key: cord-299277-ana68vpf
authors: Horn, Sarah R.; Weston, Sara J.; Fisher, Philip A.
title: Identifying causal role of COVID-19 in immunopsychiatry models
date: 2020-05-29
journal: Brain Behav Immun
DOI: 10.1016/j.bbi.2020.05.066
sha: 
doc_id: 299277
cord_uid: ana68vpf

nan

The coronavirus disease 2019 pandemic has rapidly evolved into one of the most serious public health crises in recent history. As COVID-19 continues to spread globally, the field of immunopsychiatry is correspondingly adjusting our studies while maintaining highquality research standards (Holmes et al., 2020) . This goal necessitates significant flexibility given the gaps in our knowledge base and the quickly developing shifts in available information. The primary obstacle in many parts of the world is the inability to precisely discern the proportion of our samples that have been or are currently infected with COVID-19, a problem compounded by the high rates of asymptomatic or pre-symptomatic presentations (Furukawa et al., 2020) . Currently, without widespread testing, the field must creatively adopt strategies to properly assess and then account for COVID-19.

Several factors have been shown to increase the likelihood of contracting the virus or developing a more severe presentation of the illness. In the absence of direct assessment of COVID-19, proxy variables represent a resourceful way to consider the role of the virus in germane immunopsychiatric research. Examples include personality traits (e.g., neuroticism; Kroencke et al., 2020) , geographic location (CDC, 2020), occupation, working from home versus community, socioeconomic status (SES) (Ahmed et al., 2020) , racial/ethnic and minority status (Yancy, 2020) , underlying health conditions (CDC, 2020), mental health symptomatology , and proximity to documented cases. As access to testing improves, there may be unique strategies to more accurately estimate the rates of exposure in specific locations (e.g., comparing to another location with a similar geographic or demographic background with a comparable infection rate). Expectedly, several of these factors align strongly with an immunopsychiatry framework, highlighting further the strong connections between stress, physical and mental health, and immunological processes that are particularly relevant in the wake of COVID-19.

Under these circumstances, causal models in immunopsychiatry will, for the foreseeable future, need to account for COVID-19. Furthermore, correctly specifying the role of COVID-19 in these models is critical for hypothesis testing. A central issue is whether COVID-19 is treated as a confounding variable (e.g., as causal influence on both inflammation and mental health symptoms; Wang et al., 2020) or a mediating variable. (e.g., explaining the causal link between hypercytokinemia and physical outcomes; Troyer et al., 2020) . In this paper, we examine implications for conceptualizing and analyzing measures associated with COVID-19 (e.g., diagnosis, viral load) as confounding versus mediating variables. To explore the implications of this distinction, we visualize simulated data varied according to sample and effect size, as well as the role of COVID-19 in the causal model. We consider the statistical implications of each of these scenarios.

There are several immunopsychiatry examples in which COVID-19 may be appropriately considered a confounder, including research on inflammation and mortality risk. We simulated scenarios in which the virus directly elicits a host immunological response (i.e., inflammation) with either a weak (= .10) or strong ( = .50) effect and also directly impacts mortality, again with both weak and strong effects. In all four scenarios, the direct relationship between inflammation and mortality was set to  = .30. As shown in Figure 1 , models that test the direct relationship with COVID-19 as a covariate are unbiased, but models that omit the virus are positively biased. Furthermore, bias was unmitigated by sample size. In the case when COVID-19 causes both the outcome and predictors of interest in a study, it must be treated as a confounder and statistically controlled, or estimates will be biased.

[insert Figure 1 ] Scenario 2: Do not control for COVID-19 when it is a mediator

Other immunopsychiatric-related research questions exist in which COVID-19 may be conceptualized as mediating key pathways among other variables. For example, we simulated scenarios in which low SES is a cause of COVID-19 (e.g., through pathways such as increased financial pressure to work; Ahmed et al., 2020) , and COVID-19 in turn causes respiratory distress . Like before, we simulated effect sizes leading to and away from COVID-19 as either strong or weak, and set the direct relationship between the predictor and outcome of interest (SES and respiratory distress, respectively) to a moderate effect size. As shown in Figure 2 , models that omit the virus best estimate the total effect of SES on distress (i.e., the total effect). Unless researchers are interested in the effect of all pathways from SES to respiratory distress except COVID-19, they should not include the virus as a covariate in their models, or their estimates will not capture the full pathway (see Figure 3 ). Again, larger samples do not mitigate the amount of bias.

[

A significant challenge is how to address research where a strong argument could be made to support COVID-19 as either a confounder or mediator. An example is research on the causal impact of immune functioning on depression, in which it is unclear where to incorporate COVID-19 in the model. It can be argued that those with low immune functioning are more susceptible to contracting the virus; yet, there is no doubt that those who contract COVID-19 suffer short-term declines in immune functioning as they recover. In this case, it is difficult to strictly dictate methodological choices, as doing so requires knowing the true underlying model. Establishing temporal precedence can help guide decisions. Overall, we recommend that in these circumstances, researchers should present both zero-order and partial relationship.

In summary, the challenge for immunopsychiatry researchers is to identify the causal effects of COVID-19 and appropriately incorporate these effects into theoretical causal models. The COVID-19 pandemic is an unprecedented test for the field; yet, immunopsychiatry is also uniquely poised to meet this challenge and rigorously examine the impact of the virus on the human body and mind for years to come. (Rosseel, 2012) package, with varying sample sizes; each combination of population parameters and sample size was simulated 10,000 times. The causal relationship of inflammation to mortality is set to 0.30. This association was tested using both a simple linear model and a regression model with COVID-19 as a covariate. Estimates of the parameter are depicted using boxplots; a solid, horizontal line represents the true population parameter, for reference. Code to recreate these simulations can be found at https://github.com/sjweston/PNI-covid-simulation.

Figure 2. Data were simulated from models in which COVID-19 mediates the causal relationship between socioeconomic status (SES) and respiratory distress. Values are simulated from this causal model using the lavaan (Rosseel, 2012) package, with varying sample sizes; each combination of population parameters and sample size was simulated 10,000 times. The direct causal relationship of SES to inflammation is set to 0.30. The association between SES and inflammation was tested using both a simple linear model and a regression model with COVID-19 as a covariate. Estimates of the parameter are depicted using boxplots, which horizontal lines at the true direct effectin all simulationsand also at the true indirect effect (calculated by multiplying the true causal pathways to and from COVID-19) and the true total effect (calculated by adding the direct and indirect effects), for reference. Code to recreate these simulations can be found at https://github.com/sjweston/PNI-covid-simulation. 

Why inequality could spread COVID-19. The Lancet Public Health

Response Team. Geographic Differences in COVID-19 Cases, Deaths, and Incidence-United States

Evidence supporting transmission of severe acute respiratory syndrome coronavirus 2 while presymptomatic or asymptomatic

Multidisciplinary research priorities for the COVID-19 pandemic: a call for action for mental health science

Neuroticism and Emotional Risk During the Covid-19 Pandemic

Lavaan: An R package for structural equation modeling and more. Version 0.5-12 (BETA)

Are we facing a crashing wave of neuropsychiatric sequelae of COVID-19? Neuropsychiatric symptoms and potential immunologic mechanisms

Immediate psychological responses and associated factors during the initial stage of the 2019 coronavirus disease (COVID-19) epidemic among the general population in China

Pathological findings of COVID-19 associated with acute respiratory distress syndrome. The Lancet respiratory medicine

COVID-19 and African Americans