|Home | About | Journals | Submit | Contact Us | Français|
Depression, with up to 11.9% prevalence in the general population, is a common disorder strongly associated with increased morbidity. The accuracy of non-psychiatric physicians in recognizing depression may influence the outcome of the illness, as unrecognized patients are not offered treatment for depression.
To describe and quantitatively summarize the existing data on recognition of depression by non-psychiatric physicians.
We searched the following databases: MEDLINE (1966–2005), Psych INFO (1967–2005) and CINAHL (1982–2005). To summarize data presented in the papers reviewed, we calculated the Summary receiver operating characteristic (ROC) and the summary sensitivity, specificity and odds ratios (ORs) of recognition, and their 95% confidence intervals using the random effects model.
The summary sensitivity, specificity, and OR of recognition using the random effects model were: 36.4% (95% CI: 27.9–44.8), 83.7% (95% CI: 77.5–90.0), and 4.0 (95% CI: 3.2–4.9), respectively. We also calculated the Summary ROC. We performed a metaregression analysis, which showed that the method of documentation of recognition, the age of the sample, and the date of study publication have significant effect on the summary sensitivity and the odds of recognition, in the univariate model. Only the method of documentation had a significant effect on summary sensitivity, when the age of the sample and the date of publication were added to the model.
The accuracy of depression recognition by non-psychiatrist physicians is low. Further research should focus on developing standardized methods of documenting non-psychiatric physicians’ recognition of depression.
Depression, with up to 11.9% prevalence in the general population,1,2 is a common disorder strongly associated with increased morbidity.3 It has been estimated that in 2020 depression will become the second leading cause of disability,4 which emphasizes the importance of its early detection and treatment. The accuracy of non-psychiatric physicians in recognizing depression may influence the outcome of the illness, as unrecognized patients are not offered treatment for depression. Currently, less than half of patients with depression are recognized by their primary care physicians, even after 5 years of follow-up.5 Moreover, studies show that among recognized depressed patients only a few receive appropriate care, which may further lead to poor outcome of depression and increased health service use and mortality rates in these patients.6,7
There are many potential reasons for the underrecognition and undertreatment of depression; patient, provider, and system barriers have been identified. Patients reduce the likelihood of being diagnosed by presenting with somatic rather than emotional complaints8–10 and may resist a diagnosis of depression or anxiety by attributing their symptoms to physical causes.11,12 Provider barriers include concerns about potential patient stigma,13,14 time pressures,14 a belief that such diagnoses are burdensome,15 inadequate knowledge about diagnostic criteria or treatment options,6 lack of a psychosocial orientation,16–18 and inadequate insight into different cultural presentations of mental disorders.19 System barriers include productivity pressures, limitations of third-party mental health coverage, restrictions on specialist, drug, and psychotherapeutic care,13,14 lack of a systematic method for detecting and managing such patients,20 and inadequate continuity of care.13
The accuracy of recognition of depression by attending physicians can be assessed using measures such as sensitivity, specificity, and odds ratio, when the clinical diagnosis made by the physician is compared to a gold standard diagnosis of depression. The estimates of these measures reported in the literature vary according to the method used to document the clinical diagnosis made by physicians and the definition of gold standard.
The main objectives of this literature review and meta-analysis were to describe and quantitatively summarize existing data on recognition of depression by non-psychiatric physicians in adult outpatients and inpatients. To our knowledge, this is the first systematic review of studies on recognition of depression by non-psychiatric physicians.
The search was conducted by MC (MSc in Epidemiology) using OVID search engine (2000–2005 version). MEDLINE (1966–2006), Psych INFO (1967–2006), CINAHL (1982–2006), and EMBASE (1986–2006) databases were searched for the following keywords: “depression” or “depression disorder” and “detection” or “recognition” or “identification” or “diagnosis”. Letters, editorials, reviews, and case-reports were excluded. The search was limited to studies with subjects of ages 18 and more and written in English or French. This search yielded 7,105 papers.
Titles and abstracts of these papers were reviewed to identify those of potential relevance, based on the following inclusion criteria:
Eighty-one studies of potential relevance were retained for review. The reference lists of the papers selected were searched and another 2 studies of potential relevance were retained for review. From the 85 reviewed papers, we excluded 49 for the following reasons:
The flow chart of the systematic review is presented in Figure 1.
The papers selected based on our inclusion and exclusion criteria were independently reviewed by 2 reviewers with training in Epidemiology using a review form and guidelines developed by the authors. Data abstracted included: study design, study setting, method of selecting the sample, age and gender of the sample, sample size used in the analysis, specialties and number of physicians participating in the study, blinding of physicians, criterion measure for diagnosing depression, data source for diagnosing depression, data source and indicators of recognition, and results (sensitivity, specificity, positive predictive value, negative predictive value, crude odds ratio, other) either presented in the study or possibly calculated from the data available.
Differences in data abstraction between the 2 reviewers were discussed and resolved by the authors. We reviewed the following aspects of study quality: selection of a clinically relevant cohort, the consistent use of a good gold standard, blinding of physicians to the gold standard results, and incomplete reporting.69
In all the studies reviewed, the authors consistently used a gold standard tool for detecting depression symptoms or diagnosing depression. Moreover, details pertaining to the validity and reliability of the tool used were included.
The authors of all the studies reviewed indicated the method of sampling the patients (consecutive or randomized sampling) and the health care settings where the patients were recruited. The age or gender distribution, as well as the prevalence of depression in the samples recruited from different health care settings vary considerably.
In 11 studies,70–80 the authors did not specify if the physicians were blinded to the gold standard diagnosis (incomplete reporting). The rest of the studies included references to the blinding of physicians or the chart review.
To determine whether the quality of the studies may influence the conclusions of our literature review and meta-analysis, we performed a sensitivity analysis that explored the following aspects of study quality:
We calculated missing sensitivities, specificities, and crude odds ratios (OR) using data reported in the papers reviewed. We used the random effects model83 to calculate the summary diagnostic odds ratios, sensitivity, and specificity. Because there was evidence of varying cut points (Spearman: −0.635, p=0.0005), we also plotted our main measure and recalculated the overall sensitivity using receiver operating characteristic (ROC) curves.84
We performed the Cochrane’s Q test85 to assess the presence of heterogeneity among results reported in studies included in our systematic literature review and meta-analysis. We performed a univariate metaregression analysis, testing 7 potential sources of heterogeneity: the method of documentation (chart review vs physician diagnosis), blinding of physicians (physicians were blinded vs blinding status was not specified), the gold standard diagnostic tools used (structured clinical interview vs diagnostic scale with a specific cut point), method of sampling (random vs other), age (55 and over vs all ages and younger patients only), gender of sample (62% and more vs less than 62% female), date of study publication (after 1998 vs 1998 and before) and prevalence of depression (less than 25% vs 25% or more). The categories of the last 3 covariates were created using the median value as cut point. These analyses were performed using the STATA statistical software (STATA/ SE 8.1 for Windows).
We included 36 studies in our systematic review (Tables 1, ,2,2, ,3,3, and and4)4) Twenty-three papers reported only sensitivity72–74,78–80,86–102. For 10 of these studies, we calculated the specificity and the diagnostic OR.72,73,80,87,89,91,93,99–101 In 8 papers, both the sensitivity and specificity of recognition were presented.75,77,103–108 Eleven papers reported other measures (such as positive predictive value, agreement expressed as percentage or kappa correlation coefficient, and identification index),70,76,77,79,103,104,106,107,109,110 and we calculated the sensitivity, specificity, and OR of recognition using data presented in 5 of them.70,76,77,79,109 One paper included in our systematic review105 reported a sensitivity of 0% and a specificity of 98.9%. We considered these results outliers and decided to exclude them from our analysis.
Twenty-seven (75%) of the studies included in our systematic review and meta-analysis were conducted in primary care,70–79,80,86,88,89,94–100,102,103,107–110 3 studies were conducted in the emergency department,90,104,105 three studies included patients admitted to the hospital,87,92,93 and three studies included outpatients attending specialty clinics.70,91,101
Overall, we found high specificity (83.7%, 95% CI: 77.5–90.0, Table 5), but lower sensitivity (36.4%, 95% CI: 27.9–44.8) with a resulting diagnostic OR of 4.0 (95% CI: 3.2–4.9). The sensitivity varied across the method of assessment (Table 5), with physician diagnosis (PD) method having higher pooled sensitivity than those using chart review (CR). Papers published after 1998 had higher pooled sensitivity than those before 1998 (Table 5). When we calculated the overall sensitivity, based on summary ROC curves, we found a sensitivity of 42.3% (Fig. 2).
Our results were heterogeneous. On metaregression, the method of documentation, age of the sample, and date of publication showed a statistically significant effect on the summary sensitivity (Table 6). These results indicate that the summary sensitivity of studies that used as method of documentation the physician diagnosis, were published after 1998 and had a sample of younger or all ages of patients was higher by 14.5%, 11.8%, and 12.3%, respectively, compared to those that used as method of documentation the chart review, were published in 1998 or before, and had a sample of patients aged 55 and more.
Only age of the sample and date of publication explained the heterogeneity of our pooled odds ratios. Studies that were published after 1998 and had a sample of younger or all ages of patients reported higher odds ratios of recognition compared to studies that were published in 1998 or before and had a sample of patients aged 55 and more (Table 7).
We performed a multivariate metaregression, including in the model 3 variables that had a significant effect on summary sensitivity or the pooled odds ratios in the univariate analysis: method of documentation, age of the sample, and date of publication. The multivariate analysis showed that the summary sensitivity of studies that used as method of documentation the physician diagnosis was 12.5% higher compared to those that used as method of documentation the chart review, when controlled for age of the sample and date of publication (Table 7).
Recognition of depression by nonpsychiatric physicians has received increased attention from researchers as depression became one of the most prevalent diseases of the 21st century and an important public health issue. The assessment of validity of recognition is challenged by the variety of methods used to document recognition and to diagnose depression. In this systematic literature review, we qualitatively and quantitatively summarized the accuracy of recognition of depression using data presented in the papers reviewed. Further, we tried to identify sources of heterogeneity in the results reported.
The summary sensitivity showed that less than half of the depressed patients are recognized by their physicians. This is consistent with rates of depression detection reported in previous studies.9 On the other hand, the summary specificities are reasonable (calculated using data reported in 22 studies) and are consistent with interrater agreement found in studies on the accuracy of psychiatrist interviews.111
Studies that used as method of documentation the physician diagnosis had a higher summary sensitivity compared to studies that used the chart review. This result is consistent with the conclusion of a study that compared the 2 methods,103 showing that either physicians tend to diagnose depression more frequently when they are specifically asked about this possible diagnosis or documentation of depression recognition (including diagnosis, treatment, and referral to a mental health specialist) in patients’ charts is low. Moreover, the same study found that even if the sensitivity of recognition by physician diagnosis was higher, the specificity of this method of documentation was lower than the specificity of recognition by chart review. This may explain why the method of documentation did not affect the summary odds ratio in our metaregression analysis.
We also found that the sensitivity and odds of recognition of depression by physicians are significantly higher in younger or unselected patients than in older ones. Data reported in several studies included in our literature review show no significant association between recognition of depression and age88,107 or higher recognition of depression in patients less than 35 compared to patients aged 65 and more,112 which demonstrates the variability among studies. However, in the multivariate metaregression, the age and the sample had no effect on summary sensitivity and summary OR when controlled for method of documentation and date of publication.
Another notable finding of our meta-analysis is that studies published after 1998 tend to report higher sensitivities and odds ratios of recognition than those published in 1998 and before. This may suggest that the non-psychiatric physicians training in diagnosing depression has improved over the years. However, in the multivariate metaregression, the date of publication had no effect on summary sensitivity and summary OR when controlled for method of documentation and age of the sample.
We identified several limitations of our literature review and meta-analysis. First, the literature search was restricted to two languages. Is it possible that articles written in other languages than English or French were missed. Second, it was difficult to assess the quality of the papers included, as there was a great variability in methods used. We abstracted and presented data regarding the sites of the studies, the method of sampling, the blinding of outcomes, and the specialties of physicians involved in these studies, to offer a view of the methodological quality of these studies that may have affected the validity of our results. Third, our subgroup analysis did not cover all the possible sources of heterogeneity (for example, the categorization in subgroups by the type of gold standard diagnostic tool used does not account for the variability in type of depression scales and cut points used). Fourth, many studies included as "missed" patients those with subthreshold depression, a category of depression in which treatment and placebo both have similar high rates of remission113 and which tend to be transient.9 Fifth, there was no gradation on severity of depression. Studies have shown that patients with more severe forms of depression are more likely to be diagnosed.9,79,80,107,114,115
The overall sensitivity of recognition of depression by non-psychiatric physicians reported in our systematic literature review and meta-analysis was low, although non-psychiatric physicians had good specificity. A number of variables, including the method of documentation of recognition, age of the sample, and date of study publication, had an impact on the summary sensitivity. Moreover, the last 2 factors had a significant effect on the summary odds ratio of recognition, a measure of diagnostic accuracy.
A large number of potential barriers to recognition and treatment of depression have been identified. The specific reason studies consistently find low rates of sensitivities is unclear. Given the high prevalence of the disease and its significant impact on the overall health of patients, this question deserves further research. It would be helpful if there were a standardized method of documenting non-psychiatric physicians’ recognition of depression. This method may be useful for studies that tests the efficacy of various educational programs designed to improve physicians’ accuracy in diagnosing depression. Also, a standardized method of documenting recognition may improve the quality of studies that look at various factors which may affect recognition of depression, such as age, gender or ethnicity of patients, or physicians’ characteristics.
This study was supported by a grant from Fonds de la recherche en santé du Québec (Grant number 25004-2560) and two grants from Institutes for Health Research (Grant numbers MOP-64462 and 11949).
Conflict of Interest None disclosed.