Researchers from the Centro de Investigación Biomédica en Red (CIBER), a Spanish public research consortium including CIBER Bioengineering, Biomaterials and Nanomedicine (CIBER-BBN) and CIBER Infectious Diseases networks at the University Clinical Hospital of Valladolid, Spain, have applied an explainable AI (XAI) approach to identify and prioritize genetic factors linked to a higher risk for sepsis after surgery.

The findings were published in Frontiers in Medicine in collaboration with the University of Leicester, Leicester, England, and other participating institutions.

Speaking with Medscape’s Spanish edition, Eduardo Tamayo Gómez, MD, professor and researcher from Hospital Clínico Universitario de Valladolid in Valladolid, Spain, and a study author, said that “for the first time, a genome-wide association study has been combined with an XAI approach to identify and prioritize susceptibility genes for sepsis in surgical patients.” The most important finding, he noted, “is the prioritization of three potential biomarkers with the highest contribution to predicting postoperative sepsis” — rs17653532, rs1575081785, and rs74707084 — “located in the PRIM2, RBSN, and SYNPR genes, which are involved in integrative processes including the regulation of gene expression, DNA replication, and cell proliferation.”

To highlight the significance of these findings, it is worth noting that sepsis is a serious complication caused by an uncontrolled immune response to infection, usually bacterial infection. It is the most severe form of infection, with a mortality rate ranging from 10% to 20%, and it can reach up to 40% in cases of septic shock. Sepsis causes approximately 11 million deaths each year worldwide, of which approximately 17,000 occur in Spain.

“Anticipating sepsis can make all the difference in a patient’s prognosis,” the researchers stated in an institutional press release.

Sepsis Predictors

Although prior genome-wide association studies have demonstrated their potential for identifying genetic variants linked to sepsis, they have often included heterogeneous patient populations and relied on single-locus analytical methods. The present study aimed to identify new susceptibility loci for sepsis in postoperative patients using an AI approach applied to genome-wide data.

Demographic data for individuals with sepsis included in the study showed that 17.6% of the surgical cohort (750 individuals) had a median age of 72 years, and close to two thirds were male. Among those with sepsis, 83.9% had septic shock, and the 90-day mortality rate was 42.7% (n = 320). The mean Sequential Organ Failure Assessment and Acute Physiology and Chronic Health Evaluation II scores were 9 and 18, respectively.

Additionally, 74.8% of individuals with sepsis had one or more associated comorbidities, the most frequent of which were cardiovascular disease, chronic respiratory disease, arterial hypertension, chronic renal or hepatic failure, and diabetes. Peritonitis (30.4%), pneumonia (24.7%), catheter-related infections (8.3%), and surgical wound infections (2.7%) were the principal sources of infection.

The sepsis prediction model used in this study was based exclusively on genetic information. Speaking with Medscape’s Spanish edition, Fernando Vaquerizo-Villar, PhD, and Roberto Hornero Sánchez, PhD, researchers from the CIBER-BBN, Universidad de Valladolid, Valladolid, Spain, and co-authors of the study, said, “Specifically, on single-nucleotide polymorphisms (SNPs).” “We applied a novel methodology based on AI to predict sepsis from SNPs and to prioritize new genetic variants associated with susceptibility to postoperative sepsis.”

Preoperative Genotyping

The application of XAI to predict the probability of sepsis following surgery enabled the team to prioritize three genetic variants that could be incorporated into the preoperative assessment phase of the study.

Thus, “the probability threshold of the sepsis prediction models,” Vaquerizo and Hornero explained, “allows clinicians to identify people at higher or lower risk for sepsis in this study population, prioritizing either sensitivity — to capture more high-risk patients — or specificity — to reduce false positives — depending on the intended clinical application and available resources.”

“Regarding the practical availability of this analysis, many hospitals in Spain already have genetics or clinical genomics units that offer targeted genotyping of specific variants from blood samples using well-established techniques, with relatively short turnaround times. The cost of genotyping a small number of genetic variants is very low compared with the considerable economic impact associated with the late detection of postoperative sepsis, which frequently results in prolonged stays in the ICU, greater use of hospital resources, and a substantial increase in morbidity and mortality,” the experts emphasized.

Although the systematic incorporation of this approach into the preoperative workflow would require clinical validation of the model and adaptation of care pathways and information systems, “from a technical and organizational standpoint, it represents a feasible strategy,” they said.

Detailed Findings

The study results showed that sepsis prediction performance remained high when using the top 20 SNPs (area under the curve [AUC], 0.951) or the top three SNPs (AUC, 0.886), the latter being rs17653532, rs1575081785, and rs74707084, all of which made a high contribution to sepsis detection.

The functional effects of the top 20 SNPs identified for sepsis prediction using the AI model were analyzed in silico. Among these variants, 11 were located within genes: two were missense variants, eight were intronic, and one was in a 3′ untranslated region.

The intronic variant rs17653532, which had the highest SHapley Additive exPlanations contribution score for accurate sepsis prediction, was located in the gene encoding DNA primase subunit 2 (PRIM2). The second variant with the highest contribution was the nonsense variant (rs1575081785) within the gene encoding the Rabenosyn RAB effector protein (RBSN). The third variant (rs74707084) is an intronic SNP located within the gene encoding synaptoporin (SYNPR).

Among the top 20 SNPs, rs79219127, intronic to the FAM155A gene, showed statistically significant associations with the duration of hospitalization and ICU stay in patients with sepsis, whereas rs79275514, which is intronic to the gene encoding the Parkin protein, was significantly associated with arterial hypertension and chronic hepatic failure as comorbidities. The majority of the top 20 SNPs also showed evidence of biological and regulatory effects across multiple elements related to chromatin state, changes in regulatory motifs, DNase I sensitivity, and expression quantitative trait loci in different cell lines and tissues.

Precision Surgery

Tamayo noted that using the risk genes identified in each person, in combination with their clinical profile, “could in the future facilitate the individualization of preventive or therapeutic strategies, guiding decision-making toward interventions directed at specific biological pathways involved in sepsis. This information would also enable stratification of individuals according to individual risk for sepsis, so that those at higher risk could benefit from closer monitoring and specific preventive measures, while available resources could be optimized for those at lower risk.”

Overall, this integrative approach would support more precise and personalized perioperative medicine “based on both genetic risk and the clinical characteristics of the individual,” the team noted.

Future Directions

For future clinical applications, additional research would be needed, “including in vitro and in vivo functional studies to understand the exact role of these genes and how they influence the development of sepsis, as well as analyses in large cohorts of postoperative individuals with and without sepsis that are geographically and ancestrally diverse,” researchers concluded. Such studies would allow for a more precise assessment of the baseline genetic and clinical factors that contribute to the predisposition to sepsis and would provide robust external validation of the current findings.

Furthermore, “it would be of interest to carry out cost-effectiveness studies that allow for a precise quantification of the economic impact and potential savings in healthcare cost associated with the implementation of early detection and risk stratification strategies for sepsis in the perioperative setting,” they added.

Tamayo, Vaquerizo, and Hornero reported having no relevant conflicts of interest.

This story was translated from Medscape’s Spanish edition.