Intellicens Neo

Predicting late-onset sepsis in the NICU

11 hours

median lead time before late-onset sepsis is clinically detected

Retrospective analysis · Meeus et al. 2024 · The Journal of Pediatrics

Intellicens Neo is under clinical investigation and is not yet available for clinical use. It has not been CE marked under EU MDR 2017/745 and has not been cleared by the FDA.

↓

Why earlier detection matters

Physiological changes occur before they become clinically visible

Late-onset sepsis is a bloodstream infection that hits a preterm infant against an immune system that is not ready for it. In the NICU it is common, and it can turn life-threatening within hours.

Treated early, it responds well to antibiotics. The challenge is the diagnosis: the first signals are subtle, spread across several parameters at once, and hard to separate from normal fluctuation.

1 in 5

Preterm infants develops sepsis during the NICU admission [Stoll et al. 2002]

24h

Deterioration patterns are present in the data up to 24h before diagnosis [Griffin & Moorman 2001]

177/day

Monitor alarms per patient in the NICU today. Only 3.6% critical [Li et al. 2018]

↓

The difference, in one picture

The same admission, twice.

One patient, the last 40 hours before sepsis was clinically detected. Above, what the bedside monitor made visible. Below, what the Neo index made visible on exactly the same data.

Without Intellicens Neo

10-second snapshots

Data stuck in monitors

Temperature36.3 to 37.8 °CHeart rate100 to 180 bpmSpO2alarm below 88 %Glucose2.6 to 8.0 mmol/L12345Hover a numbered moment to see both views at once: what the bedside showed, and what Neo showed.

With Intellicens Neo

Long-term pattern evolution

Remotely accessible

Interactive visualizations

Neo indexone value every 30 min1.000.00watch 0.15alert 0.3040 h before30 h20 h10 hclinical detection12345Alert raised11 hours earlier

The result?

11 hours

between the index crossing its threshold and the clinical picture arriving. Median, retrospective.

Curves are schematic. Lead time and alarm counts from retrospective validation cohorts, Meeus et al. 2024. Intellicens Neo is under clinical investigation and does not replace bedside assessment.

↓

From monitor signals to a risk score

Built from the signals your unit already records.

No new sensors. No new hardware. Neo reads what the monitors already stream, and it runs entirely inside the hospital.

01

3-lead ECG sensor

Heart rate

HR variability

02

Pulse-oximeter sensor

SpO₂

03

Clinical observations

EMR

3 routine obs

04

Patient demographics

Gestational age

Birth weight

The output

One risk score every 30 minutes, built from six signals and 28 features.

Every score comes with its reasoning: which parameters are pushing the index up, and by how much.

6

signals per score

28

features per score

30

min

update interval

↓

The Neo index, explained

One index. Every 30 minutes. Always explainable.

A patient’s clinical picture over time, replayed. Move across the timeline; the readout shows the index and the parameters driving it at that moment.

Intellicens Neo Index · 42-hour window · one score every 30 min
0.000.250.500.75−36h−30h−24h−18h−12h−6hT 0+6hblood culture orderedNeo alertsCulture ordered
Neo IndexWatch thresholdAlert thresholdBlood culture ordered
T −12.5 h
0.32
Alert
Neo Index · 0.00–1.00 · watch 0.15 · alert 0.30
What’s driving the score
lowers riskraises risk
Heart rate
SPO2
Glucose
Postnatal age
Temperature

Illustrative purposes only. Not an actual representation of the model's parameters, weighting or output.

Illustrative replay. Anonymized retrospective data. Not a real patient case.

↓

Retrospective and external validation

Validated the way it runs in the NICU

Every model alert is evaluated. No data removed. No windows excluded around events. The precision you'd truly experience at the bedside.

11 hours

Median lead time before late-onset sepsis is clinically detected

<

1

False alert per patient per week (Meeus et al. 2024)

5.2M

Individual predictions scored and evaluated retrospectively

(01)

EU

Finalized · Published, Journal of Pediatrics

Development study · UZA, Antwerp

Single-center retrospective cohort · 880 infants, GA <32 weeks · 2.49M predictions, every one evaluated.

81%

Sensitivity
severe sepsis

10.8 hrs

Time gain
median

<1/wk

Alarm rate

(02)

EU

Finalized · Under peer review, 2026

External validation · Erasmus MC & Amsterdam UMC

Multi-center retrospective cohort · 1,165 infants, GA <32 weeks · 2.74M predictions across two independent NICUs.

82%

Sensitivity
severe sepsis

12.1 hrs

Time gain
median

<1/wk

Alarm rate

(03)

EU

Ongoing · completion 2027

Prospective pre- & post-implementation study

Multi-center, two-phase design: blinded vs. standard of care, then post-implementation. Primary endpoints: time-to-antibiotic, clinical safety and efficacy.

±700 VLBW

Enrollment

50 episodes

Powered for

2027

Completion

(04)

US

Planned · Q3 2026

Prospective blinded study · 2 US sites

Prospective blinded clinical study at 2 US neonatal sites, building the US evidence base. Expected completion early 2027.

2 US sites

Setting

Blinded

Design

Early 2027

Completion

↓

Deployment inside your hospital network

Patient data stays within your hospital network.

Neo runs on a virtual machine inside your hospital, within your own security perimeter, and opens in the browser your team already uses. It reads your monitors over HL7 and your EMR over FHIR.

On-premise by design

Hosted on a virtual machine inside your hospital. No patient data or intellectual property ever leaves your network.

Easy installation

We handle the installation, configure it with your team and train your staff.

Runs in the browser

On a computer in your unit or at the bedside. No new hardware.

Data mapping handled

Sensor-agnostic. We connect to your monitors, your interoperability layer and your EMR.

↓Questions we get asked

What clinical, IT and legal teams ask about Neo

It turns the signals your NICU already records into one explainable risk score for late-onset sepsis. The index is recalculated every 30 minutes and always shows which parameters are pushing it up or down, so a nurse or neonatologist can see why the score moved and not only that it moved. In the retrospective validation it flagged late-onset sepsis a median of 11 hours before it was clinically suspected.

No. Intellicens Neo reads the same signals as the bedside monitor but changes nothing about it: the monitor keeps working exactly as it does today, and Intellicens Neo adds no bedside alarms. It is a second read on the same physiology that brings a suspicion forward in time, not a diagnosis. The clinical decision stays with the clinician.

Intellicens Neo is under clinical investigation and is not yet available for clinical use. It is not CE marked under EU MDR 2017/745 and has not been cleared by the U.S. Food and Drug Administration. It is developed under an ISO 13485 quality management system and follows the Class IIb route under EU MDR 2017/745 with BSI as notified body, and in the United States the 510(k) pathway is defined with a pre-submission filed. Installations today run as research or evaluation deployments under the agreement with the hospital.

It runs on continuous vital sign data your monitors already produce, together with context from the electronic medical record, over the HL7 and FHIR interfaces your hospital already operates. It requires no new hardware at the bedside. The platform is installed on-premise, so patient data does not leave your perimeter. Access runs through your own identity provider.

Retrospectively first, in the NICU where it was developed, published as Meeus et al. 2024. It was then validated externally in an independent multicenter cohort of more than 1,100 infants at Erasmus MC and Amsterdam UMC, which reproduced the lead time (publication under peer review). A prospective pre- and post-implementation study on times-to-antibiotics is the next step, because lead time on a curve and changed behavior at the bedside are two different claims.

It is built not to. Intellicens Neo raises no audible alarm at the bedside. The risk score and the parameters driving it are shown in a dashboard, next to the rest of the patient's data, and in the retrospective validation the score produced fewer than one alert per patient per week (Meeus et al. 2024).

It takes an on-premise installation of the Intellicens Data Platform, connectors to your monitors and your EMR, a data governance agreement with your data protection officer, and a short training round with the clinical team. Nothing is added at the bedside. Because the platform is installed once, adding further applications later does not repeat that work.