Surgeons to get a high-tech assistant: Smartphone app for assessing liver surgery risks

By Neha MathurReviewed by Danielle Ellis, B.Sc.Nov 6 2023

In a recent article published in the Annals of Surgery Journal, researchers developed a multivariable model (MVM) to predict clinically relevant posthepatectomy liver failure grade B and C (PHLF B+C), grade C, and 90-day mortality prediction. 

Background

PHLF is the most common cause of 90-day mortality after liver resection. For instance, chemotherapy-associated liver injury (CALI) in colorectal cancer liver metastases (CRCLM) patients increases PHLF risk. Moreover, there is no postoperative (postOP) treatment for PHLF. Thus, tools for personalized risk assessment of preoperative (preOP) risk of PHLF in patients undergoing hepatic resection are critically needed.

Currently, clinicians use scores of different liver function tests, such as indocyanine green (ICG-R15/PDR) clearance, albumin-ICG evaluation (ALICE) grade, and fibrosis-4 (FIB-4) index to predict PHLF, but they have not been directly compared.

About the study

Previously, researchers compared the predictive potential of the combined aspartate aminotransferase/platelet ratio index (APRI) and albumin-bilirubin grade (ALBI) scores to APRI and ALBI alone for PHLF and found it had higher predictive potential. Moreover, combined APRI+ALBI scores better predicted chronic liver diseases by comprehensively evaluating different liver functions. 

The present study developed an APRI+ALBI-based preOP MVM to predict PHLF using data from 12,056 patients enrolled in the National Surgical Quality Improvement Program (NSQIP) database. These patients underwent elective hepatic resection between 2000 and 2021, and their combined preOP APRI and ALBI scores were available.

Next, the researchers used an independent international multicenter cohort of 2,525 patients (validation cohort) to validate this MVM performance.

Another subcohort of 620 patients derived from the validation cohort was then used to directly compare how APRI+ALBI MVM trained on the NSQIP cohort performed against other models based on liver function tests, such as ALICE, ICG-clearance, and FIB4 concerning PHLF B+C prediction by comparing their respective areas under the curve (AUC). Notably, these tests are more expensive, time-consuming, and sometimes invasive. 

The best model based on the minimal Akaike information criterion (AIC) was selected during statistical analyses using stepwise backward feature elimination. It included these parameters: the APRI+ALBI score, sex, age, tumor type, and extent of resection. The prediction performance of this model was tested using receiver operating characteristic (ROC) curve analysis and validated using the validation cohort.

Results

The APRI+ALBI-based MVM predicted PHLF B+C with an AUC of 0.77 and showed similarly high performance in the validation cohort, with an AUC of 0.74.

Furthermore, APRI+ALBI-based MVM, trained in a cohort of 12,056 patients, demonstrated equal predictive potential for PHLF B+C as other prediction models based on more time-consuming and expensive liver function tests, e.g., ICG clearance and ALICE. 

The lack of association of ICG clearance with CALI partly explains why the APRI+ALBI score appeared to have a higher predictive potential than ICG clearance for PHLF B+C in this study. Moreover, the APRI+ALBI score was available at a fraction of the costs of ICG clearance measurement, exhibited none of its invasive features, and eliminated the risk of allergies to ICG dye components. 

Eventually, the researchers even designed a smartphone application to calculate the APRI+ALBI MVM scores for patient-specific PHLF risk assessment.

This study relied on crude and poorly validated cutoffs for volumetry (e.g., 30% and 40% after chemotherapy and for cirrhotic livers, respectively), which likely led to an underestimation of the relevance of quantifying hepatic function.

However, it laid a strong foundation for developing integrative models with volumetry, which might enable a patient-specific assessment of the liver remnant volume post-surgery. 

Conclusions

Researchers developed a novel APRI+ALBI score-based preOP MVM to predict multiple outcome measures after hepatic resection, particularly clinically significant for PHLF. Its validation in a large cohort (of >14,000 patients) found it performed well despite the use of routine laboratory tests to calculate the APRI+ALBI score. Moreover, it showed comparable predictive potential as other predictive tools based on more expensive and time-consuming liver function tests. 

Encouragingly, APRI+ALBI score-based preOP MVM could become universally available at an affordable price to perform a patient-specific risk assessment of PHLF before hepatectomy, facilitated by a free smartphone app.