Computers & Electronics

A General Population Comparison of the Composite International Diagnostic Interview (CIDI) and the Schedules for Clinical Assessment In Neuropsychiatry (SCAN)

A General Population Comparison of the Composite International Diagnostic Interview (CIDI) and the Schedules for Clinical Assessment In Neuropsychiatry (SCAN)
of 13
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
  Psychological Medicine , 2001,  31 , 1001–1013.   2001 Cambridge University PressDOI: 10.1017  S003329170105418 Printed in the United Kingdom A general population comparison of the CompositeInternational Diagnostic Interview (CIDI) and theSchedules for Clinical Assessment in Neuropsychiatry(SCAN)  T. S. BRUGHA,   R. JENKINS, N. TAUB, H. MELTZER    P. E. BEBBINGTON From the Department of Psychiatry and Department of Epidemiology and Public Health ,  University of Leicester; and WHO Collaborating Centre ,  Institute of Psychiatry ,  Social Survey Division ,  Office forNational Statistics and Department of Psychiatry and Behavioural Sciences ,  University College ,  London ABSTRACTBackground.  In psychiatric surveys of the general population, there has been considerablediscrepancy between diagnoses obtained by fully structured interviews and those established bysystematic semi-structured clinical evaluation. The Composite International Diagnostic Interview(CIDI) is an example of the first type of interview widely used in general population surveys. WecompareditsperformanceindiagnosingcurrentdepressiveandanxietydisorderswiththeSchedulesfor Clinical Assessment in Neuropsychiatry (SCAN), a semi-structured diagnostic interviewadministered by clinically trained interviewers. Methods.  Household addresses in Leicestershire, UK, were randomly sampled and 860 adults werescreened with the Revised Clinical Interview Schedule. Adults with too few symptoms to fulfildiagnostic criteria for study disorders were excluded to increase the proportion re-interviewed whomet such criteria. Repeat diagnostic interviews with the CIDI and SCAN, ordered randomly, weresought from eligible screen positive respondents. Recalibrated CIDI prevalence estimates werederived from the SCAN classification using Bayesian statistics. Results.  Concordance ranged between ‘poor’ and ‘fair’ across almost all types of study disorders,and for co-morbidity. Concordance was somewhat better for severity of depression and when lowerdiagnostic thresholds were used for depression. Interview order effects were suggested with lowerconcordance when CIDI followed SCAN. Recalibration reduced the prevalence of depressive oranxiety disorder from 9  0 to 6  2%. Conclusions.  Community psychiatric surveys using structured diagnostic interview data must beinterpreted cautiously. They should include an element of clinical re-appraisal so findings can beadjusted for estimation differences between fully structured and clinical assessments. INTRODUCTION The rational allocation of mental health careresources (Brugha  et al  . 1997; Murray & Lopez,   Preliminary findings from this study were communicated at: theAssociation of European Psychiatrists, Cambridge, April 1996; theWPA Section of Epidemiology and Community Psychiatry, Sydney,Australia, 19 October 1997; and the International Association of Epidemiologists Congress (IEA99), Florence, Italy, September 1999.   Address for correspondence: Professor T. S. Brugha, Universityof Leicester, Department of Psychiatry, Section of Social andEpidemiological Psychiatry, Brandon Mental Health Unit, LeicesterGeneral Hospital, Gwendolen Road, Leicester LE5 4PW. 1997) is assisted by valid information on theprevalence of disorder. However, there has beenscepticism about the validity of fully structureddiagnosticpsychiatricinterviews,notablyamongpubichealthphysiciansandhealthpolicymakers(Bartlett & Coles, 1998; Leeman, 1998). Thishas been increased by findings of substantialdifferences in the prevalence of depression in theUSA in two large-scale surveys of psychiatricdisorder employing such methods (Regier  et al  .1998). 1001  1002  T. S. Brugha and others We have argued at length that as mostpsychiatric diagnoses have no satisfactory ex-ternalvalidation,forinstancethroughdetectablepathophysiology, diagnostic instruments shouldbe validated against the systematic observationof the syndrome in a way closest to theconceptualization underlying it (Brugha  et al  .1999 a ). This evaluation should be done byclinically trained interviewers.One example of a fully structured interview isthe Diagnostic Interview Schedule (DIS)(Robins  et al  . 1981). In general populationsurveys, the DIS, used by trained interviewers,yielded poor concordance, both with clinicianadministered DSM-III checklists (Helzer  et al  .1985) and with a present state StandardizedPsychiatric Examination (Anthony  et al  . 1985;Romanoski  et al  . 1988). The results from suchvalidationexercises(Anthony et al  .1985;Helzer et al  . 1985; Kessler  et al  . 1998) may have beencompromised by their incorporation in large-scale surveys with different aims (Brugha  et al  .1999 a ). Moreover, the clinician assessmentswere always conducted after the fully structuredinterview assessments, possibly introducing biasdue to order effects. Delays between the mainsurvey interview and clinician re-appraisal mayalso have reduced concordance (McLeod  et al  .1990).The Composite International Diagnostic In-terview (CIDI) (World Health Organization1993 a )was developed asanimprovementon theDIS, whose features it largely shares (Robins  etal  . 1988). Subsequently, extensive efforts weremadetoimprovethecomprehension,motivationand recall ofrespondents, and concordance withclinical evaluations was better with this in-strument (Kessler  et al  . 1998). However, theseevaluations employed non-blinded clinicallytrained interviewers, completing single diag-nostic modules of the Structured Clinical In-terview Schedule for DSM-IV (Spitzer  et al  .1992). The validity of the CIDI as an actual casefinding instrument in community surveys there-fore remains to be evaluated. This requirementis more easily stated than achieved.In the general population, psychiatric symp-toms may be transient and relatively mild.Sources of diagnostic invalidity in communitysurveys may include respondents’ lack of under-standing of their task, or of the question asked,and a lack of ability and willingness to carry outthe task (Biemer  et al  . 1991; Kessler  et al  . 1998;Turner  et al  . 1998). Diagnostic concordance hassometimes been better in clinical settings(Wittchen, 1994), and this may arise becausepatients may learn from their physician orpsychologist the correct meaning of terms suchas anxiety, phobia, panic and obsession (Brugha et al  . 1999 a ).Assessing correctly the presence of psycho-pathological phenomena requires expert judge-ment (Brugha  et al  . 1999 a ). Trainees learn toelicit and assess respondents’ individual descrip-tions of mental states, and to distinguishpathological from normal phenomena. Thesetwo skills must be clearly specified and taught(Brugha  et al  . 1999 a ). Standardized, semi-structured interviews systematize this clinicalprocess (Wing  et al  . 1990; Spitzer  et al  . 1992)and we would argue in theory, therefore, thatthey provide more valid assessments thanwholly structured interviews (Brugha  et al  .1999 a ). Nevertheless, measurement error, forexample arising from inter-interviewer vari-ability in rating thresholds, may underminevalidity (Bromet  et al  . 1986; Wittchen  et al  .1999).The interpretation of data in mental healthcommunitysurveysderivedfromfullystructuredinterviews may be assisted by assessing overallsurvey error (Kruskal, 1991), for which directclinical assessment has been advocated (Brugha et al  . 1999 a ). According to Kruskal ‘ourmotivations for attacking errors … of measure-ments are mainly to gain improved under-standing of the process, thus making betterdecisions based on the measurements …’(Kruskal, 1991). After the completion of theBritish National Psychiatric Morbidity Survey(Jenkins  et al  . 1997 a ), a separate survey wasconducted with this in mind (Brugha  et al  .1999 b ). Ratings on SCAN, a semi-structureddiagnostic interview administered by speciallytrained interviewers (Wing  et al  . 1990), werecompared with data from a fully structureddiagnostic interview developed in Great Britain(Lewis  et al  . 1992), and from the CIDI (Robins et al  . 1988). The feasibility of prevalenceestimates derived from the fully structureddiagnostic interviews was examined by usingSCAN as the reference measure.  A general population comparison of the CIDI and SCAN   1003 METHODSample Two thousand five hundred Postcode AddressFile delivery points (Wilson & Elliot, 1987) wererandomly sampled in urban, suburban and ruralLeicestershire, England, an area chosen becauseit has socio-economic characteristics represen-tative of Great Britain as a whole (Brugha  et al  .1999 b ). Addresses were allocated randomly tointerviewers, who then randomly sampled oneeligible adult within each household (Kish,1965). Respondents had to be aged 16 to 64years and capable of an interview in English.Adults normally resident in institutions orelsewhere were excluded. Design A two-phase survey design was used (Pickles  etal  . 1995) (Fig. 1). In phase one, eligible adultswere screened with the Revised Clinical In-terview Schedule (CIS-R) (Lewis  et al  . 1992).ThefrequencyofICD-10diagnosesateach CIS-R score level was determined using nationaldata (Meltzer  et al  . 1995). Respondents withscoresbelow8ontheCIS-Rwerenotconsideredfurther because they were very unlikely to meetdiagnostic criteria on the follow-up instrument(Brugha  et al  . 1999 b ; Meltzer  et al  . 1995). Usingrandom numbers, subjects who were screenpositive were allocated either to have a SCANphase two diagnostic interview or a CIDIdiagnostic interview. The second phase twointerview, using the remaining diagnostic meas-ure, was then to be completed within 2 weeks(Fig. 1).If the population value of the primaryconcordance estimator kappa (Cohen, 1968) is0  4, between 150 and 200 successful pairs of interviews would be required to test the nullhypothesis of a zero value for kappa at the 5%significance level with 80% power. Measures The CIDI and the SCAN were developed inparallel under WHO auspices (Robins  et al  .1988;Wing  et al  .1990), and are widelyendorsedtechniques for case identification. The CIDI is afully structured diagnostic interview (WorldHealthOrganization,1993).Itemployspreciselyworded questions that cannot be rephrased butmay be repeated.The 10th version of the semi-structured,Present State Examination, is the central com-ponent of the SCAN Version 1 (Wing  et al  .1998; World Health Organization Division of Mental Health, 1992). Every symptom in SCANis defined in detail (Wing  et al  . 1990). Suggestedwording is provided for eliciting each SCANsymptom. However, interviewers must probefurther until satisfied with the informationobtained, because it is they who must decide if symptom definitions are fulfilled.Clinical knowledge is not required for theCIDI, but must be obtained in order to useSCAN, although lay survey interviewers havebeen trained to use it reliably in a clinicalpopulation (Brugha  et al  . 1999 a ). Interviewers and procedures Fourteen interviewers were recruited. Twelvewere non-medical interviewers, including fiveuniversity psychology graduates. All 14 inter-viewers were trained by the Office of PopulationCensuses and Surveys (OPCS) (National Statis-tics) in structured interviewing, including theCIDI, and in the survey sampling techniquesused in the British National Surveys (Meltzer  etal  . 1995). They were assisted by an experiencedCIDI trainer. Two interviewers were physicians,with at least 3 years postgraduate experience inclinical psychiatry, who completed a standard 5-day course in the use of the SCAN (U      stu      n  et al  .1998). They undertook further training andpilot SCAN interviews until the reliability of their ratings was fully concordant with seniorSCAN trainers rating interviews alongsidethem.The study aim required that direct com-parisons of diagnostic outputs should revealdifferencesinthemethodofinterviewassessment(Brugha  et al  . 1999 a ), not differences in di-agnostic procedures or differences in the type of interviewer. Interviewers introduced themselvesto respondents as University of Leicester surveyinterviewers(notaspsychologistsorphysicians).Both CIDI and SCAN were computer assisted:automated cut-offs minimized unnecessary orinappropriatequestions(WorldHealthOrganiz-ation,1993 a ,  b ;Der et al  .1998).Bothinterviewswere designed to collect data required by  1004  T. S. Brugha and others diagnostic criteria, and employ identical classifi-cation rules (World Health Organization,1993 a ,  b ).Interviewers visited respondents’ homes, andsought consent for the CIDI and SCAN inter-views from screen positive respondents (Brugha et al  . 1999 b ). Only 14 days were permittedbetween administration of the CIDI and SCAN.The two SCAN interviewers were allocatedcases according to their availability (non-randomly). All interviewers were blind to theresults of other interviews on the same re-spondent. Statistical analysis Sociodemographic and clinical factors poten-tially associated with non-completion of follow-up interviews, were assessed using logisticregressionmodels(Everitt&Der,1996).Becausewe did not interview screen negatives, thefindings have not been weighted for design ornon-response.ICD-10 algorithms developed by WHO forSCAN and CIDI (World Health Organization,1993 a ,  b ; Der  et al  . 1998) were used to classifyrespondentsinto ICD-10anxiety and depressioncategories (F32 to F42) within the past month(Tables1and 2).Diagnosesweregeneratednon-hierarchically; thus subjects with more than onediagnosis are classified according to each dis-order in the analysis. Different disorders werecombined into broader categories, including afinal‘catchall’categoryforrespondentsmeetingcriteria for any study disorder (Tables 1 and 2).Co-morbidity was defined as the co-occurrenceof at least two separate study disorders.Using SCAN as the reference instrument, wecalculated kappa(Cohen, 1968;Landis& Koch,1977), sensitivity, specificity and the proportionof subjects for whom a positive diagnosis wasrecordedbyeitherobserverinwhichbothagreed(the Index of Agreement) (Cicchetti & Feinstein,1990), with 95% confidence intervals. Sensitivity analyses Rigidly applied binary classifications (e.g. ‘de-pressed or not’) might conceal better agreementwhen different severity levels are considered.Kappavaries withthe frequency ofobservations(Cicchetti & Feinstein, 1990). Raising or re-ducing threshold for depression (by requiringfewer criteria for a diagnosis) allows this to beevaluated. Although developed in parallel underWHO auspices, discrepancies may yet existbetween published CIDI and SCAN algorithmsfor ICD-10, thus introducing classification error(Marcus & Robins, 1998). For these reasons anew depression algorithm was constructed com-mon to both interviews, based on all the ICD-10criteria for F32 to F33 (World Health Organiz-ation, 1993 a ,  b ). Thus, concordance was re-evaluated for: the diagnosis of depressive dis-order; a total score for depression; and a rangeof thresholds representing varying numbers of depression criteria (Table 3).Our assumption that the second stage screenmisses very few cases is well supported byevidence (Brugha  et al  . 1999 b ). Nevertheless,concordance analyses were carried out in whichboth zero and complete concordance in such‘missed cases’ was simulated. This analysiswould tell us what concordance would havebeen if the unassessed cases had yielded eitherperfect agreement or no agreement.As a check for variation between the twoSCAN interviewers’ data, key analyses wererepeated after dividing the total sample into twosubgroups A and B defined by SCAN inter-viewer. Recalibrated prevalence estimation To obtain recalibrated prevalence estimates, wesimulated a two-stage survey of 10000 adultswith CIDI used in the first stage and SCAN inthe second. A simulated rate of 900 cases (anystudydepressiveoranxietydisorderusingCIDI)was used. This rate is close to that actuallyfound in the present general population surveyin Leicestershire (Brugha  et al  . 1999 c ). TheCIDI prevalence rate was recalibrated by ap-plying the estimated positive predictive value(proportion CIDI positive cases who are SCANpositive) and negative predictive value (pro-portion CIDI negative cases who are SCANnegative).In order to estimate the precision of therecalibratedprevalencerate,confidenceintervalswere obtained from a Bayesian graphical modelusing BUGS computer software (Spiegelhalter et al  . 1995). This enabled us to take accountof assumptions made about the distribution of a variable before new empirical data areconsidered. Thus, it allowed for the differ-ent sampling procedures. For example, the  A  g e  n e r  a l     p o  p u l    a t   i   o n c  o m  p ar  i   s  o n o  f   t   h   e  C I   D I   a n d   S  CA N 1    0    0    5   Table 1.  Concordance ,  true negatives ,  false positives ,  false negatives and true positives of SCAN and CIDI  - Auto ICD -10  diagnoses present in the month before interview in second  - stage screen positive respondents in the general population  (N  172) Diagnosis ICD10 Code(s)SCAN  CIDIkappa* 95% CI             BothinterviewsnegativeOnlyCIDIpositiveOnlySCANpositiveBothinterviewspositiveDepressive disordersAny depressive episode F32.00, F32.01, F32.10, F32.11, F32.20 150 16 3 3 0  20   0  02 to 0  42Depressive episode or disorder Any F32 or any F33 144 22 3 3 0  15   0.04 to 0  33Dysthymia †  F34.1 158 5 3 4 0  48 0  17 to 0  79Any depression above (not remission) Any F32 or any F33 or F34.1 141 19 3 9 0  39 0  19 to 0  59Anxiety disordersAny agoraphobic anxiety disorder F40.00, F40.01 156 12 1 3 0  29 0  02 to 0  56Agoraphobia with panic disorder F40.01 166 4 1 1 0  27   0  17 to 0  71Social phobias F40.10 161 8 0 3 0  41 0  09 to 0  73Specific  isolated phobia F40.20 130 21 9 12 0  35 0  17 to 0  53Any phobic anxiety above F40.00, F40.01, F40.10, F40.20 120 27 8 17 0  38 0  22 to 0  54Panic disorder F41.00, F41.01 145 13 9 5 0  24 0  02 to 0  46Generalized anxiety disorder (GAD) F41.10, F41.11 152 17 3 0   0  03   0  06 to 0  00Any non-phobic anxiety disorder Any F41 (Panic or GAD) or F41.8 132 23 9 8 0  24 0  05 to 0  42Any phobia, panic, or GAD Any F40 or F41 103 33 12 24 0  35 0  20 to 0  50Obsessive–compulsive disorder (OCD) F42, (F42.1, F42.2 on SCAN only) 161 8 3 0   0  03   0  05 to 0  00Any of the above anxiety disorders Any F42, F41, or F40 99 35 13 25 0  33 0  18 to 0  47Any above ICD-10 depression or anxiety diagnosesDiagnoses covered in CIDI and SCAN Any F32, F33, F34.1, F40, F41, F41.8, F42 96 32 11 33 0  43 0  29 to 0  57* The proportion of observed agreements adjusted for the rate of agreement that would be expected just by chance. †  Sample size for dysthymia  170.
Similar documents
View more...
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks