Association of Major Depressive Symptoms With Endorsement of COVID-19 Vaccine Misinformation Among US Adults

IMPORTANCE Misinformation about COVID-19 vaccination may contribute substantially to vaccine hesitancy and resistance. OBJECTIVE To determine if depressive symptoms are associated with greater likelihood of believing vaccine-related misinformation. DESIGN, SETTING, AND PARTICIPANTS This survey study analyzed responses from 2 waves of a 50-state nonprobability internet survey conducted between May and July 2021, in which depressive symptoms were measured using the Patient Health Questionnaire 9-item (PHQ-9). Survey respondents were aged 18 and older. Population-reweighted multiple logistic regression was used to examine the association between moderate or greater depressive symptoms and endorsement of at least 1 item of vaccine misinformation, adjusted for sociodemographic features. The association between depressive symptoms in May and June, and new support for misinformation in the following wave was also examined. association between depression and spread and impact of misinformation merits further investigation.


Introduction
The potential for misinformation to impact public health behavior was recognized prior to the COVID-19 pandemic, 1 but since the onset of the pandemic, the consequences of misinformation have become even more apparent. Popular misperceptions are associated with hindering efforts to mitigate the spread and consequences of the SARS-CoV-2 virus by minimizing perceived risk of infection, discouraging masking and distancing behaviors, and reducing vaccination rates. 2,3 While misinformation is increasingly well studied, most of this work has concentrated on how and why such misinformation spreads. Less understood are individual characteristics, beyond simple demographics and political affiliation, associated with greater susceptibility to misinformation, such as examined in a study by Druckman et al. 4 Notably, misleading news stories inspiring negative emotions, such as disgust, have been found to spread more rapidly on social media. 5 A general bias toward negativity in information selection, processing, and recall 6,7 may exacerbate misinformation exposure. In the context of political misinformation, both anger and anxiety are associated with promoting beliefs in certain types of false stories. 8 During the COVID-19 pandemic, approximately one quarter of adults in the US have consistently endorsed moderate or greater depressive symptoms. 9,10 As depressive symptoms may contribute to negativity bias, we hypothesized that such symptoms would be associated with greater receptivity to misinformation, with potentially profound associations with health-related behaviors.
We used data from a 50-state US survey to examine this hypothesized association in 2 ways. First, with cross-sectional data from more than 15 000 individuals, we characterized the association between presence of depressive symptoms and endorsement of misinformation. Second, examining the subset of individuals who completed 2 waves of the survey approximately 1 month apart, we examined the extent to which depressive symptoms on the initial survey were associated with endorsement of new misinformation 1 month later. We then examined potential mediators or moderators of these associations and the association between misinformation and vaccination status.

Method
This survey study was reviewed by the institutional review board of Harvard University and determined to be exempt; all participants signed informed consent online prior to survey access. In reporting results, we follow the American Association for Public Opinion Research (AAPOR) reporting guideline for survey studies.

Study Design
The COVID States Project 11 survey has been conducted approximately once every 6 weeks since April 2020. Of note, participants are not aware that they are completing a survey focused on COVID-19 a priori, in an effort to limit selection bias. Our analysis used the 2 waves conducted between April 1 and May 3, 2021, and between June 9 and July 7, 2021, which included questions about vaccinerelated misinformation. This online survey applies nonprobability sampling and representative quotas to balance age, gender, and race and ethnicity across 50 states and the District of Columbia.
That is, instead of randomly sampling the full US population as in probability sampling (eg, by random digit dialing), for reasons of feasibility, this survey samples individuals who choose to participate in online surveys, but applies quotas and reweighting to approximate the US adult population in each state. Each adult in the population thus does not have an equivalent probability of being selected.
Survey results were weighted based on US Census data to balance on age, gender, race and ethnicity, education, region, and rural or urban area of residence.

Measures
We assessed vaccine-related misinformation using 4 statements, which respondents were asked to rate as accurate (statement is true), inaccurate (statement is not true), or not sure. We selected these statements based on misinformation prevalent on social media platforms in spring 2021. Specific statements of misinformation included "The COVID-19 vaccines will alter people's DNA," "The COVID-19 vaccines contain microchips that could track people," "The COVID-19 vaccines contain the lung tissue of aborted fetuses," and "The COVID-19 vaccines can cause infertility, making it more difficult to get pregnant." At the conclusion of this survey section, all respondents were informed which items were not true, to ensure that the survey itself did not facilitate spread of misinformation.
For cross-sectional analysis, as in our prior work, 4 we categorized any accurate responses as reflecting belief in misinformation. For longitudinal analysis, we categorized an increase in the number of statements labeled accurate as worsening belief in misinformation, for example, going from no statements labeled accurate in the first wave to 1 or more statements labeled accurate in the second wave.
Survey participants also completed the Patient Health Questionnaire 9-item (PHQ-9) as a measure of major depressive symptoms over the preceding 2 weeks. 12 In primary care settings, a value of 10 or greater represents at least moderate depression and is often applied as the threshold for treatment; therefore, we elected a priori to examine presence or absence of major depressive symptoms at this threshold, as in our prior work using these survey items, rather than assuming a linear or dose-response association between depression and misinformation.
Additional survey items asked respondents whether they used particular social media platforms and whether they had used any of a list of news sources (including MSNBC, Fox News, CNN, Newsmax, Facebook, and the Biden administration) as sources of COVID-19-related news over the prior 24 hours. Sociodemographic features, including race, ethnicity, and gender, were identified by self-report. Race and ethnicity data were collected to ensure representativeness of the US population for the survey as a whole. Region (ie, Northeast, South, Midwest, and West) and urban or rural status were assigned based on zip code. Ideology was assessed using a 7-point scale (range, 1 to 7, with 1 indicating extremely liberal; 4, moderate; and 7, extremely conservative). Political party affiliation was determined by asking, "Generally speaking, do you think of yourself as a…" with Democrat, Republican, Independent, and other as options; for analytic purposes other and Independent were combined in a single category. Respondents were also asked if they had received at least 1 COVID-19 vaccination; if they had not, they were further asked "If you were able to choose when to get a COVID-19 vaccine, would you get it…", with response options including "as soon as possible," "after at least some people I know," "after most people I know," or "I would not get the COVID-19 vaccine." The last category was considered to be vaccine resistant.

Statistical Analysis
For purposes of primary analysis, we applied multiple logistic regression to examine the association between presence of at least moderate depressive symptoms by PHQ-9 and endorsing at least 1 item of vaccine-related misinformation. These models were fit without adjustment and then with We also examined the possibility that individuals with depressive symptoms might be less confident in their responses to questions about misinformation, as indicated by presence of at least 1 not sure answer to misinformation questions. These analyses used logistic regression models, with the same covariates used to examine presence of misinformation.
In secondary analysis, we examined potential mediating or moderating associations of social media use, news sources, and trust in institutions with the association between mood and misinformation. That is, we considered the possibility that these associations could arise in association with mood and media use, news consumption, or willingness to trust institutions. We analyzed social media use via terms for self-reported use of Twitter, Facebook, TikTok, and Instagram, and news sources for receiving COVID-19 information within the past 24 hours, including CNN, MSNBC, Fox News, Newsmax, Facebook, or the Biden administration. We then examined selfreported trust in institutions (ie, the White House, the Food and Drug Administration, and the Centers for Disease Control and Prevention), in hospitals and physicians, in scientists, and in news media. For each of the trust variables, we asked, "How much do you trust the following people and organizations to do the right thing to best handle the current coronavirus (COVID-19) outbreak?" and used a 4-point scale from 1 indicating not at all to 4, a lot. To understand the association between misinformation and vaccine-related behavior, we compared rates of vaccination and rates of vaccine resistance among individuals who did or did not endorse misinformation.
Finally, we used the subset of individuals who responded to both the April to May and June to July waves to analyze whether presence of depression in April to May was associated with incident (rather than prevalent) misinformation (ie, emergence of additional misinformation from one wave to the next). We again used multiple logistic regression to adjust for sociodemographic features, with national reweighting, using noninterlocking weights. P values were 2-sided, and statistical significance was set at P < .05.  1). Individuals with moderate depression were also more likely to indicate that they were not sure about at least 1 item of misinformation, although the association was no longer significant after adjustment (crude OR, 1.25; 95% CI, 1.14-1.38; adjusted OR, 1.10; 95% CI, 0.99-1.22).

Among
We next examined potential factors associated with mediating or moderating this association by considering whether the association between misinformation and depression was meaningfully changed by addition of terms to the multiple regression models. Table 2 shows the base model and the adjusted ORs associating misinformation with depression in models incorporating social media, news source, or trust variables. While all of these variables were significantly associated with misinformation, in all additional models, the ORs changed by less than 10%, suggesting modest mediating or moderating associations at best.

Discussion
In this survey study using national data including more than 15 000 respondents, we found that presence of moderate or greater depressive symptoms was associated with greater likelihood of endorsing misinformation about vaccines, an association that persisted with adjustment for sociodemographic features as well as self-reported ideology and political party affiliation. This crosssectional study design does not allow us to investigate causation, so the nature of the association between these features remains to be determined. However, using a subset of participants from the   first wave who returned for the second, we found that depressive symptoms preceded misinformation emergence, suggesting that misinformation was unlikely to cause depression per se.
In general, negative biases are apparent in information processing even in the absence of depression. 6,7 Individuals with major depressive symptoms often exhibit a more pronounced negativity bias, a form of attentional bias in which thoughts with negative valence receive greater focus. 13 Insofar as forms of misinformation that elicit negative affect may be more likely to spread, 5 it follows that depression could facilitate uptake of misinformation at an individual level.
Alternatively, it is possible that the association between depression and misinformation could be mediated by change in trust. Individuals with depression could exhibit less willingness to trust institutions attempting to combat misinformation, such as the Centers for Disease Control and Prevention, or greater willingness to trust other institutions that distribute misinformation. However, we found that incorporating terms for trust in these institutions in regression models did not change the main association with depression, which does not support a mediating association of trust in institutions.
As anticipated, we also found that individuals who embraced health misinformation were less likely to be vaccinated or be willing to get the vaccine if available. As such, individuals already burdened with depression may be at a higher risk of COVID-19. While beyond the scope of the present work, it bears noting that individuals with depression may also exhibit a lack of positive interpretation bias, 14 ie, less optimistic beliefs, 15 which could lead them to underestimate the potential benefit of vaccination. Notably, mood disorders have been associated with worse COVID-19 outcomes among hospitalized patients. 16

Limitations
This study has some limitations. While we adjusted for a range of sociodemographic features, we cannot exclude the role of confounding in the observed association. One potential confounder could be the use of social media: it is possible that more individuals with depression are more prone to use certain forms of social media, and those platforms may be more likely to promote misinformation.
Alternatively, social media use could promote both depression [17][18][19] and misinformation 1 independently. Similarly, depression might be associated with different choices in news media.
However, adding terms for individual social media platforms or news sources to regression models did not substantially change the associations between depression and misinformation, suggesting this is less likely to be the case.

Conclusions
This survey study found that individuals with moderate or greater depressive symptoms were more likely to endorse vaccine-related misinformation, cross-sectionally and at a subsequent survey wave.
While associative by necessity, our results more broadly suggest the importance of directly testing causation in future experiments, for example, by manipulating negativity bias and measuring the receptivity to misinformation. If causation could be established, it might suggest strategies aimed at reducing the consequences of depression in terms of misinformation. To date, efforts to combat the impact of misinformation on public health predominantly emphasize reduction in supply. In parallel, it may be possible to develop interventions targeting negativity bias that reduce demand, or at least modulate the capacity of misinformation to impact health decision-making.