Unlocking the Secrets of Sepsis: AI Analysis Reveals Key Predictors

By applying AI to electronic health records of sepsis patients, researchers have unlocked extraordinary new insights into complex clinical factors linking groups of patients.

Sepsis occurs when the immune system orchestrates a massively heightened response to infection.

It causes system wide inflammation with a high risk of multiple organ failure an even death.

Sepsis is the leading cause of death worldwide, accounting for 1.7 million hospitalizations and over 350,000 deaths annually in the USA.

Despite its prevalence, and crippling impact on healthcare systems worldwide, sepsis diagnosis remains a major challenge with no gold standard.

The main hurdle in identifying sepsis is its capacity to mimic symptoms associated with numerous other illnesses.

This research, published in the Journal Artificial Intelligence in Medicine, represents a major step in understanding the complex relationship between pre-existing conditions (comorbidities) and sepsis.

After crunching electronic clinical records of 8,651 patients, the autoencoder algorithm used by the Stockholm University (Sweden) researchers  grouped cases into ‘clusters’ of patients with related characteristics.

The analysis found age and gender as critical factors influencing sepsis progression and outcomes.

Moreover, cardiovascular conditions, liver dysfunction, and renal failure were the top three comorbidities contributing to patient clusters.

These are the headline findings, but the research uncovered far more granular relationships between biological indicators and sepsis. For example, one cluster comprised entirely of adult males with 93.5% ventilation reliance, and high prevalence of multiple comorbidities: bleeding disorder 99%, liver disease 53%, and neurological conditions 31%.

Figure showing the patient clusters according to gender and age (left) and comorbidities (right). Credit: Bampa, M et al.

So, what does it mean and why is it important? Well, being able to categorize patients in this manner has the potential to significantly speed up the triage process. The quicker clinicians can understand the factors underlying someone’s condition, the quicker they can apply lifesaving treatment.

The identification of high-risk groups, such as elderly patients with multiple comorbidities, can also enable healthcare providers to develop targeted interventions and personalized treatment approaches.

Unfortunately, this approach is a long way from use in clinics. Interoperability (working in different system environments and health records) and generalisability (how applicable it is to real world data) represent major barriers to implementation. The authors state that future research should focus on addressing these factors.

Regardless, the approach will empower improved understanding of sepsis and lay a foundation for transformative work to come.