Sage Journals: Discover world-class research

Abstract

Introduction

Health and psychosocial outcomes are often interdependent for people with cancer and their informal caregivers. However, few studies have examined interdependence among U.S. Black patients and their caregivers. In this prospective study, we analyzed whether Black women with breast cancer and their caregivers demonstrated interdependence in depressive symptoms and other outcomes over time.

Methods

Patient and caregiver data were collected via telephone surveys at three timepoints (baseline, two months, four months). Actor-Partner Interdependence Models (random-intercept cross-panel models in Mplus) were used to analyze depressive symptoms (Center for Epidemiology Studies Depression scale; CES-D), physical and mental health quality of life, perceived social support, and health behaviors. We hypothesized patients and caregivers would demonstrate interdependence for these outcomes both cross-sectionally and over time.

Results

We enrolled 210 participants. All patients and 93% of caregivers identified as Black/African American; 70% of caregivers identified as female. Two-thirds (66%) of dyads did not live together. Mean baseline CES-D score was 15.0 (SD = 10.9) for patients and 11.9 (SD = 9.0) for caregivers. We did not find prospective interdependence for any outcomes. Dyad members were moderately correlated in overall levels (intercepts) of physical health quality of life (r = .25), mental health quality of life (r = .28), social support (r = .37), and health behavior scores (r = .33).

Conclusion

This study is one of the first longitudinal, dyadic examinations of outcomes for Black breast cancer patients and their informal caregivers. Although our hypothesis of prospective interdependence was not supported, we observed moderate correlations of overall levels of several measures. These results suggest that patients with lower social support, poor health behaviors, and worse quality of life may also have caregivers who could benefit from intervention.

Plain Language Summary

Why was the study done? Health and psychological outcomes are often related among people with cancer and their informal caregivers (that is, friends and family who help them deal with cancer). However, there has not been a lot of research looking at these effects among U.S. Black patients and their caregivers. We analyzed whether Black women with breast cancer and their caregivers had related symptoms of depression and other outcomes over time. What did the researchers do? Patient and caregiver data were collected by telephone surveys at three timepoints (baseline, two months, four months). Statistical models were used to analyze symptoms of depression, physical and mental health quality of life, social support, and health behaviors. We predicted these outcomes would be related for patients and caregivers both at individual timepoints and over the study period. What did the researchers find? Of the 210 participants in this study, all breast cancer patients and 93% of caregivers identified as Black/African American; 70% of caregivers identified as female. Two-thirds (66%) of dyads did not live together. We did not find that one person’s score at one time point predicted the other person’s score at the next time point. Members of a pair were somewhat correlated in their overall levels of physical health quality of life, mental health quality of life, social support, and health behavior scores. What do the findings mean? This study is one of the first to analyze outcomes for pairs of Black breast cancer patients and their informal caregivers over time. We found moderate correlations within pairs for several outcomes. These results suggest that patients with lower social support, poor health behaviors, and worse quality of life may also have caregivers who may benefit from additional support.

Keywords

informal caregivers breast cancer depressive symptoms social support interpersonal relationships Black or African American

Introduction

Clinicians and researchers have conceptualized cancer as a “family affair” affecting both patients and the people close to them.^1-3 Intimate partners, family members, and friends often serve as informal caregivers for people with cancer and provide unpaid support along a variety of dimensions (tangible, emotional, and/or social).^2,4,5 Dyadic research that analyzes both patients and informal caregivers or other partners has suggested that, in the context of cancer and other chronic diseases, health and psychosocial outcomes in patients and caregivers are often interdependent.^6-11 A systematic review¹² found that, in heterosexual spousal dyads in which the female partner had breast cancer, depression in one member of the dyad is a risk factor for depression in the other; the authors note, however, that research in more diverse samples is needed. The observation that caregiver mental health is associated with patient mental health is consistent with theories of emotional contagion¹³ and has led to the suggestion that screening and treating spouses of cancer patients for depression may improve patients’ mental health.^9,14

Although dyadic effects in the context of cancer and other chronic diseases are common, Litzelman et al¹⁵ point out that these effects are by no means uniform, with some studies showing influence from patients to caregivers, some showing influence from caregivers to patients, some showing reciprocal influence, and some showing a lack of dyadic influence. The authors suggest that variability in dyadic effects may be due to differences in context and population.¹⁵ Although much of the early work in dyadic cancer research involved spousal dyads,^16,17 more recent studies have found dyadic effects in heterogeneous samples of spousal and non-spousal caregivers.^9,11,18 One dyadic study of caregivers and patients with cancer, dementia, diabetes, and stroke suggested the possibility of differences in interdependence by the dyad’s relationship type; the authors describe the differences as substantive but not statistically significant.¹⁵ Another study that tested moderation of interdependence by relationship type in cancer patients and caregivers did not find significant differences.¹⁹ Patient-caregiver coresidence, which is often strongly associated with type of relationship (spousal vs non-spousal),²⁰ is another important factor to consider.^21,22

One major shortcoming of much dyadic cancer research to date has been a lack of racial and ethnic diversity.^23,24 In particular, there has been little work examining outcomes for Black cancer patients and their informal caregivers together, especially over time. A cross-sectional study of Black Americans with lung, breast, colorectal, or prostate cancer (74% of whom had metastatic disease) and their Black family caregivers²⁵ found that mental distress was prevalent (reported by 50% of patients and 53% of caregivers), and that greater caregiver mental distress was associated with lower quality of life for both patients and caregivers. The authors note that their study did not investigate patterns over time and may not generalize to patients with early-stage disease.²⁵ The Detroit Research on Cancer Survivors (ROCS) Study^26-28 has provided insights into the experiences of caregivers for African American patients. The authors found that most caregivers for African American patients (over 75%) were female, and the vast majority (97%) were African American themselves.²⁶ Caregivers provided more than 35 hours a week of care on average, and they often reported making changes to work due to caregiving as well as bearing caregiving-related costs.^26-28 Although important, the ROCS findings take caregivers as the unit of analysis rather than examining how patients and caregivers affect each other’s outcomes.

This lack of dyadic research focused on Black patients is particularly problematic in light of the well-documented disparities Black women with breast cancer face due to structural racism.^29-33 Among Black women in the United States, breast cancer is the most commonly diagnosed non-skin cancer²⁹ and the leading cause of cancer death.³¹ Compared to White women, Black women with breast cancer are 41% more likely to die from the disease.²⁹ They may experience lower quality of life,^34,35 deal with interpersonal discrimination during treatment,³⁶ and face disparities in receipt of guideline-based treatment and follow-up care.^30,37,38 Researchers have pointed out that the health system’s reliance on informal caregiving has the potential to exacerbate health disparities (if, for example, both patients and the people in their support networks have lower socioeconomic status or lack access to opportunities and resources).³⁹ Analyzing data from a dyadic perspective is one way to examine these patterns of mutual influence.

Recruiting samples from a range of racial and ethnic backgrounds is especially important because past work has indicated that patterns of caregiving and social support may differ by race.^5,39 Black women with breast cancer who are in partnered relationships often receive support from their spouses/intimate partners,^40,41 but the majority of older Black women in the U.S. are not married.⁴² Prior work has shown that Black women with breast cancer are often supported by a network of friends and family, especially female family members.^43-45 As a result, it is unclear whether past dyadic work with primarily spousal couples generalizes to Black cancer patients and their caregivers. In addition, there are relatively few supportive resources designed specifically for Black breast cancer patients,^45-47 their caregivers,⁴⁸ or patient-caregiver dyads that include Black patients.²³ In order to develop effective supports for such dyads, clinicians need to know more patterns of mutual influence among dyads of Black patients and caregivers.

Actor-Partner Interdependence Models (APIMs) can be used to analyze dyadic data and test whether one person’s predictors are associated with not only their own outcomes but also those of a dyad member.⁴⁹ Kenny, Kashy, and Cook⁴⁹ describe four potential sources of interdependence within dyads: compositional effects (in which a dyad shares similar characteristics), partner effects (in which one person’s characteristics or behavior affect the other person’s outcomes), mutual influence (in which both people’s outcomes affect each other through a feedback process), and common fate (in which a shared causal factor affects both members of a dyad). Longitudinal dyadic models allow for testing these effects over time while also accounting for people’s own autocorrelated measurements across time points.^49,50 In random-intercept cross-panel models, as used in this study, specific model components map cleanly onto two of these sources of interdependence. Compositional effects are represented by correlated intercepts (baseline tendencies) and momentary correlations at early time points, and partner effects are represented by prospective effects in the model.

The current dyadic analyses draw on the Dyadic Cancer Outcomes Framework (DCOF),²¹ a general model of the effects of cancer on both spousal and non-spousal dyads. The DCOF, designed to be used in conjunction with modeling approaches such as the APIM, describes how individual- and dyad-level factors may affect patient, caregiver, and dyad-level mental health, physical health, and relationship outcomes over time. The DCOF highlights the importance of the broader social context in which relationships and dyadic processes unfold. Elements of the framework helped frame the research questions, guide our inclusion criteria, and select the outcomes of interest.

This longitudinal dyadic study is, to our knowledge, one of the first to analyze longitudinal interdependence between Black patients and their caregivers. We examined interdependence between Black breast cancer patients and their caregivers over four months shortly after the patient’s diagnosis. We chose measures based on the literature and the DCOF.²¹ Our primary outcome was depressive symptoms, which are often prevalent in cancer patients and caregivers.^12,24,25 Secondary outcomes, chosen because they are commonly used to assess wellbeing in patients and caregivers, included physical and mental health quality of life,^21,25,51 perceived social support,^52,53 and health behaviors.^10,11,21 We hypothesized that patients and caregivers would demonstrate compositional effects (i.e., concurrent correlations) and prospective interdependence, on these measures—that is, that one person’s measures at one time point would predict the other person’s measures at the next time point.

Methods

In this prospective study, we used longitudinal Actor-Partner Interdependence Models (APIMs) to examine interdependence over time in Black women with breast cancer and their informal caregivers. The team followed Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines⁵⁴ in reporting results.

Sample and Procedures

Patients were eligible if they self-identified as a Black/African American woman, were 18 years of age or older, had a new first diagnosis of breast cancer (i.e., were within 8 weeks of their new patient visit), were able to answer questions in English, and had an informal caregiver whom they reported would participate in the study. Patients who reported being multiracial were eligible if they considered themselves Black/African American as part of their identity. Patients were excluded if metastatic breast cancer was diagnosed or strongly suspected at initial clinical presentation; however, those who later developed metastatic breast cancer were not excluded from continuing in the study. Caregivers were eligible if they were able to answer questions in English, were 18 years of age or older, and had been identified as a support person by a patient who was eligible for the study.

Potentially eligible patients were identified by screening patient records of collaborating surgeons at a NCI-designated Comprehensive Cancer Center in the Midwest. Patients were identified consecutively and mailed study information for themselves and a caregiver (described to patients as “your most important support person in dealing with breast cancer”). Study staff followed up with a phone call to assess interest, screen for eligibility, and obtain caregiver contact information. All participants provided verbal informed consent for data collection prior to the first survey; patients also consented to extraction of cancer-related information from their medical records. Surveys were administered by trained study staff over the telephone at baseline, two months, and four months; an additional 18-month exploratory timepoint is not included in the current analyses. Participants were considered enrolled in the study if they completed at least half of the baseline survey, and participants who enrolled were contacted by phone at every follow-up timepoint unless they notified the study team that they wished to withdraw from the study. The Qualtrics platform was used for computer-assisted data capture during survey administration. Participants received a check or gift card as a thank-you for completing each survey ($30 at baseline, $35 at 2 months, $40 at 4 months). At the end of every survey, participants were offered referrals to counseling/social work if desired. Data for the first three timepoints were collected January 2021 – July 2023. This study was approved by Institutional Review Boards at Washington University in St. Louis, where data collection took place (IRB #202010222, initial approval 12/3/2020), and the University of North Carolina at Chapel Hill, where data analysis was conducted (IRB #23-1627, initial approval 10/9/2023).

Measures

To conduct APIM analyses, the same measures must be administered to both members of a dyad. The team therefore chose measures appropriate for people with and without chronic disease. Our choice of outcome measures was guided by the Dyadic Cancer Outcomes Framework,²¹ which was developed based on the literature about patient-caregiver dyads in the context of cancer.

Depressive symptoms (primary outcome)

Depressive symptoms, which are often prevalent in cancer patients and caregivers,^12,24,25 were assessed with the Center for Epidemiologic Studies Depression scale (CES-D), a 20-item measure that has been used in populations with and without chronic disease and that has been shown to be reliable and valid in African American cancer survivors.^55-57 Scores range from 0 to 60, with greater scores indicating higher levels of depressive symptoms; a score of ≥16 is often used as a cutoff to indicate the need for further evaluation. In this sample, the CES-D demonstrated excellent reliability (baseline Cronbach’s alpha = .891 for patients, .869 for caregivers).

Physical and mental health quality of life (secondary outcomes)

General physical and mental health quality of life measures are often used to assess overall wellbeing in cancer patients and caregivers.^21,25,51 Participants completed the PROMIS Global-10 v.1.2, a 10-item patient-reported outcome measure that has shown good reliability and validity in the general U.S. population.^58-60 Four items are summed to create a raw physical health quality of life subscale score, and four items are summed to create a raw mental health quality of life subscale score; two items not included in either subscale score are not used in these analyses. Raw scores are transformed into scaled scores with a U.S. population mean of 50 and a standard deviation of 10, with higher scores indicating better health.⁵⁹

Perceived social support (secondary outcome)

Perceived social support, which has been associated with physical and mental health outcomes in cancer patients⁵² and caregivers,⁵³ was measured with the 19-item Medical Outcomes Study Social Support Survey (MOS-SSS). This measure has been shown to be reliable and valid in community-dwelling adults with chronic disease⁶¹ and has been successfully used in studies of Black women with breast cancer⁵² and in control study participants without cancer.⁶² Transformed MOS-SSS scores are on a scale of 0-100, where higher scores indicate greater social support.⁶³

Health behaviors (secondary outcome)

Health behaviors may also be interdependent in patient-caregiver pairs.^10,11,21 Participants were asked about three health behaviors that have been associated with mortality in prior research.^64-67 A composite health behavior variable was created from three items adapted from prior health surveys: smoking cigarettes (every day/some days/not at all), days of exercise per week (0-7)⁶⁸ and number of servings of fruits and vegetables per day (0, 1, 2, 3, 4, 5+).⁶⁹ Whereas some previous work has combined multiple health behaviors by dichotomizing separate indicators into healthy/unhealthy and then summing them,^64,65,67 we chose to retain as much information from the original items as possible. Because the items used different response scales, responses were transformed by dividing each by the maximum then combining, with percent of the maximum of exercise and fruits and vegetables added and percent of maximum smoking subtracted. This approach was used for combining items rather than z-scores (which would have the effect of removing any trends over time). The resulting index was conceptualized as a general measure of health behaviors.

Demographic data

Patients and caregivers were asked to provide information about age (in years), insurance status (dichotomized into some private insurance vs not), marital status (dichotomized into married/partnered vs not), work status (dichotomized into working at least part-time vs not), highest level of education (some high school or less; high school diploma/GED; some college/technical school; bachelor’s degree; graduate/professional degree), income (under $20,000; $20,000 to <$40,000; $40,000 to <$75,000; $75,000 or more), the number of children under 18 in the home, and the number of other adults in the home. Patients and caregivers reported their relationship to one another, dichotomized into spousal/romantic partner vs not; romantic partner dyads included dyads in which patients chose “another relationship” to describe the dyad and elaborated that the caregiver was their boyfriend or fiancé. Participants reported the length of their relationship in years and whether they lived together all, some, or none of the time (dichotomized into living together some/all of the time vs not). Type of dyad, dyad coresidence, and length of relationship in years were based on patient self-report, supplemented by caregiver self-report if no patient data were available.

Identifying as a Black/African American woman was an inclusion criterion for patients; no additional race or gender information was collected from them in surveys. Caregivers were asked to choose all that applied to describe their racial identities (Asian; American Indian or Alaska Native; Black or African American; Native Hawaiian or Other Pacific Islander, White, or prefer to self-describe; dichotomized as Black/African American vs not). Patients and caregivers were asked if they considered themselves Hispanic/Latino. Caregivers were asked whether they considered themselves male, female, or preferred to self-describe (dichotomized into male or female, since no caregivers self-described).

Analyses comparing patients who did and did not enroll in the study used two variables obtained from the medical record at the time of record screening: age in years at the time of the patient’s new cancer visit and marital status, which was coded as married vs another relationship status.

Cancer staging

Cancer stage was taken from the medical record at the time of the 4-month survey and was categorized as early (Stage 0, 1, or 2A), locally advanced (Stage 2B or 3), metastatic (Stage 4), or unknown. Although patients were ineligible for the study if they presented with metastatic disease at the time of their new cancer visit, the few patients who later developed metastatic disease were not excluded from continuing to participate.

Analysis Plan

Descriptive and bivariate statistics were conducted in IBM SPSS Statistics Version 29.0.1.0⁷⁰ to characterize the patients and caregivers in the sample. We also used bivariate tests to examine differences between patients who did and did not enroll in the study. Actor-Partner Interdependence Models assessed interdependence of patient and caregiver measures over time.⁴⁹ The random-intercept cross-panel model was the primary analytic method, but when there were indications of linear trends, latent growth models were also examined. Analyses were conducted in Mplus version 8 using the MLR (maximum likelihood estimator robust to violations of normality) or Bayes estimator with uninformative priors.⁷¹

The primary models contained only the outcomes of interest at every timepoint. Exploratory multiple-group models examined differences between dyads who were or were not coresident. Additional follow-up models tested the effects of adding covariates to the initial models; covariates were also added to additional models as appeared informative based on their initial results. Because many of the demographic variables were highly correlated, we included only a preplanned subset of them (type of dyad, patient income, patient education, patient insurance states, patient age, and cancer stage; see Table 1) in adjusted models; other variables are included in Results only to describe study participants. All data in these analyses are presented in such a way that the identity of individual participants may not be ascertained.

Table 1.

Analytic Changes to the Preregistration and Rationales for Those Changes

Type of change	Timing of change	Preregistered plan	Rationale/Impact
Data collection	Early in data collection	Baseline timepoint described as “within 6 weeks of new breast cancer appointment”	As we conducted initial recruitment, the team realized an enrollment cutoff of 6 weeks after the new breast cancer appointment did not allow sufficient time to recruit patients. In consultation with the clinicians on the study team, the timing for the patient baseline survey was changed to be within 8 weeks of the patient’s new cancer visit. Caregivers were allowed to complete their baseline survey up to one month after the patient completed their survey.
Eligibility	Early in data collection	Inclusion criteria for patients stated they must be “Stage 0 – 3B at time of their new cancer appointment”	The initial statement about Stage 0 – 3B cancer was deleted; an exclusion criterion was added: “Evidence of metastatic disease at initial clinical presentation.” This change was made to (1) include patients who were stage 3C (who had been omitted in error), and (2) to take into account the fact that staging around the time of the new cancer visit is often not final. The stage variable included in analyses was based on staging at the time of the 4-month surveys.
Data collection	During data collection	CES-D instructions for the baseline survey in Qualtrics inadvertently omitted a sentence from the instructions (“The following is a list of the ways you may have felt or behaved”). This error was corrected when it was noticed in Aug 2021.	The optimal way to test the effects of this error would be a multiple-group analysis of people who took the survey before and after the error was corrected, but a multiple-group model did not converge. Nevertheless, the estimates obtained from the partially-converged model did not indicate any sign of interdependence. Similarly, examining only those participants who completed the surveys after the error had been corrected revealed no signs of interdependence.
Analysis	After data collection for the analyzed timepoints was complete	The preregistration stated, “A second set of models will determine whether adding covariates including type of dyad (patient-partner vs other dyads), income, education, insurance, patient age, and treatment variables (Y/N for radiation, chemotherapy, and hormone therapy; breast conserving surgery vs bilateral or unilateral mastectomy) modify any effects found.”	Prior to starting analyses, members of the study team discussed the fact that, because patient data for these analyses included information only from the 4 months after enrollment, some of the clinical variables measured treatment that took place after the surveys were administered and thus could not plausibly affect the tested pathways; final treatment variables were extracted from the medical record 18 months after study enrollment. Instead, the team included stage as a covariate because past work has found to be correlated with types of treatment.
Analysis	After data collection for the analyzed timepoints was complete	The preregistration stated, “APIM models will be cross-lagged regression analyses in which a person’s score on a measure at T2 and T3 is predicted by their scores on that measure at the prior time point as well as their partner’s score at the prior time point.”	Due to new evidence that failure to account for correlated intercepts produces inflated cross-construct relationships⁵⁰ we used Random Intercept Cross-Lagged Panel Models (RI-CLPMs) to conduct the APIM analyses. We also provide the pre-registered model results in Results for comparison. When interpretation or convergence was problematic, we used the Bayes estimator provided the variables analyzed appeared normal (judged via skewness and kurtosis). The Bayes estimator has various helpful properties including (a) it does not estimate negative variances and (b) its estimates and standardized estimates have consistent significance estimates. In addition, instead of focusing only on correlations at each time point, we examined the correlations of random intercepts to look for stable dyadic relationships over time.
Analysis	After data collection for the analyzed timepoints was complete	Preregistered plans called for an exploration of differences by dyad type (patient-partner vs other types of dyads).	We decided to explore differences by dyad co-residence, which was strongly correlated with dyad type (r = .60), because we believed that interdependence/ emotional contagion was more plausible for dyads who live together compared to those who do not. Because including the variable as a covariate would not allow us to test differences of effects, we decided to conduct separate exploratory models for dyads who did and did not live together. We included a patient marital status variable as a covariate in adjusted models.

Power

The a priori power analysis and sample size estimate were based on the cross-lagged panel models originally planned (see Figure 1 and Table 1). We used a Monte Carlo simulation study with 1000 replications in Mplus 8 based on the following assumptions: a population correlation between patient and caregiver of 0.2 for CES-D score⁷²; a standardized regression coefficient for partner effect of 0.18;⁸ and a moderate amount of missing data (for patients: 2% at baseline, 10% at T2, and 20% at T3; for caregivers, 5% at baseline, 15% at T2, and 25% at T3).⁵² The simulation yielded a power of at least .90 for the planned tests to detect a partner effect with a sample of 100 dyads; a sample of 110 dyads was planned in case attrition was higher than anticipated.

Figure 1.

Illustration of a Random-Intercept Cross-Lagged Panel Model (RI-CLPM). Whereas a more conventional cross-lagged panel model estimates only the solid paths, a random intercept model also conceptualizes each timepoint as also being affected by a person’s overall tendency toward that construct (the dotted paths). For simplicity, patient-caregiver correlations and correlated errors for each timepoint are not shown here, but they are included in all models.

Subsequent to this power analysis, as described below and in Table 1, we changed our primary modeling strategy to a more conservative test that might in some cases reduce power. We address this issue further in the Discussion.

Missing data

These analyses include all participants who provided meaningful data at baseline (i.e., finished at least half of the first survey). The MLR and Bayes estimators generally handle missing data appropriately without listwise deletion, whereas listwise deletion would result in biased parameters.⁷³ For this reason, we included data from patients and caregivers in the rare instances when the other member of the dyad did not enroll. One exception is when covariates are missing; in this case we minimized the number of listwise deleted participants by predicting the most missing covariate (patient income, n = 5) with whether the patient had private insurance; as a result, adjusted analyses are missing two participants.

Preregistration and modifications to preplanned analyses

The original analysis plan was preregistered on OSF prior to data collection (https://osf.io/xf4ac). Table 1 provides information about changes made to the preregistered study design and data analysis plans and a rationale for all changes.

Of note, the original analysis plan stated that cross-lagged panel APIMs would be used to test for interdependence. Due to new evidence that failure to account for correlated intercepts produces inflated cross-construct relationships,⁵⁰ we used Random Intercept Cross-Lagged Panel Models (RI-CLPMs) to conduct APIM analyses. RI-CLPMs, as implied by the name, add a random intercept to the cross-panel model (see Figure 1). This allows the model to better account for individual tendencies toward higher or lower scores that affect all observations. That is, some people have a higher level of depressive symptoms generally, in such a way that all of their scores tend to be higher than other people’s. In the case of the current models, it is likely this general tendency towards higher or lower scores pre-dates the first measurement. When this influence is accounted for, correlations at each time point represent momentary correlations between patient and caregiver that are specific to that timepoint and do not merely reflect pre-existing tendencies. Notably, both correlated intercepts and momentary correlations fall under the concept of compositional effects. Correlated intercepts suggest pre-existing compositional effects, whereas momentary correlations suggest newly formed compositional effects.

Failing to take correlated intercepts into account can produce spurious relationships between variables that are better attributed to correlated random intercepts: Dyads might appear to influence each other when the more parsimonious explanation is that both dyad members are prone to having similar scores. Alternatively, dyads might appear to have a correlation at Time 1 and not later timepoints, which is misleading if their tendency to correlate is not restricted to Time 1 and is more appropriately understood as a correlated underlying tendency. Using the RI-CLPM avoids these issues, and at the same time allows us to examine two aspects of compositional effects. Thus, using RI-CLPM is consistent with our pre-registered intentions, although we originally intended to use standard cross-panel models. We also provide the pre-registered model results for comparison.

When interpretation or convergence was problematic, we used the Bayes estimator instead of MLR provided the variables analyzed appeared approximately normal (judged via skewness and kurtosis). The Bayes estimator has various helpful properties including (a) it does not estimate negative variances and (b) its estimates and standardized estimates have consistent significance estimates, whereas in MLR these can conflict. When the Bayes estimator is used, we report whether effects are statistically significant at P < .05 based on the 95% credible interval not containing zero.

Multiple group models tested whether results differed by whether patients and caregivers lived together. The Bayes estimator does not have a multiple group models implementation in Mplus, so the MLR estimator was used when possible, or separate Bayesian models were run in each group. Tests of whether covariates modified results were conducted by regressing all the outcome variables (at all time points) on the covariates directly. Tests of interdependence were then conducted on the residuals of the outcomes once the covariates and the random intercepts had been accounted for.

Results

Figure 2 provides a flow diagram of recruitment, enrollment, and survey completion. A total of 258 patients were mailed materials inviting them to participate. In all, 107 patients and 103 caregivers enrolled and provided baseline data, for a total of 210 participants (108 complete and partial dyads). Although every effort was made to enroll both members of a dyad, there were five patients who met eligibility criteria (including reporting that a caregiver was willing to participate) whose caregivers were ultimately unable to be reached for study enrollment. One patient expressed interest in study participation and her caregiver was then contacted and enrolled; the patient was subsequently unable to be reached. There were no significant differences in age or marital status between patients who did and did not enroll in the study. In the 108 dyads, survey completion was 97% at baseline (one patient and five caregivers missing), 94% at two months (four patients and nine caregivers missing), and 90% at four months (eight patients and 13 caregivers missing).

Figure 2.

Flow Diagram of Study Participants (N = 210)

Individual characteristics of enrolled patients and caregivers are described in Table 2. Two-thirds of patients (66%) did not live with the person they recommended for this study. Most patients (78%) referred a caregiver who was not a spouse or romantic partner; even among the patients who reported being married/partnered at baseline, 39% nominated a support person who was not their partner (of those, 75% were daughters, 17% were friends, and 8% were another family member). Overall, over one-third (34%) of referred caregivers were the daughter of the patient; others were a spouse/romantic partner (22%), sister (15%), friend (8%), son (7%), mother (5%), brother (1%), another family member (5%; two nieces, two aunts, one granddaughter) and another relationship (3%; one sister-in-law, one neighbor, one god-daughter). Mean relationship length was 35.5 years (SD = 15.4).

Table 2.

Descriptive Statistics for Baseline Measures From Patients and Caregivers (N = 210 Participants: 102 Complete Dyads, 5 Patients Without Caregivers, and 1 Caregiver Without a Patient)

Variable	Patients (n = 107) % (n) or mean (SD)	Caregivers (n = 103) % (n) or mean (SD)
Gender
Female	100% (107)	70% (72)
Male		30% (31)
African American/Black	100% (107)	93% (96)
Hispanic/Latino
No	98% (105)	98% (101)
Yes	2% (2)	2% (2)
Age	58.5 (11.2)	49.9 (14.2)
Annual household income
Under $20,000	23% (24)	19% (18)
$20,000 to <$40,000	27% (28)	21% (20)
$40,000 to <$75,000	25% (26)	26% (25)
$75,000 or more	24% (25)	35% (34)
Education
Some high school or less	7% (7)	4% (4)
High school diploma/GED	17% (18)	20% (21)
Some college/technical school	40% (43)	41% (42)
Bachelor’s degree	17% (18)	19% (20)
Graduate/professional degree	20% (21)	16% (16)
Insurance
At least some private/employer insurance	52% (56)	69% (70)
No private insurance	48% (51)	31% (32)
Martial status
Married/unmarried partnership	29% (31)	45% (46)
Another relationship status	71% (76)	55% (57)
Employment
Working at least part-time	47% (50)	69% (71)
Not currently working for pay	53% (57)	31% (32)
Children <18 years in the household
0	75% (80)	61% (63)
1	15% (16)	25% (26)
2 or more	10% (11)	14% (14)
Other adults living in the household
0	42% (45)	31% (32)
1	34% (36)	42% (43)
2 or more	24% (26)	27% (28)
Cancer stage (4 months after study enrollment)
Early stage (Stages 0, 1A, 1B, 2A)	83% (88)	_
Locally advanced (Stages 2B, 3)	15% (16)
Metastatic (Stage 4)	2% (2)
CES-D	15.0 (10.9)	11.9 (9.0)
PROMIS-Global 10 physical health	45.1 (8.4)	50.6 (7.5)
PROMIS-Global 10 mental health	46.6 (8.3)	49.0 (8.2)
MOS-SSS	82.4 (17.1)	82.9 (19.2)
Smoking
Not at all	88% (94)	89% (91)
Some days	6% (6)	5% (5)
Every day	7% (7)	6% (6)
Servings of fruits/vegetables per day
0	4% (4)	3% (3)
1	16% (17)	31% (32)
2	36% (38)	28% (28)
3	19% (20)	19% (19)
4	13% (14)	11% (11)
5+	11% (12)	9% (9)
Days of exercise/week
0	28% (30)	16% (16)
1	8% (8)	5% (5)
2	11% (12)	18% (18)
3	22% (23)	19% (19)
4	5% (5)	12% (12)
5	8% (8)	11% (11)
6	3% (3)	4% (4)
7	17% (18)	17% (17)

Note: Abbreviations: CES-D = Center for Epidemiologic Studies Depression scale; PROMIS = Patient-Reported Outcomes Measurement Information System; MOS-SSS = Medical Outcomes Study Social Support Scale. Percentages given here are valid percentages and may not sum to 100 due to rounding. The following data were missing for participants who completed the baseline survey: Income (4 patients, 6 caregivers), insurance (1 caregiver), cancer stage (1 patient), CES-D (2 patients, 2 caregivers), MOS-SSS (4 patients, 5 caregivers), smoking status (1 caregiver), servings of fruit/vegetables (2 patients, 1 caregiver), days of exercise/week (1 caregiver).

Mean baseline CES-D score was 15.0 (SD = 10.9) for patients and 11.9 (SD = 9.0) for caregivers. Forty-two percent of patients and 28% of caregivers had baseline CES-D scores ≥16. In 56% of complete dyads providing baseline CES-D data, one or both dyad members were at or above that threshold.

Depressive Symptoms

We found no evidence of prospective partner effects for depressive symptoms, but some evidence of compositional effects was found in exploratory tests. The planned RI-CLPM, the standard cross-lagged panel model, and the RI-CLPM with covariates included showed no indications at all of interdependence for depressive symptoms (all ps > .05; see Table 3). In a multiple group model based on whether participants lived together at least part of the time or not, one indication of interdependence was observed. Only among dyads who lived together, the residual at Time 3 correlated negatively (r = −.69, P < .001), indicating a tendency to systematically diverge in outcome at Time 3, implying a newly-formed compositional effect. When patients were more depressed than their baseline at Time 3, caregivers were less depressed. An additional set of models examined the potential effects of an administration error in the CES-D survey instructions (see Table 1) but found that the error did not affect results of any effects of interdependence.

Table 3.

Complete Standardized Coefficients for the Unadjusted Model of Patient and Caregiver Depressive Symptoms, as Measured by the CES-D

	Coefficient (SE)	P
Patient and caregiver correlation (intercept)	−.007 (.69)	.992
Patient actor effect (T1 to T2)	.54 (.23)	.017
Patient actor effect (T2 to T3)	.48 (.26)	.063
Caregiver actor effect (T1 to T2)	.57 (.30)	.058
Caregiver actor effect (T2 to T3)	.78 (.20)	<.001
Patient→caregiver partner effect (T1 to T2)	.18 (.17)	.291
Patient→caregiver partner effect (T2 to T3)	−.02 (.08)	.776
Caregiver→patient partner effect (T1 to T2)	.004 (.29)	.990
Caregiver→patient partner effect (T2 to T3)	.21 (.22)	.335

SE = standard error; CES-D = Center for Epidemiologic Studies Depression scale.

Post hoc analyses for depressive symptoms

Simple examination of means, which declined for both groups over time, revealed that it was possible that many patients and caregivers demonstrated changes in depressive symptoms over time. The presence of linear trends would confound the models above. We examined this possibility with a latent growth curve model in which both patients and caregivers had an intercept and slope, both of which were permitted to vary randomly across participants. This model revealed that whereas neither patients nor caregivers had a systematic linear tendency, caregivers did have significantly varying slopes (Variance = 9.63, P = .025), which suggests that some caregivers had systematically decreased depression over time whereas others had systematically increased depression. Prospective effects could only be modeled if effects of the same type were constrained to be equal (e.g., patient autoregressive parameter was the same from Time 1 to Time 2 as from Time 2 to Time 3), and then only using the Bayes estimator. No signs of interdependence were detected in this model. When participants were restricted to those who lived together, the correlation between residual variances reported above was present (r = −.79, P < .05).

Physical Health Quality of Life

Evidence of partner effects was found only in exploratory tests, but evidence of compositional effects was found in the primary model. The planned RI-CLPM showed inconsistency between standardized and unstandardized parameters in the MLR-estimated model for the correlation of the random intercepts, and thus a Bayesian model was used. In the Bayesian model, the random intercepts correlated significantly (r = .25, P < .05); patients and caregivers tended to have similar physical health quality of life. Adding covariates to the RI-CLPM did not change the significance of this finding (r = .27, P < .05). In contrast, the standard cross-lagged panel model instead showed a correlation between patient and caregiver at Time 1 only (r = .24, P = .009). The multiple-group model for dyads indicated that this correlation did not differ markedly across groups, in that in both groups the correlation was in the .2 range but not statistically significant, plausibly due to the smaller sample size for those analyses. In addition, dyads that lived together vs lived apart showed divergent prospective patterns of interdependence. For patients who lived apart from their caregivers, caregiver physical health quality of life at Time 2 predicted improved patient health at Time 3 (standardized coefficient = .42, P = .015), whereas this relationship was negative when dyads lived together (standardized coefficient = −1.04, P = .001). In addition, correlations were observed at different timepoints for those who lived apart vs together (apart: Time 3, r = .47, P = .005; together: Time 1, r = .73, P = .002).

Post hoc analyses for physical health quality of life

The finding of significant slope variance in depressive symptoms for caregivers above suggested the necessity of checking for a linear trend for other secondary outcomes as well. In an initial model including slope and intercept for both patient and caregiver, caregivers did not have a significant mean or variance for their slope (ps > .10) and thus their slope was omitted. Patients showed a slight tendency to have deceasing physical health, with little variance as estimated by a nonsignificant negative estimate (M = −0.68, P = .042, variance P > .30). When the small negative variance was constrained to be zero, the intercept for patient and caregiver correlated as reported above (r = .28, P = .006). Similarly, when prospective effects were added in a Bayesian version of the same model, none were statistically significant but the correlation between intercepts remained.

Mental Health Quality of Life

No evidence of partner effects was found for mental health quality of life, but evidence of compositional effects was clear. The planned RI-CLPM displayed correlated random intercepts (r = .28, P = .002); patients and caregivers tended to have similar mental health quality of life. The multiple-group model would not converge, but separate models for dyads who lived together vs apart using the Bayesian estimator did converge and revealed no additional evidence for interdependence. Significant findings were unchanged when covariates were added; only the random intercept correlation (r = .23, P = .006) suggested interdependence. In contrast, the standard cross-lagged panel model showed a correlation between patient and caregiver at Time 1 (r = .29, P = .004).

Post hoc analyses for mental health quality of life

Perceived Social Support

No evidence of partner effects was found in the preferred model for perceived social support, but evidence of compositional effects was clear. The planned RI-CLPM displayed correlated random intercepts (r = .37, P = .002); patients and caregivers tended to have correlations in overall levels of social support. An additional correlation at Time 1 was inconsistent across estimates but not statistically significant in a Bayesian version of the model. The multiple-group model would not converge, but separate models for dyads who lived together vs apart using the Bayesian estimator did converge and revealed no additional evidence for interdependence. Adding covariates did not change substantive results. In contrast, the standard cross-lagged panel model showed a marginal correlation between patient and caregiver at Time 1 (r = .16, P = .051) as well as a prospective effect such that patient social support at Time 3 was predicted by caregiver social support at Time 2 (standardized coefficient = .16, P = .040).

Post hoc analyses for social support

Similar to the linear trend effects seen in depressive symptoms, caregivers had significant variance but no significant mean for their social support (variance = 15.36, P = .012); patient variance was marginal (variance = 38.54, P = .060). As with the models above, a Bayesian model in which constrained prospective effects were estimated alongside random slopes and intercepts indicated no additional signs of interdependence and the correlated intercepts were no longer statistically significant.

Health Behaviors

No evidence of partner effects was found in the preferred model for health behaviors, but evidence of compositional effects was clear. The planned RI-CLPM displayed correlated random intercepts (r = .33, P = .026); patients and caregivers tended to have similar health behaviors. In the multiple group model, dyads living apart had correlated intercepts (r = .40, P = .008), whereas those living together showed a correlation at Time 3 (r = −.53, P = .026). With covariates added, the correlation between intercepts was reduced to a trend (r = .35, P = .052). In contrast, the standard cross-lagged panel model showed a correlation between patient and caregiver at Time 1 (r = .30, P < .001). Unlike the other outcome variables, health behaviors showed no sign of linear trends, with means and variances of slopes all non-significant (ps > .550), and therefore no follow-up models were conducted.

Discussion

This is one of the first longitudinal analyses of data from Black women with cancer and their informal caregivers. The Dyadic Cancer Outcomes Framework²¹ guided our selection of mental and physical health outcomes, as well as our decision to allow patients to nominate a range of caregivers (despite the analytic heterogeneity introduced by this choice). The DCOF also led us to be mindful of the broader social context in which caregiving takes place in the U.S. Consistent with Ellis et al,²⁵ we found depressive symptoms were prevalent among Black patients and their caregivers; in over half of the dyads in the current study, at least one member met or exceeded the cutoff for depressive symptoms that would indicate the need for further screening. The large majority (78%) of patients nominated a non-spousal caregiver as their most important support person in dealing with cancer. This is perhaps unsurprising given that only 29% of the patients reported being married or partnered at baseline. At the same time, even among patients who reported being married or partnered, 39% did not nominate their spouse. These findings, consistent with previous work showing the importance of female caregivers,^43-45 underline the importance of asking patients who is providing them with support rather than assuming that spouses are the primary source of support for most patients.

Our hypothesis that dyads would demonstrate prospective interdependence in depressive symptoms, mental and physical health quality of life, social support, and health behaviors was not supported. In other words, we did not observe what some have called partner effects⁴⁹ and what others have called spillover effects¹⁵ or emotional contagion¹³ during the 4-month study period. Findings are consistent with the assertion of Litzelman et al¹⁵ that, although dyadic effects are common, there may be heterogeneity in dyadic effects across different populations; it is important to assess dyadic interdependence in a range of populations instead of assuming prior work will generalize to people from other backgrounds. It is also possible that these null findings were due to the time interval at which we assessed patients and caregivers. Although our study design was similar to others who have measured psychosocial outcomes at 2-month intervals,⁹ it is possible that we missed dyadic processes unfolding over longer or shorter timeframes. For example, alternative methods, such as laboratory observations, ecological momentary assessment, or daily diary studies, would be able to detect shorter-term shifts in mood.^74,75

We did observe small-to-moderate compositional dyadic effects for several outcomes, meaning that patients and caregivers were similar to one another in their overall levels of physical and mental health quality of life, perceived social support, and health behaviors. The average relationship length among these dyads (35.5 years), as well as the fact that many caregivers were patients’ family members, suggests dyadic processes may have unfolded for years or decades prior to study enrollment and were therefore relatively stable by the time of the study period. In the cases of spouses or romantic partners, there may have also been assortative selection into partnerships (i.e., people choosing partners due to similar characteristics). The combination of lack of prospective dyadic effects with fairly consistent compositional effects underscores the importance of collecting and analyzing longitudinal dyadic data (as suggested by the Dyadic Cancer Outcomes Framework²¹). Correlations that might have been observed at Time 1 alone did not translate to prospective effects.

Although our study was not designed to detect the sources of these compositional effects, the observed similarity in psychosocial characteristics and health behaviors has implications for screening and intervention. Findings are consistent with the idea that patients who face challenges may also be supported by network members facing similar challenges.³⁹ Systematic patient and caregiver screening for a range of health and psychosocial outcomes could help identify dyads that might benefit from intervention.^23,76 For example, our findings are consistent with prior research^10,11 showing suboptimal health behaviors in patient-caregiver dyads; more than half of our patients and caregivers reported exercising three days a week or fewer, and more than half ate two or fewer servings of fruits and vegetables on a typical day. Such dyads may particularly benefit from interventions centered on health behaviors.^76,77 Even though dyads in the current observational study did not demonstrate prospective interdependence, dyadic interventions often incorporate support or counseling designed to strengthen a pair’s relationship and communication skills.^23,78,79 Thus, it is possible that participating in a dyadic intervention together may nurture new processes of mutual influence within a pair. Existing dyadic interventions could be adapted specifically for Black women with breast cancer and their informal caregivers in a range of ways (e.g., by using thoughtful targeted outreach/recruitment strategies, incorporating racially diverse staff, and acknowledging historic racial discrimination in healthcare).^76,80 Our team is currently analyzing qualitative interviews conducted with a subsample of dyads from this study to learn more about participants’ preferences for intervention.

In these analyses, we used RI-CLPMs because recent work has argued (convincingly, in our view) that they are superior to standard cross-lagged panel models for measuring outcomes in dyads over time.⁵⁰ On a philosophical level, standard cross-lagged panel models assume that any relationship between the outcome variables is due either to time-specific correlations or prospective effects. In contrast, the RI-CLPM acknowledges the fact that people differ in their overall tendencies in certain measurements. Standard cross-lagged panel models using data like ours might lead researchers to conclude that patients and caregivers have similar health behaviors, physical health, social support, and mental health at an initial timepoint but that this relationship disappears over time. In contrast, the RI-CLPM suggests a more plausible interpretation: People in dyads are more similar to each other than to people in other dyads, and these compositional effects likely arose prior to data collection rather than at the time of the baseline measurements. More technically, failing to account for these correlated errors can lead to findings of apparently prospective effects that are better accounted for by a combination of correlated intercepts and measurement error, as might be the case for the social support finding that the standard model suggested but that the RI-CLPM did not corroborate. It should also be noted that state-of-the-art techniques for intervention trials include mixed-effects models or similar strategies which, like the RI-CLPM, include random intercepts and thus account for participants’ stable tendencies when measuring intervention effects.^81,82 Researchers conducting future dyadic studies should carefully consider the best analytic tools for analyzing the dyadic processes they expect to observe.

Strengths and Limitations

This work has several strengths. We recruited a sample of 210 Black patients and caregivers and followed participants over three waves of data collection. Our retention rate was very high, with over 90% of patients and caregivers who enrolled completing their follow-up surveys at the third (4-month) timepoint. We tested preregistered hypotheses and included a range of outcome measures commonly used in dyadic studies that have been shown to be reliable and valid in populations with and without chronic disease. Analyses include both preplanned models and post hoc models to explore our findings more fully.

This work has several limitations. Although most of these outcome measures have been shown to be reliable and valid, our composite health behavior variable was created for this study. Our power analysis was based on using cross-lagged panel models; the RI-CLPMs that we ultimately employed are more conservative. In addition, power calculations were based on a test of our primary outcome (depressive symptoms); secondary outcomes and post hoc tests should be considered exploratory. However, results are generally consistent across modeling frameworks and variables here, such that power alone cannot explain our results. Instead, the RI-CLPMs simply offered greater conceptual clarity.

Several characteristics of the sample should also be noted. We recruited participants with nonmetastatic disease from one NCI-designated Comprehensive Cancer Center in the Midwestern United States. Results may not generalize to patients from other U.S. regions or other countries, those with an initial diagnosis of metastatic disease, or those treated in community hospitals. Although our 41% enrollment rate for patients is consistent with previous dyadic studies and is higher than the average seen across behavioral trials of patients and caregivers,⁸³ the majority of patients approached for participation did not enroll. We found that patients who chose to enroll in the study were not significantly different from those who enrolled in regard to age and marital status, but there may have been other, unmeasured differences between participants and non-participants.

Conclusions

This study is one of the first longitudinal, dyadic examinations of outcomes for Black breast cancer patients and their informal caregivers. As suggested by the Dyadic Cancer Outcomes Framework,²¹ we analyzed patient, caregiver, and relationship variables and examined interdependence longitudinally. We found correlations within dyads in overall levels of mental and physical health quality of life, perceived social support, and health behaviors. Although there was no dyadic interdependence demonstrated for CES-D scores, the high prevalence of elevated depressive symptoms in patients and caregivers indicates that many may benefit from additional support. Given the limited time and resources clinicians have for providing caregiver services,⁸⁰ it may make sense to administer brief screenings to caregivers or dyads and then offer interventions to those most in need. Researchers and clinicians should continue to build an evidence base for how best to support patients, caregivers, and patient-caregiver dyads from a range of racial and ethnic backgrounds in order to improve outcomes for all patients and their loved ones.

Footnotes

Acknowledgements

We would like to thank the study participants for their time and participation. We would also like to thank Charlene Caburnay, Christi M. Lero, Amy McQueen, and Jennifer Wolff for their assistance with this project.

ORCID iDs

Tess Thompson

Thomas L. Rodebaugh

Ethical considerations

This research was approved by Institutional Review Boards at Washington University in St. Louis, where data collection took place (IRB #202010222, initial approval 12/3/2020), and the University of North Carolina at Chapel Hill, where data analysis was conducted (IRB #23-1627, initial approval 10/9/2023). Verbal informed consent was obtained from all individual participants included in the study.

Author Contributions

Tess Thompson: conceptualization, funding acquisition, methodology, project administration, investigation, data curation, formal analysis, visualization, supervision, writing-original draft, writing-review and editing; Thomas Rodebaugh: methodology, formal analysis, visualization, writing-original draft, writing-review and editing; Polly Haun: project administration, investigation, writing-review and editing; Karin Han: project administration, investigation, writing-review and editing; Naitra Ramchander: project administration, investigation, writing-review and editing; Krutika Chauhan: investigation, writing-review and editing; Jada Meads: investigation, writing-review and editing; Julie Margenthaler: supervision; resources; writing-review and editing; Katherine Glover-Collins: resources; writing-review and editing; Graham A. Colditz: Conceptualization; funding acquisition; supervision; writing-review and editing

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the American Cancer Society (MRSG-19-086-01-CPPB, PI: Thompson). The sponsor did not play a role in study design; in the collection, analysis, and interpretation of data; in the writing of the report; or in the decision to submit the article for publication. Use of REDCap for data management was supported by the Clinical and Translational Science Award Grant (UL1 TR002345) and the Siteman Comprehensive Cancer Center and NCI Cancer Center Support Grant P30 CA091842). The Siteman Health Behavior, Communication and Outreach Core provided assistance with survey programming.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

References

Porter

Dionne‐Odom

. Supporting cancer family caregivers: how can frontline oncology clinicians help? Cancer. 2017;123(17):3212-3215.

Kent

Rowland

Northouse

, et al. Caring for caregivers and patients: research and clinical priorities for informal cancer caregiving. Cancer. 2016;122(13):1987-1995.

Institute of Medicine . From Cancer Patient to Cancer Survivor: Lost in Transition. National Academies Press; 2005.

Kayser

Scott

. Helping Couples Cope with Women's Cancers: An Evidence-Based Approach for Practitioners. Springer Science & Business Media; 2008.

Ellis

. Cancer and caregiving among older Black Americans and families. Annu Rev Gerontol Geriatr. 2022;41(1):211-248.

Kim

Carver

Shaffer

Gansler

Cannady

. Cancer caregiving predicts physical impairments: roles of earlier caregiving stress and being a spousal caregiver. Cancer. 2015;121(2):302-310.

Valle

Weeks

Taylor

Eberstein

. Mental and physical health consequences of spousal health shocks among older adults. J Aging Health. 2013;25(7):1121-1142.

Litzelman

Green

Yabroff

. Cancer and quality of life in spousal dyads: spillover in couples with and without cancer-related health problems. Support Care Cancer. 2016;24(2):763-771.

Segrin

Badger

Sikorskii

, et al. Longitudinal dyadic interdependence in psychological distress among Latinas with breast cancer and their caregivers. Support Care Cancer. 2020;28(6):2735-2743.

10.

Shaffer

Kim

Llabre

Carver

. Dyadic associations between cancer-related stress and fruit and vegetable consumption among colorectal cancer patients and their family caregivers. J Behav Med. 2016;39(1):75-84.

11.

Ellis

Janevic

Kershaw

Caldwell

Janz

Northouse

. Engagement in health-promoting behaviors and patient–caregiver interdependence in dyads facing advanced cancer: an exploratory study. J Behav Med. 2017;40(3):506-519.

12.

Streck

LoBiondo-Wood

. A systematic review of dyadic studies examining depression in couples facing breast cancer. J Psychosoc Oncol. 2020;38(4):463-480.

13.

Hatfield

Cacioppo

Rapson

. Emotional contagion. Curr Dir Psychol Sci. 1993;2(3):96-100.

14.

Litzelman

Yabroff

. How are spousal depressed mood, distress, and quality of life associated with risk of depressed mood in cancer survivors? Longitudinal findings from a national sample. Cancer Epidemiol Biomarkers Prev. 2015;24(6):969-977.

15.

Litzelman

Al Nassar

. Partner effects on caregiver and care recipient depressed mood: heterogeneity across health condition and relationship type. J Behav Med. 2022;45(5):750-759.

16.

Manne

Badr

. Intimacy and relationship processes in couples’ psychosocial adaptation to cancer. Cancer. 2008;112(S11):2541-2555.

17.

Hagedoorn

Sanderman

Bolks

Tuinstra

Coyne

. Distress in couples coping with cancer: a meta-analysis and critical review of role and gender effects. Psychol Bull. 2008;134(1):1-30.

18.

Streck

Wardell

LoBiondo‐Wood

Beauchamp

. Interdependence of physical and psychological morbidity among patients with cancer and family caregivers: review of the literature. Psychooncology. 2020;29(2):974-989.

19.

Segrin

Sikorskii

Cunicelli

Badger

. Moderators of interdependent psychological distress in cancer survivor-caregiver dyads. J Psychosoc Oncol. 2025;43:574-592. doi:10.1080/07347332.2025.2450012

20.

Pinquart

Sörensen

. Spouses, adult children, and children-in-law as caregivers of older adults: a meta-analytic comparison. Psychol Aging. 2011;26(1):1-14.

21.

Thompson

Ketcher

Gray

Kent

. The Dyadic Cancer Outcomes Framework: a general framework of the effects of cancer on patients and informal caregivers. Soc Sci Med. 2021;287:114357. doi:10.1016/j.socscimed.2021.114357

22.

Tennstedt

Crawford

McKinlay

. Determining the pattern of community care: is coresidence more important than caregiver relationship? J Gerontol. 1993;48(2):S74-S83. doi:10.1093/geronj/48.2.S74

23.

Badr

Bakhshaie

Chhabria

. Dyadic interventions for cancer survivors and caregivers: state of the science and new directions. Semin Oncol Nurs. 2019;35(4):337-341.

24.

Kent

Mollica

Buckenmaier

Smith

. The characteristics of informal cancer caregivers in the United States. Semin Oncol Nurs. 2019;35(4):328-332.

25.

Ellis

Hecht

Northouse

. Symptom distress and quality of life among Black Americans with cancer and their family caregivers. Psychooncology. 2021;30(8):1356-1365.

26.

Schwartz

Beebe-Dimmer

Hastert

, et al. Caregiving burden among informal caregivers of African American cancer survivors. J Cancer Surviv. 2021;15:630-640.

27.

Hastert

Kyko

Ruterbusch

, et al. Caregiver costs and financial burden in caregivers of African American cancer survivors. J Cancer Surviv. 2024;18(2):565-574.

28.

Hastert

Ruterbusch

Nair

, et al. Employment outcomes, financial burden, anxiety, and depression among caregivers of African American cancer survivors. JCO OP. 2020;16(3):e221-e233. https://doi.org/10.1200/JOP.19.00410

29.

Siegel

Giaquinto

Jemal

. Cancer statistics, 2024. CA: Cancer J. 2024;74(1):12-49. doi:10.3322/caac.21820

30.

Emerson

Golightly

Aiello

, et al. Breast cancer treatment delays by socioeconomic and health care access latent classes in Black and White women. Cancer. 2020;126(22):4957-4966.

31.

Giaquinto

Miller

Tossas

Winn

Jemal

Siegel

. Cancer statistics for African American/Black people 2022. CA: Cancer J. 2022;72(3):202-229. doi:10.3322/caac.21718

32.

Bailey

Feldman

Bassett

. How structural racism works—Racist policies as a root cause of US racial health inequities. NEJM. 2021;384:768-773. doi:10.1056/NEJMms2025396

33.

Bailey

Krieger

Agénor

Graves

Linos

Bassett

. Structural racism and health inequities in the USA: evidence and interventions. Lancet. 2017;389(10077):1453-1463. doi:10.1016/s0140-6736(17)30569-x

34.

Samuel

Pinheiro

Reeder-Hayes

, et al. To be young, Black, and living with breast cancer: a systematic review of health-related quality of life in young Black breast cancer survivors. Breast Cancer Res Treat. 2016;160(1):1-15.

35.

Paskett

Alfano

Davidson

, et al. Breast cancer survivors’ health-related quality of life: racial differences and comparisons with noncancer controls. Cancer. 2008;113(11):3222-3230. doi:10.1002/cncr.23891

36.

Martin

Denyse

Pageot

, et al. “I'm fighting for my life”: exploring interactions between Black women with breast cancer and healthcare providers. Qual Health Res. 2023;33(8-9):753-764. doi:10.1177/10497323231168581

37.

Daly

Olopade

. A perfect storm: how tumor biology, genomics, and health care delivery patterns collide to create a racial survival disparity in breast cancer and proposed interventions for change. CA: Cancer J. 2015;65(3):221-238.

38.

Advani

Ying

Theriault

, et al. Ethnic disparities in adherence to breast cancer survivorship surveillance care. Cancer. 2014;120(6):894-900.

39.

Ellis

Black

Baker

, et al. Racial differences in the influence of health care system factors on informal support for cancer care among Black and White breast and lung cancer survivors. Fam Community Health. 2020;43(3):200-212. doi:10.1097/fch.0000000000000264

40.

Lewis

Sheng

Rhodes

Jackson

Schover

. Psychosocial concerns of young African American breast cancer survivors. J Psychosoc Oncol. 2012;30(2):168-184.

41.

Thompson

Davis

Perez

Jonson-Reid

Jeffe

. “We’re in this together”: perceived effects of breast cancer on African American survivors’ marital relationships. J Soc Soc Work Res. 2022;13(4):789-815. doi:10.1086/713478

42.

Mouzon

Taylor

Chatters

. Gender differences in marriage, romantic involvement, and desire for romantic involvement among older African Americans. PLoS One. 2020;15(5):e0233836.

43.

Flannery

Yoo

Levine

. Keeping us all whole: acknowledging the agency of African American breast cancer survivors and their systems of social support. Support Care Cancer. 2019;27(7):2625-2632.

44.

Davis

Nyamathi

Abuatiq

Fike

Wilson

. Understanding supportive care factors among African American breast cancer survivors. J Transcult Nurs. 2018;29(1):21-29.

45.

Thompson

Coats

Croston

, et al. An exploration of the needs and experiences of informal caregivers for Black breast cancer patients. SSM-Qual Res Health. 2022;2:100097.

46.

Haynes-Maslow

Allicock

Johnson

L-S

. Cancer support needs for African American breast cancer survivors and caregivers. J Cancer Educ. 2016;31(1):166-171.

47.

Mollica

Nemeth

. Transition from patient to survivor in African American breast cancer survivors. Cancer Nurs. 2015;38(1):16-22. doi:10.1097/NCC.0000000000000120

48.

Tan

Waters

Chen

Hendricks

Coombs

Kent

. Inequities among cancer caregivers with diverse identities: a review of the literature and future directions. Curr Oncol Rep. 2023;25(7):803-812.

49.

Kenny

Kashy

Cook

. Dyadic Data Analysis. The Guilford Press; 2006.

50.

Lucas

. Why the cross-lagged panel model is almost never the right choice. Adv Meth Pract Psychol Sci. 2023;6(1):25152459231158378.

51.

Kim

Given

. Quality of life of family caregivers of cancer survivors. Cancer. 2008;112(S11):2556-2568.

52.

Thompson

Pérez

Kreuter

Margenthaler

Colditz

Jeffe

. Perceived social support in African American breast cancer patients: predictors and effects. Soc Sci Med. 2017;192:134-142.

53.

Kelley

Kent

Litzelman

Mollica

Rowland

. Dyadic associations between perceived social support and cancer patient and caregiver health: an actor‐partner interdependence modeling approach. Psychooncology. 2019;28(7):1453-1460.

54.

Von Elm

Altman

Egger

Pocock

Gøtzsche

Vandenbroucke

STROBE Initiative . The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. Ann Intern Med. 2007;147(8):573-577.

55.

Radloff

. The CES-D Scale: a self-report depression scale for research in the general population. Appl Psychol Meas. 1977;1(3):385-401. doi:10.1177/014662167700100306

56.

Ashing-Giwa

Rosales

. A cross-cultural validation of patient-reported outcomes measures: a study of breast cancers survivors. Qual Life Res. 2013;22(2):295-308.

57.

Conerly

Baker

Dye

Douglas

Zabora

. Measuring depression in African American cancer survivors: the reliability and validity of the Center for Epidemiologic Study—Depression (CES-D) Scale. J Health Psychol. 2002;7(1):107-114.

58.

Bevans

Ross

Cella

. Patient-Reported Outcomes Measurement Information System (PROMIS): efficient, standardized tools to measure self-reported health and quality of life. Nurs Outlook. 2014;62(5):339-345.

59.

Hays

Bjorner

Revicki

Spritzer

Cella

. Development of physical and mental health summary scores from the patient-reported outcomes measurement information system (PROMIS) global items. Qual Life Res. 2009;18(7):873-880.

60.

Seneviratne

Bozkurt

Patel

, et al. Distribution of global health measures from routinely collected PROMIS surveys in patients with breast cancer or prostate cancer. Cancer. 2019;125(6):943-951.

61.

Sherbourne

Stewart

. The MOS social support survey. Soc Sci Med. 1991;32(6):705-714.

62.

Thompson

Rodebaugh

Perez

Schootman

Jeffe

. Perceived social support change in patients with early stage breast cancer and controls. Health Psych. 2013;32(8):886-895. doi:10.1037/a0031894

63.

RAND Corporation. Medical Outcomes Study : Social Support Survey scoring instructions. 2025. Retrieved from. https://www.rand.org/health-care/surveys_tools/mos/social-support/scoring.html

64.

Ford

Bergmann

Boeing

Capewell

. Healthy lifestyle behaviors and all-cause mortality among adults in the United States. Prev Med. 2012;55(1):23-27. doi:10.1016/j.ypmed.2012.04.016

65.

Kvaavik

Batty

Ursin

Huxley

Gale

. Influence of individual and combined health behaviors on total and cause-specific mortality in men and women: the United Kingdom health and lifestyle survey. Arch Intern Med. 2010;170(8):711-718. doi:10.1001/archinternmed.2010.76

66.

Liu

Wen

Shrubsole

, et al. Impacts of poverty and lifestyles on mortality: a cohort study in predominantly low-income Americans. Am J Prev Med. 2024;67(1):15-23. doi:10.1016/j.amepre.2024.02.015

67.

Warren Andersen

Zheng

Sonderman

, et al. Combined impact of health behaviors on mortality in low-income Americans. Am J Prev Med. 2016;51(3):344-355. doi:10.1016/j.amepre.2016.03.018

68.

National Institutes of Health . National Cancer. Institute Health Information National Trends Survey. Retrieved from. https://hints.cancer.gov/data/survey-instruments.aspx.

69.

Thompson

McQueen

Croston

, et al. Social needs and health-related outcomes among Medicaid beneficiaries. Health Educ Behav. 2019;46(3):436-444. doi:10.1177/1090198118822724

70.

SPSS 29.0.1.0. IBM Corp; 2023. https://www.ibm.com/products/spss-statistics

71.

Muthén Lkm

. Mplus User's Guide. 8th ed. Muthén & Muthén; 1998-2017.

72.

Hinnen

Ranchor

Sanderman

Snijders

Hagedoorn

Coyne

. Course of distress in breast cancer patients, their partners, and matched control couples. Ann Behav Med. 2008;36(2):141-148. doi:10.1007/s12160-008-9061-8

73.

Graham

. Missing data analysis: making it work in the real world. Annu Rev Psychol. 2009;60:549-576.

74.

Shaffer

Chow

Glazer

, et al. Feasibility of ecological momentary assessment to study depressive symptoms among cancer caregivers. Psychooncology. 2021;30(5):756-764. doi:10.1002/pon.5627

75.

Langer

Romano

Todd

, et al. Links between communication and relationship satisfaction among patients with cancer and their spouses: results of a fourteen-day smartphone-based ecological momentary assessment study. Front Psychol. 2018;9:1843. doi:10.3389/fpsyg.2018.01843

76.

Thompson

Lero

Levitan

, et al. Insights about dyadic cancer survivorship interventions for Black women and their caregivers: a rapid qualitative analysis of collaborator perspectives. Cancer Control. 2024;31:10732748241305583.

77.

Ellis

Raji

Olaniran

Alick

Nichols

Allicock

. A systematic scoping review of post-treatment lifestyle interventions for adult cancer survivors and family members. J Cancer Surviv. 2022;16:1-24.

78.

Northouse

Mood

Schafenacker

, et al. Randomized clinical trial of a brief and extensive dyadic intervention for advanced cancer patients and their family caregivers. Psychooncology. 2013;22(3):555-563.

79.

Badger

Segrin

Sikorskii

, et al. Randomized controlled trial of supportive care interventions to manage psychological distress and symptoms in Latinas with breast cancer and their informal caregivers. Psychol Health. 2020;35(1):87-106.

80.

Odom

Young

Sterba

, et al. Developing a national implementation strategy to accelerate uptake of evidence-based family caregiver support in U.S. cancer centers. Psychooncology. 2024;33(1):e6221. doi:10.1002/pon.6221

81.

Brown

. An introduction to linear mixed-effects modeling in R. Adv Meth Pract Psychol Sci. 2021;4(1):1-19. doi:10.1177/2515245920960351

82.

Hilbert

Stadler

Lindl

Naumann

Bühner

. Analyzing longitudinal intervention studies with linear mixed models. TPM-Test Psychom Met. 2019;26(1):101-119.

83.

Song

Qan'ir

Guan

, et al. The challenges of enrollment and retention: a systematic review of psychosocial behavioral interventions for patients with cancer and their family caregivers. J Pain Symptom Manage. 2021;62(3):e279-e304.

Longitudinal Dyadic Analysis of Psychosocial and Health Outcomes in Black Breast Cancer Patients and Their Informal Caregivers

Abstract

Introduction

Methods

Results

Conclusion

Plain Language Summary

Keywords

Introduction

Methods

Sample and Procedures

Measures

Depressive symptoms (primary outcome)

Physical and mental health quality of life (secondary outcomes)

Perceived social support (secondary outcome)

Health behaviors (secondary outcome)

Demographic data

Cancer staging

Analysis Plan

Power

Missing data

Preregistration and modifications to preplanned analyses

Results

Depressive Symptoms

Post hoc analyses for depressive symptoms

Physical Health Quality of Life

Post hoc analyses for physical health quality of life

Mental Health Quality of Life

Post hoc analyses for mental health quality of life

Perceived Social Support

Post hoc analyses for social support

Health Behaviors

Discussion

Strengths and Limitations

Conclusions

Footnotes

Acknowledgements

ORCID iDs

Ethical considerations

Author Contributions

Funding

Declaration of Conflicting Interests

References