Rebuilding Consensus Around Valid Criteria for Disordered Grief

Are "Complicated Grief" (CG) criteria superior to "Prolonged Grief Disorder" (PGD) criteria in diagnosing grief disorder? – No.

PGD criteria ought to be the standard for diagnosing disordered grief

In 2009, following the completion of an NIH/NIMH-funded investigation of consensus criteria for disordered grief, we published validated criteria for a new diagnostic entity, prolonged grief disorder (PGD).¹ Based on the results of this field trial, the Yale Bereavement Study (YBS), we proposed PGD’s inclusion to DSM-5 and ICD-11. The strength of the evidence in support of PGD from the YBS, supplemented by confirmatory findings from multiple international studies, compelled both the DSM and ICD to take steps to include a new disorder of grief.

The DSM’s and the ICD’s tentative formulations of diagnostic criteria for a disorder of grief agree almost perfectly with PGD criteria.² They also share PGD’s sound psychometric properties. By contrast, complicated grief (CG) criteria ³ do not agree with DSM and ICD criteria, use abstruse syntax, produce more false positive than true positive diagnoses, and lack predictive validity.² Evidence supports the validity of PGD,¹ and its DSM and ICD derivatives,² but weighs against the validity of CG.² PGD meets established criteria for a mental disorder;⁴ CG does not. Based on existing evidence, PGD criteria ought to be the standard for diagnosing disordered grief.

Methodological misgivings

Presumably, medical journals only publish studies that employ standard, scientifically sound methods. Standard information for evaluating diagnostic criteria includes estimates of diagnostic accuracy (e.g., sensitivity and specificity) and prevalence of disorder. Curiously, this information is absent from the Cozza et al. publication “Performance of DSM-5 Criteria for Persistent Complex Bereavement Disorder…” (PCBD),⁵ which concludes that CG criteria are superior to PGD and PCBD criteria. In the YBS data,² CG criteria produce more false positive cases (63%) than true (37%) positive cases of disorder and have an unacceptably high (30%) overall rate of diagnosis. Based on YBS results, we expect that the unreported overall rate of diagnosis of CG and false positive CG test rate in the entire Cozza et al. sample are unacceptably high.

The results that Cozza et al. do report are inherently “spectrum biased.” Cozza et al. discard nearly half (n=797, 46%) of their total sample to focus on the most obvious “cases” (n=260, 15%) and “controls” (n=675, 39%). Spectrum bias ⁶ in estimates of diagnostic accuracy results from excluding less obvious, borderline cases in favor of extreme, easier to evaluate, cases and controls. Spectrum-biased designs overestimate sensitivity and specificity by omitting diagnostic errors from near-threshold cases. The real test of diagnostic performance is not in identifying extremes, but rather in discerning in-between, more difficult to evaluate, cases. Cozza et al.’s spectrum-biased design does not account for the many false positive tests for CG that would have appeared in the large excluded segment (46%) of their analytic sample. Most likely, and, if so, consistent with findings from the YBS,² there are more false than true positive tests for CG in Cozza et al.’s full sample.

Prevalence rates and rates of false positives and negatives obtained for the full sample (not a spectrum-biased group of cases and controls) are needed to determine which criteria sets have superior performance. Essential, yet missing, information and use of a spectrum-biased design raise questions concerning the soundness of the science behind the CG proposal.

CG criteria pathologize normal grief

Bereavement is a common, natural life event. It is normal to be upset following the loss of someone loved; heightened vigilance is needed to avoid pathologizing normal reactions. For this reason, diagnostic criteria for grief disorder should prioritize diagnostic specificity (minimizing false positives) over sensitivity (minimizing false negatives). CG criteria, lax both in number and severity of symptoms required for diagnosis, produce many false positives, over-diagnose grief disorder, and pathologize normal grief. Applying the “moderate” symptom severity threshold employed by Cozza et al.⁵ to CG criteria ³ that require only three symptoms for diagnosis, the positive test rate for CG in the YBS sample is 62%. Thus, CG criteria diagnose most bereaved individuals in a community sample as mentally ill – a result undermining their face validity (e.g., laypersons are likely to consider such criteria suspect).

Straw Men arguments

Suggestions that CG criteria are superior to PGD criteria in their embodiment of clinical wisdom or applicability are straw men arguments. These claims strive to shift attention away from issues of paramount importance to diagnostic assessment, i.e., diagnostic validity and accuracy. Simply put, diagnostic validity must be established before proceeding to clinical applications; it does not make sense to discuss clinical applications of invalid diagnoses.

We do not deny that clinical insight is essential to the formulation of psychiatric diagnoses. PGD, like CG, was informed by the clinical insight of prominent psychiatrists (e.g., Drs. Parkes, Horowitz, Jacobs, Shear, and Reynolds). However, clinical opinion alone is insufficient for the validation of diagnostic criteria. Clinical insight was translated into an NIH/NIMH-funded investigation of consensus criteria for PGD that produced compelling evidence of a new diagnostic entity.¹ By contrast, CG criteria were introduced in a review article³ (as opposed to an original research article) without any empirical support, and specifically without evidence of predictive validity.² Given the ease with which CG criteria can be satisfied, they also lack face validity.

The claim that CG criteria, compared with PGD criteria, identify a greater number of individuals who will benefit from treatment is another straw man argument. The main purpose of diagnostic assessment is to determine whether an individual truly has a disorder. To paraphrase DSM-5, accurate diagnosis is a prerequisite for appropriate treatment. Any argument about the superior clinical applicability of CG criteria diverts attention away from diagnostic accuracy to bereaved individuals’ presumed need for clinical care. As Spitzer wrote: “To confuse making a mental disorder diagnosis with demonstrating treatment need [is]… a serious mistake.” ⁷

Moreover, there is no evidence that CG criteria accurately identify bereaved individuals in need of, or likely to benefit from, treatment. We found that CG criteria, unlike PGD criteria, were unrelated to risk of future mental disorder, functional impairment, or diminished quality of life.² Thus, CG criteria do not identify bereaved individuals at risk of enduring dysfunction who might be helped by an intervention. In fact, there is no evidence that CG criteria are better than PGD criteria with respect to any clinical process or outcome. These lines of argument are aimed at drawing attention away from fundamental issues of diagnostic performance (e.g. prevalence, false positive rates, predictive validity) to focus, prematurely, on issues of clinical application.

Evidence suggests abandoning CG and supporting PGD criteria as the diagnostic standard

Data from multiple, independent, community-based datasets tell a consistent, compelling story – one that supports the diagnostic validity and accuracy of PGD criteria. This is not the case for CG criteria. CG criteria lack validity,² produce more false than true positive tests for disorder² and, because they are too easily satisfied, pathologize normal grief. PGD criteria reliably and validly identify bereaved individuals genuinely in need of and likely to benefit from the help of a mental health professional. CG criteria are inadequate and counterproductive, and should be withdrawn from serious consideration. It is time for scientists and clinicians to agree that PGD should be adopted as the standard for diagnosing disordered grief.