The methodology of RECursive Partition and AMalgamation (RECPAM) previously presented in Parts I and II (A. Ciampi et al., Computer. Methods Progr. Biomed. 26 (1988) 239-256 and 30 (1989) 283-296) pursues its development with an application to predict long-term effects of a disease given a set of clinical information measured at the time of illness. This paper illustrates how RECPAM deals with a situation typical in Medical Informatics applied to data on Haemophilus influenzae type b meningitis.