ebook img

Use of risk assessment instruments to predict violence and antisocial behaviour in 73 samples ... PDF

12 Pages·2012·0.42 MB·English
by  
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview Use of risk assessment instruments to predict violence and antisocial behaviour in 73 samples ...

BMJ2012;345:e4692doi:10.1136/bmj.e4692(Published24July2012) Page1of12 Research B M RESEARCH J : firs t p u b lis h e d a Use of risk assessment instruments to predict violence s 1 0 and antisocial behaviour in 73 samples involving 24 .11 3 6 827 people: systematic review and meta-analysis /b m j.e 4 OPENACCESS 6 9 2 o n Seena Fazel Wellcome Trust senior research fellow in clinical science1, Jay P Singh postdoctoral 24 J research fellow2, Helen Doll statistician3, Martin Grann professor4 uly 2 0 1DepartmentofPsychiatry,UniversityofOxford,WarnefordHospital,OxfordOX37JX,UK;2DepartmentofMentalHealthLawandPolicy,University 12 ofSouthFlorida,Tampa,FL,USA;3DepartmentofPopulationHealthandPrimaryCare,UniversityofEastAnglia,Norwich,UK;4CentreforViolence . D Prevention,KarolinskaInstitutet,Stockholm,Sweden ow n lo a d Abstract ConclusionsAlthoughriskassessmenttoolsarewidelyusedinclinical ed ObjectiveToinvestigatethepredictivevalidityoftoolscommonlyused andcriminaljusticesettings,theirpredictiveaccuracyvariesdepending fro m toassesstheriskofviolence,sexual,andcriminalbehaviour. onhowtheyareused.Theyseemtoidentifylowriskindividualswith h DesignSystematicreviewandtabularmeta-analysisofreplication highlevelsofaccuracy,buttheiruseassoledeterminantsofdetention, ttp studiesfollowingPRISMAguidelines. sentencing,andreleaseisnotsupportedbythecurrentevidence.Further ://w researchisneededtoexaminetheircontributiontotreatmentand w DatasourcesPsycINFO,Embase,Medline,andUnitedStatesCriminal management. w JusticeReferenceServiceAbstracts. .b m Introduction ReviewmethodsWeincludedreplicationstudiesfrom1January1995 j.c o to1January2011iftheyprovidedcontingencydatafortheoffending Withtheincreasingrecognitionofthepublichealthimportance m outcomethatthetoolsweredesignedtopredict.Wecalculatedthe ofviolence,12thepredictionofviolence,orviolencerisk o/ n diagnosticoddsratio,sensitivity,specificity,areaunderthecurve,positive assessment,hasbeenthesubjectofconsiderableclinicaland 1 predictivevalue,negativepredictivevalue,thenumberneededtodetain researchinterest.Sincethelate1980s,suchassessmenthas 6 M topreventoneoffence,aswellasanovelperformanceindicator—the mostlybeenconductedbystructuredinstrumentsafterseveral a numbersafelydischarged.Weinvestigatedpotentialsourcesof studiesfoundunstructuredclinicalopiniontohavelittleevidence rch heterogeneityusingmetaregressionandsubgroupanalyses. insupport.3Recentsurveyshaveestimatedthatover60%of 20 2 ResultsRiskassessmentswereconductedon73samplescomprising generalpsychiatricpatientsareroutinelyassessedforviolence 3 24847participantsfrom13countries,ofwhom5879(23.7%)offended risk,4risingtoabove80%inforensicpsychiatrichospitals.5 by overanaverageof49.6months.Whenusedtopredictviolentoffending, Thewidespreaduseofthesetoolshasbeenpartlydrivenby gu riskassessmenttoolsproducedlowtomoderatepositivepredictive publicconcernaboutthesafetyofmentallyillpatients,6research es values(median41%,interquartilerange27-60%)andhighernegative evidencethatseverementalillnessisassociatedwithviolence,7-9 t. P predictivevalues(91%,81-95%),andacorrespondingmediannumber andclinicalpracticeguidelinesinsomecountries,includingthe ro neededtodetainof2(2-4)andnumbersafelydischargedof10(4-18). UnitedKingdomandUnitedStates,1011recommendingviolence tec Instrumentsdesignedtopredictviolentoffendingperformedbetterthan riskassessmentforallpatientswithschizophrenia.Furthermore, ted thoseaimedatpredictingsexualorgeneralcrime. b y c o p y Correspondence to: S Fazel [email protected] rig h Extra material supplied by the author (see http://www.bmj.com/content/345/bmj.e4692?tab=related#webextra) t. Web figure 1: Results of a systematic search conducted to identify replication studies of nine commonly used risk assessment tools Web figure 2: Distribution of effect sizes (reported as Cohen’s d) in studies included and not included in the current systematic review Web appendix: Studies included in quantitative synthesis Web table 1: Comparison of summary accuracy estimates produced by three types of risk assessment tools when moderate risk individuals were classified as low risk Web table 2: Metaregression analyses within three classes of risk assessment tools Web figure 3:Summary receiver operator characteristic curve from bivariate analysis of individual violence risk assessment tools Web figure 4:Summary receiver operator characteristic curve from bivariate analysis of individual sexual risk assessment tools Web figure 5:Summary receiver operator characteristic curve from bivariate analysis of individual criminal risk assessment tools No commercial reuse: See rights and reprints http://www.bmj.com/permissions Subscribe:http://www.bmj.com/subscribe BMJ2012;345:e4692doi:10.1136/bmj.e4692(Published24July2012) Page2of12 RESEARCH criminaljusticesystemsinmanycountrieshavewelcomedthe antisocialoutcomes(basedonclinicalrecordsandother useofriskassessmenttoassistsentencingandreleasedecisions. measures)eveniftheydidnotleadtoconvictions.Forthesake Riskassessmenthasbeenusedtoinformindeterminate ofconsistency,however,werefertoalloutcomesasoffences. sentencingintheUK,12andhasbecomealargelyuncontested B M partofanexpandedcriminaljusticeprocessintheUS.13 Systematic search J Furthermore,a2004surveyreportedthatofthe32USstates Asystematicsearchwasconductedtoidentifystudiesthat : firs whereparoleisanoption,23hadusedsuchinstrumentsaspart measuredthepredictivevalidityoftheninetools.Wesearched t p ofthesedecisions.14 u thefollowingdatabasesbetween1January1995and1January b Thecurrentgroupofriskassessmenttoolseitherprovidea 2011usingacronymsandfullnamesoftheinstrumentsas lish probabilisticestimateofviolenceriskinaspecifiedtimeperiod keywords:PsycINFO,Embase,Medline,andUSNational ed (actuarialinstruments),orallowforaprofessionaljudgmentto CriminalJusticeReferenceServiceAbstracts.Additionalstudies as bemadeonrisklevel(forexample,low,moderate,orhigh) wereidentifiedthroughreferences,annotatedbibliographies, 10 aftertakingintoaccountthepresenceorabsenceofa andcorrespondencewithresearchersinthefield.Studiesinall .1 1 predeterminedsetoffactors(structuredclinicaljudgment languagesandunpublishedinvestigationswereconsideredfor 3 6 instruments).Over150ofthesestructuredmeasurescurrently inclusion.Weexcludedstudiesiftheymeasuredthepredictive /b einxcisotm,15eacnoduanrteriessta.1r6ti1n7gtobeimplementedinlowandmiddle vreatlriodsiptyecotfivseelleycwtsicthaoleustobflianmdienagsutoreo,uiftcionmstreusm,oernitfstwheeryewcoedreed mj.e4 6 However,thesetoolsaretimeconsumingandresourceintensive, calibrationstudiesfortheactuarialtools(whichmaygive 92 typicallytakingmanyhourstocompletebyamultidisciplinary inflatedeffects).48Whenstudiesusedoverlappingsamples,we o n groupofprofessionals.18Theycanalsobeexpensive;training usedthesamplewiththelargestnumberofparticipantstoavoid 2 4 isrequiredformosttools,andpaymentisoftenneededfor double-counting.Usingthissearchstrategy,weidentified J u individualuse.Further,andmoreimportantly,theinstruments’ 251validationstudies(webfigure1). ly predictiveaccuracyremainsasourceofconsiderableuncertainty, Tobeincludedinthemeta-analysis,studieshadtoreportrates 20 withsomereviewsrecommendingtheiruseinclinicaland oftruepositives,falsepositives,truenegatives,andfalse 12 correctionalsettingsandothersfindingthattheyleadtoan negativesatagivencut-offscorefortheoutcomewhichthe . D unacceptablyhighnumberoffalsepositivedecisions.18-22Expert instrumentwasdesignedtopredict.Apilotstudyshowedthat ow opinionisequallydivided.23-25 differentscorethresholdswereusedtoclassifypeopleasbeing nlo a Wehavethereforeconductedasystematicreviewand atlow,moderate,orhighriskoffutureoffending.Wecontacted d e meta-analysisofthepredictiveaccuracyofthemostcommonly studyauthorsandaskedthemtocompleteastandardisedform d usedriskassessmentinstruments.Toconsistentlyreport iftabulardatausingthecut-offscoresrecommendedinthemost fro m outcomesforindividualstudies,werequestedtabulardatafrom recentversionofaninstrument’smanualwereunavailable,or h primaryauthors.Wehavesynthesisedthesedataacrossarange ifthenumberofparticipantsclassifiedaslow,moderate,orhigh ttp ofaccuracyestimates,oneofwhichwasdevelopedforthe riskwasmissingfromastudyofastructuredclinicaljudgment ://w purposesofthisreview. tool.Forpublicationsinwhichmultipletoolsdesignedtopredict w thesameoutcomeweretestedonthesamesample(eight w .b Methods studies),werequestedtabulardataforalloutcomesbutonly m includedthoseforthetoolwiththefewestreplicationstudies j.c Review protocol toincreasethebreadthofthefindings.Thisprocedureprobably om Wefollowedthepreferredreportingitemsforsystematicreviews didnotbiasresults,sinceχ2testsofdifferencesinproportions o/ n andmeta-analysesstatement.26 foundnodifferencesinratesoftrueandfalsepositivesandtrue 1 andfalsenegativesinthetabulardataobtainedforincludedand 6 M Risk assessment tools excludedtoolsfromthesamestudywiththesameoutcome. a rc Standardisedoutcomedatawereavailableinthemanuscripts h Weidentifiedtheninemostcommonlyusedtoolsrisk 2 assessmentusingrecentreviews27-29andquestionnaire of30eligiblestudies(32samples).Werequestedadditional 02 Isnuvrveenytosr.y30-3R1eAvcistueadr(iLalSiIn-sRtr)u,3m2thenetPssiyncchluodpeadthtyheCLheecvkellisotf-RSeevrvisiecde d5saa2mtasptfulredosimefsotr(h6we2ha)ui.ctAhhocwcrseurowafce1yr7ee4usnstitamubdlaeiteetsos(of3rb3ot0ma)ina2nd3da5toaobfwttaehirnoeescedo2dn6avt8eartfeodr 3 by g (PCL-R),3334theSexOffenderRiskAppraisalGuide u (GSuOidReA(GV)R,3A53G6t)h.3e5S36tSattircu-c9t9u,r3e7d38calninditchaelVjuidoglemnecnetRtoisoklsAinpcplruadiseadl ptoroCdouhceend’bsydtuhsoisnegssatmanpdlaersdfomrewthhoidchs.4w9-5e1cTohueldmneodtiaonbtdaivnadluaeta est. P theHistorical,Clinical,Riskmanagement-20(HCR-20);3940the (904.6o7b,taininteerdqusaamrtiplelersan(0g.e740,.405.5t4o0to.807.)9w5)a(swsiembifliagrutroet2hasthoofwthse rote SexualViolenceRisk-20(SVR-20);41theSpousalAssaultRisk distributionofeffectsizes).Inaddition,theHodges-Lehmann cte Assessment(SARA);42-44andtheStructuredAssessmentof d ViolenceRiskinYouth(SAVRY).4546Wedividedtoolsinto percentiledifference,52themediandifferencebetweenall by possiblepairsofdvaluesfromthetwogroups,wassmall(0.01, c threecategories:thosedesignedtopredictviolentoffending o 95%confidenceinterval0.00to0.08).Finally,ofthe82studies p (HCR-20,SARA,SAVRY,andVRAG),sexualoffending y (SORAG,Static-99,andSVR-20),andanycriminaloffending forwhichtabulardatawasobtained,wewereabletoinclude rig (LSI-RandPCL-R).AlthoughthePCL-Rwasoriginally informationfrom68(73samples;referencesavailableinweb ht. appendix),sincetheother14studiesusedinstrumentstopredict developedtodiagnosepsychopathicpersonalitydisorder,ithas outcomesotherthanthoseforwhichtheyweredesigned. becomewidelyusedforriskassessmentpurposes,asnumerous studieshavefoundthePCL-Rscoretobestatistically Data analysis significantlyassociatedwithcriminalandantisocialoutcomes.47 Table1⇓reportsadditionaldetailsofalltheinstruments. WefollowedthecurrentguidanceprovidedbytheCochrane Althoughtheseinstrumentsweremostlydesignedtopredictthe collaborationforsystematicreviewsofdiagnostictest likelihoodofoffending,weincludedviolent,sexual,and accuracy.53Thestatisticalmethodsforsuchreviewsfocuson No commercial reuse: See rights and reprints http://www.bmj.com/permissions Subscribe:http://www.bmj.com/subscribe BMJ2012;345:e4692doi:10.1136/bmj.e4692(Published24July2012) Page3of12 RESEARCH twostatisticalmeasuresofdiagnosticaccuracyofthetest: allowsresearcherstoquantifytheimplicationsofrelyingona sensitivity(theproportionofoffenderswhoariskassessment riskinstrumenttomakedischargeorreleasedecisions. toolpredictedtooffend)andspecificity(theproportionof non-offenderswhoariskassessmenttoolpredictedwouldnot Tests of assumptions B M offend).Theaimoftheanalysiswastoquantifyandcompare J thesestatisticsaswellastheerrorrates(falsepositiveandfalse Sditaagnndoarsdticmoedtad-sarnaatliyotsicanpdooblointhgsaesnsusimtivpittiyonasnwdespreecmifeitciftoyr.5960 : firs negativediagnoses)foreachtypeoftest.Therequiredanalysis Sincetherewasasignificantcorrelationbetweenthesensitivities t p isabivariateanalysisofsensitivityandspecificityforeach u andspecificitiesproducedbythesamplesineachclassofrisk b studyaccountingforcorrelationbetweensensitivitiesand lis assessmenttools,poolingassumptionsforareasunderthecurve h specificities.54Theresultingmodelwithoutcovariatesisa e werenotmet.60Inaddition,becausethemedianbaserateof d differentparameterisationofthehierarchicalsummaryreceiver a operatingcharacteristicmodel.55Weusedsummaryreceiver o(vfifoelnednicneg3w2i.0th%in,ienatcehrqculaarstsiloefratonogles2v2a.r2i-e4d6c.6o%ns;idseexraubally16.9%, s 10 operatorcharacteristicplotstodisplaytheresultsofeachstudy 7.4-28.2%;criminal28.4%,20.7-46.0%),baseratedependent .1 inreceiveroperatingcharacteristicspace,plottingeachstudy 1 statisticswerenotpooled(suchasthepositiveandnegative 3 plottedasasinglesensitivity-specificitypoint.Parameter predictivevaluesandboththenumberneededtodetainandthe 6/b estimatesfromthebivariatemodelproducedasummaryreceiver m operatingcharacteristiccurvewithasummaryoperatingpoint numbersafelydischarged),andmedianswithinterquartileranges j.e werecalculated. 4 (thatis,summaryvaluesforsensitivityandspecificity),95% 6 9 confidenceregion,and95%predictionregion.Weusedthe Investigation of heterogeneity 2 o summarypointfromeachcurvetocalculatethesummary n diagnosticoddsratioandboththesensitivityandspecificity, ThestandardQandI2statistics61donotaccountfor 24 eachwith95%confidenceintervals. heterogeneityexplainedbyphenomenasuchaspositivity Ju thresholdeffects,andthenumericalestimatesoftherandom ly Sincebinarytestoutcomesaredefinedonthebasisofacut-off 2 effecttermsinthebivariateregressionarenoteasilyinterpreted. 0 valuefortestpositivity,wechosethesevaluesapriori.Risk 1 Therefore,themagnitudeofobservedheterogeneityin 2 aasssienssstmruemnetntotsoflosrairdeepnrteifdyoinmginhaignhtleyrurissekdinindicvliindiucaalls,s1i9ttuhautiso,nwse meta-analysesofdiagnosticaccuracyisinsteadbestdetermined . Do bythescatterofpointsinthesummaryreceiveroperating w combinedparticipantswhowereclassifiedasbeingatmoderate n characteristicplotandfromthepredictionellipse.53Inparticular, lo orhighriskforfutureoffendingandcomparedthemwiththose a thepredictionregiondepictsaregionwithinwhich,assuming d classifiedaslowrisk.Wedidsecondaryanalysesbycomparing e themodeliscorrect,wehave95%confidencethatthetrue d participantsclassifiedashighriskwiththoseclassifiedaslow sensitivityandspecificityofafuturestudyshouldlie.53 fro ormoderaterisk,anapproachconsistentwithscreening,and m alsobycompletelyexcludingthoseclassifiedasmoderaterisk. Sincethediagnosticoddsratiosmetpoolingassumptions,we h usedrandomeffectsmetaregressiontoinvestigatesourcesof ttp Accuracy estimates heterogeneitybetweenstudiesinsamplediagnosticoddsratios ://w foreachclassoftools.Metaregressioninvestigatestherelation w Weusedarangeofaccuracyestimatestoreportonthepredictive w betweenaccuracyestimatesanddichotomousorcontinuous .b validityoftheriskassessmenttools.Firstly,thesummary sampleorstudycharacteristics.62Weformallyexploredthe m operatingpointwasusedtoestimatethesummarydiagnostic moderatingroleofthefollowingvariables:sex(proportionof j.co oddsratioandbothsensitivityandspecificity.Weobtained m samplethatwasmale),ethnicity(proportionofsamplethatwas estimatesfortheareaunderthecurve,positivepredictivevalue, white),meanparticipantage,typeofinstrument(actuarialv on/ negativepredictivevalue,numberneededtodetain,andnumber structuredclinicaljudgment),temporaldesign(prospectivev 1 safelydischargedfromtheindividualsampleestimates. retrospective),settinginwhichassessmentwasconducted 6 M Thediagnosticoddsratioistheratiooftheoddsofapositive (correctional,forensicpsychiatric,generalpsychiatric,or arc testresultinanoffenderrelativetotheoddsofapositiveresult mixture),locationofoffendingoutcome(communityonlyv h 2 inanon-offender,andisrecommendedforusewithdiagnostic insideinstitutionorother),meanlengthoffollow-up(months), 0 instruments.56Theareaunderthecurveisanindexofsensitivity samplesize,andpublicationstatus(publishedinpeerreviewed 23 andspecificityacrossscorethresholds,andiscurrently journalvnot).Wealsoconductedsubgroupanalysesusingthe by consideredtheaccuracyestimateofchoiceinviolencerisk bivariatemodelsonthesevariables.Detailedexaminationof gu assessmentwhenmeasuringpredictiveaccuracy.57Neitherthe theoveralldifferencesbetweenindividualinstrumentshave es diagnosticoddsrationortheareaunderthecurveareaffected beenreportedinasubsetofthesamples.63Wedidallanalyses t. P bybaseratesofoffending.Thepositivepredictivevalueisthe inStata10.264usingthemetandi(forbivariatemodel ro proportionofparticipantsclassifiedasatriskwhogoonto meta-analysis),metan(randomeffectsmeta-analysis),and tec offend,whereasthenegativepredictivevaluereferstothe metareg(metaregression)commands. ted proportionofthoseclassifiedasnotatriskwhodonotgoonto b y offend.Thenumberneededtodetainisthenumberofpeople Results c o judgedtobeatriskwhowouldneedtobedetainedtoprevent p y oneincidentofsubsequentviolence.1958Thisoutcomeallows Descriptive characteristics rig somequantificationoftheimplicationsofusingriskassessment Wecollectedinformationfor24847participantsin73samples ht. toolstomakedetentiondecisions.Finally,thenumbersafely from68independentstudies(table2⇓).Standardisedoutcome dischargedisanewperformancestatisticthatwedevelopedfor informationfrom43ofthesamples(14798(59.6%) thepurposesofthisreview.Thisaccuracyestimatecalculates participants)wasnotreportedinmanuscriptsandobtained thenumberofparticipantsjudgedtobeatlowriskwhocould directlyfromstudyauthors.Of24847participants,5879 bedischargedintothecommunitybeforeasingleactofviolence (23.7%)offendedoveranaverageof49.6months(standard occurs(1÷[1−negativepredictivevalue]−1).Acomplementto deviation40.5).Studieswereconductedin13countries:Austria, thenumberneededtodetain,thenumbersafelydischarged, No commercial reuse: See rights and reprints http://www.bmj.com/permissions Subscribe:http://www.bmj.com/subscribe BMJ2012;345:e4692doi:10.1136/bmj.e4692(Published24July2012) Page4of12 RESEARCH Belgium,Canada,Denmark,Finland,Germany,theNetherlands, Discussion NewZealand,Serbia,Spain,Sweden,theUK,andtheUS. Thissystematicreviewandmeta-analysisexaminedthe Predictive validity predictivevalidityofviolenceriskassessmenttoolsfrom73 B M samplesinvolving24847individualsin13countries.Our J Wdseexepuefanoldu,inondrgadoninfyftechrreeinmtyciepnsealio)n.feOrsisvtikemraaasltsle,essbsoamfsepednretodinnicsdttiirvaugemnaoecnscttuic(rvaoicodyldesnce, ppriresirknfocairspmsaealsnsficmnedeoninft.gthIwfesuaessmetdheataotstuihnreefrsoerdmwepatesrnehadetimtnegreonogtneatnnhedeitpmyuairnnpaothgseeemofetnhte : first pu ratios,violenceriskassessmenttoolsperformedbest,andhad decisions,thentheseinstrumentsperformedmoderatelywell blis higherpositivepredictivevaluesthantoolsaimedatpredicting inidentifyingthoseindividualsathigherriskofviolenceand h e sexualoffending.Riskassessmentinstrumentsforviolenceand otherformsofoffending.However,ifusedassoledeterminants d a sexualoffendingproducedhighsensitivitiesandnegative ofsentencing,andreleaseordischargedecisions,these s predictivevalues.Inaddition,riskassessmentinstrumentsfor instrumentsarelimitedbytheirpositivepredictivevalues:41% 10 generaloffendinghadlowerdiagnosticoddsratios,areasunder ofpeoplejudgedtobeatmoderateorhighriskbyviolencerisk .11 thecurve,sensitivities,andnegativepredictivevaluesandhigher assessmenttoolswentontoviolentlyoffend,23%ofthose 36 specificitiesandpositivepredictivevaluesthantheothertwo judgedtobeatmoderateorhighriskbysexualriskassessment /bm classesofinstrument(table3⇓,figs1-3⇓⇓⇓). toolswentontosexuallyoffend,and52%ofthosejudgedto j.e 4 Forassessmentinstrumentspredictingtheriskofviolent beatmoderateorhighriskbygenericriskassessmenttools 6 9 outcomes,thesummarydiagnosticoddsratiowas6.1(95% wentontocommitanyoffence.Insampleswithlowerbase 2 confidenceinterval4.6to8.1)withmoderatelevelsof ratesthanthosethatcontributedtothereview,suchasingeneral on heterogeneity(individualpointsmoderatelyscatteredinreceiver psychiatry,positivepredictivevalueswillprobablybeeven 2 4 operatingcharacteristicspace,fig1)andamedianareaunder lower.25However,negativepredictivevalueswerehigh,and J u thecurveof0.72(interquartilerange0.68-0.78;table3).Of suggestthatthesetoolscaneffectivelyscreenoutindividuals ly thoseindividualswhowentonviolentlyoffend,92%(95% atlowriskoffutureoffending.Whetherthecautiousoptimism13 20 1 confidenceinterval88%to94%)hadbeenclassifiedasbeing thatexpertshavedescribedinrelationtotheabilitytopredict 2 atmoderateorhighriskoffutureviolence(thatis,sensitivity). violenceseemsjustifiedwilldependontheusetowhichthese . D o Ofthoseparticipantswhodidnotgoontoviolentlyoffend, instrumentsareput. w n 36%(28%to44%)hadbeenjudgedtobeatlowrisk(thatis, lo specificity).Ofthosepredictedtoviolentlyoffend,41%did Comparisons with other medical tools ad e (interquartilerange27-60%;positivepredictivevalue),which d wasequivalenttoamediannumberneededtodetainoftwo Aconmymcoomnpmaerdisiocnalodfiathgensoestriicskanadsspersosgmneonsttiscctooroelsswpoitsheostsheevreral from (two-four).Ofthosewhowerepredictednottoviolentlyoffend, difficulties.Firstly,comparisonwithdiagnostictoolsismostly h 91%didnot(81-95%;negativepredictivevalue),equivalentto inappropriatebecauseriskassessmentinstrumentsattemptto ttp amediannumbersafelydischargedof10(fourto18). predictthelikelihoodofafutureoutcome,whereasdiagnostic ://w Similarfindingswereobtainedwhenindividualsjudgedtobe instrumentattempttodetectthepresenceofacurrentcondition. w w atmoderateriskweregroupedwiththosejudgedtobeatlow Secondly,althoughitmaybepossibletocompareperformance .b riskforthesecondaryanalyses,butwithconsiderablyhigher statisticsofthesetoolswiththoseestimating,forexample, m specificitiesandlowersensitivities(webtable1).When cardiovascularrisk,theimplicationsofpositivepredictivevalues j.co moderateriskindividualswereexcludedfromanalyses, needtobeconsideredinevaluatinganycomparisons.Violence m assessmenttoolsforviolenceriskproducedconsiderablylarger riskassessmentpotentiallyleadstodetentionofindividualsfor on/ summarydiagnosticoddsratios(16.8,10.8to26.3)and longerthannecessary,withitsrelatedeconomic,65social,66and 1 6 specificities(0.72,0.63to0.80). civilrightsconsequences.67Bycomparisonwithcommon M a medicalprognostictools,itispossibletoarguethatthe rc Investigation of heterogeneity predictiveaccuracyofviolenceriskassessmentneedstobe h 2 Sincewesawmoderatelevelsofheterogeneityforthe higherbecauseoftheseconsequences,whichextendbeyond 02 instrumentsassessingviolenceriskandhigherlevelsfor thepersontootherpeople.Ontheotherhand,itisprecisely 3 b instrumentsassessingsexualandgeneraloffendingrisk(scatter becauseoftheriskstootherpeoplethatlowpositivepredictive y g ofpointsfromthelinebeinggreaterandthepredictionellipses valuesmaynotbeasimportantastheabilityoftheseinstruments u e lbairvgaerri)a,tewmedoiddelmtoetdareetegrrmesisnieonanayndposussbigbrloeuepxapnlaanlyastieosnussfionrgtthhies tpoerpfroerdmicatntcheomseetahsautrea,rethneontuamtrbiesrk.saOfeulryidnitsrochdaurcgtieodn,coofualdnohvelepl st. P heterogeneity.Theseanalysesfoundnoevidencethatsex, quantifythisinfutureresearch. rote ethnicity,age,typeofinstrument,temporaldesign,assessment Despitethesecaveats,theareasunderthecurvefoundinthis cte setting,locationofoffendingoutcome,lengthoffollow-up, review(0.66to0.74)werenotdissimilartothosefoundin d b samplesize,orpublicationstatuswasassociatedwithdifferences studiesexaminingscoresfromthemostvalidatedcardiovascular y c inpredictivevalidity(webtable2).Inaddition,wehave riskschemeinpredictingcardiovasculardiseaseevents.Areas o p presentedsummaryreceiveroperatingcharacteristiccurvesfor underthecurvefromtheFraminghamscoringsystemrange y eachtypeofinstrument(webfigures3-5).Subtypesoftools from0.57to0.86,theSCOREfrom0.65to0.85,andQRISK rig h performedsimilarly,lyingwithinthe95%predictionregion, from0.76to0.79.68Manyofthesestudiesreportassociations t. withthepossibleexceptionoftheSAVRYthatproducedhigher betweenpredictedandobservedrisks,69whichmaybehelpful levelsofpredictiveaccuracythantheotherviolencerisk forfutureresearchinviolenceriskassessment.Finally,the assessmentinstruments. standardbywhichtheseinstrumentsarecomparedwilldiffer dependingontheirsetting.Inforensicpsychiatry,amore meaningfulcomparisonwillbewithunstructuredclinical judgment,andclinicaltrialsareneededtotestwhetherstructured riskassessmentreducesadverseoutcomes. No commercial reuse: See rights and reprints http://www.bmj.com/permissions Subscribe:http://www.bmj.com/subscribe BMJ2012;345:e4692doi:10.1136/bmj.e4692(Published24July2012) Page5of12 RESEARCH Clinical implications threetimesasmanystudiesasthe2001reviewandarecent meta-analysisthatonlycomparedheadtoheadinvestigations Oneimplicationofthesefindingsisthat,evenafter30yearsof oftooluse.75 development,theviewthatviolence,sexual,orcriminalrisk B canbepredictedinmostcasesisnotevidencebased.This M Strengths and limitations J mademssiangisetriastiiomnpsowrthaontmfoarythhaevgeeunnerraelalpiustbiclice,xmpeecdtiaat,ioannsdosformisek Thestrengthsofthecurrentreviewincludetheincorporation : firs predictionforclinicians.70Thisexpectationisnotashighin ofnewtabulardata,thereportingofmultipleaccuracyestimates, t p u othermedicalspecialties,inwhichtheexpectationthatthe andameta-analysisusingbivariatemodels.Wereceivednew b doctorwillidentifytheindividualpatientwhowillhavean tabulardatafor14798people(60%ofpeopleincludedinthe lish e adverseeventisnotaprimaryissuewhereaspsychiatry,inmany review),andhencehavereportedaconsiderableamountofnew d countriessuchastheUK,hasdevelopedacultureofinquiries.71 data.Finally,byusingarangeofaccuracyestimates,wehave as Asecondandrelatedimplicationisthatthesetoolsarenot attemptedtominimisebiasesthatmaybeassociatedwith 10 sufficientontheirownforthepurposesofriskassessment.In reportingonlyoneofthem. .11 3 somecriminaljusticesystems,experttestimonycommonlyuse Limitationsincludethatwesolelyexaminedthepredictive 6 scoresfromtheseinstrumentsinasimplisticwaytoestimate qualitiesoftheseriskassessmenttools,anddidnotaccountfor /bm anindividual’sriskofseriousrepeatoffending.67However,our theirpotentialroleininformingmanagementandreminding j.e reviewsuggeststhatriskassessmenttoolsintheircurrentform clinicianstoenquireaboutpotentiallyimportantprognosticand 46 canonlybeusedtoroughlyclassifyindividualsatthegroup modifiablefactors.76Inaddition,wefoundmoderatetohigh 92 level,andnottosafelydeterminecriminalprognosisinan levelsofheterogeneity.Heterogeneitywastobeexpected,in on individualcase.Thisapproachismostlyusedinforensic viewofthedifferenttypesofsamplesincludedintheprimary 2 4 psychiatryintheUKandotherwesterncountries,wherethey studies(fromprison,securehospitals,andgeneralpsychiatric J u formpartofawiderclinicalassessmentprocess.These hospitals)andoutcomesmeasured.7778Weexploredsourcesof ly instrumentsmayalsoassistindevelopingriskmanagement heterogeneityandfoundnocleartrends.Investigating 20 1 plansinselectedhighriskgroups,assuggestedbyrecentclinical heterogeneityindiagnosticoddsratiosmeantthatincidenceof 2 guidelinesinEnglandandWales.72Furthermore,theyare theoutcomewasaccountedfor.Onepossiblesourceof . D o preferabletounstructuredclinicaljudgmentowingtotheir heterogeneitywasthepotentialeffectsofinterventionaftera w n increasedtransparencyandreliability. riskassessment,particularlyinpeopledeemedhighrisk.We lo a Anotherimplicationisthatactuarialinstrumentsfocusingon compareddiagnosticoddsratiosbetweenprospectiveand d e historicalriskfactorsperformnobetterthantoolsbasedon retrospectivestudiesthatwouldbeexpected,tosomeextent,to d clinicaljudgment,afindingcontrarytosomeprevious measurethis,sincehighriskparticipantsidentifiedin fro m reviews.2173Finally,ourreviewsuggeststhattheseinstruments prospectivestudieswouldprobablyhavebeenenrolledin h shouldbeuseddifferently.Sincetheyhadhighernegative interventionsdesignedtoreduceviolencerisk.However,we ttp predictivevalues,onepotentialapproachwouldbetousethem foundnodifferencesinmetaregressionorsubgroupanalysis. ://w toscreenoutlowriskindividuals.Researchersandpolicy Nevertheless,clinicaltrialsareneededdirectlytotestthe w makerscouldusethenumbersafelydischargedtodetermine possibleeffectsofintervention.Althoughwetestedfor w.b thepotentialscreeninguseofanyparticulartool,althoughits publicationstatusandfoundnoclearpatterns,wecannotexclude m usecouldbelimitedforcliniciansdependingontheimmediate thepossibilitythatsuchbiascouldexistinthestudiesthatwe j.c o andserviceconsequencesoffalsepositives.Afurthercaveatis wereunabletoinclude.Registersofsuchinvestigationswould m thatspecificitieswerenothigh—therefore,althoughthedecision assistfuturereviews.79Inaddition,fewsamplesreportedon o/ n makercanbeconfidentthatapersonistrulylowriskifscreened womenand,thus,thisreviewwasunderpoweredtoexamine 1 out,whensomeonefailstobescreenedoutaslowrisk,doctors whetherpredictivevaliditywasdifferentfrommen. 6 M cannotbecertainthatthispersonisnotlowrisk.Inotherwords, a manyindividualsassessedasbeingatmoderateorhighrisk Wethankthefollowingstudyauthorsforprovidingtabulardataforthe rch couldbe,infact,lowrisk.Ultimately,however,whatconstitutes analyses:AprilBeckmann,SarahBeggs,SusanneBengtsonPedersen, 20 anappropriatebalancebetweentheethicalimplicationsof Klaus-PeterDahle,RebeccaDempster,MaireadDolan,KevinDouglas, 23 detainingpeoplebasedonthepredictiveabilityofthesetools ReinhardEher,JorgeFolino,MonicaGammelgård,RobertHare,Grant b y andtheneedforpublicprotectionwillprimarilybeapolitical Harris,LeslieHelmus,AndreasHill,HildaHo,CliveHollin,Christopher g u consideration. Kelly,DrewKingston,P.RandyKropp,MichaelLacy,CalvinLangton, e s HenryLodewijks,JanLooman,KarinArbachLucioni,JeremyMills, t. P Comparison with other studies CatrinMorrissey,ThierryPham,CharlotteRennie,MartinRettenberger, ro MarnieRice,MichaelSeto,DavidSimourd,GabrielleSjöstedt,Jennifer te Previousmeta-analysesonriskassessmenthavefocusedon c Skeem,RobertSnowden,CornelisStadtland,DavidThornton,Jodi te comparinginstrumentswithoneanother,ormeasuringhow d individualtoolsperformacrosssexesandethnicgroups.74A Viljoen,ViviennedeVogel,ZoeWalkington,andGlennWalters. by systematicreviewpublishedin2001examinedtheaccuracyof Contributors:SFdevisedandcoordinatedtheproject,assistedindata co violenceriskassessmentinhighriskgroups,19andwasbased acquisitionandinterpretation,anddraftedandrevisedthemanuscript. py on21studies.Itestimatedthatsixpeopleneededtobedetained JPSassistedindataacquisition,performedthestatisticalanalyses, rig topreventoneviolentoffence,comparedwithourcurrent assistedininterpretingresults,andassistedindraftingandrevisingthe ht. report.HDassistedinstatisticalanalysisandcriticallyrevisedthe review’sestimateoftwopeopleneedingdetention.This manuscriptforimportantintellectualcontent.MGassistedininterpreting differencewasdespitethemedianbaserateofviolencebeing resultsandcriticallyrevisingthemanuscriptforimportantintellectual similarinbothreviews(currentreview,32%(interquartilerange content.SFandJPShadfullaccesstoallofthedatainthestudyand 22-46%)v2001review,26%,15-41%).Unliketheprevious takeresponsibilityfortheintegrityofthedataandtheaccuracyofthe report,thepresentmeta-analysisfocusedonstructured dataanalysis,andwillactasguarantors. assessmentinstrumentsandincludedbothinstitutionaland communitysamples.Thecurrentreportreviewedmorethan Funding:SFisfundedbytheWellcomeTrust. No commercial reuse: See rights and reprints http://www.bmj.com/permissions Subscribe:http://www.bmj.com/subscribe BMJ2012;345:e4692doi:10.1136/bmj.e4692(Published24July2012) Page6of12 RESEARCH Whatisalreadyknownonthistopic Instrumentsbasedonstructuredriskassessmentpredictantisocialbehaviourmoreaccuratelythanthosebasedonunstructuredclinical judgment B M Morethan100suchtoolshavebeendevelopedandareincreasinglyusedinclinicalandcriminaljusticesettings J Considerableuncertaintyexistsabouthowthesetoolsshouldbeusedandforwhom : firs Whatthisstudyadds t p u b Thecurrentlevelofevidenceisnotsufficientlystrongfordefinitivedecisionsonsentencing,parole,andreleaseordischargetobemade lis solelyusingthesetools h e Thesetoolsappeartoidentifylowriskindividualswithhighlevelsofaccuracy,buthavelowtomoderatepositivepredictivevalues d a Theextenttowhichtheseinstrumentsimproveclinicaloutcomesandreducerepeatoffendingneedsfurtherresearch s 1 0 .1 Competinginterests:AllauthorshavecompletedtheUnifiedCompeting 29 KemshallH.Riskassessmentandmanagementofknownsexualandviolentoffenders: 13 areviewofcurrentissues.UKHomeOffice,2001. 6 Interestformatwww.icmje.org/coi_disclosure.pdf(availableonrequest 30 ArcherRP,Buffington-VollumJK,StrednyRV,HandelRW.Asurveyofpsychological /b fromthecorrespondingauthor)anddeclare:SFisfundedbythe testusepatternsamongforensicpsychologists.JPersAssess2006;87:84-94. m WellcomeTrust;nofinancialrelationshipswithanyorganisationsthat 31 LallySJ.Whattestsareacceptableforuseinforensicevaluations?:asurveyofexperts. j.e ProfPsycholResPract2003;34:491-8. 4 mighthaveaninterestinthesubmittedworkintheprevious3years; 32 AndrewsDA,BontaJ.LSI-R:thelevelofserviceinventory-revised.Multi-HealthSystems, 69 nootherrelationshipsoractivitiesthatcouldappeartohaveinfluenced 33 H19a9re5.RD.TheHarepsychopathychecklist-revised(PCL-R).Multi-HealthSystems,1991. 2 o thesubmittedwork. 34 HareRD.TheHarepsychopathychecklist-revised.2nded.Multi-HealthSystems,2003. n 2 Ethicalapproval:Noethicsapprovalwassoughtbecauseonlysecondary 35 Qrisukin.sAemyeVrLic,aHnaPrrsisycGhTo,loRgiciceaMlAEs,sCoocriamtiioernC,1A9.9V8io.lentoffenders:appraisingandmanaging 4 J datawereused. 36 QuinseyVL,HarrisGT,RiceME,CormierCA.Violentoffenders:appraisingandmanaging u Datasharing:Datasharing:Noadditionaldataavailable. 37 Hrisakr.ri2snAdJeRd,.PAhmeneirxicAa,nHPasnyscohnolRogKi,caTlhAosrnstoocniaDti.oSn,ta2t0ic0-69.9codingrules:revised2003. ly 20 SolicitorGeneralCanada,2003. 1 2 1 BrundtlandGH.Violenceprevention:apublichealthapproach.JAMA2002;288:1580. 38 HDaenpsaortnmReKnt,TofhothrentSonolDic.itSotraGtice-n9e9r:aImlopfrCovainnagdaac,tu1a9r9ia9l.riskassessmentsforsexoffenders. . D 2 K20ru0g2;E3G60,:M10e8rc3y-8J.A,DahlbergLL,ZwiAB.Theworldreportonviolenceandhealth.Lancet 39 WebsterCD,DouglasKS,EavesD,HartSD.HCR-20:assessingriskforviolence(version ow 3 ÆThgeismdeóttatir-aSn,aWlyshiisteoMfcJli,nSicpaelnjugdlegrmPeMnt,pMroajuegcht:efrifmtya-snixAySe,aArnsdoefrascocnuLmAu,laCteodokreRsSea,recthaol.n 40 W2of).edSbasinmtgeoernrCoFuDrs,anEseeasrvseUasnnidDve,rirDsskoit.yuF,goMlareesnnKstiaScl,PHWseyianclthtrhiua,ptLriAacw.ST,eharvenidcHePCsoRCli-co2ym0Imnscsishtisteuiomtnee,o:1ft9hB9eri7tai.sshseCsoslummebnita, nloa clinicalversusstatisticalprediction.CounsPsychol2006;34:341-82. d 45 KHadhiguirglotinypsasyNRc,,hWWiaatertayts.vDePrs,TyBc,inhMdiaamtdraeBnnuJlT,l.S2U0la0sd5ee;2aM9n,:d1T3ph1eo-rr3cne.icivroefdtGut.ilAityssoefssstirnugctvuiorelednvcieolreisnkcienrgisekneral 41 B1P9oro9ef5re.sDsPio,nHaalrgtuSidDe,lKinreosppfoPraRs,sWesesbinstgerrisCkDo.fMsaenxuuaallvfoiorltehnecese.xSuimalovnioFleransceerrUisnki-v2e0r.sity, ed fro assessmentsinEnglishmediumsecureforensicunits.PsychiatrBull2009;33:129-32. MentalHealth,Law,andPolicyInstitute,1997. m 6 PescosolidoBA,MartinJK,LongJS,MedinaTR,PhelanJC,LinkBG.“Adiseaselike 42 KroppPR,HartSD,WebsterCD,EavesD.Manualforthespousalassaultriskassessment h 7 WaanlcayollhoaotchleedCre”?,pMeAnuddlleeencnacPed.,eABomufrcgJheaPsnsscyPec,hinPiaaptrlumyb2eli0rc1Sr0e,;aR1c6ut7iso:cn1hs3e2tno1a-s3Dc0h,.iBzroopwhnreenCia.,Sdeerpioreusssciorinm,iannadl 43 gKguuroiiddpeep..PB2Rnrid,tiHsehadrC.tBSorlDuit,misWhbieCaboIsnltuesmtritCbuiDtae,IoEnnsatvFiteuastmeDilo.yMnVaFionalumeanillcyfoeVr,it1oh9lee9ns4pc.eo,u1sa9l9a5s.saultriskassessment ttp://w offendingandmentaldisorder.BrJPsychiatry1998;172:477-84. 44 KroppPR,HartSD,WebsterCD,EavesD.Spousalassaultriskassessmentguide(SARA). w Multi-HealthSystems,1999. w 8 FazelS,LichtensteinP,GrannM,GoodwinGM,LangstromN.Bipolardisorderandviolent crime:newevidencefrompopulation-basedlongitudinalstudiesandsystematicreview. 45 BorumR,BartelP,ForthA.Manualforthestructuredassessmentofviolenceriskinyouth .b (SAVRY).UniversityofSouthFlorida,2002. m 9 AFarczhelGSe,nLåPnsgyscthröiamtryN,2H01je0rn;6A7:,9G3r1a-n3n8.M,LichtensteinP.Schizophrenia,substanceabuse, 46 BorumR,BartelP,ForthA.Manualforthestructuredassessmentofviolenceriskinyouth j.c (SAVRY):version1.1.UniversityofSouthFlorida,2003. o andviolentcrime.JAMA2009;301:2016-23. 47 LeisticoA,SalekinR,DeCosterJ,RogersR.Alarge-scalemeta-analysisrelatingthe m 10 NanadtiomnaanlaIngsetimtuetentfoorfHscehaizltohpahnrdenCialininicparliEmxacreyllaenndces.eCcoonredainrytecrvaeren.tiNonICsEin,2th0e09tr.eatment Haremeasuresofpsychopathytoantisocialconduct.LawHumBehav 2008;32:28-45. o/ 48 BlairPR,MarcusDK,BoccacciniMT.Isthereanallegianceeffectforassessment n 11 AmericanPsychiatricAssociation.Practiceguidelinesforthetreatmentofpatientswith instruments?Actuarialriskassessmentasanexemplar.JClinPsychol2008;15:346-60. 1 schizophrenia.APA,2004. 6 12 HarrisonK.Dangerousoffenders,indeterminatesentencing,andtherehabilitation 49 CohenJ.Statisticalpoweranalysisforthebehavioralsciences.2nded.Erlbaum,1988. M 50 RosenthalR.Parametricmeasuresofeffectsize.In:CooperH,HedgesLV,eds.The revolution.JSocWelfareFamLaw2010;32:423-33. a 1134 SHjuiasmrticoconeu.JrAt.nBRn.eAvRegeravsianLlsaotwfpfrSoeordtcuicnStieoc:nit:2hp0er0or5ef;isl1inu:3grg9,e7pn-o4cl2iec1ion.fgi,nadnivdidpuuanlirsihskinagsisneasnsmacetnutairniaclraimgein.al 5512 RHhfaauocnsdtodcgrbieosos.JoPJ.kLsAoy,cfpLhrreeoohsblmeaMabaerinlctinhthyE-2sb0yLa0n.s8tEeh;sd1et3simmi:s1ea.9atS-es3asu0gro.eef,olo1fc9ea9ftf4ee.dctbsaizsee:droobnursatnnkestessttos.bAasnenrMataetshaSntdatother rch 20 UniversityofChicagoPress,2007. 2 1156 SGsacniunhdgiYzihto,epJHmhPur,ceZSon.neiaMtrepaonenrtredoMaoft,t1htRe0eneratipinvosahnyialcsarhthbhiolaeJutr,liidncFsdbatriezsueomplradSeiedn.rtSstso:t.rauSsccschtyhusiizrzteoeodpmphhaarrsteiscBneuircsellsvp2miae0twe1ien1ont;ft3ost7hfw:ev8iiovt9hal9elr-indi9sci1ktey2,o.rrifsevkliiaoinbleilnitty, 53 M1rree9asv6cuie3altw;ss3sk.4iIo:lnl5f:P9dD8,ia-Ge6gea1nkt1oss.soJtniJci,stBeCso,tsDasecuceyutkrsPaMJcJy,,.GHCaaotrcsbhoornaridsneRCMC,oe,ldTlasab.koCwroaoctinihogrnai,Yn2e.0A1hn0aa.nlhdytbstpoin:o/g/ksarfdnotdrasp.cyrosetcsehemrnaatnitniecg. 3 by gu org/. e 17 Jovoeffvteeanrnacoenvssi.ćwPAisthAyc,phToiosaštttreryavuCsmkliinaDtNiLce,suIvtrrokeosscnsivd2ići0sM0o9r,d;D6e3ar.:m5V9jao2njn-o3ov.sićanAit,PGraešgilć2M00J.9P;6r6e:d1i3ct-i2n1g.violencein 54 RaneaitlsymsisaoJfBs,eGnslaitsiviAtySa,nRdustjpeescAifiWcit,ySpcrohdoultceensRinJf,oBrmosastiuveytsPuMm,mZawryinmdeearmsuarnesAiHn.dBiaivganroiasttiec st. P 18 ViljoenJL,McLachlanK,VincentGM.Assessingviolenceriskandpsychopathyinjuvenile reviews.JClinEpidemiol2005;58:982-90. ro andadultoffenders:asurveyofclinicalpractices.Assessment2010;17:377-95. 55 RutterCM,GatsonisCA.Ahierarchicalregressionapproachtometa-analysisofdiagnostic te testaccuracyevaluations.StatMed2001;20:2865-84. c 19 BuchananA,LeeseM.Detentionofpeoplewithdangerousseverepersonalitydisorders: 56 GlasAS,LijmerJG,PrinsMH,BonselGJ,BossuytPM.Thediagnosticoddsratio:asingle te 20 Cma20aes0myta9sp-;t3aeb6nme:al5laly6Mtit7ciAc-r9,ce0Fov.mrieepwnac.rhLisaSonn,cGoefetin2nd0srt0re1ua;mu35eP8n.t:s1T9ah5ne5dp-9mre.edtihcotidosnooffavsiosleesnscmeeinnta.CdurilmtoJffuesntidceerBs:eAhav 5578 MFiCnlldoeinismcsPaimntsogyarecnorhDfoSt.l.eA1Ns9stu9spme4es;br6sfeo2irnr:7mgn8eap3enr-edc9dee2id.c.JttiooCndlsienotaEfivpnii.odBleermnJcioePl:s2by0ec0ihn3iga;5tar6cy:c11u19r2a99t7e-;31a57b.1o:u2t8a7c.curacy.JConsult d by co 21 HansonRK,Morton-BourgonKE.Theaccuracyofrecidivismriskassessmentsforsexual p 22 LoaffregnedeMrMs:,aRmyaentaC-aJn,aSlyinsgishoSfP1,1P8aptroendMictBio,nNsietulsdsieesn.OPBsy.cThhoelAprsesdeiscstiv2e0v0a9l;u2e1:o1f-2ri1s.k 59 MintooseassLuEm,mLaitrtyenRbOerCgcBu,rSveh:adpairtoa-Da.nCaloymticbailnainpgprinodaecpheensdaenndtssotumdeiesadodfiatiodniaaglnostictest yrig considerations.StatMed1993;12:1293-316. h categorizationinschizophrenia.HarvRevPsychiatry2011;19:25-33. 60 DeeksJ.Systematicreviewsofevaluationofdiagnosticandscreeningtests.In:Egger t. 23 MadenA.Standardisedriskassessment:whyallthefuss?PsychiatrBull2003;27:201-4. M,SmithGD,AltmanDG,eds.Systematicreviewsinhealthcare:meta-analysisincontext. 24 MullenPE.Schizophreniaandviolence:fromcorrelationstopreventativestrategies.Adv BMJPublishingGroups,2001. PsychiatrTreat2006;12:239-48. 61 HigginsJPT,ThompsonS,DeeksJ,AltmanD.Measuringinconsistencyinmeta-analyses. 25 SzmuklerG.Violenceriskpredictioninpractice.BrJPsychiatry2001;178:84-8. BMJ2003;327:557-60. 26 MoherD,LiberatiA,TetzlaffJ,AltmanDG,PRISMAGroup.Preferredreportingitemsfor 62 ThompsonSG,HigginsJPT.Howshouldmeta-regressionanalysesbeundertakenand systematicreviewsandmeta-analyses:thePRISMAStatement.PLoSMed interpreted?StatMed2002;21:1559-73. 2009;6:e1000097. 63 SinghJP,GrannM,FazelS.Acomparativestudyofviolenceriskassessmenttools:a 27 BontaJ.Offenderriskassessment:guidelinesforselectionanduse.CrimJustBehav systematicreviewandmetaregressionanalysisof68studiesinvolving25,980participants. 2002;29:355-79. ClinPsycholRev2011;31:499-513. 28 DorenDM.Evaluatingsexoffenders:amanualforcivilcommitmentsandbeyond.Sage, 64 StataCorp.Statastatisticalsoftware:release10.1.StataCorpLP,2007. 2002. No commercial reuse: See rights and reprints http://www.bmj.com/permissions Subscribe:http://www.bmj.com/subscribe BMJ2012;345:e4692doi:10.1136/bmj.e4692(Published24July2012) Page7of12 RESEARCH 65 TyrerP,DugganC,CooperS,CrawfordM,SeivewrightH,RutterD,etal.Thesuccesses 76 GilliganDG.Violencepredictionandmanagement[electronicresponsetoMadenA,Scott andfailuresoftheDSPDexperiment:theassessmentandmanagementofsevere F,BurnettR,LewisGH,SkapinakisP.Offendinginpsychiatricpatientsafterdischarge personalitydisorder.MedSciLaw2010;50:95-9. frommediumsecureunits:prospectivenationalcohortstudy].BMJ2004.www.bmj.com/ 66 SzmuklerG.Riskassessment:‘numbers’and‘values’.PsychiatrBull2003;27:205-207. rapid-response/2011/10/30/violence-prediction-and-management. 67 JanusE.Sexuallyviolentpredatorlaws:psychiatryinservicetoamorallydubious 77 HigginsJPT.Heterogeneityinmeta-analysisshouldbeexpectedandappropriately B enterprise.Lancet2004;3664:50-1. identified.IntJEpidemiol 2008;37:1158-60. M 68 CooneyMT,DubinaA,GrahamI.Valueandlimitationsofexistingscoresforthe 78 DaviesS,ClarkeM,DugganC.Offendinginpsychiatricpatientsafterdischargefrom J 69 EasicshelsesrmKe,nPtuohfacnaMrdAio,vSatsecuurlearrJri,sBk.aJchAmmanCnolLlMCa.rPdrieodli2c0ti0o9n;5o4f:f1ir2s0t9co-2ro7n.aryeventswith 79 mEdeidtoiurimal.sSehcouureldupnroittso:cCoolsnfvoircotiobnserravtaetimonaaylrbeesemairsclehabdeinregg[ilsetteterer]d.?BLMaJnc2e0t0240;31209;3:67854:3.448.. : firs theFraminghamscore:asystematicreview.AmHeartJ 2007;153:722-31. 80 HartSD,KroppPR,HareRD.Performanceofmalepsychopathsfollowingconditional t p 70 GeddesJ.Suicideandhomicidebypeoplewithmentalillness.BMJ1999;318:1225-6. releasefromprison.JConsultClinPsychol 1988;56:227-32. u 71 CparitciehntotnhoJmHMici.dAe,r1e9vi9e5w-2o0f1p0u.bJlisFhoerednisnidcePpseyncdheianttrinPqsuyicrhieosli2n0E1n1g;2la2n:7d6i1n-t8o9p.sychiatric Accepted:15June2012 blis h 72 NationalInstituteforHealthandClinicalExcellence.Antisocialpersonalitydisorder: e treatment,managementandprevention.NICE,2010. d 73 HansonRK,Morton-BourgonK.Predictorsofsexualrecidivism:anupdatedmeta-analysis. Citethisas:BMJ2012;345:e4692 a s PublicWorksandGovernmentServicesCanada,2004. Thisisanopen-accessarticledistributedunderthetermsoftheCreativeCommons 1 74 SinghJP,FazelS.Forensicriskassessment:ametareview.CrimJusticeBehav 0 2010;37:965-88. AttributionNon-commercialLicense,whichpermitsuse,distribution,andreproductionin .1 75 YangM,WongSCP,CoidJ.Theefficacyofviolenceprediction:ameta-analytic anymedium,providedtheoriginalworkisproperlycited,theuseisnoncommercialand 13 comparisonofnineriskassessmenttools.PsycholBull2010;136:740-67. isotherwiseincompliancewiththelicense.See:http://creativecommons.org/licenses/by- 6 nc/2.0/andhttp://creativecommons.org/licenses/by-nc/2.0/legalcode. /b m j.e 4 6 9 2 o n 2 4 J u ly 2 0 1 2 . D o w n lo a d e d fro m h ttp ://w w w .b m j.c o m o/ n 1 6 M a rc h 2 0 2 3 b y g u e s t. P ro te c te d b y c o p y rig h t. No commercial reuse: See rights and reprints http://www.bmj.com/permissions Subscribe:http://www.bmj.com/subscribe BMJ2012;345:e4692doi:10.1136/bmj.e4692(Published24July2012) Page8of12 RESEARCH Tables B Table1|Characteristicsofnineincludedriskassessmenttools M J Instrumenttypeandname Noofitems Population Outcome Currentmanual : firs Actuarial t p u LSI-R* 54 Adultoffenders Criminaloffending AndrewsandBonta(1995)32 b lis PCL-R† 20 Non-specific Notapplicable‡ Hare(2003)33,34 h e SORAG 14 Sexualoffenders Sexualoffending Quinseyetal(2006)35,36 d a s Static-99§ 10 Sexualoffenders Sexualoffending Harrisetal(2003)37,38 1 0 VRAG 12 Mentallydisorderedviolent Violentoffending Quinseyetal(2006)35,36 .1 offenders 13 6 Structuredclinicaljudgment /b m HCR-20 20 Psychiatricpatients Violentoffending Websteretal(1997)39,40 j.e SVR-20 20 Sexualoffenders Sexualoffending Boeretal(1997)41 46 9 SARA 20 Spousalassaulters Violentoffending Kroppetal(1999)42-44 2 o SAVRY 24 Adolescentoffenders Violentoffending Borum,Bartel,andForth(2003)45,46 n 2 4 *Lowandlowtomoderateriskcategoriescombinedtomakelowriskbin.Moderatetohighandhighriskcategoriescombinedtomakehighriskbin. J u †Psychopathicpatients(score>30)consideredhighriskgroup,non-psychopathicpatients(<30)consideredlowriskgroup.PCL-RscoresareincludedinSORAG, ly 2 VRAG,HCR-20,andSVR-20,andthusthepredictivevalidityoftheseinstrumentsdesignedfordifferentoutcomesiscorrelated. 0 1 ‡PCL-Rwasdesignedasapersonalityassessment.Itstartedtobeusedasariskinstrumenttopredictcriminaloffendingfrom1988onwards.80 2 §Moderate-lowandmoderate-highriskcategoriescombinedtomakemoderateriskbin. . D o w n lo a d e d fro m h ttp ://w w w .b m j.c o m o/ n 1 6 M a rc h 2 0 2 3 b y g u e s t. P ro te c te d b y c o p y rig h t. No commercial reuse: See rights and reprints http://www.bmj.com/permissions Subscribe:http://www.bmj.com/subscribe BMJ2012;345:e4692doi:10.1136/bmj.e4692(Published24July2012) Page9of12 RESEARCH Table2|Descriptiveanddemographiccharacteristicsofsamplesinvestigatingpredictivevalidityofriskassessmenttoolsdesignedto predictviolent,sexual,andcriminaloutcomes.Dataareno(%)ofsamplesunlessstatedotherwise.SD=standarddeviation Categoryandgroup Violent(n=30) Sexual(n=20) Criminal(n=23) B M Sourceofstudy J Journalarticle 21(70) 18(90) 18(78) : firs Conference 4(13) 0 0 t p u b Dissertation 4(13) 2(10) 3(13) lis h Governmentreport 1(3) 0 2(8) e d Toolinformation a s Typeoftool 1 0 Actuarial 9(30) 16(80) 23(100) .1 1 3 Structuredclinicaljudgment 21(70) 4(20) 0 6 /b Toolused m HCR-20 9(30) — – j.e 4 6 LSI-R — — 11(48) 9 2 PCL-R — — 12(52) o n SARA 3(10) — — 2 4 SAVRY 9(30) — — Ju ly SORAG — 3(15) — 2 0 Static-99 — 13(65) — 1 2 SVR-20 — 4(20) — . D o VRAG 9(30) — — w n Demographic(mean(SD)insample) lo a d Maleparticipants(no) 137(98) 519(713) 409(590) e d Whiteparticipants(no) 92(49) 201(185) 213(165) fro Age(years) 28.3(10.0) 39.7(4.0) 35.2(4.6) m h Studydesign ttp Samplesize(mean(SD)) 148(94) 510(681) 439(720) ://w Assessmentsetting w w Correctional 9(30) 12(60) 21(91) .b m Forensicpsychiatric 11(37) 6(30) 0 j.c o Generalpsychiatric 5(17) 0 0 m Mixed 3(10) 1(5) 2(9) o/ n Unstatedorunclear 2(7) 1(5) 0(0) 1 6 Locationofoutcome M a Community 21(70) 18(90) 22(96) rc h Intra-institutional 6(20) 0 1(4) 2 0 2 Mixed 3(10) 2(10) 0 3 b Temporaldesign y g Prospective 12(40) 5(25) 14(61) u e s Retrospective 17(57) 15(75) 9(39) t. P Notstatedorunclear 1(3) 0 0 ro Lengthoffollow-up(months;mean(SD)) 39.4(29.6) 82.4(50.4) 33.9(24.8) tec te Sourceofoutcome d b Criminalregister 16(53) 17(85) 17(74) y c Institutionalrecords 6(20) 0 1(4) o p y Collateralreport 2(7) 0 0 rig h Mixed 6(20) 3(15) 5(22) t. No commercial reuse: See rights and reprints http://www.bmj.com/permissions Subscribe:http://www.bmj.com/subscribe BMJ2012;345:e4692doi:10.1136/bmj.e4692(Published24July2012) Page10of12 RESEARCH Table3|Summaryaccuracyestimatesproducedbythreetypesoftoolsforriskassessment Violentoffending(n=30)* Sexualoffending(n=20)† Criminaloffending(n=23)‡ B Summaryestimates(95%CI)fromsummaryreceiveroperatingcharacteristiccurve M DSeiangsnitoivsittiycoddsratio 60..0972((40..5888ttoo80..0954)) 30..8888((20..3863ttoo60..4902)) 20..8441((20..0298ttoo30..8586)) J: firs Specificity 0.36(0.28to0.44) 0.34(0.20to0.51) 0.80(0.67to0.89) t pu b Individualstudyestimates(median(IQR)) lis h Areaunderthecurve 0.72(0.68-0.78) 0.74(0.66-0.77) 0.66(0.58-0.67) e d Positivepredictivevalue 0.41(0.27-0.60) 0.23(0.09-0.41) 0.52(0.32-0.59) a s Negativepredictivevalue 0.91(0.81-0.95) 0.93(0.82-0.98) 0.76(0.61-0.84) 10 .1 Numberneededtodetain 2(2-4) 5(2-11) 2(2-3) 1 3 Numbersafelydischarged 10(4-18) 14(5-48) 3(2-6) 6 /b m CI=confidenceinterval;IQR=interquartilerange;n=numberofsamples. j.e *HCR-20,SARA,SAVRY,andVRAG. 46 9 †SORAG,Static-99,andSVR-20. 2 ‡LSI-RandPCL-R. on 2 4 J u ly 2 0 1 2 . D o w n lo a d e d fro m h ttp ://w w w .b m j.c o m o/ n 1 6 M a rc h 2 0 2 3 b y g u e s t. P ro te c te d b y c o p y rig h t. No commercial reuse: See rights and reprints http://www.bmj.com/permissions Subscribe:http://www.bmj.com/subscribe

Description:
management. Introduction Department of the Solicitor General of Canada, 1999. 39. Webster Ruscio J. A probability-based measure of effect size: robustness to base rates and other factors. Ann Math Stat. 1963 and failures of the DSPD experiment: the assessment and management of severe.
See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.