![[Photo]](DSC.gif)
I am a Reader in Statistics in the School of Mathematical Sciences at Queen Mary, University of London. On this page, you will find information on the following:
1. Teaching for 2009-10.
2. Recent research.
4. Publications.
5. Recent talks, both seminars and conference presentations.
6. Research projects for PhD or MPhil students.
7. Research grants for 2008-09.
8. Administrative responsibilities for 2009-10.
10. Contact details.
In 2009-10, I give one lecture course. A brief syllabus for this is given below. I also give the course Topics in the Design of Experiments jointly with B. Bogacka at the London Taught Course Centre. The web page for this course is available here.
Time Series (Semester 5): This is a third-year undergraduate course for majors in Mathematics/Statistics.
Trends and seasonality: time plots; smoothing by moving averages; differencing. Random variables: expectation; conditional distributions and independence; covariance and correlation; multivariate normal distribution. Stationary time series models: weak and strict stationarity; autocovariance and autocorrelation functions; moving average (AR) and autoregressive (AR) models. Estimation of the mean and autocovariance function: approximate confidence bounds. ARMA models: causality and invertibility; autocorrelation and partial autocorrelation functions; forecasting. Autoregressive integrated moving average (ARIMA) models: model fitting; seasonal ARIMA models.
The course material for Autumn Semester 2009 is available on the Time Series web page.
My current research is mainly in the area of sequential analysis, with particular emphasis on medical applications. The main topics that I have recently been working on are briefly described below.
1. Inference following Sequentially Designed Experiments.
My current research in this area is concerned with the construction of corrected confidence sets for an adaptive normal nonlinear model. There are many examples of such models in chemometrics, such as the Michaelis-Menten model and the first-order growth or decay model. With these models, the design points are chosen sequentially based on the previous data, which complicates the analysis. My research is joint work with M.B. Woodroofe at the University of Michigan and builds on our earlier papers on adaptive normal linear models.
2. Sequential Procedures for Multi-Armed Clinical Trials.
When there are more than two treatments being compared in a clinical trial, the use of a sequential procedure can sometimes require substantially fewer patients than a fixed-sample design to achieve the same error probabilities. My most recent research in this area, which is joint work with A. Biswas at the Indian Statistical Institute, is concerned with the development of a general elimination rule for comparing several treatments with responses which are multivariate, continuous and dependent on prognostic factors.
3. Response-Adaptive Designs in Clinical Trials.
These designs use the accumulating data in a clinical trial to skew the allocation probabilities in favour of the treatment which is performing better thus far in the trial. The simplest such designs, from a mathematical point of view, are adaptive urn designs and my most recent research in this area, which is joint work with A. Ivanova at the University of North Carolina, addresses the problem of bias following such a design. Current work includes the use of a stopping time and the consideration of more than two treatments.
4. Testing for the Number of Components in Mixture Models.
The determination of the number of components in a finite mixture distribution is an important, but difficult, problem. A number of approaches have been proposed in the literature for tackling the case of a normal mixture, such as the use of posterior Bayes factors and bootstrapping. In joint work with M.N. Goria at the University of Trento, a detailed comparison is being carried out of these methods for the normal case and we then plan to develop analogous approches for the determination of the number of components in a gamma mixture.
5. Inference for Secondary Parameters following Sequential Tests.
When carrying out estimation following a sequential clinical trial, methods are available for constructing corrected confidence intervals for primary parameters. However, in practice, there is often also interest in secondary parameters. In joint work with R.C. Weng at the National Chengchi University, corrected confidence intervals are being developed for secondary parameters. This work builds on existing work for primary parameters and complements recent work in the literature based on related techniques.
Barbáchano, Y. (2007): "Adaptive designs for clinical trials which adjust for imbalances in prognostic factors". DPhil.
Bailey, S.M. (2007): "Sequential adaptive designs for early phase clinical trials". DPhil.
Halimeh, A.A. (2004): "Sequential procedures for comparing several normal means". MPhil.
Morgan, C.C. (2003): "Group-sequential response-adaptive designs for clinical trials". DPhil.
Most of my current preprints are available as Research Reports. Please contact me for paper copies. The abstracts are given below.
Discussion on "A hybrid selection and testing procedure with curtailment for comparative clinical trials" by E.M. Buzaianu and P. Chen. Sequential Anal. 28, 26-29 (2009).
This discussion contains some comments on the paper by Professors Buzaianu and Chen. These concern the form of the two-stage design used and possible future work, such as the consideration of alternative stopping rules, the incorporation of response-adaptive sampling and the problem of estimation following the design.
"Response adaptive randomisation". In Encyclopedia of Clinical Trials, Volume 4, eds. R. D'Agostino, L. Sullivan and J. Massaro, pp. 113-119 (2008). New York: Wiley.
When two or more treatments are being compared in a clinical trial, information accrues on the treatments as the trial proceeds. However, it may become quite clear early on which treatments are more promising than others, and consequently there may be interest in applying response adaptive randomisation, which uses the current data to assign more patients to the treatments which are performing better thus far in the trial. In this way, patients have a higher probability of being assigned to these treatments. Examples of this type of randomisation are given, such as urn models and sequential maximum likelihood estimation rules, and some of their properties are described. The issue of inference following trials which use response adaptive randomisation is then discussed. For example, this approach can lead to a loss in power when using standard tests and the problem of estimation can be considerably more complicated than in a trial which uses complete randomisation. The use of stopping rules is also addressed, since, in practice, there may be interest in stopping a trial early if there is convincing evidence of a treatment difference or less promising treatments may be dropped from further consideration. An indication is given of possible future work.
"Inference following designs which adjust for imbalances in prognostic factors" (with Yolanda Barbáchano). Under revision for Clin. Trials.
When minimisation is used to balance treatment groups across prognostic factors, a problem arises at the time of analysing the results. Since minimisation is essentially a deterministic method, any statistical test that uses the assumption of randomisation cannot be used in the analysis. Previous papers have shown that analysis of variance does not produce valid results and that analysis of covariance should be used instead. In this paper, these results are extended to trials with more prognostic factors and more treatments. An alternative design to minimisation which makes use of optimum design theory is also considered, with two choices of biased coin. Simulation shows that these designs produce more extreme results than minimisation when analysis of variance is used, and are sometimes more powerful than minimisation when analysis of covariance is employed. Confidence intervals for the difference between the treatment mean effects are then calculated, and the adaptive methods based on optimum design theory are shown to be slightly conservative when the prognostic factors are taken into account. The possibility of treatment-covariate interactions is also addressed.
Report 2004-04: "A novel bivariate design for a combined phase I/II clinical trial in oncology" (with S.M. Bailey).
This paper reviews the current state of research on bivariate adaptive designs for early phase clinical trials with two competing outcomes, often known as combined phase I/II trials. Such trials aim to find the maximum tolerated dose (MTD) of a new drug, whilst assuring that the efficacious effects of the drug remain above a given level. A new Bayesian design is presented addressing shortfalls of previous designs of this type for these combined trials. The main features of the design are the use of separate response curves for toxicity and efficacy, the modelling of the joint events of toxicity and efficacy using the bivariate Gumbel model, and the use of Bayes' theorem in finding accurate estimates of the required MTD and probabilities of toxicities and efficacious responses. It is shown via simulation that the new design targets the doses most closely associated with the MTD, whose efficacious effects are above a given threshold, with high probability. A description of the techniques used within the simulation structure is also outlined. Comparisons with two recently proposed bivariate designs show that the new design performs favourably with regard to MTD targeting and patient allocation within the trials.
"Predictability of designs which adjust for imbalances in prognostic factors" (with Y. Barbáchano and D.R. Robinson). J. Statist. Plann. Inf. 138, 756-767 (2008).
"The duplicate method of uncertainty estimation: Are eight targets enough?" (with J.A. Lyn, M.H. Ramsey, A.P. Damant, R. Wood and K.A. Boon). Analyst 132, 1147-1152 (2007).
"A comparison of adaptive allocation rules for group-sequential binary response clinical trials" (with C.C. Morgan). Statist. Med. 26, 1937-1954 (2007).
"Corrected confidence intervals for secondary parameters following sequential tests" (with R.C. Weng). In Recent Developments in Nonparametric Inference and Probability: Festschrift for Michael Woodroofe, eds. J. Sun, A. DasGupta, V. Melfi and C. Page, pp. 80-104 (2006). Hayward, California: Institute of Mathematical Statistics.
"Sequential procedures for comparing several normal means" (with A.A. Halimeh). J. Statist. Comput. Simul. 76, 519-537 (2006).
"Sequential urn designs with elimination for comparing K ≥ 3 treatments" (with A. Ivanova). Statist. Med. 24 , 1995-2009 (2005).
"The use of the triangular test with response-adaptive treatment allocation" (with A. Ivanova). Statist. Med. 24 , 1483-1493 (2005).
"A general multi-treatment adaptive design for multivariate responses" (with A. Biswas). Sequential Anal. 24, 139-158 (2005).
"Corrected confidence intervals for adaptive nonlinear regression models" (with M.B. Woodroofe). J. Statist. Plann. Inf. 130, 63-83 (2005).
"Bias calculations for adaptive urn designs" (with A. Ivanova). Sequential Anal. 20, 91-116 (2001).
"Corrected confidence intervals following a sequential adaptive clinical trial with binary responses" (with Z. Govindarajulu). J. Statist. Plann. Inf. 91, 53-64 (2000).
"Corrected confidence sets for sequentially designed experiments: Examples" (with M. Woodroofe). In Multivariate Analysis, Design of of Experiments and Survey Sampling: A Tribute to Jagdish N. Srivastava, ed. S. Ghosh, pp. 135-161 (1999). New York: Marcel Dekker. Reprinted in Sequential Anal. 21, 191-218 (2002).
"A comparison of the randomised play-the-winner rule and the triangular test for clinical trials with binary responses" (with W.F. Rosenberger). Statist. Med. 18, 761-769 (1999).
"Approximate bias calculations for sequentially designed experiments" (with M.B. Woodroofe). Sequential Anal. 17, 1-31 (1998).
"Approximate confidence intervals after a sequential clinical trial comparing two exponential survival curves with censoring" (with M.B. Woodroofe). J. Statist. Plann. Inf. 63, 79-96 (1997).
"Corrected confidence sets for sequentially designed experiments" (with M. Woodroofe). Statist. Sinica 7, 53-74 (1997).
"Corrected confidence intervals after sequential testing with applications to survival analysis" (with M.B. Woodroofe). Biometrika 83, 763-777 (1996).
"Sequential allocation involving several treatments". In Adaptive Designs, eds. N. Flournoy and W.F. Rosenberger, pp. 95-109 (1995). Hayward, California: Institute of Mathematical Statistics.
"Sequential estimation for two-stage and three-stage clinical trials". J. Statist. Plann. Inf. 43, 343-351 (1994).
"Estimation following sequential tests involving data-dependent treatment allocation". Statist. Sinica 4, 693-700 (1994).
"Sequential tests with covariates with an application to censored survival data". Commun. Statist. - Theor. Meth. 23, 277-287 (1994).
"Sequential procedures for comparing several medical treatments" (with J.A. Bather). Sequential Anal. 11, 339-376 (1992).
"Some results on estimation for two-stage clinical trials". Sequential Anal. 11, 299-311 (1992).
"A comparative study of some data-dependent allocation rules for Bernoulli data". J. Statist. Comput. Simul. 40, 219-231 (1992).
"Sequential estimation with data-dependent allocation and time trends". Sequential Anal. 10, 91-97 (1991).
"Sequential tests for an unstable response variable". Biometrika 78, 113-121 (1991).
Discussion on "Second guessing clinical trial designs" by J.J. Shuster and M.N. Chang. Sequential Anal. 27, 21-23 (2008).
"Sequential testing". In Encyclopedia of Statistics in Behavioral Science, eds. B.S. Everitt and D.C. Howell, pp. 1819-1820 (2005). Chichester: Wiley.
Comment on "Randomised urn models and sequential design" by W.F. Rosenberger (with C.C. Morgan). Sequential Anal. 21, 29-32 (2002).
Comment on "Statistical and ethical issues in monitoring clinical trials" by S.J. Pocock. Statist. Med. 12, 1473 (1993).
Comment on "Investigating therapies of potentially great benefit: ECMO" by J.H. Ware (with P. Armitage). Statist. Sci. 4, 322-323 (1989).
Review of Mathematical Statistics: Basic Ideas and Selected Topics, Volume I, 2nd edition, by P.J. Bickel and K.A. Doksum. Biometrics 58, 691-692 (2002).
Review of Asymptotic Statistics by A.W. van der Vaart. Biometrics 57, 645-646 (2001).
Review of Optimal Sequentially Planned Decision Procedures by N. Schmitz. J. R. Statist. Soc. A 156, 511 (1993).
Review of Applied Multivariate Analysis by B.S. Everitt and G. Dunn. The Statistician 42, 325-326 (1993).
My most recent talks are listed below. In March, I am an invited speaker at the One-Day Briefing on Advanced Statistical Techniques for Clinical Trials in London, and, in June, I am invited speaker at the 9th Model-Oriented Data Analysis and Optimum Design Workshop at the University of Bologna in Italy.
"Corrected confidence intervals based on the signed root transformation for multi-parameter sequentially designed experiments". University of Southampton (April 2009).
"Predictability of designs which adjust for imbalances in prognostic factors". University of Warwick (April 2009).
"Inference following designs which adjust for imbalances in prognostic factors". University of Reading (February 2008).
"Estimation following adaptively randomised clinical trials". Workshop on Statistical Methods for Learning in Clinical Research, Bologna, Italy (June 2009).
"The use of group sequential tests with designs with adjust for imbalances in prognostic factors". Second International Workshop in Sequential Methodologies, Troyes, France (June 2009).
"Predictability of designs which adjust for imbalances in prognostic factors". International Workshop on Applied Probability, Compiègne, France (July 2008).
Here are details of two possible projects. As mentioned earlier, others are possible, and anyone interested is welcome to contact me. For general information about postgraduate research in Statistics and Probability in the School of Mathematical Sciences at Queen Mary, University of London, please look at our Postgraduate Admissions web page.
Most of the existing theoretical work in sequential analysis is concerned with comparisons of two treatments. In practice, there may be interest in designing a sequential clinical trial to compare several treatments in which there is group-sequential monitoring of the data, the possibility of eliminating less promising treatments and the use of response-adaptive treatment allocation. Such general designs present many interesting and difficult theoretical problems. How can the stopping boundaries be constructed for use at the interim analyses? What is the best way to allocate patients to treatments? The idea of the project would be to explore some of these problems. A good starting point would be to study the group-sequential response-adaptive design developed by Jennison and Turnbull (2001, Sequential Anal. 20, 209-234), and further studied by Morgan (2003, Control. Clin. Trials 24, 523-543), for comparing two normal means when the variances are known. The next step would be to consider a multi-treatment extension of this procedure.
During the course of the project, it would be necessary to learn about theoretical techniques in sequential analysis and to gain experience of Monte Carlo methods.
There are often a large number of hypotheses of interest in gene association analysis, but limited numbers of observations available. Consequently, multiple testing procedures are sought which make the most efficient use of the data whilst controlling the false discovery rate. Most of the proposed procedures are based on one-stage designs, which often lead to tests with poor power, since there is only a small number of observations for each hypothesis. Zehetmayer, Bauer and Posch (2005, Bioinformatics 21, 3771-3777) showed that, for normally distributed data, a two-stage design based on combining the p-values from a screening stage and a testing stage can significantly improve the power. However, there are a number of open problems. Can the power of these designs be further improved by allowing unequal allocation during the testing stage? Is there a worthwhile improvement in power if three stages are used? Can analogous designs be developed for other response types, such as binary data? A good starting point would be to study the effect of unequal allocation on a one-stage design.
During the course of the project, it would be necessary to learn about adaptive testing based on p-values and to gain experience of Monte Carlo methods.
During 2008-09, I was awarded a joint research grant with Y. Zhou at the University of Reading by the Medical Research Council (MRC). Details of this are given below.
This is a Fast Tracking Drug Development Initiative Grant to hold a three-day workshop at the University of Reading. More information about the workshop is available here. The project deliverables are as follows:
1. To provide an intensive programme of talks from world experts on current trends in dose-finding methodology for early phase clinical trials, together with small-group discussions, where open problems can be addressed and progress made.
2. Participants will be able to learn about state-of-the-art adaptive designs for early phase clinical trials and related ethical issues, exchange research ideas, establish collaborative research links and discuss how to implement the methodology in practice.
3. To provide a report on the outcomes, which will highlight the main issues raised on each of the topics covered and summarise the current status of any open problems, and thus emphasise researchable areas of interest and important methodological gaps.
4. By exchanging research ideas and establishing collaborative links, the workshop will lead to joint grant applications for funding, preparation of papers for presentation at future international conferences and publication in relevant leading statistical journals.
I am currently a member of the Editorial Board for Sequential Analysis and an Associate Editor for the Journal of Statistical Planning and Inference.
For the School of Mathematical Sciences, I am the Undergraduate Admissions Tutor and the Deputy Director of Statistics.
I have a BSc degree from Portsmouth Polytechnic, and MSc and DPhil degrees from the University of Oxford. I came to my present job at Queen Mary in 2005, and immediately before that was a Senior Lecturer in Statistics at the University of Sussex. During 1993-95, I was a Visiting Assistant Professor at the University of Michigan and was awarded a Fulbright Scholarship Grant in connection with the visit. Since 1993, I have been a Chartered Statisticican, and, since 1998, a Chartered Mathematician. I am a Fellow of the Royal Statistical Society and of the Institute of Mathematics and its Applications, and a Member of the Biometric Society, the International Society for Clinical Biostatistics and the Bernoulli Society.
One of my favourite pastimes is travelling. During July and August 2008, I made a three-week trip to Thailand and Laos. One of the highlights was a visit to Sala Kaew Ku near Nong Khai, which is a surreal park full of cement statues of every deity imaginable. There is an earlier, similar park near Vientiane. Other high points included a visit to beautiful Wat Si Saket with its 6,400 Buddha images in Vientiane, and temple-hopping and relaxing longtail boat rides along the Mekong River in the former royal capital of Luang Prabang. A panoramic view of this colourful and atmospheric city can be obtained by climbing to the summit of Phu Si. One of my other pastimes is learning Thai. I am currently living in Hove.
The contents of this Home Page are my own responsibility, not those of Queen Mary, University of London, or any of its Units.
[School of Mathematical Sciences] [Queen Mary, University of London]