Patients make trade-offs when they consume healthcare. Unlike consumption of many other goods, ethics and regulation often dictate that patients’ ability to pay for healthcare is divorced from their ability to access that healthcare. This is especially evident in emergency medicine, where point of care costs usually do not exist because such costs could deter uninformed patients from life-saving care. This lack of purchase price induces different trade-offs for patients because rationing of emergency care often occurs through extended waits. Patients then face the dis-utility of waiting in exchange for free point-of-care access to a physician (Nichols and Zeckhauser, 1982, Martin and Smith, 1999, Gravelle and Siciliani, 2009). This dis-utility can be substantial, and in some jurisdictions, like Canada, wait times to see an emergency physician can be over four hours (Vogel, 2017).

The goal behind disconnecting prices from care is to prevent deterring sick patients from seeking medical care. Patients who leave healthcare settings without actually accessing care have worse re-presentation rates and higher rates of admission to hospital (Smalley et al., 2021). However, reducing demand through this wait time dis-utility might also pose a useful policy lever: if enough of the “right” patients respond, higher wait times could act as a safety valve in highly supply-constrained healthcare environments like emergency departments. Emergency system crowding is often associated with higher mortality, and outcomes for other patients could improve if less sick patients respond to higher wait times by avoiding emergency care and preserving finite resources (Woodworth, 2020, Hsuan et al., 2023). This response could potentially be important in contexts where patients use emergency system resources for non-emergent reasons (Kelen et al., 2021, Atkinson et al., 2022).

I investigate a health system in Ontario that displays predicted emergency site wait times online and in its emergency waiting rooms. I examine what happens to patient demand for emergency care when the predicted wait times at these sites quasi-randomly change. This emergency system has three advantages. First, because the setting of this study is within Canada, there are no additional point of care costs in accessing medically necessary care. Any cost to the patient to access emergency care exists primarily through the dis-utility of waiting.

Second, the method underlying the creation and display of these wait time predictions allows me to causally infer how additional predicted wait time affects patient demand for care. In this emergency system, a machine learning algorithm predicts a new granular wait time every six minutes at each site. These predictions are made to the minute or sub-minute level (i.e. a prediction of 37.34 min of wait time). However, this emergency system publicly displays coarse versions of these predictions that are rounded into 30-min blocks. For example, a granular predicted wait time of 56 min is rounded to a coarse predicted wait time of 30 min and a granular predicted wait time of 61 min is rounded to a coarse predicted wait time of 60 min. This rounding creates a discontinuity in apparent predicted wait times where emergency sites with similar granular predicted wait times are assigned different coarse predicted wait times (i.e., that are 30 min apart). I compare instances where sites display granular predicted wait times very close to this discontinuity to construct treatment and control groups that have different coarse predicted wait times. I then estimate impulse response functions by local projections to compare how demand for emergency services evolves at the treatment sites relative to the control sites for up to three hours after the predicted wait time is publicly displayed.

Finally, this emergency system has two aspects that allow me to examine the type of patient that are affected by changes in predicted wait time information. This emergency system consists of higher-resourced emergency departments and lower-resourced urgent cares. The former have access to advanced imaging tests, laboratory services, and inpatient services. The latter are staffed by emergency physicians but only have access to limited blood-testing, x-rays and in-department nursing services. On the same web page as the online wait time prediction, the emergency system provides guidance on which patients should use each type of site with the goal of streaming lower-acuity patients to lower-resourced urgent cares. How demand changes at these differently resourced sites likely reflects the relative sickness of the patients that are responding to changes in the predicted wait time information. Similarly, this emergency system also assigns an ex-ante triage score to each patient prior to any treatment or investigatory decisions. This score is calculated by computer, is relatively objective, and reflects how quickly resources should be allocated towards a patient. I use this triage score as a proxy of each patient’s immediate health needs and examine whether changes in predicted wait times differentially impact demand by sicker or healthier patients.

I find that the overall number of waiting patients in an emergency site responds to higher predicted wait times. Three hours after display, 30 min of additional predicted wait time reduces the overall number of waiting patients by 2.4%. This effect almost exclusively occurs through deterrence of new patients arriving in the waiting room rather than patients leaving the waiting room. This response varies by site type. There are declines in the number of waiting patients at both emergency departments and urgent cares, but there are larger absolute and proportional declines at urgent cares. At urgent cares and emergency departments this amounts to a decline of 15% and 2% of the waiting patients, respectively. These are absolute declines in the number of waiting patients by 0.4 and 0.1 waiting patients. I find no overall changes in patients who are graded at the highest two levels of urgency (i.e., resuscitation and emergent-level triage scores) but find declines in the number of patients who are graded at lower triage scores (i.e., urgent, less urgent, and non-urgent levels). This response also varies by baseline predicted wait time; a difference of 30 min of predicted coarse wait time when the granular wait time is around 270 min affects demand differently relative to when the predicted granular wait time is 60 min. However, at high predicted wait times, all types of patients reduce emergency site use, including those who are triaged as being very sick.

In sum, increasing the apparent predicted wait time for patients causes declines in demand for emergency care. The prospective patient who is more likely to respond is one who otherwise would have used an urgent care and who would be triaged as having lower acuity, suggesting they have less need for health resources. However, increases in predicted wait time when wait times are already high also impacts demand for emergency services among even the sickest patients.

The most direct link between this research and previous literature relates to estimating wait time elasticities for medical care. Most of this literature examines non-acute medical care and finds modest elasticity estimates. Elective surgical wait times are estimated to have elasticities between −0.1 and −0.4 (Martin et al., 2007, Gravelle et al., 2002, Sivey, 2012, Riganti et al., 2017) depending on geographic context. In primary care, delaying appointments tends to similarly reduce demand (Lourenço and Ferreira, 2005) and shift demand into other locations (Pizer and Prentice, 2011).

However, several studies do examine wait time elasticities in emergency care. These show modest associations between the overall volume of patients presenting to emergency sites in response to increased wait times (Xie and Youash, 2011, Strobel et al., 2021). Plausibly exogenous presentations of high-needs patients - creating longer wait times - causes lower demand from low-needs patients. The wait time elasticity for these low-acuity patients is estimated as −0.25 (Sivey, 2018).

This paper also connects to a broader literature on informational and pricing problems that are more unique to medical care. This discourse originates from an initial literature on the economics of information (Stigler, 1961) and more specifically on informational inequalities between patients and physicians (Arrow, 1963). This early research spawned a theory-based literature examining the effects of information both in and out of the field of health (Akerlof, 1970, Spence, 1973).

The more dominant study of these informational issues relates to price. Estimates of price elasticities specific to emergency medicine estimate very low responses to price changes (Ellis et al., 2017). Physician responses to procedure price suggests patients cannot fully discern whether they require an intervention. For example, procedures like Cesarean-sections (Gruber et al., 1999) and pap smears (Hughes and Yule, 1992) are more often provided when the physician fee is greater, suggesting supply-induced demand. Even in cases where there is effective price transparency, price shopping is not a phenomenon consistently found among health care consumers (Brot-Goldberg et al., 2017, Chernew et al., 2018). Patients may not be able to interpret these price information signals properly (Milosavljevic et al., 2023, Pollack, 2022). There may also be a specific role that insurance plays in preventing shopping (Lieber, 2017). Qualitative evidence suggests that trust and the acuity of healthcare needs may trump any consumer impulse to price shop (Semigran et al., 2017). Price transparency can also paradoxically increase the demand for services (Kobayashi et al., 2019). Outside of health, transparency may also cause higher prices through supply-side effects in contexts where perfect competition does not exist (Albaek et al., 1997). Lack of perfect competition often characterizes healthcare, although price transparency effects in other settings depend on context and can decrease prices (Ater and Rigbi, 2022, Luco, 2022).

My contribution to these literatures is three-fold. First, I demonstrate that changes in predicted wait time information do modestly change demand in line with previous studies. I find evidence that increases in predicted wait time prevent patients from presenting to the waiting room rather than persuading patients to leave the waiting room. Uniquely though, I demonstrate that there is a negative relationship between resource needs and response: patients who seek care at urgent cares and those that are less sick are more likely to respond to changes in the wait time. However, these results are also cautionary. The natural experiment that I examine allows me to construct a wait-time demand curve and estimate elasticities for different patient types and estimated wait times. These results suggest that at very high wait times, all types of patients may be affected, including those who are very sick.