Mary McLeish

A prospective randomized single blind controlled clinical trial was undertaken to compare prophylactic therapy using a systemic antibiotic active against both aerobic and anaerobic bacteria with an oral antibiotic agent active only against anaerobic bacteria in elective colorectal surgery. One hundred and thirty‐one patients received ticarcillin and 130 received tinidazole. The wound infection rate was 8% in those patients receiving ticarcillin prophylaxis and 20% in those receiving tinidazole (P < 0.05). Multivariate analysis of the factors affecting wound infection rate showed that there were three independent factors that reached statistical significance: the prophylactic antibiotic used; the type of hospital (public or private) in which the operation was performed, and the presence of a stoma at operation. The wound infection rate in those patients receiving tinidazole prophylaxis was more than twice that reported previously by the authors. The mortality in patients receiving ticarcillin prophylaxis was 1.5% compared to 9.2% in those receiving tinidazole prophylaxis (P < 0.05). The clinical anastomotic leakage rate was similar in each antibiotic prophylactic group, 8.6% in those receiving ticarcillin and 7.3% in those receiving tinidazole.

Partitioning is a highly secure approach to protecting statistical databases. When updates are introduced, security depends on putting restrictions on the sizes of partition sets which may be queried. To overcome this problem, attempts have been made to add “dummy” records. Recent work has shown that this leads to high information loss. This paper reconsiders the restrictions on the size of partitioning sets required to achieve a high level of security. Updates of two records at a time were studied earlier, and security was found to hold if the sizes of the partition sets were kept even. In this paper an extended model is presented, allowing very general updates to be performed. The security problem is thoroughly studied, giving if and only if conditions. The earlier result is shown to be part of a corollary to the main theorem of this paper. Alternatives to adding dummy records are presented and the practical implications of the theory for the database manager are discussed.

Probabilistic logic has been discussed in a recent paper by N. Nilsson [12]. An entailment scheme is proposed which can predict the probability of an event when the probabilities of certain other connected events are known. This scheme involves the use of a maximum entropy method proposed by P. Cheeseman in [3]. The model uses vectors which represent certain possible states of the world. Only such vectors are entered into the probability scheme. As a result, entailment does not always yield an acceptable result and cannot be applied to real situations which could arise. This paper investigates a technique to overcome this problem, which involves extending the idea of probabilistic logic and the maximum entropy approach to Dempster-Shaffer theory. A new entailment scheme for belief functions is used which produces well-defined results, even when only consistent worlds are being considered. The paper also reconsiders an earlier attempt by the author [6,7] to model default reasoning (and subsequent nonmonotonicity) by adding inconsistent vectors to Nilsson's model. In the extended setting, more sensible entailment values are obtained than in the previous work.