July 19, 2017

How to best identify patients that have a high likelihood of dying in one year?

By Daniel Cabrera

Author: Caitlin Loprinzi-Brauer, MD @cloprinzi

To determine who should have goals of care discussions, encourage advanced care planning and consider palliative medicine consultation


Cardiopulmonary resuscitation (CPR) is the only medical intervention that does not need consent to be preformed. In medicine all patients are assumed to be a full code and undergo CPR and most times intubation (insertion of a breathing tube), unless otherwise documented. For many having all interventions preformed in attempt to save their life is what they want and what would be recommended by most medical professionals. As patients age or have a decline in their health secondary to medical comorbidities, many patients do not want aggressive heroic measures taken, particularly in times of critical illness. Also, as people age and medial comorbidities increase, the likelihood of survival following cardiac arrest decreases.

Value-based decision-making is about eliciting a patient’s values and goals in order to guide treatment; however, patients are not always empowered or able to communicate their wishes. Advance healthcare directives (AHD) can help patients communicate their wishes regarding their care and can guide conversations with surrogate decision-makers when the patient lacks capacity for such discussions. Understanding patients’ goals and values is especially important in the case of critically ill patients presenting to the emergency department (ED) who would receive life-sustaining treatments unless this is not what the patient wants.


We externally validated and derived a clinical prediction tool previously created by an ED group in Australia (Richardson P et al. https://www.ncbi.nlm.nih.gov/pubmed/25062815). We identified patients likely to die within one-year;, as these patients ideally should have advanced care planning (i.e. completed an advanced directive or similar document) including their code status documented (ie, Do not resuscitate or Do not intubate).

The PREDICT screening tool is composed by six variables that are easy to collect within a clinical ED visit and is easy to utilize in the fast-paced environment of the ED. Each criterion is assigned different points as seen in the table below:

PREDICT criteria and score

Feature Points
Referral to palliative care team for a non-cancer diagnosis 12
Current residence in nursing home 3
Department of intensive care unit (ICU) admission with multi-organ failure 10
Current diagnosis of cancer 10
2 medical admissions in the past year 3
Age at ED visit in years
   55-65 1
   66-75 2
   76 3


In our study, we applied the PREDICT tool to 927 patients older than 55 years of age presenting to the ED. A fourth of them (26%) were deceased at one year. Patients from the deceased group were older, had a higher PREDICT score, had an increased number of comorbidities, were more likely to live in a nursing home, and were more likely to have an AHD on file. Of the 147 patients with PREDICT scores ≥13, 55% were deceased at one year. The AUC of the PREDICT score was 0.717, sensitivity 33% and specificity was 90% to predict 1-year mortality. High specificity is preferred in this scenario, however a table with different sensitivity and specificity and ROC curves depending on the PREDICT score cutoff used are presented in the full article.


We further refined the clinical prediction tool to improve the original PREDICT criteria, and the PREDICT minus ICU admission with multiorgan failure (modified PREDICT) appeared to have a diagnostic test accuracy performance similar to the original PREDICT score. We selected this as the preferred model, as the variable of ICU admission with multiorgan failure is not always available at the time of assessment in the ED, was the hardest one to extract from the records, and will make the model easier to apply earlier in the course of an ED visit. By removing the ICU admission variable, this clinical tool, could result in earlier calculation of the PREDICT score during an ED visit. This time could be used for initiating conversations regarding AHD and, if appropriate, allow placement of palliative care consultation before inpatient admission.


PREDICT and modified PREDICT have advantages over other prognostic indices. Other indices require a patient be assessed for factors not regularly gathered in the context of a brief clinical encounter in the ED setting, require the use of the complex non-cancer hospice guidelines that may not be familiar to those who work outside of hospice and palliative medicine, incorporate laboratory values that may not be available or up-to-date, cannot be applied until the patient has final disposition from the ED, or cannot be applied to patients until they are dismissed from the hospital. PREDICT and modified PREDICT use data commonly available in the medical record or that are clinically relevant to an ED physician and are generally collected within the confines of a clinical encounter in the ED. Further, either PREDICT score could be discussed directly with the patient and their caregivers while they are in the ED and, thus, AHD, goals-of-care discussions and, when applicable, the need for palliative care consultation can be identified earlier and, possibly, initiated before the patient is admitted to the hospital.

Link to the paper



  • Moman RN, Loprinzi Brauer CE, Kelsey KM, Havyer RD, Lohse CM, Bellolio MF. PREDICTing Mortality in the Emergency Department: External Validation and Derivation of a Clinical Prediction Tool. Acad Emerg Med. 2017 Jul;24(7):822-831. doi: 10.1111/acem.13197. Epub 2017 May 29. PubMed PMID: 2840162

Tags: Uncategorized

Please sign in or register to post a reply.
Contact Us · Privacy Policy