“How long will I be in the hospital?”
This always seems to be the first question patients ask when I tell them we are recommending admission. After navigating their differential, diagnostics, therapeutic plan, and the admission process itself, this question is often the last thing on my mind, and - early in training - it caught me off guard. Worse, it often made me feel helpless, because I had no idea how to answer the one question the patient cared about most at that moment.
Especially when the patient is hesitant about being admitted, this is critical information that I felt I simply could not provide. Here, we turn to the literature to ask: How well-equipped are we as emergency physicians to answer this question, and how can we honestly respond?
It is easy to see why patients want to know how long they will be admitted – they want to know how long they’ll have to put their lives on hold, and external factors weigh considerably on their minds. How would you feel if you were told unexpectedly that you would not be able to go home, would have to call off from work, and would have to drop all of your responsibilities to friends, family, and loved ones for an unknown period of time? As just one example, Canaday et al. found that nearly half of surveyed pet owners would consider delaying hospitalization due to necessary pet care.1
Besides the patient, there are a host of other stakeholders in the health care system for whom anticipated length of stay is valuable information. Hospital administrators need to plan staffing and bed availability in order to ensure service efficiency, and insurance companies may employ different pay structures for inpatient vs. observation admissions.
In spite of these vested interests, a review of the literature suggests that predicting length of stay at time of admission remains extremely difficult, and emergency physicians in particular are poorly equipped to do so.
Factors Affecting Length-of-Stay Predictions
Mak et al. studied real-time length of stay predictions made by emergency physicians at time of admission, and found an average mean difference between predicted and actual length of stay of 3.9 days.2 Patients who were elderly or medically complex had even more poor prediction performance, and even pediatric patients (who received the most accurate predictions of any age group) had a mean difference between predicted and actual length of stay of 1.7 days.
Once the patient is admitted, accuracy in predicting the date of discharge improves as the actual date of discharge approaches.3 However, even inpatient physicians who manage their patients longitudinally and have the benefit of taking into account a patient’s longer term trajectory and response to treatment, when asked in the morning, can only identify which of their patients will be ready for discharge the following day with 33% positive predictive value.4
For comparison, modern weather forecasters are able to predict the daily high temperature within 5 °F as far as 5 days in advance.5
Clearly, accuracy in length of stay prediction leaves much to be desired. The difficulty stems from the fact that there are multiple complex systems at play, including individual patient characteristics, disease characteristics, diagnostic uncertainty, response to treatment, unanticipated complications, and logistical barriers to discharge to name a few.
Experts in Uncertainty
Although solutions incorporating advanced data modeling and artificial intelligence are becoming more sophisticated, their use and application remains narrow.6 For now, the vast majority of emergency physicians are answering patients’ questions about anticipated length of stay based on gestalt, with limited feedback as to the accuracy of their predictions.
Ultimately as emergency medicine physicians we specialize in uncertainty. We are trained to operate with limited information, and it is our duty to help guide patient decision-making in the face of uncertainty. When asked a question to which we have no good answer, the best response may be “I don’t know,” followed by an acknowledgement of the variables at play, a description of possible outcomes, and a best guess as to which outcome or outcomes are most likely.
We are accustomed to providing this type of counseling in a variety of other clinical scenarios, and we should use a similar approach when answering patients' questions regarding anticipated length of stay.
After all, the data would suggest that it is easier to predict the weather than it is to forecast length of stay from the emergency department.
References
- Canady B, Sansone A. Health Care Decisions and Delay of Treatment in Companion Animal Owners. J Clin Psychol Med Settings. 2019;26(3):313-320.
- Mak G, Grant WD, McKenzie JC, McCabe JB. Physicians' ability to predict hospital length of stay for patients admitted to the hospital from the emergency department. Emerg Med Int. 2012;2012:824674.
- Henry OP, Li G, Freundlich RE, Sandberg WS, Wanderer JP. Understanding the Accuracy of Clinician Provided Estimated Discharge Dates. J Med Syst. 2021;46(1):2.
- Sullivan B, Ming D, Boggan JC, Schulteis RD, Thomas S, Choi J, Bae J. An evaluation of physician predictions of discharge on a general medicine service. J Hosp Med. 2015;10(12):808-10.
- Rose B, Floehr E. Analysis of high-temperature forecast accuracy of consumer weather forecasts from 2005–2016 [Internet].Dublin, Ohio: Forecast Watch Report; 2017.
- Stone K, Zwiggelaar R, Jones P, Mac Parthaláin N. A systematic review of the prediction of hospital length of stay: Towards a unified framework. PLOS Digit Health. 2022;1(4):e0000017.