190 likes | 422 Views
Medical and Pharmaceutical Statistics Research Unit. Meeting on Futility Analysis London, 11 November 2008 Simple approaches to futility analysis John Whitehead. Medical and Pharmaceutical Statistics Research Unit MPS Research Unit
E N D
Medical and Pharmaceutical Statistics Research Unit Meeting on Futility Analysis London, 11 November 2008 Simple approaches to futility analysis John Whitehead Medical and Pharmaceutical Statistics Research Unit MPS Research Unit Director: Professor John Whitehead Department of Mathematics and Statistics Tel: +44 1524 592350 Fylde College Fax: +44 1524 592681 Lancaster University E-mail: j.whitehead@lancaster.ac.uk Lancaster LA1 4YF, UK
Example: A study in stroke Patients: Have suffered an ischaemic stroke no more than 6 hours earlier Treatments: E: Experimental drug, administered for 5 days C: Placebo Primary Modified Rankin score after 90 days Response: SUCCESS = score of 0 or 1 FAILURE = score of 2 – 6 Success rates on E and C are pE and pC Futility meeting
Intended analysis The data can be summarised as and will be analysed using Pearson’s c2-test: Futility meeting
It can be shown that where and To a good level of approximation, Z ~ N(qV, V) where q is the log-odds ratio Futility meeting
Power requirement E will be claimed better than C if Z u where and and qR represents a clinically worthwhile improvement Thus u and V must satisfy and Futility meeting
This leads to where zg is the 100g percentage point of N(0, 1), and to where Futility meeting
Suppose that a = 0.05, 1 – b = 0.90 and pC = 0.45 Attainment of pE = 0.55 would be clinically worthwhile Then qR = 0.40 and A sample size of 1052 could be adopted (V = 65.75) Run the trial until 1052 patients have been recruited, treated and followed up to 90 days Futility meeting
The one-stage design claim E > C Z 15.893 V 65.75 Futility meeting
The two-stage design claim E > C Z u1 u2 continue V1 V2 V 1 abandon Futility meeting
The futility design Z claim E > C u2 continue V1 V2 V 1 abandon Futility meeting
Power requirement E will be claimed better than C if Z11and Z2u2 where and Futility meeting
Let F2 denote the bivariate standard normal distribution function where Then F2 is the PROBBNRM function of SAS, and so can easily be evaluated Futility meeting
Thus 1, u2, V1 and V2 must satisfy and There are 2 equations and 4 unknowns, so some constraints can be imposed Let us require that : futility will be assessed when half of the information is available Futility meeting
Some feasible designs -u2 = 15.890 in all cases 1 n2 V2a 1 - b fut0 fut1 cpower -2.0 1052 65.75 0.0249 0.902 0.364 0.0040 0.206 -1.5 1052 65.75 0.0249 0.901 0.397 0.0052 0.232 -1.0 1052 65.75 0.0249 0.901 0.431 0.0067 0.260 -0.5 1052 65.75 0.0248 0.901 0.465 0.0085 0.289 0.0 1054 65.88 0.0248 0.901 0.500 0.0106 0.321 0.5 1056 66.00 0.0248 0.901 0.535 0.0133 0.354 1.0 1058 66.13 0.0248 0.900 0.569 0.0164 0.389 1.5 1062 66.38 0.0249 0.900 0.603 0.0201 0.426 2.0 1066 66.63 0.0248 0.900 0.636 0.0244 0.464 Futility meeting
fut0 = P(Z1 1 q = 0); fut1 = P(Z1 1 q = 0.40) u2 = 15.890 for all designs, compared with u = 15.893 for a one-stage design Search is to three decimal places in u2and to the nearest even integer in n2 No design exists for 1= 2.5 Futility meeting
cpower is the conditional power: P(Z2 u2 Z1= 1; q = 0.40) So, when q = qR: fut1is the probability of falling into the hole (of false stopping) cpoweris the probability of getting out of the hole fut1 cpoweris the probability of falling into the hole and then of getting out of it again (loss of power) For 1 = 0.0, fut1 = 0.0106, cpower = 0.321 and fut1 cpower = 0.0034 Futility meeting
Preferred design: 1 u2 n2 V2a 1 - b fut0 fut1 0.0 15.890 1054 65.88 0.0248 0.901 0.500 0.0106 Only two extra patients needed 50% chance of stopping if no effect Very simple futility criterion -stop if the treatment isn’t working Futility meeting
Recommended design: 1 u2 n2 V2a 1 - b fut0 fut1 0.0 15.893 1052 65.75 0.0247 0.8999 0.500 0.0107 This is the one-stage design with an added futility look Tiny loss of power Avoids misunderstanding and difficulties with regulators Analyse as if there was no futility analysis - conservative Futility meeting
General recommendations 1. In lengthy trials in serious conditions, perform an interim analysis half way through 2. Abandon the study if the estimated treatment effect is negative 3. Ignore the futility rule in the final analysis 4. Design the futility rule into the protocol, and consider absolute properties, not conditional ones 5. You can adjust for prognostic factors and deal with complicated endpoints 6. Also useful when there are multiple active treatments 7. Can check through exact calculation or simulation Futility meeting