Predicting theoutcome ofacutestroke: prospective evaluation offive multivariate models andcomparison withsimple methods

Fivemultivariate modelsdesigned topredicttheoutcome ofstroke were tested prospectively on 102consecutive stroke patients admittedto a district general hospital. Theresults were comparedwith predictions made usingtwo simpleclinicalvariables (theconscious levelon admission andthestateofurinary continence atfourweeks). Ofthethreemodels(developed inBelfast, Guy'sHospital andUppsala) intended foruseintheacute stagesofstroke thelasttwo were slightly more accurateintheir prediction ofdeath (75%)thanwas theadmission conscious levelalone(65%),whereastheBelfast modelhadan accuracy ofonly50°/O inthis situation. At a laterstage,thestateof urinarycontinence predicted goodand poor outcomeswithsimilar accuracy to thatofa multivariate modelfromEdinburgh.A modeldeveloped inBristol performedpoorly. Whentested prospectively, thesemultivariate modelsprovedconsiderably lessaccuratethanwhen they were firstdescribed. OnlytheUppsala modelshowedany advantage over simple clinical methods. Thismightbeofvaluein definingprognostic strataforclinical studies, butnot inthemanagement of individual patients. Simpleclinical variablesthusoffer as much toclinicians as complexmultivariate models.