Introduction

The impact on families of ICU bereavement, an increasingly [1] frequent [2] loss experience in the post-COVID-19 era [3], can be severe and prolonged. For a minority of persons, bereavement may precipitate an umbrella of severe, disturbed grief reactions, including symptoms of prolonged grief disorder (PGD) [4], posttraumatic stress disorder (PTSD) [5], and depression [6]. Distressed psychological grief reactions more often co-occur than not [7, 8]. If after ICU death at-risk family members go unnoticed, healthcare professionals miss opportunities to treat or prevent psychological distress, leading to unnecessary suffering.

Understanding the temporal relationships between grief-related psychological distress may help disentangle the mechanism of mental comorbidity [9]. Early grief-related symptoms may contribute to other grief reactions in later stages. Identifying psychological distress promptly in the preceding period and targeting it with appropriate symptom management may block subsequent development/exacerbation of long-lasting comorbid psychological distress and functional impairment. Yet, knowledge about temporal relationships between PGD, PTSD, and depression is limited with most of the focus on PTSD and depression [10,11,12,13,14,15,16], less on PGD and PTSD [17,18,19,20], and least on PGD and depression [19, 21, 22]. Only one study [19] was found to explore the temporal relationships between PGD, PTSD, and depression simultaneously following traumatic losses from an airplane crash. Indeed, temporal relationships between PGD, PTSD, and depression cannot be deduced from studies that only examined reciprocal relationships between two psychological grief-related responses.

Importantly, methodological inconsistencies were identified in these studies [10,11,12,13,14,15,16,17,18,19,20,21,22]. In previous studies [10,11,12,13,14,15,16,17,18,19,20,21,22], 2–4 waves of assessments were collected across periods from 0.5–7 months [11, 22] to 18–35 years post-event [13]. Such heterogeneity makes comparison challenging. Further, PGD is defined as a spectrum of grief symptoms occurring at least 6 months after the death as per the minimum duration criterion [23]. Assessments made earlier (1–4 [22] or 4–5 [20] months post-loss) may invalidate the PGD assessment. Therefore, the purpose of this study was to establish an appropriate time interval reflecting the PGD duration criterion and longitudinally determine the temporal reciprocal relationships between symptoms of PGD, PTSD, and depression among ICU family surrogates during their first 2 bereavement years to determine a potential mechanism of comorbidity.

Materials and methods

Study design/setting/study participants

Data for this study came from a longitudinal, observational study on associations between quality of end-of-life (EOL) ICU care and family surrogates’ bereavement outcomes, including symptoms of anxiety [24], depression [24], and PTSD [25]. Information on sampling strategy, patient and family characteristics, and study settings was reported [24, 25]. At two academically affiliated Taiwanese hospitals, a consecutive sample was recruited from family surrogates responsible for decision-making for critically ill patients at high risk of death (APACHE II scores > 20) in level III medical ICUs staffed by intensivists from January 2018 to March 2020 and followed through July 2022. Each surrogate signed informed consent for participation and review of the patient’s medical record. The study site’s research ethics committee approved this study (201700343B0).

Data collection

Participant characteristics were collected through a review of medical records and surrogate interviews at enrollment. Subsequent phone interviews to assess surrogate psychological distress at 1, 3, 6, 13, 18, and 24 months post-loss complied with the minimum duration criterion for PTSD (≥ 1 month) [26] and avoided the anniversary effect at 12 months post-loss. Only data collected since 6 months post-loss were used in this study to accommodate the PGD duration criterion [23].

Measures

PGD symptoms were measured by 11 items of the Prolonged Grief Disorder-13 [23]. The PG-11 includes one separation distress symptom, nine cognitive and emotional symptoms, and one functional impairment symptom but excludes 2 dichotomous items regarding duration and impairment criteria which measure dimensions other than grief symptoms. Participants rate how often symptoms occurred in the preceding month on a 5-point scale (1 = never, 5 = always).

PTSD symptoms were measured by the 22-item Impact of Event Scale—Revised (IES-R) [26], a screening instrument for PTSD symptoms on three subscales: intrusion, avoidance, and hyperarousal. For each item, participants rate their PTSD symptom distress level during the preceding week on a 0 (not at all)-4 (extremely) Likert scale [26].

Depressive symptoms were measured by the seven-item depression subscale of the Hospital Anxiety and Depression Scale (HADS-D) [27].HADS-D total scores range from 0 to 21 [27].

Data analysis

A series of cross-lagged panel modeling [28] (CLPM) was conducted to examine the temporal reciprocal relationships between PGD, PTSD, and depressive symptoms across the four assessment waves (6–24 months post-loss) by structural equation modeling in Mplus 8.6. CLPM is a method for describing the reciprocal relationships among multiple variables between a series of timepoints. The autoregressive coefficient quantifies inter-individual stability for a specific variable (e.g., do PGD symptoms at Tn predict PGD symptoms at Tn + 1), whereas the cross-lagged coefficient quantifies the association between the prior score of one variable and the subsequent scores of the other variables (e.g., do PGD symptoms at Tn predict PTSD or depressive symptoms at Tn + 1), thereby indicating the temporal relationship between two variables [28, 29]. Cross-lagged coefficients are calculated while controlling for autoregressive coefficients and for correlations among the examined variables at each wave [28, 29].

To obtain the most parsimonious CLPM, the statistical fit of an unconstrained model (Model 1) was compared with constrained models [19]: Model 2 assumed the autoregressive paths for each symptom variable are equal, Model 3 assumed both the autoregressive and the cross-lagged paths are equal, and Model 4 included Model 3 assumptions and assumed associations among the three symptom variables are consistent across each wave. Model fit was assessed by the following indices: (1) comparative fit index (CFI) > 0.90 [30], (2) Tucker–Lewis index (TLI) > 0.90 [30], (3) root-mean-square error of approximation (RMSEA) < 0.10 [31], (4) standardized root-mean-square residual (SRMR) < 0.05 [32], and (5) smallest Akaike information criterion (AIC), Bayesian information criterion (BIC), and sample-size adjusted BIC (SABIC) [33]. We selected the best-fitting model and controlled suggested covariates [19]: gender (0: female, 1: male), kinship to deceased (0: others, 1: spouse, 2: adult child), and educational level (0: ≤ senior high school, 1: > senior high school). Missing data on symptom levels of PGD, PTSD, and depression were handled using full information maximum likelihood (FIML) estimation which produces the least biased estimates of missing values among other imputation methods [34]. We adopted benchmark values synthesized for cross-lagged effects of CLPM: 0.03 (small effect), 0.07 (medium effect), and 0.12 (large effect) [35]. All significance tests were two-tailed at p < 0.05 level.

Results

Participant characteristics

Among the 353 patients who died in the ICUs, 321 family surrogates (90.9%) participated in bereavement surveys (Fig. 1). Of them, 303 (94.4%) provided data 6–24 months post-loss and constituted the study sample; 292, 277, 275, and 261 surrogates completed surveys at 6, 13, 18, and 24 months post-loss, respectively (Fig. 1). The reasons for attrition are detailed in Fig. 1. Characteristics of the sample and their relatives are in Tables 1 and 2, respectively. Surrogates (Additional file 1: Table S1, Additional file 2: Table S2) who completed, withdrew, or skipped surveys since 6 months post-loss and their respective relative (Additional file 3: Table S3) did not significantly differ. Examination of missing symptom data by FIML [34] showed no examinations were significant (Little’s missing completely at random testing [χ2/degree of freedom, p value] for PGD [30.557/20, p = 0.061], PTSD [12241/19, p = 0.875], and depression [26.303/19, p = 0.122]), indicating data were missing completely at random. Thus, analyses in this study were based on the total sample.

Fig. 1
figure 1

Participant flowchart

Full size image
Table 1 Comparison between surrogates who did or did not participate in 6–24 months bereavement surveys at enrollment (N = 353)
Full size table
Table 2 Comparison between patients whose family surrogate did or did not participate in 6–24 months bereavement surveys (N = 353)
Full size table

Longitudinal temporal relationships between symptoms of PGD, PTSD, and depression

The unconstrained model was the best-fitting model (Table 3) with data fitting fairly well: CFI = 0.967, TLI = 0.921, and SRMR = 0.047, except for RMSEA = 0.131. Autoregressive standardized coefficients for PGD (0.570–0.673), PTSD (0.375–0.687), and depression (0.591–0.655) symptoms were remarkably stable (Fig. 2).

Table 3 Model fit indexes for different models of temporal reciprocal relationships between PGD, PTSD, and depressive symptoms
Full size table
Fig. 2
figure 2

Significant reciprocal relationships between symptoms of depression, PTSD, and PGD by cross-lagged panel modeling from 6 to 24 months post-loss. Depressive symptoms were measured with the depression subscale of the Hospital Anxiety and Depression scale. PTSD symptoms were measured by the Impact of Event Scale—Revised. PGD symptoms were measured by 11 items of the Prolonged Grief Disorder-13. Autoregressive coefficients along horizontal arrows indicate stability of each variable, and cross-lagged coefficients along diagonal arrows indicate strength of temporal relationships between symptoms with autoregressive coefficient controlled. Cross-lagged effect sizes: 0.03 (small effect), 0.07 (medium effect), and 0.12 (large effect). Gender, kindship to the deceased, and education level were adjusted. *p < .05; **p < .01; ***p < .001

Full size image

Cross-lagged standardized coefficients showed that depressive symptoms measured at 6 months post-loss predicted subsequent symptoms of PGD (0.146) and PTSD (0.208) at 13 months post-loss (Fig. 2). PGD symptoms did not predict depressive symptoms. PTSD symptoms predicted subsequent depressive symptoms in the second bereavement year (0.175–0.278). PGD symptoms consistently predicted subsequent PTSD symptoms in the first 2 bereavement years (0.180–0.263), whereas PTSD symptoms predicted subsequent PGD symptoms in the second bereavement year only (0.190–0.214). PGD and PTSD symptoms are bidirectionally related in the second bereavement year. All cross-lagged standardized coefficients (0.146–0.278) had a large effect.

Discussion

Our study unprecedently investigated temporal relationships between PGD, PTSD, and depressive symptoms among ICU bereaved surrogates by CLPM. Most studies [10,11,12,13,14,15,16,17,18,19,20,21,22] on temporal relationships between PGD, PTSD, and/or depression examined survivors of natural disasters [12, 14,15,16] or anthropogenic traumatic events [10, 11, 13]. Studies on bereavement explored loss from natural disease [17, 21, 22] or pure/mixed unnatural disasters [18,19,20]. Temporal relationships between PGD, PTSD, and/or depression during grieving likely differ by the context of loss. This study contributes several novel observations on bereaved ICU surrogate grief reactions that implicate clinical practice.

PTSD and depressive symptoms are temporally related

Depressive symptoms measured at 6 months post-loss predicted PTSD symptoms at 13 months post-loss (Fig. 2), whereas long-lasting PTSD symptoms predicted depressive symptoms beyond the first bereavement year. Literature on the temporal relationship between symptoms of PTSD and depression in the first year is inconclusive: depressive symptoms predicted subsequent PTSD symptoms between 1 and 3 months posttraumatic accidents [11]; and PTSD symptoms predicted subsequent depressive symptoms in 1.5–12 months after a tornado [15] or earthquake [16]. For the second year, previous results are consistent with ours: more pervasive PTSD symptoms predicted depressive symptoms at 24 months post-earthquake [16]. Results beyond 2 years post-event contrast with ours: depressive symptoms predicted PTSD symptoms 17–35 years post-combat [13], and PTSD and depression were bidirectionally related among bereaved about 1–4 years post-airplane crash [19]. Further research is needed to understand how the ICU context influences the temporal relationship between symptoms of depression and PTSD in the first 2 bereavement years.

A depressogenic model theorizes how depression leads to PTSD [11]. At the entry into bereavement, permanent separation from an important source of emotional support and interpersonal security [36] triggers separation distress—missing the deceased and grief over an impossible physical reunion [37, 38]. Compounded with doubt [39] and inadequate or conflicting social support during ICU EOL care decision-making [40, 41], surrogates may experience negative self-perceptions, helplessness, anhedonia, or negative worldview. Consequently, these depressive symptoms may hinder bereaved surrogates’ motivation and ability to manage exposure to trauma-related stimuli, leading to cognitive and behavioral avoidance [42] of trauma-related reminders, especially around the anniversary.

Stress sensitivity [43] may also predispose depressed, bereaved surrogates to PTSD symptoms, especially approaching the anniversary of loss. External stressors, like loss, can leave individuals with neurobiological vulnerabilities that increase sensitivity to subsequent stressors [43], like memories of the ICU-care experience evoked by the anniversary [28, 37]. Sensitization to this circumstance-related distress [28, 37] may elicit excessive cognitive or behavioral avoidance of loss reminders, thereby increasing bereaved surrogates’ vulnerability to PTSD symptoms at 13 months post-loss [38, 44].

Avoidance that stalls processing and acceptance of loss [38, 44] can entrench PTSD symptoms beyond the anniversary and subsequently predict surrogates’ depressive symptoms in the second bereavement year (Fig. 2). A demoralization model [11] and internalized trauma [37, 38, 44] connect PTSD to subsequent depression. Internalized trauma feels like excessive guilt and self-blame (as after EOL-care decisions like withdrawing or withholding life-sustaining treatments), demoralization, meaninglessness, and negative beliefs about oneself, the world, and the future [37, 38, 44]. Thus, long-lasting PTSD symptoms preceding prolonged depressive symptoms in the second bereavement year may indicate internalized trauma.

Depressive symptoms predict PGD symptoms in the first year

Depressive symptoms measured at 6 months post-loss predicted PGD symptoms at 13 months post-loss (Fig. 2). Previously, no reciprocal relationships were observed between PDG and depression from approximately 1–4 years of bereavement [19] while other studies on loss from protracted advanced illness, often during hospice/palliative care, [21, 22] were inconsistent. Temporal relationships between PGD and depression were bidirectional at 1–4 months post-loss [22] while PGD predicted subsequent depression at 4–7 [22] and 6–24 [21] months post-loss. Future research should consider controlling characteristics of death (expected vs unexpected, chronic vs acute, clinical context) to determine their influence on different temporal relationships between PGD and depression.

We speculate that separation distress [37, 38] during early bereavement might sensitize depressed surrogates to loss reminders from the forthcoming anniversary [43], bringing on characteristic PGD symptoms: yearning for the eternally separated deceased, frequent preoccupying thoughts and memories of the deceased person, a feeling of disbelief or inability to accept the loss, and difficulty imagining a meaningful future without the deceased. The absence of social support may also prompt PGD symptoms, given the fragility of interpersonal relationships among those with profound depression [45]. With reluctance to accept outside support, isolation and loneliness may evolve into characteristic PGD symptoms like difficulty trusting people and feeling distant from others.

PGD symptoms predict PTSD symptoms; PGD and PTSD symptoms are bidirectional in the second year

PGD symptoms consistently preceded PTSD symptoms over the first 2 bereavement years (Fig. 2), consistent with three previous studies [19, 20, 22]. Separation distress [37, 38] and stress sensitization theory [43], as described above, may similarly explain why PGD symptoms consistently predicted heightened PTSD symptoms. Furthermore, from the perspective of cognitive behavioral models of PGD [46], persistent yearning for the deceased and difficulties accepting the loss may interfere with processing and managing exposure to trauma-related reminders, leading to cognitive and behavioral avoidance and the development of PTSD symptoms. Difficulty trusting others, prolonged withdrawal from meaningful social activities, and meaninglessness in life may further worsen PTSD symptoms.

We observed that symptoms of PTSD predicted subsequent PGD symptoms beyond the first bereavement year, consistent with the report for survivors of a terrorist attack [20]. A meta-analysis indicated the prevalence of PGD is much higher (49% [95% confidence interval (CI) 33.6, 65.4]) in people bereaved by unnatural or traumatic deaths [47] than by losses primarily due to diseases (9.8% [95% CI 6.8–14.0]) [48]. Glad and colleagues’ [20] also found more pervasive PTSD symptoms predicted subsequent PGD beyond the first year. Long-lasting PTSD symptoms like cognitive and behavioral avoidance [46] may impede emotional processing and prolong separation distress, thereby initiating or prolonging PGD symptoms. Therefore, suffering prolonged PTSD symptoms after a traumatic ICU loss is an important risk factor for PGD.

Study strengths and limitations

The strength of this study lies in its use of CLPM for four-wave longitudinal data measured between 6 and 24 months post-loss to explore the complex temporal reciprocal relationships between PGD, PTSD, and depressive symptoms for family surrogates who lost a critically ill relative in the ICU. Timely identification and alleviation of depression and PGD symptoms during early bereavement may enable bereaved surrogates to grieve more effectively and avoid long-lasting PGD, PTSD, and depression (via PTSD) (Fig. 2).

However, several study limitations should be recognized. Our study findings may not be generalized to (inter)national populations beyond the sampled hospitals or surrogates who did not participate in or withdrew from bereavement surveys despite no observed differences in baseline characteristics and bereavement data missed completely at random. We may overestimate bereaved surrogates’ psychological distress by measuring PGD, PTSD, and depressive symptoms with screening rather than diagnostic tools, but this measurement strategy may avoid overlooking their need for emotional support. Our measurement of PGD by the PG-11 does not fully assess prolonged grief according to the criteria in DSM-V (PCBD) or ICD-11 (PGD) [49]. Time frame of our assessments of grief reactions limits understanding the temporal reciprocal relationships between PGD, PTSD, and depression beyond the first 2 bereavement years.

Conclusions

PGD symptoms at 6 months after bereavement may contribute to the exacerbation of PTSD symptoms over time, whereas long-lasting PTSD symptoms disrupt the normal grief process and precipitate prolonged depressive and PGD symptoms beyond the first bereavement year. Healthcare professionals should assess PGD and depression among bereaved surrogates as soon as possible after an ICU death (e.g., 6 months post-loss). Elevated PGD and depressive symptoms early in bereavement can be considered a risk factor for prolonged PTSD. Our novel observation of the bidirectional relationship between symptoms of PGD and PTSD in the second bereavement year indicates that symptoms of PGD and PTSD beyond the first bereavement year may fuel each other, precluding recovery from loss. Targeting elevated PGD and PTSD symptoms may alleviate these grief-related responses.