Improving ICU patient flow through discrete-event simulation

Abstract

Massachusetts General Hospital (MGH), the largest hospital in New England and a national leader in care delivery, teaching, and research, operates ten Intensive Care Units (ICUs), including the 20-bed Ellison 4 Surgical Intensive Care Unit (SICU), a versatile unit which has a major role in perioperative and emergency care. 90% of SICU patients are eventually transferred to another unit in the hospital. Frequent and sometimes lengthy non-clinical delays in this transfer process can be primarily attributed to congestion in downstream units. Multivariable regression analysis demonstrates that additional nonclinical SICU time yields negligible downstream time savings, while consuming an average of 2.4 SICU beds per day, or 12% of total SICU capacity. In addition to exacerbating the delays of patients requiring admission to the SICU, these non-clinical SICU exit delays are responsible for a yearly attributable annual cost in excess of $2.5M. Possible ameliorating approaches include prioritizing SICU transfers, or modifying the care of delayed SICU patients to begin preparing for discharge from the hospital. Any such choices affecting capacity and resource allocation in the ICU environment involve high cost as well as potentially high risks related to quality of care. To evaluate the impact of potential operational changes, the SICU and its six primary downstream units are modeled in a highly detailed discrete event simulation. Patients are divided into ~2,700 procedural and diagnostic types. Entries (admissions) for each patient type are characterized as inhomogeneous Poisson processes, with lengths of stay drawn from probability distributions. Transfer practices and priorities are encoded in simulation logic. A simulation of twenty replication periods, each one year long, allows for calibration and validation by detailed comparison with historical data. Simulated average hourly census values are within 1% of historical averages and RMSE is below 4% for each of the five modeled areas, indicating high accuracy and low bias. The validated simulation is applied to evaluate the impact of several possible operational adjustments, including changes to discharge timing, transfer priorities, and resource allocation. Two approaches prove most promising: 1) Transferring patients as soon as possible after medical clearance, eliminating the current practice of waiting to see if other patients might need downstream beds. 2) Implementing a 24- hour rolling medical clearance process in the SICU. These interventions are predicted to lower average and peak SICU utilization by ~6%, cut SICU entrance delays by -35%, and decrease SICU exit delays by -50%, with relatively little impact on downstream floors and no additional capital expenditures. These relatively simple policy changes can save -$1 M in non-reimbursed expenditures while reducing overcrowding. If capital expenditures are approved, the simulation indicates that adding beds to downstream units would be more beneficial to the system than adding the same number of intensive care beds (at a much higher cost). Similar results are likely to be applicable to other ICUs at MGH, multiplying the potential impact of these findings several times over.