Detecting Delirium in the ED


Eight hours ago, the 80-year old female in room one arrived from home pleasant and cooperative after suffering a standing-level fall. She’d been ambulating to the bathroom during the commercials for her favorite evening TV show, Wipe Out.

Educational Objectives:

After evaluating this article, participants will be able to:
1. Incorporate strategies into practice to improve detection of delirium
2. Recognize the implications of misdiagnosing delirium and modify practice styles to improve patient safety
3. Utilize currently available decision tools to better identify delirium.


Emergency physicians miss this Dx in 87% of cases. Can CAM help?

Eight hours ago, the 80-year old female in room one arrived from home pleasant and cooperative after suffering a standing-level fall. She’d been ambulating to the bathroom during the commercials for her favorite evening TV show, Wipe Out. With a tender left hip and deformed left leg, you quickly diagnosed a hip fracture but have since been awaiting the confirmatory X-ray and cranial CT to exclude additional injuries. After 6 mg of morphine, her pain was well controlled at the time of your initial evaluation and when you twice re-evaluated her. However, one-hour ago she began to talk in an increasingly loud voice to nobody in particular. Nurse and physician efforts to calm her have been unsuccessful and you are now contemplating chemical sedation and/or physical restraints to prevent her from falling out of bed in her agitated state. You suspect delirium, but want to confirm your diagnostic impression before you restrain her. After reassessing her glucose, oxygenation, and self-reported pain control, you request that her CT be expedited to exclude an intracranial hemorrhage and then begin contemplating your next action.

What is your response to this patient’s change in mental status?
A. Ignore the patient’s behavior. This is simply dementia worsened by analgesics.
B. This is dementia; sedate the patient further.
C. This is dementia; restrain the patient to protect her.
D. Differentiate between dementia and delirium using bedside tool. If the latter, look for cause to treat appropriately.


The Question:
What bedside tools most accurately distinguish delirium from other etiologies of altered mental status?

The Evidence:
Wong CL, Holroyd-Leduc J, Simel DL, Straus SE; Does this patient have delirium? Value of bedside instruments. JAMA 2010 Aug 18; 304(7): 779-786.

The Bottom Line:
Among 11 instruments the CAM is the most time-appropriate (< 5-minutes to obtain) and accurate tool to identify delirium. EPs miss up to 87% of delirium in geriatric patients which can result in increased short-term ED recidivism, prolonged ED and hospital length-of-stay, and accelerated functional decline for our ever expanding aging population. Since older adults usually do not walk into the ED with the diagnosis of “altered mental status”, “dementia”, or “delirium” stamped on their forehead, EPs must remain cognizant of occult impairments. Prompt recognition of delirium mandates that astute clinicians focus on the detection of underlying etiologies like infections, medications, or occult trauma. So, remember the following lessons

  • Delirium kills.
  • For those it doesn’t kill, a good proportion will have accelerated cognitive / functional decline, and end up in a nursing home (not a great way to end your life).  Delirium threatens quality of life.
  • Physicians miss delirium a lot because they don’t screen for it. 
  • Missing delirium = bad consequences for the patient. We can miss a potentially life-threatening illness or infection.
  • We need to screen for delirium in order to not miss it.
  • The CAM is probably the best we have, but we have ongoing studies trying to validate more rapid tools.
  • Once delirium is diagnosed, find the underlying cause.

A survey of EPs in 1992 confirmed that 59% of us believe that the evaluation of an altered mental status in older adults is more difficult than in younger adults. Cognitive dysfunction includes mild cognitive impairment, dementia, and delirium. Dementia manifests as a gradual, generally irreversible decline in higher cognitive functions that impacts day-to-day activities including memory and organizational skills. Delirium, on the other hand, is an acute and fluctuating alteration in cognition that resolves when the underlying medical etiology is addressed. Patients can present with delirium (prevalent delirium) or develop delirium during the course of their ED evaluation (incident delirium). In addition, delirium exists in hypoactive and hyperactive forms with the former accounting for the majority (92%) of cases. For older adults in the ED common etiologies of delirium include medications, infections, pain, and physiologic stressors like sleep deprivation or myocardial infarction.


Delirium is present in 8-30% of geriatric adults in today’s ED (Naughton 1995, Lewis 1995, Ellie 2000, Hustey 2002, Han 2009), but emergency nurses and physicians miss up to 87% of cases. Really, 87%? Using the CAM or the CAM-ICU as the criterion standard for ED patients, here’s the data.

Lewis et al enrolled 385 ED patients with a delirium prevalence of 10%. A chart-review suggested that EPs only recognized 17% of these cases. Elie et al screened 447 ED patients over age 65-years noting a 9.6% prevalence of delirium and EP failed to recognize 76.5% of these. Hustey et al enrolled 297 ED patients over age 70-years noting a 10% prevalence of delirium and EP documentation of delirium, cognitive impairment, or any acceptable synonym in 13% (4 patients/30 with CAM-defined delirium) meaning that 87% were missed. When these investigators included any mention of disorientation or abnormal behavior as recognition of cognitive impairment (albeit not necessarily recognized delirium), then 43% of patients had documentation of impaired mentation. Han et al enrolled 303 ED patients over age 65-years and 25 (8%) met criteria delirium. Of those with delirium and 19 (76%) were not recognized by the EP.

Furthermore, most of these patients are not immediately recognized by inpatient services.  The consequences of failing to diagnose delirium for our patients can be dire, but many EP’s fail to alter their clinical response when the diagnosis is recognized.  We should.  Delirium is associated with increased ED length of stay, hospital length of stay, medical expens
es, functional decline, ED recidivism, short-term mortality, and accelerates the natural course of baseline dementia.

Delirium-related care costs reach $152 billion annually, but effective management models are being evaluated for some populations like geriatric adults and hip fracture patients. Prompt recognition of delirium has also been highlighted as an EM Quality Indicator and a minimal core competency for EM resident physicians meaning that delirium recognition and management priorities will probably soon be appearing on board certification exams.
One historical barrier to detecting delirium is insufficient training with available instruments. Traditionally, EP’s have used the Mini Mental State Exam to assess for cognitive impairment.

Unfortunately, the MMSE was not designed to assess delirium and has some of the worst diagnostic test characteristics of the instruments described here to rule-in the delirium (positive likelihood ratio 1.6). Since diagnostic research evidence continues to improve, clinicians should take advantage of this article’s high-quality summary to shape their practice for the detection of delirium in vulnerable populations like the elderly.

The objective of this review was to determine the diagnostic accuracy of bedside delirium instruments by assessing a systematic review of the medical literature since 1950. Eligible studies had to use a Diagnostic and Statistical Manual of Mental Disorders criterion standard (DSM-III, DSM-III-R, or DSM-IV) performed by an appropriate specialist physician (geriatrician, neurologist, or psychiatrist) with >80% capture of eligible subjects. Alcohol-related delirium and predominantly pediatric population-based studies were excluded as were studies using tests not available at the bedside.

After identifying 6570 potential studies, the authors ultimately reviewed 25 studies in detail that included over 3000 subjects with delirium prevalence ranging 9% to 63% while evaluating 11 bedside delirium instruments (Table 1). None of the instruments had been explicitly evaluated in ED settings. Among these 11 instruments, only four instruments demonstrated useful positive- and negative-likelihood ratios AND could be administered in less than 5-minutes: CAM, DOSS, GAR, and Nu-DESC.

The CAM had several advantages over the other instruments including simplicity (Table 2), enhanced external validity (most often studied in a variety of settings), and excellent diagnostic accuracy. The CAM had a summary-positive likelihood ratio of 7.3 (95% CI 1.9-27) and a summary-negative likelihood ratio of 0.08 (95% CI 0.01-0.38) when performed by nurses. This means that if one starts with a 20% pre-test probability of delirium, a positive CAM equates to a 65% post-test probability of delirium, while a negative result reduces the post-test probability to 2%. On the other hand, when the CAM is administered by physicians the summary-positive LR is 65 (95% CI 9.3-458) and negative LR 0.06 (95% CI 0.01-0.38). The CAM can also be completed in less than 5-minutes.

The Caveats:
By design the gold standard for delirium in this systematic review was the DSM, but the reliability of the criterion standard itself is far from perfect with two geriatricians demonstrating kappa between 0.72-0.74 using DSM-III to DSM-IV criteria. Clinicians without geriatric-specific training might yield even less reproducibility.

None of these studies assessed the accuracy or reliability of the CAM (or any other instrument) in ED settings. Monette et al has traditionally been referenced as an ED validation of the CAM, but this trial did not have an emergency nurse or physician CAM validated by an independent specialist diagnosis using DSM criteria so it was not included in this review and might not be considered a true validation trial. Recent ED-based geriatric delirium research has used the CAM-ICU as a surrogate diagnostic criterion because this instrument has been validated in ICU settings and can be quickly (< 2 minutes) administered with minimal training with high reliability.  Like the CAM, though, the CAM-ICU has not been validated in ED populations so the diagnostic accuracy and reliability of this instrument during emergency care remains undefined.
Optimally, the CAM (or any cognitive screening instrument) requires training to be used efficiently. Online training is available for the CAM. The CAM-ICU also has training manuals and instructional videos online. In addition to well-conducted ED-based validation trials of these (and other) instruments, future trials will need to assess the optimal duration and frequency of training for physician and nursing personnel to reliably and cost-effectively identify delirium.

Case Outcome:
Your patient meets CAM criteria for delirium by fulfilling Features 1, 2, & 4 above. The cranial CT is unremarkable, but when the labs ultimately return you note a UTI with >50 WBC’s in the UA with 3+ LE and nitrites +. You initiate appropriate antimicrobial therapy and discuss admission to the “delirium room” with your geriatrician. When you follow-up on the patient the next evening, you note that the energetic and intellectually intact octogenarian whom you first encountered has returned.


  1. What is the significance in difference in post test summary LR when nurses vs. doctors administer the CAM?
    How was analysis impacted by addition of .5 to some LRs?

  2. Stats:

    The significance of the difference between the summary positive likelihood ratios for nurses (7) and physicians (65) is that the physician post-test probability is increased quite a bit more. For example, as noted in the article a pre-test probability of 20% will be increased to a post-test probability (using a Fagan’s nomogram or multiple online calculators such as MEDCALC) to 65% with a positive CAM by nurses but will be increased to 94% with a positive result by physicians. The reasons why nurses and physicians have such different positive LR’s is not clear. Their negative LR’s (i.e. ability to rule-out the diagnosis of delirium) are essentially the same.

    Not sure what you mean by the second questions?

    Chris Carpenter, MD, MSc
    Washington University in St. Louis
    Director of Evidence Based Medicine
    Chief Clinical Editor, EP Monthly

  3. In the Methods section of the article the authors indicate that “If a study contained any zeros in the 2 by 2 table, resulting in likelihood estimates of zero or infinity, 0.5 was added to all the counts for that study for calculating the LR and respective confidence intervals (CIs).” Would this affect comparison of the studies or analysis? How do authors decide measures such as this for a systematic review?

  4. Stats:

    The JAMA Rational Clinical Exam authors added the 0.5 to the 2×2 table cells that were zero in order to compute likelihood ratios. As you know, the formula for LR- = (1-sen)/spec and the formula for LR+ = sen/(1-spec). If one conducts a study with 100% sensitivity than the LR- would be zero regardless of the specificity and if the specificity is 100% than the LR+ is infinity. Neither of these figures is useful for computing post-test probability estimates and they do not permit us to compute 95% confidence intervals for individual studies.

    No diagnostic study is truly 100% sensitive or 100% specific. A study that reports these figures just has not evaluated enough subjects to encounter a false-negative or false-positive (respectively). I cannot find a reference for the methods that the Delirium RCE authors used, but they are simply picking a number in order to be able to compute LR+ and LR- estimates with 95% CI’s. Would 0.1 or 1 be better estimates? I don’t know.

    Some readers might be interested to learn that statisticians recommend that diagnostic systematic reviews not pool likelihood ratios or compute 95% confidence intervals around the summary likelihood ratios. For more on this perspective see As I understand the statistician’s viewpoint, sensitivity and specificity are the “more natural” laboratory parameters for numerical analysis and not plagued by the same 0 to infinity problem that LR’s have. Therefore, the “add 0.5” solution is not required and 95% CI’s can be readily computed on any result (0% to 100%) for sensitivity & specificity. The LR’s can then be computed from the summary estimates of sensitivity & specificity.

    Hope this helps.

    Chris Carpenter

  5. May we please see the Venn Diagram for Abnormal Cognitive Screening Tests again? How were you able to put this together?
    Thank you.

Leave A Reply