 |
|
|
|
|
|
Research projects
The
list of research projects beneath is based on the BMMS research plan
for 2006-2010, found in the evaluation report from the thematic
research
area Norevent: Survival and Event History Analysis 2001 - February 2005.
Survival
and event history analysis
Epidemiology
Statistical
methods for genomics
Modeling
infectious disease
Model
selection, likelihood methods and nonparametric Bayesian methods
Survival
and event history analysis
As it was in Norevent, survival
and
event history analysis is
still an important area in our
research. Especially we emphasize the following fields:
Frailty
This denotes the (often
unobserved)
individual
heterogeneity which
plays an important role in models in survival analysis. In fact, a
whole statistical literature has grown up around the frailty concept.
In cooperation with groups at the Cancer Registry we plan to focus on
the following specific questions:
- Growth of
breast cancer as illuminated by data from
the screening
program. The Norwegian screening data gives an unprecedented
possibility of estimating the individual variation in rate of tumor
growth. There are indications of great individual variation in growth
rate. The project is presently well under way, being the PhD project of
Harald Weedon-Fekjær.
BMMS
researchers involved: Weedon-Fekjær,
Tretli, Aalen.
- Why do
incidence rates of cancer deviate from that
predicted by
the carcinogeneisis models? A major reason is likely to be individual
variation. This is follow-up of work on testicular cancer (see
“Research contributions”), but the method shall be
applied
to other cancer forms, e.g. colon cancer.
BMMS
researchers involved:
Aalen,
Tretli, Moger.
Causal modeling
This denotes the idea that models
for
longitudinal data analysis,
including event history models, should be based on an explicit causal
thinking. This idea has a strong international representation
presently. We have made an important advance by defining a model for
dynamic path analysis. Classical path analysis does not explicitly
include the time aspect, but our new model does so by using counting
process theory combined with an additive hazard model. More
specifically, we have the following aims.
- Completing the
ongoing
project where causal modeling
has been
applied to survival data with repeated occurrences and to survival data
with time-dependent covariates. In particular, time-dependent
covariates have been a thorny issue in survival analysis, and we
believe our approach throws new light on how to handle such covariates.
Two papers are presently being submitted.
BMMS
researchers involved:
Aalen,
Fosen, Borgan, Ferkingstad, Weedon-Fekjær.
- Secondly, we want to
compare our new approach with
the well
established Marginal Structural Model (MSM). We believe our alternative
model gives more detailed information on the relationship between the
various processes involved.
BMMS researchers involved: Aalen, Borgan,
Ferkingstad.
- Thirdly, we shall
continue
the work on estimating
transition
function for multistate models (e.g. Markov models). This is of
relevance, e.g. for estimating prognoses for patients, and it is
closely related to ongoing work on causal modeling.
BMMS
researchers involved: Gunnes,
Aalen, Borgan.
Book project
Aalen, Borgan and Gjessing are
writing a
book for Springer Verlag with
the title: Event History Analysis – a Process Point of View.
The
book partly contains result from very recent research and will be
published in 2006.
Efficient
methods
for sampling and analysis from large registries
The aim of the project is to
develop
case-cohort and case-control
methods for situations where one has a very large amount of family
survival data. The data can for instance come from Scandinavian
population registries, such as the Medical Birth Registry of Norway and
the Swedish multi generation registry. In these cases, traditional
Cox/frailty cohort analyses will be computationally very time
consuming, if at all possible to accomplish. Methods where information
on case families and only a sample of the control families is used will
clearly be beneficial. The project involves collaboration with the
Department of Medical Epidemiology and Biostatistics, Karolinska
Institutet in Stockholm, as well as the Department of Biostatistics,
University of Washington.
BMMS
researchers
involved: Moger,
Aalen, Borgan, Samuelsen.
Epidemiology
Cancer
epidemiology
Several of the statisticians at
the
Cancer Registry are active in
Norevent. They are involved in several major projects in the Cancer
Registry. Studies are being performed on the influence of a number of
factors on cancer incidence, such as height and weight, sun exposure,
and conditions in foetal life, childhood and puberty, and on
occupational risks of cancer as well as prediction of cancer incidence.
BMMS
researchers
involved: Tretli,
Haldorsen, Møller.
Cancer
screening
Norevent researchers are involved
in
analysing data from two major
screening projects at the Cancer Registry: Breast cancer screening and
cervix cancer screening. These projects have a very high international
standard. New statistical methods have to be developed, as well as
existing methods adapted to handle the challenges of evaluating cancer
screening. Mammography screening detects breast cancer tumours bigger
than 2 mm in diameter. Combining screening data, population data and
reports on cancer cases, we build mathematical models for studying
tumour growth on a population level. The main results will be estimates
of breast cancer tumour growth, including estimates of the
heterogeneity in the population. The project will use mathematical
models and simulation to study breast cancer tumour growth and related
central parts of breast cancer screening (see also above under
Frailty). Data from the cervix cancer screening are also analyzed by
methods from
survival analysis.
BMMS
researchers
involved: Tretli,
Weedon-Fekjær, Nygård, Aalen.
Estimation of
the
hidden population of illegal drug use
SIRUS – The Norwegian
Institute for Alcohol and Drug
Research. In
2002 a project was initiated to model illegal drug use in Norway,
including the estimation of problematic drug use. A stochastic model
including influx, change to problematic drug use, stays in prison and
treatment, and outflow (death, end of drug use) will be established. A
pilot Markov model based on present data and a lot of assumptions is
the first step of model building. The work will be integrated with
reporting of an indicator for problematic drug use to EMCDDA, the
European Monitoring Centre for Drugs and Drugs Addiction in Lisbon. In
2003 a so called Multiple Indicator Method was tried out using the 19
regions as the geographical unit and in 2004 a survey of illegal drug
use and intravenous drug use in the municipalities was evaluated as a
possible source of information for prevalence estimation of intravenous
drug use.
BMMS
researchers
involved: Ellen
Amundsen.
Non-trivial
designs in clinical studies
Present clinical studies frequently involve non-trivial designs that
give complex data structures. Multilevel data structures might appear
when data are collected repeatedly for each patient. Classical
statistical modelling techniques, which assume independence of the
observations, are not appropriate for data with complex structures. The
aim of the project is to apply modern statistical methods to clinical
studies. We will address the problem by studying new methodology and
show it can improve our research in this area, by applications to a few
specific problems. We will show that by extracting more information out
of the data, we will get more insight into the clinical problem. The
project is collaboration between statisticians at UiO and clinical
researchers at Rikshospitalet-Radiumhospitalet.
BMMS researchers involved: Frøslie, Veierød,
Røislien, Laake
Statistical
methods for genomics
The tools for collecting genetic
information, including detailed
knowledge of the genome, for individual persons are being developed
very fast. As genetic information on individuals is being collected to
an ever-increasing extent, it becomes necessary to relate this
information to risk of developing disease, and to prognosis of disease.
This requires further development of statistical methodology, not least
in the field of survival analysis.
Statistical
methods
for high dimensional genomic data
Modern high-throughput
technologies
produce data of an extraordinary
high dimension, for example in microarray experiments and genome-wide
SNP genotyping. This high dimensionality and the associated problem of
multiple testing of thousands of hypothesis simultaneously pose serious
methodological challenges. Several methods have been proposed to
account for multiple testing, but there is no consensus and a lack of
understanding of the power of the various methods. A problem with all
available methods is their difficulty in coping with dependencies among
covariates (SNPs or gene expressions). As the networks of interacting
genes are better understood, a challenge is to device methods that take
into account various types of causal ordering of genes. A number of
projects have been started up in this field.
BMMS
researchers
involved: Frigessi,
Borgan, Liestøl, Nygård,
Scheel, Kulle, Ferkingstad.
Association
analyses
in genetic epidemiology
The focus of the project is to
develop
statistical models and software
for analyzing associations between disease and candidate genes. The
candidate genes are represented by SNP markers, and we estimate
associations with the full SNP haplotypes from unphased data, not only
individual SNPs. In the near future the project will address issues
such as power to detect gene-environment interactions and the relative
strength of case-control designs versus case-parent triad designs when
SNP phase must be reconstructed from data using statistical
methodology. The project is in collaboration with: Prof. Rolv Terje
Lie, Section of Epidemiology and Medical Statistics, University of
Bergen, Dr. Wilcox, National Institute of Public Health, North
Carolina, Dr. Jugessur, University of Iowa, Dr. Skare, Norwegian
Institute of Public Health. The
software under development under this project, HAPLIN, is freely
available at http://www.uib.no/smis/gjessing/genetics/software/haplin/
BMMS
researchers
involved: Gjessing,
Kulle, Egeland.
Linkage
analysis
We are studying methods for
linkage
analysis of complex pedigrees. In
particular, we shall consider applications to epilepsy and X-linked
disorders. This is done in cooperation with colleagues at Medical
Genetics, Ullevaal University Hospital, as well as international
collaborators.
BMMS
researchers
involved: Egeland.
Modeling
infectious disease
The common statistical models in
epidemiology and clinical science are
constructed to handle non-infectious diseases. On the other hand,
infectious diseases are gaining more attention also in Norway due to
new threats like reemerging HIV infections, resistant tuberculosis,
SARS and bird flu. Internationally there exist a number of mathematical
and statistical methods that can be of use in evaluating outbreaks, or
study the spread of infectious disease, but in Norway there is little
knowledge concerning these methods and there are few people with
mathematical or statistical background working in the field, compared
to other countries. Norevent aims to change this through education and
through initiating research.
Structure and
dynamics of sexual networks for the
spread of sexually transmitted infections
Sexually transmitted infections
(STIs),
including HIV/AIDS represent a
serious and increasing global problem with major public health
consequences. In Norway the Chlamydia incidence is high and increasing
in young adults; the infection is the leading cause of infertility
among women in the country. We shall study mathematical models for
STIs. Such models for infectious disease in general are used
increasingly as health policy tools in risk assessment, disease
forecasting, and to study the impact of intervention strategies. The
project has received financial support from the Norwegian Research
Council for the period 2005-2008. The project is a collaboration
between the University of Oslo and the Norwegian National Institute of
Public Health, with participation of leading groups in mathematical
infectious disease epidemiology at Imperial College, London, and the
Karolinska Institute Stockholm.
BMMS
researchers
involved: De
Blasio.
Bayesian
back-calculation of HIV incidence
Back-calculation has been the
traditional approach for estimating the
incidence of HIV infection from observed AIDS cases. In recent years
new treatments make it possible to prevent AIDS disease occurring for
many years. We suggest that instead back-calculation can be based on
time to HIV diagnosis (i.e. first positive test). Bayesian methods are
ideal for this purpose. This is in collaboration with the MRC Unit of
Biostatistics in Cambridge, U.K.
BMMS
researchers
involved: Aalen.
Model
selection, likelihood methods and nonparametric
Bayesian methods
New statistical methods for
complex data
from medicine and other fields
are being developed. Professor Nils Lid Hjort at the Department of
Mathematics is a major international name in statistical research, and
he is also engaged in survival analysis and other methods for medical
research. He is cooperating with other Norevent researchers to develop
novel methods for handling complex datasets. Some of his present
projects:
- Model selection and
model
averaging, book project for
Cambridge
University Press, with Hjort and Gerda Claeskens (University of
Leuven). The intention is to provide both classical and new theory in a
coherent single volume. Will have "modern applications", e.g. to
various regression situations, including models for event history
analysis.
- Nonparametric
Bayesian
methods, for survival and
event history models.
BMMS
researchers involved: Hjort,
Frigessi, Scheel.
|
|
