Data and Participants
We used data from the Midlife Development in the United States (MIDUS), a national study of health and well-being (http://midus.wisc.edu). The MIDUS study was started in 1995 (MIDUS 1), followed by the longitudinal follow-up (MIDUS 2) in 2004. Descriptions of the MIDUS 1 and 2 studies are provided elsewhere [36, 37]. In 2011, a new national probability sample was recruited to participate in the MIDUS Refresher study. The MIDUS Refresher study was aimed to refresh and expand the overall MIDUS study by recruiting a new sample that was comparable to the participants in MIDUS 1 in terms of demographic characteristics. In addition, the MIDUS Refresher was intended to examine the impact of the Great Recession in the late 2000s on health and well-being [38, 39].
The study protocol in the MIDUS Refresher was the same as in the previous waves of MIDUS, in which participants were recruited through random dial digits and completed a 30-minute phone interview (n = 3,577, response rate 59%). Most of the MIDUS Refresher participants who completed the phone interview (73%) also completed self-administered questionnaires (SAQs). To increase the racial diversity of participants in the MIDUS Refresher, a supplemental sample comprising most Black participants was recruited from Milwaukee County, WI. The Milwaukee supplemental sample included 508 (response rate = 47.7%) participants who completed in-person interviews; among them, 59% of the participants also completed SAQs. Those who completed the phone/in-person interview and SAQs from the main and supplementary sample (n = 2,899) were eligible to participate in the biomarker assessment of the MIDUS Refresher.
The current study utilized data from 863 (ages 25–76, mean age = 50.8; 52% female; 69% non-Hispanic white) participants who completed the biomarker assessment in the MIDUS Refresher study. The biomarker assessment in the MIDUS Refresher was conducted from 2013 to 2016. Participants were invited to stay overnight at one of the three regional clinical research units (CRUs) in the East Coast, Midwest, and West Coast. The CRU selection for each participant was based on the one that imposed the least travel burden. During the stay, participants completed comprehensive biological and health assessments, including blood and urine sample collections. Participants in the MIDUS Refresher study completed the biomarker assessment between 6 months to 2 years after the completion of the baseline survey, with a follow-up time median of about one year. Financial hardship questions were part of the baseline survey questions in MIDUS Refresher, collected from 2011–2014, while information regarding inflammation was from the biomarker assessment. Participants signed informed consent documents to participate in the baseline survey and the biomarker assessment.
Measures
Financial Hardship
Data regarding financial hardship were collected during the baseline survey (2011–2014). Financial hardship was assessed using several indicators in the baseline survey. The authors sorted these items into the material, psychological, and behavioral domains previously described in the cancer prevention and survivorship research [24].
Material domain. Two indicators were used as the measures of the material domain of financial hardship: 1) “availability of money” and 2) “difficulty paying bills.” To operationalize the “availability of money”, we used the item that asked participants: “In general, would you say you (and your family living with you) have more money than you need, just enough for your needs, or not enough to meet your needs?” on a 1–3 scale (1 = more money than you need, 2 = just enough money, 3 = not enough money). To operationalize “difficulty paying bills”, we used the item that asked participants to rate their difficulty level in paying monthly bills on a 1–4 scale (1 = not at all difficult, 2 = not very difficult, 3 = somewhat difficult, 4 = very difficult).
Psychological domain. The psychological domain was measured using two indicators: 1) “financial satisfaction” and 2) “financial control”. To operationalize “financial satisfaction”, we used the item that asked participants to report their financial satisfaction with their current financial situation in a 0 (“the worst possible financial situation”) to 10 (“the best possible financial situation”) scale. To operationalize “financial control”, we used the item that asked participants: “how would you rate the amount of control you have over your financial situation these days?” on a 0 (“no control at all”) to 10 (“very much control”) scale. Responses to both questions were reverse coded; thus, higher scores indicated higher levels of hardships.
Behavioral domain. Measures for the behavioral domain were taken from a scale used in the National Survey of Unemployed Adults conducted by the Heidrich Center for Workforce Development [40]. The scale comprises 15 questions (yes/ no) that assess events/behaviors for job-related hardships (4 questions), home-related hardships (4 questions), and finances-related hardships (7 questions) that participants may experience since the Great Recession in 2008. For job-related hardship, participants were asked whether they had ever lost a job, started a job that they did not like, taken a job below their education or experience level, and whether they had ever taken an additional job. For home-related hardship, participants reported whether they had ever missed mortgage or rent payments, been threatened with foreclosure or eviction, had family or friends move in to save money, and whether they had ever moved in with family or friends to save money. Finally, for finances-related hardship, participants were asked whether they had ever declared bankruptcy, missed a credit card payment, missed other debt payment, increased credit card debt, sold possessions to make ends meet, cut back on their spending, and exhausted unemployment benefits. For each question, responses were coded 1 for yes and 0 for no. Responses from each participant were summed for each sub-domain to get a total number of hardships experienced related to job, home, and finance.
Inflammation Markers
We included three inflammation markers in this analysis, interleukin 6 (IL6), c-reactive protein (CRP), and fibrinogen. Inflammation markers were assayed from fasting blood samples that were collected before breakfast on the second day of biomarker assessment. To ensure consistency, blood samples were collected using standardized procedures [41]. IL6 was measured using the Quantikine® High-sensitivity ELISA kit #HS600B (R & D Systems, Minneapolis, MN). The assay range was 0.156-10 pg/mL, intra-assay CV was 3.73% and inter-assay CV was 15.66%. CRP was measured using a particle-enhanced immunonephelometric assay (BNII nephelometer, Dade Behring Inc., Deerfield, IL). The assay range was 0.014-216 ug/mL, intra-assay coefficients of variability (CVs) range from 2.2 to 4.1% and inter-assay CVs range from 4.72 to 5.16%. Finally, fibrinogen antigen was measured using the BNII nephelometer (N Antiserum to Human Fibrinogen; Siemens, Malvern, PA). The assay range was 2.8–4560 mg/dL, intra-assay CV was 2.7% and inter-assay CV was 4.13–6.64%. IL-6 was assayed in the MIDUS Biocore Laboratory at the University of Wisconsin, Madison, WI. CRP and Fibrinogen were assayed at the Laboratory for Clinical Biochemistry Research at the University of Vermont, Burlington, VT.
Covariates
Multiple covariates were included in the analysis: demographic characteristics, measures of socioeconomic status (SES), and health-related variables. Demographic covariates included age (years), sex (0 = female, 1 = male), and race/ethnicity (0 = racial/ethnic minority, 1 = non-Hispanic white). Two measures of SES were included as covariates, level of education (0 = no bachelor’s degree, 1 = bachelor’s degree or higher) and household-sized adjusted income-to-poverty ratio (%). Finally, health-related covariates included body mass index (BMI; score), the number of chronic conditions (28 conditions, e.g., heart disease, diabetes, alcoholism, depression), and prescription drug use (0 = no, 1 = yes).
Analysis
Analysis for this study was conducted using a structural equation modeling (SEM) framework. Pre-analysis steps for SEM were performed, including extensive data assessment and cleaning and missing data analysis [38, 42]. Data were inspected for the potential univariate (through standardized scores, |z| ≥ 3.30) and multivariate (i.e., Mahalanobis Distance p < .001 and Studentized Deleted Residual greater than ± 4.00) outliers. Although few univariate outliers were identified, we retained them as they were minimally severe (less than four standard deviations away from the mean) [43]. Furthermore, no multivariate outliers were found. Due to normality concerns, natural log-transformed data for CRP and IL6 were used for analysis.
Analysis was conducted in three steps. First, to test whether each item of financial hardship mapped into the corresponding hypothesized domains (material, psychological, and behavioral), we conducted exploratory factor analysis (EFA). Domains of financial hardship were hypothesized to be correlated with each other. Thus, we utilized goemin rotation, which is a type of oblique rotation method. We tested one-, two-, three-, and four-factor models in the analysis. We hypothesized that the three-factor model would show a better fit relative to the other solutions. We also hypothesized that in the three-factor model, financial hardship questions from the same domain would show higher loadings on the same factor. We utilized the factor loading cutoff of |0.4| to consider whether the question was meaningfully associated with the corresponding factor [44]. Second, to examine the second-order hypothesized model of financial hardship (Fig. 1A), we used confirmatory factor analysis (CFA). The goal of the CFA was to examine whether responses to the questions related to financial hardship could be explained by three first-order factors (material, psychological, and behavioral) and one second-order latent factor of financial hardship.
Third, we utilized the second-order measurement model of financial hardship from the second step of the analysis to predict IL6, CRP, and fibrinogen (Fig. 1B). Prediction of IL6, CRP, and fibrinogen by financial hardship was conducted in three separate models. To examine the association between financial hardship and markers of inflammation, we controlled for the influence of covariates on each inflammation marker. Financial hardship was freely correlated with all the covariates, and all the covariates were freely correlated with each other [45, 46]. The assessment of model fit and accuracy in all steps of the analysis was based on multiple criteria, including [45, 46]: (a) various fit indices to evaluate overall goodness of fit, (b) examining whether there were concentrated areas of strain in the solution, and (c) size of the estimates, statistical significance, and the interpretability of the model’s parameter estimates. Throughout the analysis, we used maximum likelihood as the estimation method. All the estimates reported in the results section are based on standardized results. Analysis was conducted using MPlus version 8.8 [47].