Polarization of alcohol consumption or coding changes as the cause of the increase in 100% alcohol-attributable deaths in Lithuania during the COVID-19 pandemic

INTRODUCTION

During the coronavirus disease 2019 (COVID-19) pandemic, many countries of the European Union (EU) experienced increases in fully (i.e., 100%) alcohol-attributable mortality despite decreases in population consumption levels (i.e., alcohol per capita consumption [1, 2]). The increases in Lithuania, similar to the other Baltic countries, were among the highest in the EU (change between 2019 and 2021 in Lithuania: +25%; for all three Baltic states: +46%; [2]). As fully alcohol-attributable causes of death are highly related to chronic and/or heavy episodic drinking [3], and as overall the population drinking level decreased, polarization was discussed as likely being the main underlying mechanism [4] (for an early conceptualization at the beginning of COVID-19: see [5]). Polarization of drinking has been defined as an increase in consumption among people with heavy drinking (e.g., the top 10% or 20% in drinking level in a population) while the rest of the drinking population decreases their level of consumption [4, 6]).
When the data were presented to Lithuanian stakeholders, an alternative explanation was suggested. In July 2019, shortly before the COVID-19 outbreak, new coding rules for causes of death were established [7], specifically to reduce the use of code K74.6 (unspecified or other liver cirrhosis, particularly for cryptogenic deaths in Lithuania), which had often been used to avoid “alcoholic” liver disease codes (i.e., the code used for 100% alcohol-attributable deaths (K70); the original term in the ICD-10 is being replaced by “alcohol-related” more and more to avoid stigmatization) [8]. This use of the unspecified code for liver cirrhosis (K74.6) has been widespread in Lithuania and elsewhere to avoid stigmatization of the affected person, including by insurance companies [8-10]. The new coding guideline also emphasized not only the immediate cause of death but also long-term health problems that led to the beginning of the pathological process ending in death.
If indeed the code K70 and/or its subcodes replaced the K74.6 code, this would lead to an increase in fully alcohol-attributable diseases, as was observed in Lithua­nia and elsewhere in the EU. The code K70 makes up about half of the fully alcohol-attributable causes of death (51% for years 2015 and later for the EU countries; 49% for Lithuania, both proportions for people 15 years and older [11]). This study was designed to test this alternative hypothesis for the increase in fully alcohol-attributable deaths in Lithuania during the pandemic.

MATERIAL AND METHODS

DATA

Monthly data for cause-of-death by ICD, sex, and 5-year interval age group (0-4, 5-9, 10-14 years of age, etc.) were obtained from the Lithuanian Institute for Hygiene from 2015 to 2022. Raw death numbers were converted to age-standardized rates (deaths per 100,000 people) using the European standard [12].

VARIABLES IN THE MODEL

The main dependent variable was the quotient of the raw number of K70 cases by K74.6. The higher the value of this fraction, the larger the proportion of K70 codes compared to K74.6. To test the impact of the coding change, a dummy variable was introduced to represent the time in months (the main independent variable) before and after July 2019 (0 before; 0.5 for July 2019, and then increases of 0.1 per month until it reached 1, where it remained for all months thereafter). This gradual increase was used after consultation with experts, who indicated that the full implementation of coding changes would take some months to complete. If the hypothesis of a coding change is correct, this dummy variable would need to be significant. In addition to this change, we tested any slope changes for the variable, and we conducted a sensitivity analysis with the number of individuals diagnosed. All analyses were done separately by sex.
Other control variables were the effects of alcohol control policy implementation, which included: 1) the 2017-03-01 policy that increased excise tax by 23% for ethyl alcohol, by 92-94% for intermediate products, and by 111-112% for wines and beer; and 2) the 2018-01-01 policy that reduced availability by increasing the legal mini­mum age to 20 years, as well as an almost comprehensive alcohol advertising ban. As in prior models (for a summary, see [13]), the policy variable was coded as 1 for one year following a policy implementation, and as 0 for the rest.

SENSITIVITY ANALYSES

Two sensitivity analyses were conducted. First, we conducted a sensitivity analysis with the quotient of the age-standardized K70/K74.6 rate, which adjusts for potential impact of age-specific coding. Since two national lockdowns occurred during the study period, the first from March 16, 2020 to June 16, 2020, and the second from November 7, 2020 to June 30, 2021, the second sensitivity analysis included the dummy variable for the COVID-19 lockdown as a control variable.

STATISTICAL MODELS

Generalized additive mixed models (GAMMs) [14] were used to investigate the effect of the code change on the outcomes, adjusted for by the alcohol control poli­cies. The outcomes, sex-specific mortality ratios, were first examined for skewness and were log-transformed when necessary. Seasonal trends were accounted for by the inclusion of smooth functions of time in months, while autocorrelation was accounted for by the inclusion of autoregressive and moving-average terms after examining the residuals. The normality and stationarity of the residuals was examined using Q-Q plots, as well as the residuals against the fitted values.
All analyses were done separately by sex using R version 4.0.2 (15).

RESULTS

Log transformations were used to reduce skewness of the dependent variable. The GAMM results (Tables 1 and 2; for diagnostic indicators see Supplementary Figures A1 and A2) show the coding of K70 decreased in relation to the unspecified code K74.6 for both males and females, but these coding changes were not significant for either sex. These potential decreases may narrow over time, suggested by the positive slope term seen in the models. But, again, the slope changes were not statistically significant. In addition, both models had very poor model fitting, suggested by the R² statistics (–0.004 for females and 0.071 for males). Therefore, we could not exclude chance as the cause of this change.
As for the control variables, there was a slight change downwards for the outcome variable over time, indicating that the ratio decreased, and that K74.6 was, in fact, increasingly used over time. The policies had no effect on coding preferences.
The sensitivity analysis on the age-standardized K70/K74.6 rate (Figure A3) showed decreases in raw numbers for both sexes, but further confirmed that the code change had no significant impacts. For the second sensitivity analysis, which controlled for the COVID-19 lockdown, all likelihood ratio tests showed that models that controlled for the lockdown were not statistically different from those that did not control for it (results not shown).

DISCUSSION AND CONCLUSIONS

Overall, we found no evidence that a coding change resulted in higher proportions of coding for 100% alcohol-attributable causes of liver cirrhosis deaths. The coding practices in Lithuania did not change, and the unspecified code (K74.6) is still used widely. We can only speculate that this coding may be used to avoid stigmatization of the decedents and their families, and may also avoid difficulties in receiving payouts from private life insurance policies [16, 17].
The increase in 100% alcohol-attributable causes of death in Lithuania [1, 2] thus cannot be explained by this coding change, and the original hypothesis of polarization of alcohol consumption seems the most likely explanation [4, 5], especially since survey results now support this explanation [18]. However, the impact of heavy drinking on alcohol-related liver cirrhosis [19, 20] may have been exacerbated by the lack of medical services available for treatment of alcohol use disorders and/or liver cirrhosis during the COVID-19 pandemic [21, 22].
What are the implications of this finding? First, it seems necessary to monitor not only level of alcohol consumption in general, but heavy drinking, in particu­lar. This seems especially important in times of crises (e.g., economic crises, pandemics; see [23]), when popu­lations do not seem to move up and down in concert, but rather show a polarization in their drinking levels. Second, proven alcohol policy measures against heavy drinking and 100% alcohol-attributable outcomes appear to be key during such tumultuous times: specifi­cally measures directed at reducing alcohol availability [24] and increasing the availability of screening and brief interventions [25].

ACKNOWLEEDGEMENT

Funding: National Institute on Alcohol Abuse and Alcoho­lism of the National Institutes of Health, grant number R01AA028224.

DISCLOSURE

The authors report no conflict of interest.

References