The development of a bedside algorithm capable of targeting anti‐endotoxins to the responder subpopulations

The published selection criteria for use of anti-endotoxin antibodies are modeled on the criteria used in protocol enrollment of patients. These criteria are not suitable for therapeutic decision-making, because the trials themselves prove that the enrollment criteria have both low sensitivity and low specificity. In order to develop a selection method with greater sensitivity and specificity, we examined the charts and records of 23 patients that we enrolled in the multicenter trials of E5 and HA-1A. We retrospectively determined that seven of our 23 enrolled patients were optimal candidates, based on a pattern of rapid clinical deterioration followed by improvement in 24–96 h. We then explored a variety of different modes of bedside patient selection, in search of a method to select as many of the seven optimal candidates as possible while at the same time rejecting the greatest number of the 16 who showed no benefit when treated. None of the resulting selection methods has perfect performance, but nearly all were better than the original protocol enrollment criteria. In our patients, bacteremia had 57% sensitivity and 56% specificity, which was quite similar to the findings in the HA-1A multicenter trial. Shock had 100% sensitivity and 44% specificity, while a baseline organ dysfunction score of 5 had 100% sensitivity and 69% specificity. A new algorithm that we developed based on a patient's need for vasopressors and baseline organ dysfunction had 100% sensitivity and 81% specificity. This algorithm could identify all seven of the optimal candidates, plus three more. We propose that there are bedside methods capable of considerably more specific identification of patients who would respond to treatment with anti-endotoxin antibodies.