A dyadic analysis of financial toxicity and health‐related quality of life among bone marrow transplant patients and their caregivers

Relatively few dyad‐based studies have evaluated the shared psychosocial and financial toxicity (FT) experiences of hematologic patients and their caregivers, especially those undergoing bone marrow transplantations (BMTs). This study evaluated the association of FT with health‐related quality of life (QOL) among BMT patient‐caregiver dyads.


| BACKGROUND
[12] Caregivers are especially vulnerable to experiencing FT when their financial circumstances are closely connected with those of the patient (i.e., spouse or parent). 13Caregivers of hematologic cancer patients report taking time off work, quitting or changing jobs, loss of income and savings, borrowing money, and difficulty paying bills. 14,15regivers also report neglecting their own physical and mental health to cope with the financial burden related to their loved one's cancer treatment. 11Despite this considerable caregiving burden, relatively few dyad-based studies have demonstrated the shared psychosocial experiences of hematologic patients and their caregivers, especially those undergoing BMTs.Even fewer assess the dyadic experiences of FT in this population.Understanding the individual and shared FT experiences of both patients and caregivers is key to shaping interventions to alleviate FT.
In this paper, we evaluated the association of demographics and FT with health-related QOL among patient-caregiver dyads who received BMTs (allogeneic and autologous) while participating in an oncology financial navigation (OFN) intervention consisting of oneon-one assistance with navigating costs of care from a financial navigator. 16We hypothesized that there would be an inverse relationship between FT and QOL for both patients and caregivers.

| METHODS
This study was part of a larger single-arm feasibility and acceptability trial of an OFN intervention conducted in the Division of Hematology and BMT at an NCI-designated Cancer Center (NCT05465577). 16 used patient electronic health record data and patient/caregiver survey data from participants who provided written informed consent.The study was approved by the University of Kentucky's Institutional Review Board (#61653).
Methods for this study have been described. 16Briefly, we screened patients 18þ years old with a hematology and/or BMT appointment for FT between April 2021 and January 2022.Eligible patients: screened positive for FT (indicated by Comprehensive Score for Financial Toxicity (COST) ≤ 24 17 and/or the National Comprehensive Cancer Network's (NCCN) Distress Thermometer (DT) scores >4 with selection of financial/insurance issues on DT problem list 17 ); were able to read/write in English; had a participating caregiver; and were consented by the financial navigator during inperson hospital/clinic visits.T1 surveys were collected during a hospital/clinic visit prior to participation in the OFN intervention and T2 surveys were completed post-intervention.

| Measures
Self-reported or EHR-extracted demographic variables are shown in Table 1.Employment status and time since diagnosis were dichotomized for ease of presentation.While most demographic factors were measured for each participant, time since diagnosis, caregiver relationship to patient, and caregiver time off work were dyad-level as they applied to both patient and caregiver.

| Independent variables
Financial Toxicity was measured in three domains2 : (1) psychological response which measures emotional response using the 11-item COST with lower scores indicating greater FT, 18 (2) material conditions which measures direct costs of cancer care (i.e.debt, borrowing money) using 8 items from the Medical Expenditure Panel Survey-Experiences with Cancer Survivorship Supplement (MEPS-ECSS) and two from the demographic survey, 19 and, (3) coping behaviors which measures foregoing care because of financial circumstances (i.e.skipping medications or foregoing treatment) was measured using a series of 16 yes/no MEPS-ECSS 20 (see main outcomes paper for scoring details). 16The COST was adapted for caregivers phrasing questions related to their loved one's cancer treatment. 21A total FT score was determined by (1) reverse-scoring COST; (2) dividing each of the three FT domain scores by the corresponding maximum score for that measure to obtain the proportion; and (3) summing the proportions.

| Dependent variables
Health-related QOL was measured using four Patient-Reported Outcomes Measurement Information System (PROMIS) scales: Physical health and Mental health subscales (PROMIS Scale v1.2 -Global Health), 22 Anxiety Short Form, 23 and Depression Short Form. 24CN's DT was used to measure distress. 17The scoring for the physical/mental health scales is such that higher scores are more positive, while for the remaining instruments, higher scores indicate greater mental distress.

| Data analysis
Descriptive statistics were used to summarize the study variables.
The comparisons of patients and caregivers at T1 (pre-intervention) were accomplished using paired t-tests, with differences calculated as the patient minus caregiver score so that a negative average difference indicated that caregivers tended to score higher.Tests of association for T1 outcomes between patients and caregivers within the dyads were based on Pearson's correlation.
Given the dyadic structure of the study design, the effect of FT on the dependent variables was evaluated using Actor-Partner Interdependence Modeling (APIM), 25 with separate models for each of the PROMIS measures and distress.This regression strategy allows for dependence among subsets of participants (i.e., the members of each patient-caregiver pair lack independence by virtue of their relationship) by nesting participants within dyads when fitting the models.As shown in Figure 1, the dyadic nature of both the independent and dependent variables allows for the estimation of both  work is common among caregivers of cancer patients).The second model included these variables and the actor and partner values for total FT score.We could not use additional demographics or the individual domain FT scores since the sample size allowed only six regressors. 26e APIM regressions were run using the SAS MIXED procedure, with participants nested within dyads and two responses for the dependent variable (T1, T2).Overall model significance was evaluated with the Chi-square likelihood ratio test, model fit was estimated using the Akaike Information Criterion (AIC), and the presence of multicollinearity was examined by estimating variance inflation factors (VIFs).Evaluation of the VIFs for both sets of models confirmed that all were <2, suggesting multicollinearity is unlikely to have distorted any of the parameter estimates.The sample size of at least 30 patient-caregiver dyads was chosen a priori so that the paired t-tests would have at least 85% power to detect an effect size that was slightly larger than medium (i.e., 0.6) 26 , and so that the regression that underlies the APIM analysis would have 85% power to detect an R 2 as small as 0.18 (which is close to the medium effect size of R 2 = 0.13), 26 both at a 0.05 alpha level.Analysis was done using SAS, v. 9.4.
The majority of patients and caregivers identified as female with a mean age of 47 years; nearly all were White/non-Hispanic (see All measures were significantly, positively correlated between patients and caregivers except for FT-coping behaviors and physical health QOL (Table 2).The dyad members had mean scores that were similar for FT-material conditions and FT-coping  Based on the caregiver score subtracted from the patient score so that a negative average difference indicates that caregivers tended to score higher. b The FT psychological response was measured using the COST scale; scoring is such that a lower score indicates higher FT.

| Actor and partner interdependence modeling
The multilevel repeated measures analysis to evaluate demographic factors without (Model 1) and with (Model 2) the inclusion of actor and partner effects of total FT scores as potential predictors of physical/mental QOL outcomes and distress are shown in Table 3.
Across the top, the QOL outcomes are labeled A through E, and each column represents the Model 1 and Model 2 that were estimated for that particular outcome.The chi-square likelihood ratio test was significant for all 10 models, suggesting that each was significant overall.Further, the Akaike Information Criterion (AIC) demonstrated that compared with the models that contained only demographic variables, the ones that also included the actor and partner effect of total FT scores constituted an improvement; the minimum decrease in AIC from the partial to full model was 43.7, well above the improvement threshold of ≥2.Thus, the findings summarized here are from Model 2 models only, which are shaded in gray in Table 3.
For each of the five outcomes, the actor effect of total FT was significant (see the Model 2 row, columns A-E in Table 3).For physical and mental health QOL (columns A and B), among both patients and caregivers, participants with higher total FT values tended to report their own physical and mental health as poorer.
From columns C-E, participants (i.e., patients and caregivers) with higher FT values tended to report their own anxiety, depression, and distress as higher.
There was not a significant partner effect for total FT for any of the five outcomes.That is, their partner's reported total FT score was not predictive of a participant's health-related QOL by any of the measures (A-E), and this was true for the combined group of patients and caregivers.
From the physical health model (A), the patient indicator was significant; patient scores for outcome A were lower than caregiver scores, on average.Time was significant in the mental health QOL model (B); compared with T1, T2 scores for this outcome were Indicator for time −0.785 0.9 (0. -5 of 8 The shared experiences of FT and its effects on health-related QOL among patient-caregiver dyads have not been well-characterized. Our research begins to address this gap by examining these associations using actor-partner modeling.Both patients and caregivers in this study experienced high FT; however patients reported poorer physical and mental health compared to caregivers.Actor effects were found for total FT where patients and caregivers who reported higher total FT scores (indicating greater financial distress) reported poorer physical and mental health, and higher anxiety, depression, and distress.Members of dyads whose caregiver took time off work had lower depression and distress than those whose caregiver did not.In this study, we did not find any partner effects for FT.
8][29] In this study, patient and caregiver COST, FT-material conditions, distress, mental health QOL, anxiety, and depression scores were significantly correlated.Findings are consistent with existing studies on psychosocial distress and FT levels of hematologic cancer patients and contributes to the gaps in literature on caregiver experiences of FT. [7][8][9][10]30,31 Patients and caregivers experiencing higher FT also experienced lower physical and mental QOL and higher rates of anxiety, depression and distress. High chological distress among BMT patients has been well documented and is associated with poor treatment compliance and outcomes.32 Contributing to this distress are the high costs of care with average cumulative costs for hematologic cancer treatment ranging from $200,000 to $800,000 (USD) in the first 36 months of treatment.

| Study limitations
This study was limited by a small sample size and predominantly non-Hispanic White sample.Future studies in this area will benefit from the inclusion of a larger, more diverse sample of patient-caregiver dyads, which will allow for the inclusion of the different domains of FT in the modeling, rather than the total score across domains.A larger sample size will also allow for the inclusion of a greater number of personal and clinical covariates, which will increase the robustness and generalizability of the findings.

| Clinical implications
Our study findings have several key clinical implications.Screening both patients and caregivers for psychosocial distress, specifically FT along the cancer care continuum, is key to early identification and intervention. 36Engaging in patient/caregiver-provider cost of care conversations is another best practice approach to help identify financial needs and build rapport and trust. 37,38If available, patients and caregivers should be connected with OFN services and/or social workers.Healthcare team members must be knowledgeable about connecting patients/caregivers to financial assistance resources, both internal and external to the health system.

| CONCLUSIONS
Our study addresses a significant gap in dyad-based studies, with a focus on the shared psychosocial and FT experiences of hematologic patients and their caregivers, especially those undergoing BMTs.We found that experiencing FT increases both patient and caregiver risk for poor physical and mental health outcomes.Findings underscore the need for further research to establish the individual and shared FT experiences of both patients and their caregivers as a precursor to the development of dyadic FT interventions.
actor and partner effects, using the APIM strategy.The actor effect is the estimate of the relationship of each participant's FT score with their own score on a given outcome measure.The partner effect is the estimate of how much a participant's partner's total FT score influences the participant's own outcome score.An additional feature in this study is that the models are longitudinal, so they evaluate the relationships of T1 demographics and FT with both T1 and T2 outcomes.The modeling was done in two steps: the first included the indicators for Time (1 = T1, 0 = T2) and Patient (1 = Patient, 0 = Caregiver), as well as education (a participant-level variable; this was proxy for socioeconomic status, since education was had fewer missing values than income), and an indicator for whether the caregiver took time off work (a dyad-level variable, included since missed T A B L E 1 Descriptive summary of pre-intervention (T1) demographic variables (N = 34 dyads).

1
Structure of the actor-partner interdependence repeated measures model for patient-caregiver dyads.T A B L E 2 Descriptive summary of pre-intervention (T1) FT indicators, physical and mental health QOL, and DT scores, with paired t-tests and correlations between patients and caregivers (N = 34 dyads).
significantly higher.The dyadic indictor for whether the caregiver had taken time off work was significant for the outcomes of depression (D) and distress (E); for both models, the depression and distress scores were lower, on average, when the caregiver had taken work leave.The studentized residuals for the five models (in the Model 2 section) all had a standard normal distribution, and none were associated with any variables in the model (p > 0.4 for all correlations).While the actor and partner effect of total FT scores are correlated (0.59, p < 0.001), and there is also a positive pointbiserial correlation for the indicator for time off work and total FT score (0.37, p < 0.001), these did not unduly influence estimation of model effects as evidenced by the low Model 2 VIFs (all <1.8).T A B L E 3 Actor-partner interdependence modeling: Repeated measures models for each outcome at pre-(T1) and post-intervention (T2).34 dyads) (N = 34 dyads) (N = 34 dyads) (N = 34 dyads) (N = 34 dyads)

Table 1 )
. Most patients and caregivers had some education beyond a high school diploma/GED.Compared with caregivers, relatively fewer patients were employed and a higher percentage were covered under Medicare or Medicaid.Close to one-third of patients and caregivers were at or below the federal poverty level.Most participants had COST scores that were ≤24.Slightly more patients than caregivers had a NCCN DT score that exceeded 4 and indicated financial/insurance problems on the DT.Over half of the dyads were comprised of married couples.Other caregivers included adult children (26%), friends (12%), or some other person (38%).Over half of the dyads indicated that a caregiver in the dyad had taken paid or unpaid time off to care for the patient.