Investigation of causal relationship
Introduction
This document is aimed to present an outline for the causal relationship investigation, i.e. introduce the use case requirements, summarize ideas so far and present some “real problems/questions” where the technology is needed.
Investigating causality relationships between information concepts like gene, symptoms, disease, biological process, environmental factor, etc. can be used to generate new hypothesis to explain observed correlations and identify not yet known opportunities.
The causality relationships could be derived from existing data sources or generated using new methods. [Q1, Do we have the expertise to develop and asses these methods?]
Development Plan
Iteration 1: (2010-06-30) No complicating functionality and limited depth, e.g. no ranking but focus on finding known relations..
Iteration 2:¨(2010-12-31) Include trivial ranking and if possible some kind of feedback.
Iteration 3: (2011-06-30) Include requirements of high on quality and relevant quantity. This is when the machine learning will be fully integrated in the app.
Read more in draft outline for work: Causal Relationship Investigation outline of work.pdf
Use case alignment
Improve the knowledge about disease and patients (chapter 4.1 in requirement document)
Epidemiologists work with project teams to improve the understanding of the etiology and pathophysiology of COPD. Identifying causal relationships between environmental factors, genetic heritage, biology in disease processes, patient groups and outcomes are important aspects of our COPD understanding.
Example query: It has been shown that individuals with high concentrations of white blood cells (WBC = leukocytes) have lower lung function than individuals with low WBC. This is true both in smokers and non-smokers. Moreover, in studies with repeated measurements of lung function, is has been shown that men with high WBC values have a more rapid deterioration of the lung function.
What is the nature of these associations? At least three hypotheses can be formulated, and they are not mutually exclusive.
- WBC can destroy the lung tissue.
- Destruction of the lung tissue could result in increased WBC, maybe as a physiological adaptation to the tissue injury.
- A third factor (such as smoking, air pollution, physical inactivity, obesity etc) could both reduce lung function and increase WBC concentrations.
Is there any evidence supporting one or more of these causal links in the literature or in available databases?
Evaluation: The hypotheses generated by causal relationship investigation will be evaluated scientifically by a team of scientists at ?AstraZeneca.
Clinical project team working on a new target (chapter 4.2 in requirement document)
The clinical project team (CPT) has to respond with accuracy to findings and external questions. They need to be on top of all available information to handle the uncertainty that exists in Early Clinical Development, e.g. about risk factors etc. for drug class and/or biological targets including interpreting previously performed studies internally and externally with characteristics of value for this program.
Example query: Explain the relation of Cyclooxygenase inhibitors and cardiovascular events (Background information to the scientific question can be found in this article: Biological basis for the cardiovascular consequences of COX-2 inhibition: therapeutic challenges and opportunities) Evaluation: The hypotheses generated by causal relationship investigation will be evaluated in comparison with existing knowledge and outcome from random indexing.
Identifying biomarkers and target mechanisms (chapter 4.3 in requirement document)
Not in scope for causal relationship investigation.
Signal evaluation of adverse events (chapter 4.4 in requirement document)
A safety expert manages adverse event reports and defines the safety profile of our drugs and targets. Part of that is to define the underlying COPD patient profile from a safety perspective, which links to epidemiology – co-morbidity, concomitant medication, adverse event profile in placebo patients and during use of current drugs etc. (see the first case study). An adverse event is any untoward medical occurrence in a patient or subject administrated a drug and which does not necessarily have a causal relationship with this treatment. Adverse events are collected both in clinical trials and post marketing of a drug. When the safety experts receive an adverse drug event reports they need to evaluate the medical occurrence and verify if there is a causal relationship between the drug and the event. One method for drug causality assessment is called the RUCAM and consists of 7 criteria of which 4 is particularly suitable for this method; risk factor(s) for drug reaction, concomitant drug(s) and non-drug related causes of event, previous information on the drug.
Example query: (Anders is this question relevant?) What causal relationships exist between "Tyrosine kinase inhibitors" and "visual disturbances" (Peter is this the question you used for ?BriefMe? Can we use your results for the evaluation?)
Evaluation: The hypotheses generated by causal relationship investigation will be evaluated through competition with result from internal experiment (?BriefMe).
Collaborative ideas to address requirements
Generating hypothesis with help of causal relationship investigation will require methods for:
- Identify casual relation chains (5-8 in order)
- Summaries the findings personalized for the user, e.g. good ranking
Could the technology be made available in a “simpler” form to give “advanced advice” in different situations (pop-up functionality).
Questions to discuss:
- Do we have the expertise to develop and asses these methods?
- Evaluation strategy, can we use two different methods for evaluation to be able to adjust for objectivity?
Meeting details
Organize phone call to start hypothesis generation/validation collaboration (due 15 February AZ)
Requirements summary and data repository