Rebecca A Betensky
Rebecca Betensky
Scroll
Chair of the Department of Biostatistics
Professor of Biostatistics
-
Professional overview
-
Prior to NYU, Dr. Betensky was Professor of Biostatistics at the Harvard T.H. Chan School of Public Health. She was director of the Harvard Catalyst (Clinical and Translational Science Award) Biostatistics Program; director of the Data and Statistics Core for the Massachusetts Alzheimer’s Disease Research Center; and director of the Biostatistics Neurology Core at Massachusetts General Hospital. Previously, she was the Biostatistics Program Leader for the Dana-Farber/Harvard Cancer Center.
Dr. Betensky’s research focuses on methods for the analysis of censored and truncated outcomes and covariates, which frequently arise from the subsampling of cohort studies. She has a long-time interest in clinical trials, and has written on the evaluation of biomarkers and the use and interpretation of p-values. She has collaborated extensively in studies in neurologic diseases, and serves as statistical editor for Annals of Neurology.
Dr. Betensky was awarded, and directed for 15 years, an NIH T32 training program in neurostatistics and neuroepidemiology for pre- and post-doctoral students in biostatistics and epidemiology and for clinician-scientists. She previously directed Harvard’s Biostatistics programs to promote and support diversity at all levels in the field of quantitative public health. She was also a member of the BMRD Study Section for review of NIH statistical methodology grants; on committees for the Institute of Medicine; and a co-chair of the technical advisory committee for the scientific registry of transplant recipients.
Dr. Betensky an elected Fellow of the American Statistical Association and of the International Statistical Institute, and is a past recipient of the Spiegelman Award from the American Public Health Association. She currently serves as a member of the Board of Scientific Counselors for Clinical Science and Epidemiology at the National Cancer Institute.
-
Education
-
AB, Mathematics, Harvard University, Cambridge, MAPhD, Statistics, Stanford University, Stanford, CA
-
Areas of research and study
-
BiologyBiostatisticsNeuroepidemiologyNeurologyNeurostatisticsTranslational science
-
Publications
Publications
Alternative derivations of a rule for early stopping in favor of H0
AbstractBetensky, R. A. (n.d.).Publication year
2000Journal title
American StatisticianVolume
54Issue
1Page(s)
35-39AbstractIt is often desirable to stop a large clinical trial before its planned end if a null result seems inevitable. This early stopping can save considerable resources. It is especially appealing when an experimental treatment is being compared to a standard treatment. Three procedures for early stopping, all with different interpretations and derivations, are described and shown to produce identical rules for normal data and certain parameters. In some cases, this is unexpected and informative. The procedures differ in which of their parameters are adjusted from the fixed sample values to maintain the desired Type I error in this setting of multiple looks at the data.Alternative derivations of a rule for early stopping in favor of HO
AbstractBetensky, R. A. (n.d.).Publication year
2000Journal title
American StatisticianVolume
54Issue
1Page(s)
35-39AbstractIt is often desirable to stop a large clinical trial before its planned end if a null result seems inevitable. This early stopping can save considerable resources. It is especially appealing when an experimental treatment is being compared to a standard treatment. Three procedures for early stopping, all with different interpretations and derivations, are described and shown to produce identical rules for normal data and certain parameters. In some cases, this is unexpected and informative. The procedures differ in which of their parameters are adjusted from the fixed sample values to maintain the desired Type I error in this setting of multiple looks at the data.Approximating the distribution of maximally selected McNemar's statistics
AbstractRabinowitz, D., & Betensky, R. A. (n.d.).Publication year
2000Journal title
BiometricsVolume
56Issue
3Page(s)
897-902AbstractIt is common in epidemiologic analyses to summarize continuous outcomes as falling above or below a threshold. With paired data and with a threshold chosen without reference to the outcomes, McNemar's test of marginal homogeneity may be applied to the resulting dichotomous pairs when testing for equality of the marginal distributions of the underlying continuous outcomes. If the threshold is chosen to maximize the test statistic, however, referring the resulting test statistic to the nominal χ2 distribution is incorrect; Instead, the p-value must be adjusted for the multiple comparisons. Here the distribution of a maximally selected McNemar's statistic is derived, and it is shown that an approximation due to Durbin (1985, Journal of Applied Probability 22, 99-122) may be used to estimate approximate p-values. The methodology is illustrated by an application to measurements of insulin-like growth factor-I (IGF-I)in matched prostate cancer cases and controls from the Physicians' Health Study. The results of Simulation experiments that assess the accuracy of the approximation in moderate sample sizes are reported.Multiple imputation for simple estimation of the hazard function based on interval censored data
AbstractBebchuk, J. D., & Betensky, R. A. (n.d.).Publication year
2000Journal title
Statistics in MedicineVolume
19Issue
3Page(s)
405-419AbstractA data augmentation algorithm is presented for estimating the hazard function and pointwise variability intervals based on interval censored data. The algorithm extends that proposed by Tanner and Wong for grouped right censored data to interval censored data. It applies multiple imputation and local likelihood methods to obtain smooth non-parametric estimates for the hazard function. This approach considerably simplifies the problem of estimation for interval censored data as it transforms it into the more tractable problem of estimation for right censored data. The method is illustrated for two real data sets: times to breast cosmesis deterioration and times to HIV-1 infection for individuals with haemophilia. Simulations are presented to assess the effects of various parameters on the estimates and their variances. Copyright (C) 2000 John Wiley and Sons, Ltd.On nonidentifiability and noninformative censoring for current status data
AbstractBetensky, R. A. (n.d.).Publication year
2000Journal title
BiometrikaVolume
87Issue
1Page(s)
218-221AbstractThe event times and examination times that produce current status data are typically assumed to be independent. Here, an increasing sequence of nested models is considered for current status data, namely independence models, 'constant sum' models and models for which the conditional probability of the occurrence of the event prior to the examination time, given the examination time, is nondecreasing in the examination time. In the class of constant sum models, the distribution of the event time is identifiable and the examination times are noninformative. In the class of models with nondecreasing conditional probability, the distribution of the event time is nonidentifiable. Outside this class, the examination times cannot be ignored.Redistribution algorithms for censored data
AbstractBetensky, R. A. (n.d.).Publication year
2000Journal title
Statistics and Probability LettersVolume
46Issue
4Page(s)
385-389AbstractEfron (Proceedings of the Fifth Berkeley Symposium, Vol. 4, University of California Press, Berkeley, CA, pp. 831-853) and Dinse (Amer. Statist. 39, 1985, 299-300) proposed redistribution of mass algorithms for survivor function estimation from right censored data. Dinse's algorithm is easily extended to survivor function estimation from interval censored data and is further extended to incorporate information on disease markers.Shipment impairs lymphocyte proliferative responses to microbial antigens
AbstractBetensky, R. A., Connick, E., Devers, J., Landay, A. L., Nokta, M., Plaeger, S., Rosenblatt, H., Schmitz, J. L., Valentine, F., Wara, D., Weinberg, A., & Lederman, H. M. (n.d.).Publication year
2000Journal title
Clinical and Diagnostic Laboratory ImmunologyVolume
7Issue
5Page(s)
759-763AbstractLymphocyte proliferation assays (LPAs) are widely used to assess T-lymphocyte function of patients with human immunodeficiency virus infection and other primary and secondary immunodeficiency disorders. Since these assays require expertise not readily available at all clinical sites, specimens may be shipped to central labs for testing. We conducted a large multicenter study to evaluate the effects of shipping on assay performance and found significant loss of LPA activity. This may lead to erroneous results for individual subjects and introduce bias into multicenter trials.Simple approximations for the maximal transmission/disequilibrium test with a multi-allelic marker
AbstractBetensky, R. A., & Rabinowitz, D. (n.d.).Publication year
2000Journal title
Annals of Human GeneticsVolume
64Issue
6Page(s)
567-574AbstractSpielman et al. (1993) popularized the transmission/disequilibrium test (TDT) to test for linkage between disease and marker loci that show a population association. Several authors have proposed extensions to the TDT for multi-allelic markers. Many of these approaches exhibit a 'swamping' effect in which a marker with a strong effect is not detected by a global test that includes many markers with no effect. To avoid this effect, Schaid (1996) proposed using the maximum of the bi-allelic TDT statistics computed for each allele versus all others combined. The maximal TDT statistic, however, no longer follows a chi-square distribution. Here, a refinement to Bonferroni's correction for multiple testing provided by Worsley (1982) based on maximal spanning trees is applied to calculate accurate upper bounds for the type I error and p-values for the maximal TDT. In addition, an accurate lower Bonferroni bound is applied to calculate power. This approach does not require any simulation-based analysis and is less conservative than the standard Bonferroni correction. The bounds are given for both the exact probability calculations and for those based on the normal approximation. The results are assessed through simulations.Two double-blinded, randomized, comparative trials of 4 Human Immunodeficiency Virus type 1 (HIV-1) envelope vaccines in HIV-1-infected individuals across a spectrum of disease severity: AIDS clinical trials groups 209 and 214
Failed generating bibliography.AbstractPublication year
2000Journal title
Journal of Infectious DiseasesVolume
182Issue
5Page(s)
1357-1364AbstractThe potential role of human immunodeficiency virus type 1 (HIV-1)-specific immune responses in controlling viral replication in vivo has stimulated interest in enhancing virus-specific immunity by vaccinating infected individuals with HIV-1 or its components. These studies were undertaken to define patient populations most likely to respond to vaccination, with the induction of novel HIV-1-specific cellular immune responses, and to compare the safety and immunogenicity of several candidate recombinant HIV-1 envelope vaccines and adjuvants. New lymphoproliferative responses (LPRs) developed in <30% of vaccine recipients. LPRs were elicited primarily in study participants with a CD4 cell count >350 cells/mm3 and were usually strain restricted. Responders tended to be more likely than nonresponders to have an undetectable level of HIV-1 RNA at baseline (P =.067). Induction of new cellular immune responses by HIV-1 envelope vaccines is a function of the immunologic stage of disease and baseline plasma HIV-1 RNA level and exhibits considerable vaccine strain specificity.Using conditional logistic regression to fit proportional odds models to interval censored data
AbstractRabinowitz, D., Betensky, R. A., & Tsiatis, A. A. (n.d.).Publication year
2000Journal title
BiometricsVolume
56Issue
2Page(s)
511-518AbstractAn easily implemented approach to fitting the proportional odds regression model to interval-censored data is presented. The approach is based on using conditional logistic regression routines in standard statistical packages. Using conditional logistic regression allows the practitioner to sidestep complications that attend estimation of the baseline odds ratio function. The approach is applicable both for interval-censored data in settings in which examinations continue regardless of whether the event of interest has occurred and for current status data. The methodology is illustrated through an application to data from an AIDS study of the effect of treatment with ZDV+ddC versus ZDV alone on 50% drop in CD4 cell count from baseline level. Simulations are presented to assess the accuracy of the procedure.A non-parametric maximum likelihood estimator for bivariate interval censored data
AbstractBetensky, R. A., & Finkelstein, D. M. (n.d.).Publication year
1999Journal title
Statistics in MedicineVolume
18Issue
22Page(s)
3089-3100AbstractWe derive a non-parametric maximum likelihood estimator for bivariate interval censored data using standard techniques for constrained convex optimization. Our approach extends those taken for univariate interval censored data. We illustrate the estimator with bivariate data from an AIDS study.An extension of Kendall's coefficient of concordance to bivariate interval censored data
AbstractBetensky, R. A., & Finkelstein, D. M. (n.d.).Publication year
1999Journal title
Statistics in MedicineVolume
18Issue
22Page(s)
3101-3109AbstractNon-parametric tests of independence, as well as accompanying measures of association, are essential tools for the analysis of bivariate data. Such tests and measures have been developed for uncensored and right censored failure time data, but have not been developed for interval censored failure time data. Bivariate interval censored data arise in AIDS studies in which screening tests for early signs of viral and bacterial infection are done at clinic visits. Because of missed clinic visits, the actual times of first positive screening tests are interval censored. To handle such data, we propose an extension of Kendall's coefficient of concordance. We apply it to data from an AIDS study that recorded times of shedding of cytomegalovirus (CMV) and times of colonization of mycobacterium avium complex (MAC). We examine the performance of our proposed measure through a simulation study.Clinical trials using HIV-1 RNA-based primary endpoints: Statistical analysis and potential biases
AbstractMarschner, I. C., Betensky, R. A., DeGruttola, V., Hammer, S. M., & Kuritzkes, D. R. (n.d.).Publication year
1999Journal title
Journal of Acquired Immune Deficiency Syndromes and Human RetrovirologyVolume
20Issue
3Page(s)
220-227AbstractClinical trial endpoints based on magnitude of reduction in HIV-1 RNA levels provide an important complement to endpoints based on percentage of patients achieving complete virologic suppression. However, interpretation of magnitude of reduction can he biased by measurement limitations of virologic assays, particularly lower and upper limits of quantification. Using data from two AIDS Clinical Trials Group (ACTG) studies, widely used crude methods of analyzing HIV-1 RNA reductions were compared with methods that take into account censoring of HIV-1 RNA measurements. Such methods include Kaplan- Meier and censored regression analyses. It was found that standard crude methods of analysis consistently underestimated treatment effects. In some cases, the bias induced by crude methods masked statistically significant differences between treatment arms. Although statistically significant, adjustment for baseline HIV-1 RNA levels had little effect on estimated treatment differences. Furthermore, convenient parametric analyses performed as well as more complex nonparametric analyses. It is concluded that conveniently implemented censored data analyses should be conducted in preference to widely used crude analyses of magnitude of HIV-1 RNA reduction. To obtain complete information about virologic response to antiretroviral therapy, such analyses of magnitude of virologic response should be used to complement analyses of the percentage of patients having complete virologic suppression.Local EM estimation of the hazard function for interval-censored data
AbstractBetensky, R. A., Lindsey, J. C., Ryan, L. M., & Wand, M. P. (n.d.).Publication year
1999Journal title
BiometricsVolume
55Issue
1Page(s)
238-245AbstractWe propose a smooth hazard estimator for interval-censored survival data using the method of local likelihood. The model is fit using a local EM algorithm. The estimator is more descriptive than traditional empirical estimates in regions of concentrated information and takes on a parametric flavor in regions of sparse information. We derive two different standard error estimates for the smooth curve, one based on asymptotic theory and the other on the bootstrap. We illustrate the local EM method for times to breast cosmesis deterioration (Finkelstein, 1986, Biometrics 42, 845-854) and for times to HIV-1 infection for individuals with hemophilia (Kroner et al., 1994, Journal of AIDS 7, 279-286). Our hazard estimates for each of these data sets show interesting structures that would not be found using a standard parametric hazard model or empirical survivorship estimates.Maximally selected χ2 statistics for k x 2 tables
AbstractBetensky, R. A., & Rabinowitz, D. (n.d.).Publication year
1999Journal title
BiometricsVolume
55Issue
1Page(s)
317-320AbstractIt is common in epidemiologic analyses to summarize continuous outcomes as falling above or below a threshold. With such a dichotomized outcome, the usual χ2 statistics for association or trend can be used to test for equality of proportions across strata of the study population. However, if the threshold is chosen to maximize the test statistic, the nominal χ2 reference distributions are incorrect. In this paper, the asymptotic distributions of maximally selected χ2 statistics for association and for trend for the k x 2 table are derived. The methodology is illustrated with data from an AIDS clinical trial. The results of simulation experiments that assess the accuracy of the asymptotic distributions in moderate sample sizes are also reported.Predictive value of CD19 measurements for bacterial infections in children infected with human immunodeficiency virus
AbstractBetensky, R. A., Calvelli, T., & Pahwa, S. (n.d.).Publication year
1999Journal title
Clinical and Diagnostic Laboratory ImmunologyVolume
6Issue
2Page(s)
247-253AbstractWe investigated the predictive value of CD19 cell percentages (CD19%) for times to bacterial infections, using data from six pediatric AIDS Clinical Trials Group protocols and adjusting for other potentially prognostic variables, such as CD4%, CD8%, immunoglobulin (IgA) level, lymphocyte count, prior infections, prior zidovudine treatment, and age. In addition, we explored the combined effects of CD19% and IgG level in predicting time to infection. We found that a low CD19% is associated with a nonsignificant 1.2-fold increase in hazard of bacterial infection (95% confidence interval: 0.97, 1.49). In contrast, a high IgG level is associated with a nonsignificant 0.87-fold decrease in hazard of infection (95% confidence interval: 0.68, 1.12). CD4% was more prognostic of time to bacterial infection than CD19% or IgG level. Low CD19% and high IgG levels together lead to a significant (P < 0.01) 0.50-fold decrease in hazard (95% confidence interval: 0.35, 0.73) relative to low CD19% and low IgG levels. Similarly, in a model involving assay result changes (from baseline to 6 months) as well as baseline values, the effect of CD19% by itself is reversed from its effect in conjunction with IgG. In this model, CD19% that are increasing and high are associated with decreases in hazard of infection (P < 0.01), while increasing CD19% and increasing IgG levels are associated with significant (at the P = 0.01 level) fourfold increases in hazard of infection relative to stable CD19% and decreasing; stable, or increasing IgG levels. Our data suggest that CD19%, in conjunction with IgG level, provides a useful prognostic tool for bacterial infections. It is highly likely thai T-helper function impacts on B-cell function; thus, inclusion of CD4% in such analyses may greatly enhance the assessment of risk for bacterial infection.Thymocyte development in Ah-receptor-deficient mice is refractory to TCDD-inducible changes
AbstractHundeiker, C., Pineau, T., Cassar, G., Betensky, R. A., Gleichmann, E., & Esser, C. (n.d.).Publication year
1999Journal title
International Journal of ImmunopharmacologyVolume
21Issue
12Page(s)
841-859AbstractThe arylhydrocarbon receptor (AhR), a ligand-activated transcription factor, is differentially distributed in tissues and abundant in the thymus epithelium. The activated AhR can induce the transcription of an array of genes, including genes of cell growth and differentiation. Neither the physiological function of the AhR nor its putative natural ligand is known. 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) is a xenobiotic high-affinity activator of the AhR, and appears to be essential for most of the multifold toxic effects of TCDD. Activation of the AhR by even low doses of TCDD results in general immunosuppression and thymus hypoplasia. TCDD exposure interferes with thymocyte development; for instance, it reduces the proliferation rate of the very immature (CD4-CD8- and CD4-CD8+HSA+) thymocytes, leads to preferential emigration of very immature cells, and drastically skews the differentiation of thymocyte subpopulations towards mature CD4-CD8+ αβTCR(high) thymocytes. As shown here, in fetal thymi of AhR-deficient mice, thymocyte differentiation kinetics as defined by CD4 and CD8 surface markers, was comparable to AhR(+/+) C57BL/6 mice. Also, the cell emigration characteristics were similar to AhR(+/+) mice. These parameters were refractory to TCDD exposure in the AhR(-/-) mice, but not in the C57BL/6 mice. However, in AhR deficient mice at gestation day 15 more CD4-CD8- immature cells bore high amounts of the (αβ-T-cell receptor. Also, fetal thymocyte numbers were significantly lower, as compared to strain C57BL/6. Thus, the AhR is the mediator of thymotoxic effects of TCDD.A boundary crossing probability for the bessel process
AbstractBetensky, R. A. (n.d.).Publication year
1998Journal title
Advances in Applied ProbabilityVolume
30Issue
3Page(s)
807-830AbstractAnalytic approximations are derived for the distribution of the first crossing time of a straight-line boundary by a d-dimensional Bessel process and its discrete time analogue. Themain ingredient for the approximations is the conditional probability that the process crossed the boundary before time m, given its location beneath the boundary at time m. The boundary crossing probability is of interest as the significance level and power of a sequential test comparing d + 1 treatments using an O Brien–Fleming (1979) stopping boundary (see Betensky 1996). Also, it is shown by DeLong (1980) to be the limiting distribution of a nonparametric test statistic for multiple regression. The approximations are compared with exact values from the literature and with values from a Monte Carlo simulation.A boundary crossing probability for the Bessel process
AbstractBetensky, R. A. (n.d.).Publication year
1998Journal title
Advances in Applied ProbabilityVolume
30Issue
3Page(s)
807-830AbstractAnalytic approximations are derived for the distribution of the first crossing time of a straight-line boundary by a d-dimensional Bessel process and its discrete time analogue. The main ingredient for the approximations is the conditional probability that the process crossed the boundary before time m, given its location beneath the boundary at time m. The boundary crossing probability is of interest as the significance level and power of a sequential test comparing d + 1 treatments using an O'Brien-Fleming (1979) stopping boundary (see Betensky 1996). Also, it is shown by DeLong (1980) to be the limiting distribution of a nonparametric test statistic for multiple regression. The approximations are compared with exact values from the literature and with values from a Monte Carlo simulation.Construction of a continuous stopping boundary from an alpha spending function
AbstractBetensky, R. A. (n.d.).Publication year
1998Journal title
BiometricsVolume
54Issue
3Page(s)
1061-1071AbstractLan and DeMets (1983, Biometrika 70, 659-663) proposed a flexible method for monitoring accumulating data that does not require the number and times of analyses to be specified in advance yet maintains an overall Type I error, α. Their method amounts to discretizing a preselected continuous boundary by clumping the density of the boundary crossing time at discrete analysis times and calculating the resultant discrete-time boundary values. In this framework, the cumulative distribution function of the continuous-time stopping rule is used as an alpha spending function. A key assumption that underlies this method is that future analysis times are not chosen on the basis of the current value of the statistic. However, clinical trials may be monitored more frequently when they are close to crossing the boundary. In this situation, the corresponding continuous-time boundary should be used. Here we demonstrate how to construct a continuous stopping boundary from an alpha spending function. This capability is useful also in the design of clinical trials. We use the Beta-Blocker Heart Attack Trial (BHAT) and AIDS Clinical Trials Group protocol 021 for illustration.Effect of shipment, storage, anticoagulant, and cell separation on lymphocyte proliferation assays for human immunodeficiency virus-infected patients
AbstractWeinberg, A., Betensky, R. A., Zhang, L. I., & Ray, G. (n.d.).Publication year
1998Journal title
Clinical and Diagnostic Laboratory ImmunologyVolume
5Issue
6Page(s)
804-807AbstractLymphocyte proliferation assays (LPA), which can provide important information regarding the immune reconstitution of human immunodeficiency virus (HIV)-infected patients on highly active antiretroviral therapy, frequently involve shipment of specimens to central laboratories. In this study, we examine the effect of stimulant, anticoagulant, cell separation, storage, and transportation on LPA results. LPA responses of whole blood and separated peripheral blood mononuclear cells (PBMC) to different stimulants (cytomegalovirus, varicella-zoster virus, candida and tetanus toxoid antigens, and phytohemagglutinin) were measured using fresh specimens shipped overnight and frozen specimens collected in heparin, acid citrate dextrose (ACD), and citrate cell preparation tubes (CPT) from 12 HIV-infected patients and uninfected controls. Odds ratios for positive LPA responses were significantly higher in separated PBMC than in whole blood from ACD- and heparin-anticoagulated samples obtained from HIV-infected patients and from ACD-anticoagulated samples from uninfected controls. On separated PBMC, positive responses were significantly more frequent in fresh samples compared with overnight transportation for all antigens and compared with cryopreservation for the candida and tetanus antigens. In addition, viral antigen LPA responses were better preserved in frozen PBMC compared with specimens shipped overnight. CPT tubes yielded significantly more positive LPA results for all antigens, irrespective of the HIV patient status compared with ACD, but only for the candida and tetanus antigens and only in HIV- negative controls compared with heparin. Although HIV-infected patients had a significantly lower number of positive antigen-driven LPA responses compared with uninfected controls, most of the specimen processing variables had similar effects on HIV-positive and -negative samples. We conclude that LPA should be performed on site, whenever feasible, by using separated PBMC from fresh blood samples collected in either heparin or ACD. However, if on-site testing is not available, optimal transportation conditions should he established for specific antigens.Multiple imputation for early stopping of a complex clinical trial
AbstractBetensky, R. A. (n.d.).Publication year
1998Journal title
BiometricsVolume
54Issue
1Page(s)
229-242AbstractIt is desirable to have procedures available for stopping a clinical trial early if there appears to be no treatment effect. Conditional power procedures allow for early stopping in favor of the null hypothesis if the probability of rejecting H0 at the planned end of the trial given the current data and a value of the parameter of interest is below some threshold level. Lan, Simon, and Halperin (1982, Communications in Statistics C1, 207- 219) proposed a stochastic curtailment procedure that calculates the conditional power under the alternative hypothesis. Alternatively, predictive power procedures incorporate information from the observed data by averaging the conditional power over the posterior distribution of the parameter. For complex problems in which explicit evaluation of conditional power is not possible, we propose treating the problem of projecting the outcome of a trial given the current data as a missing data problem. We then complete the data using multiple imputation and thus eliminate the need for explicit calculation of conditional power. We apply this method to AIDS Clinical Trials Group (ACTG) protocol 118 and to several simulated clinical trials.An examination of methods for sample size recalculation during an experiment
AbstractBetensky, R. A., & Tierney, C. (n.d.).Publication year
1997Journal title
Statistics in MedicineVolume
16Issue
22Page(s)
2587-2598AbstractIn designing experiments, investigators frequently can specify an important effect that they wish to detect with high power, without the ability to provide an equally certain assessment of the variance of the response. If the experiment is designed based on a guess of the variance, an under-powered study may result. To remedy this problem, there have been several procedures proposed that obtain estimates of the variance from the data as they accrue and then recalculate the sample size accordingly. One class of procedures is fully sequential in that it assesses after each response whether the current sample size yields the desired power based on the current estimate of the variance. This approach is efficient, but it is not practical or advisable in many situations. Another class of procedures involves only two or three stages of sampling and recalculates the sample size based on the observed variance at designated times, perhaps coinciding with interim efficacy analyses. The two-stage approach can result in substantial oversampling, but it is feasible in many situations, whereas the three-stage approach corrects the problem of oversampling, but is less feasible. We propose a procedure that aims to combine the advantages of both the fully sequential and the two-stage approaches. This quasi-sequential procedure involves only two stages of sampling and it applies the stopping rule from the fully sequential procedure to data beyond the initial sample which we obtain via multiple imputation. We show through simulations that when the initial sample size is substantially less than the correct sample size, the mean squared error of the final sample size calculated from the quasi-sequential procedure can be considerably less than that from the two-stage procedure. We compare the distributions of these recalculated sample sizes and discuss our findings for alternative procedures, as well.Conditional power calculations for early acceptance of H(O) embedded in sequential tests
AbstractBetensky, R. A. (n.d.).Publication year
1997Journal title
Statistics in MedicineVolume
16Issue
4Page(s)
465-477AbstractFor ethical and efficiency concerns one often wishes to design a clinical trial to stop early if there is a strong treatment effect or if there is strong evidence of no treatment effect. There is a large literature to address the design of sequential trials for detecting treatment differences. There has been less attention paid to the design of trials for detecting lack of a treatment difference and most of the designs proposed have been ad hoc modifications of the traditional designs. In the context of fixed sample tests, various authors have proposed basing the decision to stop in favour of the null hypothesis, H(O), on conditional power calculations for the end of the trial given the current data. Here I extend this procedure to the popular sequential designs: the O'Brien-Fleming test and the repeated significance test. I derive explicit boundaries for monitoring the test statistic useful for visualizing the impact of the parameters on the operating characteristics of the tests and thus for the design of the tests. Also, they facilitate the use of boundary crossing methods for approximations of power. I derive appropriate boundaries retrospectively for two clinical trials: one that concluded with no treatment difference (AIDS Clinical Trials Group protocol 118) and one that stopped early for positive effect (Beta-Blocker Heart Attack Trial). Finally, I compare the procedures based on the different upper boundaries and assess the impact of allowing for early stopping in favour of H(O), in numerical examples.Early stopping to accept H(o) based on conditional power: Approximations and comparisons
AbstractBetensky, R. A. (n.d.).Publication year
1997Journal title
BiometricsVolume
53Issue
3Page(s)
794-806AbstractIt is intuitively appealing to clinicians to stop a trial early to accept the null hypothesis H0 if it appears that this will be the likely outcome at the planned end of the trial. We consider procedures that calculate at each time point the conditional probability of rejecting H0 at the end of the trial given the current data and some value of the parameter of interest. Lan, Simon, and Halperin (1982, Communications in Statistics C1, 207-219) calculate this probability under the design alternative, and Pepe and Anderson (1992, Applied Statistics 41, 181-190) use an alternative based solely on the current data. We investigate a modification to Pepe and Anderson's (1992) procedure that has a more satisfying interpretation. We define all of these procedures as formal sequential tests with lower stopping boundaries and study them in this context. This facilitates an improved understanding of the interplay of parameters by introducing visual displays, and it leads to an approximation for power by treating it as a boundary crossing probability. We use these tools to compare the performances of the different designs under a variety of parameter configurations.