Catálogo de publicaciones - libros

Studies of prognosis have received rather little attention by those carrying out systematic reviews. Such reviews are increasingly being attempted but the poor quality of published ‘primary’ studies leads to serious difficulties. Thus there have been calls for such reviews to be based on individual patient data (IPD) but such studies are as yet rare. We consider the advantages of IPD for reviews of prognostic variables and describe in detail a systematic review of microvessel density counts as a prognostic variable for patients with non-small cell lung cancer. We show that such a study is feasible, but note that it may not be cost-effective to attempt to obtain all relevant data.

Various statistical methods to analyse prognostic marker studies are available and are used in practice. Issues of multivariable model building will be discussed in the framework of regression models and classification and regression trees (CART). It is shown that the choice of one specific statistical method has a strong influence on the results and, therefore, on the interpretation of a prognostic marker. Within regression models we compare the full model with models obtained by backward variable selection considering also transformations of continuous covariates. We discuss problems caused by the uncritical application of CART and outline advantages of small and simple trees. Furthermore, we show how to form risk groups with different prognoses and we illustrate the necessity to validate results in an independent study. Data of two breast cancer studies are used for illustration.

Prognostic markers can help to identify patients with different risks of specific outcomes, facilitate treatment choice, and aid patient counselling. Unfortunately, within any given disease area, the wealth of conflicting and heterogeneous evidence makes it difficult for the clinician to ascertain the overall evidence about specific markers and how to use them in practice. The application of formal methods (e.g., a systematic review and meta-analysis) of obtaining and synthesising evidence is therefore greatly needed in the prognostic marker field. However, in this chapter we illustrate and discuss the reasons why currently poor standards of design, clinical relevance, and reporting in primary studies limit statistically reliable and clinically relevant evidence-based results for prognostic markers. These problems add to those issues for the statistical analysis in primary studies that are discussed in another chapter in this volume. To help overcome the problems we highlight guidelines for conducting and reporting primary prognostic research, and we particularly discuss why the availability of individual patient data would help realise the evidence-based use of prognostic markers in clinical practice.

We propose an approach to monitoring unanticipated adverse events in a clinical trial. For a specific type of event, the methods allow the first occurrence, the first few occurrences, or an elevated rate to trigger a formal monitoring plan to identify whether that event occurs more frequently in the treated than in the control arm. The methods can apply either to rare events or to common events not expected to occur at an elevated rate in the treated group. We offer some simple models, emphasizing that, by the very nature of its rarity, a rare event is quite difficult to monitor.

Pharmacovigilance aims at detecting adverse effects of marketed drugs. It is generally based on a Spontaneous Reporting System (SRS) that consists of the spontaneous reporting, by health professionals, of events that are supposed to be adverse effects of marketed drugs. SRS supply huge databases, the human-based exploitation of which cannot be exhaustive. Automated signal generation methods have been proposed in the literature but no consensus exists concerning their efficiency and applicability due to the difficulties in evaluating the methods on real data. The objective is to propose SRS modelling in order to simulate realistic data sets that would permit completion of the methods’ evaluation and comparison. In fact, as the status of the drug-event relationships is known in the simulated data sets, generated signals can be labelled as “true” or “false.” The spontaneous reporting is viewed as a Poisson process depending on: the drug’s exposure frequency, the delay from the drug’s launch, the adverse events’ background incidence and seriousness, and the reporting probability. This reporting probability, quantitatively unknown, is derived from the qualitative knowledge found in the literature and expressed by experts. This knowledge is represented and exploited by means of a set of fuzzy rules. Then, we show that the SRS modelling permits to evaluate the automatic signal generation methods proposed within pharmacovigilance and contribute to generate a consensus on drugs’ postmarketing surveillance strategies.

Study of multivariate data in situations where a variable of interest is unobservable (latent) and only measured indirectly is widely applied. Item response models are powerful tools for measurement and have been extended to incorporate latent structure. The (log-linear) Rasch model is a simple item response model where tests of fit and item parameter estimation can take place without assumptions about the distribution of the latent variable. Inclusion of a latent variable as predictor in standard regression models such as logistic or Poisson regression models is discussed, and a study of the relation between psychosocial work environment and absence from work is used to illustrate and motivate the results.

Early stopping of clinical trials either in the case of beneficial or deleterious effect of treatment on quality of life (QoL) is an important issue. QoL is usually evaluated using self-assessment questionnaires and responses to the items are combined into scores assumed to be normally distributed (which is rarely the case). An alternative is to use item response theory (IRT) models such as the partial credit model (PCM) for polytomous items which takes into account the categorical nature of the items. Sequential analysis and mixed partial credit models were combined in the context of phase II noncomparative trials. The statistical properties of the sequential probability ratio test (SPRT) and of the triangular test (TT) were compared using mixed PCM and traditional average scores methods (ASM) by means of simulations. The type I error of the sequential tests was correctly maintained for both methods, the mixed PCM being more conservative than the ASM. Although remaining a bit underpowered, the mixed PCM displayed higher power than the ASM for both sequential tests. Both methods allowed substantial reductions in average sample numbers as compared with fixed sample designs. Overlapping of item category particularly affected the ASM by inflating the type I error and power. The use of IRT models in sequential analysis of QoL endpoints is promising and should provide a more powerful method to detect therapeutic effects than the traditional ASM.

A parametric degradation model based on the Wiener process is studied. The best unbiased estimators are constructed for the parameters of this model.

Agreement studies, where several observers may be rating the same subject for some characteristic measured on an ordinal scale, provide important information. The weighted Kappa coefficient is a popular measure of agreement for ordinal ratings. However, in some studies, the raters use scales with different numbers of categories. For example, a patient quality of life questionnaire may ask ‘How do you feel today?’ with possible answers ranging from 1 (worst) to 7 (best). At the same visit, the doctor reports his impression of the patient’s health status as very poor, poor, fair, good, or very good. The weighted Kappa coefficient is not applicable here because the two scales have a different number of categories. In this paper, we discuss Kappa coefficients to measure agreement between such ratings. In particular, with R categories of one rating, and C categories of another, by dichotomizing the two ratings at all possible cutpoints, there are ( R −1)( C −1) possible (2×2) tables. For each of these (2×2) tables, we estimate the Kappa coefficient for dichotomous ratings. The largest estimated Kappa coefficients suggest the cutpoints for the two ratings where agreement is the highest and where categories can be combined for further analysis.

Frailty models are becoming more and more popular in multivariate survival analysis. Shared frailty models in particular are often used despite their limitations. To overcome their disadvantages, numerous correlated frailty models were established during the last decade. In this study, we present different variants and extensions of the bivariate correlated gamma frailty model with special focus on the analysis of bivariate breast cancer onset data from the Swedish Twin Registry. Points of discussion are parametric versus semi-parametric approaches, the inclusion of observed covariates, dependence of the frailty distribution on observed covariates, and a possible cure fraction.