1. My aim was to investigate, by mathematical simulation, the errors inherent in the measurement by the primed infusion method of the rate of appearance of glucose in man when turnover was as low or lower than in overnight-fasted normal subjects (control subjects). The simulations were based on published data for means and variances of turnover rates and concentrations in non-diabetic subjects and diabetic patients.
2. Systematic errors (bias) were shown to be considerable whether or not the Steele equation was used, unless run-times were longer than is customary. Errors were greater the lower the turnover rate, and were greatest in patients with diabetes, owing to insulin resistance. Studies of, for example, control subjects, age, obesity, exercise, sepsis and injury, are, however, all likely to be affected.
3. Estimates of variance, within-group means, between-group differences and slopes of rate—concentration relationships were all biased. Entirely spurious results appeared statistically significant.
4. When the Steele equation was not used, run-times had to exceed 3 h in control subjects and 10 h in some diabetic patients to reduce bias to acceptable levels. The nature of the bias depended on how the priming dose/infusion rate ratio was chosen. Each choice implies a particular hypothesis about the values of the rate of appearance of glucose, their variance, and how they are related to concentration. The bias was always such as to favour that hypothesis.
5. When the Steele equation was used, the accessible glucose space (pool fraction × distribution volume) had to be correct to 20–30 ml to avoid unacceptable bias in some patients in runs 4 h long. The space is not known this accurately. Theoretically, in the near-steady metabolic states considered, the pool fraction should be near 1.00, i.e. the accessible space should be near the glucose distribution volume of 200–300 mg/kg. There is some confirmatory experimental evidence.
6. Large random errors from variance of specific (radio)activity measurements when the Steele equation is used can be reduced by a suitable choice of protocols.
7. The propagation of errors is too complex to permit correction of results. It is essential to choose protocols that can be shown to give results that are acceptably bias-free. Ways of doing this are discussed.