The theme of the project is causal inference: How to make inferences about causation from statistical data? We are dealing with comparison of the effects of different treatment strategies used in patients with a specific disorder. This is a key issue in assuring the quality of care provided by the health care system. Usually the effect of a treatment is evaluated in controlled clinical trials. It will increasingly be possible to also make use of data from health registries in Norway to consider this. One reason is that conducting a controlled clinical trial requires a lot of resources and time. In some cases, the implementation of clinical trials may also lead to ethical problems.
To quantify treatment effects from registry data leads one quickly to difficult problems because the usual summary statistics typically gives wrong answers. We have in this project worked with mathematical methods to simulate clinical trials, and with statistical methods for determining appropriate treatment effects based on registry data.
We have published several papers with new statistical methods in this field. Currently, we are developing software that will make it easier to take the methods in general use. A new program package has been posted on the statistical system R, and an article is submitted to the "R Journal". We think many could benefit from this program.
We are also working on a larger paper where we compare several key methods from the field causal inference.
We want to build a viable research group in causal modeling for clinical and epidemiological studies. One aim is to create interest and disseminate knowledge about the tools of this new area.
The main research objective is to develop new methods for caus al modeling that explicitly include time.
Dynamic path analysis:
In particular we wish to extend the concept of a DAG (directed acyclic graph) to explicitly include time. The tool is dynamic path analysis which we want to develop into a general method fo r longitudinal data. This is a combination of our own work in this area and the work of a British group (Diggle, Farewell and Henderson). Using our ideas of additive modeling Diggle et al recently presented what they termed a new approach to analyzing lon gitudinal data with missing observations. This focused on the changes (or increments) of the observed processes. Dynamic path analysis allows for assessment of direct and indirect effects, and how processes influence each other causally. Practical methods of analysis will be developed.
Dynamic treatment regimes:
We have a fruitful cooperation with the Swiss HIV Cohort Study and have been developing new tools for analyzing dynamic treatment regimes. we wish to extend our methods to answer the following qu estions: (i) what determines whether the treatment shall be started (which is a question of local independence), (ii) how is the treatment effect mediated through the effect on other parameters, like CD4 count (using tools like dynamic path analysis, dire ct and indirect effects). The methods will be developed further to handle more general dynamic treatment regimes.