Under funding from the National Library of Medicine authorized by the American Recovery and Reinvestment Act, Applied Biomathematics has undertaken a research project entitled "Compensating for uncertainty biases in health risk judgments". The project has produced several websites and programs. The long term objective of this project is to enable patients and physicians to have a clear appreciation and understanding of complex clinical data for interpreting medical test results and for choosing among possible medical interventions. Proper interpretation is commonly hindered by two problems: (1) statistical data are often presented in formats that are incompatible with known quantitative perceptual abilities in humans, and (2) recognition of uncertainty in the underlying data can often change perceptions of risks and benefits, therefore changing preferences for particular decisions. The project addresses these critical barriers to informed consent in clinical decision-making. The research has three specific aims:
1) Develop best practices for presenting statistical clinical data
The builders of the Parthenon corrected for geometric optical illusions so that the lines of the building appear straight. In much the same way, techniques will be developed that will correct for cognitive biases and convey risks in the proper light so that humans can more easily understand their implications. The efficacy of these practices will be tested by conducting computer simulations and web surveys to test the accuracy of interpretation of laypersons in situations with known probability relationships.
2) Develop algorithms using robust Bayesian analysis for imprecise data
Correct interpretation of test results and other frequency-based clinical data requires elementary application of Bayes’ rule (for instance, to estimate the probability of having a disease given that one has tested positive for it). Clinical data is by nature rife with uncertainty. In order to incorporate this underlying uncertainty into clinical interpretations, methods of robust Bayesian analysis will be used. Estimation of risk in the face of uncertainty can be made more valuable to patients with a version of logistic regression that can accept interval uncertainty in predictor variables. We will develop such methods to estimate outcome probabilities represented as intervals or bounds on distributions, rather than as point estimates when uncertainty is substantial. Decision making based on maximizing expected utility is generalized to account for imprecise probabilities with the E-admissability criterion. This approach to characterizing epistemic uncertainty and these analyses are consistent with traditional Bayesian sensitivity analysis but are considerably simpler computationally.
3) Implement the solutions as a mobile application for personal use and as an internet module for institutional use
The software will accept a wide range of data so that it may be easily used by different institutions and for different specific applications. The software will output graphical presentations of the probabilities involved that naturally incorporate uncertainty in the data and that bypass known perceptual biases concerning quantitative information. Two tasks will demonstrate the usefulness of the software. A public dataset will be assembled for the risk factors associated with a single illness, coronary heart disease, and case studies based on the professional experience of a leading cardiologist will be used to show how use of the software improves the decision-making ability of the patient.
Return to the list of Applied Biomathematics' NIH ARRA project websites and programs.