Pattern analysis of epidemiological variables applied in the study of weanling diarrhoeal disease

The need for and requirements of a computer-based pattern analysis approach to the investigation of longitudinal epidemiological records have been considered; it is argued that the objective specification of relevant variables is of first priority. The development and implementation of suitable techniques is reported in context of a continuing study of a large-scale international epidemiological problem-weanling diarrhoeal disease of infants. These techniques have been used for the objective specification of variables measured in an epidemiological field study of an ensemble of 139 weanling infants followed for the first two years of life; rationale, methods, and results are reported here for the simplest of the available variables-the daily defaecation rate (or diarrhoeal signal, DS). Two separate detailed findings have emerged. Three patterns of spontaneous defaecation behaviour have been confirmed and located objectively by a numerical matched filter method; their occurrence statistics were compiled and the indicated inter-linkages investigated. A two-way classification of the DS records-by initial short-term DS experience and by computing the average DS-profile and testing each DS record against it-was derived. A detailed DS specification can be based on these findings and using this model an objective identification of abnormal DS behaviour has become feasible. It is argued that these results on basic mechanisms and variables offer a stable starting point for further epidemiological study of this problem and that the utility of these computer-based signal techniques suggests their more general applicability in epidemiology.