Symptom Presentation of Rural Crisis Patients: Dimensions of Clinician Ratings
Steven Thurber and Eugene Bonynge
During a fifteen month time period, licensed clinicians rated the symptom presentation of patients (185 males; 373 females) evaluated in a crisis unit of a rural community mental health center. There were 618 crisis episodes evaluated on degree of severity (0-4) in the areas of danger to self, danger to others, confusion, depression, and functional decline. Exploratory and confirmatory factor analyses indicated that the patient episodes were mainly related to the severity of depression and suicide proclivities (factor 1) but patients also tended to show general distress across the four rating scales, yielding an oblique two-factor solution. Although infrequently endorsed, the fifth scale, “danger to others” (aggressive tendencies and violence) had loadings on the secondary factor and was considered to be a part of the confirmatory model with the most practical significance.
Keywords: Rural crisis unit, symptom presentation, confirmatory factor analysis
Procedures for dealing with patients in crisis have been developed and evaluated in urban and metropolitan areas and may not apply in rural vicinities in which population density issues limit available options. But regardless of rural-urban considerations, persons in acute distress may have difficulties in verbal communication and in their capacity to complete questionnaires. Oftimes mental health professionals working with such persons cannot obtain standardized test data and hence must rely on behavioral observations, oral descriptions of presenting problems by patients, or reports from significant others (See Bonynge, Lee, & Thurber, 2005).
In an effort to aid clinicians working with patients in a rural crisis delivery service, we relied on several years of crisis experience suggesting that persons who enter a crisis unit tend to have acute problems in at least one of the following domains: Suicidal proclivities, (termed “danger to self”), danger to others, confusion, depression and/or functional decline. In an earlier article (Thurber & Bonynge, 2008), we discussed the functional utility of transforming these five problem areas into 0 (absence)-4 (severe) rating scales. It was found that rural mental health professionals working with patients in crisis used their ratings configurationally in arriving at dispositional recommendations (e.g., referrals to inpatient versus outpatient services). This suggests that these clinicians had considered and perhaps differentially weighted all the rated problem areas in arriving at a dispositional decision.
In the current study, we subjected the rating scales to exploratory and confirmatory factor analysis with the guiding hypothesis that the results would be unidimensional or oblique in nature, corresponding to holistic information processing by crisis clinicians.
The participants were 185 males (Mean age = 36.62) and 373 females (Mean age = 36.47) admitted to a rural community mental health center’s crisis unit over a fifteen month period. There were 618 admissions among these 418 individuals during this time period. Each participant met individually with a state licensed mental health professional who then rated him or her on each of the five distress dimensions. In 387 crisis episodes, 156 assessments were completed by clinical social workers, 106 by doctoral level psychologists, 66 by master level psychologists, and 59 by marriage and family therapists.
An exploratory factor analysis was initially conducted, suggesting an oblique two-factor structure. The primary factor was composed of “danger to self” and “depression” and accounted for 37% of the variance. The three other scales comprised the second factor, accounting for 24% of scale variance. The correlation between the factors indicates the dimensions are associated because of their relationship with another latent variable, a general factor that might be termed “distress.”
We conducted a confirmatory factor analysis that evaluated the two factor oblique structure as well as the possibility of a unitary factor. In addition, the possibility of confirmation of a two factor-orthogonal structure was tested. Further, a series of analyses were computed with and without the variable “danger to others” (dto) because of the low endorsement on this scale by clinical raters. Goodness of fit statistics were computed and the various factor models were compared to each other, to a perfectly fitting “saturation” model, and to an independence model, in which all variables are assumed to be uncorrelated. Because of the skewed distribution (low endorsement) in dto, an unweighted least squares confirmatory method was used.
With reference to table 1, the absolute fit was assessed by the root mean square residual (RMR; Arbuckle, 1997) and the goodness of fit statistic (GFI; Joreskog & Sorborn, 1984). The former refers to the discrepancy between the values in the original variance-covariance matrix and the reproduced matrix based on the proposed model, with values of zero indicating perfect congruence (Arbuckle, 1997). The latter represents the comparison of squared prediction residuals to the actual data, with values close to 1 indicating better fit. Both indices suggested better fit for the two factor oblique models, particularly for the model in which dto was eliminated. Constraining the correlation between the two factors to zero resulted in a significantly poorer absolute fit. Comparative or incremental fit measures compared the hypothesized models to an independence or saturation model. Again, the normed fit index (NFI; Bentler & Bonett, 1980) produced better fit for the two factor oblique models, especially the model without dto. The value of .996 indicates that the hypothesized model strongly departs from independence and is almost congruent (99.6%) with the perfectly fitting saturation model. Regarding parsimony, the adjusted goodness of fit measure (AGFI; Joreskog & Sorborn, 1984) modifies the fit of the model to the degrees of freedom and the number of variables, with values of .90 and above representing acceptable levels. Only the oblique models attained this standard. The parsimony ratio or PRATO refers to the ratio of degrees of freedom in the hypothesized models to the degrees of freedom in the model that assumes complete independence of variables. The orthogonal and unitary models were superior on this measure.
|2 factor oblique||.050||.989||.959||.400||.944|
|2 factor orthogonal||.125||.933||.831||.600||.653|
2 factor oblique without dto
|2 factor orthogonal without dto||.144||.943||.858||.667||.744|
|General factor without dto||.088||.975||.877||.333||.874|
dto=danger to others: RMR= rootmean square residual: GFI = goodness of fit index
AGFI = adjusted goodness of fit index: PRATO=parsimony ratio:
NFI = normed fit index: RFI = relative fit index
Both two-factor oblique models evinced stronger confirmation on most indices except PRATO. A chi square difference test indicated that the two-factor oblique model without dto was superior to the other oblique model (x2 (3) = 21.94, p< .001). However, the latter model is more parsimonious and may be more important in a practical sense. That is, in addition to covariation with confusion and functional decline, any crisis patient showing evidence of violence or homicidal proclivities will likely engender immediate action above and beyond other symptomatology.
These results suggest that persons entering a rural community mental health crisis unit are best differentiated on the basis of depression and suicidal inclinations and less so with respect to confusion and functional decline. Danger to others is rarely endorsed by clinicians but when it is found to be extant, it obviously has significant ramifications. It does load on the secondary factor and has pronounced practical significance. Nevertheless, the oblique nature of the factor structure suggests that the prototypic crisis patient may in the main show evidence for depression and suicide but may also display lesser degrees of confusion and functional decline and occasionally aggressive danger to others.
Arbuckle, J. L. (1997). Amos users’ guide. Chicago: SmallWaters.
Bentler, P. M. & Bonett, D. G. (1980). Significance tests and goodness of fit in the
analysis of covariance structures. Psychological Bulletin, 88, 588-606.
Bonynge, E. R, Lee, R. G., & Thurber, S. (2005). A profile of mental health crisis
response in a rural setting. Community Mental Health Journal, 41, 675-685.
Joreskog, K. G., & Sorborn, D. (1984). LISREL-VI user’s guide (3rd ed.) Mooresville:
IN: Scientific Software.
Thurber, S., & Bonynge, E. R. (2008). Symptom ratings and clinical decision-making
in a rural crisis-response unit. Psychology Journal
First Published April 2009
Copyright Priory Lodge Education limited 2009-