Article Text

Download PDFPDF

Admission prediction rules: some limited promise, but far from proven
  1. Elaine Rabin
  1. Correspondence to Dr Elaine Rabin, Emergency Medicine, Icahn School of Medicine at Mount Sinai, New York City, NY 10029, USA; elaine.rabin{at}mssm.edu

Statistics from Altmetric.com

The internationally shared problem of emergency department crowding appears to have generated an interest in admission prediction rules. These are not rules to predict which patients would benefit from an admission—a clinical question—but rather who will be admitted—an administrative question that has repercussions for the entire hospital system. These rules are conceptually different from the prediction rules usually encountered in medicine, which aim to provide clinicians with an evidence basis to better target testing and treatment in order to improve patient outcomes. The immediate goal of the admission prediction rules presented here is to improve the efficiency of care processes. So could these rules still benefit patients, or only hospital administrators? What is the potential utility of such rules? And do those presented here live up to that potential?

In their recent study from Australia, Ebker-White et al validate the Sydney Triage to Admission Risk Tool (START) previously designed for early identification of ED patients who will be discharged.1 The authors then develop an extended tool (START+) to identify patients expected to have lengths of stay of under 48 hours, that is, discharged directly from the ED or from short-stay units. Also in this issue, Kraaijvanger et al report on the development and validation of another admission prediction rule.2 In a previous issue of EMJ, Lucke et al proposed separate rules for patients older and younger than 70 years to predict admissions.3

Though these rules appear primarily concerned with administrative processes, very real improvements in patient outcomes could theoretically result if the rules’ implementation successfully decreases crowding in the ED. The international epidemic of hospital crowding continues, and a large literature demonstrates not only its association with increased morbidity, but even mortality effects.4 ED throughput problems are a minor contributor to crowding compared with impairments in overall patient flow through the hospital, which create backups.5 However, inefficient ED processes do contribute to crowding, and are often the only piece of the puzzle over which the ED has control, so they remain worthy targets for improvement. Increasing the efficiency of throughput processes is the goal set by the authors of admission prediction rules.

The authors of the START+ tool suggest that it be used to stream patients early to short-stay units, and Kraaijvanger et al suggest that bed requests could be placed at triage for patients expected to be admitted. Lucke and colleagues did not describe how their tool would be implemented. No study yet, including by these authors, has demonstrated that an admission prediction rule can reduce crowding in practice. (Kraaijvanger et al and Lucke et al indicate that implementation studies are planned). There are several hefty obstacles which could thwart any admission prediction rule’s success: requirements for significant accuracy and generalisability, and cultural norms in admission processes. It may also be difficult to design a rule that performs better than the judgement of triage nurses and other clinicians, who already incorporates knowledge of local practices and resources.

Regarding cultural norms, an increasing amount of imaging and other testing has come to be expected in the ED at many institutions prior to the placement of a bed request.6 This information is often used to confirm that the patient requires inpatient admission, and to sort patients into services (medical, surgical, neurological, and so on). Similarly, some workup may also be expected in the ED prior to placement in a short-stay unit. Without redesigning systems in these institutions to provide inpatient or short-stay care for unsorted patients, earlier bed requests are unlikely to expedite patient movement to these beds.

Accuracy is required because admitting patients unnecessarily could actually decrease patient flow and worsen crowding by filling the hospital. This is especially true if it causes hospitals to exceed 85% capacity at which efficiency has been shown to decrease.7 Flow can similarly be impeded if a short-stay unit becomes congested with patients who are eventually admitted and boarding there.

The accuracy of the rules described here varies. The results of the START+ study reveal that it is likely more useful to identify short-stay/discharged patients than admitted patients: Only 2.5% classified as ‘very likely’ or ‘likely’ discharge/short stay were admitted, while 63.1% of patients classified as ‘likely’ or ‘very likely’ admission were discharged.1 Of note, the rule does not distinguish between short stay and discharge, calling into question its utility in streamlining patients to short-stay units. In the study from Ebker-White, using optimal cut-offs, 2.4% of patients at the community hospitals and 6.6% at the academic hospital were ‘wrongly admitted’, while 14.1% and 19.7% respectively could be identified early for admission.2 Lucke et al show a positive predictive value of 0.71 for the ‘top 10% most likely to be admitted patients’ under age 70, and 0.87 for those 70 and older. They do not provide more general accuracy information.3

Whether these results are precise enough to improve flow in a hospital depends on the capacity at which a given hospital usually operates. In addition to the negative effects on a system, inappropriate admissions incur costs to the patient too: financial, time, exposure to hospital-acquired infection, risks associated with deconditioning and mobility, and so on.

Next is the issue of generalisability, which in the case of admission rules is likely to be significant. Nearly universally, a young healthy patient presenting with a upper respiratory infection (URI) will not benefit from an admission, while an elderly dialysis patient with sepsis will. In fact, in deriving Ebker-White et al’s rule, the authors excluded these ‘obvious’ cases from analysis as unlikely to require application of a rule.2 However, for patients with moderate illness, comorbidities and social issues, admission decisions are based on a complex web of factors, which vary locally and individually. Resources available to the individual as an inpatient versus as an outpatient are considered, and the decision often will depend on the hospital’s and community’s resources, and on the individual’s access to care given their mobility, insurance status, and so on. Local standards of practice are also incorporated into the admission decision, and especially in litigious areas, the risk tolerance of the provider may come into play. As such, a priori, it is a daunting challenge to design a general rule that could predict admission decisions across patients, hospitals and communities. The rules developed by Kraaijvanger and colleagues were derived at institutions that do not care for cardiac emergencies, introducing an obvious potential issue with generalisability.

Lucke et al’s rule, like most of those reported in the past, was derived and validated at a single institution, Leiden University Medical Center in the Netherlands, though the authors have a multi-institutional study planned. The other two studies provide somewhat stronger, but still limited, evidence of generalisability: The START tool was developed using regional data and START+ is validated at two hospitals in Australia. Ebker-White’s rule was prospectively validated in three hospitals in that country, finding decent but varying efficacy across them, which confirms the challenge.

One reason for optimism in finding a generalisable admission prediction rule is that the rules in these studies seem to identify similar predictors. Age, triage category, chief complaint and mode of arrival are incorporated into all three rules as significant predictors, and have also appeared in other rules.8–11 However, these factors all seem to be proxies for the concept that older and sicker patients are more likely to be admitted. Older patients are frailer and more likely to have comorbidities. (Frailty and comorbidities are directly incorporated into START+. Medical history was not found to be a significant predictor by Kraaijvanger et al 2 and was not examined by Lucke et al 3). Does a clinician really need a rule to understand this?

Which leads to the third challenge: whether any of these rules outperform an a priori assessment by a triage nurse or screening physician. This information is crucial to any assessment of the utility of an admission prediction rule. A study examining triage nurses’ ability to predict admissions, performed at a single Australian hospital, found a positive predictive value of 48.5%, implying possible room for improvement. (Physicians fared little better).12 However, Peck et al 8 modelled several admission prediction rules including triage nurse opinion, and did not find drastic differences in performance, while Kim et al 9 found no difference between their rule and triage nurse opinion.

In conclusion, a growing number of admission prediction rules are being proposed, and seem to identify a common list of factors associated with admission. However, admission prediction rules have not yet proven their worth. In addition to the need to be accurate, generalisable and superior to clinician judgement, evidence is needed that they can actually improve efficiency and patient care. Demonstrating a rule can satisfy these requirements is a crucial next step in justifying further investigations in this area.

References

Footnotes

  • Contributors ER was exclusively involved in the planning and writing of this editorial.

  • Funding The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.

  • Competing interests None declared.

  • Patient consent Not required.

  • Provenance and peer review Commissioned; internally peer reviewed.

  • Author note ER is an associate professor in the Department of Emergency Medicine, associate programme director for the Residency in Emergency Medicine, and director of Resident Research at the Icahn School of Medicine at Mount Sinai in New York, New York. She studies and has published on several health policy topics including hospital crowding and emergency department boarding. She previously served as the co-chair of the Crowding Interest Group for the Society for Academic Emergency Medicine and currently serves as co-chair of the Choosing Wisely Sub-committee, which focuses on limiting low-value care, of the Quality and Patient Safety Committee for the American College of Emergency Physicians. ER is also a member of the New York Department of Health’s Emergency Department Advisory Group for Non-fatal Opioid Overdoses.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Linked Articles