Improved computer-assisted digoxin therapy. A method using feedback of measured serum digoxin concentrations

Automated feedback control methods were applied to a medical problem, in a computer program that used measured serum digoxin concentrations (as feedback) to predict future concentrations and to achieve desired concentrations. The system was validated by comparing its ability with the corresponding ability of physicians to regulate digoxin dosage. The prospective, randomized study included 51 patients. In the presence of varying amounts of feedback (serum digoxin concentration) information, the computer always predicted future digoxin concentrations as accurately as did physicians. For both computer and physician, the decrease in the prediction errors when two concentrations were known against that when no concentrations were known was significant: mean absolute error decreased from 0.40 to 0.25 ng/ml for the physicians and from 0.45 to 0.27 ng/ml for the computer. Thus the computer system is capable of simulating and reproducing a sophisticated aspect of physician behavior: "learning" about individual patient responses. The computer achieved desired concentrations more accurately than did physicians, especially when two or more previous digoxin concentrations were abailable (mean absolute achievement error for computer, 0.28 ng/ml; for physicians, 0.50 ng/ml).