Kaiser Sepsis Calculator

Illustrative bands and example scenarios

Bands are this page’s own teaching labels, not Kaiser thresholds and not management categories. A lower band never means a baby is safe, and a higher band never means a baby has sepsis.

Related tools

How it works in 3 steps

Deep dive

What this page is, and what it is not

This page is an educational illustration of how the published risk factors behind the Kaiser Permanente early-onset sepsis approach combine, and it is not the official Kaiser calculator, its fitted model, or its score. That distinction is stated in the page’s own result heading, “Illustrative result — not a Kaiser score”, and it is repeated in the note beneath the result, which says the figure is an arithmetic illustration from the multipliers printed on the page, not a validated clinical risk, and not a reason to seek or skip evaluation. The reason for the distinction is stated in the page’s own frequently asked questions: the official Kaiser model coefficients are not published in a replicable form, so no page here can honestly compute the official score. This page instead multiplies a visitor-entered baseline incidence by fixed teaching multipliers that are printed on the page itself, so that every step of its arithmetic is visible and checkable. A reader who understands that method understands both what the page can teach and what it cannot decide.

The method is deliberately transparent, and the default result shows every step. The visitor-entered baseline incidence is 0.60 per 1,000 live births, and every multiplier at the default inputs is 1.00 except the antibiotics multiplier, which is 0.80 for the default selection of broad-spectrum antibiotics given four or more hours before birth. Multiplying 0.60 by 0.80 gives the default illustrative index of 0.48 per 1,000, which the page places in the Lower illustrative band with zero risk flags counted. Each multiplier is printed in the result’s “How the index was built” breakdown, so a reader can reproduce the default result by hand: 0.60 times 1.00 for gestational age, times 1.00 for maternal temperature, times 1.00 for rupture of membranes, times 1.00 for Group B Streptococcus status, times 0.80 for antibiotics, times 1.00 for clinical presentation, equals 0.48. The bands themselves are this page’s own teaching labels, not Kaiser thresholds, and the page’s own section heading says so. A band on this page is a way of grouping illustrative results for teaching, never a clinical category.

What the page refuses to do is as important as what it does. It gives no diagnosis, no test advice and no antibiotic advice, and its own frequently asked questions state that only the baby’s own clinician, with examination and laboratory results, can make those decisions. A higher illustrative band means only that more of the published risk factors are present in the inputs: it never means the baby has sepsis, and a lower band never means the baby is safe. Those two sentences are the page’s own wording, and they are the correct frame for every figure in this guide. The page also carries safety messaging that appears with the result: when clinical illness is selected as the presentation, an urgent banner appears stating that the presentation selected is clinical illness, that the page gives no care decision, and that immediate clinical evaluation should be sought now, and when the presentation is equivocal, a warning appears stating that the page gives no care decision and that prompt clinician review is the safe course. Those banners are part of the page’s design, not decorations, and they should be read as the page’s own standing advice whenever they appear.

The baseline incidence deserves special attention, because it is the one input the page does not supply. The default of 0.60 per 1,000 is a visitor-entered starting value, not a published constant, and the official tool uses the institution’s own early-onset sepsis incidence, which varies by setting. Changing the baseline to 1.20 per 1,000 doubles the default index to 0.96 per 1,000, because every other multiplier at the defaults is unchanged, and that doubling shows why the baseline matters so much: the index is only as meaningful as the incidence figure fed into it, and a reader who uses a baseline from a different setting is computing an illustration for that setting, not a universal risk. The honest use of this page is therefore to enter a locally relevant baseline if one is known, or to treat the default as a teaching example, and never to treat the resulting index as a measured risk for any individual baby.

Each input, and what it does to the index

The calculator’s seven inputs are the same seven the official tool uses, and the page’s own frequently asked questions list them: the institution’s early-onset sepsis incidence, gestational age, highest maternal temperature, rupture-of-membranes duration, maternal Group B Streptococcus status, intrapartum antibiotic type and timing, and the infant’s clinical presentation. On this page each input contributes one multiplier, and the index is the baseline multiplied by all seven, so changing one input changes the index only through its multiplier, and the result breakdown prints each multiplier so the effect is visible. The table below records every multiplier verified by running the page’s own calculator, with the default multipliers in the first row and each single-input change in the rows that follow. Every figure in the table is a displayed figure from the calculator, not an estimate.

Three patterns in that table are worth noticing, because they correct common misreadings. First, most single changes leave the band at Lower, including a doubled baseline, a doubled gestational-age multiplier, and every 1.50 multiplier: the band is a grouping of illustrative results, and a single raised factor does not by itself move the grouping, which is why the band should never be read as a risk verdict. Second, the antibiotics multiplier works in the opposite direction from the others: the default broad-spectrum antibiotics given four or more hours before birth carry a multiplier of 0.80, which lowers the index, while selecting no antibiotics raises the multiplier to 1.00 and the index to 0.60, because the protective factor is removed rather than a risk factor added. Third, clinical presentation is the strongest single factor on the page: equivocal presentation doubles the index to 0.96, and clinical illness triples it to 1.44 and moves the band to Middle, which is the only single-input change in the table that changes the band. That ordering, presentation strongest, gestational age next, and the 1.50 factors grouped together, is a property of this page’s teaching multipliers, not a claim about the official model’s weights, and it should not be quoted as one.

The risk-flag count follows a simpler rule than the multipliers: it counts how many inputs sit in a raised-factor state, not how large their multipliers are. At the defaults, with every multiplier at 1.00 except the protective antibiotics multiplier, the count is zero. Changing any one raised factor, whether it multiplies by 1.50 or by 2.00 or by 3.00, raises the count to one, because the count records the number of raised factors, not their combined size. The count is therefore a checklist summary, useful for seeing at a glance how many factors are raised, and not a second risk score: two inputs with multipliers of 1.50 and 3.00 would both contribute one flag each, and their combined effect lives in the index, not in the flag count. A reader who adds the flag count to the index, or who treats a higher flag count as a separate risk, is double-counting the same inputs, and the page’s own breakdown keeps the two apart precisely to prevent that.

Worked scenarios, and how to discuss them with a clinician

The scenarios below combine the verified single-input changes, and each one was checked by running the page’s own calculator. The teaching default, baseline 0.60 with all multipliers at 1.00 except antibiotics at 0.80, gives 0.48 per 1,000 in the Lower band with zero flags, and it is the baseline against which the other scenarios should be compared. Removing the antibiotics protection alone, by selecting no antibiotics, gives 0.60 per 1,000, still Lower with zero flags, because removing a protective factor returns the index to the baseline rather than adding a risk factor. Adding a raised gestational age alone, at 34 weeks, gives 0.96 per 1,000, still Lower with one flag, because the gestational-age multiplier of 2.00 doubles the default index. Selecting clinical illness as the presentation alone gives 1.44 per 1,000 in the Middle band with one flag, because the presentation multiplier of 3.00 triples the default index, and it is the only single change that moves the band. Those four scenarios together show the page’s range from its teaching default to its most strongly raised single factor, and they show that the band changes only at the top of that range.

Discussing any of these scenarios with a clinician means discussing the inputs, not the index. The productive conversation is about the seven inputs themselves: what the local incidence figure is, what the gestational age and maternal temperature were, how long membranes were ruptured, what the Group B Streptococcus status was, what antibiotics were given and when, and how the baby is presenting on examination. Those are the facts a clinician can verify and act on, while the index is an arithmetic illustration built from teaching multipliers that the page itself says are not the official model’s coefficients. A parent or reader who brings this page to a clinician should bring the input list and the page’s own disclaimer with it, and should expect the clinician to rely on examination and local protocol rather than on the page’s number, because the page’s own note says exactly that. If a baby looks ill, the correct action does not involve this page at all: seek immediate clinical evaluation now, as the page’s own frequently asked questions and its urgent banner both state, without waiting for or relying on any number computed here.

What the official approach adds that this page cannot

The official Kaiser Permanente approach and this page share an input list, but they differ in everything that turns inputs into a decision, and that difference is the reason this page is labelled illustrative. The official tool uses a fitted statistical model, which means its weights were estimated from real outcomes in a real population, and its output was validated against those outcomes before it was used for care. This page uses fixed teaching multipliers printed on the page, which were chosen to illustrate how risk factors combine, and its output has no such validation. The distinction is not a criticism of either side: a teaching illustration is useful precisely because its arithmetic is visible and hand-checkable, while a fitted model is useful precisely because its weights come from data, and the two serve different purposes. A reader who treats the teaching multipliers as if they were fitted weights is using the wrong tool for the job, and a reader who treats the fitted model as if it were visible arithmetic is expecting a transparency the official model does not offer, since its coefficients are not published in replicable form.

The second difference is examination and laboratory results, which the official approach uses and this page cannot. The official tool’s presentation input is a clinical assessment made by a clinician who has examined the baby, and its decisions are made alongside laboratory results that this page never sees. This page’s presentation input is a selection from three options, well, equivocal or clinical illness, made by the reader, and its result is computed from that selection alone, without any examination behind it. That gap is why the page’s own guidance is so firm: only the baby’s own clinician, with examination and laboratory results, can decide about tests or antibiotics, and a reader’s selection of “clinical illness” on this page is not a clinical finding, even though it triggers the page’s urgent banner. The banner’s advice, to seek immediate clinical evaluation now, is the page’s way of closing that gap: it directs the reader toward the examination that the page cannot perform.

The third difference is timing and setting, which the official tool fixes and this page leaves open. The official tool uses the institution’s own incidence figure, which reflects the population and practices of that institution, and its antibiotic input includes both type and timing, which matter because antibiotics given closer to birth have had less time to act. This page lets the reader enter any baseline and select from four antibiotic categories, broad-spectrum four or more hours before birth, broad-spectrum two or more hours, Group B Streptococcus-specific two or more hours, or none, and it applies the same teaching multipliers regardless of setting. A reader who enters a baseline from one setting and antibiotic details from another is mixing contexts, and the resulting index illustrates a combination that may not correspond to any real population. The page’s transparency about its multipliers helps here, because a reader who prints the breakdown can see exactly which baseline and which multipliers produced the index, and can say so when discussing it, rather than presenting the index as a standalone figure.

Common misreadings follow from those three differences, and each one has a simple correction. Treating the Lower band as “safe” is corrected by the page’s own sentence: a lower band never means the baby is safe. Treating the Middle band as “has sepsis” is corrected by the companion sentence: a higher band never means the baby has sepsis. Treating the index as a probability for one baby is corrected by its units: it is an illustrative index per 1,000, built from teaching multipliers, not a validated individual risk. Treating the flag count as a second score is corrected by the breakdown: the count records how many factors are raised, and the index carries their combined size. And treating a selection of clinical illness as a diagnosis is corrected by the urgent banner itself, which states that the page gives no care decision and directs the reader to immediate clinical evaluation. A page that is read with those corrections is a useful teaching tool; a page that is read without them is a source of false reassurance or false alarm, and the difference lies entirely in the reading.

Risk as a gradient, not a label

The Kaiser EOS calculator’s central idea is that risk is a gradient, not a binary label. Two babies can both carry risk factors — prolonged membrane rupture, maternal fever — yet face very different probabilities of early-onset sepsis, and a single “treat everyone with a risk factor” rule over-treats the lower-risk baby and under-distinguishes the higher-risk one. By combining gestational age, rupture duration, maternal temperature, Group B Streptococcus status, and intrapartum antibiotics into a prior risk, then adjusting for whether the baby looks well, equivocal, or ill, the calculator gives clinicians a number per 1,000 births to match management to — routine care, observation with evaluation, or empiric antibiotics — instead of one rule for all.

A clinical tool, with clinical limits

The calculator does not diagnose sepsis, does not apply beyond the first 72 hours, and never overrides the examination: a baby who looks ill is treated as ill whatever the score says, and a deteriorating baby is reassessed rather than held to an earlier estimate. Its value — fewer unnecessary antibiotics and separations — comes from the whole system working: accurate inputs from the record, reliable observation, and rapid escalation. Parents should never use it to assess a baby at home; any concern about a newborn goes immediately to a clinician.

Frequently asked questions

More calculators