Bone marrow transplant patients having severe, prolonged oral mucositis pain (expected to last for one to three weeks) used a computer-controlled infusion system to self-administer morphine for pain control. Individual patient pharmacokinetic information, derived from a pretreatment bolus morphine dose, was used in a new bolus-elimination transfer algorithm to produce rapid adjustments of steady plasma morphine concentrations when the patient requested more or less drug. We evaluated the performance characteristics (bias and precision) of this pharmacokinetically based patient-controlled analgesic infusion system (PKPCA) in a group of 15 cancer patients over six to 14 days. Although we found a three- to fivefold pharmacokinetic variability in the tailoring morphine dose data, the PKPCA system was free of systematic bias (insignificant overall prediction error) during the patient-controlled infusions in this study population. The absolute prediction error was 19.9% for the group on the first study day and 25.6% over the entire study period (aggregate results; 6-14 days of continuous use). Two-thirds of the patients exhibited no bias throughout the study period, and individual bias in the others was symmetrically distributed (three patients with underpredictions and two overpredicted). Magnitude of prediction error during the patient-controlled morphine infusions was not related to the magnitude of pharmacokinetic deviation of individual subjects from group parameters. Our results indicate that this PKPCA system provides accurate control of plasma morphine concentration when used by patients to self-administer opioid for prolonged pain relief continuously over 1 to 2 weeks.(ABSTRACT TRUNCATED AT 250 WORDS)