When a patient needs the highest level of medical care, a hospital’s intensive care unit (ICU) offers crucial life saving resources. But ICUs are among the most in-demand units at hospitals, with a constrained number of beds available. Adding ICU capacity is neither quick nor cheap, and hospitals must prioritize treatment with the ICU beds and nurse staffing they have on hand. To make room for new ICU patients, existing ICU patients are sometimes transferred to other units that are not designed for intensive care before it is medically optimal. So-called demand-driven discharges increase the risk of complications, which result in readmission to the ICU, and increase length of stay (LOS) in the ICU.
Current medical protocols are not as helpful in alleviating the strain on ICUs as one might expect — they tend to be subjective, with physicians not always agreeing on the best course of action. “Of course doctors want to discharge the least critical patients,” Professor Carri Chan says. “But no systematic framework exists to measure that.”
Chan worked with Vivek Farias of MIT, Nicholas Bambos of Stanford, and Gabriel Escobar of Kaiser Permanente to analyze patient and ICU data and recommend a discharge protocol that would reduce ICU readmission rates while preserving the quality of patient care. The researchers examined 60,000 patient data points at 17 different hospitals in the Northern California Kaiser Permanente Medical Care Program during the course of one year. Kaiser has extensive patient data available electronically, giving the researchers access to a rich data set that included patient age, sex, type of admission (for instance, if the episode was an emergency), the medical reason for admission, and whether a patient had any prior conditions prone to causing complications.
The researchers winnowed their large data set down to focus on 6,600 ICU patients at seven hospitals during one year. To determine if capacity played a role in a patient’s discharge, researchers examined the occupancy level of each ICU throughout the course of the year. If an ICU was full when a patient was discharged, it was more likely that capacity issues prompted the decision.
Next, the researchers combined this discharge information with an assessment created by Kaiser that uses patient lab results taken within 72 hours prior to admission to describe the severity of a patient’s condition. Using these patient severity scores, researchers placed patients into nine different classes. For each of the classes, researchers determined patients’ likely LOS in the hospital if there were no capacity issues, the likelihood of readmission if subject to demand-driven discharge, and the likely LOS if readmitted to the ICU.
Using the Kaiser dataset, the researchers determined that discharge strategies that take a patient’s risk of readmission to the ICU into consideration are safer and more efficient than other discharge strategies. Although physicians may lean toward first discharging patients least likely to be readmitted, Chan stresses the role that LOS plays in readmissions. “We want to discharge the patient who will use the fewest resources upon readmission,” she explains. “We have to think about not just whether they’re going to be readmitted, but what happens when they come back — how long will they need intensive care the second time around?”
It doesn’t always work out that that a patient with a higher likelihood of readmission is the sickest patient, Chan points out. “It’s common for a patient who just had surgery to experience bleeding from the wound. But that complication doesn’t usually take very long to recover from compared to many other complications, so readmitting that patient wouldn’t redirect too many ICU resources. On the other hand, another patient may face a lower likelihood of readmission, but if readmission is required, that patient is more likely to suffer from a more severe complication and require a much longer stay in the ICU.” It’s this distinction between which patients are or aren’t likely to experience lengthy LOS upon readmission that the researchers’ model can help hospitals determine.
“Compared to other ICU discharge strategies, we expect the strategy we have defined through our study to be associated with similar or lower mortality rates and with a lower readmission load,” Chan says. Reduced readmissions can translate into shorter wait times for new ICU patients, less need to shuffle ICU patients to other units, and a lower average LOS. Chan is now expanding her collaboration with Kaiser so that more refined models can be developed from additional patient data. These models, in turn, will be used by Kaiser to define better operational strategies for its ICUs.
“There is sometimes hesitation in the medical community to discuss space-capacity issues,” Chan says. “But this model suggests that it is possible to make better use of critical care resources while maintaining or improving the quality of care and saving lives.”
Professor Chan teaches the core MBA class, Decision Models. Her primary research interests are in modeling complex stochastic systems, dynamic optimization, scheduling and queueuing with applications in information technology systems and health-care operations management. Her recent work has focused on the development of effective admission and discharge strategies to improve patient flow in hospital intensive care units.
Read the Research
Carri Chan, Vivek Farias, Nicholas Bambos, Gabriel Escobar
"Maximizing throughput of hospital intensive care units with patient readmissions"