AI algorithms could make it easier to predict them – breaking latest news

Researchers at the Neuroscience Center of the Children’s Hospital of Philadelphia (CHOP) have developed a prediction model, through a
machine learning algorithm
which can determine which infants may undergo
convulsions
while they are hospitalized in the Neonatal Intensive Care Unit (TIN).

According to the authors, this model could be incorporated into routine care to help the clinical team decide which children will need a
electroencephalogram (EEG)
and which babies can be safely managed in the NICU without EEG monitoring. This would allow families and caregivers to take care of children without invasive and unnecessary procedures. The results of the retrospective cohort study were published in Lancet Digital Health.

Hypoxic ischemic encephalopathy

The seizures remain an important risk factor for
morbidity
and mortality in infants with a critical illness. In particular, infants with
hypoxic ischemic encephalopathy (Hie) have a high incidence of seizures (approximately 30%), which have been associated with an increased risk of subsequent neurobehavioral problems and
epilepsy
. Most of these seizures can only be detected through EEG monitoring and not simply through clinical observation.

Detecting and treating seizures is important to reducing injury seizure-induced, thereby improving outcomes for infants with early seizures. The gold standard for their detection is the continuous electroencephalogram (CEeg). In infants with hypoxic ischemic encephalopathy, the guidelines recommend that they undergo this diagnostic test during periods of therapeutic hypothermia and return to normothermia, that is, even for 4-5 days.

Limited resources

However, this approach is not always feasible, as many of these children (even in Western countries) receive ICU care who do not have access to continuous EEG. Even intensive care units in large healthcare networks often have only limited EEG resourcesespecially since the interpretation of the EEG tracing takes a lot of time for the entire service team, including doctors and specialized technicians. Predicting which newborns will experience seizures is complex and previous attempts to predict future seizures using clinical and EEG data have not produced highly accurate results.

Collected data from over a thousand newborns

To help address these issues, the Philadelphia researchers used data from a recently developed model that they use to build prediction models using machine learning methods. “In this study, we used EEG data from over 1,000 newborns to build models to predict neonatal crises,” explains Jillian McKee, first author of the study. “These data helped us understand which newborns should receive EEG monitoring in the NICU.”

The results

Researchers have built their own crisis prediction models based on standardized EEG characteristics reported in electronic medical records. The retrospective study found that these models could predict seizures, and specifically seizures in infants with Hie, with a more than 90% accuracy. According to the authors, this is the first study to report a seizure prediction model based on standardized clinically derived reports. The US team made it the publicly available model online.

The merits and limits of the study

«The study of the Children’s Hospital of Philadelphia is interesting for the use of information technology techniques which are and will be used more and more also in healthcare, assisting and facilitating the decision of clinicians at the patient’s bed”, comments the Professor Daniele DeLuca, professor of Neonatology – Paris Saclay University and president of the European Society for Pediatric and Neonatal Intensive Care (Espnic) -. It is a topical issue and we already have other examples of the use of these technologies at the service of clinical practice”.

« However, there are two very important points on which to raise our guard. In the first place it is something that must assist and facilitate the clinical decision and not replace the work of the clinicians. Although the work starts from the real impossibility of having continuous EEG monitoring in intensive care (and this is frequently the case everywhere due to a lack of child neuropsychiatrists and technical personnel), one must not run the risk of reducing the competence of the clinician (and in this case of the intensivist) with the simple use of IT tools that predict, albeit with high accuracy, convulsions (or other complications). The price would be an increasingly poor competence of clinicians to the detriment of global patient management”.

«The authors then completely forget the other solution of the problem which is that of point-of-care (Poc). By Poc we mean any diagnostic or rapid monitoring tool, at the patient’s bed, which does not allow refined diagnoses, but the clinical management of emergencies. Poc technologies are used every day in intensive care of both adults and newborns, such as ultrasound, monitoring of vital signs, inflammatory indices or hemostasis.
Poc technology to overcome the lack of human resources for continuous EEG monitoring has existed for some time and is called Amplitude-integrated-Eeg (aEeg).

The aEeg allows the neonatologist to detect seizures, estimate the level of sedation and decide whether an infant after birth pain needs hypothermia treatment. All quickly, accurately and simply because placement and interpretation of an aEeg is much easier than a traditional Eeg and require only 2 or 5 electrodes and a short introduction. This solution is commonly adopted in neonatal intensive care units in Italy and Europe (see the document of the Italian network for the study of seizures “INNESCO” Clin Neurophysiol.) and is visibly better because there allows you to have an electroencephalographic evaluation while we actually observe the clinical event (data integration), while the Children’s Hospital proposal is limited to predicting a future event without observing it. The Poc is not considered at all by the authors because it is completely extraneous to the American context, d

ove is culturally difficult to implement as it sees the intensivist as having more and more varied skills and responsibilities».