Gonzalo Sirgo1, Federico Esteban2, Josep Gómez3, Gerard Moreno4, Alejandro Rodríguez5, Lluis Blanch6, Juan José Guardiola7, Rafael Gracia8, Lluis De Haro9, María Bodí10. 1. Intensive Care Unit, Hospital Universitario Joan XXIII, Instituto de Investigación Sanitaria Pere Virgili, Rovira i Virgili University, Tarragona, Spain. Electronic address: gsirgo.hj23.ics@gencat.cat. 2. Intensive Care Unit, Hospital Universitario Joan XXIII, Instituto de Investigación Sanitaria Pere Virgili, Rovira i Virgili University, Tarragona, Spain. Electronic address: festeban.hj23.ics@gencat.cat. 3. Intensive Care Unit, Hospital Universitario Joan XXIII, Instituto de Investigación Sanitaria Pere Virgili, Rovira i Virgili University, Tarragona, Spain. Electronic address: Josep.gomez@urv.cat. 4. Intensive Care Unit, Hospital Universitario Joan XXIII, Instituto de Investigación Sanitaria Pere Virgili, Rovira i Virgili University, Tarragona, Spain. Electronic address: murenu77@hotmail.com. 5. Intensive Care Unit, Hospital Universitario Joan XXIII, Instituto de Investigación Sanitaria Pere Virgili, Rovira i Virgili University, Tarragona, Spain. Electronic address: arodri.hj23.ics@gencat.cat. 6. Critical Care Centre, Hospital Universitari Parc Taulí, Institut de Investigació i Innovació Parc Taulí (I3PT), Universitat Autònoma de Barcelona, Sabadell, Spain; Centro de Investigación Biomédica en Red de Enfermedades Respiratorias (CIBERES), Instituto de Salud Carlos III, Majadahonda, Spain. Electronic address: lblanch@tauli.cat. 7. Department of Pulmonary, Critical Care and Sleep Medicine, University of Louisville, Louisville, KY, USA. Electronic address: Juan.Guardiola@va.gov. 8. Management Department, Camp de Tarragona Region, Institut Català de la Salut, Tarragona, Spain. Electronic address: rgracia@gencat.cat. 9. Functional Competence Center, Information Systems, Institut Català de la Salut, Barcelona, Spain. Electronic address: ldeharo@gencat.cat. 10. Intensive Care Unit, Hospital Universitario Joan XXIII, Instituto de Investigación Sanitaria Pere Virgili, Rovira i Virgili University, Tarragona, Spain. Electronic address: mbodi.hj23.ics@gencat.cat.
Abstract
BACKGROUND: Big data analytics promise insights into healthcare processes and management, improving outcomes while reducing costs. However, data quality is a major challenge for reliable results. Business process discovery techniques and an associated data model were used to develop data management tool, ICU-DaMa, for extracting variables essential for overseeing the quality of care in the intensive care unit (ICU). OBJECTIVE: To determine the feasibility of using ICU-DaMa to automatically extract variables for the minimum dataset and ICU quality indicators from the clinical information system (CIS). METHODS: The Wilcoxon signed-rank test and Fisher's exact test were used to compare the values extracted from the CIS with ICU-DaMa for 25 variables from all patients attended in a polyvalent ICU during a two-month period against the gold standard of values manually extracted by two trained physicians. Discrepancies with the gold standard were classified into plausibility, conformance, and completeness errors. RESULTS: Data from 149 patients were included. Although there were no significant differences between the automatic method and the manual method, we detected differences in values for five variables, including one plausibility error and two conformance and completeness errors. Plausibility: 1) Sex, ICU-DaMa incorrectly classified one male patient as female (error generated by the Hospital's Admissions Department). Conformance: 2) Reason for isolation, ICU-DaMa failed to detect a human error in which a professional misclassified a patient's isolation. 3) Brain death, ICU-DaMa failed to detect another human error in which a professional likely entered two mutually exclusive values related to the death of the patient (brain death and controlled donation after circulatory death). Completeness: 4) Destination at ICU discharge, ICU-DaMa incorrectly classified two patients due to a professional failing to fill out the patient discharge form when thepatients died. 5) Length of continuous renal replacement therapy, data were missing for one patient because the CRRT device was not connected to the CIS. CONCLUSIONS: Automatic generation of minimum dataset and ICU quality indicators using ICU-DaMa is feasible. The discrepancies were identified and can be corrected by improving CIS ergonomics, training healthcare professionals in the culture of the quality of information, and using tools for detecting and correcting data errors.
BACKGROUND: Big data analytics promise insights into healthcare processes and management, improving outcomes while reducing costs. However, data quality is a major challenge for reliable results. Business process discovery techniques and an associated data model were used to develop data management tool, ICU-DaMa, for extracting variables essential for overseeing the quality of care in the intensive care unit (ICU). OBJECTIVE: To determine the feasibility of using ICU-DaMa to automatically extract variables for the minimum dataset and ICU quality indicators from the clinical information system (CIS). METHODS: The Wilcoxon signed-rank test and Fisher's exact test were used to compare the values extracted from the CIS with ICU-DaMa for 25 variables from all patients attended in a polyvalent ICU during a two-month period against the gold standard of values manually extracted by two trained physicians. Discrepancies with the gold standard were classified into plausibility, conformance, and completeness errors. RESULTS: Data from 149 patients were included. Although there were no significant differences between the automatic method and the manual method, we detected differences in values for five variables, including one plausibility error and two conformance and completeness errors. Plausibility: 1) Sex, ICU-DaMa incorrectly classified one male patient as female (error generated by the Hospital's Admissions Department). Conformance: 2) Reason for isolation, ICU-DaMa failed to detect a human error in which a professional misclassified a patient's isolation. 3) Brain death, ICU-DaMa failed to detect another human error in which a professional likely entered two mutually exclusive values related to the death of the patient (brain death and controlled donation after circulatory death). Completeness: 4) Destination at ICU discharge, ICU-DaMa incorrectly classified two patients due to a professional failing to fill out the patient discharge form when thepatients died. 5) Length of continuous renal replacement therapy, data were missing for one patient because the CRRT device was not connected to the CIS. CONCLUSIONS: Automatic generation of minimum dataset and ICU quality indicators using ICU-DaMa is feasible. The discrepancies were identified and can be corrected by improving CIS ergonomics, training healthcare professionals in the culture of the quality of information, and using tools for detecting and correcting data errors.
Authors: Francisco Javier Pérez-Benito; Carlos Sáez; J Alberto Conejero; Salvador Tortajada; Bernardo Valdivieso; Juan M García-Gómez Journal: PLoS One Date: 2019-08-07 Impact factor: 3.240