Clinical diagnosis of depression in primary care: a meta-analysis

Articles Clinical diagnosis of depression in primary care: a meta-analysis Alex J Mitchell, Amol Vaze, Sanjay Rao Summary Background Depression is a major burden for the health-care system worldwide. Most care for depression is delivered by general practitioners (GPs). We assessed the rate of true positives and negatives, and false positives and negatives in primary care when GPs make routine diagnoses of depression. Methods We undertook a meta-analysis of 118 studies that assessed the accuracy of unassisted diagnoses of depression by GPs. 41 of these studies were included because they had a robust outcome standard of a structured or semistructured interview. Findings 50 371 patients were pooled across 41 studies and examined. GPs correctly identified depression in 47·3% (95% CI 41·7% to 53·0%) of cases and recorded depression in their notes in 33·6% (22·4% to 45·7%). 19 studies assessed both rule-in and rule-out accuracy; from these studies, the weighted sensitivity was 50·1% (41·3% to 59·0%) and specificity was 81·3% (74·5% to 87·3%). At a rate of 21·9%, the positive predictive value was 42·0% (39·6% to 44·3%) and the negative predictive value was 85·8% (84·8% to 86·7%). This finding suggests that for every 100 unselected cases seen in primary care, there are more false positives (n=15) than either missed (n=10) or identified cases (n=10). Accuracy was improved with prospective examination over an extended period (3–12 months) rather than relying on a one-off assessment or case-note records. Interpretation GPs can rule out depression in most people who are not depressed; however, the modest prevalence of depression in primary care means that misidentifications outnumber missed cases. Diagnosis could be improved by re-assessment of individuals who might have depression. Lancet 2009; 374: 609–19 Published Online July 28, 2009 DOI:10.1016/S01406736(09)60879-5 See Editorial page 587 See Comment page 589 Leicestershire Partnership Trust, Leicester General Hospital, Leicester, UK (A J Mitchell MRCPsych, A Vaze MRCPsych, S Rao MRCPsych); and Department of Cancer and Molecular Medicine, Leicester Royal Infirmary, Leicester, UK (A J Mitchell) Correspondence to: Dr Alex J Mitchell, Leicester General Hospital, Leicester LE5 4PW, UK Alex.Mitchell@leicspart.nhs.uk Funding None. Introduction The burden of depression is considerable in terms of missed workdays1 and disability.2 The WHO study on psychological problems in general health care (PPGHC) across 14 countries found that 14% of individuals suffered from major depression.3 Similar rates were reported from the longitudinal investigation of depression outcomes in primary care (LIDO) study4 and the Depression 2000 study from Germany.5 However, rates are usually higher in urban than in rural settings.6,7 Most care for depression is delivered by general practitioners (GPs) and individually many GPs have considerable expertise in managing depression.8 Yet, under-recognition of depression in primary care has been extensively described.9,10 Evidence shows that clinicians in all medical specialties have difficulty recognising mental disorders.11 Recognition rates vary by practitioner, study, and country. The WHO primary care study3 found that 54·2% of those who met the criteria for depression were judged by their treating physician as having a psychological illness, although rates of accurate diagnosis of depression ranged from 19·3% in Nagasaki (Japan) to 74·0% in Santiago de Chile (Chile). Over-detection—ie, the generation of false positives—is less discussed but nevertheless important. Clinicians can overestimate or underestimate levels of distress of their patients.12 Under-detection could lead to not enough treatment. Studies suggest that active treatwww.thelancet.com Vol 374 August 22, 2009 ment takes place in 15–60% of those identified as depressed.13,14 Conversely, over-detection (misidentification) can lead to too much treatment. For example, in the European study of the epidemiology of mental disorders (ESEMeD) about 13% of individuals presenting to the GP with symptoms of depression did not have any mental disorder.15 In the national comorbidity replication study, almost 40% received an intervention for depression without a concurrent diagnosis.16 Whether under-recognition or over-recognition of depression is most problematic in routine clinical practice is not clear. We hypothesised that the absolute number of false positives (over-detections or misidentifications) would outnumber that of false negatives (underdetections). Our aim was to assess, with a quantitative analysis, the rate of true positives, true negatives, false positives, and false negatives in primary care when GPs make routine diagnoses of depression. Methods Literature search We did a systematic literature search, critical appraisal, and pooled analysis. We searched Medline from January, 1966, to April, 2009; PsycINFO from January, 1887, to April, 2009; Embase from January, 1980, to April, 2009; and Scopus from January, 1980, to April, 2009. The search terms were “general practi$ or primary care or family practice$” and 609 Articles 157 physician diagnosis of depression in primary care 39 no unassisted assessment by primary care physicians 118 physician diagnosis of depression in primary care 4 non-validated comparison 61 diagnostic accuracy of depression in primary care by severity scale 5 duplicate data 7 insufficient data 41 interview-based diagnostic accuracy of depression in primary care Sample size* 33 mid-life sample 8 late-life sample 22 sensitivity alone 19 specificity data Outcome measure 94 to 199 (n=13) 9 CIDI 200 to 499 (n=12) 5 SCID ≥500 (n=7) 10 DIS ≥1000 (n=9) 2 CIE Outcome criterion Case ascertainment 21 DSMIII, IIIR, or IV 6 cumulative notes or electronic record 9 ICD 5 cross-sectional notes or electronic record 11 other standardised interview or expert diagnosis 2 cumulative prospective questionnaire 28 contemporaneous prospective questionnaire 2 SCAN/PSE 12 others Figure 1: Quorum diagram of studies CIDI=composite international diagnostic interview. CIE= Canberra interview for the elderly. DIS=diagnostic interview schedule. DSM=Diagnostic and Statistical Manual. ICD=International Classification of Diseases. SCAN/PSE=schedules for clinical assessment in neuropsychiatry/present state examination. SCID=structural clinical interview for DSM. *Refers to raw data extracted. “diagnos$ or detect$ or case-finding or recogni$” and “depress$ or mood or affect$”. Refinement for psychiatric interviews was achieved with “ICD”, “ICD10”, “ICD9”, “DSM”, “DSMIV”, “DSMIII”, “SCID”, “CIDI”, “PSE”, “DIS”, “interview”, or “structured”. In full-text collections including Science Direct, Ingenta Select, Ovid Full text, and Wiley-Blackwell Interscience, we used the same search terms but as a full-text search and citation search. We also searched the abstract database Web of Knowledge (4.0, ISI) with the above terms as a text word search and with key publications in a reverse citation search. 610 Inclusion and exclusion criteria The main inclusion criterion was the unassisted diagnostic ability of GPs to identify depression. Unassisted means that the practitioner is trying to diagnose depression without specific help from severity scales, diagnostic instruments, education programmes, or other organisational approaches. Of those studies with a randomised intervention to improve detection (eg, an education package), we only included data from GPs without (or before) intervention. To define robust studies, we included those with a minimum sample size www.thelancet.com Vol 374 August 22, 2009 Articles of 50 cases and excluded all vignette studies. We used a robust outcome definition in the form of interview-based diagnoses, using a psychiatric expert diagnosis or validated structured or semi-structured interviews applied by a research interviewer. If a study assessed detection rates of different types of depression, we used data for major depression alone. Data extraction Data were extracted with a standardised spreadsheet by two authors (AV and SR) and re-examined by another author (AJM) independently. We followed the review guidelines for diagnostic tests recently outlined.17 Questions for each report included the sample (eg, age of patients), data integrity, choice of reference criterion, method of case ascertainment, and duration of clinician assessment of depression (cumulative or cross-sectional). At least two authors rated each publication on a 5-point scale. Statistical analysis To assess diagnostic accuracy, we extracted and pooled raw data with proportion meta-analysis to correct for variations in sample size.18 Considerable heterogeneity was present. I² was 94·0% (95% CI 93·0% to 94·8%) from the full sample of 41 studies and 99·5% (99·5% to 99·5%) in the 19 studies that assessed both rule-in and rule-out accuracy; therefore, we used random-effects meta-analysis. We adjusted positive predictive values and negative predictive values for variations in prevalence from a Bayesian curve of all post-test probabilities.19 We calculated overall accuracy with fraction correct, which is the number of correctly identified individuals in every 100 patients screened. The number of screening applications to rule in or rule out without error is known as the number needed to screen (NNS),20 where NNS=1/(fraction correct–fraction incorrect) We also undertook a bivariate meta-analysis. This method fits a two-level model, with independent binomial distributions for true positives and true negatives subject to sensitivity and specificity in each study, and a bivariate normal model for the logit transformations of sensitivity and specificity between studies.21 We then constructed a summary Receiver Operator Characteristic curve (sROC), in which each datapoint represents a separate study, using the bivariate model to produce a 95% confidence ellipse within ROC space. We also examined predictors of accuracy according to duration of assessment (cumulative vs cross-sectional), method of case ascertainment (case notes vs contemporaneous form), country of origin, and patient age. Role of the funding source This study did not have any sponsor. The corresponding author had full access to all the data in the study and had www.thelancet.com Vol 374 August 22, 2009 final responsibility for the decision to submit for publication. Results 118 studies assessed the ability of GPs to make an unassisted diagnosis of depression, but only 41 satisfied criteria for valid studies measured against a robust outcome standard (figure 1 and table). One study that used the 10th International Classification of Diseases (ICD10) criteria was excluded because both recognition and criterion interview were done by the same GPs. One study explored anxiety alone. 26 were excluded because they examined broadly defined distress or mixed mental disorders. Two studies included diagnosis by nursing colleagues working in primary care. One study that examined outcome after recognition but not detection sensitivity was excluded. We excluded five reports containing data reported in previous publications. Thus, 19 studies assessed diagnostic specificity and had both rule-in (case-finding) and rule-out (exclusion) accuracy. 13 of these 19 studies examined22–33 general adult patients (including one of adolescents34) and six examined elderly (>65 years) people.35–40 The remaining 22 studies only assessed diagnostic sensitivity. Of these, two were done in elderly (>65 years) people,41,42 leaving 20 that were done in adult patients.43–62 The total pooled sample was 50 371 patients and the mean sample size per study was 1228 (median 318), but with a large SD of 4004·5. The mean number of depressed patients was 135 (median 70; SD 179·4) and the mean number of GPs was 53·7 (median 26·5; SD 75·5). Criterion methods of diagnosis of depression included a composite international diagnostic interview (n=9), diagnostic interview schedule (n=10), structural clinical interview for DSMIIIR (n=5), Canberra interview for the elderly (n=2), schedules for clinical assessment in neuropsychiatry/present state examination (n=2), schedule for affective disorders and schizophrenia (SADS) (n=2), and one study for each of the MINI international neuropsychiatric interview and geriatric mental state examination. Eight studies used research interviews of unspecified design to diagnose depression. Regardless of method of interview, an ICD-based diagnosis of depression was used in nine studies, and a diagnosis based on Diagnostic and Statistical Manual (DSM) in 21 studies. In all 41 studies, all severities of depression were included. 30 studies used a simple questionnaire for GPs to assess whether they could diagnose depression (defined as contemporaneous ratings). Of these studies, two asked GPs to rate their clinical impression over the previous 6–12 months;42,43 11 examined the notes (or electronic records) entered by GPs (defined as case-note method). In six of 11 case-note studies, depression was assessed over a period of 3–12 months (cumulative recognition). The overall prevalence of depression (adjusted) was 19·5% (95% CI 15·7% to 23·7%). The corresponding 611 612 Methodological summary Statistical summary Participating Mean patient GPs age (number) (years) Description of case ascertainment Gold standard Rating* Analysed Depressed Rate (I–V) sample with GS Sensitivity Specificity PPV NPV Fraction NNS correct Prospective case ascertainment (adults) Aben et al22 Netherlands 39 Aragones et al23 Spain 23 Balestrieri et al24 Italy 25 Borowsky et al25 USA 349 Christensen et al26 Denmark Gerber et al27 USA 4 ·· Gledhill et al28 UK 10 15·6 38 ·· Blind recognition of depression by GP SCID DSMIV III 58 29 50·0% 37·9% 89·7% 78·6% 59·1% 63·8% 3·63 Prospective questionnaire of patient depression SCID DSMIV IV 306 120 39·2% 71·7% 59·7% 53·4% 76·6% 64·4% 3·48 ·· Prospective questionnaire of patient depression ICD10 I 2093 283 13·5% 38·9% 93·7% 49·1% 90·7% 86·3% 1·38 ·· Prospective questionnaire about the past 12 months (cumulative) DIS for DSMIII II 661 661 ·· 35·2% ·· ·· ·· ·· ·· Prospective questionnaire SCAN for ICD10 II 301 70 23·3% 45·7% 89·2% 56·1% 84·4% 79·1% 1·72 Prospective GP recording of any depression SADS IV 268 84 31·3% 57·1% 84·2% 62·3% 81·2% 75·7% 1·94 Prospective depression checklist K-SADS V 38 10 26·3% 20·0% 85·7% 33·3% 75·0% 68·4% 2·71 48·8 38·8 Henkel et al29 Germany 18† Blind prospective recognition form CIDI III 431 73 16·9% 64·4% 74·0% 33·6% 91·1% 72·4% 2·23 Jones et al30 USA 20 44·0 Structured physician interviews and chart audits DIS for DSMIII V 94 21 22·3% 33·3% ·· ·· ·· ·· ·· Klinkman et al31 USA 50 39·6 Prospective clinician rating form with yes/no for presence of depression SCID (DSM-III-R) IV 425 52 12·2% 34·6% 78·8% 18·6% 89·6% 73·4% 2·14 Lecrubier32 France Menchetti et al33 Italy ·· 164 ·· Uncertain MINI III 2419 238 9·8% 26·1% ·· ·· ·· ·· ·· 191 49·1 Prospective GP rating form ICD10 III 1896 299 15·8% 73·9% ·· ·· ·· ·· ·· ·· Prospective GP rating form PSE III 1994 184 9·2% 65·8% ·· ·· ·· ·· ·· CIDI (primary health care version) (ICD10, current depression or dysthymia) IV 189 66 34·9% 71·2% 68·3% 54·6% 81·6% 69·3% 2·59 148 ·· 62·8% ·· ·· ·· ·· ·· Ormel et al34 Netherlands 25 Pini et al35 Italy 46† 44·5 Prospective physician encounter form for psychiatric caseness (5-point scale) Ronalds et al36 UK 13 35·0 GP rated the reason for the index consultation Psychiatric and medical records were examined assessment schedule for DSMIIIR IV ·· Prospective depression tool SCID for DSMIIIR III 1580 92 5·8% 31·5% ·· ·· ·· ·· ·· Prospective depression tool CIDI III 970 64 6·6% 64·1% ·· ·· ·· ·· ·· www.thelancet.com Vol 374 August 22, 2009 Schwenk et al37 USA 50 Simon and VonKorff38 USA ·· 38·9 Simon et al39 International ·· 40·3 Prospective depression tool on a 5-point scale CIDI for ICD10 III 25 916 948 3·7% 42·5% ·· ·· ·· ·· ·· Strauss et al40 USA 2 42·0 Retrospective contemporaneous form symptoms and syndrome of depression SCID (DSMIV) by psychiatrist II 327 52 15·9% 61·5% ·· ·· ·· ·· ·· Tiemens et al41 Netherlands 10 40·5 Prospective physician encounter form for distress CIDI-PC for ICD10 IV 518 136 26·3% 70·6% ·· ·· ·· ·· ·· Tiemens et al42 International 61 ·· Prospective physician encounter form for distress CIDI-PC for ICD10 II 709 174 24·5% 40·2% 85·8% 47·9% 81·5% 74·6% 2·03 Tylee et al43 UK 36 38·4 Prospective encounter form GHQ30 and then combined clinical interview V 1756 430 24·5% 41·6% ·· ·· ·· ·· ·· Van WeelBaumgarten et al44 Netherlands 46·0 Routine (ICHPPC-2 criteria) applied consecutively CIDI for DSMIV MDD III 99 33 33·3% 78·8% 89·4% 78·8% 89·4% 85·9% 1·39 4† ·· (Continues on next page) Articles Country www.thelancet.com Vol 374 August 22, 2009 Methodological summary Country Statistical summary Participating Mean patient GPs age (number) (years) Description of case ascertainment Gold standard Rating* Analysed Depressed Rate (I–V) sample with GS Sensitivity Specificity PPV NPV Fraction NNS correct (Continued from previous page) Case-notes ascertainment (adults) Von Korff et al45 USA 45 Füredi et al46 Hungary 12† 40·5 ·· Kirmayer et al47 Canada 2† ·· Miller and McCrone48 USA ·· Munitz et al49 USA ·· Nuyen et al50 Netherlands Rost et al51 USA Schmaling and Hernendez52 USA Schulberg et al53 USA Williams et al54 USA Crawford et al55 UK Volkers et al56 Netherlands 59·8 ·· Prospective rating of emotional problems DIS for DSMIIIR V 49 49 3·9% 75·5% ·· ·· ·· ·· ·· Medical notes DIS for DSMIV I 1211 60 5·0% 6·7% ·· ·· ·· · ·· Cumulative diagnosis of depression or anxiety DIS for DSMIIIR in the medical chart for index presentation or subsequent 12 months (cumulative) II 685 65 9·5% 33·8% ·· ·· ·· ·· ·· Retrospective case notes Interview by mental health clinicians III ·· 48 Unknown 72·9% ·· · ·· ·· ·· Blind cumulative 3 months medical notes (retrospective) (cumulative) ICD10 checklist I 1197 134 11·2% 21·6% ·· ·· ·· ·· ·· 195 45·4 Electronic case records from a 12 month period CIDI I ·· 191 ·· 28·8% ·· ·· ·· ·· ·· NR 46·3 Recognition from medical notes together with prescribing habits over 1 year (cumulative) DIS for DSMIV III ·· 98 ·· 67·3% ·· ·· ·· ·· ·· 2† 45·5 Retrospective notes review for 6 month prevalence of depression DSMIV II 447 146 32·7% 20·5% 90·4% 50·8% 70·1% 67·6% 2·85 3† 33·4 Recognition from medical notes over 6 months DIS for DSMIII II 294 27 9·2% 25·9% 98·1% 58·3% 92·9% 91·5% 1·20 80 56·0 Recognition from medical notes DIS for DSMIIIR II 276 38 13·8% 28·9% ·· ·· ·· ·· ·· 28 ·· Retrospective notes examination and GP interview for 6 months (cumulative) Short CARE II 318 62 19·5% 51·6% 71·9% 30·8% 86·0% 67·9% 2·79 Prospective electronic medical record over 12 months (cumulative) CIDI for DSMIV II 237 53 22·4% 20·8% ·· ·· ·· ·· ·· Prospective clinician questionnaire (4-point scale) DIS for DSMIIIR IV 101 15 14·9% 60·0% 62·8% 22·0% 90·0% 62·4% 4·04 Blind structured questionnaire 1 year after the PRIME-MD for initial screen (cumulative) DSMIV III ·· 73 ·· 32·9% ·· ·· ·· ·· ·· 4·13 ·· 63·6 Prospective case ascertainment (elderly) Bowers et al57 Australia Licht-Strunk et al58 Netherlands 11 ·· ·· 64·0 O’Conner et al59 Australia 30 78·2 Prospective depression tool CIE III 995 61 6·1% 75·4% 61·2% 11·3% 97·4% 62·1% Pond et al60 Australia 13 82·5 Prospective questionnaire CIE for ICD10 IV 105 12 11·4% 50·0% 77·0% 21·9% 92·3% 73·9% 2·09 Turrina et al61 Italy 14 72·8 Prospective contemporaneous form on 4-point Likert scale Geriatric mental state examination for DSMIIIR III 255 89 34·9% 71·9% 71·1% 57·1% 82·5% 71·4% 2·34 Van Marwijk et al62 Netherlands 13 73·6 Prospective depression tool DIS for DSMIII III 580 46 7·9% 26·1% 92·7% 23·5% 93·6% 87·4% 1·34 Table: Studies of GPs’ ability to diagnose depression Articles CARE=comprehensive assessment and referral evaluation. CIDI=composite international diagnostic interview. CIE=Canberra interview for the elderly. DIS=diagnostic interview schedule. GDS=geriatric depression scale. GHQ=general health questionnaire. GPs=general practitioners. GS=gold standard. K-SADS=Kiddie-Sads diagnostic interview. NNS=number needed to screen. NPV=negative predictive value calculated from data. PPV=positive predictive value calculated from data. PRIME-MD=primary care evaluation of mental disorders. PSE=present state examination. *Ratings: I=blind, sample >1000; drop-outs accounted for. II=blind, sample >99; minor design issues. III=blind but underpowered (<100) or with substantial design issues or unblind or unclear allocation concealment, but otherwise minor issues. IV=not blind or unclear with sample <1000. V=not blind or unclear with sample <100 and/or other substantial methodological weaknesses. †Refers to number of GP practices (not individuals). 613 Articles Van Weel-Baumgarten et al (2000) 0·79 (0·61–0·91) Von Korff et al (1987) 0·76 (0·61–0·87) Menchetti et al (2009) 0·74 (0·69–0·79) Miller et al (2005) [CN] 0·73 (0·58–0·85) Aragones et al (2004) 0·72 (0·63–0·80) Pini et al (1997) 0·71 (0·59–0·82) Tiemens et al (1999) 0·71 (0·62–0·78) Rost et al (1998) [CN] 0·67 (0·57–0·76) Ormel et al (1990) 0·66 (0·58–0·73) Henkel et al (2003) 0·64 (0·52–0·75) Simon and VonKorff (1995) 0·64 (0·51–0·76) Ronalds et al (2003) 0·63 (0·55–0·71) Strauss et al (1995) 0·62 (0·47–0·75) Gerber et al (1989) 0·57 (0·46–0·68) Christensen et al (2003) 0·46 (0·34–0·58) Simon et al (1999) 0·43 (0·39–0·46) Tylee et al (1996) 0·42 (0·37–0·46) Tiemens et al (1999) 0·40 (0·33–0·48) Balestriere et al (2004) 0·39 (0·33–0·45) Aben et al (2003) 0·38 (0·21–0·58) Borowsky et al (2000) 0·35 (0·32–0·39) Klinkman et al (1997) 0·35 (0·22–0·49) Kirmayer et al (1993) [CN] 0·34 (0·23–0·47) Jones et al (1987) 0·33 (0·23–0·57) Schwenk et al (1996) 0·32 (0·22–0·42) Williams et al (1999) [CN] 0·29 (0·15–0·46) Nuyen et al (2005) [CN] 0·29 (0·22–0·36) Lecrubier (2007) 0·26 (0·21–0·32) Schulberg et al (1985) [CN] 0·26 (0·11–0·46) Munitz et al (2000) [CN] 0·22 (0·15–0·30) Schmaling and Hernendez (2005) [CN] 0·21 (0·14–0·28) Gledhill et al (2003) 0·20 (0·03–0·56) Füredi et al (2003) [CN] 0·07 (0·02–0·16) O’Connor et al (2001) [E] 0·75 (0·63–0·86) Turrina et al (1994) [E] 0·72 (0·61–0·81) Bowers et al (1990) [E] 0·60 (0·32–0·84) Crawford et al (1998) [CN] [E] 0·52 (0·39–0·65) Pond et al (1994) [E] 0·50 (0·21–0·79) Licht-Strunk et al (2008) [E] 0·33 (0·22–0·45) Van Marwijk et al (1996) [E] 0·26 (0·14–0·41) Volkers et al (2004) [CN] [E] 0·21 (0·11–0·34) Combined 0·47 (0·42–0·53) 0·0 0·2 0·4 0·6 Proportion (95% CI) 0·8 1·0 Figure 2: Meta-analytic sensitivity of unassisted diagnosis of depression in primary care Proportion meta-analysis plot (random effect). CN=case-note method. E=elderly sample. prevalence was 18·4% (13·5% to 23·9%) in studies recruiting strictly defined adult patients (aged 18–65 years); 27·6% (17·4% to 39·1%) for mixed adults and older people (≥65 years); and 15·9% (8·8% to 24·5%) for people aged 65 or more. Prevalence varied by country of origin, being highest in the Netherlands (22·6%, 13·5% to 33·3%) and lowest in Australia (10·2%, 5·0% to 17·0%). In 17 DSM-based studies, depression 614 prevalence was 17·3% (12·2% to 23·1%) and for ICD-based studies it was 17·2% (10·4% to 25·4%). Across 41 studies, GPs correctly identified depression in 2514 out of 5534 true cases (45·4% uncorrected identification). With the random-effects weighted proportion meta-analysis, this detection rate was corrected to 47·3% (41·7% to 53·0%) (figure 2). However, diagnostic sensitivity varied greatly between individual studies, ranging from 6·6% to 78·8%. Contemporaneous ratings were more accurate than case-note methods (52·6% [46·5% to 58·6%] vs 33·6% [22·4% to 45·7%]; χ²=24·7, p<0·001). Diagnostic sensitivity was greater in case-note studies using cumulative recognition than in those using cross-sectional methods (36·8% [21·1% to 54·0%] vs 29·9% [14·3% to 48·3%]; χ²=11·9, p<0·001). Diagnostic sensitivity seemed to be slightly greater when GPs attempted to diagnose depression in elderly people (>65 years) than in younger people (49·6% vs 45·1%; χ²=2·97, p=0·08), which was significant when restricted to prospective non-cumulative studies (61·4% vs 49·3%; χ²=12·3, p=0·001). Diagnostic sensitivity did not differ for ICD-defined or DSM-defined depression, but differed according to country of origin. To assess other possible predictors, we analysed the association between diagnostic sensitivity and sample size, effect of blinding, year of publication, prevalence, practice size, and mean patient age. No significant associations were found. Data were insufficient to analyse the effect of severity of depression. Across 19 studies reporting full data for rule-in and rule-out accuracy, GPs could correctly exclude 5408 out of 6560 non-depressed individuals, giving a raw detection specificity of 82·4% (81·5% to 83·3%). The adjusted meta-analytic (random-effects) specificity was 81·3% (74·5% to 87·3%; figure 3). In this restricted dataset, re-calculation of the detection sensitivity showed a meta-analytic sensitivity of 50·1% (41·3% to 59·0%). The overall fraction correct was 74·2% (69·2% to 78·9%). Figure 4 shows the results of the bivariate meta-analysis with the 95% CIs and a 95% prediction line (the region within which we could expect the results of a future study to fall). In this smaller sample, data were insufficient to analyse the effect of cumulative recognition. Regarding method of case ascertainment, studies using contemporaneous ratings had inferior specificity compared with those using the case-note method (79·7% vs 88·8%), although this difference was not significant. When comparing accuracy in the dataset of 19 studies for individuals of different ages, no difference existed in GPs’ ability to diagnose depression in elderly compared with nonelderly samples (sensitivity elderly 56·3% vs adult 47·5%; specificity elderly 73·5% vs adult 84·5%). The overall fraction correct was 70·7% for depression in elderly people and 78·6% for that in younger people. Accuracy varied by country of origin, being highest in Italy and lowest in the UK (fraction correct 82·5% vs 50·0%). www.thelancet.com Vol 374 August 22, 2009 Articles From the findings of the 19 studies (sensitivity 50·1%, specificity 81·3%, prevalence 21·9%), the corrected positive predictive value was 42·0% (39·6% to 44·3%) and negative predictive value was 85·8% (84·8% to 86·7%). The positive likelihood ratio was 2·37 (2·21 to 2·54) and the negative likelihood ratio was 0·64 (0·61 to 0·67). To test our hypothesis, we calculated the misclassification rates for each 100 consecutive presentations (figure 5). In general, a motivated GP in an urban setting (where the rate of depression is 20%) would correctly diagnose ten out of 20 cases, missing ten true positives. The GP would correctly reassure 65 out of 80 non-depressed individuals, falsely diagnosing 15 people as depressed. Thus, the fraction correct would be 75%, with a misclassification rate of 25% and a number needed to screen of 2·0. Out of every five cases judged as depressed, only two would be true cases but six out of seven non-depressed individuals would be correctly reassured. With a Bayesian analysis to apply this accuracy to a low-risk sample (such as a rural setting) in which the prevalence is 10%, a GP would correctly identify five out of ten cases, missing five true positives. A typical GP would correctly reassure 73 out of 90 non-depressed individuals, but might falsely diagnose 17 people as depressed. Thus, the fraction correct would be 78%, with a misclassification rate of 22% and a number needed to screen of 1·77. Therefore, every five cases judged as depressed, only one is a true case (positive predictive value 22·9%), and every 20 cases judged as healthy, about 19 are correctly reassured (negative predictive value 93·6%). A www.thelancet.com Vol 374 August 22, 2009 0·98 (0·96–0·99) 0·94 (0·92–0·95) 0·90 (0·86–0·93) 0·90 (0·73–0·98) 0·89 (0·79–0·96) 0·89 (0·84–0·93) 0·86 (0·83–0·89) 0·86 (0·67–0·96) 0·84 (0·78–0·89) 0·79 (0·74–0·83) 0·74 (0·69–0·78) 0·68 (0·59–0·76) 0·60 (0·52–0·67) 0·93 (0·90–0·95) 0·77 (0·68–0·85) 0·72 (0·66–0·77) 0·71 (0·64–0·78) 0·63 (0·52–0·73) 0·61 (0·58–0·64) 0·81 (0·75–0·87) 0·5 B Van Weel-Baumgarten et al (2000) Aragones et al (2004) Pini et al (1997) Henkel et al (2003) Gerber et al (1989) Christensen et al (2003) Tiemens et al (1999) Balestriere et al (2004) Aben et al (2003) Klinkman et al (1997) Schulberg et al (1985) [CN] Schmaling and Hernendez (2005) [CN] Gledhill et al (2003) O’Connor et al (2001) [E] Turrina et al (1994) [E] Bowers et al (1990) [E] Crawford et al (1998) [E] [CN] Pond et al (1994) [E] Van Marwijk et al (1996) [E] Combined C Schulberg et al (1985) [CN] Balestriere et al (2004) Van Weel-Baumgarten et al (2000) Christensen et al (2003) Gerber et al (1989) Tiemens et al (1999) Klinkman et al (1997) Henkel et al (2003) Pini et al (1997) Gledhill et al (2003) Schmaling and Hernendez (2005) [CN] Aragones et al (2004) Aben et al (2003) Van Marwijk et al (1996) [E] Pond et al (1994) [E] Turrina et al (1994) [E] Crawford et al (1998) [E] [CN] Bowers et al (1990) [E] O’Connor et al (2001) [E] Combined Discussion We identified 41 studies that assessed the unassisted ability of GPs to diagnose depression against a robust outcome standard of psychiatric interviews. The overall prevalence of depression was 19·5% in various mainly urban primary care practices across more than ten countries. 19 studies examined the ability of GPs to accurately rule in a diagnosis of depressed people and rule out a diagnosis of non-depressed people. We found a diagnostic sensitivity of 47·3–50·1%, suggesting that GPs can generally identify about half of true cases. However, documentation of depression in medical notes occurred for only one in three depressed individuals; indeed, when case notes were examined retrospectively, identification was as low as one in five true cases.49,52 GPs might be reluctant to regard patients as depressed even if the criteria for the disorder are met.63,64 At the same time, GPs could accurately exclude 81·3% of non-depressed individuals. In this meta-analysis, sensitivity is higher but specificity lower than that in large-scale studies that used severity scales to rate depression.65 Our hypothesis of misidentifications being frequent was confirmed by our data that suggest that, in a typical urban practice with motivated staff, out of 100 unselected people the rate of false positives would outnumber that of true positives by about 50%. In a rural practice, the rate of false positives would Schulberg et al (1985) [CN] Balestriere et al (2004) Schmaling and Hernendez (2005) [CN] Aben et al (2003) Van Weel-Baumgarten et al (2000) Christensen et al (2003) Tiemens et al (1999) Gledhill et al (2003) Gerber et al (1989) Klinkman et al (1997) Henkel et al (2003) Pini et al (1997) Aragones et al (2004) Van Marwijk et al (1996) [E] Pond et al (1994) [E] Crawford et al (1998) [E] [CN] Turrina et al (1994) [E] Bowers et al (1990) [E] O’Connor et al (2001) [E] Combined 0·0 0·5 0·7 0·9 1·1 0·79 (0·61–0·91) 0·72 (0·63–0·80) 0·71 (0·59–0·82) 0·64 (0·52–0·75) 0·57 (0·46–0·68) 0·46 (0·34–0·58) 0·40 (0·33–0·48) 0·39 (0·33–0·45) 0·38 (0·21–0·58) 0·35 (0·22–0·49) 0·26 (0·11–0·46) 0·21 (0·14–0·28) 0·20 (0·03–0·56) 0·75 (0·63–0·86) 0·72 (0·61–0·81) 0·60 (0·32–0·84) 0·52 (0·39–0·65) 0·50 (0·21–0·79) 0·26 (0·14–0·41) 0·50 (0·41–0·59) 0·2 0·4 0·6 0·8 1·0 0·91 (0·88–0·94) 0·86 (0·85–0·88) 0·86 (0·77–0·92) 0·79 (0·74–0·84) 0·76 (0·70–0·81) 0·75 (0·71–0·78) 0·73 (0·69–0·78) 0·72 (0·68–0·77) 0·69 (0·62–0·76) 0·68 (0·51–0·82) 0·68 (0·63–0·72) 0·64 (0·59–0·70) 0·64 (0·50–0·76) 0·87 (0·84–0·90) 0·74 (0·65–0·82) 0·71 (0·65–0·77) 0·68 (0·62–0·73) 0·62 (0·52––0·72) 0·62 (0·59–0·65) 0·74 (0·69–0·79) 0·7 0·9 Proportion (95% CI) 1·1 Figure 3: Meta-analytic specificity of unassisted diagnosis of depression in primary care (A) Specificity. (B) Sensitivity. (C) Fraction correct. Proportion meta-analysis plot (random effect). CN=case-note method. E=elderly sample. 615 Articles 1·0 Study estimate HsROC curve 95% prediction region 95% confidence region Summary point 0·8 Sensitivity 0·6 0·4 0·2 0 1·0 0·8 0·4 0·6 0·2 0 Specificity Figure 4: Bivariate meta-analysis of unassisted diagnosis of depression in primary care HsROC=Hierarchical summary Receiver Operator Characteristic. Prevalence 27·3% False negatives (%) True positives (%) True negatives (%) 13·6 False positives (%) Non-depressed Depressed 13·7 59·1 13·6 Prevalence 20% 10·0 Prevalence 10% Fraction correct 5·0 5·0 10·0 15·0 65·0 Non-depressed Depressed 73·2 16·8 Depressed 0 5·0 10·0 Non-depressed 15·0 20·0 25·0 30·0 35·0 40·0 45·0 50·0 55·0 60·0 65·0 70·0 75·0 80·0 85·0 90·0 95·0 100·0 Figure 5: Real-world proportions correctly and incorrectly diagnosed per 100 consecutive primary care assessments outnumber that of true positives by three to one. The rate of false positives is high because even a modest rule-out error rate becomes significant when the number of nondepressed participants is large. Assuming these detection rates hold true, a tipping point occurs at a prevalence of 27·3%. Below this value, the number of false positives outweigh the number of false negatives (figure 5). 616 The highest risk of creating more false positives is in those with known risk factors for depression.66 Indeed, the clinical significance of a high false-positive rate depends on the action taken after a diagnosis of depression—eg, whether the next step is active treatment, watch and wait, referral, or re-assessment. The risk of both false positives and false negatives diminishes if few people are offered help. A simple way of managing a high false-positive rate on initial screening is a multistep assessment process applied to routine clinical examinations. Assuming 50% sensitivity and 80% specificity, if each suspected case is re-assessed then the overall accuracy of the clinician would increase from 77·2% to 89·5% if two assessments were offered in a rural setting, because a second assessment allows reexamination of not just true positives but also false positives. Additionally, cumulative assessment over a period of time was associated with a 35% increase in detection sensitivity, although data were inadequate to comment on specificity. This is consistent with previous work.61 For example, in the MAGPIE study,64 80·2% of cases seen five or more times during the previous year were correctly identified compared with 28·8% of those not seen in the previous year. Why do clinicians have difficulty diagnosing depression in routine practice? One issue is the modest rate of depression, which is between 10% and 20% in primary care.3–5 A low rate of depression favours identification of non-depressed cases, whereas a high rate favours diagnosis of depression. Not all depressed individuals spontaneously express emotional symptoms,67–69 but those depressed individuals who volunteer psychological complaints receive higher detection rates.5,70 Another issue is the complexity of fitting the continuous variation in depression severity into a categorical diagnosis. When allowed to choose between definite depression, possible (subthreshold) depression, and unclear cases, clinicians rate over a third of their decisions as not definitive.14 Many errors are incorrect estimates of severity—for example, when symptoms are diagnosed but judged clinically insignificant.31,64,66 For example, non-depressed individuals in primary care thought to have a psychiatric disorder were twice as likely to develop depression in the following year compared with non-depressed individuals who were not diagnosed as depressed.71 In the Hampshire depression project,65 72% of people who were missed had very mild depression, according to the hospital anxiety and depression scale (HADS). Thus, more-severe cases of depression are diagnosed more reliably than less-severe forms.22,50,59 Similarly, many false positives have related disorders, such as anxiety or adjustment disorders or subthreshold mood disorder, and in some cases re-assessment allows revision of the incorrect diagnosis or simply improvement of the therapeutic alliance. A further factor is consultation time. Short appointments seem to compromise www.thelancet.com Vol 374 August 22, 2009 Articles diagnosis in difficult cases.72 Patients with depression might feel under time pressure during consultations to the extent that they might often be inhibited from fully disclosing their problems.73 Another issue is whether GPs prioritise the appropriate signals for major depression. Stressful life and occupational, financial, and housing issues were the most influential prompts to a diagnosis.74 Other predictors of diagnostic sensitivity (recognition) include a better therapeutic relationship, presentation with psychological symptoms, clinician experience, and contact with the patient.14,50,64,75 Our analysis has several limitations. First, we have no standardised method to rate a health professional’s opinion as to whether a person is depressed or not. In a study from van Weel-Baumgarten and colleagues44 in the Netherlands, GPs were trained to use the criteria for the International Classification of Health Problems in Primary Care (ICHPPC-2). Although this was a routine diagnostic method, training might explain the unusually high sensitivity in this study. Examination of medical notes is not optimal because written records might not indicate the up-to-date opinion of the doctor.76 A contemporaneous (exit) rating of the preferred diagnosis of a GP, together with blind re-interview of the patient, is the most robust method but is subject to a hawthorn effect. In most cases, allocation concealment existed but this was not explicitly declared in every study. However, the two assessments must not occur too far apart to avoid clinical change contaminating the result. In this analysis, prospective ratings had better sensitivity but less specificity than case-note ascertainment. Another limitation is that a health professional might revise his/ her opinion at a later stage on the basis of further examination where diagnosis is initially uncertain. Thus, the accuracy of GPs is likely to vary by the number of contacts. Few studies have been properly designed to account for this variation and we found only one study that used prospective GP case ascertainment in a cumulative manner.25 Finally, data were insufficient to adjust for the effect of severity of depression, which has been strongly associated with detection accuracy in previous work. Many interventions have been aimed at improving the accuracy of the diagnosis of depression in primary care. Overall, although diagnosis seems to be modestly improved by use of scales, a positive effect on overall patient outcomes has not been shown.77 Only when screening is paired with organised systems of depression care can outcomes be improved. For example, Christensen and colleagues26 recruited 38 GPs in Aarhus county, Denmark. They randomly assigned 900 patients to severity questionnaires with feedback of results and 885 to routine care. The rate for unassisted diagnosis of depression was 35% and disclosure of screening questionnaire increased GPs’ recognition of mental disorders by only 6·6%, with the greatest effect in those GPs with moderate or low recognition rates. www.thelancet.com Vol 374 August 22, 2009 Our results should not be interpreted as a criticism of GPs for failing to diagnose depression but rather a call for better understanding of the problems that nonspecialists face. No data suggest that GPs do worse than other non-psychiatric medical colleagues.11 Depression is not common in primary care with at least eight out of ten attendees not suffering serious depression. Furthermore, depression rarely occurs on its own (>90% of people with depression have another physical or mental disorder).78 Finally, depressed patients often present with physical symptoms and hence their mental health is often overlooked.49 In primary care, time and resources are limited, and psychological or even structured self-help programmes are often not available. Antidepressants are often the treatment of choice for clinicians, but frequently not the first choice for patients, and hence management of depression can be difficult or unfulfilling. Only about 50% of true cases are diagnosed and 15% treated.61 Conversely, about 80% of non-depressed individuals are correctly reassured. Because one-off brief assessments only facilitate identification of about half of those with depression, we suggest that additional consultation time should be available for those likely to have depression. Repeated assessments by the GP or other professional in a collaborative model with a case manager might help to reduce diagnostic errors and improve overall quality of care.79,80 Contributors AJM did data extraction, critical appraisal, data analysis, data interpretation, and writing of the report. AV and SR did data extraction, critical appraisal, and data interpretation. Conflicts of interest We declare that we have no conflicts of interest. Acknowledgments We thank Alex Sutton for advice on bivariate meta-analysis. We also thank the staff of the postgraduate library, Leicester General Hospital (Leicester, UK). References 1 Greenberg P, Stiglin L, Finkelstein S, Berndt E. Depression: a neglected major illness. J Clin Psychiatry 1993; 54: 419–24. 2 Broadhead WE, Blazer DG, George LK, Tse CK. Depression, disability days, and days lost from work in a prospective epidemiologic survey. JAMA 1990; 264: 2524–28. 3 Ustun TB, Von Korff M. Primary mental health services. In: Ustun TB, Sartorius N, eds. Mental illness in general health care: an international study. Chichester, UK: John Wiley & Sons, 1995: 347–60. 4 Herrman H, Patrick DL, Diehr P, et al. Longitudinal investigation of depression outcomes in primary care in six countries: the LIDO study. Functional status, health service use and treatment of people with depressive symptoms. Psychol Med 2002; 32: 889–902. 5 Wittchen HU, Pittrow D. Prevalence, recognition and management of depression in primary care in Germany: the Depression 2000 study. Human Psychopharmacol 2002; 17 (suppl 1): S1–11. 6 Gilchrist G, Gunn J. Observational studies of depression in primary care: what do we know? BMC Fam Pract 2007; 8: 28. 7 Friedman B, Conwell Y, Delavan RL. Correlates of late-life major depression: a comparison of urban and rural primary care patients. Am J Geriatr Psychiatry 2007; 15: 28–41. 8 Harman JS, Veazie PJ, Lyness JM. Primary care physician office visits for depression by older Americans. J Gen Intern Med 2006; 21: 926–30. 617 Articles 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 618 Hirschfeld R, Keller MB, Panico S, et al. The National Depressive and Manic-Depressive Association consensus statement on the undertreatment of depression. JAMA 1997; 277: 333–40. Depression Guideline Panel. Depression in primary care. Treatment of major depression. Rockville, MD: US Department of Health and Human Services, Agency for Health Policy and Research, 1993. Cepoiu M, McCusker J, Cole MG, Sewitch M, Belzile E, Ciampi A. Recognition of depression by non-psychiatric physicians— a systematic literature review and meta-analysis. J Gen Intern Med 2008; 23: 25–36. Zastrow A, Faude V, Seyboth F, Niehoff D, Herzog W, Löwe B. Risk factors of symptom underestimation by physicians. J Psychosom Res 2008; 64: 543–51. Maginn S, Boardman AP, Craig TKL, Haddad M, Heath G, Stott J. The detection of psychological problems by general practitioners. Influence of ethnicity and other demographic variables. Soc Psychiatry Psychiatr Epidemiol 2004; 39: 464–71. Wittchen HU, Hofler M, Meister W. Prevalence and recognition of depressive syndromes in German primary care settings: poorly recognized and treated? Int Clin Psychopharmacol 2001; 16: 121–35. Alonso J, Codony M, Kovess V, et al. Population level of unmet need for mental healthcare in Europe. Br J Psychiatry 2007; 190: 299–306. Druss BG, Wang PS, Sampson NA, et al. Understanding mental health treatment in persons without mental diagnoses: results from the National Comorbidity Survey replication. Arch Gen Psychiatry 2007; 64: 1196–203. Pai M, McCulloch M, Enanoria W, Colford JM. Systematic reviews of diagnostic test evaluations: what’s behind the scenes? ACP J Club 2004; 9: 101–03. Irwig L, Macaskill P, Glasziou P, Fahey M. Meta-analytic methods for diagnostic test accuracy. J Clin Epidemiol 1995; 48: 119–30. Macaskill P, Glasziou P, Irwig L. Meta-analysis of diagnostic tests In: Armitage P, ed. Encyclopedia of Biostatistics, 2nd edn, 2005. Mitchell AJ. Predictive summary index and other summary measures of diagnostic accuracy. In: Kattan MW, ed. Encyclopedia of Medical Decision Making, 2009. Reitsma JB, Glas AS, Rutjes AWS, Scholten RJPM, Bossuyt PM, Zwinderman AH. Bivariate analysis of sensitivity and specificity produces informative summary measures in diagnostic reviews. J Clin Epidemiol 2005; 58: 982–90. Aben I, Verhey F, Beusmans G, Lodder J. Recognition and treatment of post-stroke depression in general practice. Huisarts en Wetenschap 2003; 46: 487–92. Aragones E, Pinol JL, Labad A. The overdiagnosis of depression in non-depressed patients in primary care. Fam Pract 2006; 23: 363–68. Balestrieri M, Carta MG, Leonetti S, Sebastiani G, Starace F, Bellantuono C. Recognition of depression and appropriateness of antidepressant treatment in Italian primary care. Soc Psychiatry Psychiatr Epidemiol 2004; 39: 171–76. Borowsky SJ, Rubenstein LV, Meredith LS, Camp P, Jackson-Triche M, Wells KB. Who is at risk of nondetection of mental health problems in primary care? J Gen Intern Med 2000; 15: 381–88. Christensen KS, Toft T, Frostholm L, Ornbol E, Fink P, Olesen F. The FIP study: a randomised, controlled trial of screening and recognition of psychiatric disorders. Br J Gen Pract 2003; 53: 758–63. Gerber PD, Barrett J, Barrett J, et al. Recognition of depression by internists in primary care A comparison of internists and ”gold standard” psychiatric assessments. J Gen Intern Med 1989; 4: 7–13. Gledhill J, Kramera T, Iliffe S, Garrald ME. Training general practitioners in the identification and management of adolescent depression within the consultation: a feasibility study. J Adolesc 2003; 26: 245–50. Henkel V, Mergl R, Kohnen R, Maier W, Möller HJ, Ulrich Hegerl U. Identifying depression in primary care: a comparison of different methods in a prospective cohort study. BMJ 2003; 326: 200–01. Jones LR, Badger LW, Ficken RP, Leeper JD, Anderson RL. Inside the hidden mental health network. Examining mental health care delivery of primary care physicians. Gen Hosp Psychiatry 1987; 9: 287–93. 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 Klinkman MS, Coyne JC, Gallo S, Schwenk TL. Can case finding instruments be used to improve physician detection of depression in primary care. Arch Fam Med 1997; 6: 567–73. Lecrubier Y. Widespread underrecognition and undertreatment of anxiety and mood disorders: results from 3 European studies. J Clin Psychiatry 2007; 68 (suppl 2): 36–41. Menchetti M, Belvederi M, Murri B, Bertakis K, Bortolotti B, Berardi D. Recognition and treatment of depression in primary care Effect of patients’ presentation and frequency of consultation. J Psychosom Res 2009; 66: 335–41. Ormel J, Van Den Brink W, Koeter MWJ, et al. Recognition, management and outcome of psychological disorders in primary care: a naturalistic follow-up study. Psychol Med 1990; 20: 909–23. Pini S, Berardi D, Rucci P, et al. Identification of psychiatric distress by primary care physicians. Gen Hosp Psychiatry 1997; 19: 411–18. Ronalds C, Creed F, Stone K, Webb S, Tomenson B. Outcome of anxiety and depressive disorders in primary care. Br J Psychiatry 1997; 171: 427–33. Schwenk TL, Coyne JC, Fechner-Bates S. Differences between detected and undetected patients in primary care and depressed psychiatric patients. Gen Hosp Psychiatry 1996; 18: 407–15. Simon GE, VonKorff M. Recognition, management, and outcomes of depression in primary care. Arch Fam Med 1995; 4: 99–105. Simon GE, Goldberg D, Tiemens BG, Ustun TB. Outcomes of recognized and unrecognized depression in an international primary care study. Gen Hosp Psychiatry 1999; 21: 97–105. Strauss PR, Gagiano CA, van Rensburg PH, de Wet KJ, Strauss HJ. Identification of depression in a rural general practice. S Afr Med J 1995; 85: 755–59. Tiemens BG, Ormel J, Jenner JA. Training primary-care physicians to recognize, diagnose and manage depression: does it improve patient outcomes? Psychol Med 1999; 29: 833–45. Tiemens BG, VonKorff M, Lin EH. Diagnosis of depression by primary care physicians versus a structured diagnostic interview. Understanding discordance. Gen Hosp Psychiatry 1999; 21: 87–96. Tylee AT, Freeling P, Kerry S. Why do general practitioners recognize major depression in one woman patient yet miss it in another? Br J Gen Pract 1993; 43: 327–30. van Weel-Baumgarten EM, van den Bosch WJ, van den Hoogen HJ, Zitman FG. The validity of the diagnosis of depression in general practice: is using criteria for diagnosis as a routine the answer? Br J Gen Pract 2000; 50: 284–87. Von Korff, Shapiro S, Burke JD, et al. Anxiety and depression in a primary care clinic. Arch Gen Psychiatry 1987; 44: 152–56. Füredi J, Rózsa S, Zámbori J, Szádóczky E. The role of symptoms in the recognition of mental health disorders in primary care. Psychosomatics 2003; 44: 402–06. Kirmayer LJ, Robbins JM, Dworkind M, Yaffe MJ. Somatization and the recognition of depression and anxiety in primary care. Am J Psychiatry 1993; 150: 734–41. Miller MJ, McCrone S. Detection of depression in primary care. Military Med 2005; 170: 158–63. Munitz H, Valevski A, Weizman A, et al. Recognition and treatment of depression in primary care settings in 6 different countries: a retrospective file analysis by WHO. Eur J Psychiatry 2000; 14: 85–93. Nuyen J, Volkers AC, Verhaak PFM, et al. Accuracy of diagnosing depression in primary care: the impact of chronic somatic and psychiatric co-morbidity. Psychol Med 2005; 35: 1185–95. Rost K, Zhang M, Fortney J, et al. Persistently poor outcomes of undetected major depression in primary care. Gen Hosp Psychiatry 1998; 20: 12–20. Schmaling K, Hernendez D. Detection of depression among lowincome Mexican Americans in primary care. J Health Care Poor Undeserved 2005; 16: 780–90. Schulberg HC, Saul M, McClelland M, Ganguli M, Christy W, Frank R. Assessing depression in primary medical and psychiatric practices. Arch Gen Psychiatry 1985; 42: 1164–70. Williams JW, Mulrow C.D, Kroenke K, et al. Case-finding for depression in primary care: a randomized trial. Am J Med 1999; 106: 36–43. www.thelancet.com Vol 374 August 22, 2009 Articles 55 56 57 58 59 60 61 62 63 64 65 66 67 68 Crawford MJ, Prince M, Menezes P, Mann AH. The recognition and treatment of depression in older people in primary care. Int J Geriatr Psychiatry 1998; 13: 172–76. Volkers A, Nuyen J, Verhaak P, Schellevis F. The problem of diagnosing major depression in elderly primary care patients. J Affect Disord 2004; 82: 259–63. Bowers J, Jorm AF, Henderson S, Harris P. General practitioners’ detection of depression and dementia in elderly patients. Med J Aust 1990; 153: 192–96. Licht-Strunk E, Beekman ATF, de Haan M, van Marwijk HWJ. The prognosis of undetected depression in older general practice patients. A one year follow-up study. J Affect Disord 2009; 114: 310–15. O’Conner DW, Rosewarne R, Bruce A. Depression in primary care 2: general practioners’ recognition of major depression in elderly patients. Int Psychogeratrics 2001; 13: 367–74. Pond CD, Mant A, Kehoe L, Hewitt H, Brodaty H. General practitioner diagnosis of depression and dementia in the elderly: can academic detailing make a difference? Fam Pract 1994; 11: 141–47. Turrina C, Caruso R, Este R, et al. Affective disorders among elderly general practice patients: a two-phase survey in Brescia, Italy. Br J Psychiatry 1994; 165: 533–37. van Marwijk HWJ, de Bock GH, Hermans J, Mulder JD, Springer MP. Prevalence of depression and clues to focus diagnosis: a study among Dutch general practice patients 65+ years of age. Scand J Primary Health Care 1996; 14: 142–47. Rost K, Smith R, Matthews DB, Guise B. The deliberate misdiagnosis of major depression in primary care. Arch Fam Med 1994; 3: 333–37. Bushnell J. Frequency of consultations and general practitioner recognition of psychological symptoms. Br J Gen Pract 2004; 54: 838–42. Thompson C, Ostler K, Peveler RC, Baker N, Kinmonth A-I. Dimensional perspective on the recognition of depressive symptoms in primary care: the Hampshire depression project 3. Br J Psychiatry 2001; 179: 317–23. Balestrieri M, Baldacci S, Bellomo A, et al. Clinical vs structured interview on anxiety and affective disorders by primary care physicians. Understanding diagnostic discordance. Epidemiol Psichiatr Soc 2007; 16: 144–51. Prior L, Wood F, Lewis G, Pill R. Stigma revisited, disclosure of emotional problems in primary care consultations in Wales. Soc Sci Med 2003; 56: 2191–200. Cape J, McCullough Y. Patients’ reasons for not presenting emotional problems in general practice consultations. Br J Gen Pract 1999; 49: 875–79. www.thelancet.com Vol 374 August 22, 2009 69 70 71 72 73 74 75 76 77 78 79 80 Stromberg R, Wernering E, Aberg-Wistedt A, Furhoff A-K, Sven-ErikJohansson S, Backlund LG. Screening and diagnosing depression in women visiting GPs’ drop in clinic in primary health care. BMC Fam Pract 2008; 9: 34. Aragones E, Pinol JL, Labad A, et al. Detection and management of depressive disorders in primary care in Spain. Int J Psychiatry Med 2004; 34: 331–43. Barkow K, Maier W, Üstün TB, Gänsicke M, Wittchen H-U, Heun R. Risk factors for new depressive episodes in primary health care: an international prospective 12-month follow-up study. Psychol Med 2002; 32: 595–607. Hutton C, Gunn J. Do longer consultations improve the management of psychological problems in general practice? A systematic literature review. BMC Health Serv Res 2007; 7: 71. Pollock K, Grime J. Patients’ perceptions of entitlement to time in general practice consultations for depression: qualitative study. BMJ 2002; 325: 687. Saltini A, Mazzi MA, Del Piccolo L, Zimmermann C. Decisional strategies for the attribution of emotional distress in primary care. Psychol Med 2004; 4: 729–39. Hickie IB, Davenport TA, Scott EM, Hadzi-Pavlovic D, Naismith SL, Koschera A. Unmet need for recognition of common mental disorders in Australian general practice. Med J Austr 2001; 175 (suppl): S18–24. Rhodes A, Fung K. Self-reported use of mental health services versus administrative records: care to recall? Int J Methods Psychiatr Res 2004; 13: 165–75. Gilbody S, Sheldon T, House A. Screening and case-finding instruments for depression: a meta-analysis. CMAJ 2008; 178: 997–1003. Vuorilehto M, Melartin T, Isometsa E. Depressive disorders in primary care: recurrent, chronic, and co-morbid. Psychol Med 2005; 35: 673–82. Gilbody S, Whitty P, Grimshaw J, et al. Educational and organizational interventions to improve the management of depression in primary care—a systematic review. JAMA 2003; 289: 3145–51. Christensen H, Griffiths KM, Gulliver A, et al. Models in the delivery of depression care: a systematic review of randomised and controlled intervention trials. BMC Fam Pract 2008; 9: 25. 619