Validation of the new DSM‐5‐TR criteria for prolonged grief disorder and the PG‐13‐Revised (PG‐13‐R) scale (2023)

Abstract

Although the concept of pathological grief dates back at least as far as Freud’s “Mourning and Melancholia”, there has been opposition to its recognition as a distinct mental disorder. Resistance has been overcome by evidence demonstrating that distinctive symptoms of prolonged grief disorder (PGD) – an attachment disturbance featuring yearning for the deceased, loss of meaning and identity disruption – can endure, prove distressing and disabling, and require targeted treatment. In acknowledgement of this evidence, the American Psychiatric Association Assembly has recently voted to include PGD as a new mental disorder in the DSM‐5‐TR. We tested the validity of the new DSM criteria for PGD and of an adapted version of our PG‐13 scale, the PG‐13‐Revised (PG‐13‐R), designed to map onto these criteria, using data from investigations conducted at Yale University (N=270), Utrecht University (N=163) and Oxford University (N=239). Baseline assessments were performed at 12‐24 months post‐loss; follow‐up assessments took place 5.3‐12.0 months later. Results indicated that the PG‐13‐R grief symptoms represent a unidimensional construct, with high degrees of internal consistency (Cronbach's alpha = 0.83, 0.90 and 0.93, for Yale, Utrecht and Oxford, respectively). The DSM PGD diagnosis was distinct from post‐traumatic stress disorder (phi=0.12), major depressive disorder (phi=0.25) and generalized anxiety disorder (phi=0.26) at baseline. Temporal stability was remarkable for this diagnosis (r=0.86, p<0.001). Kappa agreement between a PG‐13‐R threshold symptom summary score of 30 and the DSM symptom criterion for PGD was 0.70‐0.89 across the datasets. Both the DSM PGD diagnosis and the PG‐13‐R symptom summary score at baseline were significantly associated (p<0.05) with symptoms and diagnoses of major depressive disorder, post‐traumatic stress disorder and/or generalized anxiety disorder, suicidal ideation, worse quality of life and functional impairments at baseline and at follow‐up, in the Yale, Utrecht and Oxford datasets. Overall, the DSM‐5‐TR criteria for PGD and the PG‐13‐R both proved reliable and valid measures for the classification of bereaved individuals with maladaptive grief responses.

Keywords: Prolonged grief disorder, DSM‐5‐TR, PG‐13‐R, ICD‐11, pathological grief, bereavement, post‐traumatic stress disorder

Although the concept of pathological grief dates back at least as far as Freud’s Mourning and Melancholia1, there has been public and professional opposition to its recognition as a mental disorder2, 3, 4, 5. For example, a 2015 international online survey of public attitudes revealed that approximately 25% of respondents did not endorse the position that grief could be a mental disorder2. More recently, an online survey on public opinion in China found that about 40% of participants did not agree that grief could be a mental disorder, even under circumstances such as threat of harm to self or others4. Concerns about “pathologizing” grief are reported to be rooted in the belief that all grief is normal and an expected response to the death of a loved one. Thus, a diagnosis of pathological grief is considered to be tantamount to stigmatizing, medicalizing and/or pathologizing love2, 4.

Himself wary of pathologizing grief, Freud conceptualized mourning (grief) as a normal, natural reaction to loss of a loved one, and even deemed working through grief as necessary to bereavement adjustment – the hard, often painful, work a mourner must do to withdraw emotional attachment to the deceased person. In fact, Freud considered medical interference in “grief work” to be “inadvisable if not even harmful”1. By contrast, he considered melancholia (i.e., depression) the pathological response to bereavement, and noted that this condition, not grief, posed a risk for suicide, and warranted medical attention.

Research over the past quarter century has shown not only that a small but substantial proportion of grief reactions can be severe, disabling, and endure beyond normal expectations, but that they may respond only to specialist treatment. Specifically, studies have documented that certain grief symptoms are distinct from those of bereavement‐related depression6, 7, 8, 9, have idiosyncratic neurobiological10 and clinical11, 12, 13 correlates, can persist unabated for months or even years8, 14, prove distressing and dysfunctional14, 15, 16, and may only respond to targeted intervention17, 18. Thus, there exists a substantial and mounting body of evidence in support of a psychiatric syndrome of maladaptive grief.

The ICD‐11 Workgroup on Stress‐Associated Disorders found the available evidence for prolonged grief disorder (PGD) sufficiently compelling to recommend its recognition as a new mental disorder19. The DSM‐5 had included “persistent complex bereavement disorder” (PCBD) in Section III (i.e., among “conditions for further study”). In response to the ICD’s inclusion of PGD and the accumulated evidence, the DSM Steering Committee convened a workshop in June 2019. An invited panel of researchers presented their data to the Committee, who concluded that these data supported moving the disorder to Section II (i.e., among recognized mental disorders). A provisional PGD criteria set was then drafted, and the researchers were tasked with using the best data available to inform the parameters of the PGD diagnostic algorithm, and then to evaluate that algorithm's reliability and validity. The researchers submitted their reports, which found the same PGD diagnostic algorithm to be optimal. The Steering Committee then posted that PGD algorithm online on the American Psychiatric Association (APA)’s website and opened a period for public commentary between April and May 2020. After reviewing the research reports and submitted comments, the Steering Committee released the proposed criteria, and on November 7, 2020, the APA Assembly approved the inclusion of PGD in the DSM‐5‐TR (see Table1).

In order to be sensitive to the concern expressed in the public commentary about pathologizing normal grieving and diagnosing a grief‐related disorder “too soon” after the death, the DSM‐5‐TR PGD criteria specify that 12 months must elapse since the death. This time frame contrasts with the ICD‐11 diagnostic guidelines for PGD, requiring a period of 6 months20. Unlike the PCBD criteria, the DSM‐5‐TR criteria for PGD acknowledge the possibility of delayed onset of symptoms at or beyond 12 months post‐loss. Furthermore, the PGD criteria require that three of eight C criteria (compared to PCBD's six of 12) be met for a diagnosis, and focus more on “yearning for” and preoccupation with the deceased person and less on “preoccupation with the circumstances of the death” – the latter of which could be captured by a post‐traumatic stress disorder (PTSD) diagnosis. Lastly, the PGD diagnosis allows for fewer combinations of symptoms to meet the criteria compared to the PCBD diagnosis. An empirical analysis of the performance of these new DSM criteria for PGD has not been published, nor has the psychometric performance of a scale that maps onto these diagnostic criteria been evaluated.

The PG‐13 scale22 was introduced in the process of developing PGD diagnostic criteria proposed for inclusion in the DSM‐5 and ICD‐118. The scale contains 13 items that can be used for the dual purposes of assessing grief intensity continuously on a dimensional scale and of diagnosing PGD according to the proposed criteria. Items in the PG‐13 are a subset of those in the Inventory of Complicated Grief ‐ Revised23, which is a revision of the Inventory of Complicated Grief7. Included items were those that we found to be informative and unbiased with respect to gender, relationship to the decedent, and time from loss in item response theory‐based item analysis, and which mapped onto our criteria for PGD proposed in 20098.

The present paper has two primary objectives. First, it aims to introduce and validate the PG‐13‐R, a revised version of the PG‐13 scale that corresponds to the new DSM‐5‐TR criteria for PGD. Second, it aims to validate these new DSM criteria for PGD. Data from the US (the Yale Bereavement Study), the Netherlands (the Utrecht Bereavement Study), and the UK (the Oxford Grief Study) were used to evaluate the psychometric properties of the PG‐13‐R, determine its agreement with the new DSM criteria for PGD, assess the PG‐13‐R and DSM criteria’s predictive validity, and establish a threshold PG‐13‐R score to identify syndromal level PGD.

METHODS

RESULTS

Table2 summarizes the demographic characteristics of the three study samples. The Yale sample was older (mean age: 61.8±13.5 years) than the Utrecht (mean age: 56.2±13.3 years) and Oxford (mean age: 46.9±13.3 years) ones. All three samples were primarily female (73.0 to 79.1%), and most survived a death from natural causes (compared to unnatural causes such as suicide or homicide or accidental) (>90%). The Yale and Oxford samples had higher levels of educational attainment (college or above >60%) than the Utrecht sample (college or above <40%).

The mean scores for each PG‐13‐R symptom item at T1 are presented in Table3. They ranged from 1.3 to 2.9 in the Yale study; from 1.9 to 3.8 in the Utrecht study; and from 1.8 to 3.2 in the Oxford study. In general, most item means were located around the center of the range, which is an indication of desirable variability. The avoidance (Q7) and preoccupation (Q4) items were infrequent in the Yale study, where mean scores in general were low. Variances for most items across the datasets were reasonably high, confirming the scale’s discriminating ability.

Across studies, the PG‐13‐R symptom items cohered well (Cronbach’s alpha = 0.83 for Yale, 0.90 for Utrecht, 0.93 for the Oxford study) (see Table3). This analysis revealed that the deletion of the avoidance item in each of the three datasets resulted in either the same or an improved overall Cronbach's alpha (deleted alpha = 0.84, 0.91, 0.93 for the Yale, Utrecht and Oxford, respectively). Similarly, while all the other items had high item‐total correlations (r ≥ 0.50, 0.56 and 0.69 for the three datasets, respectively), the avoidance item was an exception, with lower item‐total correlations (r=0.25, 0.33, 0.52, respectively).

As illustrated in Figure2, principal components factor analysis in combination with parallel analysis for each dataset supported the conclusion that the PG‐13‐R grief symptoms represent a unidimensional construct. In fact, in each dataset, a single factor emerged whose eigenvalue was substantially larger than 1 and greater than would be expected by chance. This primary factor explained 40.3%, 53.5% and 61.8% of the variance in the Yale, Utrecht and Oxford studies, respectively.

Results in Table4 support the external validity of the PG‐13‐R symptom score, not including the impairment item (Q13). PG‐13‐R symptom scores at T1 were significantly associated with PTSD, MDD and/or GAD diagnoses or symptomatology and suicidal ideation, both concurrently (p<0.001) and predictively (p<0.05), in the Yale, Utrecht and Oxford data. PG‐13‐R symptom scores were significantly associated with poorer role‐emotional and mental health domains of quality of life both concurrently and predictively in the Yale data (p<0.005), and with work and social adjustment difficulties both concurrently and predictively in the Oxford data (p<0.001).

PG‐13‐R symptom threshold scores of 29, 32 and 30 maximized agreement with meeting DSM symptom criteria for PGD in the Yale (kappa=0.77), Utrecht (kappa=0.86), and Oxford (kappa=0.89) study data, respectively. Overall, a symptom threshold score of 30 optimized agreement with meeting DSM symptom criteria for PGD across the three datasets (kappa ≥0.70 across the datasets).

Results in Table5 illustrate that using a PG‐13‐R symptom threshold score of 30 in combination with the impairment criterion demonstrated excellent external validity. The prevalence of PGD using the PG‐13‐R score ≥30 at T1, including impairment, was 6.3%, 16.6% and 11.3% for the Yale, Utrecht and Oxford samples, respectively. The PG‐13‐R threshold‐based diagnoses of PGD at T1 were significantly (p<0.05) associated with PTSD, MDD and/or GAD diagnoses or symptomatology and suicidality in the Yale, Utrecht and Oxford data, concurrently and predictively (except for suicidality in the Utrecht study, where the association was significant only concurrently). PG‐13‐R threshold‐based diagnoses of PGD were significantly associated with poorer role‐emotional and mental health domains of quality of life both concurrently and predictively in the Yale data (p<0.05), and with work and social adjustment difficulties both concurrently and predictively in the Oxford data (p≤0.001).

Results in Table6 illustrate that the DSM diagnosis of PGD demonstrated excellent external validity. The prevalence of PGD using DSM criteria at T1 was 4.4%, 15.3% and 10.9% for the Yale, Utrecht and Oxford samples, respectively. DSM diagnoses of PGD at T1 were significantly (p<0.05) associated with PTSD, MDD and/or GAD diagnoses or symptomatology concurrently and predictively in the Yale, Utrecht and Oxford data. Interestingly, in the Yale sample, DSM diagnoses of PGD were significantly associated with suicidality predictively (at T2) but not concurrently (at T1). DSM diagnoses of PGD were significantly associated with poorer vitality, role‐emotional and mental health domains of quality of life both concurrently and predictively in the Yale data (p<0.05), and with work and social adjustment difficulties both concurrently and predictively in the Oxford data (p≤0.001).

In the Yale data (T1, N=270), the DSM PGD diagnosis was found to be distinct from PTSD (phi=0.12), MDD (phi=0.25) and GAD (phi=0.26). Temporal stability (T1, T2 correlation; N=48) was greatest for DSM PGD (r=0.86, p<0.001), significant for MDD (r=0.31, p=0.030), and not significant for GAD (r=–0.07, p=0.653). We could not estimate the temporal stability for PTSD because no participants with T2 data met criteria for PTSD at T1 (and only one study participant met criteria for PTSD at T2).

DISCUSSION

Results of analyses of data from independent Yale, Utrecht and Oxford bereavement studies suggest that both the PG‐13‐R and the DSM‐5‐TR PGD diagnostic criteria possess desirable performance characteristics. The symptoms were uniformly higher in the Utrecht sample, which is unsurprising given that this sample was recruited via mental health professionals. Across all three datasets, the preoccupation item was infrequently reported at syndromal levels. This was most noticeable in the Yale data, where syndromal level preoccupation was found in <3% of the sample. Such low prevalence is an undesirable property for a “gatekeeper” item, which suggests that it might have been preferable to have only “yearning” in the B criterion for PGD in the DSM.

The weakest performing item across all the datasets was “avoidance of reminders that the deceased is dead”. Item‐total correlations for this item were the lowest of all items examined, and Cronbach’s alpha improved in the Yale and Utrecht datasets when the avoidance item was removed. It may be the case that avoidance is more a function of fear, with roots in psychological trauma, than a function of grief, with roots in an attachment disturbance. Alternately, there may be a need to revise the item to focus on what aspect of the loss is avoided (e.g., avoidance of reminders of the death as an event may be more a traumatic stress response, while avoidance of reminders that the deceased is truly gone may be the most relevant to disturbed grief). Future studies are needed to confirm whether the avoidance item should be retained, revised or discarded.

In accordance with the high internal consistency of the PG‐13‐R symptom items, factor analyses revealed that the scale is unidimensional. These results are consistent with those reported for the Inventory of Complicated Grief7 and its Dutch version35, and for the original PG‐138 and its Swedish36, Chinese37, Portuguese38 and many other translated versions^e.g.,39. Though some studies have found multiple factors in this set of grief symptoms40, these exceptions occurred only in highly comorbid treatment‐seeking and treatment‐receiving samples and a military family study, not in community‐based samples. The preponderance of evidence supports the unidimensional nature of PGD symptomatology as found in the three studies examined here.

Because the Yale data alone included structured clinical interviews that yielded diagnoses of mental disorders, only these data could be used to assess PGD’s overlap with other disorders and to compare diagnostic stability over time. The results demonstrated minimal overlap between PGD and competing diagnoses (i.e., PTSD, MDD and GAD) (phi=0.12‐0.26), suggesting its distinctness from mental disorders already included in Section II of the DSM. In addition, the PGD diagnosis proved remarkably stable between the T1 and T2 assessments approximately 7.4 months apart (r=0.86, p<0.001) and much more stable than MDD (r=0.31, p=0.030) or GAD (r=–0.07, p=0.653). These results suggest that PGD fills a diagnostic gap left open by other mental disorders secondary to bereavement. Furthermore, they show that PGD is likely not to remit with the passage of time and to require specialized treatment.

With respect to concurrent and predictive validity, we first sought to determine if the intensity of PGD symptoms alone (excluding impairment, the DSM criterion D) would predict distress and dysfunction. The PG‐13‐R symptom score proved to be highly predictive of both concomitant and future distress and dysfunction, indicating that the severity of these symptoms themselves is pathological even without “stacking the deck” by requiring the fulfillment of an impairment criterion.

Next, we sought to determine the threshold score of these symptoms that optimized agreement with meeting the B and C symptom criteria for PGD in the DSM. We found that the PG‐13‐R symptom score of 30 was the optimal threshold score across the three datasets. Finally, we sought to evaluate and compare the concurrent and predictive validity of diagnoses for PGD using the PG‐13‐R threshold diagnostic score, and, separately, using the DSM criteria B and C, each in combination with meeting the impairment criterion. Results indicated that both performed extremely well in predicting substantial current and future maladaptive behaviors and outcomes.

A strength of this study was the use of three independent community‐based bereavement cohort samples. A possible weakness was the fact that the wording for the PG‐13‐R questions was slightly different in the three studies. The Utrecht sample was uniformly more distressed than the Yale and Oxford samples, which is understandable given that Utrecht participants were recruited via mental health care providers, who are more likely to encounter distressed bereaved individuals. The Yale and Utrecht samples were predominantly comprised of widowed persons, which was not the case for the Oxford sample (~80% to ~30%, respectively). With respect to ethnicity, all three samples nearly entirely consisted of people of Caucasian ethnicity.

In conclusion, three independent community‐based samples showed that the PG‐13‐R is a reliable tool for assessing grief symptoms on a dimensional scale. A PG‐13‐R symptom score of 30 or greater identifies syndromal‐level PGD symptomatology. The dimensional PG‐13‐R symptom score, the diagnosis of PGD using the PG‐13‐R threshold symptom score of 30 plus the impairment criterion, and the diagnosis of PGD using the new DSM‐5‐TR criteria all predict enduring distress and dysfunction. Thus, the PG‐13‐R and the new DSM‐5‐TR criteria for PGD appear to be reliable and valid measures for the classification of bereaved individuals with maladaptive grief responses. Future research is needed to confirm their psychometric performance in more ethnically diverse samples.

ACKNOWLEDGEMENTS

This work was supported by grants from the US National Cancer Institute (nos. CA197730 and CA218313), the US National Institute of Minority Health and Health Disparities (no. MD007652), the US National Institute of Nursing Research (no. NR018693), the US National Institute on Aging (no. AG049666), the US National Institute of Mental Health (no. MH121886), the US National Center for Advancing Translational Science (no. TR002384), the Wellcome Trust (no. 200796), the National Institute for Health Research (NIHR) Biomedical Research Centre, based at Oxford University Hospitals National Health System (NHS) Trust (no. NIHR‐INF‐0085), and the Oxford Health NIHR Biomedical Research Centre. The views expressed are those of the authors and not necessarily those of the supporting institutions. The authors are grateful to A. Ehlers for her support and supervision of the Oxford Grief Study. A smart pdf version of the PG‐13‐R is available at https://endoflife.weill.cornell.edu/sites/default/files/file_uploads/pg‐13‐r.pdf.

REFERENCES