Skip to main navigation
Skip notifications
Access to POEMs and Essential Evidence Plus will no longer be included in CMA membership as of Dec. 1, 2023.
Search
"Surprise question" is moderately accurate for predicting likelihood of death
Clinical Question
How accurately does the "surprise question" identify patients who are near the end of life?
Bottom line
If you answer "No" to the question "Would you be surprised if this patient died in the next month (or year)," nearly half of those patients will indeed die in the specfiied interval (positive predictive value [PPV] = 46%). If your answer is "Yes, I would be surprised," the patient's chance of dying is just slightly more than 6%. 2a
Reference
Study design: Systematic review
Funding: Foundation
Setting: Various (meta-analysis)
Synopsis
"Would you be surprised if this patient died in the next x days or months?" These authors performed a systematic review to identify studies evaluating the accuracy of this question for predicting the likelihood of death for a range of time frames (up to 30 days, up to 6 months, and up to 12 months). The study was methodologically sound, with a thorough search, appropriate data abstraction, and assessment of the quality of these observational studies. The authors identified a total of 26 studies that met their inclusion criteria, 20 of which were set in the United Kingdom or the United States, with the oldest dating back to 2008. Study size ranged from 39 to 6642 patients, with a total of 25,718 mortality estimates in total (they were unable to extract usable data from 4 studies, which begs the question of how they met their inclusion criteria… but I digress). Overall, if the clinician said they would not be surprised by the patient's death, the patient in fact did die 2979 of 6495 times (positive predictive value = 46%). On the other hand, if they said they would be surprised, only 1238 of 19,223 patients died during the specified interval (negative predictive value = 94%). The pooled sensitivity was 71%, the pooled specificity 84%, and the overall pooled accuracy was 75%. Accuracy was a bit higher for physicians than for nurses, was not affected by the time frame (ie, death in the next week, month, or year), and was somewhat higher for patients with cancer.
Reviewer
Mark H. Ebell, MD, MS
Professor
University of Georgia
Athens, GA
Comments
Good poem
C’est plus amusant qu’utile pour donner un pronostic au patient et à sa famille . Ça ne fait pas sérieux !
Ce n’est d’aucune utilité pour demander et organiser des services en soins palliatifs
This is important for advance care planning
Great tool for #palliative care early identification
Trust your instincts. Then what? Do we use it for teaching, inspiring change, or perhaps helping family members prepare.
i wonder then if I shoild be discussing this with the patients in terms of palliative care and also billing a K023 for them.I believe DNR discussions shoild be had with some of these patients and are not done often enough
Surely the original article’s discussion has some proposals for what use this is. The use that comes to my mind is not as a “preparation” for the family, but as a physician self-check. Eg. Is the intervention I was considering still valid for a person not likely to survive the year? (Eg. How important is that repeat cholesterol or bone density test at this stage for this person)
In my Active practice I used statistic from book but added that statistic may be to old for these days and it was true at least for GBM .