CDOoDocuments.StdDocumentDescDocuments.DocumentDescContainers.ViewDescViews.ViewDescStores.StoreDescDocuments.ModelDescContainers.ModelDescModels.ModelDescStores.ElemDesc TextViews.StdViewDescTextViews.ViewDescTextModels.StdModelDescTextModels.ModelDesc,$TextModels.AttributesDesc'*G ,*PS +ArialG <(Times New RomanPS*uTTextRulers.StdRulerDescTextRulers.RulerDescTextRulers.StdStyleDescTextRulers.StyleDescZTextRulers.AttributesDesc$ ZGo*uTY?$ ZGo/ WinBUGS document - DEPRESSION STUDY (Paul Miller's depression data) Binary effectiveness measure Analyse efficacy and cost data assuming that the responses are bernoulli distributed and that costs are lognormally distributed. Assume non-informative priors for means Assume exchangable prior information for variances of log costs Includes the patient with zero cost as a two-stage model. A. The model model { for(t in 1:M[1]) { e1[t] ~ dbern( phi[1] ) ep1[t] <- e1[t] + 1 z1[t] ~ dbern(p[1,ep1[t]])} for(t in 1:M[2]) { e2[t] ~ dbern( phi[2] ) ep2[t] <- e2[t] + 1 z2[t] ~ dbern(p[2,ep2[t]])} for(j in 1:N[1] ) { dp1[j]<-e1[j] +1 d1[j] ~ dnorm( lambda[1,dp1[j]], siginv[1,dp1[j]] ) } for(j in 1:N[2] ) { dp2[j]<-e2[j] +1 d2[j] ~ dnorm( lambda[2,dp2[j]], siginv[2,dp2[j]] ) } for(i in 1:2) { for(c in 1:2) { p[i,c] ~ dbeta( a,b) lambda[i,c] ~ dnorm( ml, vl ) siginv[i,c] ~ dgamma( ds, omega[i]) gm[i,c] <- exp(lambda[i,c] + 1/(siginv[i,c]*2)) } omega[i] ~ dgamma(do[i], ao[i]) phi[i] <- 1 / (1 + exp( psi[i] )) gam[i] <- (phi[i] * gm[i,2]*p[i,2]) + (1-phi[i]) * gm[i,1]*p[i,1] } psi[1:2] ~ dmnorm( mp[], vp[,] ) for(k in 1:NK) { Q[k] <- step( (gam[1] - gam[2]) - K[k] * (phi[1] - phi[2]) ) } effic <- step(phi[2] - phi[1]) cheap <- step(gam[1] - gam[2]) dominant <- effic * cheap } Note: The quantities gam1 and gam2 define the expected costs. B. Efficacy measure is Good Response; Costs are the log total depression related costs. Outcome e1 = 1 if GLOBAL2 score is one, otherwise e1 = 0; z1 = 1 if the cost is non-zero, otherwise z1 = 0; d1 = log cost. list( M = c(39, 42), N=c(38,42) ) z1[] e1[] d1[] 1 1 4.90527 1 0 5.41610 1 0 6.39846 1 0 5.72031 1 0 6.12678 1 0 7.03040 1 1 6.23441 1 0 5.78837 1 0 5.59842 1 1 4.78749 1 0 5.43808 1 0 5.93904 1 0 6.40643 1 0 5.53197 1 0 5.39884 1 0 5.90574 1 0 6.07594 1 1 5.35413 1 0 6.45567 1 0 5.70870 1 1 4.78749 1 0 5.40699 1 0 7.30139 1 0 5.40722 1 0 6.12641 1 0 4.71438 1 0 5.01064 1 1 5.72997 1 1 6.33798 1 1 6.58398 1 0 5.79161 1 0 5.01064 1 1 4.78749 1 1 5.87698 1 1 6.25217 1 1 5.10595 1 0 5.79454 1 1 5.65767 0 0 0 z2[] e2[] d2[] 1 0 5.35541 1 1 6.32599 1 0 7.20935 1 0 6.79365 1 0 4.88432 1 0 5.89167 1 1 5.41044 1 0 6.41696 1 0 6.34553 1 1 4.73961 1 0 6.10014 1 1 5.58259 1 0 5.11572 1 1 3.96594 1 1 4.99903 1 1 5.87541 1 0 6.30363 1 0 7.88273 1 0 7.30828 1 0 4.79761 1 0 5.65547 1 0 5.61819 1 1 4.42891 1 0 5.70508 1 0 5.74524 1 1 3.43881 1 0 4.85266 1 1 5.92029 1 1 5.93063 1 1 5.67583 1 0 6.00808 1 1 2.96476 1 1 4.89328 1 1 4.75703 1 1 5.98944 1 1 4.22829 1 1 4.81931 1 1 5.82553 1 1 5.58086 1 1 4.31749 1 0 3.80666 1 0 5.32003 k 3. The prior. ds gives correlation between variances of log costs. mp and vp give information about efficacies. list( ml = 0, vl = 0.00001, ds = 10, do = c(0.001,0.001), ao = c(0.001, 0.001), a = 1, b = 1 mp = c(2.086, 4.014), vp = structure( .Data=c(0.0000001, 0 , 0, 0.0000001), .Dim=c(2,2) ), ) 4. Specifying the C/E acceptability curve list( NK = 35 K = c(1, 3, 10, 30, 100, 200, 300, 500, 700, 1000, 1500, 2000, 2500, 3000, 4000, 5000, 7500, 10000, 15000, 20000, 25000, 30000, 40000, 50000, 75000, 100000, 150000, 200000, 250000, 300000, 400000, 500000, 750000, 1000000, 5000000) ) 5. Starting values list( psi = c(0, 0), p = structure( .Data=c(1,1,1, 1), .Dim=c(2,2)), omega = c(0.001, 0.001), lambda = structure( .Data=c(0, 0, 0, 0), .Dim=c(2,2) ), siginv = structure( .Data=c(0.001, 0.001, 0.001, 0.001), .Dim=c(2,2) ) ) TextControllers.StdCtrlDescTextControllers.ControllerDescContainers.ControllerDescControllers.ControllerDesc aY?$0ZGo *G <[ @Documents.ControllerDesc hcs8 fh