Why Doctors Still Guess About Newborn Infections — and How a New Scoring Chart Tries to Help

A look at research from a children's hospital in Suzhou, China, that combined ordinary bedside observations and two routine blood tests into a single score for newborn blood infection

Doctors caring for newborns face a problem that has never had a clean solution: the early signs of a serious blood infection look almost exactly like the ordinary, harmless unsettledness of a baby in the first days of life. A team at a children's hospital in China has now built two scoring charts that combine simple bedside observations — and, in one version, two everyday blood tests — into a single estimate of how likely infection is. The charts performed well on the hospital's own records, but the researchers are clear that they are not ready for use with real babies until they have been tested elsewhere.

The Problem Behind the Research

Infection of the bloodstream in a newborn, called neonatal sepsis, is one of the most serious things that can happen in the first month of life. Worldwide it is estimated to affect somewhere between about 2,800 and 3,900 babies out of every 100,000 born, and together with premature birth it accounts for close to half of all deaths in children under five [1]. The burden is heaviest in countries with the fewest resources, where most newborn deaths occur [2].

What makes it so difficult is that a newborn cannot describe how she feels, and the signs of infection are quiet and general. A slightly unstable temperature. Faster breathing. Refusing feeds. Being floppy or less responsive than yesterday. Every one of those things also happens in perfectly well babies adjusting to life outside the womb [3]. Doctors and nurses have been trained for generations to recognise these patterns, and every review of the subject concedes the same thing: recognition based on appearance alone is unreliable [4].

For decades the intended answer was the blood culture — a sample of the baby's blood placed in a bottle to see whether bacteria grow. It remains the reference test, but it does not solve the problem at the moment the decision has to be made. Growing bacteria takes 24 to 48 hours, and the decision about antibiotics has to be made in the next hour or two. Worse, the test misses a great many real infections: often fewer than half are caught, because only a very small amount of blood can safely be taken from a newborn, because antibiotics given to the mother during labour can suppress the bacteria, and because some organisms are simply hard to grow [5].

The result is a familiar and uncomfortable compromise. Because missing an infection can be fatal within hours, and starting antibiotics is comparatively low-risk, hospitals start antibiotics on suspicion in far more babies than turn out to be infected. That is not carelessness — it is a reasonable response to an unfair asymmetry. But it means many well babies receive drugs they do not need, along with drips, blood tests and time separated from their parents. It also contributes to antibiotic resistance and disturbs the developing community of helpful bacteria in a baby's gut.

Attempts to solve this with a single blood test have repeatedly fallen short. Markers of inflammation such as C-reactive protein (a protein that rises when the body is fighting infection, usually shortened to CRP) and the white blood cell count are useful but not decisive; no single one of them reliably confirms or rules out infection, and the numbers that count as "abnormal" shift depending on how premature the baby is and when the sample was taken [6]. The genuine progress of the last decade came from a different idea: instead of asking one question, combine many small clues into one calculated estimate. A tool built on that principle, the early-onset sepsis calculator, has been shown across many hospitals to cut unnecessary antibiotic use without babies coming to harm [7]. But it was designed only for the first three days of life in babies born at or near full term. For everyone else — the baby who becomes unwell on day six, the premature baby who deteriorates in week three — nothing comparable existed.

What the Researchers Did

The team, based at the Children's Hospital of Soochow University in Suzhou, China, looked back through hospital records from October 2019 to October 2025 [8]. This is called a retrospective study: no baby was given a new treatment, and nothing was done differently at the time. The researchers simply examined what had already been recorded.

They identified 483 babies who had been diagnosed with sepsis, and compared them with 200 babies admitted during the same period who did not have sepsis. That comparison group was deliberately mixed: 97 babies had a different kind of infection (a skin infection, a urine infection, or pneumonia), and 103 had no infection at all and were in hospital for something like jaundice or feeding difficulty. Mixing the comparison group this way was intentional — it reflects the real range of babies a doctor has to tell apart.

They then built two scoring charts, called nomograms. A nomogram is a printed chart where each finding earns a number of points; you add up the points and read off an estimated probability at the bottom. The first chart, Model A, uses nine pieces of information that need no laboratory at all: where the mother lives, whether she had bleeding threatening miscarriage during pregnancy, whether the baby is breastfed, whether the mother took antibiotics before delivery, the baby's temperature, whether the baby seems dull or unresponsive, whether the baby's newborn reflexes are strong, whether antibiotics had already been given before arriving, and how the baby's current weight compares with birth weight. The second chart, Model B, adds two ordinary blood tests: the white blood cell count and CRP.

To check the charts fairly, the researchers set aside 30% of the records, built the charts using only the other 70%, and then tested them on the set-aside records they had not used.

What They Found

Both charts worked better than chance by a wide margin [8]. Researchers measure this with a number between 0.5 (useless) and 1.0 (perfect). Model A, using no blood tests, scored 0.839 and 0.852 on the two sets of records. Model B, with the two blood tests added, scored 0.905 and 0.923.

The practical difference between them is worth understanding in concrete terms. Model A, at the cut-off the researchers chose, correctly identified about 65 to 71 out of every 100 babies who really had sepsis, while correctly clearing about 87 to 91 out of every 100 who did not. In other words, it rarely raised a false alarm, but it missed roughly three infected babies in ten. Model B, with the blood tests, found about 86 to 91 out of every 100 infected babies — catching most of the ones Model A missed — but at the cost of wrongly flagging about one in four healthy babies.

The researchers were careful to report the results that complicate their own story. When they narrowed the comparison to the harder, more realistic question — telling sepsis apart from other infections, rather than from healthy babies — both charts got noticeably worse. And Model B, despite being better at ranking babies from lower to higher risk, produced probability figures that did not match reality closely enough in one of the two sets of records.

What This Means for Families

If your baby is in a neonatal unit, this research does not change anything about her care today, and the researchers say so plainly. These charts have been tested only on records from the one hospital where they were built. Whether they work in a different country, with different patients and different practices, is unknown.

There is one point that matters more than the charts themselves, and it is worth knowing. Because of the way the study selected which records to include, the percentages the charts print out are estimates relative to this particular group of babies, not a true probability for any individual child [8]. A number from one of these charts should never be quoted to a parent as "your baby has an 80% chance of infection." It does not mean that.

What the study does support is something more reassuring and more general: that the things nurses and doctors watch for — how alert your baby is, how she feeds, how her reflexes respond, how her weight is tracking against her birth weight — genuinely carry information, and carry more of it when weighed together than when considered one at a time. Those bedside observations are not filler. In this study they did most of the work.

And if antibiotics were started for your baby before anyone was certain, that is not a mistake or an oversight. It is the deliberate choice hospitals make because waiting two days for a culture result is more dangerous than a short course of antibiotics that turns out to have been unnecessary. Teams routinely stop those antibiotics once the picture becomes clear.

What Researchers Are Working On Next

The next steps are well defined. The charts need to be tested at other hospitals, in other countries, on babies the researchers have never seen — a process called external validation. They need to be adjusted so the printed percentages reflect how common sepsis actually is in each place. And then, if they survive that, they need to be tried out in real time to see whether using them actually leads to fewer unnecessary antibiotic courses and faster treatment for the babies who need it — the same long road the early-onset sepsis calculator travelled before it changed practice [7], [9].

A related effort uses machine learning to attempt the same task, and it faces the same unfinished journey: many promising models, very few tested in real hospitals with real babies [10]. The wider picture is that newborn infection remains a substantial global problem, both worldwide [11] and in China specifically, where the number of affected babies remains large even as death rates have fallen [12]. Better tools for spotting it earlier and treating fewer well babies unnecessarily would matter enormously. This study is a careful, honest step along that road — and its authors are refreshingly clear that it is a step, not an arrival.

References

  1. Fleischmann C, Reichert F, Cassini A, et al. Global incidence and mortality of neonatal sepsis: a systematic review and meta-analysis. Arch Dis Child. 2021;106(8):745–752. doi:10.1136/archdischild-2020-320217 ↩
  2. Milton R, Gillespie D, Dyer C, et al. Neonatal sepsis and mortality in low-income and middle-income countries from a facility-based birth cohort: an international multisite prospective observational study. Lancet Glob Health. 2022;10(5):e661–e672. doi:10.1016/S2214-109X(22)00043-200043-2) ↩
  3. Strunk T, Molloy E, Mishra A, Bhutta Z. Neonatal bacterial sepsis. Lancet. 2024;404(10449):277–293. doi:10.1016/S0140-6736(24)00495-100495-1) ↩
  4. Crocker L, White A, Heaton P, Horta D, Paul S. Recognition and management of neonatal sepsis. Br J Nurs. 2021;30(7):410–415. doi:10.12968/bjon.2021.30.7.410 ↩
  5. Puopolo K, Mukhopadhay S, Frymoyer A, Benitz W. The term newborn: early-onset sepsis. Clin Perinatol. 2021;48(3):471–484. doi:10.1016/j.clp.2021.05.003 ↩
  6. Bethou A, Bhat B. Neonatal sepsis—newer insights. Indian J Pediatr. 2022;89(3):267–273. doi:10.1007/s12098-021-03852-z ↩
  7. Achten NB, Klingenberg C, Benitz WE, et al. Association of use of the neonatal early-onset sepsis calculator with reduction in antibiotic therapy and safety: a systematic review and meta-analysis. JAMA Pediatr. 2019;173(11):1032–1040. doi:10.1001/jamapediatrics.2019.2825 ↩
  8. Yu Y, Wang H, Huang F, Ding X. Construction and validation of neonatal sepsis diagnostic nomograms: a single-site retrospective study. Front Pediatr. 2026;14:1841983. doi:10.3389/fped.2026.1841983 ↩
  9. Collins GS, Moons KGM, Dhiman P, et al. TRIPOD+AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods. BMJ. 2024;385:e078378. doi:10.1136/bmj-2023-078378 ↩
  10. O'Sullivan C, Tsai D, Wu I, et al. Machine learning applications on neonatal sepsis treatment: a scoping review. BMC Infect Dis. 2023;23(1):441. doi:10.1186/s12879-023-08409-3 ↩
  11. Li J, Xiang L, Chen X, et al. Global, regional, and national burden of neonatal sepsis and other neonatal infections, 1990–2019: findings from the Global Burden of Disease Study 2019. Eur J Pediatr. 2023;182(5):2335–2343. doi:10.1007/s00431-023-04911-7 ↩
  12. Wu Y, Xia F, Chen M, et al. Disease burden and attributable risk factors of neonatal disorders and their specific causes in China from 1990 to 2019 and its prediction to 2024. BMC Public Health. 2023;23(1):122. doi:10.1186/s12889-023-15050-x ↩