Abstract
Assessing depression in the medically ill risks unreliability when many depressive symptoms (e.g. insomnia, appetite loss, anergia) may be more due to the medical illness itself. To redress such confounding, many options have been proposed [see 1], including the ‘exclusive approach’ of ignoring features common to those with medical illness, and the ‘substitutive approach’ of substituting psychological symptoms (e.g. tearfulness) for somatic features, but with each approach having intrinsic limitations. Adopting a different approach, we developed a measure of depression focusing on cognitive features of depression and thus obviating any need to make judgement about somatic or other items that are more likely to be primary manifestations of the medical illness.
Two development studies [2, 3] were undertaken, generating both brief (DMI-10) and extended (DMI-18) versions of our Depression in the Medically Ill (DMI) measure. They established that discrimination of depressed and non-depressed medically ill subjects was improved by including several cognitive anxiety items rather than limiting the set to pure ‘depression’ ones, suggesting that the nature of ‘depression’ may be somewhat more diffuse in medically ill subjects. In the final development study, we established very high classification accuracy for the DMI-10 measure's cut-off score of 9 or more.
There is also a need for an efficient depression screening measure in those who present to their general practitioner, both for identifying those currently depressed and those at high risk. The General Health Questionnaire [4] has long been used as a measure of state psychological morbidity in general practice and, while possessing excellent psychometric properties as a screening measure for generic psychological morbidity, it is not a screening measure for depression per se. The 9-item PRIME-MD [5] mood module has been shown to have superior properties in detecting current cases of major depression but, as it includes a high percentage of somatic items (i.e. sleep disturbance, anergia, appetite change, concentration problems), risks false positive assignment. As any general practice screening measure needs to be brief, we therefore test the utility of our shorter DMI measure (the DMI-10) in general practice settings.
Testing utility or usefulness is clearly not the same as testing validity. A valid measure of state depression would be expected to have high sensitivity (in detecting true ‘cases’) and high specificity (in excluding noncases), with such properties usually tested by comparing measure scores against an independent ‘case-finding’ interview or related procedure. On the basis of our previous development studies, we note an immediate caveat to this approach, as it is reliant on the subject providing valid information. If a depressed individual elects to deny or minimize depression, or if a non-depressed individual admits to a set of depressive features, scoring on a measure or in responding to an interviewer, the seeming validity of the measure will be promoted, but at the risk of being spurious. A good example emerges from a study by Chochinov and colleagues [6], who compared a single item (‘Are you depressed?’) with interview diagnoses generated from Research Diagnostic Criteria, and with the single item having perfect sensitivity and specificity–a finding leading many people to suggest that only that single question is required to determine whether depression is clinically present or absent. While this is not a sufficient argument for rejecting such validity studies, we examine the utility of our measure in a number of ways. We suggest, that if it is to be a useful screening measure in general practice settings for those with depression or at high risk to lifetime depression, then such individuals should be able to be identified by higher state DMI-10 measure scores.
Method
Six Sydney single and group general practices assisted with data collection–from July to October 2000, with the majority of subjects recruited from a middle class lower North Shore practice, and the remainder from the Inner City of Sydney and a somewhat poorer region of the eastern suburbs. Patients were approached by the practice secretary or by a research assistant, subject to the patient being at least 16 years, having adequate English skills, and not being severely physically unwell. Forms were completed anonymously and then ‘posted’ into boxes near the reception desk. The refusal rate was not formally recorded, but few patients declined. However, as many patients were called in to see the practitioner before completing the forms, there was a significant percentage of returned forms that were incomplete. Thus, while more than 900 questionnaires were returned, we restrict principal analyses to the 638 fully completed questionnaires. It is of interest to note that no patient objected (verbally or in writing to this survey), while there were several dozen appreciative comments written on the forms, with patients stating that the survey had encouraged them to raise either past or current depression with their general practitioner.
Subjects were first required to complete our extended state depression measure (the DMI-18), with the four rating options (‘not true at all’, ‘slightly true’, ‘moderately true’ and ‘very true’ scored 0, 1, 2 and 3, respectively). They then completed a (similarly rated and scored) 70-item measure of personality styles clinically observed in those who present with non-melancholic depression [7], provided sociodemographic details, rated the severity and duration of their medical problem (on 4-point and 3-point scales, respectively), rated their current level of stress (3-point scale) and answered questions pursuing lifetime depression and its treatment. Specifically, subjects were asked whether, over their lifetime, they had ever had a period where they had been depressed (i.e. ‘felt significantly depressed, hopeless and pessimistic about things, had a drop in self-esteem or self-worth, and not been able to cope as well as usual for a period of at least two weeks’). If so experienced, they were to judge whether such episodes were most likely to be ‘normal blues’, a distinct disorder at times, or always a distinct disorder. Details on length of the longest episode and episode impairment (preventing them working) were collected, and subjects were asked whether they had ever consulted a general practitioner, a psychiatrist, psychologist or other professional for their depression or ever received antidepressant medication.
Results
The subjects had a mean age of 38.3 (SD ± 16.5) years and 55% were female. Forty-five per cent were single, 43% married and 12% separated, divorced or widowed. Two-thirds had been born in Australia. Two-thirds were either in full-time or part-time employment, 6% in home duties, 12% students, 4% on sickness benefits or a pensioner, 3% unemployed and 7% retired.
Psychometric properties of the DMI-10.
By the time the study data were being analysed, independent analyses (of a general hospital sample) had suggested that the DMI-10 was almost as efficient as a discriminator of depressed and non-depressed patients as the DMI-18. Thus, in light of its comparative brevity, we then elected to pursue properties associated with the DMI-10 (items listed in Table 1). The internal consistency (alpha = 0.92) of item scoring was extremely high. Table 1 reports the correlation coefficient of each item with the total DMI-10 score (minus that item), establishing the strongest links existing between scale items assessing ‘depression’ and ‘feeling less worthwhile’, supporting its construct validity as a depression measure.
Items in the 10-item Depression in the Medically Ill (DMI-10) measure and corrected item-total correlation
Sociodemographic influences
Age and social class had weak influences on DMI-10 scores, with respective coefficients (r =–0.13 and 0.11) indicating that depression scores decreased slightly with age and increased slightly in higher social class groups. A trend for females to score higher than males (8.3 vs 7.3, t = 1.85) was not significant. Marital status had an influence on scores (F = 9.13, p < 0.001), with those separated or divorced (means = 12.7 and 10.3, respectively) returning higher scores than single (mean = 8.5), widowed (mean = 7.0) or married (mean = 6.4) subjects. Employment status was influential (F = 11.85, p < 0.001), with higher mean scores returned by those unemployed (mean = 17.5) or receiving a pension or sickness benefits (mean = 12.5), compared to students (mean = 6.6), those performing home duties (mean = 7.9) and those in part-time (mean = 7.5) and full-time (mean = 7.5) employment.
Influences of other study variables
Patients' ratings of severity of their medical disorder were associated with DMI-10 scores (r = 0.24, p < 0.001), indicating that higher depression levels were returned by those with more severe medical conditions or that a higher order variable (such as anxiety or worry) inflated both DMI-10 and subjectively judged medical illness severity scores. Patients' ratings of their current stress levels were also associated with DMI-10 scores (r = 0.38, p < 0.001). When those ‘stress’ scores were partialled out of the association between judged severity of medical disorder and DMI-10 scores, the original association was reduced to 0.23, suggesting that a direct link existed between severity of medical illness and state depression scores, and was not merely determined by stress levels. A longer duration of the medical disorder was linked with higher DMI-10 scores (r = 0.14, p < 0.001).
In our development study we determined a DMI-10 score of 9 or more as the most efficient cut-off score for discriminating those who had or did not have clinical depression. In this sample, 35.7% (38.9% of the females and 34.6% of the males) exceeded that criterion and are now regarded as ‘cases’ for further analyses. Such analyses identified ‘cases’ as younger, more likely to be female, single and to be on benefits and unemployed. Those rating as ‘cases’ were distinctly more likely to affirm having had a lifetime depression (and to rate such episodes as more being at the disorder level), been unable to work when depressed, previously received an antidepressant and to have consulted a general practitioner, a psychiatrist and a psychologist for depression. On the personality measure, ‘cases’ scored distinctly higher on anxious worrying, avoidant and irritability dimensions, somewhat higher on self-criticism and self-centred dimensions, and somewhat lower on the obsessive dimension.
A forced entry logistic regression analysis assessed the independent contribution of all identified univariate ‘case’ predictor variables. That analysis identified three variables as significant: younger age (Wald = 14.1, p = 0.001) and personality styles of anxious worrying (Wald = 53.1, p < 0.001) and self-criticism (Wald = 11.5, p < 0.001), with 74.2% of the cases being correctly classified. This analysis also established that other sociodemographic and previous episode and treatment variables were subsumed by this condensed variable set (see Table 2).
Comparison of 10-item Depression in the Medically Ill (DMI-10) measure assigned ‘cases’ and ‘non-cases’ on sample variables
Discussion
The refined DMI-10 comprises two anxiety items (i.e. stewing and feeling vulnerable) and eight depression items. The latter items assess fundamental ‘cognitive’ aspects of depression (as against ‘physical’ ones such as fatigue and sleep disturbance) and are thus less likely to be confounded by physical illness itself. In the development study of hospitalized medically ill subjects, the cut-off score of 9 or more was derived to distinguish depressed ‘cases’ from non-cases, with validation undertaken against both a standardized and computerized case measure, the Composite International Diagnostic Interview [8] and judgements made by a clinical psychiatrist. The respective estimates of sensitivity (i.e. detecting true cases) were 100% and 93%, specificity (i.e. true noncases) were 65% and 70%, giving overall classification rates of 69% and 74%, respectively.
Our DMI-10 measure is brief and was shown in the development studies to be highly acceptable. While we cannot identify its specific acceptability in this study (as it was one component of a lengthy survey), we note the positive annotations and the lack of any negative judgement about the overall survey. In our development study of general hospital patients, 36.4% scored above our derived cut-off score, almost identical to the current 35.7% case rate.
This report is the first study of its utility in a general practice setting. Importantly, age and social class had trivial influences on DMI-10 scores. Even more importantly, females did not return higher scores than males, supporting the developmental strategy of excluding sex-dimorphic items such as crying, and appetite and weight changes. Age, marital and employment status influences on assignment as currently ‘depressed’ were consistent with literature on the epidemiology of depression. For example, in the ABS study of the Australian general population [9], 12-month rates of depression decreased with age, were highest in the separated/divorced and never married (marital status) groups, and highest in the unemployed (occupational status) group.
As noted, 36% of attenders exceeded the predetermined cut-off. While there have been a number of studies examining psychological morbidity in Australian general practices, few studies have estimated depression rates, and with most using dimensional self-report scales lacking firm cut-off assignment rules. Using the dimensional Zung scale in a study of 251 routine general practice attendees, a Sydney study [10] established that 21% (24% females, 15% males) scored above a suggested cut-off of 40 or more. In a second Sydney study of 564 routine general practice attendees [11], 19% scored above a suggested Beck scale cut-off (25% of females and 17% of males). Our rates differ by estimating a higher point prevalence and in failing to identify a distinct female preponderance. Our DMI-10 point prevalence rate could suggest a temporal increase in ‘depression’ over the last two decades, the impact of depression being progressively destigmatized and individuals being more prepared to acknowledge depressive symptoms, or our current cut-off score being set too low. However, if we raise the DMI-10 cut-off to 10 or more or to 11 or more, such increases lower the point prevalence rates in the current sample to 32.3% and 29.6%, respectively; relatively minimal decreases. The imposition of a cut-off score for any dimensional measure will always risk imprecision and therefore we do not regard our measure as necessarily deriving a precise prevalence rate of ‘current depression’. As our screening measure seeks to capture all those with ‘true depression’, a percentage of ‘false positive’ subjects will have scored above cut-off, so inflating the overall ‘case’ rate. However, if a screening measure can narrow the need for focused clinical assessment of potential depression, there are distinct advantages.
While there is no necessity for those who are currently ‘depressed’ to have higher lifetime rates of depression, we argue that an overall association would be expected as a consequence of depression being a persistent condition, whether intermittent or chronic, and by persistence of depression risk factors. Those assigned by our measure as potential current ‘cases’ differed by returning distinctly higher depression disorder rates, in having had previous episodes prevent them from working and being more likely to have consulted a professional and received medication for depression. A multivariate analysis assessed the independent contribution of each individual predictor of case status, identifying a dominant contribution from personality style (with high anxious worrying and self-criticism) in addition to younger age. Analyses suggest that, while patients presenting to a general practitioner with ‘depression’ are more likely to have a past history of depression and to differ on a number of sociodemographic variables, personality style may well be determining their depression trajectory. Such analyses indicate that, while a state depression measure, the DMI appears useful in identifying those who are at socioeconomic and personality style risk to depression, and thus speaks to its screening capacity to identify those at distinct depression risk–both currently and at risk of such episodes. Further studies will need to formally assess DMI-10 scores against a case-finding interview of general practice patients, particularly to confirm whether the cut-off score established in our hospital-based study applies in a general practice setting.
Footnotes
Acknowledgements
The study was funded by National Health and Medical Research Council grant number 993208. The assistance of the many general practitioners and of Kerrie Eyers is appreciated.