A GENERAL USE OF THE POISSON APPROXIMATION FOR BINOMIAL EVENTS, WITH APPLICATION TO BACTERIAL ENDOCARDITIS DATA.
SYSTEM DEVELOPMENT CORP SANTA MONICA CALIF
Pagination or Media Count:
Use of past data to predict future results is the goal of much medical research. Unfortunately, even partial attainment of this goal is ordinarily difficult. For two special types of situations involving binomial events, however, usable methods can often be developed. Let success and failure denote the outcomes for an event. Estimation of the distribution for the number of successes is considered for new groups containing at least moderately large numbers of events. The probabilities of success not necessarily equal are known to be small for one of the special situations. The failure probabilities are small for the other special situation. For each situation, a Poisson approximation to the distribution can often be used, even for mild dependence among the events. Estimation of the distribution is then reduced to estimation of a parameter that has an elementary interpretation. The estimation method developed contains adjustment factors to allow for differences between the new group and the past data. Cases where the special situations both occur are also considered. Stratification of characteristics is used to obtain events of the two special types and to relate past data to these events. This methodology was originally developed for a mortality study of persons who had bacterial endocarditis. Author