Modeling inpatient flow from hospital information systems

Hospitals are increasingly challenged to improve care quality and decrease operating costs. There is a recognized need for research regarding the operational aspects of healthcare delivery. Patient flow, the progression of patients through a hospital, is a key element in efficient care delivery. Flow paths have provided insight into resource allocation procedures and medical care. Although there has been a considerable amount of research in modeling patient flow, the majority of these studies focus on a single hospital department, such as the Emergency Department, or on a particular patient class, such as cardiology. In reality, care delivery requires the coordination and allocation of resources across the entire hospital to a heterogeneous patient population. Patient flow studies are further exasperated by the lack of a comprehensive data source. Extant models tend to focus on steady state behavior with almost no transient analysis, and few existing analytical models consider the effects of finite resource capacity. As demonstrated in a review of the literature, a gap in patient flow research exists. The objective of this research was to identify, develop, and implement inpatient flow models from information systems message exchanges. This method addresses the current paucities in the research by (1) recognizing the impact of different patient diagnoses on hospital resources, (2) considering resource usage across multiple hospital departments, and (3) identifying a suitable data source. For every inpatient in a hospital, communication exchanges between departments drive much of what happens to the patient and represent a rich source of electronic data. Information from HL7-formatted information system messages was extracted to create a time-ordered sequence of activities a patient underwent during his/her stay. The unique combination of resources required for each activity, referred to as Resource Bundles, was identifed. A patient’s path through the hospital, the patient sequence, is thus represented as a sequence of resource requirements. Patients were grouped according to Diagnostic Related Groups (DRG). Patient sequences for three patient types [DRG 14 (18 models), DRG 544 (19 models), and DRG 558 (25 models)] were developed. The granularity of information contained in a patient sequence can be controlled through the application of the Term Discrimination Algorithm. Petri net and regular expression techniques synthesized a set of patient sequences into a representative structure for a given DRG. These techniques are well suited to address the operational dynamics of systems with interacting and concurrent components such as hospitals. Patient sequences were successfully validated through a patient chart review. The correctness of a DRG flow model was verified using regular expressions and the properties of Petri nets. The verification exercises identified and corrected structural errors in the models. The usefulness of this research was illustrated through a case study examining stroke care compliance. Nine patient sequences for stroke patients were compared against an idealized care pathway to measure (1) deviations from the pathway, (2) adherence to the pathway, and (3) patient outcomes. From the deviation and adherence measures, potential sources for feedback on quality of care were identified. The patient outcome measure quantified the relationship between deviations from the pathway and a patient’s outcome. Based upon the results of this dissertation, three future research avenues are discussed: identifying operational policies to avoid undesirable hospital states, identifying suitable admission profiles, and developing patient-condition driven pharmaceutical inventory management