Bayesian Approaches to Clinical Trials and Health-Care Evaluation

• David J. Spiegelhalter
• Keith R. Abrams
• Jonathan P. Myles

---

Errors and Omissions

Many apologies for all of these and we hope they have not caused too much chaos and confusion. Last updated December 2008

• Page 24, line 20: there should be an m in the numerator, so the expression reads exp[-(y-theta)^2 m / 2 sigma^2]
• Page 33, lines 10 and 11: ‘30%’ should be ‘70%’, ‘330’ should be ‘142’, ‘165’ should be ’71’
• Page 36, formula (2.45), the power of (1-y) should be b-1, not a-1
• Page 41, line -2. `0.773′ should be `0.675′.
• Page 43, Caption to Figure 2.13. `(b)’ should be `(a)’.
• Page 62, line 3. `b+20-5′ should be `b+20-15′.
• Page 62, last formula. In this expression both occurrences of theta_0 should be mu.
• Page 72, Legend to Figure 3.7, line 2. `centred on 0′ should be `centred on 1′.
• Page 82, line 13. “‘the probability that the event will occur at the next trial is (m+1)/(m+2)”‘, NOT m/(m+1) as for some reason was originally written
• Page 86, line 1. `Lindley (1975)’ should be `Lindley (1985)’
• Page 93, line 16. `trial’ should be `trials’
• Page 109, Figure 3.18: the Doodle should have no arrow between n and P.crit, a double arrow pointing from y.pred to P.crit, a rectangular node labelled n.crit, and a double arrow from n.crit to P.crit. (In fact double arrows pointing into logical nodes are ignored when writing WinBUGS code from Doodles, and only the explicit expression given for the logical node is used).
• Page 109, first line of WinBUGS code: the file name should be drug-dat.txt and not data.txt
• Page 145, Figure 5.1. The heading on the right-hand-side should be ‘CHART better than standard by %’
• Page 164, Example 5.4.,line -8. `0.047′ should be `0.049′
• Page 169, line 14. `1.09′ should be `0.954′.
• Page 169, Table 5.2. line 3 of legend and heading of third column: `1.09′ should `0.954′. This influences all the entries in the third column, which should now read ‘1.00, 1.10 , 1.21, 1.33, 1.46, 1.61, 1.77, 1.95, 2.14, 2.36, 2.60, 4.18, 6.74’
• Page 174, line 13. `analagous’ should be `analogous’
• Page 208, Example 6.6. the tail areas given in Figures 6.7 and 6.8 will not match an approximate analysis derived from the number of events recorded in Tables 6.3 and 6.4.
• Page 235, para 2, line 8. `odds ratio of 0.55 (95% interval from 0.39 to 0.77)’ should be `odds ratio of 0.34 (95% interval from 0.18 to 0.62)’
• Page 240, Table 6.12., last line. The formula should be exp(mu_sigma+tau^2_sigma/2), with estimate 3.17 (95% interval 1.44 to 12.00). The corrected models are on the website.
• Page 261, line 5. `effected’ should be `affected’
• Page 279, Example 8.3. The expressions for the Poisson means should both have the left-hand-side enclosed in log( ).
• Page 284, Table 8.5. There is a missing result: in Study 5, treatment C(j=3), there should be an entry m=32, Mean = 7.78, SD = 6.78. The WinBUGS code includes this data.
• Page 285, Example 8.4. The first entry in Table 8.6 should be labelled mu_phi, not mu_delta. See below for revised Table 8.6
• Page 337, line 7. `inevitable’ should be `inevitably’
• Page 337, line 8. `preferably’ should be `preferable’

The results in Table 8.6 were derived from a different formulation to that described in the text, using a model in which the control mean was unconstrained and the random-effects mean set to 0. The correct table for the model described in the text (as obtained from the code on the website) is as follows:

node	median	sd	2.5%	97.5%
mu.phi	4.039	0.4959	3.08	4.983
theta_1 (A)	5.329	0.8712	3.661	7.081
theta_2 (B)	2.098	0.9094	0.2181	3.79
theta_3 (C)	2.865	1.136	0.6679	5.158
AvB	3.231	1.154	1.088	5.645
sigma	8.182	0.2139	7.792	8.627
tau.phi	0.4455	0.4919	0.01582	1.761