Red-flag symptoms: when an outpatient visit becomes an ED visit

Calculators for this article: qSOFA score, SIRS criteria, the Canadian CT Head Rule and the ABCD2 score.

The decision to send a patient from the office to the emergency department is one of the highest-stakes calls in outpatient medicine, and one of the least structured. It is usually made in minutes, often at the end of a full clinic, from a mental list of "red flags" that most of us were taught as facts. Some of them hold up. Many have never been tested.

The scale of the problem

Diagnostic error in outpatient care is common. Combining three large US data sets, one analysis estimated an outpatient diagnostic error rate of 5.08%, about 12 million US adults a year, and judged that about half of those errors could potentially be harmful.^[1] Missing the patient who needed the emergency department is part of that number. So is sending patients who did not.

Single red flags are weaker than their reputation

Low back pain is the best-studied example. A systematic review of red flags for malignancy and fracture concluded that many red flags in current guidelines provide virtually no change in the probability of fracture or malignancy, or have untested diagnostic accuracy. The most informative single flag for malignancy was a history of malignancy (post-test probability 33%), and for fracture the probability rose to 90% when multiple red flags were present together.^[2]

Two lessons carry beyond back pain. A single flag in isolation often means little. Several together, or one strong one, can mean a great deal.

Rules validated as a set do better

Headache shows the contrast. The SNNOOP10 list collects 15 red and orange flags for secondary headache (systemic symptoms, neoplasm history, neurologic deficit, sudden onset, onset after 65, pattern change, positional headache, precipitation by Valsalva or exercise, among others) and proposes using it to screen new headache patients. Its authors note that the lack of prospective studies on red flags leaves many questions open.^[3]

The Ottawa subarachnoid hemorrhage rule, by contrast, was tested prospectively in alert patients with headache peaking within an hour and no neurologic deficit. Any one of age 40 or older, neck pain or stiffness, witnessed loss of consciousness, onset during exertion, thunderclap onset, or limited neck flexion on examination makes the rule positive. It had 100% sensitivity (95% CI 97.2% to 100%) and 15.3% specificity for subarachnoid hemorrhage. The authors were explicit that implementation studies were still needed before routine use.^[4]

High sensitivity with low specificity is the usual trade for a rule-out tool. It is useful when it is negative, in the population it was built for.

Screening scores are prompts, not diagnoses

Sepsis is the clearest example of a bedside score being asked to do too much. qSOFA counts three findings: respiratory rate of 22 or more, altered mentation, and systolic blood pressure of 100 mm Hg or less. Patients with two or more are more likely to have the poor outcomes typical of sepsis.^[5] The validation study supported its use as a prompt to consider possible sepsis.^[6] The 2021 Surviving Sepsis Campaign guideline recommends against using qSOFA, compared with SIRS, NEWS or MEWS, as a single screening tool for sepsis or septic shock (strong recommendation, moderate-quality evidence).^[7]

A negative qSOFA does not clear a patient who looks septic.

A practical way to decide

None of this reduces to a single list. What I find useful, and what we built into Pogosh, is to separate three questions.

1. Is it new or changed? The same finding means different things in a first presentation and in a known, stable, unchanged condition. A first seizure, a new neurologic deficit or a change in a chronic symptom raises the tier. A long-standing, unchanged finding that has already been worked up often does not.
2. Is there a time-critical diagnosis that this presentation fits? Not "is there a red flag", but "is there a specific emergency that these findings support": subarachnoid hemorrhage, sepsis, acute coronary syndrome, ectopic pregnancy, cauda equina. Where a validated rule exists for that diagnosis, use it in its own population.
3. What happens after the ED? "Go to the ED" is not a plan. The useful question is whether this patient needs admission, observation, a procedure or a specialist, or same-week follow-up, and that determines whether the ED is the right door at all.

Write down why

Whatever the decision, the reasoning belongs in the note: which specific diagnosis you considered, which findings argued for or against it, which rule you applied and in what population, and what the patient was told to return for. Those are the same items a colleague, a patient or a reviewer will look for later. A sentence such as "no red flags" says nothing about which ones were checked.

What Pogosh does with this

Pogosh reads the note and returns structured next steps, including an urgency tier and a disposition. The tier is higher for new or changed presentations than for known, stable ones, and Pogosh shows that caveat rather than giving one tier for a chronic condition. A separate, strict emergency check looks for specific time-critical diagnoses and is designed not to fire on weak or indirect evidence, because an alarm on every headache would be ignored. When the request comes from the emergency department, the disposition describes what happens next (admit, observe, operate, or discharge with follow-up) rather than "ED". Relevant scores such as qSOFA are offered to the clinician to run, never computed silently.

It is decision support. The clinician who has seen the patient makes the call, and can override every part of it.