Sepsis without a fever: 2,506 real intensive-care charts, hour by hour, before sepsis was dated

Each patient’s highest temperature in the 12 hours before sepsis

Above, each dot is one of the 2,445 patients who developed sepsis and had at least one temperature charted in those hours: the highest value charted from 12 hours before the dated moment up to it. Below, for comparison, one patient from the same hospital who did not develop sepsis, in intensive care at the same hours since admission.

In hospital A, 64.8% of 1,535 patients had no reading of 38 °C or more; in hospital B, 66.4% of 910. A charted fever was more common in those with sepsis than in the others (35.2% against 21.1% in A, 33.6% against 11.4% in B), but most of those with sepsis had none charted. A normal temperature is no reassurance: 15.8% (A) and 18.0% (B) of them had a reading below 36 °C, also a warning sign.

Temperature was taken rarely: the median is 3 readings across the 13 hours of the window, from 12 hours before up to the dated hour. Between two readings the chart says nothing, and several readings in one hour appear in the data as their median, so a short fever may have gone uncharted. For 3.3% (A) and 1.0% (B) of the patients with sepsis no temperature was charted in those hours; they are left out of the percentages.

What a simple rule catches, and how many false alarms it raises

Over a window of about 15 hours (from 12 hours before the dated moment to the end of the chart), for 2,491 patients with sepsis and 2,504 without, at the same hours of their stay: how many each rule would have raised the alarm for at least once.

Picture 1,000 patients like these, 73 of whom develop sepsis. The rule “temperature ≥ 38 °C”, applied over that window, would raise the alarm for 27 of the 73 and, over the same hours, for 167 of the 927 who do not develop sepsis. Over a whole stay there would be more false alarms still. A pulse above 90 would flag 74% of the patients with sepsis, but also 58% of the others.

No single vital sign on its own separates the two groups well. The sepsis moment was dated from lab tests and from the team’s decisions, blood cultures and antibiotics, which the chart here does not show.

What “the moment sepsis was dated” means

The moment is neither the start of the illness nor the minute a doctor made the diagnosis. The organisers of the 2019 PhysioNet challenge computed it with the Sepsis-3 criteria3: the earlier of two moments. One is suspicion of infection: blood cultures taken and intravenous antibiotics given within a short interval of each other, with the antibiotics continued for at least three days. The other is a rise of at least two points in the SOFA score, which measures how badly the organs are affected, from 24 hours before the suspicion to 12 hours after2. The three-day antibiotic condition uses information from later on, so the moment was not known at the bedside at the time.

The charts in the data stop a few hours after that moment (median: 3 hours), which is why this page looks at the hours before. For 426 patients the sepsis label was already on at their first recorded hour, so the moment cannot be pinned down; they are left out.

What the data cannot say: whether patients were given fever-lowering drugs or cooled down, why they had no fever, or what sepsis looks like at home. These are people already in intensive care.

What other studies found

The finding is not new. A study of 5,532 patients diagnosed with sepsis in the emergency department, from the MIMIC-IV database of Beth Israel Deaconess in Boston, sorted the course of their temperature into four patterns. The largest group, 74.1%, followed a normal-temperature pattern. The smallest, 0.9% with a low temperature, had the highest death rate: 37.5%4.

In the emergency department of the same hospital, 45% of patients with septic shock had no fever (neither reported nor measured above 38 °C). They were given antibiotics in the department less often (81% against 94%) and died in hospital more often (33% against 11%)5. That does not prove the missing fever caused the difference, but it shows why it matters.

Both studies come from Beth Israel Deaconess, which is hospital A in the data here. The independent check is hospital B, Emory in Atlanta, where the share is almost the same.

At home: the signs are not only fever

The data above come from intensive care and do not describe what sepsis looks like at home. Guidance does. The NHS lists these signs in adults: confusion or slurred speech, uncontrollable shivering, muscle pain, difficulty breathing, blue, pale, grey or blotchy skin, lips or tongue, a high or low temperature (a high temperature is less common in older adults), and not peeing all day or very little in the past 18 hours6. Someone may not have every sign.

If you think someone may have sepsis, call 999 (in the UK; 112 across Europe) or go to the emergency department now. The game above takes points for false alarms to show what a rule that goes off for everyone costs. At home it is the other way round: when in doubt, get help.

The guide for patients and families Sepsis Clarity explains the warning signs, what to ask in hospital and what recovery looks like.

In Romania: how many people die of infections (pneumonia, sepsis, C. difficile) and how that compares with the rest of the EU.