally decline after approximately 18–25 years of age through social processes such as employment, marriage, or parenthood [22]. However, in the case of Korean adults, they decline only after age 45. This is believed to be related to the social value of drinking. The social roles acquired throughout the life development process (e.g., employment or parental roles) influence to restraint in drinking behavior. However, the manifestation patterns of this restraint can vary depending on the societal values associated with alcohol consumption [21]. In the ROK, where the belief that drinking together can facilitate social interaction predominates, the instrumental value of alcohol holds a strong influence [11]. Therefore, it decreases only after the age of 45. It is considered that at earlier ages, the likelihood of alcohol consumption or high-risk drinking is higher depending on the level of social and economic activities. However, after that stage, it is believed that individuals may refrain from high-risk drinking due to increased concerns about their health and the burden of chronic disease risks. The decline in drinking behavior at age 45 in the ROK is similar to that of Organisation for Economic Co-operation and Development (OECD) countries. According to an OECD report, weekly drinking rates in these countries were the highest among people aged 45–59 [23].

However, for women, the monthly drinking rate continues to increase even after age 45. This needs to be explained in relation to the characteristics of the monthly drinking rate indicator. Unlike high-risk drinking rates, the monthly drinking rate reflects whether a person drinks at least once a month, which is different from high-risk drinking. In societies where women were traditionally not permitted to consume alcohol, there has been a permissive shift in the societal values and judgments regarding woman drinking. This change is believed to have spread within the population from the younger to older age groups. While the age effect can be explained by the interaction of period and cohort effects, the increase in monthly drinking rates among women as they age can be attributed to the societal trend of increased acceptance of woman drinking, which spreads from the latest birth cohorts to earlier ones. When examining the variation in age effects, men remained stable at around 67% after age 45, while women showed a slight increase of approximately 23–25%. Although the monthly drinking rate for women increases with age, the monthly drinking rate is lower in absolute terms compared to men, and the rate of increase is not significant. This can be seen as reflecting the continued social constraints on women’s drinking [21].

Period effects can also be observed in monthly and high-risk drinking rates, but the pattern is slightly different. Regarding monthly drinking rates, the RR was higher for both men and women in 2009–2011, after which it declined for a while, then increased around 2015, and tended to decrease after 2018. The RR for high-risk drinking in women during this period was also higher than that in other periods. The fluctuations in monthly drinking rates were larger for women than for men, and for men, high-risk drinking remained almost unchanged during this period, which may be related to the socio-economic changes that our society has experienced [16]. The increase in imported alcoholic beverages due to free trade agreements (especially the spread of low-alcohol beverages such as wine or beer, which women prefer); introduction of new alcoholic products (e.g., low alcohol by volume [ABV] spirit or fruit-flavored alcohol beverages targeting women); and impacts of aggressive marketing strategies for these products may also be contributing factors. Wine imports increased in around 2010, and during this period, also the consumption of lower-proof sprits of Soju with below 20% ABV became more popular and its aggressive marketing to women was expanded in its size [24]. In addition, in 2015, a new product called “fruit-flavored Soju” was launched. Moreover, the proliferation of alcohol portrayal in mass media that glorify drinking has led to a significant change in woman drinking [25]. The so-called “drinking room,” a talk show during which participants are encouraged to drink alcoholic beverages to intoxication, first aired in 2016 on the general programming cable TV channel, and the number of such shows has increased significantly since then.

The magnitude of the period effect on monthly drinking rates and high-risk drinking rates differs between men and women. The RR of monthly drinking rates, which reflects only the frequency (or presence) of drinking, is larger for women than for men. However, when considering both frequency and quantity, as in the case of high-risk drinking rates, the RR for men is lower than 1 (approximately 0.9). There was little change in this RR during the observation period. However, it is clear that changes in alcohol production, distribution, and marketing affect men and women differently [21]. This has policy implications as improvements can be made in this regard.

The cohort effect of monthly drinking rates differs between gender, tending to decline slightly for men as they enter the recent birth cohort, while for women, it tends to increase. The birth cohort effect can be observed in the high-risk drinking rate. For both gender, the latest birth cohort shows significant differences in the trend of change in high-risk drinking compared to past birth cohorts. Men have a slight decrease in trend from the 1980s, while women exhibit a linearly increasing trend with the most recent birth cohorts. Men show a decrease in high-risk drinking rates starting from the 1960 birth cohort, and the rates decrease even more from the 1985 cohort onwards compared to the reference cohort born in 1923. Contrarily, women exhibit a more than two-fold increase compared to the 1923 cohort for those born in 1958, followed by a three-fold increase from the 1974 cohort, a four-fold increase from the 1981 cohort, and more than a five-fold increase from the 1987 cohort. This indicates that changes in drinking behavior among the Baby Boomer generation and their children differ between gender. It is clear from the cohort effect that while high-risk drinking rates are decreasing among men, they are increasing among women. The increased high-risk drinking of women in recent birth cohorts has also been observed in studies in Germany [26], Sweden [27], and other countries [21]. If this trend continues, the high-risk drinking rate of men and women will reach a similar level, confirming the possibility of a convergence hypothesis of drinking behavior between men and women [28]. The trend changes in the gender ratio of high-risk drinking rates support the convergence hypothesis between men and women. Similarly, in the trend changes of the gender ratio of monthly drinking rates, excluding outliers (1998 and 2001), there is a decreasing trend, which supports the convergence hypothesis. If the analysis period is extended and monitored further, this trend could be clarified and confirmed. The changes observed in women’s monthly drinking rates and high-risk drinking rates are believed to be associated with changes in the availability of alcoholic beverages [16], along with changes in women’s human rights and roles [29] and expansion of social participation [30], and delayed marriages and childbirth [21]. Since differential influence of these social and cultural changes on each birth cohort [21], rates of increase in drinking behaviors has accelerated in recent birth cohorts.

Although this study represents a significant attempt to distinguish whether long-term changes in drinking behavior are due to age, cohort, or period effects, it has several limitations. First, there is a possibility of systematic errors due to fluctuations in response rates or changes in survey methods regarding drinking behavior questions words over the years as it is a survey that has been repeatedly conducted. Considering that the NHANES relies on self-reporting, there is a possibility that the estimates of cohort effects may have been distorted due to variations in whether individuals honestly report their drinking behaviors or differences in understanding the survey questions, which may vary by age group [16]. In the case of questionnaires on high-risk drinking, it would be difficult to say that older cohort and the recent cohort were equally honest in reporting drinking behavior. Second, the constrained generalized linear model adopted in this study has statistical limitations. Specifically, the estimates may be biased due to the assumption of equality constraints, where the model assumes that age, period, or cohort effects are homogeneous [20]. Therefore, to ensure the robustness of the APC estimates in this study, it is necessary to base them on relevant theories or accumulated empirical research. However, due to the lack of such evidence in research conducted in the ROK, caution is needed when interpreting the estimates.

Nonetheless, this study is significant because it is the first APC analysis that considers both monthly and high-risk drinking rates based on data from nationally representative survey, KNHANES. Given that APC analysis is one of the most crucial analytical tools for monitoring non-communicable disease risk factors such as alcohol consumption [21], research should continue as part of national disease surveillance efforts. The challenge lies in exploring how the observed changes in alcohol consumption behavior regarding age, cohort, and period effects can be reflected in non-communicable disease prevention and public health policies. Policies that can respond to changes in drinking behavior, especially the increase in women’s drinking, should be more highly prioritized. In the long term, the observed decrease in the gender gap in monthly drinking and high-risk drinking rates suggests the so-called “convergence hypothesis,” which posits that the differences in drinking will eventually disappear. The main reason for this trend is the increase in woman drinking behavior. In this context, it is necessary to prioritize the prevention and reduction of alcohol-related harm nationwide to align policies with the needs of women. Efforts should be expanded to improve alcohol health literacy—which is knowledge about the dangers of alcohol use and the capacity to understand and respond to the essence of strategies employed by the alcohol industry—to address the alcohol industry’s marketing strategies aimed at normalizing or routinizing alcohol consumption among women. Additionally, there is a need to emphasize the production and dissemination of information highlighting women’s relative vulnerability to alcohol-related harm. At the same time, the effectiveness of policy alternatives such as restricting alcohol availability (e.g., designating a public place as alcohol-free zone), limiting alcohol advertising or marketing, and reducing the portrayal of drinking scenes in the mass media has been verified to impact both men and women. Therefore, there should be active efforts to expand the implementation of these measures and raise alcohol taxes. In cases where raising alcohol taxes is challenging, amending laws to allocate tax revenues for alcohol harm prevention and reduction could be an alternative solution.

Funding Statement

Funding Source: This work was supported by Korea Disease Control and Prevention Agency (grant number 2023-11-002).