INTERVENTION TIME-SERIES ANALYSIS - SAMPLING STRATEGIES

Goldschmidt, H.M.J., Ten Voorde, L.J.F., Leijten, J.F. and Lent, R.W., 1990. Intervention time series analysis - sampling strategies. Chemometrics and Intelligent Laboratory Systems, 9: 83-94. The application of process control characteristics to the quality assessment of analytical and clinical chemical data as well as to series of monitoring results is described extensively. These applications presume a steady state, i.e., no interventions or actions are allowed other than those induced by the process itself or by random analytical errors. However, in the study of physiological time series, interventions are common; e.g., a new type of insulin is administered, a new level of anticoagulant is prescribed, etc. A type of statistical analysis that is suited for this type of data is intervention time series analysis. We measured blood glucose, insulin and C-peptide in two diabetic patients every two hours for six days. In one patient (V) the blood glucose varied from 5 to 24 mmol/l, the C-peptide remained within the reference interval (0.2-1.0 nmol/l) while the insulin level varied from 20-140 μIU/ml, depending on the amount of insulin administered. There was no direct correlation between blood glucose level and total insulin. In the second case (VA), the blood glucose varied from 5 to 15 mmol/l, the C-peptide was always greater than 2.0 nmol/l along with a total insulin of 40-220 μIU/ml. A retrospective analysis of the blood glucose level for both patients using univariate and multivariate statistical analysis proved to be applicable to only one patient (VA). The data in the other case (V) could only be fitted by means of an intervention time series analysis. Time series analysis not only provides a good description of physiological time tracks but, with the application of intervention analysis, it also offers the possibility of handling medical actions. The accurate forecasting of results, e.g. in medical settings, is possible with these techniques. The optimal sample frequency can be also estimated. In the future, with the growing need for cost-effective information management tools, these methods will gain in importance. © 1990.