Correspondence analysis is a multivariate method for studying association between categorical variables. It transforms a multidimensional data table (contingency table, Burt table etc.) into a graphical display. Together with accompanying numerical indicators, that display is the basis for analysing the studied phenomenon. Examples of application of correspondence analysis will be presented from banking, socio-economic analyses, diabetes screening and genetics.
Slides: PDF (Part 1, in Slovenian), Slides: PDF (Part 2, in Slovenian), Slides: PowerPoint (example of application)
DOSE FINDING STUDIES IN CANCER
6 November, 2006
Prof. John O'Quigley, PhD
Lancaster University, Department of Mathematics and Statistics
The purpose of a Phase I dose finding study is to identify the so-called MTD (maximum tolerated dose) from a range of potential doses. For cancer and other chronic conditions such as HIV the idea is to give as high a dose as possible without inflicting unacceptable side effects known as dose limiting toxicities (DLTs). The only sensible, and workable, definition that we can reasonably give to the concept MTD is that of a percentile from some unknown dose-toxicity distribution. The statistical purpose is then to identify this percentile in as efficient a way as possible bearing in mind some practical constraints such as not deliberately either undertreating or overtreating the patients in the study. Ideally, all patients in any study would be treated at the MTD itself. If this were possible though there would be no need for the study! Nonetheless this idea can provide guidance. In clinical practice this simple situation becomes very quickly much more complex due to patient heterogeneity. More recently dose finding studies also include some measure of efficacy and we discuss how to address these kinds of study.
Slides: PDF
COMPUTATIONAL GENOMICS
24 October, 2006
Prof. Blaz Zupan, PhD
University of Ljubljana, Faculty of Computer and Information Science; Baylor College of Medicine, Department of Molecular and Human Genetics, Houston
A standard definition of a phenotype associates it to the organism's morphological, biochemical or physiological properties. From an engineering perspective, a phenotype describes the state of the organism at a distinct time: if the genotype and the environment are inputs, then the phenotype is the output of the system under study. Observations of relations between environmental/genetic changes and corresponding phenotypes provide ground for studies in functional genomics. Yet, if phenotypes are limited to a certain manifestation and report only on the effects of specific type (e.g., growth, sporulation, etc.), they are insufficient for genome-wide studies pertinent to system's biology. We argue that for systems biology classical phenotypes should be complemented with surrogates that can encode the state of the entire organism and provide increased resolution. In the talk I will present a number of recently proposed global phenotypes, discuss about computational tools for their analysis and talk about their utility in functional genomics.
Slides: PDF
APPLIED STATISTICS 2006 CONFERENCE
From 17 to 20 September, 2006, the Applied Statistics 2006 international conference was held in Ribno near Bled. It was traditionally interesting and pleasant; our distinguished biostatistical guest was Professor Per Kragh Andersen.
PROPOSALS FOR SAMPLE SIZE CALCULATION PROGRAMS
13 June, 2006
Assist. Prof. Harald Heinzl, PhD
Medical University of Vienna, Core Unit for Medical Statistics and Informatics, Section of Clinical Biometrics
Sample size calculation is a vital issue in the planning stage of a prospective two-armed clinical trial. In order to study a rare disease some researchers might consider performing a notably underpowered clinical trial. They might trade off the risk for a statistically not significant result with the chance for a significant one. In the latter case, they could expect that any discussion concerning the small power of the study would be suppressed in the case of a statistically significant result. Actually, the latter argument has to be used with great care in the case of directional two-sided testing in small-sized studies.
A directional two-sided test can be considered as a three decision problem. Besides the null hypothesis there are two one-sided alternatives each of them favouring either of the two therapies. Now the well known Type I error (falsely rejecting the null) and Type II error (falsely stick to the null) are accompanied by another rather unpleasant type of error, that is falsely favouring a therapy which is truly worse. Consequently, this kind of error is called Type III error. The so-called q-value is defined as conditional probability of a Type III error given statistical significance.
We suggest that sample size calculation programs should be able to compute the Type III error probability values and q-values in order to provide additional useful support in the planning phase of a clinical trial. A further requirement on sample size calculation programs can be established which concerns the feasibility to compute the detectable effect size of the so-called auxiliary alternative for a given sufficiently large power.
Photos: 1, 2, 3, 4, 5, 6, 7, 8, 9
NUMBER OF STEM CELLS REPOPULATING THE MARROW FOLLOWING TRANSPLANTATION
25 May, 2006
Prof. John O'Quigley, PhD
Lancaster University, Department of Mathematics and Statistics
It was once thought that very few, possibly no more than one or two, stem cells were responsible for repopulating the marrow following transplantation. Techniques of cell biology were able to show this hypothesis to be unlikely. In this talk we consider some simple statistical approaches which can be used to investigate the number of stem cells repopulating the marrow. The data consist of pairs of donors/patients both for whom some indirect measurement was taken of a marker on the cell. If all of the donor stem cells were marked in some way then the patient would also be marked subsequently in the same way. If, say, half of the donor stem cells were marked, then, making some simple realistic assumptions concerning the dynamics of cell replication, we would anticipate half of the patient's cells to be likewise, all the more so the greater the number of cells involved in the repopulation process. Should few cells be involved we might anticipate greater variation between the marked percentage in the donor group compared with this percentage in the patient group. In this talk we investigate how to use this variation to obtain efficient estimates of n, the number of stem cells involved in repopulation.
Slides: PDF
References: MEDLINE (PubMed), JSTOR
STATISTICAL METHODS FOR MICROARRAY DATA ANALYSIS: AN OVERVIEW WITH EXAMPLES FROM CANCER RESEARCH
24 April, 2006
Lara Lusa, PhD
National Institute for the Study and Treatment of Tumors, FIRC Institute of Molecular Oncology (IFOM), Milan
Gene expression microarrays are nowadays a widely used tool in biological and medical research. The peculiarity of experiments using microarrays is that the expression of thousands of genes can be simultaneously quantified, while the number of subjects measured in each experiment is still often very small. Microarrays have been used to discover disease subclasses that were previously not known, to compare two or more phenotypes in terms of gene expression, identifying genes that are differentially expressed between them, and to develop classifiers based on expression profiles that predict disease outcome. Many statistical methods have been developed (or, often, rediscovered) for microarray data analysis. The aim of this talk is to give a critical overview of these methods, using examples from cancer research.
Slides: PowerPoint
Photos: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13
TELECARE AND HOME HEALTH AT A DISTANCE - POLITICS AND EXPERIENCES OF THE UK AND IRELAND
12 April, 2006
Malcolm Fisk, PhD, MA, BSc, DipSoc, MCIH
Insight Social Research Ltd., Newport, Wales
Mr. Fisk presented experiences in setting strategic policies and establishing networks for higher quality of life of older people, people with reduced capabilities and the sick in home environment using modern organisational models and telecommunication and information technologies. The lecture provided an encouraging opportunity for taking more decisive steps in this field here in Slovenia.
The lecturer was invited to Ljubljana by the Slovene Medical Informatics Association. He is a sociologist and expert in the field of use of modern technologies for older people and people with reduced capabilities, and Chair of the National Partnership Forum for Older People in Wales, established by the Welsh Assembly. His responsibility there is preparation of strategies for improving life of older people in Wales. He is also the Vice Chair of the British Telecare Services Association (whose members include over 300 home assistance centers and about 20 equipment suppliers) representing equipment supplier interests. He is also Managing Director of Insight Social Research Ltd., a research consultancy serving government institutions, local authorities and charities across the UK and Ireland.
The lecture was organised by Drago Rudel, PhD, and held at the Faculty of Medicine in Ljubljana.
THE thorax.6 MEDICAL INFORMATION SYSTEM
28 March, 2006
Tomaz Stupnik, MD, graduate student of statistics
Ljubljana Clinical Centre
While thorax v.3 from 1998 was just a prototype, thorax.6 is a fully functional physician information system characterised by a host of buzzwords: Oracle 10g, record versioning, multilayer transaction design, intranet, https, biometric authentication, XML, HL7 etc. Unlike the information systems of the majority of Slovenian hospitals, which were developed by IT people more or less exclusively following requests and needs of the management, thorax.6 was designed to meet the needs of the physicians - as a platform for testing statistical and information technologies in treatment and patient care.
In connection with statistical software packages (like SAS and R) and data mining, the final aim of thorax.6 is simplified collection of research data, as well as a standardised an maximally automated treatment process with integrated quality control.
The lecture will present thorax.6 and outline possible directions of its further development.
PREDICTIVE CLUSTERING TREES WTH APPLICATIONS TO BIO[MEDICAL] INFORMATICS
22 February, 2006
Prof. Saso Dzeroski, PhD
Jozef Stefan Institute, Department of Knowledge Technologies
Predictive clustering trees are a data-analysis approach that combines clustering with predictive modeling. They enable conceptual clustering, which yields cluster descriptions in addition to cluster membership, while predicting several target variables at the same time.
The lecture will present predictive clustering trees, methods for learning them from data, and examples of their application in the field of bio[medical] informatics (gene function prediction, microarray analysis).
GOODNESS OF FIT AND EXPLAINED VARIATION IN REGRESSION MODELS
24 January, 2006
Prof. Janez Stare, PhD
University of Ljubljana, Faculty of Medicine, Institute of Biomedical Informatics
(inaugural on the occassion of being elected full professor)
Fit and explained variation of a regression model are often misunderstood, with many statisticians still seeing R2 as a measure of goodness of fit. Therefore, I will first explain what one should understand under these notions, and then give a short overview of my work in the area. This contains definition of explained variation for survival models, assumptions of linear models for censored data, and goodness of fit of relative survival models.