[Computer aided dosage management of phenprocoumon anticoagulation therapy. Clinical validation]. / Rechnergestützte Dosierung in der Antikoagulationstherapie mit Phenprocoumon. Validierung für den klinischen Einsatz.

UNLABELLED: A recently developed multiparameter computer-aided expert system (TheMa) for guiding anticoagulation with phenprocoumon (PPC) was validated by a prospective investigation in 22 patients. The PPC-INR-response curve resulting from physician guided dosage was compared to INR values calculated by "twin calculation" from TheMa recommended dosage. Additionally, TheMa was used to predict the optimal time to perform surgery or invasive procedures after interruption of anticogulation therapy. RESULTS: Comparison of physician and TheMa guided anticoagulation showed almost identical accuracy by three quantitative measures: Polygon integration method (area around INR target) 616.17 vs. 607.86, INR hits in the target range 166 vs. 161, and TTR (time in therapeutic range) 63.91 vs. 62.40 %. After discontinuation of anticoagulation therapy, calculating the INR phase-out curve with TheMa INR prognosis of 1.8 was possible with a standard deviation of 0.50 ± 0.59 days. CONCLUSION: Guiding anticoagulation with TheMa was as accurate as Physician guided therapy. After interruption of anticoagulant therapy, TheMa may be used for calculating the optimal time performing operations or initiating bridging therapy.

Compensation type algorithms for neural nets: stability and convergence.

Plasticity of synaptic connections plays an important role in the temporal development of neural networks which are the basis of memory and behavior. The conditions for successful functional performance of these nerve nets have to be either guaranteed genetically or developed during ontogenesis. In the latter case, a general law of this development may be the successive compensation of disturbances. A compensation type algorithm is analyzed here that changes the connectivity of a given network such that deviations from each neuron's equilibrium state are reduced. The existence of compensated networks is proven, the convergence and stability of simulations are investigated, and implications for cognitive systems are discussed.