In the previous section, we examined some unsupervised and supervised machine learning methods to determine the top predictors of alcohol consumption habit changes among healthcare workers in the United States. Some of the questions from the survey (predictors) have been selected by the most accurate supervised models. Chi-squared test, mutual information support, and K-modes clustering support some results obtained by supervised methods. In this section, we discuss the top predictors, and we explore some non-machine-learning based articles whose results agree with our analysis.
One of the most important questions, or features, that appears in all unsupervised learning methods and supervised learning algorithms is Question 11 that reads, “Are children home from school in the house?” It raises a question about the relationship between COVID-19-associated school closure and alcohol consumption habit changes among healthcare workers. There is a strong relationship between COVID-related school closure, parenting stress, and alcohol consumption that requires further research of the topic. Based on our findings, the school closure associated with the COVID-19 pandemic can be associated with an increase in alcohol consumption as a coping mechanism. To alleviate the hidden effects of COVID-19, e.g., parenting stress, on healthcare workers, some operational strategies might be used in schools to reduce the spread of COVID-19 and maintain safe operations in school during the COVID-19 pandemic. Boschuetz et al. [54] discuss how alcohol use patterns differ from the effects of social distancing caused by Covid-19. They use the Alcohol Use Disorders Identification Test (AUDIT-C) to measure alcohol use. A higher score on the test indicates higher levels of drinking in patients and that drinking could lead to harm within the patient.
A Chi-squared test was used to compare the original results from before social distancing to results obtained after social distancing was implemented. Additional tests such as the paired t-test and McNemar test were used to compare the different variables. They conclude that having children at home was deemed statistically significant as values for the AUDIT-C test were on average 0.47 for having children at home and 0.1 for not having children at home. The p-value for having children at home was 0.01, deeming the presence of children at home statistically significant. In a multivariable analysis, the presence of children at home remained statistically significant.
Another interesting point brought forth by this paper was the difference in AUDIT-C scores between men and women. The researchers in this study believe the difference in values of women and men with children at home could be due to the notion that women are usually responsible for mainly raising the children. Homeschooling the children could also be a factor increasing the stress in women over men as the study notes that oftentimes women are tasked to homeschool their children while working from home. The difference between males and females had a p value of 0.03, making the data statistically significant. This could be why in the study Question 11 is one of the most influential questions in determining alcohol consumption. Acuff et al. [55] compare the results of various empirical studies conducted between the Summer and Fall of 2020 that dealt with drinking practices at the start of the pandemic. A subset of the studies considered other factors such as sex, age, and children at home in determining the results of the study. The number of articles included in the final meta-analysis was 128 papers with data on the drinking effects that were caused by the pandemic. The study concluded that each additional person that was in a household was associated with a greater increase in drinking. Parents living with children displayed a higher increase in drinking as well, which displays how living with more people in the house can lead to higher levels of stress.
Both the Chi-Squared test and Mutual-Information method, as well as some supervised learning algorithms imply that Question 15: “Have you varied your work schedule?” is not independent of alcohol consumption habit changes among healthcare workers in the US. Our results support a study by the Ohio State University College of Nursing [56], which reveals that nurses working long shifts due to the COVID-19 pandemic have experienced negative effects. In another study by Cooper et al. [57], it is shown that work stressors lead to increased distress, which in turn promotes problematic alcohol use. They also reveal the relationship between work-related stress and an increase in alcohol consumption among healthcare professionals during the COVID-19 pandemic.
Ahn [58] examines the relationship between work hours and drinking habits and concludes shortening work hours increases the probability of participating in drinking but does not significantly increase daily drinking habits. Frone [59] investigates a pattern between job support and alcohol consumption. From the Cause-and-Effect Model, they have suggested that jobs low in complexity and control and high in demand are related to increased employee alcohol use. However, the results from the Cause-and-Effect Model are inconsistent, and that is why they used the Moderation Model. Then, they found a positive association between job demands and lack of a clearly defined role at the workplace (role ambiguity) and heavy drinking.
As the COVID-19 pandemic continues, healthcare workers are more in need of working shifts, and consequently, healthcare workers who work shifts may consume more alcohol than day workers. Therefore, using results from [59] and our ML-based results, we can conclude that work stress related to Covid-19 and a change in work schedule is positively associated with job dissatisfaction, which causes an increase in alcohol use among vulnerable people.
Question 28: “Has the amount of food you have been eating per day changed?” is selected by AdaBoost-8 and K-modes clustering as one the most important top predictors. This supports the findings in [60] and [61]. Huber et al. [60] surveyed 1980 students about their eating habits during Covid-19. Questions were asked about the setting where participants eat their food, the amount of food consumed, and the healthiness of the food. Demographic information such as age, height, and gender was recorded in the survey.
In the second part of the study, participants were asked about their food consumption along with smoking and alcohol consumption since the start of the pandemic. SPSS 25 was used to perform a statistical analysis of the data. Categorical variables were compared using Chi-squared tests to determine a relationship between data points, and non-categorical variables were compared using Kruskal-Wallis tests. Their results concluded that people’s eating habits had drastically changed during the pandemic. People were eating an increased number of homecooked meals and were eating at restaurants very rarely. Almost 50 percent of people reported eating in restaurants or cafeterias before the virus, but now less than 3% of people reported eating in these places. Half of the participants stated that the amount of food consumed had not changed during the lockdown, while 31% said their food intake had increased and 16% stated that their food intake had decreased. Using a bivariate analysis, the study concluded that alcohol consumption showed a significant correlation with the change in food consumed. The study concluded that changes in alcohol drinking patterns led to an increase in food intake in their sample population.
Yeomans et al. [61] discuss the relationship between alcohol use and food consumption in individuals. They noted that previous findings also determined that alcohol consumption prior to a meal leads to a short-term increase in food intake. They found that the short-term effects of alcohol have both direct and indirect effects on the amount of food consumed. According to their findings, food intake is increased in the period short after consuming alcohol due to the pharmacological and energetic effects alcohol induces on the body.
Question 1: “What is your age?” is selected as an important feature by AdaBoost-8 and XGBoost-4. Villanueva-Blasco et al. [62] showed that as age increases, the percentage of consumers of alcohol who increased their consumption during the pandemic is higher. In daily alcohol consumption the lower age ranges had a higher decrease in daily alcohol consumption than the older age ranges. They concluded that age is a large factor in alcohol consumption during the pandemic. The limitations of alcohol consumption by younger adults can be explained by the closure of venues due to the pandemic. There are less social drinking opportunities, and therefore younger adults have less reason to drink.
For older adults, alcohol consumption is more associated with their home and, therefore could be a reason for the increase in consumption. Härkönen et al. [63] utilized six Finnish Drinking Habits Surveys, which sampled a population between the ages of 15–69. All the surveys were conducted as interviews and asked questions to measure the frequency of light and binge drinking occasions. Then they used a binomial linear regression to model the distribution after the number of light and binge drinking occasions per year. They concluded that light drinking increased between the ages of 15–29 in both women and men. After the age of 29, light drinking in men continued to increase until about age 54 before plateauing until age 69. However, in women, light drinking seemed to have a slight decrease after the age of 29. Similar to light drinking, binge drinking increased within the male and female cohorts until around age 37. In males, binge drinking habits continued to increase until around age 61, while in females, these habits decreased after the age of 37. This will provide insight into why age is a pivotal factor in determining the amount of alcohol consumption, especially during the Covid-19 pandemic. Therefore, regarding the relationship between age and alcohol consumption changes, we may utilize our findings, which agree with the findings of [64], to make a conclusion: the susceptibility to stress differs between young and old people, and may lead to different stress responses e.g., turning to alcohol to cope with stressful situations.
Question 2: “What is your gender?” is selected as an important feature by AdaBoost-8 and XGB-4. Villanueva-Blasco et al. [65] discussed the patterns of alcohol consumption during the Covid-19 pandemic based upon gender in Spain. Their study used convivence sampling through online surveys that surveyed a group of 3779 participants. 70% of the participants were female and 30% of the participants were male that were surveyed. The average age of all participants was 37.76 years old.
Data collection began after confinement and continued until measures eased up. AUDIT-C was used to measure alcohol consumption in the study. Participants were asked about their drinking habits 6 months into confinement compared to their drinking habits before the onset of the pandemic. Different data analysis strategies were employed, such as the Chi-squared test and frequency analysis. Their results indicated that in both males in females, there was an increase in daily average consumption of 1–2 drinks per day but a decrease in daily consumption of 3–4 drinks per day. However, for consuming alcohol more than 4 days a week, the number of females during confinement doubled, whereas for males the number only increased by a factor of 1.5. For both males and females, the frequency and average daily consumption decreased at a similar rate; however, for females, the frequency of intensive consumption and the average SDUs or drinks per day decreased at a lower rate than that of males. This leads to a possible correlation between gender and alcohol consumption during the pandemic.
The researchers possibly attribute this discrepancy between gender to the fact that females drink more within the home and males in more social settings. This would lead to an increase in alcohol consumption in females compared to males. The main subgroup affected by the pandemic was heavy drinkers as both males and females started to consume 4 or more drinks per week. This can be attributed to buying alcohol as a weekly purchase and having the pleasure of drinking at home because most people had to work at home during this time. So, regarding the relationship between gender differences and alcohol consumption changes during the first wave of Covid-19 pandemic, we may make a conclusion, which agrees with the results of [66]: men and women tend to react differently with stress, both psychologically and biologically, and may lead to different stress responses.
Question 3, which is another important predictor, asks where the healthcare workers live (50 States, D.C., or Puerto Rico). Klein et al. [67] describe the regional differences in alcohol consumption. The study attempts to determine whether the region of residence influences alcohol consumption. Results indicated that geographic region has little determining effect on people’s alcohol use. However, the study did find a small impact between geographic region and drinking-related attitudes. Brenner et al. [68] found that there exists a considerable regional variation in the amount of alcohol consumed. Their results showed that there was a significant state of residence impact on alcohol use. They also found a correlation between rural versus urban residence. They concluded that living in a small town compared to a metropolitan area was associated with less frequent drinking. In addition, living in a rural area compared to an urban area was associated with more frequent alcohol consumption. Our results, besides results in [67] and [68], can justify the importance of where a healthcare worker lives on their alcohol use habit changes. During the first wave of Covid-19, most of the hospitals in populous states were mostly occupied by Covid-related patients. Hence, the healthcare workers in populous states experienced more work-related stresses which were associated with an increase in alcohol use consequences for some healthcare workers.
Question 12: “approximately how many hours did you sleep on an average work night in January 2020?” is selected as an important predictor. Imaki et al. [69] analyzed the relationship between the hours of sleep and lifestyle factors in Japanese factory workers. In this study, they compared lifestyle factors for factory workers who got 6 or fewer hours of sleep, 6.1–8.9 hours of sleep, and 9 or more hours of sleep, and they concluded that there is no significant difference between alcohol drinking between the groups who got 6 or fewer hours of sleep and 6.1–8.9 hours of sleep. The study notes, however, that the group receiving more than 9 hours of sleep was excluded from the analysis because they only made up 1% of the participants. Miller et al. [70] conducted a study that examined if adequate sleep is associated with drinking quantity and its consequences. In their study, College students reported drinks consumed per week, average sleep quality, and sleep adequacy. They found significant interactions between adequate sleep and weekly drinking quantity. In addition, this interaction was associated with alcohol-related consequences. Based on our ML results, as well as results in [69] and [70], we may conclude that the Covid-related issues caused an increase in alcohol use, and it subsequently led to changes in sleep patterns in healthcare workers.
Question 13: “How many hours do you sleep on an average night?” is another predictor that is selected as an important predictor. In their study, Du et al. [71] sampled students from various countries who were over the age of 18. The studies were conducted during times when the countries were under strict shelter-in-place orders. Demographic data such as age, sex, class status, height, and weight were all recorded in the studies. Perceived stress, alcohol consumption, and sleep duration were all recorded through questionnaires of the various students to determine relationships between these variables and the ongoing Covid-19 pandemic. Alcohol misuse was measured using the AUDIT-C questionnaire, and sleep was assessed using the Pittsburgh Sleep Quality Index. Questions were asked about how the Covid-19 pandemic affected various factors in their life including sleep quality and alcohol consumption. IBM SPSS version 26 was used to perform descriptive statistics, and statistical significance tests.
A total of 28 comparisons were used between the data in order to determine relationships between sleep quality and alcohol consumption during the pandemic. A total of 2254 students were analyzed for this study. 67% of the students were female and 58.7% of the students were living in the United States. Their results concluded that alcohol use was positively correlated with greater perceived stress and poor sleep quality. Poorer sleep quality meant a shorter duration of sleep which was increased with alcohol use. Improving sleep quality in these students would lead to a decrease in alcohol use which could explain why the question of how many hours one sleeps per night affected alcohol consumption. A decrease in sleep quality also leads to a positive correlation with stress, and the association between stress and alcohol misuse was statistically significant which could also impact the use of alcohol during the pandemic.
Question 23: “Has the amount of news you are consuming increased since the end of Feb, 2020?” is another important predictor that is selected by XGB-4. Chartier et al. [72] investigated the relationship between social media use and alcohol consumption during the first onset of the Covid-19 pandemic. Their data was taken from the Understanding American Study which is a group of 5,874 individuals who are 18 years or older that are meant to resemble the United States Population. The participants completed an original survey in March of 2020, and a follow-up survey one month later. The data were analyzed to determine any relationships between social media use and alcohol use frequency. Their analysis compared the social media use of participants in March 2020 with their alcohol use a month after. Hierarchical logistic regression was employed to examine how increased and decreased alcohol use altered with social media consumption. Their results concluded that social media use did not affect drinking habits during March 2020; however, alcohol use was higher for Wave 3 in individuals who had higher levels of social media use.
Alcohol use was also higher for individuals who used multiple social media platforms than those who only had a single social media account. They believe that individuals who had higher amount of social media activity would be exposed to more news content. This content especially surrounding the virus had been previously shown to lead to higher levels of stress in individuals which could account for the increase in alcohol use. Also, during the pandemic, many workers were forced to work from home and had more free time to scroll social media which could have sparked a major increase in social media use. This increase in social media activity would lead to greater consumption of news which could account for the reason so many individuals reported an increase in news consumption. Engels et al. [73] found that the portrayal of alcohol in the movie and commercials directly affected alcohol consumption during the movie.
On average, people who were presented with alcohol portrayal in the movies and in the commercials consumed around 1.5 more drinks than those presented with no alcohol portrayal. This could provide insight into how consuming the news affects alcohol use because when watching the news, viewers would be susceptible to alcohol advertising which could lead to increased use. Strainback et al. [74] also explore COVID-related news effects on mental health. Their findings show that COVID-19 media consumption leads to psychological distress, which may cause an increase in alcohol consumption. Thus, we can make a conjecture: the stress resulting from COVID-19 related news exposure may produce changes in drinking behavior. Question 28: “Has the amount of food you have been eating per day changed?” is another important feature that is selected by LightGBM and CatBoost. An increase in alcohol consumption can cause changes in a healthcare worker's diet. People who drink alcohol more frequently may choose less healthy food options that are high in fat and sugar [75]. Moreover, each gram of pure alcohol has 29 kilojoules of energy [76], and when it is mixed with sugary drinks, it contains even more calories. Since alcohol is absorbed directly in the bloodstream [77] that can lead to immediate changes in the amount of food that people eat.
Question 25: “Have you had more screen time around Bedtime?” is also selected by AdaBoost-8 and XGB-4. Tebar et al. [78] investigated how the increase in screen time that was caused by Covid-19 isolation policies influences the consumption and desire for alcohol in the U.S. population. In their study, a survey was conducted on 1897 adults with a mean age of 38. Participants were asked questions about their screen time habits during the Covid-19 pandemic, along with questions detailing their alcohol consumption and smoking habits. Demographic information detailing their gender, age, race, and working conditions were recorded in the survey as well. Overall, the survey was comprised of 70 questions to assess the various levels of screen time on different devices and their addictive behaviors. Binary logistic regression models were employed to discover the relationship between an increase in screen time and various addictive behaviors.
Their results showed that individuals who reported an increase in television time had an increased desire to drink during the pandemic. However, individuals who reported that computer time increased during the pandemic were less likely to report alcohol consumption. Increased cell phone time was associated with an increase in alcohol consumption during the pandemic as well. The study concluded that increases in cell phone and television consumption were factors influencing the increase in alcohol consumption and desire to consume alcohol. The study attributes these increases mainly to advertisements that contain alcohol being displayed on both television and cell phone. Overall, screen time was concluded to lead to an increase in addictive behaviors such as smoking and drinking.
Question 8: “Are you currently conducting your job mostly from home now?” is another important predictor. This correlates to having children at home because in [54] they found that when working from home parents were often tasked with raising and teaching their kids during the pandemic which leads to extra stress. Schmits et al. [79] used a sample of 2871 adults were asked to report on an online questionnaire. Participants were between the ages 18 and 85 and mainly lived in France and Belgium. As of the Covid Lockdown 36.5% of participants in the study were working from home. Demographic data was captured along with their living environment and how it changed during the lockdown. The frequency and quantity of alcohol use of the respondents was surveyed and assessed through an adapted version of the AUDIT-C questionnaire. An additional item was incorporated into the studies to evaluate how their alcohol use changed from before the lockdown. They were asked a variety of questions about who they consume alcohol with, how much alcohol they consume, and what factors lead to them consuming alcohol. Then, they used SPSS 26 software to perform various statistical tests such as Chi-squared, descriptive statistics, Kruskal-Wallis, and Spearman’s correlation tests. They finally concluded that individuals that were working from home tended to increase their alcohol use during lockdown by an average of 36.4%. The research also concluded that patients who were busier during lockdown increased their alcohol intake as well. The researchers attribute this increase in alcohol consumption at home to the absence of workplace regulations and the availability of the substance at home.
Children could also lead to increased stress among parents, where working at home would lead to increased time under stress from their children. Finally, the researchers hypothesize that the consumption increase could be in response to shifting from a work environment to a home environment. In a work environment, alcohol is less regulated, and therefore when one is working from home, they are around a higher-risk environment. In another research by Caluzzi et al. [80], it was found that people who lost their jobs or switched to working from home had increased their alcohol consumption. Their findings suggest that the normal obligations and restrictions that would limit one’s drinking had been reduced by working from home and their daily routine had been completely changed. This quick change can lead to increased stress and allows individuals to lose accountability for their actions. The subjects felt that while working from home the weekends and weekdays were blurred together, which could often lead to drinking habits during the week instead of only on the weekends. These are all reasons why working from home has increased the amount of alcohol consumed.
We found that healthcare workers whose children stayed home during the first wave in the US consumed more alcohol. We also found that the work schedule changes due to the Covid-19 pandemic led to a change in alcohol use habits. Changes in food consumption, age, gender, geographical characteristics, changes in sleep habits, the amount of news consumption, and screen time are also important predictors of an increase in alcohol use among healthcare workers in the United States.