País Vasco

The study of alternative proposals for statistical modeling of patient-perceived outcomes, such as health-related quality of life, is one of the areas of great interest whose use has proven to be highly valued in different areas of Medicine (Epidemiology). Some of the extensions that we propose to analyze within this area are: a) beta-binomial regression model; b) mixed-effects beta-binomial regression model that allows for any possible correlation structure in the data by random effects; c) multidimensional beta-binomial regression model; development of joint models for longitudinal variables with beta-binomial distribution.

Clinical prediction models play a fundamental role in decision making for hospital clinical practice. The development and validation of predictive models with a high predictive and discriminative capacity would allow medical practitioners to make decisions with a good level of certainty. The development of predictive models is based on statistical modeling, where the characteristics of the response variable, the relationship between predictors and response, longitudinal data and the treatment of non-random data losses are factors that determine the type of models to be used and their validity. The development of reliable and useful clinical prediction models requires prior experience in the statistical modeling of outcomes, both crude outcomes such as mortality or survival, and patient-perceived outcomes, and in the statistical treatment of loss to follow-up in longitudinal studies. Moreover, the real success of the development of predictive models lies in their applicability in clinical practice, for which it is necessary to implement them in computer tools with an intuitive graphical interface that is easy to use and manage, so that the results are obtained automatically and transparently for the clinician and can be used reliably in daily decision making.

Complex survey data are increasingly used today, especially in the context of official statistics. In this framework, complex survey data are usually obtained by sampling the population of interest for the survey, following some particular complex sampling design. This sampling process can be carried out in one or more sampling stages, for which the combination of techniques such as stratification and clustering is a very common practice. One of the most special features of complex sampling data, as compared to data obtained from simple random sampling, is that each individual in the population has a probability (other than 0) of being included in the sample. In this context, the application of usual statistical methods is not trivial, and therefore, it is necessary to develop new approaches that take into account the complex design from which the data have been obtained.

In the analysis of the genetic basis of infectious diseases of viral etiology, it is important to develop tools for the detection and analysis of polymorphisms that can be used as markers for disease progression. Data analysis and statistical modeling in this context requires the treatment of small samples, due to the low frequency of occurrence of some genes.

Mathematical modeling of the natural history of neurological diseases by means of stochastic simulation models with discrete events is a current topic of great interest in epidemiology. Knowing the appropriate distribution for each variable and calculating its parameters from the primary data poses difficulties due to the very nature of the variable. The validity of the models depends not only on the original data, but also on the distributions applied and their parameterization. In this sense, our work aimed at the mathematical representation of diseases requires the use of statistics so that the models meet the objective of translating epidemiological data into mathematical language.

Among the current lines of research of this node (see poster) are:

• Development of regression models based on the beta-binomial distribution for multidimensional and longitudinal modeling of patient-perceived outcomes (PRO).
• Development of a theoretical framework for joint modeling of PRO and time-to-event data based on non-exponential family distributions, especially the beta-binomial distribution.
• Development of predictive models of clinical course, mortality, survival or changes in health-related quality of life. Creation of technological tools for use in hospital clinical practice.
• Proposal of variable selection for the development of predictive models with data from complex design surveys.
• Development and comparison of scoring methods for questionnaires.
• Discriminative capacity in survival models.
• Analysis and evaluation of patient-centered strategies, interventions and clinical outcomes.
• Detection of genetic markers for viral disease progression using statistical modeling methods.
• Mathematical modeling of the natural history of neurological diseases using stochastic simulation models with discrete events.

We believe it is relevant to mention that, in addition to the methodological objectives that we have previously mentioned in each of the areas of interest in this node, many of these objectives have arisen as a result of research in the different areas of knowledge that justify them and that, in our opinion, should imply an interest in transferring this knowledge to society in general and to the different fields of application of Biostatistics in particular. Therefore, our main objectives are as follows:

(a) Dissemination of statistical techniques developed through publications of a general nature or those containing applications of the proposed methodology in the different fields of Medicine.

(b) Dissemination of statistical techniques developed through summer courses or general courses in hospitals or medical institutions.

(c) Organization of conferences for the dissemination of developed statistical techniques oriented to a non-specialized public.

Development of software that can be used in a friendly way by all those professionals of Medicine and Biology interested in the application of the developed statistical techniques.