A conceptually-derived, normally-distributed overall liver dysfunction index (OLDI) predicts short- and long-term outcomes after hepatectomy for hepatocellular carcinoma: Single-institution validation in 844 HCC patients

Abstract

Introduction: We developed the OLDI index: a non-empirically-derived and normally-distributed statistic which summarizes measures of liver synthetic function and hepatocyte death using the concept of geometric averages. The geometric average is superior to the arithmetic or harmonic average for representing central tendencies as it is unit-agnostic i.e. preserves the underlying statistical distribution during scale conversion (e.g. U/L to mmol/L). Method: Most prognostic scores that have been proposed are empirically-calibrated using regression techniques (i.e. have coefficient terms), and are therefore sensitive to the sample size and representativeness of the derivation set, and often fail to be reproduced in external validation sets. We showed that the geometric average of commonly-reported liver dysfunction biomarkers (“OLDI”) can be mathematically-rearranged as: (Albumin (g/L))/((Prothrombin time (s)×Bilirubin (μmol/L)×(AST [U/L]×ALT [U/L])1/2)1/3). The OLDI index was validated in the Singapore General Hospital cohort of 844 HCC patients who underwent hepatectomy between 2001 and 2016. Result: OLDI was found to be normally-distributed in the overall cohort (mean[SD]=2.104 [0.671], median[range]=2.105[0.205-4.485]), and in clinically-partitioned subsets of our cohort. In univariable analyses, OLDI predicted Clavien-Dindo grade≥2 complications (OR=0.686, 95% CI: 0.543-0.866, P=0.0016), post-hepatectomy liver failure using the ISGLS Grade B/C criteria (OR=0.631, 95% CI: 0.444-0.897, P=0.0103) or 50-50 criteria (OR=0.520, 95% CI: 0.290-0.933, P=0.0283), length of hospital stay (beta=-1.29 days; 95% CI: -2.04 to -0.54, P=0.0008 for every additional 1-point increase in OLDI), recurrence-free survival (HR=0.700, 95% CI: 0.605-0.811, P<0.0001), and overall survival (HR=0.655, 95% CI: 0.551-0.780, P<0.0001). The OLDI score remained an independent predictor of all outcomes in multivariable models adjusted for clinically-relevant confounders. Crucially, OLDI was found to outperform popular risk-stratification systems including Child-Pugh points, MELD score, TNM staging, and albumin-bilirubin score for all outcomes examined. Conclusion: OLDI outperformed many widely-used risk-stratification systems for predicting a multitude of short-term and long-term outcomes after hepatectomy for HCC, and is also easily-interpretable because of its normal distribution.