Showing papers in "Statistical Methods in Medical Research in 1999"

Measuring agreement in method comparison studies

Abstract: Agreement between two methods of clinical measurement can be quantified using the differences between observations made using the two methods on the same subjects. The 95% limits of agreement, estimated by mean difference +/- 1.96 standard deviation of the differences, provide an interval within which 95% of differences between measurements by the two methods are expected to lie. We describe how graphical methods can be used to investigate the assumptions of the method and we also give confidence intervals. We extend the basic approach to data where there is a relationship between difference and magnitude, both with a simple logarithmic transformation approach and a new, more general, regression approach. We discuss the importance of the repeatability of each method separately and compare an estimate of this to the limits of agreement. We extend the limits of agreement approach to data with repeated measurements, proposing new estimates for equal numbers of replicates by each method on each subject, for unequal numbers of replicates, and for replicated data collected in pairs, where the underlying value of the quantity being measured is changing. Finally, we describe a nonparametric approach to comparing methods.

Multiple imputation: a primer:

Abstract: In recent years, multiple imputation has emerged as a convenient and flexible paradigm for analysing data with missing values. Essential features of multiple imputation are reviewed, with answers to frequently asked questions about using the method in practice.

Applications of multiple imputation in medical studies: from AIDS to NHANES

Abstract: Rubin's multiple imputation is a three-step method for handling complex missing data, or more generally, incomplete-data problems, which arise frequently in medical studies. At the first step, m (>...

The interpretation of diagnostic tests.

Abstract: Laboratory diagnostic tests are central in the practice of modern medicine. Common uses include screening a specific population for evidence of disease and confirming or ruling out a tentative diagnosis in an individual patient. The interpretation of a diagnostic test result depends on both the ability of the test to distinguish diseased from nondiseased subjects and the particular characteristics of the patient and setting in which the test is being used. This article reviews statistical methodology for assessing laboratory diagnostic test accuracy and interpreting individual test results, with an emphasis on diagnostic tests that yield a continuous measurement. The article begins with a summary of basic concepts and terminology, then briefly discusses study design and reviews methods for assessing the accuracy of a single diagnostic test, comparing the accuracy of two or more diagnostic tests and interpreting individual test results.

Maximum likelihood analysis of generalized linear models with missing covariates.

Abstract: Missing data is a common occurrence in most medical research data collection enterprises. There is an extensive literature concerning missing data, much of which has focused on missing outcomes. Covariates in regression models are often missing, particularly if information is being collected from multiple sources. The method of weights is an implementation of the EM algorithm for general maximum-likelihood analysis of regression models, including generalized linear models (GLMs) with incomplete covariates. In this paper, we will describe the method of weights in detail, illustrate its application with several examples, discuss its advantages and limitations, and review extensions and applications of the method.

Calculating reference intervals for laboratory measurements

Abstract: Reference intervals are used in laboratory medicine to detect measurements which are extreme, possibly abnormal. Methods of estimating reference intervals and age-specific reference intervals (where the measurement is dependent on a covariate, typically age) are reviewed. The issues of calculating confidence bands, determining appropriate sample sizes and assessing goodness-of-fit are discussed.

Population modelling in drug development

Abstract: In this paper we discuss the vital role that population (hierarchical) modelling can play within the drug development process. Specifically, population pharmacokinetic/pharmacodynamic models can provide reliable predictions of an individualized dose-exposure-response relationship. A predictive model of this kind can be used to simulate and hence design clinical trials, find initial dosage regimens satisfying an optimality criterion on the population distribution of responses, and individualized regimens satisfying such a criterion conditional on individual features, such as sex, age, etc. Throughout we emphasize prediction and advocate mechanistic as opposed to empirical modelling, and argue that the Bayesian approach is particularly natural in this setting.

Parametric models for incomplete continuous and categorical longitudinal data.

Abstract: This paper reviews models for incomplete continuous and categorical longitudinal data. In terms of Rubin's classification of missing value processes we are specifically concerned with the problem of nonrandom missingness. A distinction is drawn between the classes of selection and pattern-mixture models and, using several examples, these approaches are compared and contrasted. The central roles of identifiability and sensitivity are emphasized throughout.

Modelling method comparison data

Abstract: We explore a range of linear regression models that might be useful for either: (a) the relative calibration of two or more methods or (b) to evaluate their precisions relative to each other Ideally, one should be able to use a single data set to carry out the jobs (a) and (b) together Throughout this review we consider the constraints (assumptions) needed to attain identifiability of the models and the possible pitfalls to the unwary in having to introduce them We also pay particular attention to the possible problems arising from the presence of random matrix effects (reproducible random measurement 'errors' that are characteristic of a given method when being used on a given specimen or sample, ie specimen specific biases or subject by method interactions) Finally, we stress the importance of a fully-informative design (using replicate measurements on each subject using at least three independent methods) and large sample sizes

The impact of compliance in pharmacokinetic studies

Abstract: In population pharmacokinetic (PK) studies, one observes just a few concentration measures spread out in time, on a sizable sample of the target population. Common-sense dictates that for estimation of a drug exposure-plasma concentration relationship, one needs accurate information on drug intake history besides the concentration measures. The population PK literature is well aware of this. Studies of simulated compliance behaviour have helped quantify the problem with naive compliance estimators and pointed towards a solution. In this paper we look at actually observed compliance patterns recorded via electronic monitoring. We simulate a documented pharmacokinetic model from the hypertensive literature on top of these and come to some interesting findings. In this clinical trial the problem of noncompliance is much more dramatic than simulated compliance patterns suggested so far. The systematic errors made by compliance naive estimators can be corrected when using timing explicit hierarchical nonlinear models and accurate information on a number of previous dose timings. When it is possible to observe irregular drug intake times in a well-controlled study, a substantial amount of precision is retrieved from the same number of data points. In general, the estimators of PK parameters benefit greatly from information that enters through greater variation in the drug-exposure process. Here we find support for the claim that noncompliance as a rich natural experiment of dosing variation can be a blessing rather than a curse from the information/learning point of view.

StatView for Windows, Version 5.0

Abstract: StatView may be installed from a CD and comes with two manuals `Using StatView' and `StatView reference'. According to the manual, StatView was designed to `be simple, ̄exible and powerful' and to do everything `starting from data spreadsheets and going all the way through your project to fullcolor presentation'. We review the package in view of these aims by concentrating on the following topics: the StatView environment, data entry and manipulation, statistical analyses and graphics. So far we have used this package for the purpose of the review only and so may have missed some strengths or weaknesses.

A regulatory perspective on pharmacokinetic/pharmacodynamic modelling.

Abstract: We present an integrated summary from statistical and pharmacological perspectives of pharmacokinetic/pharmacodynamic (PK/PD) modelling and its use in drug development and regulation for guiding appropriate dosing. An overview of the technical aspects of PK/PD modelling describes how structural models are constructed and refined using pharmacokinetic and pharmacodynamic principles and how random effects models are used to account for individual differences in desired (and undesired) responses due to patient characteristics. Lastly, we describe applications of PK/PD modelling for the purposes of drug labelling, for resolving a safety concern, and for improving therapeutic monitoring of anaesthetic depth during surgery.

Individually randomized intervention trials for disease prevention and control.

Abstract: It is argued that randomized, controlled trials should fulfil a critical role in the identification of practical approaches to the prevention and control of chronic diseases. Because of the great public health potential of chemopreventive and behavioural approaches to chronic disease prevention there is need for a major interdisciplinary scientific effort aimed at intervention development. Because of the cost and duration of controlled trials to evaluate specific interventions there is a need for well-conducted feasibility, pilot and intermediate outcome trials, to inform and to justify corresponding full-scale trials having clinical disease outcomes. Compared to therapeutic trials, prevention trials need to have a greater emphasis on overall benefit versus risk assessment. Such trials need to be large enough, and of sufficient duration, to yield powerful tests of key hypotheses, and informative benefit versus risk summary statements. These requirements have a range of implications for intervention trial ...

Pharmacokinetic and pharmacodynamic modelling in drug development.

Abstract: The aim of a clinical development programme of a new drug is to provide relevant information on the safety and ef®cacy of the compound to enable the prescribing physician to treat individual patients optimally. Frequently, however, dosing recommendations that emerge from such studies are found inappropriate and when individual dose adjustment is needed, the recommendations provided may not be informative enough to allow the adjustment to be undertaken in an optimal manner. The dose±effect relationship is basic to the process of identifying doses to be used in phase III clinical trials and in clinical practice. It is derived in a piece-wise fashion from the various phases of drug development. Pharmacokinetic and pharmacodynamic modelling plays a pivotal role in de®ning the dose±effect relationship. Pharmacokinetics, from the literal Greek meaning of the name, is the study of drug movement in the body. More speci®cally it is the study of the processes of drug absorption, distribution and elimination and frequently the de®nition is extended to encompass the relationship between plasma concentration and pharmacological effect, which more correctly is described as pharmacodynamics. Pharmacokinetics and pharmacodynamics are multidisciplinary subjects, drawing on expertise from physiology, pharmacology, pharmaceutics, analytical chemistry, mathematics and statistics. It is the purpose of the present issue to expose the statistical community to the issues and problems in pharmacokinetic/pharmacodynamic modelling, with the hope that more statisticians will become involved in this important area. The article by Sheiner and Wake®eld gives an introduction to population pharmacokinetics. Frequently, and particularly in the later stages of drug development, only relatively sparse observational concentration and effect data are available. The analysis of sparse observational data, which is called the population approach, has been implemented in phase III studies to obtain additional information about the pharmacokinetic/pharmacodynamic model in a representative sample of patients. Population pharmacokinetics is the study into the pharmacokinetic similarity and differences between individuals from measurements of drug levels in biological ̄uids of subjects or patients arising from some population of interest. In contrast to a traditional study, a population pharmacokinetic study involves large numbers of patients with very heterogeneous characteristics. Study control is dif®cult: many studies involve outpatients and many centres are involved. The reason for adopting a population approach to pharmacokinetic studies is that it has become increasingly obvious that one should study the drug in the target population, which may be different from a normal population. For example, pharmacokinetics may be altered by pathophysiological factors such as renal disease. In addition, apart from estimating mean pharmacokinetic parameters it is important to both quantify and explain

Estimating the population impact of an intervention: a decision-analytic approach.

Abstract: The aim of this paper is to highlight the role for decision analysis in assessing outcomes of medical interventions at a population level. The basic steps of decision analysis are introduced and an illustrative hypothetical preventive intervention is examined. Specific modelling challenges that arise when estimating the population impact of an intervention are described and each is accompanied by an example. Decision analysis can provide useful information for health policy decision makers by identifying the intervention(s) with the largest beneficial impact on health over a wide range of assumptions. In addition, by focusing attention on the parameters with the greatest influence on projected outcomes, decision analysis can aid in identifying critical areas for future research.

Clinical cross-over trials in phase I

Abstract: We review the role of cross-over trials in pharmacokinetic and pharmacodynamic studies, in particular as applied in phase I. Design and analysis considerations are covered. We also consider the use of pharmacokinetic and pharmacodynamic theories in planning cross-over trials. Finally some practical considerations are covered.

The Bayesian analysis of a pivotal pharmacokinetic study.

Abstract: The aim of this paper is to carry out a detailed Bayesian population pharmacokinetic analysis of a three-period cross-over study of the drug fluticasone propionate carried out in 12 healthy male volunteers. The study was carried out to characterize the pharmacokinetics of the drug, in particular to investigate dose proportionality. We examine the appropriateness of modelling assumptions via a variety of diagnostic techniques. We also examine the effect of deleting time points at which the concentration was recorded as below the limit of quantification, as opposed to including these points as censored observations. We assess dose proportionality before carrying out a final combined analysis of data from all three doses.

The role of statisticians in intervention trials.

Abstract: The planning, conduct, analysis and reporting of randomized trials to reliably evaluate an intervention create a complex intellectual and logistical endeavour. Success generally requires active collaboration of a well-trained, experienced and committed statistician. A good intervention trial asks an important question, gets a reliable answer and is honestly reported. Many trials fail on one of these components and some of these failures could have been avoided with better statistical involvement. Many subject matter investigators do not understand how to effectively collaborate with statisticians. The objective here is to outline some of the important responsibilities of the statistician in intervention trials.

Book Review: Improving efficiency by shrinkage: the James-Stein and ridge regression estimators:

Abstract: tated. Also, topics ranging from A to Z are covered in each update. The editors are to be congratulated on the outstanding editorial job they did in putting this volume together. It reads as if a single author wrote it instead of the over 100 who actually contributed articles. It is, perhaps, even more surprising that there are so few typographical errors and so few missed cross-references. Still, I did have some issues with the style of presentation. Each page of text is divided into two columns. While this may work well in a nontechnical encyclopaedia, it was often a distraction here since, in many cases, complicated equations had to be extended over two or three lines. This may be especially confusing to the nonstatistician. Another quibble was the presentation of the illustrations. Some of them appeared to be of poor quality, making it dif®cult on the eyes and dif®cult to appreciate the usefulness of the graphic. A change to highresolution colour graphics as a standard format for all but the simplest of illustrations would seem a nice improvement for future updates. It would also be useful to include a listing of the notation used in the entries. Finally, in this age of increasingly more powerful desktop computers, it would be nice to see a CD-ROM version of not only this update but all future updates and the ESS itself. Seeing and listening to Svante Wold talk about the contributions of his father, Herman Wold, would be priceless. A complete inventory of the entries is not practical. A partial (and entirely subjective) listing of the entries which the readers of SMMR may ®nd of interest is the following: Aalen's Additive Risk Model, Bioavailability and Bioequivalence, Chaos, Classi®cation, Clinical Trials, Comparisons with a Control, Computer-Intensive Statistical Methods, Current Population Survey, Data Augmentation, DNA Fingerprinting, (The) EM Algorithm, Fractals, Galton±Walton Process, Generalized Additive Models, Generalized P-Values, Gibbs Sampling, Group Sequential Tests, Initial Data Analysis, Kaplan±Meier Estimator (Update), Latent-Variable Modeling, (Genetic) Linkage, Markov Chain Monte Carlo Algorithms, (Combination of) PValues, Statistical Packages, Stochastic Curtailment, and (Multivariate) Studentized Range. In summary, I recommend this volume as either a nice addition to the ESS or as a standalone supplement. Of course in the latter case, the reader cannot take advantage of the meticulous level of cross-referencing found in most of the entries. Although primarily geared towards statisticians, other scientists, especially those who collaborate closely with statisticians, will also ®nd it to be an indispensable source of information on new research and established methods. With the extensive bibliographies found in each entry, the reader will be able to locate quickly most of the key articles required for a thorough literature search or just a better understanding of the topic.

Book Review: Applied survival analysis: regression modeling of time to event data

Abstract: works I was unaware of, but I wish he had included Frank and Freidman and Tibshirani. In Chapter III, Gruber notes that CR Rao's suggestion of adding an arbitrary positive semide®nite matrix H to X0X is more general than the Hoerl±Kennard suggestion of adding a matrix of the special form that Gruber writes as QKQ0, where K is diagonal and the columns of Q are the `direction cosine' vectors for the `principal axes' of a centred X. Unfortunately, choices for H that are not of this special QKQ0 form seem to arise naturally only in the `dynamic coef®cient' situations treated in Gruber's Chapter X. Obenchain restricted interest to H matrices of the QKQ0 form and derived a variety of MSE risk results, including a unique `inferior direction,' the `2/rank' rule for limiting the extent of shrinkage, risk optimality of X0y in the direction parallel to b, etc. These are much more complete risk characterizations of shrinkage estimators than the general but vague results provided in Section 3.2. Gruber's equation (3.5.6b) for C(q) on page 163 is wrong, and C(q) should be maximized by choice of q to locate the path most likely to contain the minimum risk shrinkage estimator under normal distribution theory. Chapter VII on simultaneous estimation, Chapter IX on multivariate linear models and Chapter X on SURE and Kalman ®ltering all address potentially quite new, interesting and important applications of shrinkage estimation. A reader can easily get lost in this mass of material (about 150 pages). Luckily, each chapter starts with a wellwritten introduction, including an outline, and ends with a brief summary. Gruber's ®nal chapter (XI: Summary and Conclusion) contains sections on each chapter that I also found to be extremely helpful guideposts to both published results and topics for future research.

Book Review: Statistical issues in drug development

Abstract: which might be taken on a given issue. But I believe for some topics, the reader can tell which position(s) the author favours or rejects. He expresses his preferences in various tones from implicitly polite and suggestive to more adamantly depending upon his passion for a particular topic. The book is up to date. In fact several times Senn presents issues in a temporal context as `... at the time of the writing of this book ...'. A reader cannot ask much more from an author who covers such a wide range of controversial issues in a single text. However, this reviewer would like to see more information in Chapters 20±22 and 24 concerning dose-®nding, pharmacokinetics and pharmacodynamics, bioequivalence and pharmaco-economics and portfolio management, respectively. In all fairness, the author did indicate that these chapters were not currently his strengths and some of the chapter contents are still evolving. In preparing an extended or second edition of his book, it is my hope that the author will place a greater emphasis on these chapters. With respect to the presentation and production of the book, chapter sections are clearly labelled and logically ordered; bold-face and italic type are used effectively; graphics are used to visualize issues, although I suggest using more given the intended readership; bits of humour are sprinkled throughout the text often highlighted by a shaded box and each chapter begins with pieces of literary history which can be fully appreciated by the more well read. The book's notable ̄aw in its presentation is the number and nature of typographical errors that can be found throughout. The author is quite articulate so I am perplexed as to how these errors crept into the book. In summary, this book is a welcome addition to the literature on statistical science in drug development. It contains a wealth of information and opinion ± both theoretical and practical. It is somewhat unique in that it focuses squarely on issues, with algebraic details relegated to chapter appendices. This format will be greatly appreciated by its intended readership ± scientists and statisticians working in drug development or studying medical statistics. For a pharmaceutical industry physician's opinion see Brown. The book will also be insightful for regulatory personnel. Every pharmaceutical company and regulatory agency should have copies of this book available for its employees. In addition, the book will be quite informative in the academic arena as one of two required texts in graduate-level statistics courses in clinical trials and drug development methodology. If you are involved in the drug development and approval process, make time to read this book and consider the statistical issues contained within.