Development and internal validation of the Laboratory-based Liver Frailty Index (Lab-LFI)-a prediction model for frailty in patients with cirrhosis.

BACKGROUND AND AIMS: Frailty, as measured by the Liver Frailty Index (LFI), is a well-established predictor of outcomes in cirrhosis, but testing requires in-person encounters. We sought to develop a blood-based biomarker index that correlates with frailty. APPROACH AND RESULTS: Seven serum/plasma analytes (myostatin, IL-6, growth differentiation factor-15 [GDF-15], testosterone, leucine-rich alpha-2 glycoprotein [LRG1], fetuin-A, and IGF-1) were measured by ELISA in 527 ambulatory cirrhosis patients. "Frail" was defined as LFI≥4.4. The cohort was split in half; the "Lab-LFI" was developed using LASSO regression in the training set and validated in the test set. Other machine learning methods (generalized additive model [GAM], Random Forest, and XGBOOST) were considered to compare discrimination. Of 527 patients (median age 59, MELD 14), 114 (21.6%) were frail. All seven biomarkers were associated with frailty. In the test set, Lab-LFI's c-statistic was 0.80 (95% CI 0.74-0.87) for frailty. The GAM, random forest, and XGBOOST approaches produced similar discrimination (c-statistics 0.79-0.82); as such, the LASSO model was chosen for maximum interpretability. Addition of clinical variables to the LASSO model, including MELD labs and sex, did not improve discrimination (c-statistic 0.79, 95% CI 0.72-0.85). Patients were stratified by Lab-LFI using a cutpoint maximizing Youden's index; after adjusting for MELD and liver disease etiology, the high-risk group had higher pre-transplant death/waitlist dropout (HR 1.77, 95% CI 1.17-2.68, p=0.007). CONCLUSIONS: We developed and internally validated the Lab-LFI, a frailty prediction model consisting of 7 laboratory-based biomarkers. Lab-LFI enables blood-based assessment of physical function, as MELD does with liver function.