سوگ و داغداری پیچیده به عنوان یک اختلال پاسخ به استرس: ارزیابی معیارهای تشخیصی در یک نمونه آلمانی

Abstract Background Complicated grief has been described as a diagnosis candidate for DSM-V. On the basis of the stress response theory, Horowitz et al. [Am J Psychiatry 154 (1997) 904–10] characterized complicated grief as a combination of sustained intrusion, avoidance, and maladaptation symptoms following the loss of a close person. This study aimed at evaluating diagnostic criteria based on the stress response model of complicated grief. Methods We administered a symptom list derived from Horowitz et al.'s operationalization to a sample of bereaved persons and evaluated the psychometric properties of the symptom criteria and symptom category subscales. Using this symptom list and other self-report measures of psychopathology and normal grief reactions, we examined a German sample consisting of 75 participants who had lost either siblings, children, parents, or spouses, on average, 5.4 years prior to the study. Results Analyses confirmed the classification of symptoms into intrusion, avoidance, and failure-to-adapt categories with only minimal reordering (two symptom criteria). The symptom category subscales showed favourable psychometric characteristics, receiver operating characteristic (ROC) analyses indicated high diagnostic accuracy of the symptom criteria, and predictive validation revealed a meaningful correlational pattern to standard measures of divergent psychopathology and normal grief reactions. Conclusions The application of a stress response operationalization of complicated grief is supported.

Introduction For some time, various authors have been proposing the introduction of the new diagnosis of complicated grief, which could help clinicians to properly recognize and treat a syndrome that is not adequately covered in the established nosology of DSM-IV [2], [3], [4] and [5]. Recent studies using independent bereavement samples demonstrated that the symptoms of complicated grief form a syndrome that is distinct from normal grief reactions, as well as depression and anxiety (e.g., Refs. [6] and [7]). Horowitz and his collaborators [1], [8] and [9] developed a concept for the diagnosis of complicated grief that is based on the stress response theory. Accordingly, complicated grief is, in part, similar to posttraumatic stress disorder (PTSD), characterized by processes that result in three sets of symptoms: intrusions, avoidance, and failure to adapt to the loss. This model is supported by research on the role of intrusions and avoidance in the mourning process [10]. The stress response concept and its operationalization seeks to identify symptoms deviating from normal mourning processes, resulting in pathologically impaired functional capacity. So far, only one study has explicitly examined and confirmed this concept by using latent class model analysis and signal detection procedures [1]. A stress response operationalization was chosen as outcome variable in a series of studies dealing with the grieving process [11], [12], [13], [14], [15], [16], [17] and [18]. Tomita and Kitamura [19] published supportive evidence for the stress response formulation of complicated grief in a Japanese sample. Kersting et al. [20] applied a modified version for diagnosing complicated grief in women after induced abortion. The present study analyses additional evidence for the stress response model of complicated grief in a German sample. To this end, we used the list of symptoms constituting the three symptom categories of Horowitz et al. [1], extended by four additional symptoms suggested by Horowitz (personal communication, 2001). It was examined (1) whether the classification of symptoms into the three categories could be replicated; (2) whether a useful and psychometrically sensible reduction of symptom criteria could be achieved; (3) whether there was support for Horowitz et al.'s diagnostic algorithm consisting of seven symptoms; (4) whether meaningful associations of the symptom category subscales with other standard measures of psychopathology (anxiety, depression, posttraumatic stress, and general) and a measure for normal grief reactions could be found; and (5) whether there were any interpretable predictive associations between the symptom category subscales and age, sex, kind of relationship, predictability of death, and time since death.

Results Kolmogorov–Smirnov tests indicated that scores of several scales were not normally distributed. Therefore, nonparametric methods were generally employed for statistical analyses. Item characteristics, item reduction and reordering, and scale properties First, we examined the extended 34-item CGM as to its item characteristics and efficiency of the symptom category scales. For evaluation, the following parameters were used in this order: (i) Cronbach's alpha coefficient of each symptom category compared with the alpha coefficient of each symptom category if any of this category's items was deleted; (ii) effect size (ES) of extreme-group comparisons of each item as a measure of internal construct validity; (iii) item discriminations and the correlations of each item with the scores of the other two symptom categories; and (iv) the interitem correlations. The aims of these psychometric analyses were to examine the assignment of items to the symptom categories and to exclude any item from the CGM that did not suffice our demands on the parameters used. The distributional parameters and ES of all items are given in Table 1. (i) Coefficient alphas were satisfactory to good for each CGM symptom category (αCGM-Intrusions=.87; αCGM-Avoidance=.77; αCGM-Failure=.86) and could not significantly be improved on by deleting any single item. (ii) The between-group ES was calculated as a measure of how well each item discriminates between participants, with high versus low grief severity. The two groups to be discriminated were formed as follows: Participants were divided by the median split of each symptom category score. Those participants whose scores in all three symptom categories were below the respective median formed the “low-symptomatic group” (LSG, n=23), and those with scores equal to or above the median in all three categories formed the “highly symptomatic group” (HSG, n=23). The between-group ES of each item was computed using the following formula: (meanHSG−meanLSG)/pooled standard deviation [30]. Items with an ES below d=0.8 were excluded from the CGM, which applied to three of them (see Table 1). (iii) To further examine the internal structure of the CGM, item discriminations and correlations with the other two subscales were calculated for each item. Two items originally belonging to the Failure-to-Adapt category (“Keeping deceased's possessions the same” and “Difficulty concentrating”) were assigned to different symptom categories (Intrusions and Avoidance, respectively) because their correlations with these categories' scores were higher by .05 than the ones with their original categories (see Footnote d in Table 1). Three items (“Seeing others who look like deceased”, “Feeling alienated from others”, and “Feeling worthless”) correlated similarly highly with a second subscale, apart from their own (difference of Rs<.05; see Table 1), but were retained in their original symptom categories on theoretical grounds. Items with a low discrimination (i.e., a correlation with their own symptom category of R<.50) and low correlations with any other symptom category (Rs<.50) were excluded from further analyses, which applied to eight of them (including the three items previously excluded on the basis of their low ES; see Table 1). (iv) By examining interitem correlations, we checked on possible redundancies, but the highest single correlation between two items (“Feeling alone or empty” and “Feeling worthless”) was R=.71. Thus, no item needed to be excluded for reasons of efficiency. The resulting 26-item version of the CGM can be found in the right part of Table 1. The following numbers of items remained in the symptom categories after the analyses: Intrusions, 13 items; Avoidance, 7 items; and Failure to Adapt, 6 items. Altogether, eight items were excluded from the extended version of the CGM and further analyses according to the selection criteria described above. The internal consistency, as measured by Cronbach's alpha, of each symptom category of the shortened CGM version can be considered good: αCGM-Intrusions=.88; αCGM-Avoidance=.82; αCGM-Failure=.84. None of these coefficient alphas would have increased if any item had been deleted. To further examine the internal structure of the shortened CGM, we recalculated item-subscale Spearman correlations for each item and symptom category (see Table 1, middle columns). The result shows relatively high item discrimination values but also substantial correlations with the other two subscales for most items, respectively. This bears out the notion of three distinct but overlapping symptom categories representing three facets of one disorder. This overlap between the categories is further underlined by the relatively high CGM subscale intercorrelations (see Table 2). Table 2. Spearman correlations among the shortened CGM symptom categories, other psychopathology measures, and predictors of complicated grief 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 1. CGM Intrusions .76 .73 .55 .58 .60 .42 .63 .55 .53 .12 −.06 .04 2. CGM Avoidance .74 .61 .64 .56 .47 .68 .65 .55 .15 .04 .02 3. CGM Failure to Adapt .70 .65 .55 .39 .75 .62 .56 .30 .12 −.12 4. BDI .74 .55 .47 .77 .77 .46 .34 .27 −.14 5. BAI .45 .29 .72 .73 .47 .24 .18 −.03 6. IES-R Intrusions .49 .78 .51 .55 .46 .11 −.27 7. IES-R Avoidance .51 .37 .49 .07 .02 .00 8. IES-R Hyperarousal .75 .61 .30 .04 −.19 9. SCL-90-R (GSI) .40 .18 .13 −.16 10. TRIG .00 .20 −.22 11. Own age at death .57 −.33 12. Deceased's age at death −.28 13. Time since death BDI=Beck Depression Inventory; BAI=Beck Anxiety Inventory; IES-R=Impact of Event Scale-Revised; SCL-90-R=Symptom Checklist-90-Revised; GSI=global severity index; TRIG=Texas Revised Inventory of Grief. All correlations R>.36 are significant at P≤.01; all correlations R>.26 are significant at P≤.05. Table options Examination of the diagnostic algorithm by Horowitz et al. [1] In a next step, we aimed to replicate the results of the study of Horowitz et al. [1] concerning the selection of a short set of diagnostic criteria. By means of a latent class model analysis, Horowitz et al. selected a “gold standard” (i.e., the best diagnostic criterion) from four different measures of grief-specific distress. Their analysis revealed that an above-median number of CGM items scored as severe was the best criterion for diagnosing complicated grief. It correctly classified 100% of cases at 14 months after the bereavement. Subsequent signal detection analyses resulted in the selection of four CGM items (“Strong yearning”, “Feeling alone or empty”, “Avoiding reminders of deceased”, and “Trouble sleeping”), whose combination by an “OR” rule achieved considerable diagnostic power (sensitivity=1.00; specificity=0.71; total predictive value=0.86). Additional analyses further improved specificity to 0.78 by adding three more items (“Low interest in important activities”, “Emotional spells”, and “Unbidden memories”) and applying an algorithm that combined the presence of any three of these seven symptoms. All seven items are marked in Table 1 (far right column). Utilizing the gold standard of Horowitz et al. [1] for the diagnosis of complicated grief, we calculated the sensitivity, specificity, and total predictive value of each item of the shortened CGM, employing receiver operating characteristic (ROC) analyses. The total predictive value is given by the nonparametrically estimated area under the ROC curve (AUC). The AUC value corresponds to the probability that a randomly selected pair of observations drawn from the two underlying distributions (complicated vs. normal grief reactions) will be classified correctly [31]. A value of .50 indicates chance level. The results of our ROC analyses are shown in the right part of Table 1. In accordance with Horowitz et al., symptoms had lower sensitivity than specificity, which indicated that, while any one symptom occurred relatively infrequently, symptoms tended to be rated as severe by those participants who reported multiple symptoms of complicated grief. The total predictive values of all items significantly exceeded chance performance (AUC values ranging from .69 to .84), supporting the diagnostic value of each symptom of the shortened CGM. The application of the “OR” combination rule to the abovementioned four diagnostic criteria in our sample yielded results comparable with that of Horowitz et al. ([1]; sensitivity=0.85; specificity=0.82; total predictive value [AUC]=0.93). Our implementation of their algorithm of combining any three of the selected seven symptoms (as described above) resulted in an expected shift of the diagnostic parameters (sensitivity=0.60; specificity=0.99; AUC=0.96). The increase in specificity corresponds to the data of Horowitz et al., who did not report the sensitivity and total predictive value of this “three-out-of-seven” combination algorithm but expected a decrease in sensitivity due to an overinclusion of diagnostic criteria. This is further underscored by the high total predictive value, as expressed by the AUC. In sum, we successfully replicated the findings of Horowitz and his colleagues, which supports their approach to selecting criteria for the diagnosis of complicated grief. Predictive validity of CGM subscales For the assessment of the predictive validity of the CGM, Spearman correlations between CGM symptom categories and various measures of psychopathology were computed. These measures included the BDI, BAI, IES-R, and SCL-90-R. The results of these analyses, as can be seen in Table 2, showed moderate to high correlations between the three CGM symptom categories and all other measures, reflecting a substantial covariation of the different psychopathological constructs. The correlations between the CGM subscales and the TRIG were moderate (Rs=.53–.56). To estimate the influence of depression and posttraumatic stress symptoms on sustained grief reactions, we calculated regression analyses for predicting each CGM subscale from the BDI and the respective “corresponding” IES-R subscale scores. The BDI and IES-R Intrusion scores explained 44.8% of variance of the CGM Intrusion subscale, the BDI and IES-R Avoidance scores explained 32.8% of variance of the CGM Avoidance subscale, and 54.0% of variance of the CGM Failure to Adapt subscale were explained by the BDI and IES-R Hyperarousal scores. Thus, the construct of complicated grief appears to be lying somewhere in the range of depression and PTSD symptomatology but, at the same time, can be considered a distinct phenomenon. Furthermore, associations of the symptom category scores with the participants' or deceased's age at the bereavement, the time (number of years) passed since the death, possible influences of gender, predictability of the loss, and kind of relationship to the deceased were examined. A significant correlation was found between the participants' age at the death and the CGM Failure to Adapt symptom domain (see Table 2). No associations were found between symptom severity and the deceased's age at the bereavement, or time passed since the death, respectively (see Table 2). Comparing female with male participants, Mann–Whitney U tests did not reveal any significant gender differences in both complicated grief severity (CGM Intrusions: U=174.0, P=.56; CGM Avoidance: U=180.0, P=.65; CGM Failure to Adapt: U=179.5, P=.64) and all other measures of psychopathology or normal grief reactions (BDI: U=176.0, P=.43; BAI: U=193.5, P=.76; IES-R Intrusions: U=118.5, P=.11; IES-R Avoidance: U=183.5, P=.98; IES-R Hyperarousal: U=144.5, P=.34; SCL-90-R [GSI]: U=206.0, P=.90; TRIG: U=125.5, P=.15). The comparison of participants who anticipated the loss with those who did not yielded no significant differences in the severity of grief symptoms (CGM Intrusions: U=149.5, P=.19; CGM Avoidance: U=165.5, P=.35; CGM Failure to Adapt: U=156.0, P=.25). There were also no significant differences between these subgroups in the IES-R subscale scores (IES-R Intrusions: U=256.0, P=.94; IES-R Avoidance: U=225.0, P=.48; IES-R Hyperarousal: U=214.0, P=.35), which showed that the different circumstances of foreseeable versus unforeseeable deaths did not result in different posttraumatic symptoms, which, in turn, might have overlapped the grief reaction. When comparing participants who had lost a brother or sister with those who had lost a different member of their immediate families, it was found that the former generally rated the severity of their symptoms lower than the latter did (CGM Intrusions: U=294.5, P=.01; CGM Avoidance: U=236.0, P=.001; CGM Failure to Adapt: U=265.0, P=.003). No differences were found between participants who had lost a sibling of the same sex and those who had lost a sibling of the opposite one (CGM Intrusions: U=163.0, P=.48; CGM Avoidance: U=169.5, P=.59; CGM Failure to Adapt: U=151.0, P=.29).