How Delirium Is Detected With the Confusion Assessment Method

The Confusion Assessment Method, or CAM, detects delirium by converting a formal psychiatric definition into four observable features that a trained clinician can check at the bedside in a few minutes. A patient screens positive only when there is acute onset with a fluctuating course and inattention, plus either disorganized thinking or an altered level of consciousness. In its original 1990 validation the instrument reported sensitivity of 94 to 100 percent and specificity of 90 to 95 percent, but later pooled data are more modest, and those figures describe how the tool performs across studies rather than a guarantee for any single patient.

Turning a definition into an algorithm

The CAM came out of a 1990 study by Sharon Inouye and colleagues in Annals of Internal Medicine. The problem it addressed was practical. Delirium is common in older hospitalized patients and carries real risk, yet the reference standard for diagnosis was a psychiatric interview against the Diagnostic and Statistical Manual, which most bedside clinicians were not positioned to perform. The authors took the DSM-III-R criteria for delirium and operationalized them into nine items, then singled out four features they hypothesized in advance would carry the diagnostic weight. That last step is the heart of the design. Rather than asking a clinician to hold an entire diagnostic manual in mind, the CAM specifies which observations to make and exactly how to combine them.

The four features and the rule that binds them

The diagnostic algorithm rests on four features: acute onset with a fluctuating course, inattention, disorganized thinking, and an altered level of consciousness. The combining rule is deliberately restrictive. Delirium is registered only when the first two features are both present, together with either the third or the fourth. Acute change and fluctuation are usually established from a family member or nurse who knew the patient's baseline. Inattention is tested directly, for example by asking the patient to recite something backward or by noting distractibility during conversation. Disorganized thinking shows up as rambling, illogical, or unpredictable speech, while altered consciousness ranges from drowsiness to hypervigilance. The instrument works as a logic gate rather than a symptom checklist, and that structure is what makes it reproducible across different raters.

What the reported accuracy means

Two features of the accuracy data deserve attention. The original validation, done on 56 patients aged 65 and older at two academic sites, reported very high sensitivity and specificity. A later systematic review and meta-analysis by Shi and colleagues, published in 2013 and pooling 22 studies with 2,442 patients, found pooled sensitivity of about 82 percent and pooled specificity of about 99 percent. The specificity figure is the more reassuring one. High specificity means a positive CAM is unlikely to be a false alarm, so a positive result strongly argues that delirium is present and should trigger a search for the cause. Sensitivity of roughly 82 percent tells a different story. It means that across those studies, close to one in five patients with delirium screened negative. A negative CAM lowers the probability of delirium but does not exclude it, which is why the meta-analysis authors framed the tool as a complement to clinical judgment rather than a substitute for it.

Why the numbers move with training

The gap between the near-perfect original figures and the more modest pooled estimates is not a flaw in the instrument so much as a fact about how instruments behave once they leave the study that created them. Sensitivity in particular falls when the CAM is applied without structured cognitive testing or without training in what each feature actually requires. Work on standardizing the CAM for multicenter trials, reported in BMC Geriatrics in 2019, describes the training and calibration needed to keep assessments consistent across raters and sites. The lesson generalizes beyond this one tool. A reported sensitivity belongs to the conditions under which it was measured, including who administered the test and how carefully. Used as a quick yes-or-no glance without formal attention testing, the CAM misses more cases than its headline numbers suggest.

The limits worth naming

A few boundaries follow directly from how the tool works. Hypoactive delirium, in which a patient is quiet and withdrawn rather than agitated, is easy to overlook and accounts for a share of the missed cases. Baseline dementia complicates the picture because inattention and disorganized thinking can predate the acute illness, which is why establishing the patient's usual baseline is built into the first feature. And the CAM detects a syndrome; it does not explain why the syndrome is there. A positive result is the start of an evaluation for infection, medication effects, metabolic disturbance, and other reversible contributors, not the end of one. Read this way, the instrument does what it was designed to do: it makes a difficult diagnosis reproducible enough for non-psychiatrists to act on, while leaving the harder clinical reasoning in place.

This article is educational and not medical advice.