Application of Neural Net Analysis to Predict Wound Healing with Dermagraft in the Treatment of Diabetic Foot Ulcers
By Frank S. Casciani, Advanced Tissue Sciences, Inc., La Jolla CA, and M. E. Parham, Parham Analysis, Bedford, MA
Case Study
Neural Net Analyses (NNA) have been used by Advanced Tissue Sciences, Inc., (ATS), to predict the outcome of a clinical trial. ATS is a tissue engineering company utilizing its proprietary core technology to develop and manufacture human tissue products for tissue repair. ATS performed a clinical trial to establish the safety and effectiveness of Dermagraft, a metabolically active tissue engineered dermis, to heal diabetic foot ulcers. At the completion of the clinical study, NNAs were used to generate a nested series of models that simulated the outcome of the clinical trial for individual patients. The models were used to explain and visualize the complicated interactions between patient demographics and specific Dermagraft product parameters. Highly predictive models have been generated which describe diabetic ulcer healing as a complex function of product and patient characteristics.
Results
Conclusion
Model Development
The data were based on a recently concluded clinical study. The NNA used specific product and patient parameters as inputs to calculate the clinical outcome: ulcer closure by 12 weeks. Outputs included healing rates (calculated from planimetry performed on wound tracings obtained by week) and ulcer closure according to the clinical protocol. These inputs and outputs were used in test/train/validate sets in various architectures for numerous NNAs. Highly predictive nets were selected on the basis of R2, calculated from actual vs. predicted outcomes. Sensitivity analyses, using the models generated, indicated the importance of both specific product and patient parameters. Highly predictive models were generated which could calculate 1) the amount of healing due to the first application of Dermagraft and 2) the healed/non-healed result in the time prescribed in the clinical protocol. A dichotomous accuracy (heal/not heal) of 93% was achieved in the final network analysis.
Dermagraft parameters consistently associated with highly predictive models were 1) the metabolic activity or viability, 2) protein concentration, and 3) the frozen storage time prior to implantation; important patient parameters were ulcer duration, and ulcer area prior to first Dermagraft application.
In a pair of NNAs, product and patient parameters were used to calculate the amount of healing following the first application of Dermagraft. This result was then used as a separate input into a NNA which calculated the dichotomous healing result. Based on changes in the sensitivity analyses for each NNA, it was observed that the calculation of the dichotomous result required either 1) product parameters for the first application, i.e., viability and frozen storage time or 2) the healing resulting from the first application, but not both. This was a critical observation, which indicated that healing resulted directly from the Dermagraft application.
The healing rate following the application of Dermagraft can be visualized by holding all input parameters except metabolic activity and frozen storage time at, for example, their average values, and plot the variation in healing as a function of viability and freezer storage time. These results are shown below.
It is interesting to note that the amount of healing in the control population could not be calculated based solely on the same set of patient parameters, i.e. no model could be developed with comparable accuracy using only patient parameters.
Conclusion
NNAs have proven invaluable in the identification of those characteristics of Dermagraft, which are intimately associated with the healing of diabetic ulcers. The most important Dermagraft parameters associated with the healing process are the metabolic activity and the frozen storage time. Variations in either of these parameters result in variations in healing.
The healing of diabetic ulcers is related to both the viability and metabolic activity and the frozen storage time of Dermagraft. The understanding of these parameters and their effect on healing has led to an improved understanding of the healing of diabetic ulcers with and without Dermagraft.
For more information: Mark Parham, president, Parham Research, 6 Ruben Duren Way, Bedford, MA 01730-1666. Telephone: 781-271-0836.