L'applicazione dei metodi Bayesiani nella...
Transcript of L'applicazione dei metodi Bayesiani nella...
L’applicazione dei metodi Bayesiani nella
Farmacoeconomia
Gianluca Baio
Department of Statistical Science, University College London (UK)Department of Statistics, University of Milano Bicocca (Italy)
Razionalizzazione della spesa dei farmaci ad alto costo
Torino, 5 Ottobre 2012
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 1 / 27
Outline of presentation
1 What is Bayesian statistics?
– Relationship with standard statistical procedures– Prior distributions– Bayesian computation
2 How to implement Bayesian statistics in Health Economics?
– Probabilistic assumptions– Decision-theory– Sensitivity analysis
3 Example
– Modelling– Cost-effectiveness analysis– Probabilistic sensitivity analysis
4 Conclusions
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 2 / 27
Outline of presentation
1 What is Bayesian statistics?
– Relationship with standard statistical procedures– Prior distributions– Bayesian computation
2 How to implement Bayesian statistics in Health Economics?
– Probabilistic assumptions– Decision-theory– Sensitivity analysis
3 Example
– Modelling– Cost-effectiveness analysis– Probabilistic sensitivity analysis
4 Conclusions
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 2 / 27
Outline of presentation
1 What is Bayesian statistics?
– Relationship with standard statistical procedures– Prior distributions– Bayesian computation
2 How to implement Bayesian statistics in Health Economics?
– Probabilistic assumptions– Decision-theory– Sensitivity analysis
3 Example
– Modelling– Cost-effectiveness analysis– Probabilistic sensitivity analysis
4 Conclusions
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 2 / 27
Outline of presentation
1 What is Bayesian statistics?
– Relationship with standard statistical procedures– Prior distributions– Bayesian computation
2 How to implement Bayesian statistics in Health Economics?
– Probabilistic assumptions– Decision-theory– Sensitivity analysis
3 Example
– Modelling– Cost-effectiveness analysis– Probabilistic sensitivity analysis
4 Conclusions
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 2 / 27
Variability and statistical models
• Size N = 10
• Mean µ
• Standard deviation σ
• Size n = 5
• Mean x̄
• Standard deviation sx
In reality we observe only one such sample (out of the many possible — in factthere are 252 different ways of picking at random 5 units out of the population!)and we want to use the information contained in that sample to infer about thepopulation parameters (e.g. the true mean and standard deviation)
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 3 / 27
Deductive vs inductive inference
Hypothesis 1 Hypothesis 2 Hypothesis 3
∆ = 0% ∆ = 5% ∆ = 10%
c −5% 0% 5% 10% 15% c
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 4 / 27
Deductive vs inductive inference
Deduction Hypothesis 1 Hypothesis 2 Hypothesis 3
∆ = 0% ∆ = 5% ∆ = 10%
c −5% 0% 5% 10% 15% c
• “Standard” (frequentist) methods set the value of the parameters(hypotheses) and by deduction infer about the plausibility of the observeddata
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 4 / 27
Deductive vs inductive inference
Deduction Hypothesis 1 Hypothesis 2 Hypothesis 3 Induction
∆ = 0% ∆ = 5% ∆ = 10%
c −5% 0% 5% 10% 15% c
• “Standard” (frequentist) methods set the value of the parameters(hypotheses) and by deduction infer about the plausibility of the observeddata
• Conversely, Bayesian statistics “conditions” on the observed data and byinduction makes inference on the unobservable parameters (hypotheses)
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 4 / 27
Bayesian inference
Prior(subjectiveknowledge)
p(θ)
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 5 / 27
Bayesian inference
Data(observedevidence)
Prior(subjectiveknowledge)
p(y | θ) p(θ)
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 5 / 27
Bayesian inference
Data(observedevidence)
Prior(subjectiveknowledge)
Bayestheorem
p(y | θ) p(θ)
p(θ)p(y | θ)
p(y)
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 5 / 27
Bayesian inference
Data(observedevidence)
Prior(subjectiveknowledge)
Bayestheorem
Posterior(updatedknowledge)
p(y | θ) p(θ)
p(θ)p(y | θ)
p(y)
p(θ | y)
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 5 / 27
Bayesian inference — updating knowledge
θ
0.0 0.2 0.4 0.6 0.8 1.0
PriorLikelihoodPosterior
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 6 / 27
Choice of the prior distribution
• Non-informative prior
– Attempts to include minimal information in the prior to “let the data speak forthemselves” (sometimes known as “minimally informative”)
– Need to be careful in defining the scale in which non-informativeness is selected– Sometimes helpful as preliminary approximation — often leads to essentially
the same inference as using maximum likelihood
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 7 / 27
Choice of the prior distribution
• Non-informative prior
– Attempts to include minimal information in the prior to “let the data speak forthemselves” (sometimes known as “minimally informative”)
– Need to be careful in defining the scale in which non-informativeness is selected– Sometimes helpful as preliminary approximation — often leads to essentially
the same inference as using maximum likelihood
• Conjugate prior
– Convenient mathematical formulation– Prior and posterior in the same family
E.g. Prior = Normal(m0, s0) + Data = Normal(µ, σ2) ⇒E.g. Posterior = Normal(m1, s1)
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 7 / 27
Choice of the prior distribution
• Non-informative prior
– Attempts to include minimal information in the prior to “let the data speak forthemselves” (sometimes known as “minimally informative”)
– Need to be careful in defining the scale in which non-informativeness is selected– Sometimes helpful as preliminary approximation — often leads to essentially
the same inference as using maximum likelihood
• Conjugate prior
– Convenient mathematical formulation– Prior and posterior in the same family
E.g. Prior = Normal(m0, s0) + Data = Normal(µ, σ2) ⇒E.g. Posterior = Normal(m1, s1)
• Informative prior
– Proper Bayesian model: express (subjective) knowledge by means of a suitableprobability distribution
– Can be based on hard evidence– NB: Informative priors are not necessarily conjugated!
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 7 / 27
Non conjugated models
• Despite their usefulness in computational terms, non-informative andconjugated models are not always the best
– Too restrictive, might not encode the actual level of prior information– Non-informative priors are generally not invariant to scale transformations– When more complex (and realistic!) structures considered — for instance
multiparametric, or generalised linear models — conjugacy rarely hold
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 8 / 27
Non conjugated models
• Despite their usefulness in computational terms, non-informative andconjugated models are not always the best
– Too restrictive, might not encode the actual level of prior information– Non-informative priors are generally not invariant to scale transformations– When more complex (and realistic!) structures considered — for instance
multiparametric, or generalised linear models — conjugacy rarely hold
• Since the 1990’s the development of MCMC methods has allowed the use ofsimulation techniques for Bayesian computation. Software like BUGS1 orJAGS
2 can be used to perform analysis on most “real-life” problems
• If the model is well specified, the level of accuracy of the approximationprovided by the simulation technique is very good
1http://www.mrc-bsu.cam.ac.uk/bugs/2http://mcmc-jags.sourceforge.net/
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 8 / 27
MCMC methods
−2 0 2 4 6 8
23
45
67
After 10 iterations
µ
σ
1
2
3
4
56
7
89
10
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 9 / 27
MCMC methods
−2 0 2 4 6 8
23
45
67
After 30 iterations
µ
σ
1
2
3
4
56
7
89
10
11
12
13
14
15
1617
18
19 2021
22
23
24
25 26
27
28
29
30
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 9 / 27
MCMC methods
−2 0 2 4 6 8
23
45
67
After 1000 iterations
µ
σ
1
2
3
4
56
7
89
10
11
12
13
14
15
1617
18
19 2021
22
23
24
25 26
27
28
29
30
31
32
33
34
35
36
37
38 39
40
41
42
43
44
45
46
47
484950
51
52
53
5455
56
57
58
59
60
61
62 63
646566
67
68
69
70
71
72 73
74
75
76
77
78
79
80
81
82
83
8485
86
87
88
89
90
91
92
93
9495
96
97
9899100
101
102
103
104105
106
107 108
109
110
111
112
113
114
115116117118119
120
121122
123
124
125126
127
128129
130
131132
133
134
135136
137
138
139140141142
143
144
145
146
147
148
149150
151152
153 154
155156
157158
159
160
161
162 163164
165
166
167
168169
170
171
172
173
174
175
176
177
178
179180181182
183
184
185
186187
188 189
190
191 192
193
194
195
196197
198199
200201
202203
204
205
206
207
208
209
210
211 212213
214
215
216
217
218219
220
221
222
223
224225
226
227
228
229
230
231
232
233
234
235
236237238
239
240
241
242
243
244
245
246
247
248
249
250251252
253254
255
256257
258
259
260
261262
263
264
265
266
267
268
269
270
271
272
273
274
275
276277
278279
280
281
282
283
284
285
286287288
289
290291
292
293
294
295296
297
298
299300
301
302 303
304305
306307
308
309
310
311312
313
314
315
316
317
318
319
320
321
322
323
324 325326327
328
329
330
331
332
333
334335336
337338339
340
341
342
343
344
345
346
347
348349
350
351
352
353
354355
356
357
358
359360
361362
363
364
365
366
367
368
369
370
371
372373
374
375
376377378
379
380
381382
383
384
385
386
387
388
389390391
392
393
394
395
396
397398
399
400
401402403
404
405
406
407
408409
410
411
412
413
414415
416
417418419
420
421
422423
424
425426 427
428
429
430431
432433
434435
436
437438439
440
441
442
443 444445446
447
448449
450
451
452453
454
455
456
457
458
459460
461
462463
464
465
466
467
468
469
470
471472 473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488489490
491
492493
494
495
496 497
498
499
500501
502503
504505
506
507
508509
510
511
512513
514515
516
517
518
519520
521522
523
524
525
526
527
528
529530
531532
533534
535
536
537538
539
540541
542
543
544
545
546
547
548
549550
551
552
553
554
555
556
557
558
559560
561
562
563
564
565
566
567
568
569570
571572
573
574575
576577
578
579
580
581
582
583
584
585
586
587
588589
590
591592
593594595
596
597
598
599
600
601602603604
605
606
607608609
610
611
612
613
614
615
616
617618
619
620621
622623
624
625
626
627
628629
630
631
632
633
634
635636637
638
639640
641
642
643
644
645
646
647
648
649
650
651
652
653
654 655
656
657658
659
660
661
662663
664
665
666
667
668
669
670671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689690
691
692
693
694
695
696
697
698699
700701
702703
704
705706707
708709
710
711712
713
714715716
717
718
719
720
721722 723
724
725
726727
728
729
730
731
732
733
734735736737
738
739740
741
742
743
744
745
746 747
748
749
750
751
752
753754
755
756
757758
759
760
761
762
763764
765766
767 768
769
770771
772
773
774
775
776
777
778779
780
781
782
783
784
785
786
787
788789
790791
792
793
794
795796797
798799
800
801
802
803804805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830831
832
833
834835
836837
838
839840
841
842843
844
845
846
847
848
849
850
851852
853
854
855
856857858
859
860
861
862
863
864
865
866
867
868869
870
871
872
873
874875
876
877878
879
880
881
882
883
884885886 887
888
889
890
891892
893
894895
896
897
898
899 900
901902
903
904
905
906907908
909
910
911
912 913
914
915 916917
918919 920921
922
923
924925
926
927
928929
930
931
932933
934
935
936
937
938
939
940941
942
943944
945
946
947
948
949950
951
952953954
955
956957
958
959
960
961
962963964
965
966
967
968
969970
971972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 9 / 27
MCMC methods
0 100 200 300 400 500
−10
0−
500
5010
015
020
0
Iteration
Chain 1Chain 2
Burn−in Sample after convergence
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 9 / 27
(Bayesian) Decision-making process
• Typically, we define a “health economic response” (e, c), where for eachintervention (treatment) t
– e represents a suitable measure of clinical benefits (e.g. QALYs)– c are the costs associated with a given intervention
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 10 / 27
(Bayesian) Decision-making process
• Typically, we define a “health economic response” (e, c), where for eachintervention (treatment) t
– e represents a suitable measure of clinical benefits (e.g. QALYs)– c are the costs associated with a given intervention
• The variables (e, c) are usually defined as functions of a set of relevantparameters θt which represent some population-level features of theunderlying process
– Probability of some clinical outcome– Duration in treatment– Reduction in the rate of occurrence of some event
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 10 / 27
(Bayesian) Decision-making process
• Typically, we define a “health economic response” (e, c), where for eachintervention (treatment) t
– e represents a suitable measure of clinical benefits (e.g. QALYs)– c are the costs associated with a given intervention
• The variables (e, c) are usually defined as functions of a set of relevantparameters θt which represent some population-level features of theunderlying process
– Probability of some clinical outcome– Duration in treatment– Reduction in the rate of occurrence of some event
• There are (at least) two sources of uncertainty
– Sampling variability is modelled using an intervention-specific distributionp(e, c | θt)
– Parametric uncertainty is modelled using a (possibly subjective) priordistribution p(θt | D), based on some background data D
– NB: Sometimes, we can (should!) consider also structural uncertainty, i.e.about the modelling assumptions used
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 10 / 27
(Bayesian) Decision-making process
• In addition, we define a utility function to describe the quality of t
– The function u(e, c; t) describes the value associated with applyingintervention t, in terms of the future (uncertain) outcomes
– Uncertainty is expressed through p(e, c,θ) = p(e, c | θ)p(θ | D)
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 11 / 27
(Bayesian) Decision-making process
• In addition, we define a utility function to describe the quality of t
– The function u(e, c; t) describes the value associated with applyingintervention t, in terms of the future (uncertain) outcomes
– Uncertainty is expressed through p(e, c,θ) = p(e, c | θ)p(θ | D)
• NB: typically, the utility function chosen is the monetary net benefit
u(e, c; t) := ket − ct
– k is the “willingness to pay”, i.e. the cost per extra unit of effectiveness gained
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 11 / 27
(Bayesian) Decision-making process
• In addition, we define a utility function to describe the quality of t
– The function u(e, c; t) describes the value associated with applyingintervention t, in terms of the future (uncertain) outcomes
– Uncertainty is expressed through p(e, c,θ) = p(e, c | θ)p(θ | D)
• NB: typically, the utility function chosen is the monetary net benefit
u(e, c; t) := ket − ct
– k is the “willingness to pay”, i.e. the cost per extra unit of effectiveness gained
• Decision making is based on
– Computing for each intervention t the expected utility
Ut = E[u(e, c; t)]
(computed with respect to both individual and population uncertainty)
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 11 / 27
(Bayesian) Decision-making process
• In addition, we define a utility function to describe the quality of t
– The function u(e, c; t) describes the value associated with applyingintervention t, in terms of the future (uncertain) outcomes
– Uncertainty is expressed through p(e, c,θ) = p(e, c | θ)p(θ | D)
• NB: typically, the utility function chosen is the monetary net benefit
u(e, c; t) := ket − ct
– k is the “willingness to pay”, i.e. the cost per extra unit of effectiveness gained
• Decision making is based on
– Computing for each intervention t the expected utility
Ut = E[u(e, c; t)]
(computed with respect to both individual and population uncertainty)
– Treating the entire homogeneous (sub)population with the most cost-effectivetreatment, i.e. that associated with the maximum expected utility
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 11 / 27
(Bayesian) Decision-making process
• In addition, we define a utility function to describe the quality of t
– The function u(e, c; t) describes the value associated with applyingintervention t, in terms of the future (uncertain) outcomes
– Uncertainty is expressed through p(e, c,θ) = p(e, c | θ)p(θ | D)
• NB: typically, the utility function chosen is the monetary net benefit
u(e, c; t) := ket − ct
– k is the “willingness to pay”, i.e. the cost per extra unit of effectiveness gained
• Decision making is based on
– Computing for each intervention t the expected utility
Ut = E[u(e, c; t)]
(computed with respect to both individual and population uncertainty)
– Treating the entire homogeneous (sub)population with the most cost-effectivetreatment, i.e. that associated with the maximum expected utility
– Performing sensitivity analysis (to parameter and/or structural uncertainty)to investigate the impact of underlying uncertainty on the decision process
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 11 / 27
Example: Chemotherapy
t = 0: Old chemotherapy
A0
Ambulatory care(γ)
SE0
Blood-relatedside effects
(π0)
H0
Hospital admission(1 − γ)
cdrug0 L99
NStandardtreatment
A0
Ambulatory care(γ)
N − SE0
No side effects(1 − π0)
H0
Hospital admission(1 − γ)
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 12 / 27
Example: Chemotherapy
t = 0: Old chemotherapy
A0
Ambulatory care(γ)
99K camb
SE0
Blood-relatedside effects
(π0)
H0
Hospital admission(1 − γ)
99Kchosp
cdrug0 L99
NStandardtreatment
A0
Ambulatory care(γ)
99K camb
N − SE0
No side effects(1 − π0)
H0
Hospital admission(1 − γ)
99Kchosp
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 12 / 27
Example: Chemotherapy
t = 1: New chemotherapy
A1
Ambulatory care(γ)
99K camb
SE1
Blood-relatedside effects(π1 = π0ρ)
H1
Hospital admission(1 − γ)
99Kchosp
cdrug1 L99
NNew
treatment
A1
Ambulatory care(γ)
99K camb
N − SE1
No side effects(1 − π1)
H1
Hospital admission(1 − γ)
99Kchosp
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 12 / 27
Prior information vs prior distributions
• Often, we can use the prior information and formalise it in order toapproximate it with a suitable probability distribution
Mean 2.5% Median 97.5% Distribution
π0 0.241 0.1633 0.2378 0.3295 Beta(27.12, 85.88)ρ 0.8004 0.4058 0.7913 1.1702 Normal(0.8, 0.2)γ 0.619 0.570 0.616 0.667 Beta(5.80, 13.80)camb 120.11 86.15 118.78 160.31 logNormal(4.77, 0.17)chosp 5483.36 3744.44 5394.53 7703.16 logNormal(8.60, 0.18)
cdrug0 110 — — — —
cdrug1 520 — — — —
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 13 / 27
Prior information vs prior distributions
0.0 0.2 0.4 0.6 0.8 1.0
π0 ∼ Beta(27.12, 85.88)
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 14 / 27
Prior information vs prior distributions
0.0 0.2 0.4 0.6 0.8 1.0
Pr(π0 < 0.1633) = 0.025
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 14 / 27
Prior information vs prior distributions
0.0 0.2 0.4 0.6 0.8 1.0
Pr(π0 < 0.1633) = 0.025 Pr(π0 > 0.3295) = 0.025
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 14 / 27
Prior information vs prior distributions
0.0 0.2 0.4 0.6 0.8 1.0
Pr(π0 < 0.1633) = 0.025 Pr(π0 > 0.3295) = 0.025
Pr(0.1633 ≤ π0 ≤ 0.3295) = 0.95
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 14 / 27
Bayesian model — specification
• The assumptions underlying the model can be coded in BUGS/JAGSmodel {
pi[1] ~ dbeta(a.pi,b.pi) # Baseline probability of side effects (t=0)pi[2] <- pi[1]*rho # Decreased probability of side effects (t=1)rho ~ dnorm(m.rho,tau.rho) # Decrement rate in side effects for t=1
gamma ~ dbeta(a.gamma,b.gamma) # Probability of ambulatory carec.amb ~ dlnorm(m.amb,tau.amb) # Unit cost of ambulatory care
c.hosp ~ dlnorm(m.hosp,tau.hosp) # Unit cost of hospitalisationfor (t in 1:2) {
SE[t] ~ dbin(pi[t],N) # Predicted no. patients with side effects
A[t] ~ dbin(gamma,SE[t]) # Predicted no. patients needing ambulatory careH[t] <- SE[t] - A[t] # Predicted no. patients needing hospitalisation
}}
and then the MCMC analysis can be performed
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 15 / 27
Bayesian model — specification
• The assumptions underlying the model can be coded in BUGS/JAGSmodel {
pi[1] ~ dbeta(a.pi,b.pi) # Baseline probability of side effects (t=0)pi[2] <- pi[1]*rho # Decreased probability of side effects (t=1)rho ~ dnorm(m.rho,tau.rho) # Decrement rate in side effects for t=1
gamma ~ dbeta(a.gamma,b.gamma) # Probability of ambulatory carec.amb ~ dlnorm(m.amb,tau.amb) # Unit cost of ambulatory care
c.hosp ~ dlnorm(m.hosp,tau.hosp) # Unit cost of hospitalisationfor (t in 1:2) {
SE[t] ~ dbin(pi[t],N) # Predicted no. patients with side effects
A[t] ~ dbin(gamma,SE[t]) # Predicted no. patients needing ambulatory careH[t] <- SE[t] - A[t] # Predicted no. patients needing hospitalisation
}}
and then the MCMC analysis can be performed
• The MCMC procedure will generate samples from the posterior distributionsof the relevant quantities
θt = (πt, γ, ρ, SEt, At, Ht, c
amb, chosp, cdrug)
• These can be combined to compute the variables of cost and benefit, andperform the economic analysis
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 15 / 27
Bayesian model — convergence10
015
020
025
0A[1]
iteration
A[1
]
25 50 75 100 125 150 175 200 225 250
5010
020
0
A[2]
iteration
A[2
]
25 50 75 100 125 150 175 200 225 250
5010
015
020
0
H[1]
iteration
H[1
]
25 50 75 100 125 150 175 200 225 250
5010
015
020
0
H[2]
iteration
H[2
]
25 50 75 100 125 150 175 200 225 25010
020
030
0
SE[1]
iteration
SE
[1]
25 50 75 100 125 150 175 200 225 250
100
200
300
400
SE[2]
iteration
SE
[2]
25 50 75 100 125 150 175 200 225 250
3000
6000
9000
c.hosp
iteration
c.ho
sp
25 50 75 100 125 150 175 200 225 250
8012
016
0
c.amb
iteration
c.am
b
25 50 75 100 125 150 175 200 225 250
0.4
0.6
0.8
gamma
iteration
gam
ma
25 50 75 100 125 150 175 200 225 250
0.15
0.25
pi[1]
iteration
pi[1
]
25 50 75 100 125 150 175 200 225 250
0.1
0.2
0.3
0.4
pi[2]
iteration
pi[2
]
25 50 75 100 125 150 175 200 225 250
0.2
0.6
1.0
1.4
rho
iteration
rho
25 50 75 100 125 150 175 200 225 250
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 16 / 27
Bayesian model — posterior distributionsπ0
0.15 0.20 0.25 0.30 0.35
020
4060
8010
0
γ
0.3 0.4 0.5 0.6 0.7 0.8
020
4060
8010
012
0
ρ
0.2 0.4 0.6 0.8 1.0 1.2 1.4
020
4060
8010
0
SE0
100 150 200 250 300 350
020
4060
8010
0
SE1
0 100 200 300 400
050
100
150
A0
100 150 200 250
050
100
150
A1
0 50 100 150 200 250
020
4060
8010
0
H0
50 100 150 200
050
100
150
chosp
2000 4000 6000 8000 10000
050
100
150
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 17 / 27
Measures of cost & benefits
• The total cost associated with each treatment can be computed bymultiplying the unit cost of each clinical resource (drug, ambulatory care andhospital admission) by the number of patients consuming it. Thus:
ct := cdrugt (N − SEt) + (cdrugt + camb)At + (cdrugt + chosp)Ht
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 18 / 27
Measures of cost & benefits
• The total cost associated with each treatment can be computed bymultiplying the unit cost of each clinical resource (drug, ambulatory care andhospital admission) by the number of patients consuming it. Thus:
ct := cdrugt (N − SEt) + (cdrugt + camb)At + (cdrugt + chosp)Ht
• Similarly, the measure of effectiveness can be computed as the total numberof patients who do not experience side effects
et := (N − SEt)
• NB: we can (should) extend this to consider QALYs, instead of the “hard”effectiveness measure in terms of events averted
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 18 / 27
Expected incremental benefit
0 10000 20000 30000 40000 50000
050
0000
1000
000
1500
000
2000
000
Expected Incremental Benefit
Willingness to pay
EIB
k* = 6497.1
EIB = U1 − U0
Based on the current evidence, choose old chemotherapy if k < 6 500monetary units and new chemotherapy otherwise
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 19 / 27
Probabilistic sensitivity analysis (PSA)
The quality of the current evidence is often limited
• During the pre-market authorisation phase, the regulator should decidewhether to grant reimbursement to a new product — and in some countriesalso set the price — on the basis of uncertain evidence, regarding bothclinical and economic outcomes
• Although it is possible to answer some unresolved questions after marketauthorisation, relevant decisions such as that on reimbursement (whichdetermines the overall access to the new treatment) have already been taken
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 20 / 27
Probabilistic sensitivity analysis (PSA)
The quality of the current evidence is often limited
• During the pre-market authorisation phase, the regulator should decidewhether to grant reimbursement to a new product — and in some countriesalso set the price — on the basis of uncertain evidence, regarding bothclinical and economic outcomes
• Although it is possible to answer some unresolved questions after marketauthorisation, relevant decisions such as that on reimbursement (whichdetermines the overall access to the new treatment) have already been taken
• This leads to the necessity of performing (probabilistic) sensitivity analysis(PSA)
– Formal quantification of the impact of uncertainty in the parameters on theresults of the economic model
– Standard requirement in many health systems (e.g. for NICE in the UK), butstill not universally applied
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 20 / 27
PSA to parameter uncertaintyParameters Model structure Decision analysis
0.0 0.2 0.4 0.6 0.8 1.0
0.0 0.5 1.0 1.5 2.0
0.0 0.2 0.4 0.6 0.8 1.0
0 2000 6000 10000
π0
ρ
γ
chosp
Old chemotherapy
A0
Ambulatory care(γ)
99K camb
SE0
Blood-relatedside effects
(π0)
H0
Hospital admission(1 − γ)
99Kchosp
cdrug0 L99
NStandardtreatment
A0
Ambulatory care(γ)
99K camb
N − SE0
No side effects(1 − π0)
H0
Hospital admission(1 − γ)
99Kchosp
New chemotherapy
A1
Ambulatory care(γ)
99K camb
SE1
Blood-relatedside effects(π1 = π0ρ)
H1
Hospital admission(1 − γ)
99Kchosp
cdrug1 L99
NNew
treatment
A1
Ambulatory care(γ)
99K camb
N − SE1
No side effects(1 − π1)
H1
Hospital admission(1 − γ)
99Kchosp
Old chemotherapyBenefits Costs
743.1 656 644.6
New chemotherapyBenefits Costs
743.1 656 644.6
ICER =20 000
1QALY
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 21 / 27
PSA to parameter uncertaintyParameters Model structure Decision analysis
0.0 0.2 0.4 0.6 0.8 1.0
0.0 0.5 1.0 1.5 2.0
0.0 0.2 0.4 0.6 0.8 1.0
0 2000 6000 10000
π0
ρ
γ
chosp
x
x
x
x
Old chemotherapy
A0
Ambulatory care(γ)
99K camb
SE0
Blood-relatedside effects
(π0)
H0
Hospital admission(1 − γ)
99Kchosp
cdrug0 L99
NStandardtreatment
A0
Ambulatory care(γ)
99K camb
N − SE0
No side effects(1 − π0)
H0
Hospital admission(1 − γ)
99Kchosp
New chemotherapy
A1
Ambulatory care(γ)
99K camb
SE1
Blood-relatedside effects(π1 = π0ρ)
H1
Hospital admission(1 − γ)
99Kchosp
cdrug1 L99
NNew
treatment
A1
Ambulatory care(γ)
99K camb
N − SE1
No side effects(1 − π1)
H1
Hospital admission(1 − γ)
99Kchosp
Old chemotherapyBenefits Costs741 670 382.1
New chemotherapyBenefits Costs732 1 131 978
ICER =20 000
1QALY
⇒ ⇒
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 21 / 27
PSA to parameter uncertaintyParameters Model structure Decision analysis
0.0 0.2 0.4 0.6 0.8 1.0
0.0 0.5 1.0 1.5 2.0
0.0 0.2 0.4 0.6 0.8 1.0
0 2000 6000 10000
π0
ρ
γ
chosp
x
x
x
x
Old chemotherapy
A0
Ambulatory care(γ)
99K camb
SE0
Blood-relatedside effects
(π0)
H0
Hospital admission(1 − γ)
99Kchosp
cdrug0 L99
NStandardtreatment
A0
Ambulatory care(γ)
99K camb
N − SE0
No side effects(1 − π0)
H0
Hospital admission(1 − γ)
99Kchosp
New chemotherapy
A1
Ambulatory care(γ)
99K camb
SE1
Blood-relatedside effects(π1 = π0ρ)
H1
Hospital admission(1 − γ)
99Kchosp
cdrug1 L99
NNew
treatment
A1
Ambulatory care(γ)
99K camb
N − SE1
No side effects(1 − π1)
H1
Hospital admission(1 − γ)
99Kchosp
Old chemotherapyBenefits Costs741 670 382.1699 871 273.3
New chemotherapyBenefits Costs732 1 131 978664 1 325 654
ICER =20 000
1QALY
⇒ ⇒
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 21 / 27
PSA to parameter uncertaintyParameters Model structure Decision analysis
0.0 0.2 0.4 0.6 0.8 1.0
0.0 0.5 1.0 1.5 2.0
0.0 0.2 0.4 0.6 0.8 1.0
0 2000 6000 10000
π0
ρ
γ
chosp
x
x
x
x
Old chemotherapy
A0
Ambulatory care(γ)
99K camb
SE0
Blood-relatedside effects
(π0)
H0
Hospital admission(1 − γ)
99Kchosp
cdrug0 L99
NStandardtreatment
A0
Ambulatory care(γ)
99K camb
N − SE0
No side effects(1 − π0)
H0
Hospital admission(1 − γ)
99Kchosp
New chemotherapy
A1
Ambulatory care(γ)
99K camb
SE1
Blood-relatedside effects(π1 = π0ρ)
H1
Hospital admission(1 − γ)
99Kchosp
cdrug1 L99
NNew
treatment
A1
Ambulatory care(γ)
99K camb
N − SE1
No side effects(1 − π1)
H1
Hospital admission(1 − γ)
99Kchosp
Old chemotherapyBenefits Costs741 670 382.1699 871 273.3. . . . . .726 425 822.2743.1 656 644.6
New chemotherapyBenefits Costs732 1 131 978664 1 325 654. . . . . .811 766 411.4794.6 991 804.0
ICER =335 159.4
51.6
ICER= 6 497.1
⇒ ⇒
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 21 / 27
Cost-effectiveness plane
−200 −100 0 100 200
−20
0000
020
0000
6000
00
Cost effectiveness plane contour plot New Chemotherapy vs Old Chemotherapy
Effectiveness differential
Cos
t diff
eren
tial
Pr(∆e > 0, ∆c > 0) = 0.756Pr(∆e ≤ 0, ∆c > 0) = 0.19
Pr(∆e ≤ 0, ∆c ≤ 0) = 0 Pr(∆e > 0, ∆c ≤ 0) = 0.054
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 22 / 27
Cost-effectiveness plane
The economic analysis depends on the willingness-to-pay, which determines thesustainability area
Cost effectiveness plane New Chemotherapy vs Old Chemotherapy
Effectiveness differential
Cos
t diff
eren
tial
−200 −100 0 100 200
−20
0000
020
0000
6000
00
• ICER=6497.10
k = 25000
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 23 / 27
Cost-effectiveness plane
The economic analysis depends on the willingness-to-pay, which determines thesustainability area
Cost effectiveness plane New Chemotherapy vs Old Chemotherapy
Effectiveness differential
Cos
t diff
eren
tial
−200 −100 0 100 200
−20
0000
020
0000
6000
00
• ICER=6497.10
k = 25000
Cost effectiveness plane New Chemotherapy vs Old Chemotherapy
Effectiveness differential
Cos
t diff
eren
tial
−200 −100 0 100 200
−20
0000
020
0000
6000
00
• ICER=6497.10
k = 1000
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 23 / 27
Summarising PSA
• For any given value of the willingness-to-pay k, we can analyse the “possiblefutures”
• For example, consider k = 25 000 monetary units
Parameters simulations Expected Incrementalt = 0 t = 1 utility benefit
Iter/n Benefits Costs Benefits Costs U(θ0) U(θ1) IB(θ)
1 741 670 382.1 732 1 131 978 19 214 751 19 647 706 432 955.8
2 699 871 273.3 664 1 325 654 17 165 526 17 163 407 -2 119.3
3 774 639 071.7 706 1 191 567.2 18 710 928 16 458 433 -2 252 495.5
4 721 1 033 679.2 792 1 302 352.2 16 991 321 18 497 648 1 506 327.0
5 808 427 101.8 784 937 671.1 19 772 898 18 662 329 -1 110 569.3
6 731 1 168 864.4 811 717 939.2 17 106 136 18 983 331 1 877 195.1
. . . . . . . . . . . .
1000 739 431 079.0 699 1 004 195.0 18 043 921 16 470 805 -1 573 116.0
U0=18 659 238 U
1=19 515 004 EIB= 855 766
• One way of summarising PSA is to compute the cost-effectivenessacceptability curve
CEAC = Pr(IB(θ) | D) > 0
• Upon varying k, this is the probability that the “optimal” decision would notbe reversed by reduced uncertainty
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 24 / 27
Summarising PSA
0 10000 20000 30000 40000 50000
0.0
0.2
0.4
0.6
0.8
1.0
Cost Effectiveness Acceptability Curve
Willingness to pay
Pro
babi
lity
of c
ost e
ffect
iven
ess
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 24 / 27
Summarising PSA
• NB: CEACs only quantify the probability of cost-effectiveness, but do not sayanything about the payoffs associated with “taking the wrong decision”
Parameters simulations Expected Maximum Opportunityt = 0 t = 1 utility utility loss
Iter/n Benefits Costs Benefits Costs U(θ0) U(θ1) U∗(θ) OL(θ)
1 741 670 382.1 732 1 131 978 19 214 751 19 647 706 19 647 706 —
2 699 871 273.3 664 1 325 654 17 165 526 17 163 407 17 165 526 2 119.3
3 774 639 071.7 706 1 191 567.2 18 710 928 16 458 433 18 710 928 2 252 495.5
4 721 1 033 679.2 792 1 302 352.2 16 991 321 18 497 648 18 497 648 —
5 808 427 101.8 784 937 671.1 19 772 898 18 662 329 19 772 898 1 110 569.3
6 731 1 168 864.4 811 717 939.2 17 106 136 18 983 331 18 983 331 —
. . . . . . . . . . . .
1000 739 431 079.0 699 1 004 195.0 18 043 921 16 470 805 18 043 921 1 573 116.0
EVDI= 226 585
• At each iteration, the OL is the difference between the maximum utility andthe value associated with the intervention with the maximum utility overall
• The expected value of information is the average opportunity loss
EVDI = E[OL(θ)]
and quantifies the “value” of getting more information to reduce uncertainty
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 24 / 27
Summarising PSA
0 10000 20000 30000 40000 50000
010
0000
2000
0030
0000
4000
00
Expected Value of Information
Willingness to pay
EV
PI
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 24 / 27
Conclusions
• Bayesian modelling allows the incorporation of external, additionalinformation to the current analysis
• This can come in the form of
– Previous studies– Elicitation of expert opinions
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 25 / 27
Conclusions
• Bayesian modelling allows the incorporation of external, additionalinformation to the current analysis
• This can come in the form of
– Previous studies– Elicitation of expert opinions
• In general, Bayesian models are more flexible and allow the inclusion ofcomplex relationships between variables and parameters
– This is particularly effective in decision-models, where information fromdifferent sources may be combined into a single framework
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 25 / 27
Conclusions
• Bayesian modelling allows the incorporation of external, additionalinformation to the current analysis
• This can come in the form of
– Previous studies– Elicitation of expert opinions
• In general, Bayesian models are more flexible and allow the inclusion ofcomplex relationships between variables and parameters
– This is particularly effective in decision-models, where information fromdifferent sources may be combined into a single framework
• Using MCMC methods, it is possible to produce the results in terms ofsimulations from the posterior distributions
• These can be used to build suitable variables of cost and benefit
– Particularly effective for running probabilistic sensitivity analysis
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 25 / 27
More info (and shameless marketing)
• It is possible to standardise the economic analysisderived from the output of a Bayesian model, forexample using the R package BCEA
• BCEA features heavily in the brilliant, forthcomingbook on Bayesian methods in health economics(written by me )
• In the book, I describe the entire process ofmaking Bayesian analysis in health economics,including pre-processing of the data and runningthe model
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 26 / 27
More info (and shameless marketing)
• It is possible to standardise the economic analysisderived from the output of a Bayesian model, forexample using the R package BCEA
• BCEA features heavily in the brilliant, forthcomingbook on Bayesian methods in health economics(written by me )
• In the book, I describe the entire process ofmaking Bayesian analysis in health economics,including pre-processing of the data and runningthe model
• More info is available at the webpageswww.statistica.it/gianluca/BMHE andwww.statistica.it/gianluca/BCEA
• Also, some discussion (and more to come) in afew posts on gianlubaio.blogspot.co.uk
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 26 / 27
Thank you!
Gianluca Baio ( UCL) Metodi Bayesiani in Farmacoeconomia Torino, 5 Ottobre 2012 27 / 27