THE NUMBER OF CENTERS IN A MULTICENTER CLINICAL STUDY: EFFECTS ON STATISTICAL POWER
|
|
- Gary Davis
- 5 years ago
- Views:
Transcription
1 Drug Information Journal, Vol. 34, pp , /2000 Printed in the USA. All rights reserved. Copyright 2000 Drug Information Association Inc. THE NUMBER OF CENTERS IN A MULTICENTER CLINICAL STUDY: EFFECTS ON STATISTICAL POWER ZHENGNING LIN, PHD Senior Biostatistician, Medical Information Processing and Statistics, Novartis Pharmaceuticals Corporation, East Hanover, New Jersey In the pharmaceutical industry it is often important to ensure fast patient enrollment. A common practice to speed up patient enrollment is to use more clinical centers. One of the concerns regarding this practice is that increasing the total number of centers to a certain degree may decrease the statistical efficiency of treatment comparisons. This is mainly because a study with too many clinical centers usually includes quite a few small centers and very often they do not have enough patients to represent all treatment groups. These small centers often carry little information on treatment differences. This paper utilizes a statistical model to quantify the relationship between statistical efficiency and the number of centers under typical clinical trial settings. Results in this paper provide useful statistical knowledge in choosing the number of clinical centers when planning multicenter studies. Key Words: Clinical trial; Multicenter; Number of clinical centers; Center size; Statistical power INTRODUCTION trial conduct at all study centers which may affect the study quality as more and more IN THE PHARMACEUTICAL industry it centers join the study. The approach in this is often important to ensure fast patient enpaper focuses on the following statistical perrollment. A common practice to speed up spective: How much, if any, statistical effipatient enrollment is to use more clinical cenciency is affected by the increase of study ters. One concern about this practice is centers, while the study conduct and the whether there is any statistical penalty assooverall quality of study centers remains unciated with the increase of the total number changed? I use a statistical model to quantify of study centers. What needs to be answered the relationship between the number of cenis how many study centers we can use in a ters and the statistical efficiency under typiclinical study without losing statistical effical clinical trial settings. Statistical efficiency ciency. Actually, there is no simple and genis discussed in terms of actual statistical eral answer that is appropriate for all studies, power. Results in this paper provide useful because it depends on many aspects of the statistical knowledge regarding how many centers we may use when planning a multicenter clinical study. For simplicity of expla- Reprint address: Zhengning Lin, Senior Biostatistician, nation, the discussion focuses on multicenter Medical Information Processing and Statistics, Novartis studies with the objective of comparing two Pharmaceuticals Corporation, 59 Route 10, East Hantreatments, while the conclusions are appliover, NJ zhengning.lin@pharma.novartis.com. cable to clinical studies with multiple parallel 379
2 380 Zhengning Lin treatment groups. This paper assumes that where n A and n B are the total number of patients both treatment and center effects are controlled for A and B, respectively, n Ai (or n Bi ) using a two-way (treatment-by-cen- is the number of patients in Group A (or B) at ter) analysis of variance (ANOVA) model. center i, and c is the total number of centers. Following recommendations in recent publi- Suppose that the treatment effect δ is one. cations, including those by Lin (1), Senn For 80% power we need a total of 34 patients (2,3), and Källén (4), this paper just considers based on standard (or commonly used) sample the weighted analysis of the ANOVA model size calculation. The maximum power to assess treatment differences. based on 34 patients is 80.7% when n A = n B It is well known that with the same number = 17 in a one-center study. Suppose that the of total patients, treatment comparison is block size for randomization is two per treat- most efficient when the number of patients ment group, and only one center is used. The is balanced between the two treatment imbalance of treatment allocation is controlled groups. For multicenter clinical studies this by the block size and the minimum balance is usually made at the center level, power is 80.6% when the larger group has because based on the weighted analysis of 18 patients and the smaller group has 16 pa- the ANOVA model, treatment main effect is tients. The maximum power loss due to the assessed by a weighted average of the treatment imbalance is only 0.1% in this one-center comparisons at individual centers. Al- case. If, however, we use, say, 7 centers in- though blocked randomization is usually stead, and the distribution of the 34 patients applied at study center level to balance treat- among the 7 centers is as shown in Table 1. ment allocations, it does not guarantee the Then the power is only 68.7%, a 12.0% balance in small centers when each of them drop from the maximum of 80.7% with one has few or even less than one block of patients. center. Among the 12.0% power loss, only As a result, studies using too many 1.5% is due to decreased degrees of freedom centers can lose statistical efficiency, or statistical and the other 10.5% is due to the imbalance power, due to this imbalance in a large of treatment allocations. number of small centers. The following example If we increase the number of centers, or illustrates this potential loss of statisti- the block size of the randomization in the cal power. example in Table 1, the potential power loss Consider a multicenter randomized clini- can be dramatically increased. A practical cal study to compare two treatments, A and question is: how often can we see an imbal- B, with response variable X quantitative and ance similar to this and what is the actual approximately normally distributed. Let u A risk of power loss when a study is designed and u B be the mean of X and σ be the standard to include a large number of clinical centers? deviation. Let us also define δ *u A u B */σ In the following, I use a statistical model to as the treatment effect. The statistical power of detecting a statistically significant treatment effect, with two-tailed 0.05 level test, TABLE 1 Distribution of 34 can be estimated as: Patients Among 7 Centers Center i n Ai n Bi Power = Pr t na +n B 2c > t 0.025,(nA +n B 2c) δ n Ai n Bi c i=1 n Ai + n Bi
3 The Number of Centers in a Multicenter Clinical Study 381 quantify the relationship between the number Now let us consider the calculation of the of centers and the actual statistical power of statistical power. In clinical studies, the sta- a study. tistical power is the probability of detecting a statistically significant treatment effect. METHOD Usually the power is calculated based on fixed sample sizes. In multicenter clinical Let us describe the process of patient enrollment studies, the fixed sample sizes are at the study at Center i by a Poisson process with center level for each treatment group. This parameter λ i, i = 1,..., c, where c is the is illustrated in the power calculation formula total number of centers in the study. Then provided in the previous section, where n Ai the probability that a newly recruited patient and n Bi are both fixed values for center i. belongs to Center i is π i =λ i /Σλ j. Different At the designing stage of typical multicenter sites recruit patients with different speeds studies, however, we do not know the sample and, therefore, in general, the λ i s are not sizes at each study center and can only specify equal. Usually, the study centers are not close the total sample sizes of the treatment to each other and, therefore, we can assume groups. Therefore, the power calculation that the λ i s are statistically independent. In conditional on fixed sample sizes at the cen- the following analysis I assume that λ 1,..., ter level is often not realistic. In the following λ c are independent and identically distributed. power calculation, I define the actual statisti- While the common distribution of the cal power as the probability of detecting a λ i s may vary from study to study, one possible, significant treatment effect conditional on and maybe the simplest, distribution for the total sample sizes for the treatment λ i s is the exponential distribution. In this case groups, and the sample sizes at the study the π i s follow D(1,...,1), the Dirichlet dis- center level is not fixed. tribution with all parameters equal to one. The calculations used Monte Carlo simu- To analyze the robustness against the as- lations based on the model assumptions specified sumption of patient distribution among study earlier, with replications for each centers, the analysis also consider the case of power calculation. Statistical significance is equal λ i s. Under this assumption, all centers defined as p < 0.05 using a two-tailed test. I have the same characteristics in recruiting consider the cases when the treatment effect, patients and therefore π 1 =π 2 =...=π c = δ, is 1/1, 1/2, and 1/3, which requires a total 1/c. In either case, the mean of the λ i s does of 17, 64, and 143 patients per group, respectively, not affect the distribution of patients across to achieve 80% power using a standard the centers. It only affects the overall recruit- (or common) sample size calculation ing speeds. Table 2 shows two examples of method. As mentioned earlier, treatment 34 patients distributed across 7 centers based comparison is most efficient when the number on simulation, assuming that the πs either of patients is equal between the two comon follow D(1,..., 1), or they are all equal to paring treatments at each of all centers, and 1/c, and the randomization block size is two in this case the maximum power (with fixed per group. total sample size) is achieved. Although the Note that in (a) the total number of centers sample sizes are powered at 80%, this maxithat contributed study patients is only six, mum power sometimes can be slightly less as Center 2 failed to recruit any patient, a than 80% because of the reduced degrees of situation frequently encountered in multicen- freedom due to increased number of centers ter studies. In practice, most multicenter in the ANOVA model, which the standard studies have center imbalance similar to the sample size calculation does not take into example in (a), while some others are more account. Nevertheless, the power reduction balanced between study centers and are due to loss of degrees of freedom is negligible closer to the example in (b). for studies with large sample sizes.
4 382 Zhengning Lin TABLE 2 Two Examples of 34 Patients Distributed Across 7 Centers (a): Example of s D(1,...,1),34patients, 7 centers Center All In Group A In Group B π (%) (b): Example of i 1/7 (14.3%), 34 patients, 7 centers Center All In Group A In Group B π (%) RESULTS AND CONCLUSIONS Table 3 displays the power calculation results using a randomization block of two per treatment group. The last column, Power Loss, calculates the difference between the maximum power and the actual power, which measures the power reduction due to the imbalance of sample sizes between the two treatments at each of the study centers. From Table 3 it is evident that no matter which assumption we use for the π i s, the power losses due to the imbalance in sample sizes between the two treatments are similar and relatively small, no more than 2% when the block size is two per group. The overall difference between the actual power and the targeted power (80%) is no more than 3%. Is this conclusion true for any block size? For comparison purposes Table 4 and Table 5 list the same results based on block sizes of four and one per group, respectively. Block sizes of more than four per group are not common and are usually unnecessary. Results in Tables 4 and 5 demonstrate that the power loss can be around 4% when the block size is four per group, while the power loss with block size of one per group is around 1.5% or less. Since the power loss is no more than several percentage points, it can be controlled by increasing the sample size to make up the TABLE 3 Actual Powers with Block Size Two per Group Actual power (%) Power loss a (%) Average n A + n B size per Maximum πs πs δ c group power D(1,..,1) π i 1/c D(1,..,1) π i 1/c / / / a Power loss due to imbalance within centers = Max power - Actual power
5 The Number of Centers in a Multicenter Clinical Study 383 TABLE 4 Actual Powers with Block Size Four per Group Actual power (%) Power loss a (%) Average n A + n B size per Maximum πs πs δ c group power D(1,..,1) π I 1/c D(1,..,1) π i 1/c b 0.1 1/ / / a Power loss due to imbalance within centers = Max power - Actual power b Negative number due to approximation is rounded to 0.0 power loss. For example, the actual power for the case of 64 patients per arm with 27 centers and a block size of four per group (Table 5) is 75.5 with πs D(1,..,1) while the sample size is based on 80% power using standard sample size calculation. To make up the 5% deficit a sample size of 73 per arm is needed to have 85% power using the standard sample size calculation. If the same average sample size per center is kept, the total number of centers becomes 31. The actual power with this D(1,..,1) assumption is 81.0%, which is sufficient for the 80% requirement. Since the power calculation re- sults are similar no matter which assumption we make for the π i s, we only need to pay attention to the average number of patients per center, and also the block size, in determining the extra number of patients needed for the study to make up the power loss. This can be done easily in practice by increasing the specified power in the standard sample size calculation using the above tables for reference. The above calculations are all based on sample sizes with 80% power. Is the power loss pattern similar for higher-powered studies? In fact, for studies with the same sample TABLE 5 Actual Powers with Block size One per Group Actual power (%) Power loss a (%) Average n A + n B size per Maximum πs πs δ c group power D(1,..,1) π i 1/c D(1,..,1) π i 1/c b 1/ / b 0.0 b 1/ a Power loss due to imbalance within centers = Max power - Actual power b Negative number due to approximation is rounded to 0.0
6 384 Zhengning Lin size, a higher power corresponds to a larger using a relatively large number of centers. δ, that is, a larger treatment effect. For example, Understanding this fact is important in deteris a study with 34 total patients (n A + n B ) mining the sample sizes for studies with powered at 80% for δ=1, and it is powered many study centers. For most Phase III stud- at 90% for δ=1.15. More calculations re- ies we may want to ensure that studies are vealed that for the same sample sizes, studies adequately powered and, therefore, the sam- with larger δs (compared to those for 80% ple sizes should not be calculated by the standard power) tend to have smaller power losses. As method without adjusting for the power an illustration, when δ=1.15, the maximum loss. Although in most cases there is no accurate power for a study with 34 total patients and way of knowing the exact statistical 7 centers is 89.0%. The actual power with power, at least this paper quantified one πs (D(1,..,1) and block size of four per source of power loss and provided a tool to group is 86.3%. Therefore, the power loss adjust the sample size. is 89.0% 86.3% = 2.7%, smaller than the Results in this paper by no means suggest power loss of 3.6% when the study is pow- that no matter how many centers we use, ered at 80% (δ=1, see Table 4). This interesting the reduction of statistical power is always finding is mathematically explained in limited by a couple of percentages. Actually, Appendix 1. even if we only increase the number of cen- From the above we can conclude the fol- ters to, say, 10, for the example of n A + n B = lowing: In general the actual power of a 34 and block size of 4 per group, the actual multicenter study can be up to 5% lower than power would reduce to 69.8% for π i 1/10. what the study is targeted for. The power In this case the maximum power is 77.3% loss caused by the imbalance of sample sizes and the power loss due to the imbalance is within study centers: already a substantial amount of 7.5%. It is not common, however, to have average center Increases with the block size, sizes less than two per group and also it Increases with the number of centers, is usually not desirable. Is closely related to the average center size, Results in this paper also indicate that, in as opposed to the total sample size, general, there is limited statistical penalty to Is around 2% or less when the block size pay if we use as many centers as logistically is no more than two per group, desirable. The advantage of speeding up the Is negligible (<1%) when the average center patient enrollment and exposing the investigroup, size is at least seven per treatment gating drug to more physicians and clinical centers is often well worth the little extra Is less than 4% when the average center cost of slightly increasing the total number size is at least 2.4 per treatment group, and of patients to make up the power loss. The Can be made up by increasing the power statistical model in this paper, however, is used in the standard sample size calculation. based on the assumption that the treatment effect (δ) at each study center is independent of the center size. Although this assumption Note that the results based on the two ex- is used in virtually all parametric statistical treme scenario, πs D(1,..,1) and π i 1/ c, are similar and therefore, these conclusions are not scenario dependent. DISCUSSION not have the opportunity to pass this learning curve. Also, clinical centers serving larger patient populations often are more experienced than those serving smaller patient pop- This paper demonstrates that standard power calculation overestimates the actual power by a few percentage points ( 5%) for studies analyses for controlled multicenter clinical studies, it often can be incorrect in practice. For studies using new treatment assessment procedures, there is often a learning curve for each clinical center and small centers do
7 The Number of Centers in a Multicenter Clinical Study 385 Acknowledgments I would like to thank an anonymous referee for helpful comments and suggestions. REFERENCES ulations. In these cases, larger centers may provide more reliable data than smaller ones, although there are always exceptions. These issues may be too complicated to be incorporated in statistical analyses. The best way to handle them is to select and monitor all study centers carefully to assure the quality of the clinical data. For studies using well-estab- lished treatment assessment procedures, the model assumption in this paper can be appropriate when study centers are carefully selected and monitored. 1. Lin Z. An issue of statistical analysis in controlled multi-centre studies: How shall we weight the centers? Stat Med. 1999; 18: Senn S. Some controversies in planning and analyzing multi-center trials. Stat Med. 1998; 17: Senn S. Statistical Issues in Drug Development. Multi-center trails. Wiley;1997: Källén A. Treatment-by-center interaction: What is the issue? Drug Inf J. 1997; 31: APPENDIX 1 Why the Power Loss for a Higher Powered Study is Smaller The overall study power is estimated by: Power = Pr t na +n B 2c > t α/2,(na +n B 2c) δ c i=1 n Ai n Bi n Ai + n Bi = E (nji ) Pr t na +n 2c > t α/2,(na +n 2c) δ n Ai n Bi c B B i=1 n Ai + n Bi *n ji = E (nji ) 1 F t t α/2,(na +n 2c) δ n Ai n Bi c B i=1 n Ai + n Bi *n ji Power max 1 F t t α/2,(na +n 2c) δ n A + n B B 2 where Power max is the maximum power when the numbers of patients are exactly the same between the two treatments at each of all centers, and F t is the cumulative probability distribution of t na+nb 2c. Therefore, the power loss is: Loss Pwr (δ) = Power max Power = E (nji ) F t t α/2 δ c i=1 = E (nji )[G(δ)*n ji ] n Ai n Bi n Ai + n Bi F t t α/2 δ n A + n B 2 *n ji where G(x) = F t t α/2 x c i=1 n Ai n Bi n Ai + n Bi F t t α/2 x n A + n B 2 = F t (t α/2 xm) F t (t α/2 xn) where M = Σn Ai n Bi /(n Ai + n Bi ),andn = n A + n B /2. The power loss corresponding to a higher powered study (eg,
8 386 Zhengning Lin 90% power) with the same sample size is simply the power loss corresponding to a higher δ. Therefore, if Loss pwr (x) is generally a decreasing function of x then the power loss for higher powered study is less than that for a 80% powered study with the same sample size. To show that Loss pwr (x) is decreasing it is sufficient if we can prove that G(x) is decreasing. Actually, G (x) = f t (t α/2 xn)n f t (t α/2 xm)m = H(N) H(M) where f t (a) =Γ[(r + 1)/2] rπ Γ(r/2) 1 (1 + a 2 /r) (r 1)/2 is the density function of the t distribution with r degrees of freedom, and H(y) = f t (t α/2 xy)y. To show that G (x) < 0 so that G(x) is decreasing in the region of interest, we only need to prove that H(y) is decreasing for M y N while x corresponds to a test of at least 80% power. Actually, it can be shown that where H (y) = f t (xy t α/2 )r(r + (xy t α/2 ) 2 ) 1 L(y) L(y) = xy r 1 t α/2 + (r + 1)t α/2 + 1 Therefore if L(y) < 0 then H(y) has to be decreasing for M y N. Actually, L(y) < 0 xy r 1 t α/2 > (r + 1)t α/2 + 1 xy t α/2 > r + 1 t α/2 + (r + 1)t α/2 + 1 xy t α/2 > (r + 1)t α/2 (r + 1)t α/ However, (r + 1)t α/2 (r + 1)t α/2 1 1 z α/2 z α/2 = where Z α/2 is the corresponding percentile for standard normal distribution and α=0.05. However, only statistical tests with power = 66.3 can have xy t α/2 > 0.42 (when x =δ and y = M), which is usually outside the area for studies powered at 80% or more with relatively small randomization block size. Therefore, generally xy t α/2 = 0.42 and thus L(y) < 0. Therefore, the actual power loss for a higher powered study with the same total sample size is generally smaller.
SAMPLE SIZE RE-ESTIMATION FOR ADAPTIVE SEQUENTIAL DESIGN IN CLINICAL TRIALS
Journal of Biopharmaceutical Statistics, 18: 1184 1196, 2008 Copyright Taylor & Francis Group, LLC ISSN: 1054-3406 print/1520-5711 online DOI: 10.1080/10543400802369053 SAMPLE SIZE RE-ESTIMATION FOR ADAPTIVE
More informationComparison of Three Calculation Methods for a Bayesian Inference of Two Poisson Parameters
Journal of Modern Applied Statistical Methods Volume 13 Issue 1 Article 26 5-1-2014 Comparison of Three Calculation Methods for a Bayesian Inference of Two Poisson Parameters Yohei Kawasaki Tokyo University
More informationMantel-Haenszel Test Statistics. for Correlated Binary Data. Department of Statistics, North Carolina State University. Raleigh, NC
Mantel-Haenszel Test Statistics for Correlated Binary Data by Jie Zhang and Dennis D. Boos Department of Statistics, North Carolina State University Raleigh, NC 27695-8203 tel: (919) 515-1918 fax: (919)
More informationSensitivity study of dose-finding methods
to of dose-finding methods Sarah Zohar 1 John O Quigley 2 1. Inserm, UMRS 717,Biostatistic Department, Hôpital Saint-Louis, Paris, France 2. Inserm, Université Paris VI, Paris, France. NY 2009 1 / 21 to
More informationEnquiry. Demonstration of Uniformity of Dosage Units using Large Sample Sizes. Proposal for a new general chapter in the European Pharmacopoeia
Enquiry Demonstration of Uniformity of Dosage Units using Large Sample Sizes Proposal for a new general chapter in the European Pharmacopoeia In order to take advantage of increased batch control offered
More informationProbability Distributions.
Probability Distributions http://www.pelagicos.net/classes_biometry_fa18.htm Probability Measuring Discrete Outcomes Plotting probabilities for discrete outcomes: 0.6 0.5 0.4 0.3 0.2 0.1 NOTE: Area within
More informationCHL 5225H Advanced Statistical Methods for Clinical Trials: Multiplicity
CHL 5225H Advanced Statistical Methods for Clinical Trials: Multiplicity Prof. Kevin E. Thorpe Dept. of Public Health Sciences University of Toronto Objectives 1. Be able to distinguish among the various
More informationImpact of Stratified Randomization in Clinical Trials
Impact of Stratified Randomization in Clinical Trials Vladimir V. Anisimov Abstract This paper deals with the analysis of randomization effects in clinical trials. The two randomization schemes most often
More informationPrecision of maximum likelihood estimation in adaptive designs
Research Article Received 12 January 2015, Accepted 24 September 2015 Published online 12 October 2015 in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/sim.6761 Precision of maximum likelihood
More informationSTAT5044: Regression and Anova
STAT5044: Regression and Anova Inyoung Kim 1 / 18 Outline 1 Logistic regression for Binary data 2 Poisson regression for Count data 2 / 18 GLM Let Y denote a binary response variable. Each observation
More informationA Note on Bayesian Inference After Multiple Imputation
A Note on Bayesian Inference After Multiple Imputation Xiang Zhou and Jerome P. Reiter Abstract This article is aimed at practitioners who plan to use Bayesian inference on multiplyimputed datasets in
More informationDistribution Theory. Comparison Between Two Quantiles: The Normal and Exponential Cases
Communications in Statistics Simulation and Computation, 34: 43 5, 005 Copyright Taylor & Francis, Inc. ISSN: 0361-0918 print/153-4141 online DOI: 10.1081/SAC-00055639 Distribution Theory Comparison Between
More informationStatistical Practice
Statistical Practice A Note on Bayesian Inference After Multiple Imputation Xiang ZHOU and Jerome P. REITER This article is aimed at practitioners who plan to use Bayesian inference on multiply-imputed
More informationBest Linear Unbiased Prediction: an Illustration Based on, but Not Limited to, Shelf Life Estimation
Libraries Conference on Applied Statistics in Agriculture 015-7th Annual Conference Proceedings Best Linear Unbiased Prediction: an Illustration Based on, but Not Limited to, Shelf Life Estimation Maryna
More informationSample Size Calculations for Group Randomized Trials with Unequal Sample Sizes through Monte Carlo Simulations
Sample Size Calculations for Group Randomized Trials with Unequal Sample Sizes through Monte Carlo Simulations Ben Brewer Duke University March 10, 2017 Introduction Group randomized trials (GRTs) are
More informationEstimating terminal half life by non-compartmental methods with some data below the limit of quantification
Paper SP08 Estimating terminal half life by non-compartmental methods with some data below the limit of quantification Jochen Müller-Cohrs, CSL Behring, Marburg, Germany ABSTRACT In pharmacokinetic studies
More informationLecture Notes for BUSINESS STATISTICS - BMGT 571. Chapters 1 through 6. Professor Ahmadi, Ph.D. Department of Management
Lecture Notes for BUSINESS STATISTICS - BMGT 571 Chapters 1 through 6 Professor Ahmadi, Ph.D. Department of Management Revised May 005 Glossary of Terms: Statistics Chapter 1 Data Data Set Elements Variable
More informationSTAT 5500/6500 Conditional Logistic Regression for Matched Pairs
STAT 5500/6500 Conditional Logistic Regression for Matched Pairs The data for the tutorial came from support.sas.com, The LOGISTIC Procedure: Conditional Logistic Regression for Matched Pairs Data :: SAS/STAT(R)
More informationOn the efficiency of two-stage adaptive designs
On the efficiency of two-stage adaptive designs Björn Bornkamp (Novartis Pharma AG) Based on: Dette, H., Bornkamp, B. and Bretz F. (2010): On the efficiency of adaptive designs www.statistik.tu-dortmund.de/sfb823-dp2010.html
More informationVersion 1.0. General Certificate of Education (A-level) June Statistics SS06. (Specification 6380) Statistics 6. Final.
Version 1.0 General Certificate of Education (A-level) June 013 Statistics SS06 (Specification 6380) Statistics 6 Final Mark Scheme Mark schemes are prepared by the Principal Examiner and considered, together
More informationType I error rate control in adaptive designs for confirmatory clinical trials with treatment selection at interim
Type I error rate control in adaptive designs for confirmatory clinical trials with treatment selection at interim Frank Bretz Statistical Methodology, Novartis Joint work with Martin Posch (Medical University
More informationWhy Correlation Matters in Cost Estimating
Why Correlation Matters in Cost Estimating Stephen A. Book The Aerospace Corporation P.O. Box 92957 Los Angeles, CA 90009-29597 (310) 336-8655 stephen.a.book@aero.org 32nd Annual DoD Cost Analysis Symposium
More informationLecture 9: Learning Optimal Dynamic Treatment Regimes. Donglin Zeng, Department of Biostatistics, University of North Carolina
Lecture 9: Learning Optimal Dynamic Treatment Regimes Introduction Refresh: Dynamic Treatment Regimes (DTRs) DTRs: sequential decision rules, tailored at each stage by patients time-varying features and
More informationQED. Queen s Economics Department Working Paper No Hypothesis Testing for Arbitrary Bounds. Jeffrey Penney Queen s University
QED Queen s Economics Department Working Paper No. 1319 Hypothesis Testing for Arbitrary Bounds Jeffrey Penney Queen s University Department of Economics Queen s University 94 University Avenue Kingston,
More informationContinuous Probability Distributions
1 Chapter 5 Continuous Probability Distributions 5.1 Probability density function Example 5.1.1. Revisit Example 3.1.1. 11 12 13 14 15 16 21 22 23 24 25 26 S = 31 32 33 34 35 36 41 42 43 44 45 46 (5.1.1)
More informationImproving Efficiency of Inferences in Randomized Clinical Trials Using Auxiliary Covariates
Improving Efficiency of Inferences in Randomized Clinical Trials Using Auxiliary Covariates Anastasios (Butch) Tsiatis Department of Statistics North Carolina State University http://www.stat.ncsu.edu/
More informationUse of Coefficient of Variation in Assessing Variability of Quantitative Assays
CLINICAL AND DIAGNOSTIC LABORATORY IMMUNOLOGY, Nov. 2002, p. 1235 1239 Vol. 9, No. 6 1071-412X/02/$04.00 0 DOI: 10.1128/CDLI.9.6.1235 1239.2002 Use of Coefficient of Variation in Assessing Variability
More informationREFERENCES AND FURTHER STUDIES
REFERENCES AND FURTHER STUDIES by..0. on /0/. For personal use only.. Afifi, A. A., and Azen, S. P. (), Statistical Analysis A Computer Oriented Approach, Academic Press, New York.. Alvarez, A. R., Welter,
More informationSupporting Information for Estimating restricted mean. treatment effects with stacked survival models
Supporting Information for Estimating restricted mean treatment effects with stacked survival models Andrew Wey, David Vock, John Connett, and Kyle Rudser Section 1 presents several extensions to the simulation
More informationConflicts of Interest
Analysis of Dependent Variables: Correlation and Simple Regression Zacariah Labby, PhD, DABR Asst. Prof. (CHS), Dept. of Human Oncology University of Wisconsin Madison Conflicts of Interest None to disclose
More informationHow do we compare the relative performance among competing models?
How do we compare the relative performance among competing models? 1 Comparing Data Mining Methods Frequent problem: we want to know which of the two learning techniques is better How to reliably say Model
More informationPubh 8482: Sequential Analysis
Pubh 8482: Sequential Analysis Joseph S. Koopmeiners Division of Biostatistics University of Minnesota Week 12 Review So far... We have discussed the role of phase III clinical trials in drug development
More informationImplementing Response-Adaptive Randomization in Multi-Armed Survival Trials
Implementing Response-Adaptive Randomization in Multi-Armed Survival Trials BASS Conference 2009 Alex Sverdlov, Bristol-Myers Squibb A.Sverdlov (B-MS) Response-Adaptive Randomization 1 / 35 Joint work
More informationDistributed Optimization. Song Chong EE, KAIST
Distributed Optimization Song Chong EE, KAIST songchong@kaist.edu Dynamic Programming for Path Planning A path-planning problem consists of a weighted directed graph with a set of n nodes N, directed links
More informationAPPENDIX B Sample-Size Calculation Methods: Classical Design
APPENDIX B Sample-Size Calculation Methods: Classical Design One/Paired - Sample Hypothesis Test for the Mean Sign test for median difference for a paired sample Wilcoxon signed - rank test for one or
More informationC A R I B B E A N E X A M I N A T I O N S C O U N C I L REPORT ON CANDIDATES WORK IN THE CARIBBEAN ADVANCED PROFICIENCY EXAMINATION MAY/JUNE 2009
C A R I B B E A N E X A M I N A T I O N S C O U N C I L REPORT ON CANDIDATES WORK IN THE CARIBBEAN ADVANCED PROFICIENCY EXAMINATION MAY/JUNE 2009 APPLIED MATHEMATICS Copyright 2009 Caribbean Examinations
More informationCalibration Verification Linearity in the Clinical Lab. Did I Pass or Fail? 2017 ASCLS New Jersey
Calibration Verification Linearity in the Clinical Lab. Did I Pass or Fail? 2017 ASCLS New Jersey Presentation Topics & Objectives Calibration Verification Key Definitions Why do I need to perform CV?
More informationAlternative Presentation of the Standard Normal Distribution
A2 A APPENDIX A Alternative Presentation of the Standard Normal Distribution WHAT YOU SHOULD LEARN How to find areas under the standard normal curve INSIGHT Because every normal distribution can be transformed
More informationEFFECT OF THE UNCERTAINTY OF THE STABILITY DATA ON THE SHELF LIFE ESTIMATION OF PHARMACEUTICAL PRODUCTS
PERIODICA POLYTECHNICA SER. CHEM. ENG. VOL. 48, NO. 1, PP. 41 52 (2004) EFFECT OF THE UNCERTAINTY OF THE STABILITY DATA ON THE SHELF LIFE ESTIMATION OF PHARMACEUTICAL PRODUCTS Kinga KOMKA and Sándor KEMÉNY
More informationGroup Sequential Tests for Delayed Responses
Group Sequential Tests for Delayed Responses Lisa Hampson Department of Mathematics and Statistics, Lancaster University, UK Chris Jennison Department of Mathematical Sciences, University of Bath, UK Read
More informationThe Design of a Survival Study
The Design of a Survival Study The design of survival studies are usually based on the logrank test, and sometimes assumes the exponential distribution. As in standard designs, the power depends on The
More informationWhat is Experimental Design?
One Factor ANOVA What is Experimental Design? A designed experiment is a test in which purposeful changes are made to the input variables (x) so that we may observe and identify the reasons for change
More informationBy Keith Chrzan, Division Vice President, Marketing Sciences Group, Maritz Research
Monte Carlo Forecasting: Safer than it Sounds By Keith Chrzan, Division Vice President, Marketing Sciences Group, Maritz Research What do oil well exploration, the Dow Jones and hurricane wind probabilities
More informationReport on the Examination
Version 1.0 General Certificate of Education (A-level) January 01 Mathematics MPC4 (Specification 660) Pure Core 4 Report on the Examination Further copies of this Report on the Examination are available
More informationTwo-stage Adaptive Randomization for Delayed Response in Clinical Trials
Two-stage Adaptive Randomization for Delayed Response in Clinical Trials Guosheng Yin Department of Statistics and Actuarial Science The University of Hong Kong Joint work with J. Xu PSI and RSS Journal
More informationAdvanced Machine Learning
Advanced Machine Learning Bandit Problems MEHRYAR MOHRI MOHRI@ COURANT INSTITUTE & GOOGLE RESEARCH. Multi-Armed Bandit Problem Problem: which arm of a K-slot machine should a gambler pull to maximize his
More informationFinal. Mark Scheme. Mathematics/Statistics MS/SS1A/W. (Specification 6360/6380) Statistics 1A. General Certificate of Education (A-level) June 2013
Version 1.0 General Certificate of Education (A-level) June 2013 Mathematics/Statistics MS/SS1A/W (Specification 6360/6380) Statistics 1A Final Mark Scheme Mark schemes are prepared by the Principal Examiner
More informationAnalysing data: regression and correlation S6 and S7
Basic medical statistics for clinical and experimental research Analysing data: regression and correlation S6 and S7 K. Jozwiak k.jozwiak@nki.nl 2 / 49 Correlation So far we have looked at the association
More informationPrerequisite: STATS 7 or STATS 8 or AP90 or (STATS 120A and STATS 120B and STATS 120C). AP90 with a minimum score of 3
University of California, Irvine 2017-2018 1 Statistics (STATS) Courses STATS 5. Seminar in Data Science. 1 Unit. An introduction to the field of Data Science; intended for entering freshman and transfers.
More informationPhase I design for locating schedule-specific maximum tolerated doses
Phase I design for locating schedule-specific maximum tolerated doses Nolan A. Wages, Ph.D. University of Virginia Division of Translational Research & Applied Statistics Department of Public Health Sciences
More informationProbability and Probability Distributions. Dr. Mohammed Alahmed
Probability and Probability Distributions 1 Probability and Probability Distributions Usually we want to do more with data than just describing them! We might want to test certain specific inferences about
More informationLecture 3: Introduction to Complexity Regularization
ECE90 Spring 2007 Statistical Learning Theory Instructor: R. Nowak Lecture 3: Introduction to Complexity Regularization We ended the previous lecture with a brief discussion of overfitting. Recall that,
More informationPQL Estimation Biases in Generalized Linear Mixed Models
PQL Estimation Biases in Generalized Linear Mixed Models Woncheol Jang Johan Lim March 18, 2006 Abstract The penalized quasi-likelihood (PQL) approach is the most common estimation procedure for the generalized
More informationMathematics Ordinary Level
L.16/19 Pre-Leaving Certificate Examination, 018 Mathematics Ordinary Level Marking Scheme Paper 1 Pg. Paper Pg. 36 Page 1 of 56 exams Pre-Leaving Certificate Examination, 018 Mathematics Ordinary Level
More informationCOMPLEMENTARY EXERCISES WITH DESCRIPTIVE STATISTICS
COMPLEMENTARY EXERCISES WITH DESCRIPTIVE STATISTICS EX 1 Given the following series of data on Gender and Height for 8 patients, fill in two frequency tables one for each Variable, according to the model
More informationSIMULTANEOUS CONFIDENCE BANDS FOR THE PTH PERCENTILE AND THE MEAN LIFETIME IN EXPONENTIAL AND WEIBULL REGRESSION MODELS. Ping Sa and S.J.
SIMULTANEOUS CONFIDENCE BANDS FOR THE PTH PERCENTILE AND THE MEAN LIFETIME IN EXPONENTIAL AND WEIBULL REGRESSION MODELS " # Ping Sa and S.J. Lee " Dept. of Mathematics and Statistics, U. of North Florida,
More informationStatistics in medicine
Statistics in medicine Lecture 3: Bivariate association : Categorical variables Proportion in one group One group is measured one time: z test Use the z distribution as an approximation to the binomial
More informationAdaptive Designs: Why, How and When?
Adaptive Designs: Why, How and When? Christopher Jennison Department of Mathematical Sciences, University of Bath, UK http://people.bath.ac.uk/mascj ISBS Conference Shanghai, July 2008 1 Adaptive designs:
More informationConfidence Intervals of the Simple Difference between the Proportions of a Primary Infection and a Secondary Infection, Given the Primary Infection
Biometrical Journal 42 (2000) 1, 59±69 Confidence Intervals of the Simple Difference between the Proportions of a Primary Infection and a Secondary Infection, Given the Primary Infection Kung-Jong Lui
More informationMcGill University. Faculty of Science MATH 204 PRINCIPLES OF STATISTICS II. Final Examination
McGill University Faculty of Science MATH 204 PRINCIPLES OF STATISTICS II Final Examination Date: 20th April 2009 Time: 9am-2pm Examiner: Dr David A Stephens Associate Examiner: Dr Russell Steele Please
More informationAdaptive designs beyond p-value combination methods. Ekkehard Glimm, Novartis Pharma EAST user group meeting Basel, 31 May 2013
Adaptive designs beyond p-value combination methods Ekkehard Glimm, Novartis Pharma EAST user group meeting Basel, 31 May 2013 Outline Introduction Combination-p-value method and conditional error function
More information11. Bootstrap Methods
11. Bootstrap Methods c A. Colin Cameron & Pravin K. Trivedi 2006 These transparencies were prepared in 20043. They can be used as an adjunct to Chapter 11 of our subsequent book Microeconometrics: Methods
More informationPubH 7405: REGRESSION ANALYSIS INTRODUCTION TO LOGISTIC REGRESSION
PubH 745: REGRESSION ANALYSIS INTRODUCTION TO LOGISTIC REGRESSION Let Y be the Dependent Variable Y taking on values and, and: π Pr(Y) Y is said to have the Bernouilli distribution (Binomial with n ).
More informationSlides for Data Mining by I. H. Witten and E. Frank
Slides for Data Mining by I. H. Witten and E. Frank Predicting performance Assume the estimated error rate is 5%. How close is this to the true error rate? Depends on the amount of test data Prediction
More information4 Integration. Copyright Cengage Learning. All rights reserved.
4 Integration Copyright Cengage Learning. All rights reserved. 4.1 Antiderivatives and Indefinite Integration Copyright Cengage Learning. All rights reserved. Objectives! Write the general solution of
More informationa Sample By:Dr.Hoseyn Falahzadeh 1
In the name of God Determining ee the esize eof a Sample By:Dr.Hoseyn Falahzadeh 1 Sample Accuracy Sample accuracy: refers to how close a random sample s statistic is to the true population s value it
More information6 Sample Size Calculations
6 Sample Size Calculations A major responsibility of a statistician: sample size calculation. Hypothesis Testing: compare treatment 1 (new treatment) to treatment 2 (standard treatment); Assume continuous
More informationRobust covariance estimator for small-sample adjustment in the generalized estimating equations: A simulation study
Science Journal of Applied Mathematics and Statistics 2014; 2(1): 20-25 Published online February 20, 2014 (http://www.sciencepublishinggroup.com/j/sjams) doi: 10.11648/j.sjams.20140201.13 Robust covariance
More informationAdaptive Extensions of a Two-Stage Group Sequential Procedure for Testing a Primary and a Secondary Endpoint (II): Sample Size Re-estimation
Research Article Received XXXX (www.interscience.wiley.com) DOI: 10.100/sim.0000 Adaptive Extensions of a Two-Stage Group Sequential Procedure for Testing a Primary and a Secondary Endpoint (II): Sample
More informationReliability of Acceptance Criteria in Nonlinear Response History Analysis of Tall Buildings
Reliability of Acceptance Criteria in Nonlinear Response History Analysis of Tall Buildings M.M. Talaat, PhD, PE Senior Staff - Simpson Gumpertz & Heger Inc Adjunct Assistant Professor - Cairo University
More informationFULL LIKELIHOOD INFERENCES IN THE COX MODEL
October 20, 2007 FULL LIKELIHOOD INFERENCES IN THE COX MODEL BY JIAN-JIAN REN 1 AND MAI ZHOU 2 University of Central Florida and University of Kentucky Abstract We use the empirical likelihood approach
More informationBootstrap. Director of Center for Astrostatistics. G. Jogesh Babu. Penn State University babu.
Bootstrap G. Jogesh Babu Penn State University http://www.stat.psu.edu/ babu Director of Center for Astrostatistics http://astrostatistics.psu.edu Outline 1 Motivation 2 Simple statistical problem 3 Resampling
More informationA Simulation Study on Confidence Interval Procedures of Some Mean Cumulative Function Estimators
Statistics Preprints Statistics -00 A Simulation Study on Confidence Interval Procedures of Some Mean Cumulative Function Estimators Jianying Zuo Iowa State University, jiyizu@iastate.edu William Q. Meeker
More informationAn Integral Measure of Aging/Rejuvenation for Repairable and Non-repairable Systems
An Integral Measure of Aging/Rejuvenation for Repairable and Non-repairable Systems M.P. Kaminskiy and V.V. Krivtsov Abstract This paper introduces a simple index that helps to assess the degree of aging
More informationBayesian Optimal Interval Design for Phase I Clinical Trials
Bayesian Optimal Interval Design for Phase I Clinical Trials Department of Biostatistics The University of Texas, MD Anderson Cancer Center Joint work with Suyu Liu Phase I oncology trials The goal of
More informationPermutation tests. Patrick Breheny. September 25. Conditioning Nonparametric null hypotheses Permutation testing
Permutation tests Patrick Breheny September 25 Patrick Breheny STA 621: Nonparametric Statistics 1/16 The conditioning idea In many hypothesis testing problems, information can be divided into portions
More informationSurvival Regression Models
Survival Regression Models David M. Rocke May 18, 2017 David M. Rocke Survival Regression Models May 18, 2017 1 / 32 Background on the Proportional Hazards Model The exponential distribution has constant
More information6 Single Sample Methods for a Location Parameter
6 Single Sample Methods for a Location Parameter If there are serious departures from parametric test assumptions (e.g., normality or symmetry), nonparametric tests on a measure of central tendency (usually
More informationON THE FAILURE RATE ESTIMATION OF THE INVERSE GAUSSIAN DISTRIBUTION
ON THE FAILURE RATE ESTIMATION OF THE INVERSE GAUSSIAN DISTRIBUTION ZHENLINYANGandRONNIET.C.LEE Department of Statistics and Applied Probability, National University of Singapore, 3 Science Drive 2, Singapore
More informationSemi-parametric predictive inference for bivariate data using copulas
Semi-parametric predictive inference for bivariate data using copulas Tahani Coolen-Maturi a, Frank P.A. Coolen b,, Noryanti Muhammad b a Durham University Business School, Durham University, Durham, DH1
More informationRerandomization to Balance Covariates
Rerandomization to Balance Covariates Kari Lock Morgan Department of Statistics Penn State University Joint work with Don Rubin University of Minnesota Biostatistics 4/27/16 The Gold Standard Randomized
More informationTECHNICAL REPORT # 59 MAY Interim sample size recalculation for linear and logistic regression models: a comprehensive Monte-Carlo study
TECHNICAL REPORT # 59 MAY 2013 Interim sample size recalculation for linear and logistic regression models: a comprehensive Monte-Carlo study Sergey Tarima, Peng He, Tao Wang, Aniko Szabo Division of Biostatistics,
More informationA-LEVEL Further Mathematics
A-LEVEL Further Mathematics Statistics Mark scheme Specimen Version 1.1 Mark schemes are prepared by the Lead Assessment Writer and considered, together with the relevant questions, by a panel of subject
More informationTime: 1 hour 30 minutes
Paper Reference(s) 6684/0 Edexcel GCE Statistics S Silver Level S Time: hour 30 minutes Materials required for examination papers Mathematical Formulae (Green) Items included with question Nil Candidates
More informationST 371 (IX): Theories of Sampling Distributions
ST 371 (IX): Theories of Sampling Distributions 1 Sample, Population, Parameter and Statistic The major use of inferential statistics is to use information from a sample to infer characteristics about
More informationApproximating the negative moments of the Poisson distribution
Approximating the negative moments of the Poisson distribution C. Matthew Jones, 2 and Anatoly A. Zhigljavsky School of Mathematics, Cardiff University, CF24 4YH, UK. 2 Cardiff Research Consortium, Cardiff
More informationChapter 1 Statistical Inference
Chapter 1 Statistical Inference causal inference To infer causality, you need a randomized experiment (or a huge observational study and lots of outside information). inference to populations Generalizations
More informationSample Size/Power Calculation by Software/Online Calculators
Sample Size/Power Calculation by Software/Online Calculators May 24, 2018 Li Zhang, Ph.D. li.zhang@ucsf.edu Associate Professor Department of Epidemiology and Biostatistics Division of Hematology and Oncology
More informationNon-Parametric Weighted Tests for Change in Distribution Function
American Journal of Mathematics Statistics 03, 3(3: 57-65 DOI: 0.593/j.ajms.030303.09 Non-Parametric Weighted Tests for Change in Distribution Function Abd-Elnaser S. Abd-Rabou *, Ahmed M. Gad Statistics
More informationA-LEVEL STATISTICS. SS04 Report on the Examination June Version: 1.0
A-LEVEL STATISTICS SS04 Report on the Examination 6380 June 2016 Version: 1.0 Further copies of this Report are available from aqa.org.uk Copyright 2016 AQA and its licensors. All rights reserved. AQA
More informationChapter 14: Repeated-measures designs
Chapter 14: Repeated-measures designs Oliver Twisted Please, Sir, can I have some more sphericity? The following article is adapted from: Field, A. P. (1998). A bluffer s guide to sphericity. Newsletter
More informationData Analysis. with Excel. An introduction for Physical scientists. LesKirkup university of Technology, Sydney CAMBRIDGE UNIVERSITY PRESS
Data Analysis with Excel An introduction for Physical scientists LesKirkup university of Technology, Sydney CAMBRIDGE UNIVERSITY PRESS Contents Preface xv 1 Introduction to scientific data analysis 1 1.1
More informationIntroduction to Rare Event Simulation
Introduction to Rare Event Simulation Brown University: Summer School on Rare Event Simulation Jose Blanchet Columbia University. Department of Statistics, Department of IEOR. Blanchet (Columbia) 1 / 31
More informationSurvival Analysis for Case-Cohort Studies
Survival Analysis for ase-ohort Studies Petr Klášterecký Dept. of Probability and Mathematical Statistics, Faculty of Mathematics and Physics, harles University, Prague, zech Republic e-mail: petr.klasterecky@matfyz.cz
More informationExam details. Final Review Session. Things to Review
Exam details Final Review Session Short answer, similar to book problems Formulae and tables will be given You CAN use a calculator Date and Time: Dec. 7, 006, 1-1:30 pm Location: Osborne Centre, Unit
More informationSample size re-estimation in clinical trials. Dealing with those unknowns. Chris Jennison. University of Kyoto, January 2018
Sample Size Re-estimation in Clinical Trials: Dealing with those unknowns Christopher Jennison Department of Mathematical Sciences, University of Bath, UK http://people.bath.ac.uk/mascj University of Kyoto,
More informationTable of z values and probabilities for the standard normal distribution. z is the first column plus the top row. Each cell shows P(X z).
Table of z values and probabilities for the standard normal distribution. z is the first column plus the top row. Each cell shows P(X z). For example P(X.04) =.8508. For z < 0 subtract the value from,
More informationThe information complexity of sequential resource allocation
The information complexity of sequential resource allocation Emilie Kaufmann, joint work with Olivier Cappé, Aurélien Garivier and Shivaram Kalyanakrishan SMILE Seminar, ENS, June 8th, 205 Sequential allocation
More informationLecture 22: Variance and Covariance
EE5110 : Probability Foundations for Electrical Engineers July-November 2015 Lecture 22: Variance and Covariance Lecturer: Dr. Krishna Jagannathan Scribes: R.Ravi Kiran In this lecture we will introduce
More informationPERCENTILE ESTIMATES RELATED TO EXPONENTIAL AND PARETO DISTRIBUTIONS
PERCENTILE ESTIMATES RELATED TO EXPONENTIAL AND PARETO DISTRIBUTIONS INTRODUCTION The paper as posted to my website examined percentile statistics from a parent-offspring or Neyman- Scott spatial pattern.
More information