INTERNATIONAL JOURNAL OF METHODS IN PSYCHIATRIC RESEARCH. VOL 5 wn DEVELOPMENT OF A SHORT QUESTIONNAIRE FOR USE IN EPIDEMIOLOGICAL STUDIES OF DEPRESSION IN CHILDREN AND ADOLESCENTS ADRIAN ANGOLD* MACPoych, ELIZABETH J. COSTELLO PAO and STEPHEN C. MESSER PD Developmental Epidemiology Program. Duke University Medical Center. Box 3434, Durham, North Caroling 27710-2484, USA ‘ANOREW PICKLES PhD, FRANCES WINDER BSc and DANA SILVER BS Department of Child adolescent Paychiairy, MRC Child Prychiatry Unit, Insitute of Psychiatry, 16 DeCrespigny Park, Denmark Hill, London SES BAF UK SUMMARY The purpose of the present study was to describe the development and assess the psychometric properties of the Short Mood and Feelings Questionnaire (SMFQ). The SMFQ is a brief, easy-to-administer, self-report measure of childhood and adolescent depression, designed for the rapid evaluation of core depressive symptomatology or for use in epidemiological studies, The SMFQ's content and criterion-elated validity were examined in a sample of 172 S16 year-olds, comprised of both psychiatric and unselected pediatric controls. Results revealed substeatial correlations between the SMFQ, the Children's Depression Inventory (CDI) and the Diagnostic Interview Schedule for Children (DISC) depression scale. The SMFQ successfully discriminated the clinically-referred, psychiatric subjects from the pediatric controls. Within the pediatric (general population) sample, the SMFQ disenminatel DISC-diagnosed children with depressive disorder from non-depressed subjects. Exploratory factor analyses, along. witha high internal consistency, suggesed that the SMFQ was a unifactorialsaie. In sum, the SMFQ appears to be 8 promising tool for both the swift assessment of core depressive symptomatology and ai a screening mecsure for ‘depression in child psychiatric epidemiological studies. KEYWorDs—depression; children and adolescents; assessment have the disorder, so that a brief screening questionnaire that’ identified children likely to hhave the disorder at interview would be very useful. Such @ questionnaire should be available in both self-report and parent-report forms. It also needs to be as short and simple as possible, both to reduce subject burden and to maximize further participation in later stages of a project. We have Pointed out elsewhere that very high degrees of INTRODUCTION Sao but they are the central pillar of the clinical and research psychiatric assessment of both adults and children. Wi (Weissman er al, jepression has a six-month prevalence of only 2-5% (Anderson et al, 1987; Boyd and Weissman 1981; Fleming and Offord, 1990; Guyer et al, 1989; Kashani er al,, 1983, McGee et al., 1990; McGee and Williams, 1988), which means that most children interviewed in Beneral population studies of depression will not ‘Author to whom correspondence should be addeesed CCC 1o4se8951 95 0902 E1995 by Joha Wiley € Sons. Lid sensitivity and specificity cannot be expected from such instruments and that they might be better regarded as ‘nets’ than as ‘screens’ (Costello and Angold, 1988). Even though nets will miss 2 umber of cases of interest and pick up a good deal of other material, they can be expected to increase the ‘concentration’ of cases in the interviewed sample sufficiently to be cost-effective. ‘Such a questionnaire could also serve a useful Purpose in providing for the Accepted 10 Nivember (984SHORT MOOD AND FECLINGS WI EsTIONNAIRE SEQ! Sex 318) ban Aye ear) 3B 2s Race 318) Black 2118) 162) Massing 4 t ot central to the concept of depression as it is usually understood and were not included. A parallel version was written for parents, containing the same items, with one addition that made sense only as something reported about the child by another person (‘S/he was not as happy as usual ‘even when you praised or rewarded him/her’). In order to make the questionnaire as simple as possible, each item consisted of a single sentence 10 which the ‘subject could respond ‘Not true’, "Sometimes’ or ‘True’. ‘We refer to this initial item pool as the Mood and Feelings Questionnaire (MFQ). The suffix C indicates the child self-report version, while P indicates the parent-report version. Samples Symptom scores wer that were already available to us Involvement in other studies, Psychiatrie services an Psychiairic group. This group comprised 48 consecutive child psychiatric outpatient referrals attending the ra Psychiatric Institute and Clinic in Pittsburgh and aged between 6 and 17 years. Pediatrie group. This group comprised 125 6-11 year-olds (only one was younger than seven) who ‘had been brought to the primary care pediatric clinic of a large Health Maintenance Organization for a wide variety of general health problems. This group was 4 consecutive subsample 239 oof childeen involved in a study of psychiatric morbidity and service use that included around 80% of children enrolled in the pediatric group. Costello ev al. (1988) have shown that this group was a close approximation to a general population sample both in terms of its demographics and patterns of psychiatric disorder. Table | shows the demographic characteristics Of the two groups. Measures In addition to the initial item pool, the Children’s Depression Inventory (CDI) (Kovacs, 1983) was used as one point of comparison. The CDI is currently the most widely used depression questionnaire for children, designed with a format and content similar to that of the Beck Depression Inventory (BDI) (Beck and Beamesderfer, 1974). Each item consists of three statements about symptom area, encompassing different levels of severity (for example; I hate myself; Ido not like myself; I like myself.) The child endorses the one closest to how he or she has felt or thought during the preceding 2 weeks. Though this approach works well enough for many purposes, it is somewhat more complex than the method of questioning adopted for the SMFQ, in that it requires the subject to compare three statements for each item, rather than making a decision about a single statement, In the pediatric group, the criterion used in making a diagnosis of DSM-III depression was a detailed psychiatric assessment of the child using the Diagnostic Interview Schedule for Children (DISC) (Costello et al, 1982). This highly structured instrument consists of two. parallel interviews about the child's emotional and behavioral problems: the DISC-C for children and the DISC-P for parents and caretakers Responses to the DISC guestions are coded as 0 (no), | (Sometimes or maybe) or 2 (yes). They are Computer scored to yield symptom sores, and also DSM-III diagnoses, at two levels of severity, for a wide range of psychiatric disorders. Level 1 Cpossibie’ diagnosis) is. a _precise Operationalization of the DSM-III criteria. A weighted summary score generated by the DISC diagnostic algorithms was derived from the responses to the DISC and is referred to as the Total DISC Depression Score, This consisted of the sum of the scores on each of the depression, subsections of the DISC (affective. cognitive,SHORT Wo9D ANO FEELIN: QUESTIONNAIRE \SMeO) 241 Tuble $ Parent nitisl tem pool: stem sotal predictive power Fotal MEQ? score given sem score in) IRL MEQ iets a i SPE gk Comnatica tiem 1060) Miserable Ot unhappy 201831691868) OF 2.0001 076 Dwr’ enjoy anything 3.9115) 120 10) 014 043 aoa! 033 ess nungey 391) 98013) Ot 036 cose 038 Aw tore 26(10H loots 236) at enna 034 37 108) 13a) 038 onto 050 24098 $90 17908) LIS o.goo1 0 No good ay W243) 300) 118 0.0001 on Blumec self BAUM eccioD¥C1l) 5298001) Ot 038 noot 0. Ingecive 3008) 68) 98) aaa, 032 Festable 16168) 36053) 98@) 02g ot 033 “Talking less 3700) 870) 300) os, ass anor 0.50 Cried 3 ot 29(t0) 10918) 158) oss aeactin cinerea) is) © sO) «age og ane ose Not wort living 41 (1) 234G) ry) on Thoughts of death 43.18) ao) 80) 08 oe naz 049 Bester off without 3801) 156) 020 © 068 0.0007 083 Suicidal choughee 452) 133) 010 038 ngs 046 Not se frends 4311) 1040) 010 © 0300337 032 fe EAS sain DintS (I) GOs) 03k. Or ae 057 Fee pina MnnT SO 1970) azae Oar esiaane 04s Hated myself 3611) 300) 325) 03s 13a 09 ‘Bad person Looked ugly 32009" 0307 25) oom eot 046 Lonely 2760 5G = 35) ast kato on Usloved 3000) 24a) = 5000) 032 oe ooo 07 Never be as good 3300) 6S) 3500) 02k ost .ooat 087 Did everything wrong Poot sleep 3601) 81d) 3G) ws ast 0002 ost Slept more 4501) 78 09501802631 o3t Not cheered up 3515) 159) 3000) 038 coor 0.70 ‘PE, parameter esimate; ‘SPE, RESULTS The analyses broadly fall into three parts: (1) assessments of the internal structure of the initial item pool: (2) an assessment of the criterion- related validity of the initial item pool, as measured by its ability to predict clinical group status, and in the form of comparisons with the subjects" DISC seale score and CDI responses, and. (3) an assessment of the performance of the subset of empirically derived screening items emerging from (1) and (2), Content validity of the MEQ In dex one sense, content validity is ‘ermined by looking at the items in the simply e scale standardized parameter estimate; "Pearson product ‘moment correlation coefficient. and noting that they do indeed cover the depressive phenomenology embodied in both the DSM and ICD diagnostic systems. However, we hhave extended our consideration of this concept somewhat to include our findings on the acceptability of the instruments in use. Tt was our impression that, as part of a long assessment, the MFQ functioned well with both the younger children and the oldest subjects in the study. Its format presented no obvious disadvantages and seemed simpler in use than that of the CDI. The parent version was also quite Satisfactory in these respects. However, as the result of a typographical error while creating the arent version of the scale, the wording for items 24 (‘was a bad person’) and 32 (‘did everything wrong’) in the MFQ-P was incorrect. The resultsSHORT MUOD AND FEELINGS QUESTIONNAIRE ISMQs le § Initial stem poi: tems us predictors of sample membership MEQ sem Child se-epors Parent ceporis oR Miseable 190087 35 Fiest factor loadings of SMFQ-C and SMFQ-P Table 9. In lations among depres SMFOC SMFQP SMFQ-C SMFQ-P MFQ-C MFQ-? CDI DISC-C Miserable or untappy ass on SMFQC sajoy anvening 035 ost SMFQP 020 a3 MEQC 086 034 07 MFQP 028 091 No good 0.30 cor oer os Cred 3 lot ost DISC 06s 038 019 ous Poor concentration os7 DIP 032 a3 040 023 027 Hated myself 076 Bad person . Lonely on. Unloved on. 056 Never be at good Did everthing wrong to have identical items in the parent and child versions to allow direct comparisons between parent and child reports of individual items in the future, but since two of the items which performed well in the child reports were those with faulty questions in the parent version, the analyses of the parent data include only the 11 usable parent items, Internal structure of the SMFQ Internal reliability. Coefficient « was 0.85 for the SMFQ-C and 0.87 for the SMFQ-P. These levels suggest that both scales could be shortened still further. However, at this stage, given the limited size of our samples, we wished to avoid narrowing. the scale too severely, since this would carry a risk of reducing the replicability of our results in other populations. Maximum likelihood factor analysis. Given that one contributor to our decisions about items for inclusion in the SMFQ was a high item loading on the first factor from the principal components analysis of the original MFQ item pool, we expected that the SMFQ would be a unifactorial scale with high item loadings on the first principal factor. This is exactly the pattern that emerged (see Table 8). For both the SMFQ-C and SMFQ-P only a single factor had an eigenvalue greater than 17.98 and 9.92, respectively). Criterion validity Comparisons with the CDI and DISC depression scores. Table 9 shows the correlations amongst the child and parent total inital item pool scores (MFQ-EC), the SMFQ-C, the SMFQ-P, the CDI and the DISC depression scores, The correlations between the MFQ and the SMFQ are so high as to indicate that litle information has been lost in cutting out over half of the initial item pool items. The SMFQ-C correlates moderately highly with both the CDI score and the DISC depression score, and its correlation with the latter is similar to that of the CDI. Correlations between parent and child reports are at the expected level of around 0.2-0.4 for all the scales. Clinical status. Table 10 shows the mean scores of the pediatric and psychiatric groups on the various depression scales. The SMFQ-P and SMFQ-C discriminated clearly between the psychiatrically referred and the non-referred children. DISC depression diagnoses. It can be seen from Table 11 ‘that all of the scales discriminated between those with a DISC diagnosis of depres- sion and those without. The SMFQ scales did as well as the CDI. The DISC depression scales are included for comparison pusposes, and it can be seen that the other scales performed nearly as well as the DISC data on which the diagnoses themselves were based. Since it was possible that the MFQ might be detecting psychiatric disturbance in general, rather than specifically depression, we then comparedSHORT MCOD AND FEELINGS QUESTIONNAIRE (SMF) 247 0 0.2 04 os os 1 ‘One minus Specificity fia 1 ROC curves for BMEQ-C, SMFQ-P, SMFQ-C~P, CDI and DISC total depression sore agsnst DISC ression diagnosis.SHORT MOOD AND FEELINGS QUESTIONNAIRE (SMFQ) Fleming. J. £ and Offord. D. R. (1990) Epidemiology of childhood depressive disorders & critical review. J. dm, dead Child Adolescent Psychiat. 29, $71~580. Guyer, B.. Lescoher. L, Gallagher, &, S., Hausman, A. and Azzara, C. V. (1989) Intentional injuries among children and adolescents in Massachusetts. N. Eng. J Med. 7, 1$85-1589, Kashani, J. HL. McGee, 8. ©. Clarkson, S. Anderson, J. C., Walton, L. A.. Williams, §,, Silva, P. A. Robins. A. J.. Cytryn, L. and McKnew, D. H. (1983) Depression in a sample of 9-year-old children’ prevalence and associated characteristics. Arch. Gen Paychiat. 40, 1217-1223, Kazdin, A. E. (1987) Children's Depression Scale: validation with child psychiatric inpatients. J. Child Psychol. Psychiat. 28, 29-4, Kovacs, M. (1983) The Children's Depression Inventory 2 SelfeRated Depression Scale for School-Aged Youngsters. University of Pittsburgh School of Medicine, (Unpublished manuscrip). Lang, M. and Tisher, M. (1978) Children's Depression ‘Seale. Australian Council for Educational Research, Victoria, Australia, McGee, Rand Williams, S. (1988) A longitudinal study ‘of depression in nine-year-old children. J, Am. Acad. Child Adolescent Psychiat. 27, 342-348, McGee, R., Feehan, M., Williams, S., Partridge, F., Silva, P. A. and Kell, . (1990) DSMCIT disorders in 249 4 large sample of adolescents. J am. stead. Child Adotescent Psychiat. 27, 342-348 Reynolds, C. R. (1982) Covergent and divergent validity Of the Revised Children’s Manifest Anxiety Scale Edueat. Prychol. Measurement 42, 1205-121 Rotundo, N. and Hensley, V. R. (1984) The Children's Depression Scale: a study of its validity. J. Child Prychol. Psychiat 26, 917-92 Ryan, N. D., Puig-Antich, J, Cooper. T. By Rabinovich, H.. Ambrosini, P. j.. Fried. J. Davies, M,, Torres, D.’ and Suckow, R.'F. (1987) Relative safety of single versus divided dose Imipramine in adolescent major depression. J. dm. dead. Child Adolescent Psychiat. 26, (3), 400-106, Saylor, C.F., Finch Jr, A... Baskin, C. H., Saylor, C.B., Darnell, G. and "Furey, W. '(i984a) Children's Depression Inventory: investigation of procedures and correlates. J. dm. Acad. Child Adolescent Psychiat. 3, 626-63 Saylor, C. F., Finch, A. J., Spirito, A. and Bennett, B (19886) ‘The Children’s Depression Taventory: a systematic evaluation of psychometric properties. J Consult. Clin. Psychol. $2, 955-967. Weissman, M. M., Wickramaraine, P, Warmer, V. John, K., Prusoff, B. A., Merikangas, KR Gammon, G. D., Weise, J. R. and Cicchetti, D! (1987) Assessing ‘psychiatric disorders in children: ‘discrepancies between mothers" and children’s reports Arch. Gen. Prychiat, 44, 747-753,but where at least a fepression is required. A further Potentia| use is as a brief measure suitable for ‘more frequent follow-ups than the usual annual or semi-annual major assessments employed in many psychiatric epidemiological studies However. a net or screen that does not work would not be costeffective under any circumstances and may prove counter-productive (Kashani er al., 1983). Attention to some practical Psychometric ‘considerations (particularly the important distinction between ‘criterion-related and content validity) suggests that many of the available depression questionnaires for children and adolescents are likely to be suboptimal in their netting properties (Costello and Angold, 1988; Kazdin, 1987; Reynolds, 1982). In this’ paper, therefore, we discuss some of the important design considerations for a netting questionnaire describe the development and psychometric properties of the Short Mood and Feelings Questionnaire (SMFQ), which we have developed for use in future epidemiological studies, CRITERION-RELATED VERSUS CONTENT VALIDITY Researchers are often interested in at least two aspects of the phenomenology of depression. The first is the range of symptoms displayed by the subjects under study. They therefore require that their measures adequately reflect the range of phenomena that are associated with the disorder in Which they are interested. Depressive “symptomatology is quite wide- ranging, and most depression questionnaires include’ a variety of items covering areas as diverse as appetite disturbance, guilt and suicidal thinking. The degree to which a questionnaire, interview or test covers the full range of the material of interest is often referred to as its “content validity However, researchers often want to sake a step beyond the symptoms and to define disorders of iseases. If a questionnaire is being used to select Subjects who are likely to have a particular disorder. as defined by some criterion measure, then the degree to which it does so is referred to as the instrument's ‘rterion-related valiity’ If this ability to detect disorder isthe only consideration, A ANGOLD Ar 1c then the best instrument will usually be the one that performs the task of selecting cases and rejecting non-cases most efficiently and in the Shortest time: in other words with as few items as Possible. “Concurrent validity’ (a subset of Griterion validity) reters to the degree to which a ew measure agrees (or correlates) with assessments of the same consicuct by other (usually pre-existing) measures ‘As Cronbach (1970) has pointed out. the requirements of content validity and criterion validity are often contradictory. since a measure that covers the arca fully is likely to be relatively Jong and may be a poor predictor of a particular criterion of interest, while one that is short but highly predictive may provide poor content coverage. The Children's Depression Inventory (CDI) (Kovacs, 1983; Saylor et al., 1984a, 1984b) and the Children’s Depression Scale (CDS) (Lang and Tisher, 1978; Rotundo and Hensley, 1984: Kazdin, 1987), which are possible candidates for the role of first-stage netting questionnaire in epidemiological studies fall between the two stools of content- and criterion-related validity, combining moderate coverage of the area with relatively low predictive power for the diagnosis of depression (Costello and Angold, 1988). For instance, they have been reported to have very high levels of internal reliability (Cronbach's alphas frequently>0.9), suggesting a great deal of redundancy in the items as measures of the latent depression variable (Boyle, 1985). In other words, these scales are probably longer than they need to be for netting purposes. We therefore set ut to produce a netting questionnaire (the Short Mood and Feelings Questionnaire, SMFQ) that was empirically designed to minimize length and ‘maximize criterion validity. METHOD Initial item selection We began by selecting 30 symptom items (see Table 2) reflecting current clinical and taxonomic thinking about childhood depression (e.g. Ryan 2t al, 1987; American Psychiatric Association, 1980), which tapped affective, cognitive, vegetative and Suicidal aspects of depression. Symptoms such as ‘enuresis, wandering behavior and school phobia, whereas they might be related to depression in childhood and adolescence and are contained in some other depression scales 'Angold, 1988) are20 Table 2 ‘Total MEQLC score given item sore (a) A ANGOLD er Child initial tem pook: tem total predictive power ea at or! MFQ items ' Pe P Correlation! item —otal Pee Ea 3 hier SACOG THs Moan, a Dida'enjoy anything 9 9306) 1900) 3101) atte, GL Osa Less hungry 7605 136G8) 1610 009 ods onnt D4 Ate more 5916) 10907) 14001) 00 =a) os 012 Tired 626) 426) 172083 ln 058) gaan 048 Restless 82) .12300 040) a? aaa ata 039 No good 81¢0 20) 1G) ol tok ocomot 0.32 Blamed set M69») 4209-— 17909 018-039 oooot 0a Indecisive $5($ 11.7042) 166) = at2 sat 0.3 lentable 7503) 1520) 16803) 10st ooo os Talking lees Zo) 18605) saci) ato 0s2 oot 043 Cred 2 lot 9300 174117) 0) oma O46 Fa caine pape 68 ls nea) na 04s Not wert living $(10) 19311) 268) 020088 oot 036 eenebaset aiatied) (TOD UTS AD LobaseKsion suig bus BERIOAOECT out Sere eit 26 89 008) 8206 Mic 36, horns “OHV UNG “ApoE 047 Suicidal thoughes 3901) 217. 25@) ag 8s teed 038 Not se fends 8700) 19.70) 209 att 0ss anon 033 Totes SRBN sect GO ZB TIO. Oa 084) mortar re Fe ey ac TRO seo Nhl Ge Serie Stay eo) cna 083 Hated myself 84009 2070 270) as aaa 047 Bad person S100) 214@) 273) asker oss, Looked ugly 2109 13.708) 1830) ooT 37 og009 032 Lonely 720) 1230) 385013) oar 036 Unloved 24010) 17701) 25900) oles acot 055 Never be as good 8900 153) 2630) 03st ot oe Pe everthing wrong! = TEED W9SEH.- 934) 9s. tae” Get cr) Poor seep 730) 128@) 183@ ~—030at 09 Slept more 4501) 10060) 18317308 ast ga018 030 'PE, parameter estimate; ‘SPE, standardized parameter imate; Nesetative and suicidal) cach standardized to range between 0 and 50, Procedure The initial version of the MFQ was completed as part of a larger psychiatric assessment of the pediatric and psychiatric groups. The children cither completed the form on their own of had it read to them, depending upon their age and reading competence. The answers recorded were the subjects" seif reports and not the examiners” opinions about them. One parent was also asked to complete the parent version of the questionnaire Pearson product-moment correlation coefficient about the target child. The children also completed the CDI. The order of administration of the MFQ and CDI was randomized within each sample. The Parents and children in the pediatric group were then seen separately by different research interviewers, blind to the MFQ responses, wio completed the DISC with them, The psychiatric sample was also seen by an experienced psychologist or social worker, and by a child Psychiatrist. However, no structured diagnostic Interview was performed and the well-known unceliability of clinical diagnoses (see, eg. Cantwell, 1988) precluded comparisons of MFO seores between different diagnostic yroups within this sample.Tuble + Results trom principal components analysis of the parent and child initial item pools Them loadings om fist component MFO.C MEQ Miserable or unhappy oss on Did’: enjoy anything 046 048 eae hunaey 040 o2 Ate more ous om Tiree os oa Restless 064 on No goed on on Blamed self 099 os Incisive 030 034 Irsable oa 036 “Talking lest 039 03 Cred 2 lot 046 0.65 No good in future ost 0.82 [Not wort ving on 056 ‘Thoughts of death ose 013 Becer off without on os Suicidal thoughes 0% 07 [Not see frends 034 028 Poot concentration og 08 Bad things happen 08 034 Hated nyse 082 O78 ‘Bad person 0s ~ Looked ugly 040 06 Lonely ost on Usloved 08 on [Never be as good 0.67 os Did everything wrong. 075 4 Poor sleep ost oat Slept more 030 0.06 Not cheered up 074 for these items are therefore excluded from all analyses of the parent data. Internal consistency of the MFQ item pool As we expected, initial item pools for both Parents and children had high internal reliabilities (Cronback’s alpha~0.90 for both), suggesting 00d deal of redundancy in the measurement of the latent depression “variable. This finding encouraged our search for the subgroup of items that would be most closely linked to the depression variable measured by the scale, so that we could reduce redundancy and increase the criterion- ‘elated performance of the final SMFQ. AANGOLO ET ie divedual stems us predictors uf total MEQ trem pool score The most useful items in a homogeneous scale are those that are strongly associated with elevations in the total score. We compared the total initial item pool scores of those who reported a particular item present with the total initial item ool scores of those who reported that item absent. The total initial item pool score was adjusted to take account of the contribution of, the item under consideration by subtracting each subject's score on that item ftom his/her total initial item pool score. Tables 2 and 3 show the mean total MFQ score for those whose reports contained a0, | or 2 on each item for parent and shild reports respectively (where the n values in columns 0, | and 2 do not sum to 125, data for that item were missing for one or more cases). Since the initial item pool scores were far from being normally distributed (all had ‘reverse J" shaped distributions), the parameter estimates and significance values shown in Tables 2 and 3 represent the findings from maximum likelihood logistic regressions for ordinal response data of, total MFQ score on item score. ‘The first point to note is that, overall, the rates of positively reported symptoms were much higher for the child self-reports than for the parent reports. In particular, parents rarely used codings of 2. Secondly, nearly all of the items were substantially associated with the adjusted total score in both the parent and child data. Another way of looking at the relationships between individual items and the total scale scores is to compute item-total correlations, with the contribution of the item under consideration removed from the total. Pearson correlation coefficients are also shown in Tables 2 and 3. Factor analysis The methods of assessing the relationships between individual items and the scale’s overall measure of depression discussed so far have used the total score minus the score on the item under consideration as the overall measure. However, it may be more appropriate to look at individual item loadings on the latent variable (or variables) measured by the scale as 2 whole using factor analysis, since this will exclude a certain ammount of error variance that is included ia the totalMe Tuble 7 Initial tem pook items as predictors of OISC depression diagnosis Seerepors Parent ceperts ont FOR Ne Miserabe or 57 0m 32 tt uskappy Didn't enioy 22 0m 66 Oar anything Less hungry 12 0s98 100966 Ate more Wome 17 o3a7 Tired 47 oon 25 os Restless 25 aor 218s No good 49 000343020 Blamed seit 1S oz 34 ots Indesisive 17 0291s aes mtabe 230 00714 ats Talking lest 21 oom 91 <0001 Coed alot 11 ass 33 ons Nogood in fuure 25008025 0.000 Notworth ving 28-0065 Thoughts ofdeath = 41 gost Bewer off without 3.10006 8.4 0.007 Suicidal thoughts = 22022 = 24 Not see ends 32 om 21.836 Poor concentration 27 00232223 Bad things happen 32 000s tw Hated myselt 30° 002690003 Bad person 36 one e Looked ugly 17 0 38 ons Lonely 1s 0385 26 089 Unloved 41 <0on 520.009 Neverbeasgood = 3.1001, 3.1 <0.01 Did everything 37 0m - = wrong Poor sleep 30 oms 29 ss Slept more 12y wiagys we [Not cheered up 40 oes "Solution would aot converge. criteria were available to us for this purpose: clinical status (pediatric versus psychiatric) and depression status as measured by the DISC in the pediatric group. Comparisons of individual item frequencies by study group. Table 5 shows the odds ratios (OR) and significance levels from maximum likelihood logistic regressions of individual item scores on group status (pediatric versus psychiatric). An a ANGOLD er 1c odds ratio of 2 indicates that for each increment of | iffom a score of D to | ur trom a score of | to 2) the odds of being in the psychiatric group doubled, ‘The parent items proved to be far bewer predictors of group status than the child self-ceport items, which is perhaps not surprising, given that parents are usually responsible for bringing their children for treatment. However, several of the child self Teport items were associated with substantial increases in the odds of being in the psychiatric group. In order to determine whether the poor item-by- item prediction of clinical status was 2 particular feature of the MFQ-C or a more general feature of self-reports of depressive symptoms, we conducted a similar analysis on the items of the CDI (Table 6). The pattern of results proved to be very similar for the CDI and initial item pool, Prediction of DISC depression status in the pediatric group. The following diagnoses gener- ated by the DISC were pooled as ‘depression diagnoses’: dysthymia; major depression; manic- depressive disorder; cyclothymia. There were six Parent reports of possible or probable depres- sion diagnoses from the 125 subjects in the pediatric sample and five possible or probable Giagnoses from the child reports, in 10 children altogether. To determine the ability of individual items to predict DISC depression status we conducted a further series of logistic regressions of item scores on DISC depression status. The results of these analyses are shown in Tabie 7. In this case, the child self-reports proved to be better predictors of depression status than the parent reports, even though both parent and child DISC results were included in the determination of depression status. ‘Selection of items for the SMFQ The foregoing analyses were used as the basis serve as the 3¢ affective and cognitive items in the MFQ item Pool, but it also included tiredness, restlessness and poor concentration. ‘The remainder of this paper deals with the Psychometric properties of the SMFQ. We wanted246 4 Fabie 10. Discrimination clinical group ANGOLD EF 4 OR ? Peduaine mean (SD) Prychiauie mean (SD) SMFS ‘103 0.008 $68 4.55) 1146.19 SuFQ? is <0.001 LIsa2 579 (5.80) cor Los 0.026 558657) 814697) Table 11. Discrimination of DISC depression status in pediatric group Not depresed on DISC M(SD) __Depretsed on DISC M(SD) oR ? SMEQ-C 417 29) 10.30 (5.56) 126 <0001 SMFG-P 132.56) 580 (636) 136 0.003, cor 486 (533) 13.80 23) Lie <0.001 Disc 645 4676), 1am 792) 125 <0.001, Disc 593 3.66) 1530 (1042) 130 <0.001 those in the depressed group with the 29 other selection. A sensitivity of 70% and specificity of individuals who had any other diagnosis but not depression. Although the sample sizes are so smaall, the MFQ-C still significantly discriminated (OR=1.15, p=0.049), but the MFQ-P did not (OR=1.09, p=0,30). Thus we have some evidence for the child self-reports that the MFQ is sensitive to depressive symptomotology specifically, rather than just disturbance in general. How useful are the SMFQ scale scores as Predictors of depression? Receiver operating characteristic (ROC) curves provide a means of examining the efficiency of a screening instrument at different cut-off levels. Fig. | shows ROC curves for the parent and child initial item pools, SMFQ-C and SMFQP, with DISC depression diagnosis as the criterion. Combined scores on the SMFQ-C and SMFQ-P (ie, the SMFQ.C score plus the SMFQ-P score), DISC depression score and CDI score are also plotted against DISC diagnosis for comparison. Overall, “the child self-reports discriminated depression status better than parent reports, but the combination of parent and child reports from the SMFQ did better than either alone. It can be seen that the combined SMFQ score provided a Substantial improvement over random sample 85% resulted from a cut-off score of 12 or more on the combined scale. However, the 13 items of the SMFQ-C alone achieved 60% sensitivity and 85% specificity at a cut-off score of 8 or more. This translates into a positive predictive power of 80% and a negative predictive value of 68%, DISCUSSION ur results suggest iS seale that taps an underlying construct of general depression similar to those measured by the CDI ‘and the DISC depression scale scores. It appears to be usable with children and adolescents from the age of six to 17, and, at least in 6-1! year-olds, hhas useful screening properties. [t can be administered in five minutes or less and can easily be scored on the spot by an interviewer, [tems addressing the affective and cognitive components of depression tended to be the best predictors of depressive status, and are, therefore, heavily represented in the SMFQ( should be noed that these results may overestimate the screening efficiency of the SMFQ. One of the criteria for the inclusion of items in the final scale was success in predicting ISC depression status u this sample, and so Turther testing in other sampies will be required to determine how much of the success ofthe seale is a result of capitalization on chance covariation. On the other hand, given the strength of the correlations between the SMFQ, CDI and DISC depression scale scores, it is unlikely that this represents the whole story. It is striking that, in predicting depression status, both the SMFQ and the CDI were almost as efficient as the DISC depression scores themselves, despite the fact that the DISC depression diagnoses were obviously based on the same data as the depression scores. A further caveat concerns the small numbers involved in the study. In particular, there were few depressed individuals, necessitating the inclusion of DISC ‘possible’ cases of depression. It would have been preferable to have been able to adapt a more restrictive definition of depression (using, for instance, only DISC ‘probable diagnoses’), On the other hand, the finding of significant predictive effects and significant discrimination between those with depression diagnoses and those with other diagnoses with such small numbers is indicative of a large Predictive effect size at least for child self-reports. These results also suggest that a still shorter scale ‘might well be used with little los of screening efficiency, ‘but further testing ofthis idea awaits the availabilty of other data sets. In particular, comparisons of these results with those of studies using different criterion diagnostic instruments are needed. This study did not include a test-retest stability ‘component, but data from another study (Costelloet al, 1991) found I-week stability in an in-patient adolescent psychiatric population to be quite high Gntraclass correlation=0.75) considering that the fume frame of the questionnaire is only 2 weeks. However, as yet, no SMFQ test-retest stability data from a general population sample are available. Overall, we conclude thatthe SMFQ shows promise asa shor, easly administered screening questionnaire {or general population studies of depression. However, ACKNOWLEDGEMENT This work was supporied by Contract 278-83- 0006(DB) from the National fasticute of Mental A-ANGOLD Er 4c Health, by the Medical Research Council and by the Leon Lowenstein Foundation. The authors wish to thank the patients and staf? of, HealthAmerica Inc. the Western Psychiatrie Institute and Clinic. and. in particular, A. J Costello MD. for help with this study REFERENCES American Psychiatrie Association (1980) Diagnostic ‘and Siausiical Manual of Mental Disorders, 3rd ed. American Psychiatric Association, Washington, Dc. ‘Anderson, J. C., Williams, S., McGee, R. and Silva, P. A, (1987) DSMAIIT disorders in preadolescent children: prevalence ina large sample from the general population 4rch. Gen, Psychiat. 44, 69-77 ‘Angola, A. (1988) Childhood and adolescent depression epidemiological and aetiological aspects. Br. J Paychiat. 152, 601-617. Beck, A. T. and Beamesderfer, A. (1974) Assessment of depression: the depression inventory. In Psychological ‘Measuremenis in Psychopharmacology. Modern Problems in Pharmacopsychiatry, Tth ed. (P. Pichot, ed.) pp. 151-169. Karger and Basel, Paris, Boyd, J. H. and Weissman, M. M. (1981) Epidemiology of affective disorders. a re-examination and future directions. Arch, Gen. Psychiat. 38, 1039-1086. Boyle, G. J. (1985) Self-report measures of depression: some psychometric considerations. Br. J. Clin. Paychol. 24, 45-39, ‘Cantwell, D. P. (1988) DSM-IM studies. In Diagnosis in Child Pochopathology (M. Rutter, A. Hussain-Tuma and J. S. Lan, eds), pp. 46-52. Tae Guilford Press, New York. Costello, E. J. and Angold, A. (1988) Scales to assess child and adolescent depression: checklist, screens and nets. J. Am. Acad. Child Adolescent Peychiat. 21, 726-737, Costello, A.J, Edelbrock, C., Kalas, R., Kessler, M. D. and Klarie, S. H. (1982) The National Jnctitute of Mental Health Diagnostic Interview Schedule for Children (DISC). National Institute of Meatal Health: Rockcille, MD. Costco. £.., Costello, A ., Edelbrock, C., Burns, B. J. Dulean, MK, Brent, D. and Janiszewski, S. V. (1988) Payehistnic disorders in pediatric and primary care: prevalence and risk factors arch. Gen. Psychiat. 5, Tortie. Costello, EJ. Benjamin, R., Angold, A. and Silver, D. (1991) Moed variability in adolescents: a study of ‘depressed, nondepressed and comorbid. patients. J Affective Disord, 23, 199-212. Cronbach, L. J. (1970) The two disciplines of seientific avchology. dm. Peychol, 671-684Settreport version SHORT MOOD AND FEELINGS QUESTIONNAIRE This form is about how you might have been feeling or acting recently. For each question, please check how much you have felt or acted this way in the past two weeks. If a sentence was true about you most of the time, check TRUE. If it was only sometimes true, check SOMETIMES. If a sentence was not true about you, check NOT TRUE. 1. [felt miserable or unhappy ..........-20e0eeee eee Deoet sl atall . 3. [felt soltired | just sat around and did nothing ....... Ab alias Very MMM coca cals atttttnihentie nals Se a 10. felt lonely 11. I thought fobody really loved me. 12. I thought | could never be as good as other kids) EA TRUE SOME (Nor TIMES TRUE {Copyright Adrian Angold & Elizabeth |. Costello, 1987; Developmental Epidemiology rogram; Duke UnivesityParentreport version SHORT MOOD AND FEELINGS QUESTIONNAIRE This form is about how your child may have been feeling or acting recently. For each question, please check how much she or he has felt or acted this way in the past two weeks. If a sentence was true about your child most of the time, check TRUE. If it was only sometimes true, check SOMETIMES. If a sentence was not true about your child, check NOT TRUE. TRUE SOME NOT TIMES TRUE 1. Shhe felt miserable or unhappy .......--220ee22005 2. Shhe didn't enjoy anything at all 3. Sihe felt so tired that s/he just sat around and did nothing. 4. Sthe was very restless 5. Sihe felt s/he was no good any more... .....--+ i Ge Ste ctied alot ifort [ 7. She found it hard to think properly or concentrate... . 8. She hated him/herself ......00.c00ceeeeeeeeeees 9. She felt s/he was a bad person pe) 10. She felt lonely ..... Pivie pieent tee ee eee 11. S/he thought nobody really loved hinvher ........ 12. Sihe thought s/he could never be as good as other kids . 13. Sihe felt s/he did everything wrong .......... 4.05 Let Lt Copyright Acian Angeld & Elizabeth J. Costello, 1987; Developmental Epidemiology Program: Duke University