Profiling COVID‐related experiences in the United States with the Epidemic‐Pandemic Impacts Inventory: Linkages to psychosocial functioning

Abstract The COVID‐19 pandemic has had a profound impact on the lives of individuals, families, and communities around the world with constraints on multiple aspects of daily life. The purpose of the present study was to identify specific profiles of pandemic‐related experiences and their relation to psychosocial functioning using the 92‐item Epidemic‐Pandemic Impacts Inventory (EPII). Data were collected as part of a cross‐sectional, online survey of adults (18+) residing in the Northeast region of the United States (N = 652) and recruited via online advertisements. Person‐centered latent class analysis (LCA) was applied to 38 pandemic‐related experiences that showed a significant bivariate correlation with perceived stress. Measures of psychosocial risk were also obtained. Results revealed five unique profiles of respondents based on patterns of pandemic‐related experiences. Three profiles representing about 64% of the sample were characterized by moderate to high exposure to adverse experiences during the pandemic and were more likely to screen positive for depression, anxiety, and posttraumatic stress. These profiles were differentiated by sociodemographic differences, including age, caregiving, and employment status. Two profiles differentiated by age and caregiver status represented about 36% of the sample and were characterized by relatively low exposure to adverse experiences and lower risk for psychosocial impairment. Findings support the EPII as an instrument for measuring tangible and meaningful experiences in the context of an unprecedented pandemic disaster. This research may serve to identify high‐risk subpopulations toward developing public health strategies for supporting families and communities in the context of public health emergencies such as the COVID‐19 pandemic.


| INTRODUC TI ON
In the first three-month period, the COVID-19 pandemic has had a profound impact on individuals, families, and communities across the world. Intensive health precautions have created constraints on mobility (e.g., sheltering in place, self-quarantining), work and schooling (e.g., virtual commuting, homeschooling), family life (e.g., more intensive contact in primary household relationships, separation from extra-household family members), and interpersonal relationships (e.g., social distancing, wearing masks). A number of survey studies have found evidence of an increase in self-reported emotional and behavioral problems (e.g., anxiety, depression, stress disorders, insomnia) in the pandemic's immediate wake (Breslau et al., 2021;Fu et al., 2021;García-Fernández et al., 2020;Murata et al., 2021;Petzold et al., 2020;Wu et al., 2020). However, few studies have examined the impact of specific pandemic-related experiences or patterns of experiences, both negative and positive, on functioning, which is necessary for understanding the origins of burden on families toward developing public health interventions. Some exceptions include a study associating limitations in mobility with higher psychosocial distress in a United States sample (Devaraj & Patel, 2021) and an international study associating COVID exposure, government-imposed quarantine level, and lifestyle changes with increased reports of depression and anxiety , as well as increases in domestic conflict with self-reported sleep difficulties .
The goal of the present study was to employ a person-centered analytic approach for empirically identifying specific profiles of pandemic-related experiences and their relation to psychosocial functioning with information from the novel Epidemic-Pandemic Impacts Inventory (EPII; Grasso et al., 2020). The EPII is a comprehensive, 92-item inventory of experiences that extend across five thematic domains including adverse experiences specific to work/ employment, home life, social activities and quarantine, and emotional/physical health and infection, as well as positive changes.
The EPII is currently maintained in the National Institute of Health (NIH) Disaster Research Response (DR2) Repository of  Research Tools (https://dr2.nlm.nih.gov). Recent studies using the EPII have associated specific pandemic-related experiences with increased risk for depression and anxiety Yuksel et al., 2021), cumulative counts of adverse experiences with psychosocial distress and coping difficulties among teachers , and positive experiences with better psychosocial health in Scottish adults (Williams et al., 2021).
The present study sought to determine whether patterns of cooccurring pandemic-related experiences on the EPII define unique profiles of individuals that also differ on sociodemographic characteristics and psychosocial functioning. Profiles were empirically determined using latent class analysis (LCA), an exploratory, person-centered, datadriven approach for clustering individuals on a set of characteristics.
LCA was applied to a subset of EPII items showing a significant correlation with a separate measure of perceived stress. Demonstrating unique profiles of pandemic-related experiences that differentially predict psychosocial risk would support the construct validity of the EPII. Notably, traditional factor analytic methods grounded in classical test theory are not appropriate for evaluating the validity of instruments that inventory event-type data (Felix et al., 2019), as is the case for the EPII. Classical methods treat items as indicators of latent constructs and assume an underlying, normally distributed latent variable or variables comprised of correlated indicators. LCA is not bound by these assumptions and provides the means to identify items or experiences that probabilistically co-occur to characterize unique profiles or subgroups of individuals.
LCA also provides information beyond what is possible by summing event-type data to create a cumulative count of experiences that may associate with risk. While this cumulative count approach is practical and statistically robust in predicting outcomes, drawbacks include erroneous assumptions that: (a) all items are equally associated with a particular outcome, (b) distances between sum scores are proportionately associated with an outcome, and (c) equivalent sum scores representing different combinations of items convey the same risk on an outcome (Netland, 2001). As such, the cumulative count approach, while informative, offers little to be learned about risk specific to individual exposures or unique constellations of cooccurring exposures on outcomes.
In contrast, LCA uses maximum likelihood methods to empirically classify individuals into profiles or classes based on probabilistic patterns of co-occurring exposures. LCA is not bound to linear assumptions and can be used to test the significance of different combinations of exposures on outcomes. The trauma exposure field has seen a burgeoning of studies using LCA to identify unique subgroups of individuals with different combinations of trauma exposures (Dierkhising et al., 2019;Ford et al., 2013;Goldstein et al., 2020;Grasso et al., 2013;Grasso, Dierkhising, et al., 2016;Grasso, Petitclerc, et al., 2016). Additionally, to our knowledge, only one study has applied LCA to disaster-specific experiences to examine the impact of flooding on families (Felix et al., 2019). The latter study identified four unique profiles that were differentially associated with depression, anxiety, and PTSD

symptoms.
To this end, the current study applied LCA to stress-related pandemic experiences assessed with the EPII in a cross-sectional survey conducted in the Northeast region of the U.S, the location of the initial epicenter of the pandemic in the U.S. The first aim was to use exploratory, person-centered LCA to examine whether unique profiles of individuals could be identified based on different patterns of probabilistically co-occurring stress-related pandemic experiences endorsed on the EPII. A second aim examined whether identified profiles of individuals would significantly differ on sociodemographic characteristics and psychosocial indicators.
Identifying unique profiles of individuals with distinct patterns of pandemic-related experiences that differentially associate with psychosocial risk would support the validity of the EPII as an inventory of experiences relevant to understanding the impact of the pandemic on daily life, health, and well-being. The availability and efficient and validated measure of the specific impacts of mass disasters is critical for both current and future prevention and intervention efforts.

| Procedures
An anonymous online survey using Qualtrics Survey Software was deployed via advertisements posted on social media (Facebook, Twitter, Instagram, Reddit), listservs, and ResearchMatch.org to re-

| Measures
2.3.1 | The Epidemic-Pandemic Impacts Inventory  The Epidemic-Pandemic Impacts Inventory (EPII) is a 92-item inventory of pandemic-related experiences across several life domains: Work Life (12-items), Home Life (19 items), Social Activities and Isolation (18-items), Emotional/Physical Health and infection (24-items), and Positive Change (19-item). All domains except for the Positive Change domain index negative or adverse experiences. Each item has a response set of "Yes, Me", "Yes, Person in Home", "No", and "Not Applicable", except for items 42, 43, and 65, which pertain to the household more globally. The first two responses can be mutually inclusive. The second response ("Yes, Person in Home") can pertain to family or non-family living in the home and is conceptualized as having a potential impact on the respondent. For the purposes of this paper, the two "Yes" responses were collapsed, as were the "No" and "N/A" responses, which resulted in dichotomous indicators. (Cohen et al., 1983) The Perceived Stress Scale (PSS) is a 10-item measure of one's perception of life is unpredictable, uncontrollable, and overloaded (0 = "Never," 1 = "Almost Never," 2 = "Sometimes," 3 = "Fairly Often," 4 = "Very Often"). The total score is the sum of all items (α = .80).

| The Patient
Health Questionnaire-9 (Kroenke et al., 2001) The Patient Health Questionnaire-9 (PHQ-9) is a 9-item self-report measure of depressive symptoms over the past two weeks that range from 0 ("Not at All") to 3 ("Nearly Every Day"). Total score ranges from 0 to 27. It has established construct validity and excellent test-retest reliability (r = .84; Kroenke et al., 2001). In the present study, internal consistency was .88. The average number of non-disclosed/imputed items across participants was 0.15%.

| The Generalized
Anxiety Disorder-7 (Spitzer et al., 2006) The Generalized Anxiety Disorder-7 (GAD-7) is a 7-item self-report measure of generalized anxiety disorder symptoms over the past two weeks that range from 0 ("Not at All") to 3 ("Nearly Every Day").
Total score ranges from 0 to 21. It has good convergent validity with other anxiety scales and excellent test-retest reliability (intra-class correlation = .83; Spitzer et al., 2006). In the present study, the The Primary Care PTSD Screen for DSM-5 (PC-PTSD-5) is a selfreport measure of DSM-5 defined PTSD symptoms. Five Yes/No items assess symptoms yielding a continuous symptom score ranging from 0 to 5. Previous research has demonstrated that the PC-PTSD-5 predicts PTSD diagnosis with a high degree of accuracy and has good test-retest reliability (Prins et al., 2016). In the present study, internal consistency was .75. The average number of nondisclosed/imputed items across participants was 0.01%.
2.3.6 | The Duke-UNC Social Support Questionnaire (Broadhead et al., 1988) The Duke-UNC Social Support Questionnaire is a 5-item self-report measure assessing one's perception of the availability of support or assistance to fulfill needs. Each item assesses the degree/quantity to which a person feels that they have access to different indicators of social support using a 5-point Likert scale ranging from 0 ("None of the Time") to 5 ("All of the Time"). The Social Support Questionnaire has convergent validity with other measures of social support and general health and good two-week test-retest reliability (r = .66; Broadhead et al., 1988). In the present study, the internal consistency was .88. The average number of non-disclosed/imputed items across participants was 0.15%.

| Analytic approach
Descriptive statistics were calculated with Mathworks Inc. Matlab to determine optimal fit. Information criterion indices include the Bayesian information criteria (Schwartz, 1978), Sample Size Adjusted Bayesian Information Criterion (Sclove, 1987), Consistent Akaike Information Criterion (Bozdogan, 1987), and Approximate Weight of Evidence (Banfield & Raftery, 1993), which are interpreted such that lower values convey better fit. Several relative fit indices were also examined. The Vuong-Lo-Mendell-Rubin Likelihood Ratio Test (Lo et al., 2001) provides comparisons between models, such that nonsignificant values indicate the model with one additional class is not a statistically improved fit over the current model. The Bayes Factor (Wagenmakers, 2007;Wasserman, 2000) is interpreted such that BF less than three is considered weak evidence that the model with one fewer class is superior over the model with one additional class, BF greater than three but less than 10 conveys moderate evidence, and BF greater than 10 conveys strong evidence for the model with one fewer class. The approximate correct model probability (Schwartz, 1978) provides an estimate of the probability that a given model is "correct" among the set of tested models under the assumption that one of the models is "correct". Entropy values were used to evaluate the quality of classes and ranged from 0 to 1, with values closer to one representing better separation of classes (Ramaswamy et al., 1993). Univariate entropy scores were examined to evaluate the relative contribution of individual items in separate classes.
To examine associations between classes and continuous variables, we used the Mplus DU3STEP procedure described by Vermunt (2010) and Asparouhov and Muthén (2014). Associations between classes and dichotomous variables were examined using the Mplus DCAT procedure described by Lanza et al. (2013). These procedures follow 3-steps: (1) the LCA is estimated without covariates or distal outcomes, (2) the highest probability of class membership is used to assign classes, and (3)  sufficient (i.e., entropy > 0.60). Alpha was adjusted for pairwise comparisons using the Bonferroni procedure.   Table 4). Although secondary information criterion fit indices supported a 6-class model, the BIC is the most commonly used and relied upon fit index for comparing models (Masyn, 2017;Nylund-Gibson & Choi, 2018) and both the BF and cmP identified the more parsimonious 5-class model as superior.   Table 6 presents class differences on sociodemographic characteristics, psychosocial risk, perceived stress and social support, and cumulative counts of experiences across all thematic domains from the full EPII. Conditional item probabilities and class differences on proximal variables were used to further characterize and label the five classes.

| Class 1 "Parents -high exposure/high risk"
This class represents about a fifth of the sample and is characterized by a greater probability of living with a partner and caring for a child in the home. Individuals in this class were less likely to report caring for an older adult in the home. This class was differentiated

TA B L E 2 (Continued)
from Classes 4 and 5 by a having greater probability of reporting cumulative pandemic-related experiences in the home life and emotional/physical health domains. Specifically, individuals in this class were relatively more likely to report needing to continue to work despite the risk, experiencing childcare issues, having to take over teaching at home, using harsher discipline, observing an increase in child behavior problems, and experiencing an increase in verbal conflict with a partner. This class was also differentiated from Classes 4 and 5 by having a greater probability of screening positive for possible PTSD, depression, and anxiety and reporting higher levels of perceived stress, but also from Classes 3 and 5 in reporting a higher level of positive experiences related to the pandemic.

| Class 2 "Young adult -high exposure/high risk"
This class represents about 14% of the sample and is characterized by a relatively greater probability of being a young adult. This class was also differentiated from Classes 4 and 5 by having a higher

| Class 3 "Older adult -moderate exposure/ high risk"
This class is the largest class representing about 30% of the sample. Class 3 was differentiated from other classes by having a higher probability of being over the age of 60 and being retired.
This class also had a relatively lower probability of cumulative pandemic-related experiences in the work, home life, and social/ isolation domains, but a relatively higher probability of experiences in the emotional/physical health domain, similar to Classes 1 and 2 but with lower probability of reporting barriers to accessibility of mental health treatment. This class was differentiated from Classes 4 and 5 by having a higher probability of screening positive for PTSD, depression, and anxiety, and reporting higher levels of perceived stress, and from Classes 1, 2, and 4 in reporting fewer positive pandemic-related experiences.

| Class 4 "Young parents -high work/low risk"
This class represents about 17% of the sample. Individuals in this class had a high probability of being a young adult and caring for a child in the home. Individuals in this class were also relatively more likely to be employed, but also to have public or no insurance. This class had a relatively high probability of reporting cumulative

| Class 5 "Older adult -low exposure/low risk"
This class represents about 19% of the sample and was differentiated from Classes 1, 2, and 4 by having a higher probability of being an older adult and being retired. Individuals in this class were least likely, relative to other classes, to report pandemic-related experiences in all domains. This class was differentiated from Classes 1, 2, and 3 by having a lower probability of screening positive for PTSD, depression, or anxiety, reporting lower levels of perceived stress, and reporting higher levels of social support, and from Classes 1, 2, and 4 in reporting fewer positive experiences.

| Summary of Class Differences
Overall, classes with the most negative pandemic-related experiences included Classes 1, 2, and 3. These classes also tended to include a greater proportion of individuals screening positive for PTSD, depression, and anxiety, reporting higher levels of perceived stress, and reporting lower levels of social support. Additionally, two of those classes that had a higher probability of younger or mid-life adults (Classes 1 and 2) along with the class with a high probability of young adults (Class 4) reported more positive pandemic-related experiences than the two classes that had a higher probability of older adults (Classes 3 and 5).

| D ISCUSS I ON
The current findings demonstrate the utility of the EPII in identifying unique profiles (classes) based on patterns of stress-related pandemic experiences using a person-centered analytic approach. Five unique profiles were identified as the best-fitting solution. Findings also provide evidence of differential associations between identified profiles and sociodemographic characteristics and psychosocial functioning.
Caregivers of children and adolescents were more likely to be classified into two profiles differentiated by exposure level and psychosocial risk. Individuals in the "Parents -High Exposure/High Risk" (Class 1) profile comprised about a fifth of the sample and were likely to report cumulative pandemic-related experiences.
This profile was specifically differentiated from other profiles by having a greater probability of caring for a young child and reporting difficulties with childcare and teaching at home, increased child emotional and behavioral problems, and perhaps consequently, increased use of harsh discipline. In contrast, individuals classified with the "Parents -High Exposure/High Risk" profile more likely to report high perceived stress and to screen positive for possible PTSD, depression, and anxiety, and less likely to report high levels of social support relative to individuals classified in the "Young Parents -High Work/Low Risk" profile.
The two 'caregiver' profiles tell different narratives of family experiences over the pandemic, with one profile clearly set apart by a heavier burden including caring for young children, fewer social resources, and greater risk for psychosocial impairment. Although the current study did not collect information about child functioning, the psychosocial risk of the "Parents -High Exposure/High Risk" profile likely extends to children in the home. For example, a recent study using the EPII demonstrated that parents' emotional availability and ability to maintain a stable home routine served to buffer the impact of pandemic-related stress on children's emotional and behavioral problems (Cohodes et al., 2021). Membership in the high-risk class might indicate a need for family support programs and services to assist with parent management and home education.
Older adults were more likely to be classified in two profiles that were differentiated by exposure level and psychosocial risk.

Individuals classified in the "Older Adult -Low Exposure/Low Risk"
profile represented about a fifth of the full sample and were likely to be retired and to report relatively fewer pandemic-related experiences across thematic domains. Despite their age difference, Individuals with this older adult profile were similar to those classified in the "Young Parents -High Work/Low Risk" profile in that they tended to report the lowest levels of perceived stress and the highest social support, while also less likely to screen positive for PTSD, depression, or anxiety. In contrast, individuals classified in the "Older Adult -Moderate Exposure/High Risk" profile represented the largest proportion of the sample (30%) and were more likely than other older adults to report cumulative pandemic-related adverse experiences in the work, home life, and emotional and physical health domains, with a high probability of reporting increased social isolation, mental health, sleep, and alcohol/substance use problems, as well as increased negative lifestyle behaviors (e.g., less physical activity, unhealthy eating). Like the other high-risk profiles, individuals with this profile were likely to report high levels of perceived stress and to screen positive for PTSD, depression, and anxiety. Both the lower risk and higher risk older adult classes were notably less likely to report positive pandemic-related experiences than the young or midlife adult classes.
The physical, psychological, and social vulnerabilities that come with older age may make managing life with COVID particularly challenging for older individuals and lead to compound risk for psychosocial impairment (Banerjee, 2020 TA B L E 6 (Continued) for older adults. The two profiles representative of older adults in the current study reflects two distinct narratives of pandemic life -one in which there may be insufficient resources, increased social isolation, and high risk for adverse outcomes, and one in which there may be safeguards and resources in place to help older individuals adapt to increased restrictions, lifestyle changes, and disruptions in services and to remain socially connected despite social distancing measures. These patterns might suggest the need for intensive efforts to help older individuals adapt to the many changes to their daily activities and implement strategies for risk prevention while still maintaining critical social connections. These individuals may include "essential" workers in healthcare, social and environmental services, and commercial settings suggested to be a heightened risk for psychosocial impairment due to work-related stress (Kang et al., 2020). These individuals may also include those with preexisting mental health difficulties or life stressors that may have exacerbated the impact of the pandemic on emotional health (Mukhtar & Rana, 2020).
This profile highlights the need for specialized services geared toward serving the highest risk individuals and those that might need assistance in multiple life domains, including work, family life, and personal health.
The current study has limitations. Notably, the sample represents a convenience sample recruited through social media advertisements, listservs, and utilization of a research participant database.
As such, the sociodemographic composition of the sample is quite homogenous and not representative of lower income and racially/ ethnically diverse populations. It is quite possible that experiences with low base-rates in the current sample may have been more prevalent in a sample representative of a more diverse, under-resourced population. Despite this important limitation, the value of this preliminary study lies in its capacity to substantiate the EPII's ability to identify meaningful profiles of pandemic-related experiences with linkages to indicators of perceived stress and impairment, even in a non-diverse sample. Future research will need to extend these findings to more diverse populations.
Another consideration is that rates of pandemic-related experiences on the EPII may change over time. The current study surveyed individuals during and after peak rates of COVID-19 in the Northeast region. It is possible that response patterns on the EPII may have been different if the sample was surveyed later, as more time passed provides greater opportunity for these experiences to occur. Ideally, the EPII would be administered at multiple time points across the pandemic so as to track changes in rates of different types of experiences.
In summary, the current study represents the first investigation to use a person-centered, data-driven approach to characterizing and contextualizing negative and positive experiences of individuals during an unprecedented pandemic crisis. It serves to provide an example of an approach that may be suitable for studying future disasters and public health emergencies, as well as support a novel instrument for measuring tangible, pandemic-related experiences. More research will be necessary to understand how pandemics such as COVID-19 impact personal and social experiences over time and how patterns and impacts might differ in various populations.

CO N FLI C T O F I NTE R E S T S
None declared.

PEER R E V I E W
The peer review history for this article is available at https://publo ns.com/publo n/10.1002/brb3.2197.

DATA AVA I L A B I L I T Y S TAT E M E N T
Research data are not shared.