[Expert systems-urgently needed for early diagnosis of sepsis : Results of retrospective clinical validation of the expert system FLORIDA].

The expert system FLORIDA (Fuzzy Logic Orientated Rule Interpreter for Diagnostic Applications) is equipped with a knowledge base applying linguistic rules of clinical experts according to the pathophysiologic conception of the sepsis-3 definition and the mathematics of fuzzy logic. It works independently of any subjective factors. FLORIDA detects sepsis by an increase of a second incoming value ≥ 25% compared with the reference value. It requires dynamics of the parameters used in knowledge base; thus, if the dynamics are absent, the system is unable to detect sepsis. In a retrospective clinical validation study, FLORIDA was used to scan 498 consecutive patients in a medical intensive care unit, which corresponded to 1700 patient-days. During the study period, the prevalence of sepsis was 10%. In all, 423 patients were identified as not having sepsis, which was confirmed by clinical experts. Among the 48 patients identified as having sepsis, 26 were confirmed (true positive). In 22 patients, sepsis could not be confirmed by the clinician (false positive). FLORIDA did not detect sepsis in 4 patients, but was diagnosed by the clinician (false negative). In the 22 false-positive patients, a life-threatening disease existed requiring intensive care. Because of the system philosophy, FLORIDA is unable to recognize patients with full blown sepsis at admission. With this in mind, the sensitivity was 1.0 and specificity was 0.95. Thus, FLORIDA is qualified for the early detection of a developing sepsis. Sepsis is detected on average 12.5 ± 8.6 h after the start of sepsis has been determined by a clinical expert. FLORIDA should be used on normal wards and should contribute to early detection of sepsis and potentially earlier therapeutic intervention in order to decrease hospital mortality. However, prospective validation is needed.