SBIR-STTR Award

In the drive to increase quality and decrease health care costs for elderly populations, providers are collecting data on outcomes of medical procedures to guide decision making, identify opportunities for process improvements, and demonstrate the value of interventions. The key barrier to using this collected information is the lack of powerful user-friendly programs to facilitate data analysis. Conventional statistical analysis applications have proven inadequate to identify important patterns that may be hidden in the recorded demographic procedural, and health outcomes data elements. Also, the application of conventional statistical techniques is difficult because of the enormous combination of recorded variables. We will apply proprietary neural networks (NN) techniques to develop user-friendly data analysis systems which will facilitate exploration of complex datasets for outcome studies. The novel MSI approach is based on transformation of a decision tree into a layered NN. After training, this special NN can also be used to generate a fixed set of expert system compatible rules suitable for utilization in guiding clinical decision making and employment in public health population studies. To demonstrate concept feasibility, we will focus on hip replacement data that has been collected by the Henry Ford Health System.Awardee's statement of the potential commercial applications of the research:Results will be integrated into a complete hip replacement outcomes data analysis computer system. Any algorithms developed could be modified to assist in the analysis of other common procedures and conditions afflicting the elderly such as cataract surgery, knee replacement, and asthma and diabetes management. This computer system will be adopted by the widest possible audience because it will be far easier to use than conventional statistical packages and by virtue of being designed for compatibility with standard outcomes tools. Therefore, the whole hospital management profession will be a large potential market for incorporation of this technology.National Institute on Aging (NIA)

Mandala Sciences (MSI) CODA project has 2 main objectives: (1) develop analysis tools to test hypotheses regarding effectiveness of surgical procedures and patient outcomes and (2) generate proprietary decision support prediction models for hip and knee replacements. MSI hybrid Neural Network/Expert System methodology uses an Entropy NN TM structure which has the innovative ability to generate a rule base. The discovered rules will be used to create "portable" Expert System predictive modules. Phase II progress is built upon successful MSI collaboration with Henry Ford Health System to show NN techniques can generate and evaluate prognostic models using outcomes data. Consultation with orthopedic surgeons identified 13 patient-provided variables as potential predictors of hip replacement surgery failure. An NN trained on these data predicted the 1-year post-surgical change in the patient's self-assessed pain and physical function scores. Comparison with standard statistical analysis techniques showed superior accuracy of NN-based predictions. Phase II research will generalize the product by adopting the ASTM-E-1238 interface standard for data collection from multiple sources. NN/Expert prediction models will be improved by pooling data from geographically diverse sites and field trial performance to evaluate physician-rated adoption, usefulness, and influence on their actual decision making. Proposed commercial applications: MSI will build a user-friendly, stand-alone outcomes database analysis tool. By using new proprietary neural network techniques, the system will have the predictive power of NN combined with the explanatory capabilities of an expert system. This computer system will be adopted by the widest possible audience because it will be far easier to use than conventional statistical packages, and by virtue of being designed for compatibility with the ASTM-E-1238 standard for outcomes data transmission.