Neonatal Early-Onset Sepsis Risk Calculator
Neonatal Early-Onset Sepsis Risk Calculator
The Kaiser Permanente early-onset sepsis calculator turns gestational age, the highest maternal intrapartum temperature, the duration of rupture of membranes, maternal GBS status and intrapartum antibiotics into a risk per 1,000 births, then multiplies it by the infant’s clinical examination. This page does not reproduce that arithmetic and does not try to. Run the official Kaiser Permanente neonatal early-onset sepsis calculator, then bring the number it gives you back here: what follows is the management band that number falls in, what the examination step actually did to it, and where the published misses are concentrated.
What the EOS risk figure means
Calculator output → management bandAn infant born at 39 weeks. You have run the official Kaiser Permanente calculator with the maternal factors and your unit’s local incidence, and then entered the clinical examination; it returned an EOS risk after clinical examination of 1.6 per 1,000 births. The infant is well appearing at four hours. There was no clinical diagnosis of chorioamnionitis.
What this page does, and what it deliberately does not do
What it does do. It reads the number that tool gives you against the two management thresholds the implementation study published:
3 or more per 1,000 after the clinical examination → blood culture and empiric antibiotics.
1 to under 3 per 1,000 → blood culture and four-hourly observations for 24 hours, without routine antibiotics.
Under 1 per 1,000 → routine care.
And it adds the three things the number does not carry: whether the model applies at this gestation, whether the examination step has been done, and whether this is the situation in which the published misses cluster.
- risk per 1,000 births
- a rate in the model’s reference population, not a probability about your patient. It answers "of 1,000 infants with this combination of maternal factors and this examination, how many had culture-confirmed early-onset sepsis in the derivation cohort". A figure of 2 per 1,000 is a one-in-five-hundred rate, which is why the treatment threshold sits where it does: treating everybody above 3 per 1,000 means treating several hundred infants for each case
- the examination term
- the largest single factor on the page. Likelihood ratios of 0.41, 5.0 and 21.2 for well appearing, equivocal and clinical illness span a factor of about fifty; no maternal variable in the model comes close. The corollary is that the quality of the examination, and the fact that it is an observation over hours rather than a glance, is what the whole estimate rests on
- local incidence
- an input to the official tool that people routinely leave at its default. It sets the intercept, and the published options span 0.3 to 0.6 per 1,000 — a two-fold range that scales every output. A unit with a genuinely different incidence and a default setting is reading numbers that are systematically wrong in one direction
- 34 weeks
- the floor of the derivation cohort. Implementation and validation were at 35 weeks or more. Below 34 there is no model, and this page refuses rather than extrapolating — early-onset sepsis is commoner and more lethal in the preterm infant, so an extrapolated reassuring number would be wrong in the dangerous direction
- 0.19 to 0.31
- the pooled probability that the calculator would have delayed or missed treatment in a culture-positive case, best case to worst case, across eleven studies and 75 cases. It is not a reason to abandon the tool — the comparator misses cases too, and treats far more infants to do it — but it is a reason to keep examining the infant rather than the number
Worked example
An infant born at 39 weeks. You have run the official Kaiser Permanente calculator with the maternal factors and your unit's local incidence, and then entered the clinical examination; it returned an EOS risk after clinical examination of 1.6 per 1,000 births. The infant is well appearing at four hours. There was no clinical diagnosis of chorioamnionitis.
Gestation 39 weeks, so the model applies — it was implemented and validated at 35 weeks and above
The examination has been applied, so 1.6 per 1,000 is the posterior figure and not the risk at birth
No clinical chorioamnionitis, so the branch where the published misses cluster does not apply
1.6 is below 3, so empiric antibiotics are not recommended on the risk alone
1.6 is at or above 1, so the recommendation is a blood culture with four-hourly observations for 24 hours
What would change it: any new abnormality moves the infant out of the well appearing category, which multiplies the risk by about twelve if it becomes equivocal and about fifty if it becomes clinical illness — so the answer is re-examined, not re-calculated
The three management bands, and what each one asks of the team
| EOS risk after examination | Blood culture | Empiric antibiotics | Observation |
|---|---|---|---|
| 3 or more per 1,000 | Yes | Yes | As clinically indicated |
| 1 to under 3 per 1,000 | Yes | No | Every 4 hours for 24 hours |
| Under 1 per 1,000 | No | No | Routine newborn care |
Why this page does not print a calculated risk
| Status | |
|---|---|
| The published logistic equation | Available on the model owner’s own implementation page, with the coefficients and four incidence-specific intercepts |
| Whether it is current | No. The model was re-estimated in 2024 on 412,595 infants born 2010–2020, with 113 EOS cases; sensitivity rose from 0.76 to 0.80 and the clinical-status likelihood ratios were re-estimated too |
| The 2024 coefficients | Not retrievable. The owner’s page points to a separate tab for them, which could not be fetched |
| Published worked examples to verify against | None found. No accessible source gives both the inputs and the output of a calculator run |
| Consequence of building it anyway | A page printing the superseded model’s number would disagree with the tool the clinician is actually looking at, for a decision about giving a newborn antibiotics |
A risk per 1,000 is not a decision, and the examination is most of the arithmetic
Early-onset sepsis — invasive bacterial infection in the first 72 hours, in practice almost always group B Streptococcus or Escherichia coli — is uncommon and dangerous, which is the worst combination for a decision rule. At an incidence in the region of 0.3 to 0.6 per 1,000 live births in populations with universal antenatal GBS screening, any strategy that treats every infant with a risk factor treats hundreds for each case found. The categorical approach that preceded the calculator did roughly that: a maternal temperature, a prolonged rupture of membranes, an unknown GBS status, and the infant was cultured and started on antibiotics.
The multivariable approach asks a different question. Instead of counting risk factors it estimates a probability from all of them at once, and then — the part that matters most and is most often skimmed — multiplies that probability by what the infant actually looks like over the first hours of life. The likelihood ratios the tool applies are 0.41 for a well appearing infant, 5.0 for an equivocal one and 21.2 for clinical illness. That is a factor of about fifty from end to end, which is more than the entire range of the maternal model. In the implementation cohort of 204,485 infants across 14 hospitals, moving to this approach cut empirical antibiotics in the first 24 hours from 5.0 to 2.6 per cent and blood cultures from 14.5 to 4.9 per cent, with culture-confirmed early-onset sepsis unchanged and readmissions for it rare throughout.
Two honest limitations belong beside that. The first is the published miss rate: across eleven studies and 75 culture-positive cases, the pooled probability that the calculator would have delayed or missed treatment was 0.19 in the best case and 0.31 in the worst, with the misses concentrated among infants exposed to chorioamnionitis. The comparator misses cases too, and treats several times as many infants to do it, so this is a trade rather than a verdict — but it is a trade a unit should make knowingly. The second is that the model has moved: it was re-estimated in 2024 on a contemporary cohort of 412,595 infants, with the likelihood ratios re-estimated alongside the coefficients, and the official tool now runs the updated version.
That second point is why this page reads the calculator’s output rather than recomputing it. The superseded equation is published; the current one is not accessible here, and a page that printed a number disagreeing with the tool on the clinician’s screen would be worse than a page that prints none. What it can do instead is the part the output does not carry: whether the model applies at this gestation, whether the examination step has been done at all, and whether this is the one situation in which the published misses cluster. For the differential that presents the same way in the first hours, the neonatal hypoglycaemia threshold interpreter, the cord blood gas interpreter, the neonatal hyperammonaemia interpreter and the arterial blood gas interpreter are the pages that settle it quickly.
Frequently asked questions
Does this page calculate the Kaiser early-onset sepsis risk?
No, deliberately. The model was re-estimated in 2024 on a contemporary cohort of 412,595 infants — both the regression coefficients and the clinical-examination likelihood ratios — and the official tool now runs that updated version. The coefficients for the current model are not publicly retrievable, and no published worked example exists to verify an implementation against. A page that printed the superseded model’s figure would show a clinician a number that disagrees with the calculator on their own screen, in a decision about whether to give a newborn antibiotics. So this page reads the official tool’s output instead, against the management thresholds the implementation study published, and adds the three things the output does not carry.
What EOS risk means antibiotics, and what means just a blood culture?
In the implementation study that put the calculator into practice across 14 hospitals, empirical antibiotics were recommended at an EOS risk of 3 or more per 1,000 live births after the clinical examination, and a blood culture at 1 or more per 1,000 with observations every four hours for 24 hours. Below 1 per 1,000, routine care with no culture and no antibiotics. The same three-band structure is described elsewhere as green, amber and red. Note that all three bands are read off the figure after the examination, not the risk at birth.
Can the calculator be used in a baby born before 35 weeks?
The model was derived in infants born at 34 weeks or more, and the implementation and validation work was done at 35 weeks or more. Below 34 weeks it has no evidence base: the incidence, the organisms, the informativeness of the clinical examination and the consequences of a missed infection are all different, and extrapolating a reassuring number into that group would be wrong in the dangerous direction. Preterm infants below 34 weeks are managed on unit protocol. Between 34 and 35 weeks the model exists but the management thresholds were validated above it, which is a real if smaller caveat worth stating to the team rather than absorbing silently.
How often does the sepsis calculator miss a case?
A systematic review and meta-analysis of eleven studies covering 75 culture-positive early-onset sepsis cases found that 14 to 22 of them would have had delayed or missed treatment under the calculator compared with categorical management, giving a pooled probability of missing a case of 0.19 in the best case and 0.31 in the worst. The misses were concentrated among infants exposed to chorioamnionitis. The comparison is not one-sided: the categorical approach treats several times as many infants and misses cases of its own. But the concentration of misses in chorioamnionitis-exposed infants is specific enough to act on, and this page carries a dedicated branch for it.
Why does the clinical examination change the number so much?
Because it is the strongest predictor in the system. The tool applies a likelihood ratio of 0.41 to a well appearing infant, 5.0 to an equivocal one and 21.2 to clinical illness — a span of about fifty — whereas the entire maternal model moves the prior far less. The practical consequence is that the estimate is only as good as the examination behind it, and the calculator’s categories are defined by persistence: equivocal means a single physiological abnormality lasting over four hours, or two lasting over two hours. That is a period of structured observation by somebody, not a glance at delivery, and recording an unobserved infant as well appearing is the commonest way the output is made falsely reassuring.
Does a low EOS risk mean the baby can be discharged early?
No. The figure is a probability at a moment, conditioned on an examination that has already happened. It says nothing about the next twelve hours, and any new abnormality of feeding, temperature, respiratory effort, tone or colour moves the infant out of the well appearing category and multiplies the risk by about twelve or about fifty depending on how far. Discharge timing follows the unit’s normal practice and the infant’s feeding and observations, not the calculator. The right response to a change is to re-examine the infant, not to re-run the model.
Related calculators
References
- Kuzniewicz MW, Puopolo KM, Fischer A, Walsh EM, Li S, Newman TB, Kipnis P, Escobar GJ. A quantitative, risk-based approach to the management of neonatal early-onset sepsis. JAMA Pediatr. 2017;171(4):365–371. doi:10.1001/jamapediatrics.2016.4678. The source of the two management thresholds used on this page: blood culture at an EOS risk of 1 or more per 1,000 live births, empirical antibiotics at 3 or more. 204,485 infants born at 35 weeks or more across 14 hospitals, 2010–2015; empirical antibiotics in the first 24 hours fell from 5.0% to 2.6% and blood culture use from 14.5% to 4.9%, with culture-confirmed early-onset sepsis unchanged at 0.03%, 0.03% and 0.02% across the three periods.
- Kuzniewicz MW, Puopolo KM, Walsh EM, et al. Update to the neonatal early-onset sepsis calculator utilizing a contemporary cohort. Pediatrics. 2024;154(4):e2023065267. doi:10.1542/peds.2023-065267. 412,595 infants born at 35 weeks or more at 14 hospitals between January 2010 and December 2020, with 113 EOS cases. Both the model coefficients and the likelihood ratios for clinical status were re-estimated; sensitivity 0.76 for the original model against 0.80 for the updated one, and recommended empirical antibiotic use 3.5% against 3.7%. This is the paper that makes reproducing the older published equation unsafe.
- Pettinger KJ, Mayers K, McKechnie L, Phillips B. Sensitivity of the Kaiser Permanente early-onset sepsis calculator: a systematic review and meta-analysis. EClinicalMedicine. 2020;19:100227. doi:10.1016/j.eclinm.2019.11.020. Eleven studies, 75 culture-positive EOS cases; 14 to 22 would have had delayed or missed treatment under the calculator compared with categorical management. Pooled probability of missing a case 0.19 (best case) to 0.31 (worst case), with misses concentrated among infants exposed to chorioamnionitis.
- Kaiser Permanente Division of Research. Neonatal early-onset sepsis calculator and EMR implementation FAQ. neonatalsepsiscalculator.kaiserpermanente.org. The source of the clinical examination likelihood ratios quoted on this page — 21.2 for clinical illness, 5.0 for equivocal and 0.41 for well appearing — and of the published logistic equation and its four incidence-specific intercepts, which this page deliberately does not implement.
- Puopolo KM, Benitz WE, Zaoutis TE; American Academy of Pediatrics Committee on Fetus and Newborn, Committee on Infectious Diseases. Management of neonates born at ≥35 0/7 weeks’ gestation with suspected or proven early-onset bacterial sepsis. Pediatrics. 2018;142(6):e20182894. doi:10.1542/peds.2018-2894. The clinical report that sets multivariate risk assessment alongside categorical risk assessment and enhanced clinical observation as one of three acceptable approaches, and that fixes the 35-week floor for the implementation evidence.
Medical Disclaimer: The tools and content provided here are for educational and reference purposes only. They are not intended to substitute for professional medical advice, diagnosis, or treatment. Clinical decisions should always be based on the comprehensive assessment of a qualified healthcare professional.