Data for this secondary-analysis study were from two institutional review-board-approved (98-0476B, 101-0898A3) longitudinal studies on associations of EOL caregiving with and the effectiveness of an advanced care planning intervention on caregivers’ depressive symptoms before and after patient death [31]. Details have been reported on participant eligibility, subject enrollment and participation in postloss surveys (including validation of between-study and between-participation-status homogeneity), measures, and identification of distinct depressive-symptom trajectories from EOL caregiving through the first 2 bereavement years [31]. In brief, adult caregivers of terminally ill cancer patients (recognized as having a progressing, metastatic cancer by their oncologist) were recruited by convenience from a medical center in Taiwan from 2010 to 2017 and followed through their first 2 bereavement years (end July 2020). Detailed participant characteristics are Table 5 in Appendix 1. As reported in our published article [31], among those 661 caregivers who participated in bereavement surveys, numbers (bereavement follow-up rates) of bereaved caregivers who participated at 1, 3, 6, 13, 18, and 24 months postloss are as follows: 616 (93.2%), 520 (78.7%), 510 (77.2%), 471 (71.3%), 418 (63.2%), and 382 (57.8%), respectively. Among these 661 caregivers, 187 (28.3%), 227 (34.3%), 280 (42.4%), 361 (54.6%), 470 (71.1%), and 625 (94.6%) provided data during each of the last 6 months preloss, respectively.

Caregivers were interviewed in person for preloss caregiving experiences (i.e., depressive symptoms, stressors, stress appraisal, internal/external resources, and prognostic awareness) at enrollment and approximately monthly thereafter while providing EOL caregiving until they declined to participate. Time invariant covariates were assessed at baseline. Bereaved caregivers were surveyed for depressive symptoms at 1, 3, 6, 13, 18 and 24 months postloss, whereas internal/external resources were measured at 1-month postloss only.

Depressive symptoms were measured with the 20-item Center for Epidemiologic Studies Depression Scale (CES-D) [32], with four subscales: positive emotions, depressive emotions, physical activities, and social difficulties. Items are scored for frequency during the past week by a 4-point Likert scale from 0 to 3. Total scores range from 0 to 60; higher scores indicate more depressive symptoms [32]. CES-D scores >16 are considered the threshold for severe depressive symptoms [32]. Prevalence of severe depressive symptoms among the 661 participants increased substantially over the patient’s last 6 months as the patient’s death approached from 9.1% to 69.0% and decreased steadily over the first 2 bereavement years from 69.1% at 1 month postloss to 8.6% at 24 months postloss (data not shown).

We previously [31] used latent-class growth analysis (LCGA) to identify 7 distinct gender- and age-adjusted depressive-symptom trajectories by their timing, intensity, and duration of depressive symptoms (prevalence): minimal-impact resilience (20.4%), recovery (34.0%), preloss-depressive only (21.6%), delayed symptomatic (9.1%), relief (5.9%), prolonged symptomatic (6.5%), and chronically persistent distressed (2.5%) (Supplemental Fig. 1, Appendix 2).

Drawing from the stress-appraisal-coping model, modifiable factors associated with caregivers’ membership in the seven depressive-symptom trajectories were examined in three domains [25]: (1) stressors, (2) stress appraisal, and (3) available resources. Caregiver responses to variables in these categories could differ at various data-collection times, making them time-varying variables.

Stressors indicated as objective caregiving demands were measured by the amount of assistance provided in personal care, homemaking, transportation, and health care on a 4-point scale (1=none at all to 4=a lot) [33]. Caregiving demand scores range=4-16; higher scores indicate greater caregiving demands. To comprehensively reflect the dynamic nature of objective caregiving demands throughout the patient’s last 6 months, objective-caregiving-demand trajectories were identified using LCGA with a continuous latent-class indicator (total objective caregiving demand score) to identify stressor-trajectories experienced by caregivers over their entire EOL caregiving period. Four objective-caregiving-demand trajectories were identified by their intensity and patterns of change: profound, decreasing, increasing, and low objective caregiving demands using the LCGA procedures and criteria for selecting the optimal number of classes prior indicated [31].

Stress appraisal refers to appraisal of caregiving impact. The 24-item Caregiver Reaction Assessment (CRA) [34] measures EOL-caregiving impact on caregivers’ schedule, health, and finances, as well as caregiver esteem in providing care (rewarding or causing resentment), and lack of family support. Total scores range from 24 to 120. Higher scores represent stronger negative appraisal of EOL-caregiving impact. Similarly, to reflect the dynamic nature of subjective appraisal of caregiving impact, we identified caregivers’ CRA trajectories using LCGA with a continuous latent-class indicator (total CRA score) to identify stress-appraisal trajectories experienced by caregivers over their entire EOL caregiving period. Four CRA trajectories were identified and were named low, moderate, high, and substantial caregiving burden using the LCGA procedures and criteria for selecting the optimal number of classes previously indicated [31]. These four trajectories differed in their CRA-score levels. Detailed information about use of latent class-growth analysis to identify optimal number of distinct objective-caregiving-demand and CRA trajectories is available upon request via the corresponding author.

Caregivers’ available resources were internal coping capacity and external social support. Caregivers’ coping capacity was measured by the 13-item sense of coherence scale [35]. Sense of coherence, an important coping capacity for adjusting to stressors to restore homeostasis, has three components: comprehensibility (belief that life is structured and predictable), meaningfulness (life is challenging but worthwhile), and manageability (resources suffice to manage challenges) [35]. Total sense of coherence scale scores range from 13 to 91; higher scores indicate stronger sense of coherence.

Caregivers’ perceived social support was measured by the 19-item Medical Outcomes Study Social Support Survey (MOS-SSS) [36], which assesses emotional, informational, tangible, and affectionate support, as well as positive social interaction. Total scores are computed on a 0-100 scale; higher scores indicate stronger perceived social support.

Contextual factors [25] were treated as covariates and included caregivers’ marital status (married vs unmarried), relationship with the patient (spouse/adult child/other), educational attainment (< vs > senior high school), financial status (making ends meet [yes/no]), comorbidity (yes/no), and awareness of the patient’s prognosis. Caregiver prognostic awareness was measured at each assessment by asking caregivers whether they knew their relative’s prognosis, and if so, whether the disease (1) was curable; (2) might recur in the future, but the patient’s life was not currently in danger; or (3) could not be cured and the patient would probably die soon. Caregivers were recognized as accurately knowing the patient’s prognosis only if they chose option 3; inaccurate prognostic awareness reflected not knowing the patient’s prognosis or choosing option 1 or 2.

Potential factors associated with caregivers’ membership in the seven distinct depressive-symptom trajectories (see Independent variables) were examined by multinomial logistic regressions while controlling for covariates/contextual factors. Multinomial logistic regressions are flexible enough to allow the researcher to alter the reference category, allowing for all possible group comparisons, while the overall model statistics remain unchanged. Preloss scores of internal and external resources were measured at the first assessment within patient’s last 6 months, whereas postloss scores were measured 1-month postloss. This timeline reflected the two-phase piecewise estimation used to model changes in depressive symptoms from EOL caregiving through the first 2 bereavement years, with time of transitioning into EOL caregiving and time of loss serving as major life transitions for caregivers [31]. The regression estimate for each independent variable in the multinomial logistic-regression models was exponentiated to transform to adjusted odds ratio (AOR) with 95% confidence interval (CI).