Risk Factors and Consequences of Financial Hardship among Family of Critically Ill Patients

Abstract

Rationale

Major knowledge gaps persist in understanding factors that contribute to family financial hardship and its association with other family-reported outcomes when patients experience critical illness.

Objectives

We sought to identify factors associated with family financial hardship and associations with family-reported outcomes.

Methods

We conducted a prospective cohort study at three hospitals within a large academic healthcare system in the Pacific Northwest of the United States. Participants included family members of patients who were admitted to the intensive care unit (ICU) (4/2019–2/2023) and diagnosed with a chronic life-limiting illness or severe acute illness. Clustered bivariate and multiple linear regression models evaluated potential predictors of financial hardship. Clustered linear and probit regression models, adjusted for prespecified covariates, measured associations of 3 hardship with family-reported outcomes.

Results

A total of 783 family participants provided information on financial hardship 1 month after patients’ ICU admission by responding to a modified version of the Comprehensive Score for financial Toxicity instrument. Financial hardship was greater for respondents who were younger (P = 0.019), had less education (P = 0.034), were the patient’s spouse (P = 0.016), lived with the patient (P = 0.018), had dependent children at home (P < 0.001), experienced work-status changes related to patient’s illness (P < 0.001), had poorer health (P < 0.001), lacked emotional support (P = 0.001), assessed hardship while the patient was alive (P = 0.005), or had been in the hospital longer (P = 0.011). Reported hardship was lower when the patient had a cancer diagnosis (P = 0.030), and it varied by racial–ethnic group (P = 0.003) and insurance type (P = 0.045). Financial hardship was associated (P < 0.001) with increased family depression and anxiety and with reduced emotional preparedness and quality of life at 1 and 3 months after the patient’s ICU admission.

Conclusions

For family of critically ill patients, social and demographic factors may play a larger role in financial hardship than clinical factors. Financial hardship is predictive of lower family quality of life. These findings add to the literature demonstrating the prevalence and adverse consequences of financial hardship, its inequitable distribution, and the need for evidence on supports and interventions to mitigate poor outcomes and inequities.

Family members of critically ill patients experience poor health-related quality of life (1, 2) and psychological distress (1–5). These poor outcomes for family members have been well studied in the critical care literature (6, 7). By contrast, the financial burden that family members experience when a loved one is critically ill has only recently started to surface as an important family-centered outcome that needs further investigation. Data suggest that family members’ financial hardship associated with critical illness is widely prevalent, with adverse consequences (8, 9). Major knowledge gaps, however, persist in our understanding of factors that contribute to family financial hardship and its association with other key family-reported outcomes.

Our objectives were to identify factors associated with family financial hardship and associations with family-reported outcomes in a large, prospective longitudinal study of the family members of patients who experienced critical illness. To measure financial hardship, with permission from FACIT.org, we modified the Comprehensive Score for financial Toxicity instrument (10, 11), which has demonstrated reliability and validity in measuring financial toxicity and is widely utilized in the oncology population (hereinafter referred to as the modified FACIT-COST), and we then computed a measure of financial hardship based on the standard toxicity score.

Methods

Design Overview and Setting

Data for this study were collected and aggregated from two studies funded by the National Institutes of Health: the Financial Worries and Health Study (FWHS) and the Facilitating Communication Study (FCS2) (12, 13). FWHS was designed as a prospective, longitudinal, cohort study, whereas FCS2 was a randomized trial of a nurse-led communication intervention to teach communication strategies that enable patients and families to align medical care with the patient’s goals (13). Overlapping eligibility criteria for study participants and inclusion of participants from the same healthcare system (UW Medicine) enabled us to efficiently and effectively achieve large sample sizes to conduct these analyses. We decided on the procedures to collect and aggregate data before the start of both studies. As previously reported, eligible critically ill patients and their families were identified between April 2019 and February 2023 from three hospitals within a large academic healthcare system in the Pacific Northwest of the United States: a quaternary care facility for surrounding regions (79 intensive care unit [ICU] beds); a university-operated, county-owned, safety-net hospital that serves as the only Level 1 trauma center for five surrounding states (89 ICU beds); and a nonprofit community hospital that serves a large number of retirement homes and skilled nursing facilities (15 ICU beds) (14). All family participants provided informed consent for study activities. The study was approved by the University of Washington Institutional Review Board (Institutional Review Board nos. STUDY00008703 and STUDY00005369).

Patient and Family Participants

Patients

Using the electronic health record (EHR) and daily ICU rounds (Monday– Friday), trained study staff identified consecutive patients in the ICU who met the following eligibility criteria: Patients had to 1) be at least 18 years old, 2) be English speaking, 3) have a family member who is available to participate, and 4) have a chronic life-limiting illness suggesting a median survival of 2 years or a severe acute illness with a risk of hospital mortality of 15% or higher during the index hospitalization. Chronic life-limiting illnesses are those associated with a median survival (without critical illness) of 2 years: dementia, cancers with a poor prognosis, chronic pulmonary disease, coronary artery disease, heart failure, chronic liver disease, chronic kidney disease, diabetes with end-organ damage, and peripheral vascular disease (15–17). Acute illness criteria include those with a risk of hospital mortality of at least 15% (for a list of diagnoses, see Figure E1 in the data supplement) and/or an Acute Physiology and Chronic Health Evaluation (APACHE-II) score of 13 or higher. Once deemed eligible, study staff made in-person or telephone contact with the patient or, if the patient did not have decisional capacity, a legal next of kin. Decisional capacity was determined by the attending physician (or designee) to align with clinical practice (18). Patients with neither capacity nor eligible family were excluded.

Family

Family participants were identified by the patient if the patient had capacity. If the patient did not have decisional capacity, family participants were identified through the EHR or by a legal surrogate decision maker. Any family member or friend was eligible if they were at least 18 years old and identified by the patient or legal surrogate as involved in care of the patient, provided they had English-language proficiency and no cognitive deficits limiting their ability to complete surveys (19–22). Family participants were recruited by phone or in person. We did not limit the number of family participants per patient. (For a participant flow diagram, see Figure E2)

Data Collection

Surveys were administered to participants who were members of the patient’s family by phone, in person, or online (the latter by means of the Research Electronic Data Capture, or REDCap system). Surveys from the two studies contained similar measures, including the modified FACIT-COST instrument (10, 11) (see Figure E3). Baseline demographic data collection occurred at the time of study enrollment for FCS2 and 1 month after ICU admission for FWHS. Both studies collected modified FACIT-COST scores and other outcome measures at approximately 1 month after the patient’s ICU admission and again approximately 3 months after the ICU admission. Demographic and clinical information for the associated patients was abstracted from the EHR.

Measures

The primary measure for this study was a measure of financial hardship that is based on the standard toxicity score from the modified FACIT-COST instrument (10) and administered to family participants approximately 1 month after the patient’s ICU admission. (See Figure E3 for wording and response options for the modified instrument. For this study, we imputed a value of “not at all” for respondents who responded “does not apply” to Items 9, 10, or 11.) Computed as a sum of the 11 items, weighted for number of valid responses, the standard toxicity score resulted in a sum that could range from 0 (maximum toxicity level) to 44 (no toxicity). Respondents who lacked responses to all 11 items were excluded. Rather than using the standard toxicity score, however, in our study, we used financial hardship, scored as the reverse of the toxicity score, with values ranging from 0 (no financial hardship) to 44 (maximum financial hardship).

Also included in the study were four outcome variables from family participants’ 1-month and 3-month questionnaires, representing emotional states to which we expected financial hardship to contribute. Two were composite measures from the Hospital Anxiety and Depression Scale (23, 24): a seven-item sum score measuring anxiety, and a seven-item sum score measuring depression. We computed each of these as the sum of the valid responses, weighted for the number of valid responses and required valid responses to at least half of the relevant items to compute the sum (25). The other two were Likert-scaled single items. Emotional preparedness for the future, no matter what happens, is from a questionnaire measuring the quality of family experience—the QUAL-E (Fam) (26)—with ordinal responses reversed in our analyses to measure reduced preparedness, ranging from 0 (a great deal prepared) to 4 (not at all prepared). Quality of life is based on responses to a question from the QUAL-E questionnaire, “How would you rate your overall quality of life?” (27). Our two studies used different response codes for the quality-of-life item. One study used the response codes endorsed by the QUAL-E developers (very poor, poor, fair, good, and excellent); and the other used response codes designated for a companion item in the QUAL-E (Fam) questionnaire (26) (poor, fair, good, very good, and excellent). To compensate for this inconsistency, we revised the numeric coding to encompass four categories, with higher codes indicating poorer quality of life: 0 (excellent), 1 (good or very good), 2 (fair), and 3 (poor or very poor).

We tested 32 variables as potential predictors of financial hardship (see Figure E4 for a depiction how predictors were conceptualized). Ten were based on family questionnaire responses at baseline: their age, gender, racial–ethnic identification, level of education, relationship to the patient, whether they lived with the patient, whether they had dependent children living at home, whether they had experienced any change in work status related to the patient’s illness or treatment, a self-rating of their current health (28), and whether they lacked emotional support (e.g., someone with whom they could talk over problems or who would help them make difficult decisions). The other 22 predictors were drawn from the patient’s medical record or from study administrative data. Nineteen were already known at the time the family member’s baseline questionnaire was completed and the financial hardship was computed: whether the patient was a participant in the randomized trial and whether the patient was assigned to the intervention group in that trial; whether the patient had died by the time the baseline questionnaire was completed; the patient’s age, insurance coverage, and number of inpatient stays during the year before the index ICU admission; documentation of each of the nine Dartmouth Atlas chronic conditions (poor prognosis cancer, chronic pulmonary disease, coronary artery disease, heart failure, peripheral vascular disease, end-stage liver disease, diabetes with end-organ damage, renal failure, and dementia) (15–17) during the 2 years before ICU admission and a count of those diagnoses; the APACHE II score during the first 24 hours of the index ICU stay; and the number of days that had elapsed since the index hospitalization and index ICU stay at the time the baseline questionnaire was completed. The other three were still unknown for some patients at the time the family member’s financial hardship was determined, and family participants with unknown values were excluded from the analyses: 21 were excluded for unknown values on the number of days of mechanical ventilation by the time the questionnaire was completed, 50 were excluded for unknown values regarding whether palliative care consultations had occurred by the time of questionnaire completion, and 87 were excluded for unknown values regarding whether goals-of-care discussions (29) had occurred by questionnaire completion.

Several additional variables were used for descriptive purposes only: family participants’ baseline responses to questions assessing their current work status, their difficulty in paying regular household bills, the amount of financial strain they were experiencing, and their assessments of the patient’s frailty (fatigue level, difficulty walking up stairs, unintentional weight loss in the previous year, and the extent of frailty and its impact on quality of life) (30). Descriptive variables were also taken from patients’ medical records: The variables were gender, racial and/or ethnic identification, mortality status at the time of index hospitalization discharge, documentation of the patient’s number of days in the complete index hospitalization and index ICU stay, total days of MV administration, and whether palliative care consultation or goals-of-care discussion occurred during the index stay. We also computed for each patient the number of family participants who provided data for the study.

Statistical Analyses

The standard toxicity construct, developed with data from patients with cancer, comprises 11 Likert-scaled items found by the instrument developers to represent a single factor based on exploratory factor analysis (EFA) (10). In this study, we conducted preliminary analyses using EFA with the 11 indicators defined as ordered categorical variables and estimated with weighted least squares with mean and variance adjustment, both with our full sample of family respondents and with those whose associated patient had been diagnosed with poor-prognosis cancer. Our data did not produce, in either group, any EFA solutions with either a single factor or multiple factors having acceptable fit, based on the chi-square test of fit. In addition, we were unable to locate a subset of the 11 modified FACIT-COST items that measured a unidimensional financial toxicity construct with invariant meaning between subgroups or over time. However, to provide information about the practical utility of the traditional summed toxicity score as a predictor of expected outcomes, we used a financial hardship measure derived from it for analyses included in this report.

We first tested associations between the 32 predictors and financial hardship, defined as a linear outcome. These analyses were based on clustered linear regression, estimated with robust maximum likelihood. After examining bivariate associations between each predictor and the financial hardship score, we ran a multiple regression model containing all predictors that had P < 0.20 in the bivariate models

In addition, we tested associations between financial hardship as measured at 1 month and four preselected outcomes. Two were composite measures from the Hospital Anxiety and Depression Scale, with analysis based on clustered (family respondents within patients) linear regression with robust maximum likelihood estimation. The other two were Likert-scale outcomes, defined as ordered categorical variables, with analysis based on clustered probit regression and estimated with weighted least squares with mean and variance adjustment. We conducted these analyses two ways. First, we used the full sample and the values of both financial hardship and the outcomes as measured at 1 month, acknowledging that there is no way to determine the actual direction of causation between the predictor and “outcome” when the two variables are measured at the same time point. For these models, we adjusted the association for the family member’s age, gender, and racial and/or ethnic minority status, and for the patient’s APACHE II score. In addition, we conducted the analyses using the value of financial hardship at 1 month as a predictor of the outcome as measured at 3 months. For the analyses involving the 3-month outcome, we added the respondent’s value on the outcome at 1 month as a covariate. Because the FCS2 study involved randomization of participants, we excluded from the 3-month analyses FCS2 participants who were randomized to the intervention condition. For both the 1-month and 3-month analyses, we examined results for the full sample and also for two groups: family members of patients who were alive and those of patients who had died when the outcome was measured.

We used Mplus software (https://www.statmodel.com/) for all factor analysis and regression modeling. Descriptive information for family participants and their associated patients was produced with IBM SPSS Statistics (https://www.ibm.com/products/spss-statistics).

Results

Sample Characteristics

Out of the 927 family participants in the studies providing data for this report, we computed financial hardship for 783 (84%), by virtue of their having a value for at least one of the 11 modified FACIT- COST items on the 1-month survey. The 783 participants had characteristics similar to those of the full sample of family participants in the two studies (Table 1). Almost 70% were female. The median age was 57, ranging from 19 to 92 years. More than 75% were White, non-Hispanic. Most (over 80%) had some post–high school education. Almost 40% were the patient’s spouse or partner, with an additional 24% being a child of the patient. More than half lived with the patient. About 30% had dependent children at home. More than 65% either worked full time or were retired, and over 20% had experienced changes in their work status related to the patient’s illness. The 783 family participants in our analyses represented 715 unique patients (Table 2). Distributions of financial hardship and family-reported outcomes are presented in Table 3.

Table 1.

Characteristics of family participants

Characteristic	Valid Financial Hardship Score*		Total Sample†
	Valid n‡	Statistic§	Valid n‡	Statistic§
Family participants, n‖	783	—	927	—
Age, median (IQR)	745	57 (20)	885	55 (21)
Female, n (%)	770	536 (69.6)	913	635 (69.6)
Racial and/or ethnic group, n	766	—	905	—
White non-Hispanic, n (%)	—	589 (76.9)	—	682 (75.4)
Racial and/or ethnic minority, n (%)	—	177 (23.1)	—	223 (24.6)
Asian	—	28 (3.7)	—	32 (3.5)
Black	—	37 (4.8)	—	50 (5.5)
Hispanic (White)	—	47 (6.1)	—	57 (6.3)
Indigenous	—	19 (2.5)	—	26 (2.9)
Pacific Islander	—	5 (0.7)	—	6 (0.7)
Biracial or multiracial¶	—	41 (5.4)	—	52 (5.7)
Education level, n	774	—	918	—
Eighth grade or less, n (%)	—	2 (0.3)	—	2 (0.2)
Some high school, n (%)	—	25 (3.2)	—	31 (3.4)
High school diploma or equivalent, n (%)	—	115 (14.9)	—	135 (14.7)
Trade school or some college, n (%)	—	304 (39.3)	—	372 (40.5)
Four-year college degree, n (%)	—	158 (20.4)	—	192 (20.9)
Some graduate school, n (%)	—	36 (4.7)	—	41 (4.5)
Graduate or professional degree, n (%)	—	134 (17.3)	—	145 (15.8)
Relationship to patient, n	778	—	922	—
Spouse or partner, n (%)	—	305 (39.2)	—	349 (37.9)
Patient’s child, n (%)	—	183 (23.5)	—	224 (24.3)
Sibling, n (%)	—	103 (13.2)	—	121 (13.1)
Patient’s parent, n (%)	—	145 (18.6)	—	175 (19.0)
Friend, n (%)	—	9 (1.2)	—	11 (1.2)
Other relationship, n (%)	—	33 (4.2)	—	42 (4.6)
Lived with patient, n (%)	774	410 (53.0)	917	470 (51.3)
Had dependent children at home, n (%)	776	226 (29.1)	920	286 (31.1)
Current work status, n	691	—	693	—
Employed full-time, n (%)	—	261 (37.8)	—	261 (37.7)
Employed part-time, n (%)	—	80 (11.6)	—	80 (11.5)
Unemployed, n (%)	—	65 (9.4)	—	65 (9.4)
Disabled, n (%)	—	34 (4.9)	—	34 (4.9)
Retired, n (%)	—	193 (27.9)	—	195 (28.1)
Other, n (%)	—	58 (8.4)	—	58 (8.4)
Any work status change related to the patient’s illness, n (%)	769	164 (21.3)	771	165 (21.4)
Extent current health was poor, median (IQR)**	772	2 (1)	774	2 (1)
Lacked emotional support, n (%)	738	64 (8.7)	740	64 (8.6)
Extent of difficulty paying bills, median (IQR)††	757	1 (2)	759	1 (2)
Extent of illness-related financial strain, median (IQR)‡‡	762	2 (2)	771	2 (2)
Respondent-assessed patient frailty
Frequency of fatigue in the past 4 wk, median (IQR)§§	566	3 (2)	680	3 (2)
Any difficulty in walking up stairs, n (%)	578	329 (56.9)	703	389 (55.3)
Any unintentional weight loss in past yr, n (%)	571	264 (46.2)	697	307 (44.0)
Extent of frailty, median (IQR)‖‖	640	2 (2)	781	2 (2)
Impact of frailty on quality of life, median (IQR)¶¶	634	3 (3)	775	3 (3)

Definition of abbreviations: ICU = intensive care unit; IQR = interquartile range; QUAL-E (Fam) = quality of family experience questionnaire.

^*Family participants for whom a valid financial hardship score could be computed 1 month after the patient’s ICU admission.

^† All family participants who completed questionnaires for either of the two included studies.

^‡ Number in the group named in the column for whom we had valid data for the characteristic named in the row.

^§ Unless otherwise specified, number of participants (and percentage with valid data) having the characteristic.

^‖ The total number of family participants in the sample named in the column.

^¶ Includes combinations of minority groups, as well as combinations of White non-Hispanic and one or more minority categories.

^**Median is based on responses to the question “In general, how would you describe your health?” with the following response codes: 0 (excellent), 1 (very good), 2 (good), 3 (fair), and 4 (poor).

^†† Median is based on the following response codes: 0 (after paying bills, I still have enough money for special things), 1 (I have enough money to pay the bills but little spare money to buy special things), 2 (enough to pay bills, but only because I have cut back on things), and 3 (difficulty paying the bills, no matter what I do).

^‡‡ Median is based on responses to the QUAL-E (Fam) question “How much of the time do you feel financial strain related to [the patient’s] illness?” with the following response codes: 0 (none of the time), 1 (only a little of the time), 2 (some of the time), 3 (a good bit of the time), and 4 (a great deal of the time).

^§§ Median is based on the question “How much of the time during the past 4 weeks did [the patient] feel tired?” with the following response codes: 0 (none of the time), 1 (a little of the time), 2 (some of the time), 3 (most of the time), and 4 (all of the time).

^‖‖ Median is based on the question “In your opinion, how frail is [the patient]?” with the following response codes: 0 (not at all frail), 1 (a little frail), 2 (somewhat frail), 3 (very frail), and 4 (extremely frail).

^¶¶ Median is based on the question “In your opinion, how much has frailty interfered with [the patient’s] quality of life?” with the following response codes: 0 (not at all), 1 (only a little), 2 (somewhat), 3 (a good bit), and 4 (a great deal).

Table 2.

Characteristics of unduplicated patients associated with family respondents

Characteristic	Valid Financial Hardship Score*		Total Sample†
	Valid n‡	Statistic§	Valid n‡	Statistic§
Unduplicated patients, n‖	715	—	838	—
Age, median (IQR)	715	60 (24)	838	59 (26)
Female, n (%)	715	282 (39.4)	838	338 (40.3)
Racial and/or ethnic group, n	688	—	801	—
White non-Hispanic, n (%)	—	558 (81.1)	—	642 (80.1)
Racial and/or ethnic minority, n (%)	—	130 (18.9)	—	159 (19.9)
Asian	—	26 (3.8)	—	26 (3.2)
Black	—	32 (4.7)	—	44 (5.5)
Hispanic (White)	—	32 (4.7)	—	43 (5.4)
Indigenous	—	13 (1.9)	—	17 (2.1)
Pacific Islander	—	5 (0.7)	—	6 (0.7)
Biracial or multiracial¶	—	22 (3.2)	—	23 (2.9)
Insurance coverage, n	715	—	838	—
Private or commercial, n (%)	—	238 (33.3)	—	270 (32.2)
Medicare, n (%)	—	321 (44.9)	—	364 (43.4)
Medicaid, n (%)	—	134 (18.7)	—	176 (21.0)
Military, n (%)	—	10 (1.4)	—	14 (1.7)
Other insurance, n (%)	—	12 (1.7)	—	14 (1.7)
Number of inpatient stays in yr before index ICU admission, median (IQR)	715	0 (0)	838	0 (0)
Number of EHR-documented conditions, median (IQR)	715	1 (2)	838	1 (2)
Diagnoses documented during 24 mo before ICU admission, n (%)	715	—	838	—
Poor prognosis cancer	—	102 (14.3)	—	113 (13.5)
Chronic pulmonary disease	—	102 (14.3)	—	118 (14.1)
Coronary artery disease	—	119 (16.6)	—	135 (16.1)
Heart failure	—	169 (23.6)	—	193 (23.0)
Peripheral vascular disease	—	55 (7.7)	—	64 (7.6)
End-stage liver disease	—	74 (10.3)	—	83 (9.9)
Diabetes with end-organ damage	—	57 ( 8.0)	—	64 (7.6)
Renal failure	—	117 (16.4)	—	130 (15.5)
Dementia	—	40 (5.6)	—	45 (5.4)
APACHE II score, median (IQR)	561	25 (10)	663	25 (10)
Days in hospital (index hospitalization), median (IQR)	715	21 (25)	838	21 (25)
Days in intensive care (index hospitalization), median (IQR)	715	8 (10)	837	8 (11)
Days of MV (index hospitalization), median (IQR)	715	3 (8)	837	4 (8)
Any palliative care consultation, index hospitalization, n (%)	715	174 (24.3)	837	207 (24.7)
Any goals-of-care discussion, index hospitalization, n (%)	715	451 (63.1)	837	527 (63.0)
Died during index hospitalization, n (%)	715	193 (27.0)	838	234 (27.9)
Number of associated family respondents, n	715	—	838	—
1, n (%)	—	655 (91.6)	—	761 (90.8)
2, n (%)	—	52 (7.3)	—	67 (8.0)
3, n (%)	—	8 (1.1)	—	8 (1.0)
4, n (%)	—	0 (0.0)	—	2 (0.2)

Definition of abbreviations: APACHE II = Acute Physiology and Chronic Health Evaluation; EHR = electronic health record; ICU = intensive care unit; IQR = interquartile range; MV = mechanical ventilation; QUAL-E (Fam) = quality of family experience questionnaire.

^*Patients associated with family respondents for whom a valid financial hardship score could be computed 1 month after the patient’s ICU admission.

^† Patients associated with family participants who completed questionnaires for either of the two included studies.

^‡ Number in the group named in the column for whom we had valid data for the characteristic named in the row.

^§ Unless otherwise specified, number of patients (and percentage with valid data) having the characteristic.

^‖ The total number of patients in the sample named in the column.

^¶ Includes combinations of minority groups, as well as combinations of White non-Hispanic and one or more minority categories.

Table 3.

Distribution of financial hardship scores and family-reported outcomes

Outcome	Valid n	Median (IQR)	Range
Financial hardship, baseline*	783	16 (17.05)	0–44
HADS depression score†
Baseline	783	7 (6.5)	0–21
Follow-up	575	6 (6)	0–21
HADS anxiety score‡
Baseline	782	9 (6)	0–21
Follow-up	575	8 (7)	0–21
Lower emotional preparedness§
Baseline	783	2 (1)	0–4
Follow-up	478	2 (1)	0–4
Poorer current quality of life‖
Baseline	783	1 (1)	0–3
Follow-up	573	1 (0)	0–3

Definition of abbreviations: FACIT-COST = modified version of the Comprehensive Score for financial Toxicity; HADS = Hospital Anxiety and Depression Scale; IQR = interquartile range.

^*Computed as 44− (sum of 11 modified FACIT-COST items, weighted for the number of items with valid responses); scores ranged from 0 (lowest financial hardship) to 44 (highest financial hardship).

^† Traditional HADS depression scale score, weighted for the number of valid responses; missing if there were fewer than four valid responses.

^‡ Traditional HADS anxiety scale score, weighted for the number of valid responses; missing if there were fewer than four valid responses.

^§ Responses to the question “How prepared are you emotionally for the future, no matter what happens?” (0 = great deal, 1 = good bit, 2 = somewhat, 3 = only a little, and 4 = not at all).

^‖ Responses to the question, “How would you rate your overall quality of life?” (0 = excellent, 1 = good or very good, 2 = fair, and 3 = poor or very poor).

Associations of Financial Hardship with Preselected Potential Predictors

Of the 32 variables preselected as potential predictors of financial hardship, 21 had P < 0.20 in bivariate models (see Table E1) and were included in a multiple regression model (Table 4). These 21 variables explained 39.5% of the variance in hardship. Of the 21 predictors, 14 had strong independent associations with the score (P < 0.05). Nine were variables collected from family respondents. Reported hardship was substantially lower among older respondents (P = 0.019), those with higher levels of education (P = 0.034), and those who were not spouses or partners of the patient (P = 0.016). Reported hardship was notably higher among family participants who lived with the patient (P = 0.018), had dependent children at home (P < 0.001), had experienced changes in work status related to the patient’s illness (P < 0.001), had poorer health (P < 0.001), or lacked emotional support (P = 0.001). Hardship levels also differed considerably (P = 0.003) between racial and/or ethnic groups, with Hispanic and Pacific Islander respondents reporting greater hardship than those from other racial and/or ethnic groups. Five characteristics of the respondent’s associated patient had independent associations with financial hardship: There was less reported hardship when the patient was in the randomized trial (P = 0.006), died before the family member completed the questionnaire (P = 0.005), or had a cancer diagnosis (P = 0.030); and hardship was greater if the patient had been in the hospital longer (P = 0.011) when the respondent’s questionnaire was completed. In addition, the patient’s insurance coverage was independently associated with financial hardship (P = 0.045 for differences between insurance types), with those covered by private or commercial insurance reporting the greatest hardship and those covered by military insurance reporting the least.

Table 4.

Association of predictors with financial hardship

Predictor	Multipredictor Model
	β	P	95% CI
Family-reported predictors
Age	−0.110	0.019	−0.201, −0.018
Racial and/or ethnic identity*	—	0.003	—
White	0.000	—	—
Asian	−1.482	0.534	−6.155, 3.191
Black	−0.805	0.595	−3.776, 2.165
Hispanic	3.231	0.032	0.277, 6.186
Indigenous	−2.859	0.230	−7.522, 1.805
Pacific Islander	11.717	<0.001	5.301, 18.134
Biracial or multiracial	1.930	0.289	−1.634, 5.495
Education†	−0.582	0.034	−1.121, −0.043
Relationship to patient*	—	0.016	—
Spouse or partner	0.000	—	—
Patient’s child	−4.340	0.011	−7.687, −0.994
Sibling	−5.076	<0.001	−7.897, −2.255
Patient’s parent	−3.605	0.045	−7.130, −0.079
Friend	−4.437	0.070	−9.229, 0.355
Other relationship	−3.131	0.069	−6.500, 0.238
Lived with patient	2.712	0.018	0.471, 4.953
Had dependent children at home	3.055	<0.001	1.385, 4.726
Any illness-related work status change	3.764	<0.001	1.691, 5.837
Extent current health was poor‡	3.717	<0.001	2.888, 4.546
Lacked emotional support	4.178	0.001	1.807, 6.550
Patient predictors
In randomized trial	−2.752	0.006	−4.709, −0.796
In intervention group	0.141	0.898	−2.010, 2.292
Died before family questionnaire	−2.580	0.005	−4.397, −0.763
Patient age	0.020	0.661	−0.070, 0.110
Insurance coverage*	—	0.045	—
Private or commercial	0.000	—	—
Medicare	−1.048	0.294	−3.006, 0.911
Medicaid	−1.201	0.240	−3.205, 0.804
Military	−8.676	0.005	−14.717, −2.635
Other insurance type	−3.203	0.238	−8.526, 2.121
Specific diagnoses
Cancer	−2.268	0.030	−4.321, −0.215
Heart failure	−0.195	0.822	−1.902, 1.511
Peripheral vascular disease	0.063	0.961	−2.470, 2.596
End-stage liver disease	0.221	0.859	−2.229, 2.671
Days in hospital before family questionnaire, index stay	0.077	0.011	0.017, 0.136
Palliative care consultation before family questionnaire, index stay§	−0.458	0.640	−2.375, 1.460
Goals-of-care discussion before family questionnaire, index stay§	1.055	0.255	−0.762, 2.871

Definition of abbreviations: CI = confidence interval.

n = 605. R²= 0.395. Results are based on a clustered linear regression model (family participants clustered within patients) estimated with robust maximum likelihood. The model includes all predictors from bivariate associations with associated P < 0.20 (see the data supplement).

The bold-faced items are those that have P-values less than 0.05.

^*The P value for the full set of predictor categories was based on Wald’s test of parameter constraints.

^† Education was modeled as an ordinal variable: 0 = high school diploma or less, 1 = some post–high school education, 2 = 4-year college degree or some graduate school, and 3 = graduate degree.

^‡ Current health was modeled as an ordinal variable: 0 = excellent, 1 = very good, 2 = good, 3 = fair, and 4 = poor.

^§ Respondents were excluded if the consultation and/or discussion occurred at some point during the index hospitalization but the stay was still in progress at the time of completion of the 1-month questionnaire, so we could not determine whether the event had preceded questionnaire completion.

Associations of Financial Hardship with the Four Outcomes

Regression models adjusted for preselected covariates showed strong associations of financial hardship with all four family-reported outcomes at both 1 month and 3 months after the patient’s ICU admission, with greater hardship associated with more negative outcomes (Table 5).

Table 5.

Associations between financial hardship and family-reported outcomes

Outcome	Valid n	β	P	95% CI	R 2 *
Outcome measured 1 mo after ICU admission†
Total sample
HADS depression score‡	572	0.215	<0.001	0.187, 0.244	0.299
HADS anxiety score§	572	0.212	<0.001	0.183, 0.241	0.325
Less emotionally prepared for the future‖	571	0.040	<0.001	0.032, 0.049	0.181
Poorer quality of life¶	568	0.053	<0.001	0.044, 0.063	0.257
Patient alive at time of interview
HADS depression score‡	410	0.215	<0.001	0.184, 0.247	0.314
HADS anxiety score§	410	0.216	<0.001	0.182, 0.249	0.325
Less emotionally prepared for the future‖	410	0.042	<0.001	0.032, 0.052	0.207
Poorer quality of life¶	408	0.057	<0.001	0.046, 0.068	0.290
Patient died before interview
HADS depression score‡	162	0.233	<0.001	0.170, 0.295	0.311
HADS anxiety score§	162	0.196	<0.001	0.135, 0.257	0.323
Less emotionally prepared for the future‖	161	0.046	<0.001	0.028, 0.063	0.195
Poorer quality of life¶	160	−0.051	<0.001	0.031, 0.070	0.241
Outcome measured 3 mo after ICU admission**
Total sample
HADS depression score‡	264	0.094	<0.001	0.057, 0.131	0.586
HADS anxiety score§	264	0.100	<0.001	0.056, 0.144	0.607
Less emotionally prepared for the future‖	233	0.028	0.001	0.012, 0.045	0.365
Poorer quality of life¶	260	0.025	0.001	0.010, 0.040	0.434
Patient alive at time of interview
HADS depression score‡	169	0.093	<0.001	0.046, 0.139	0.581
HADS anxiety score§	169	0.100	<0.001	0.051, 0.150	0.596
Less emotionally prepared for the future‖	168	0.032	0.001	0.012, 0.051	0.435
Poorer quality of life¶	166	0.017	0.094	−0.003, 0.036	0.452
Patient died before interview
HADS depression score‡	95	0.077	0.021	0.011, 0.143	0.616
HADS anxiety score§	95	0.079	0.080	−0.009, 0.167	0.627
Less emotionally prepared for the future‖	65	0.010	0.677	−0.039, 0.059	0.253
Poorer quality of life¶	94	0.038	0.023	0.005, 0.071	0.461

Definition of abbreviations: APACHE II = Acute Physiology and Chronic Health Evaluation; FCS2 = Facilitating Communication Study; HADS = Hospital Anxiety and Depression Scale; ICU = intensive care unit; MLR = robust maximum likelihood; WLSMV = weighted least squares with mean and variance adjustment.

Regression models clustered family respondents within patients, and all included adjustment for the respondent’s age, gender, and racial–ethnic minority status and for the patient’s APACHE II score. Financial hardship was treated as a continuous linear predictor. The HADS depression and anxiety outcomes were defined as continuous variables and analyzed with linear regression, estimated with MLR. The other outcomes were defined as ordered categorical variables and analyzed with probit regression, estimated with WLSMV.

^*Proportion of variance in the outcome jointly explained by financial hardship and the covariates.

^† Participants randomized to the intervention condition in the FCS2 study were retained in models predicting 1-month outcomes, as both financial hardship and the outcome included any effects of the portion of the intervention that had occurred by the time of the 1-month assessment.

^‡ The depression score (potential range = 0–21) was computed as the sum of the responses to the seven HADS depression items, weighted by 7 divided by number of valid responses. Respondents with valid answers to fewer than four of the items were excluded from the model.

^§ The anxiety score (potential range = 0–21) was computed as the sum of the responses to the seven HADS anxiety items, weighted by 7 divided by number of valid responses. Respondents with valid answers to fewer than four of the items were excluded from the model.

^‖ Response codes: 0 (great deal prepared), 1 (good bit), 2 (somewhat), 3 (a little), and 4 (not at all prepared).

^¶ Response codes: 0 (excellent), 1 (good or very good), 2 (fair), and 3 (poor or very poor).

^**Models of 3-month outcomes differed in two ways from models of 1-month outcomes. They included an additional covariate (the respondent’s 1-mo value on the outcome); and they excluded participants randomized to the intervention condition in FCS2, owing to potential confounding of the association between financial hardship and the outcome by intervention activities that occurred between 1 and 3 months.

At 1 month, financial hardship and the covariates explained between 18.1% and 32.5% of the variance in the outcomes for the total sample. When the model was stratified by the patient’s mortality status at the time of the 1-month assessment, results were similar for the two strata (Table 5). We also investigated whether the associations of financial hardship with the four outcomes differed for respondents who were family members of patients who had been randomized to receive an intervention and those who were family members of patients who had not. Associations with all four outcomes remained positive, with P values <0.001 in both groups (data not shown).

At 3 months, with the exclusion of respondents who were assigned to the intervention group, financial hardship continued to have strong independent positive associations with the 3-month outcomes after adjustment for the 1-month outcome and the other covariates (Table 5). Adjustment for the outcome value at 1 month understandably reduced the coefficients for financial hardship in the total sample (although P values were ⩽0.001 for all outcomes), and the set of covariates explained larger percentages of variance in the outcomes than was the case with the 1-month outcome, explaining 36.5–60.7% of the variance. In models stratified by the timing of the 3-month assessment (before or after the patient’s death), all of the associations remained positive, although some of the P values were >0.05, owing to both decreases in coefficient strength and reduced sample sizes.

Discussion

In this large, prospective longitudinal study of family of critically ill patients, we found that living with the patient, being the spouse or partner of the patient, having dependent children at home, younger age, lacking emotional support, poorer health and work status changes as a result of the patient’s illness were associated with higher levels of financial hardship. Additionally, family financial hardship levels differed by racial and ethnic groups and education level. It is interesting that only three patient characteristics were associated with family financial hardship: A longer hospital stay was associated with higher family financial hardship, and family of patients who died before questionnaire completion or with a cancer diagnosis had lower levels.

These findings suggest that, for the families of critically ill patients, social and demographic factors may play a larger role in experiencing financial hardship than patient characteristics related to disease characteristics, severity of illness, and intensity of care. Although this finding conflicts with reports from other illnesses where patient clinical factors were a key driver of financial hardship (31–34), our findings are consistent with prior studies among critically ill patients (8). Our results suggest that clinical measures may not reliably predict who experiences financial hardship (8) and that, notably, social and demographic factors cannot be ignored as key determinants of outcomes for patients with critical illness and their families.

The benefits of post-ICU support groups and peer support groups for the families of critically ill patients have been increasingly recognized and studied (35–37). However, data are lacking on the potential role and benefit these groups may have in addressing financial hardship and who may be in most need of this added support. Even in countries with publicly funded healthcare systems, informal family caregivers of patients with cancer have highlighted the need for family support as a strategy to mitigate financial toxicity (38), suggesting that financial hardship encompasses more than out-of-pocket costs and may be modifiable.

We found it interesting that patients with cancer and their families reported experiencing less financial hardship than did patients with other life-limiting, chronic illnesses. This may be due, in part, to the fact that these families may have received financial guidance in the outpatient setting, at the time of the patient’s diagnosis. As a result of the adverse health consequences caused by financial hardship, the National Cancer Institute and the American Society of Clinical Oncology have recommended that interventions to decrease financial hardship be urgently designed, tested, and incorporated into high-quality cancer care (39). Promising interventions in the outpatient oncology setting have been developed and successfully piloted. Examples include interventions designed to increase financial literacy for patients with cancer and integrating trained financial navigators into the outpatient oncology care team (40–43). Although the outpatient oncology setting is very different than the acuity of the ICU and the often less predictable course characterizing other serious illnesses, similar types of interventions may be effective in reducing financial hardship for family members and may warrant further investigation.

Last, our findings demonstrated a strong association between financial hardship and subsequent family member anxiety, depression, and quality of life, regardless of whether the respondent’s associated patient had died before questionnaire completion. These findings are consistent with those of previous studies demonstrating a mediating role of financial hardship in these outcomes that are of great importance to the critical care community (8). This once again highlights that addressing financial hardship may be a key piece to improving family-centered outcomes for this vulnerable population (44). Interventions targeted at improving support systems for family members of patients who experience critical illness during and after an ICU stay require additional, rigorous research.

Our study has several notable strengths. First, to our knowledge, this is the largest prospective longitudinal study in family members of critically ill patients that specifically examined the risk factors and outcomes of financial hardship, using a measure that captures the multidimensional construct of financial hardship. Key components of financial hardship include both objective financial burden (i.e., out-of-pocket costs) and subjective financial burden (i.e., worries about real or perceived financial hardship) (45, 46). Second, we did not limit our participants to the family members of survivors of critical illness, but we also included family members of decedents, providing a more comprehensive assessment of family burden. Last, our broad scope of predictors was intentionally designed to capture patient-level factors, along with family social and demographic factors.

Our study also has several limitations. First, participants were recruited from one health system in the Pacific Northwest, which may limit the generalizability of findings. Second, our study design, which incorporated data from two studies, presented some unmeasured differences, although the intervention itself in the randomized controlled trial was not significantly associated with the extent of financial hardship. However, we feel that the ability to efficiently accrue such a large sample size outweighed potential biases introduced by combining data collection efforts. Third, our data collection period overlapped with the height of the coronavirus disease (COVID-19) pandemic, and additional background or economic factors may have influenced the findings (e.g., changes in work status due to patient’s illness). Finally, we note that this study was an exploratory one, representing an early attempt to investigate levels of financial hardship, and the potential precursors, correlates, and outcomes of this hardship, among the families of seriously ill patients. This exploration involved a very large number of inferential tests, and we have tried to avoid imputing formal “statistical significance” to the results, given the risks associated with multiple comparisons. Nonetheless, we have reported P values throughout the report, and we urge readers to use caution in interpreting their significance, in light of these considerations.

Conclusions

For the families of critically ill patients, social and demographic factors may play a larger role in experiencing financial hardship than patient clinical factors. It is important to note that financial hardship is independently associated with key family-reported outcomes, including anxiety, depression, emotional preparedness, and quality of life. These findings add to the growing literature demonstrating the prevalence and adverse consequences of financial hardship, the inequitable distribution of this hardship, and the need for evidence on supports and interventions to mitigate poor outcomes and inequities.