key: cord-0291975-in0gc1y3
authors: Ye, J.; Ren, Z.
title: Using multivariate models to examine the impact of COVID-19 pandemic and gender differences on health and health care
date: 2021-09-05
journal: nan
DOI: 10.1101/2021.09.02.21263055
sha: c5b8b3d25be8211c3d2019570359adff03a2b31d
doc_id: 291975
cord_uid: in0gc1y3

Objective: To examine the effect of the COVID-19 pandemic, the effect of sex, and the joint effect of sex and the COVID-19 pandemic in relation to health communication, physical activity, mental health, and behavioral health. Methods: We drew data from the National Cancer Institute's 2020 Health Information National Trends Survey (HINTS). We described and compared the characteristics of social determinants of health, physical activity, mental health, alcohol use, patterns of social networking service use, and health information data sharing. Analyses were weighted to provide nationally representative estimates. Multivariate models (multiple linear regression, multiple logistic regression, and multinomial logistic model) were used to assess the sole and joint effect of sex and pandemic. In addition, we applied the Bonferroni correction to adjust p-values to decrease the risks of type I errors when making multiple statistical tests. Results: Women are more likely to use mobile health and health communication technologies. The effect of sex after the COVID-19 pandemic is significant on mental health, and women are more possible to have depression or anxiety disorders. The effect of sex is also significant before and after the pandemic regarding seeking health or medical information. Women have a smaller quantity and intensity of physical activity, which has a negative effect on health. Conclusion: Gender differences exist regardless of the COVID-19 pandemic and the pandemic amplifies the differences in some health and health care domains. Intersectional gender analyses are integral to addressing issues that arise and mitigating the exacerbation of inequities. Responses to the pandemic should consider diverse perspectives, including sex and gender.

Data used in this study were from the third round of data collection for HINTS 5 (Cycle 4), which began in February 2020 and concluded in June 2020. A binary variable indicating whether the participants returned their survey before or after the COVID-19 pandemic was provided. This variable facilitates the examination of responses before and after COVID-19 in the United States. The final HINTS 5, Cycle 4 (2020) sample consists of 3,865 respondents.

To understand the effect of the pandemic, the effect of sex, and the joint effect of sex and pandemic, we characterize the response of interest via a multivariate model. [13] Specifically, If is continuous, we assume it to be generated from the following linear model:

The Pandemic is a dichotomous variable indicating whether the response was obtained before or after the declaration of the pandemic; Gender is a dichotomous variable with 0 referring to the male and 1 the female; is a zero-mean normal random variable. Sociodemographic variables are incorporated in the model to account for possible confounding factors, where Age and Income are treated as continuous variables; BMI, Ethnicity, Education Marital Status are categorical variables. In model 1, is the difference between the mean of the response after and before the pandemic for the males (adjusting the demographic covariates); is the same difference but in the female population. represents the difference of the mean of the response between females and males (adjusting the demographic covariates) before the pandemic, and is the difference after the pandemic. The coefficient stands for the change of the gender difference before and after the pandemic, that is, . When the response is dichotomous (Y=0 is the reference), we build model 2, where we assume to be generated from a logistic regression model:

In model 2, the is the odds ratio (OR) of the male population after and before the pandemic adjusting the demographic covariates, and the is for the female population. The stands for the OR between females and males before the pandemic, and is the OR between females and males after the pandemic. The is the ratio of the gender odds ratio after and before the pandemic, that is, . When the response is a categorical variable with more than two classes, we consider a multinomial logistic regression model. [14] Supposing the categories of the are denoted by the set= , we set the first category as the baseline, for each and build model 3:

The interpretation of model 3 is the same as model 2 except that we are comparing category k to the reference category.

Statistical analyses were performed using R version 4.0.5 (R Foundation, Vienna, Austria). For continuous responses, ordinary least squares (OLS) estimators of the coefficients in the linear model are estimated with the lm function in R and reported with the p-values; for categorical responses, adjusted odds ratios (aORs) are estimated with the glm package in R, and reported with the p-values. The p-values for the sum of two coefficient estimators are obtained using the multcomp package in R, and all statistical testing was 2-tailed. In this study, 50 characteristics were categorized into four domains (health communication, physical activity, mental health, and behavioral health). Within each domain, there were five hypotheses to be tested: the effect of sex before the pandemic, the effect of sex after the pandemic, the effect of the pandemic in male population, the effect of the pandemic in female population, and the interacting effect of gender and pandemic.

Commonly, p=0.05 is used as the significance threshold for a single hypothesis testing. In this study, 50*5=250 hypotheses are tested simultaneously. Therefore, it is necessary to adjust for the multiplicity of the hypotheses, to avoid type I errors. [15] Assuming that all the 250 hypotheses are null, if we are still using the 0.05 threshold, then the probability of making at least a false discovery is the probability of at least one p-value exceeding 0.05. When the pvalues are mutually independent, the value is 1-(0.95) ^ 250 ≈ 1.00, which means we almost surely make at least one false discovery. Meanwhile, the expected number of discoveries is 250*0.05=12.5, and these discoveries are all, by definition, false discoveries. To address this issue, we need to adjust the p-value threshold based on the number of hypotheses being tested. In particular, we apply the Bonferroni correction [16] to adjust p-values to decrease the risks of type I errors, which proceeds as the following steps: when testing m hypotheses simultaneously, a p-value threshold of 0.05/m is utilized. With the adjusted threshold, the 0 denotes the number of null hypotheses, and the probability of making at least one false discovery can be bounded as:

The above derivation illustrates that the probability of making at least one false discovery is no more than 0.05, which is desirable. As a result, a p-value < 0.05/250=0.0002 is designated as statistically significant in our statistical analyses. We indicated the significant variables in the results section and all the p-values for Table 2 . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted September 5, 2021. Table 1 demonstrates the sociodemographic characteristics of the participants in this study. We present the weighted and unweighted prevalence of subcategories for each variable, and the weighted prevalence of the subcategories before and after the COVID-19 pandemic. There were more younger participants and white participants after the pandemic, but the other sociodemographics didn't show significant differences. ✝Other includes single (never been married), separated, living as married or living with a romantic partner. Table 2 presents the estimates of the multivariate model in terms of the variables corresponding to the health communication. For the characteristic made appointments with a health care provider online, the effect of sex before the pandemic is not significant but becomes significant after the pandemic, where the aOR between the female and male is e -0.36 =0.70 (the other aORs in the tables can be interpreted in the same way), which means the odds of making appointments with a health care provider online for males is 0.7 times that of females; the effect of the pandemic is not significant either before or after the pandemic, and the interacting effect of the pandemic and sex is not significant either. As for characteristic looked for health or . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted September 5, 2021. ; https://doi.org/10.1101/2021.09.02.21263055 doi: medRxiv preprint medical information for oneself, the effect of sex is significant before the pandemic, with an aOR=e -0.65 =0.52, which means the odds of looking for health or medical information for males is 0.52 times of the female; the effect of sex is also significant after the pandemic, with an aOR between the female and male population is e -0.55 = 0.58. Similarly, we see a significant effect of sex both before and after the pandemic in characteristics including looked up medical test results, shared health information on social networking sites, participated in an online forum or support group, visited social networking sites, used smartphone to track progress on a healthrelated goal and used an electronic wearable device to monitor or track health or activity. We also see a significant effect of sex after the pandemic with regard to the characteristic have "apps" related to health and wellness. We see gender differences in using mobile health and social networking services, and females are more likely to use these tools. Table 3 presents the model estimates for the characteristics corresponding to mental health. For Depression or Anxiety, the effect of sex is significant before and after the pandemic, with . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted September 5, 2021. ; estimated effects being -0.09 and -0.13, respectively. This means the difference of the levels of having depression or anxiety disorder in females and males are -0.09 and -0.31 before and after the pandemic respectively. For Feeling nervous, anxious, or on edge, the effect of sex is significant before and after the pandemic, with estimated coefficient -0.15 and -0. 19 respectively, suggesting that females are more likely to feel nervous or anxious. The effect of sex demonstrates a significant effect for not being able to stop or control worrying after the pandemic, with an expected difference -0.19. Sex also shows a significant effect for characteristics like When I feel threatened or anxious, I find myself thinking about my strengths before the pandemic, with an expected difference -0.17. The characteristic Most important value is categorical with more than two classes. As introduced in the Methods section, we set Making my own decisions as the reference, and compare other values to it. The effect of sex is significant after the pandemic for the characteristics being happy, having a deep connection to my religion and assuring my family is safe and secure (when compared with the reference), and the aORs are 3.53 and 1.93 respectively. Overall, females are more possible to have mental health issues Table 4 presents the model estimates for variables corresponding to behavioral health. In this table, Number of days drinking alcohol and Average drinks per day are treated as continuous variables, and Frequency of alcohol use per month is treated as a categorical variable. For the number of days drinking alcohol, the effect of sex is significant before and after the pandemic, with estimated coefficients -0.59 and -0.39, respectively. Similarly, there is a significant effect of sex for average drinks per day, before and after the pandemic. The . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted September 5, 2021. ;  characteristic Frequency of alcohol use per month is a categorical variable with 5 classes. We set Never as the reference and compare other classes to the reference, and there are no significant effects either from sex or the pandemic. We also examined the effects of sex and pandemic on physical activity. Figure 1 shows the three measurements of physical activities in the male and female population, before and after the pandemic. Figure 1 (b) compares the frequency of strength training per week, and we see that males tend to have a higher frequency of strength training regardless of the pandemic. Figure 1 . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted September 5, 2021. ;  (c) compares the sitting time per day, showing that males sit slightly longer than females regardless of the pandemic.

Gender disparities in health care could be defined as the dominance of one gender over the other in particular contexts, such as among primary social roles or occupations. Stereotypes appear when the less dominant genders are always underrepresented. [17] The biological, behavioral, social, and systemic factors underlying the differences in how women and men may experience COVID-19 and its consequences cannot be oversimplified. [18] Responses to the pandemic should consider diverse perspectives, including sex and gender. Individuals seek information for informed decision-making, and they consult a variety of information sources nowadays. Although we do not see a significant effect of the interaction term for all the characteristics, suggesting that the pandemic has not statistically affected the two genders differently. However, the results show that the pandemic amplifies some existing gender differences in health information seeking, communication, sharing, physical activity, and mental health. To avoid perpetuating gender and health inequities in responses to public health emergencies such as COVID-19 is not perpetuate gender and health inequities, it is important that gender norms, roles, and relations that influence women's and men's differential health communication behaviors are considered and addressed. [19] Early evidence is mixed regarding whether the widespread transition to telehealth during the pandemic is creating additional fractures in society by disproportionately influencing health equity. [3] Our results show that implementation of remote health care or using mobile health needs to be grounded in strong gender analysis and ensure meaningful participation of affected groups, including women.

The pandemic induces a slight decrease in the quantity and intensity of physical activity in both genders, as well as an increase in alcohol drinking, which has a negative effect on health. [20] We found that women have a lower level of physical activity than men, and this gap has slightly increased during the pandemic. The effect of sex is significant before and after the pandemic regarding alcohol assumption, and men drink more than women. However, it is necessary to highlight that the pandemic has a greater negative impact on women.

In terms of mental health, the COVID-19 pandemic can be experienced in many different ways, including feelings of depression, fear, panic, and anxiety. Stress-related disorders and the long-term consequences of COVID-19 on health outcomes highlight another important effect of sex and gender. COVID-19 acts as a potent stressor, with millions of individuals experiencing fear and social isolation over a prolonged period. [10] In this study, we found that women are more likely to have increased vulnerability to and severity of stress-related psychiatric disorders than men. These observations may be helpful to developing and implementing prevention and treatment interventions that are able to address the acute and long-term effects of the pandemic on the health and well-being of both male and female populations.

While the mailing and the reminder postcards were sent out on schedule and without any issues, the World Health Organization's announcement on March 11 of the COVID-19 pandemic impacted the rest of the field period in HINTS 5 Cycle 4. Given the limitations of the dataset, we . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted September 5, 2021. ;  did not have information about when the participants filled out the survey, which may have impacted the results. Because the survey was cross-sectional, we could not examine causality among variables. Despite these limitations, this study provides nationally representative estimates and contributes to a better understanding of the effects of the gender differences and COVID-19 pandemic on common characteristics of health and health care.

The COVID-19 pandemic has devastated marginalized communities, exposing the deep inequities of the health care system. This study used multivariate models to understand the effect of COVID-19 pandemic, the effect of sex, and the joint effect of sex and COVID-19 pandemic in relation to health communication, physical activity, mental health, and behavioral health. The findings of this study demonstrate that the COVID-19 pandemic amplifies existing gender disparities in some health and health care domains. Intersectional gender analyses are integral to addressing issues that arise and mitigating inequities. Responses to the pandemic should consider diverse perspectives, including sex and gender. Decisions that are informed by accurate data and include sex and gender perspectives are essential.

Z.R. acknowledges the support from ONR grant N00014-20-1-2337.

J.Y. conceived and designed the study, and contributed to the analyses. Z.R. contributed to the analyses. J.Y. and Z.R. contributed to the interpretation of the results, drafting, and revision of the manuscript. All the authors read and approved the final version of the manuscript.

All data referred to in the manuscript are available on: https://www.cancer.gov/, and the relevant code and analyses are available at: https://github.com/zhimeir/hints_analysis

None.

. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted September 5, 2021. ;

Patient characteristics associated with telemedicine access for primary and specialty ambulatory care during the COVID-19 pandemic

The impact of electronic health record-integrated patient-generated health data on clinician burnout

Racial and ethnic differences in self-reported telehealth use during the COVID-19 pandemic: a secondary analysis of a US survey of internet users from late March

CDC health disparities and inequalities report-United States, 2013. Foreword. MMWR supplements

Prevalence and patterns of tobacco smoking among Chinese adult men and women: findings of the 2010 national smoking survey

World Health Organization, Gender and COVID-19: advocacy brief

COVID-19 exacerbating inequalities in the US

Acceptability of app-based contact tracing for COVID-19: Crosscountry survey study. JMIR mHealth and uHealth

53% of Americans Say the Internet Has Been Essential during the COVID-19 Outbreak: Americans with Lower Incomes Are Particularly Likely to Have Concerns Related to the Digital Divide and the Digital" Homework Gap

Sex and gender differences in health: what the COVID-19 pandemic can teach us

The role of health technology and informatics in a global public health emergency: practices and implications from the COVID-19 pandemic

The health information national trends survey (HINTS): development, design, and dissemination

A multivariate model of gender differences in adolescents' internalizing and externalizing problems. Developmental psychology

Categorical data analysis

Why most published research findings are false

When to use the B onferroni correction. Ophthalmic and Physiological Optics

An analysis of gender bias studies in natural language processing

Sex and gender disparities in the COVID-19 pandemic

COVID-19: the gendered impacts of the outbreak. The lancet

The effects and patterns among mobile health, social determinants, and physical activity: a nationally representative cross-sectional study