Usefulness of an expert software for hemodynamic evaluation: results of the HEMODYN- Survey

Abstract

In an earlier study, we clinically validated an expert computer program (Hemodyn-) designed to assist in interpreting pulmonary artery catheterization data [1]. The present study in seven European centers assessed the influence of Hemodyn- on the therapeutic strategies of experienced residents. Patients with a pulmonary artery catheter inserted were included in the study if they had hemodynamic disorders unresponsive to standard therapy and/or if their condition raised a therapeutic problem that PAC was expected to solve. Each resident examined and collected PAC data from the study patients in their center, under the supervision of the local study coordinator. Then, the resident completed a data form and made one or more choices among the diagnostic possibilities listed on the form. The same PAC data were then entered into the software. Based on the software's diagnostic evaluation, the resident could either maintain or change his or her diagnosis and treatment. Finally, a senior intensivist accepted or rejected the resident's final diagnosis and treatment plan. Agreement between the residents' initial evaluation and the software's evaluation was poor (kappa < 0.6). Sixty-four hemodynamic profiles from 44 patients were used for the study covering a broad spectrum of critical situations. Before computer assistance, the residents suggested at least one treatment change (to improve hemodynamics) in 83% of patients and a mean of 2.2 treatment changes per patient. After computer assistance and evaluation by the local study coordinator, the residents changed their treatment suggestions in 94% of patients, making a mean of 1.9 changes per patient. Therapeutic agreements before and after computer assistance are shown in the Table (*P < 0.05 vs other comparisons). Table After computer assistance, agreement was very good between residents and seniors and between seniors and the software. Computer assistance led the residents to change at least one suggested treatment in 63% of cases; in 55% of cases, the change was not minor. Analysis of the points of disagreement showed that the residents often underused fluids and vasodilators: these two points contributed 42% of changes after computer assistance. In only 20% of the cases in which the patient received a vasodilator was this treatment suggested initially by the resident. For the other treatments, the proportions of inappropriate use and inappropriate absence of use were similar. Expert software capable of helping residents to interpret PAC data properly may improve the quality of care given to critically ill patients.