Liver decompensation / PHLF risk (HCC, major hepatectomy)

Preoperative estimate of ISGLS grade B or C PHLF after major hepatectomy for HCC using the Cheng et al. 2022 multivariable model (age, sex, total bilirubin, PT, CSPH). For liver reserve context, see also Child-Pugh, MELD-Na, and FIB-4.

Age (years)

Sex

Total bilirubin (preoperative)

Bilirubin unit

Prothrombin time (seconds)

Use laboratory PT in seconds as in the derivation cohort, not INR.

Clinically significant portal hypertension (CSPH) — esophageal varices on endoscopy, or splenomegaly (major diameter >12 cm) with platelets <100,000/mm³ (paper definition).

Disclaimer: This calculator supports shared decision-making for adults being evaluated for major liver resection for HCC. It does not replace multidisciplinary assessment, imaging-based volumetry, dynamic liver tests, or guideline-based contraindications. If the patient does not meet the original study inclusion criteria, interpret with extra caution.

Cheng et al. nomogram — equation and scope

Overview

This tool implements the preoperative logistic model from Cheng N et al., Frontiers in Oncology 2022, for ISGLS grade B or C post-hepatectomy liver failure (PHLF) in patients with hepatocellular carcinoma (HCC) undergoing major hepatectomy (resection of three or more Couinaud segments). Clinicians often describe severe PHLF as a form of postoperative liver decompensation. Primary reference: DOI 10.3389/fonc.2022.817895.

Logistic probability

Multivariable odds ratios in Table 3 of the paper are used as per-unit (or per-category) multiplicative effects on the odds. The linear predictor is LP = β₀ + Σ βᵢxᵢ with βᵢ = ln(ORᵢ) for each term. Then:

P(PHLF grade B/C) = 1 / (1 + exp(−LP))

The published manuscript does not print raw regression coefficients; this implementation uses the reported multivariable ORs. The intercept β₀ is set so that, at the training cohort’s mean covariate pattern (Table 1), the predicted probability equals the observed 13.5% incidence of grade B/C PHLF (93 of 688 patients).

Terms (multivariable model, Table 3)

Sex: male vs female reference, OR 2.75.
Age: per year, OR 1.03.
Total bilirubin: per mg/dL, OR 1.23 (enter value in mg/dL or µmol/L; mg/dL is the unit in the paper).
Prothrombin time: per second, OR 1.51 (laboratory PT in seconds, not INR).
Clinically significant portal hypertension (CSPH): yes vs no, OR 2.09. Per the paper: esophageal varices on endoscopy, or splenomegaly (major diameter >12 cm) with platelet count <100,000/mm³.

Risk groups (published tertiles on predicted risk)

Category	Predicted probability
Low	≤ 8.6%
Medium	> 8.6% and ≤ 13.9%
High	> 13.9%

These cutoffs correspond to the authors’ tertile split on nomogram-predicted risk (Figure 2 narrative in the paper). They describe group-level risk in the derivation cohort; individual calibration may differ in other populations (the authors note limited Western validation).

Exclusions in the derivation cohort

The model was developed in selected patients (e.g. no preoperative TACE, no prior hepatectomy in the study, major resection only). If your patient falls outside that context, interpret cautiously and rely on multimodal assessment (volume, indocyanine green clearance, centre pathways, and guidelines).

Why this topic matters

Partial hepatectomy remains a central curative-intent option for selected patients with hepatocellular carcinoma (HCC). When a large volume of functional liver must be removed—major hepatectomy, commonly defined as resection of three or more Couinaud segments—the remnant liver must regenerate and sustain synthetic, metabolic, and detoxification workloads under perioperative stress. In some patients, that reserve is exceeded, and the liver fails to meet physiological demands in the early postoperative period. That syndrome is usually discussed as post-hepatectomy liver failure (PHLF). In everyday language, severe PHLF overlaps with what clinicians describe as postoperative liver decompensation: coagulopathy, hyperbilirubinemia, encephalopathy, ascites, renal dysfunction, and need for escalated support.

Because PHLF is a leading driver of morbidity and mortality after liver resection, there is persistent interest in preoperative risk estimation that is specific enough to support counseling, intensity of monitoring, and multidisciplinary decision-making—while remaining practical at the bedside. The calculator on this page implements a published preoperative logistic model developed in a multicenter HCC cohort undergoing major hepatectomy, with an endpoint of clinically consequential PHLF rather than mild biochemical deviation alone.

Defining the outcome: ISGLS grading of PHLF

Not every postoperative liver enzyme elevation represents “liver failure” in a management-changing sense. The International Study Group of Liver Surgery (ISGLS) framework stratifies PHLF by severity. In brief, the diagnosis rests on elevated bilirubin and/or INR relative to preoperative values together with a clinical picture consistent with liver dysfunction, with grades reflecting whether care must be escalated and whether organ support is required.

Grade A PHLF may be associated with laboratory deterioration but typically does not force a fundamental change in clinical management and is not the focus of aggressive prognostic modeling in many surgical studies. Grade B implies insufficient liver function such that the usual postoperative trajectory cannot be maintained without modification of care (for example, prolonged need for medical support or intervention-directed management). Grade C denotes need for organ support or intensive care–level measures. The model implemented here targets grade B or C PHLF as the composite “severe PHLF” endpoint—aligning with analyses that aim to capture complications that materially alter hospital course and risk.

Study context: major hepatectomy for HCC

The derivation work summarized by this tool enrolled consecutive HCC patients treated with laparoscopic or open major hepatectomy across participating centers, with explicit exclusions that materially change risk profiles or confound preoperative labs (examples emphasized in the primary report include staged resection strategies such as portal vein embolization pathways, preoperative transarterial chemoembolization in the study window, certain biliary drainage scenarios, prior hepatectomy, distant metastasis, and incomplete data). The cohort therefore represents a selected surgical population rather than all-comers with HCC.

Baseline characteristics in the published cohorts were notable for a high prevalence of chronic hepatitis B–related disease, which is typical of East Asian hepatobiliary centers. That epidemiologic skew matters for transportability: models trained in one liver-disease ecosystem may require revalidation where non-viral metabolic liver disease, alcohol-related cirrhosis, or different referral patterns predominate.

What the model uses (and why each variable is plausible)

After variable clustering to reduce collinearity and a penalized selection procedure to narrow candidates, the final multivariable model retained five preoperative inputs:

Age (per year): Older patients may tolerate major resection less well due to reduced regenerative reserve, altered drug handling, and higher burden of comorbidity—effects that can manifest as prolonged coagulopathy, infection susceptibility, and renal vulnerability in the postoperative window.
Sex (male vs female): The published multivariable analysis identified male sex as associated with higher odds of severe PHLF in this dataset. Mechanistic explanations are uncertain and may partly reflect residual confounding or baseline severity differences; nonetheless, the association was retained in the adjusted model reported by the authors.
Total serum bilirubin (per mg/dL): Bilirubin integrates secretory/excretory capacity and severity of cholestasis; rising values often accompany impaired hepatic clearance and can signal limited reserve before resection.
Prothrombin time in seconds (not INR): The derivation cohort modeled PT as reported by the laboratory in seconds. While INR is ubiquitous in transplant and cirrhosis scoring, the original publication’s effect estimate is tied to the PT scale as analyzed; users should enter the same quantity their local laboratory reports for consistency.
Clinically significant portal hypertension (CSPH): Portal hypertension marks hemodynamic remodeling of cirrhosis and is associated with splanchnic congestion, collateral circulation, and vulnerability after parenchymal loss. In the source study, CSPH was operationalized as endoscopic esophageal varices or the combination of marked splenomegaly (major diameter >12 cm) with thrombocytopenia (platelets <100,000/mm³). This definition is not interchangeable with every informal use of “portal hypertension” in clinic notes.

Model performance and what “probability” means here

In the training cohort, the reported discrimination for grade B/C PHLF was in the moderate range (apparent AUROC approximately 0.73 with internal bootstrap validation), with external validation AUROC approximately 0.72. Calibration plots in the primary report suggested reasonable agreement between predicted and observed risk across much of the spectrum, although any single patient’s probability should be interpreted as a group-derived estimate, not a guaranteed individual forecast.

The authors also reported that this preoperative model outperformed conventional liver reserve scores such as MELD and the albumin–bilirubin (ALBI) grade for this specific endpoint in their cohorts. That comparison supports the intuition that PHLF after major resection is not fully captured by generic cirrhosis severity metrics alone; nonetheless, those scores remain valuable for complementary assessment and guideline-aligned pathways.

How this site’s calculator implements the published odds ratios

The primary manuscript presents multivariable odds ratios for the retained predictors. In a logistic model, each odds ratio corresponds to a multiplicative effect on odds per unit change (for continuous variables) or for the contrasted category (for binary variables). Translating ORs into a probability therefore requires a linear predictor on the log-odds scale and a logistic transform to convert log-odds into a predicted probability.

Because the paper does not print the raw intercept term, this implementation sets the intercept so that—when evaluated at the published cohort’s mean covariate pattern (approximated from reported summary statistics)—the predicted probability matches the observed marginal incidence of grade B/C PHLF in the training set. This calibration step anchors absolute risk levels to the derivation population while preserving the relative weighting implied by the multivariable ORs. Predicted probabilities may differ slightly from values obtained by manually reading a printed nomogram line-by-line, but the intent is to remain faithful to the same statistical ingredients described in the report.

Risk bands used in the tool

For communication, the source analysis described tertile-style groups based on predicted probability: approximately ≤8.6% (low), between 8.6% and 13.9% (medium), and >13.9% (high). These thresholds are useful for stratifying follow-up intensity and for framing discussions, but they are not universal treatment rules. A patient just below or above a boundary is not inherently “safe” versus “unsafe”; resection candidacy depends on oncologic stage, expected remnant function, portal pressure assessment, anesthesia risk, and center expertise.

Clinical integration: what to do alongside the number

Responsible use pairs the predicted risk with volume assessment (future liver remnant), dynamic tests where available (for example indocyanine green clearance in centers that standardize it), evaluation of portal hypertension, infection control, nutritional status, and review of medications that stress the liver or alter coagulation. For patients with advanced fibrosis or cirrhosis, society guidelines emphasize careful selection for resection when portal hypertension is clinically significant; this model’s CSPH term aligns conceptually with that concern but does not replace endoscopic or hemodynamic evaluation when indicated.

Postoperatively, teams should continue to apply ISGLS-oriented monitoring and escalate care when PHLF criteria emerge, independent of whether preoperative risk was labeled low. The calculator is a decision support adjunct, not a substitute for surgeon judgment or multidisciplinary tumor board recommendations.

Limitations and cautions

Retrospective derivation: Like many surgical prediction tools, the model reflects care patterns, referral filters, and outcome ascertainment from its era and institutions.
Disease etiology and geography: Performance may shift when applied to populations with different causes of chronic liver disease or different surgical volumes.
Resection extent: The model was developed among patients undergoing major hepatectomy; applying it to minor resections or to non-HCC indications is extrapolation.
Missing intraoperative factors: Blood loss, ischemia time, bile duct injury risk, and extent of parenchymal injury can modify postoperative course but are not part of this preoperative score.
Endpoint specificity: The tool estimates grade B/C PHLF risk, not overall mortality, length of stay, or need for transplant rescue—though those outcomes correlate with severe PHLF.

Using this page responsibly

Enter preoperative values that reflect the patient’s status at the time resection is being planned. If CSPH criteria are uncertain, resolve them with appropriate testing rather than guessing. When probabilities suggest elevated risk, consider whether alternative oncologic strategies, staged approaches, or referral to a high-volume hepatobiliary center are appropriate. Always document shared decision-making, especially when patient values prioritize aggressive cancer surgery despite higher predicted hepatic risk.

CalcMD.

Liver Decompensation Risk after Hepatectomy for HCC

Preoperative ISGLS grade B/C post-hepatectomy liver failure (PHLF) risk after major hepatectomy for HCC: age, sex, bilirubin, prothrombin time, and CSPH. Based on Cheng et al., Frontiers in Oncology 2022.