ICD-11 Prolonged Grief Disorder Criteria: Turning Challenges Into Opportunities With Multiverse Analyses

IntroductionRecently, prolonged grief disorder (PGD), a diagnosis characterized by severe, persistent and disabling grief, was formally included in the 11th revision of the International Classification of Diseases (ICD-11; World Health Organization, 2018: Table 1). To meet PGDICD-11 criteria one needs to experience persistent and pervasive longing for the deceased and/or persistent and pervasive cognitive preoccupation with the deceased, combined with any of 10 additional grief reactions assumed indicative of intense emotional pain for at least six months after bereavement. Contrary to the 5th revision of the Diagnostical and Statistical Manual of Mental Disorders (DSM-5; American Psychiatric Association, 2013) and the 10th revision of the International Classification of Diseases (ICD-10, World Health Organization, 1993), the ICD-11 only uses a typological approach, implying that diagnosis descriptions are simple and there is no strict requirement for the number of symptoms one needs to experience to meet the diagnostic threshold. Some researchers have argued that PGDICD-11’s typological approach is helpful, as it will lead to greater sensitivity in case identification in clinical practice and increased cross-cultural applicability (Killikelly and Maercker, 2017). Others have highlighted that the typological approach allows for flexible diagnostic algorithms in research, so that PGDICD-11 criteria can be adapted to resemble the characteristics of both stricter and more lenient precursor criteria (Simon et al., 2020). In the current contribution, we take a different, complementary position. We highlight a series of challenges in using the PGD criteria for research purposes and discuss the application of a method that employs flexibility of PGDICD-11 diagnostic approach to address these challenges, which may help in working towards the unbiased, structured, and transparent identification of optimal criteria for disturbed grief. 2. A critique of PGD ICD-11 for research purposesA first challenge to researchers applying PGDICD-11 criteria is that they were completely new when first introduced and differed substantially from previously proposed diagnostic criteria sets (Eisma and Lenferink, 2017). For example, PGDICD-11 contains multiple symptoms not found in any prior proposed criteria set, such as guilt, blame and the inability to experience positive mood (for a full criteria set comparison: Lenferink et al., in press). Furthermore, oft-used measures to assess disturbed grief responses, such as versions of the Inventory of Complicated Grief (e.g., ICG; Prigerson et al., 1995) do not fully assess PGDICD-11 criteria (Eisma for a recent review illustrating this point: Treml, Kaiser, Plexnies, Mauro et al., 2019). It soon became apparent that applying these minimal criteria led to much higher prevalence rates for PGDICD-11 than for prior proposed criteria of PGD (PGD2009; Prigerson et al. 2009) and persistent complex bereavement disorder (PCBD; DSM-5, American Psychiatric Association, 2013). This algorithm is thus relatively lenient, and applying it may lead to overdiagnosis and limited generalizability of findings on two of the most-studied grief disorder proposals (i.e., PGD2009; PCBD) to PGDICD-11 (Eisma and Lenferink, 2017). This elicits the question: If the diagnostic algorithm directly derived from the ICD-11 text is too liberal, which diagnostic rules are then optimal for research? 3. Multiverse analyses in research on PGD ICD-11In summary, a fundamental challenge for grief researchers in using the PGDICD-11 diagnosis is that its criteria are open for multiple interpretations and that the only diagnostic algorithm mapping one-on-one on the diagnosis description is too lenient. While the current criteria cannot easily be amended, their systematic investigation can make them more useful to researchers, for instance by providing a basis for achieving consensus on symptom interpretation, algorithms, and future PGD criteria. We propose that multiverse analyses can be particularly helpful in achieving such goals. Multiverse analyses typically consist of a procedure wherein one performs similar analyses across multiple datasets generated by making reasonable but variable choices on excluding, transforming and coding data (Steegen et al., 2016). For example, when using a reaction time task with skewed data, one may perform analyses based on the median or the mean or analyze the data using parametric or non-parametric statistical tests. By comparing outcomes of multiple analyses, one can establish the degree of uncertainty about the conclusions one arrives at and the robustness of findings to arbitrary decisions made in data preparation and analysis. For example, one may discover that the direction and significance of effects is similar regardless of these decisions or that some decisions lead to significant effects, whereas others do not. The first scenario would allow for strong conclusions and the second scenario would signal caution is warranted in the interpretation of findings. We advocate a similar but conceptually distinct procedure wherein empirical research examining the characteristics of PGDICD-11 systematically vary certain aspects of these criteria (e.g., using a more stringent cut-off for longing) or the diagnostic algorithm (e.g., varying the number of additional symptoms). A comparison of results obtained with multiple interpretations of criteria can help illuminate how robust specific results are dependent on multiple interpretations of the PGDICD-11 criteria. For example, one may be able to investigate the robustness of group differences between people with and without PGD on risk-factors and protective factors or treatment effectiveness (e.g., percentage with and without diagnosis after treatment) dependent on different interpretations of PGDICD-11. Additionally, critical information can be gathered on the influence of variations in symptom interpretations and algorithms on PGDICD-11 characteristics and how these characteristics compare to other proposed criteria sets (e.g., the newly developed PGD criteria for the upcoming text revision of DSM-5, American Psychiatric Association, 2020). That is, multiverse analyses can be applied to shed light on a variety of clinically relevant characteristics of PGDICD-11 (e.g., retest and interrater reliability, specificity and sensitivity of classification, distinctiveness from other disorders, associations with functional impairment) when systematically modifying interpretations of its criteria.Only a handful of studies have thus far applied such analyses, which have predominantly narrowly focused on examining characteristics of PGDICD-11 and their comparability against external standards when varying the number of additional criteria (see Table 1 for a summary). It has been observed that minimal PGDICD-11 criteria yield a similar prevalence as the relatively lenient Shear et al. (2011) criteria for complicated grief (Mauro et al., 2019), but almost two times higher prevalence than relatively strict PCBD criteria (Boelen and Lenferink, 2020; Boelen et al., 2019b; O’Connor et al., 2019). Multiverse analyses in community samples demonstrated that similar prevalence estimates and good diagnostic agreement with PCBD and PGD2009 appears to be achieved with five additional criteria for PGDICD-11 (Boelen and Lenferink, 2020; Boelen et al., 2019a; Bonanno and Malgaroli, 2020; cf. Mauro et al., 2019). Similarly, in treatment-seekers, minimal PGDICD-11 criteria correctly classified people against a relatively lenient standard of 30 or higher on the ICG (Cozza et al., 2019; Mauro et al., 2019), yet as many as six additional symptoms were necessary to yield comparable prevalence and good diagnostic agreement with PGD2009 and PCBD (Comtesse et al., 2020). Moreover, one study demonstrated the influence of the number of additional symptoms on symptom heterogeneity, theoretically demonstrating that the number of ways to meet PGDICD-11 criteria ranges from 3.069 for one to 528 for seven additional symptoms (Boelen and Lenferink, 2020). A less heterogeneous diagnosis is clearly preferable, as it would lead to less variability within groups of people meeting grief disorder criteria, making the distinction between these people more useful for research and practice (Lenferink and Eisma, 2018). Taken together, these examples illustrate that the properties of PGDICD-11 depend on both the chosen diagnostic rule and the stringency of the comparison standard, and that the number of additional symptoms is critical in determining prevalence, clinical classification, and symptom profiles. For future research, we recommend multiverse analyses varying not only the algorithm, but additionally symptom interpretations of single-item criteria and cognitive preoccupation and cut-offs for the presence of the longing criterion. We also advise to substantially expand the current focus of multiverse analyses of PGDICD-11 to establish the robustness of clinically-relevant findings (e.g., on treatment efficacy) and the variety of other aspects relevant to the validity of a diagnosis (for reviews: Killikelly Simon et al., 2020). The latter includes - but is not limited to: reliability of classification (e.g., Lichtenthal et al., 2018), the structure of symptoms (e.g., Golden & Dalgleish, 2010), distinctiveness from related disorders (e.g., Malgaroli et al., 2018), and relationships with functional impairment (e.g., Maccallum & Bryant, 2019). We further advocate transparency in applying multiverse analyses and recommend: open access publication, data accessibility (e.g., through availability in repositories), fully specifying the origins and formulations of items used to assess PGDICD-11 and, if applicable, other disturbed grief criteria, and complete reporting of the variations of PGDICD-11 and outcomes under investigation. 4. DiscussionIn the absence of clearly defined criteria and diagnostic rules for PGDICD-11, researchers should broadly apply structured methods to examine the characteristics of this disorder and compare it against past and future proposed grief disorders. Multiverse analyses can be a powerful tool to determine the validity and clinical usefulness of the PGDICD-11 criteria. By systematically varying the number of core and additional symptoms, the interpretation of symptoms, and the standards for meeting symptoms, we can evaluate how such decisions influence the characteristics of PGDICD-11, also in relation to different external standards. This will create a comprehensive research base enabling us to enhance our understanding of PGDICD-11 and of disturbed grief more generally. Creating this research base is no panacea: it cannot undo the inherent weaknesses of the PGDICD-11 criteria. However, the systematic evaluation of this information will help clarify under which circumstances diagnoses behave similarly or differently, providing a stepping stone to harmonize PGDICD-11 criteria with other criteria sets and to develop more optimal future disturbed grief criteria.