Clinical implications of a dynamic nomogram for predicting sepsis in acute liver failure

Abstract

This letter comments on a web-enabled, dynamic nomogram developed for early sepsis-risk estimation in adults with acute liver failure (ALF) admitted to the intensive care unit. The study successfully established and validated the sepsis in ALF model using five routinely available variables: Age, total bilirubin, lactate dehydrogenase, albumin, and mechanical ventilation. Across cohorts, the model demonstrated strong discrimination and outperformed traditional scores. We commend the inclusion of both Western and Chinese intensive care unit cohorts, which enhances the cross-population generalizability of the findings. This letter highlights the strengths of the model, including its web-based dynamic calculator and effective risk stratification, while also acknowledging limitations such as reliance on baseline admission data, restriction to intensive care unit populations, and the absence of infection-related biomarkers. We encourage further prospective, multicenter investigations to refine the sepsis in ALF model and expand its clinical utility.

Key Words: Acute liver failure; Risk prediction; Predictive model; Intensive care; Biomarkers; Dynamic nomogram

Core Tip: This letter highlights the development and validation of a dynamic nomogram for predicting sepsis risk in patients with acute liver failure. The sepsis in acute liver failure model shows robust discrimination, consistently outperforming the sequential organ failure assessment and systemic inflammatory response syndrome scores, and has been validated across both Western and Chinese intensive care unit cohorts. Future work should incorporate longitudinal clinical variables, integrate infection-related biomarkers, and pursue multicenter validation to further enhance clinical utility.

TO THE EDITOR

Qi et al[1] report the development and external validation of the sepsis in acute liver failure (SIALF) nomogram, designed to provide individualized sepsis-risk estimates for patients with acute liver failure (ALF). This article addresses a clinically important challenge, as sepsis is among the most lethal complications of ALF and contributes substantially to early mortality, making accurate and early risk stratification critical for guiding timely intervention[2,3]. External validation in a Western critical-care database (Medical Information Mart for Intensive Care IV) alongside a Chinese intensive care unit (ICU) cohort from the Fifth Medical Center of the Chinese PLA General Hospital supports the model’s generalizability across populations and aligns with the increasing adoption of nomogram-based prediction tools in hepatology and gastroenterology[4].

Several strengths of the study deserve emphasis. First, the SIALF nomogram draws on five variables that are typically available in routine care: Age, total bilirubin, lactate dehydrogenase, albumin, and mechanical ventilation. Together, these measures capture hepatic injury/clearance, systemic tissue damage, nutritional or inflammatory reserve, and the intensity of organ support; each has been independently associated with sepsis risk in patients with liver dysfunction[5-10]. As the original rationale of the SIALF predictors is detailed in the original article, we summarize it here and focus our commentary on the calibration, external generalizability, integration of dynamic/biomarker data, and bedside utility of the nomogram. In addition to discrimination, the original article reported strong calibration of the SIALF model (Hosmer-Lemeshow P = 0.989; Brier score = 0.128), with calibration curves closely matching observed outcomes, a key strength of the work. To further aid clinical interpretation, future reports could include subgroup calibration [such as according to age bands or baseline sequential organ failure assessment (SOFA)/severity strata] or calibration-in-the-large/slope within strata, enabling readers to assess performance across patient categories.

While Sepsis-3 requires pathogenically positive infection, the original study does not report microbiological data (including blood/urine culture results or pathogen classes such as gram-negative vs gram-positive). Briefly reporting these data, or clarifying their absence, would strengthen the diagnostic framing and antimicrobial stewardship in future iterations of the SIALF, and would help readers assess model performance across pathogen categories.

Second, the model demonstrated excellent discrimination, with area under receiver operating characteristic curve values of 0.849 (derivation), 0.847 (internal validation), and 0.835 (external validation), consistently outperforming the SOFA and systemic inflammatory response syndrome (SIRS) criteria (Table 1). Notably, SIRS criteria, although convenient for systemic inflammation screening, are often nonspecific and overly sensitive in cirrhosis or acute liver dysfunction, where physiological disturbances such as hyperdynamic circulation, hypersplenism, and altered mental status can mimic inflammatory signs, confounding their interpretation. For example, Philips et al[11] and Bruns et al[12] highlighted how SIRS scores may lack specificity and reliability in liver failure and cirrhosis contexts, underscoring the need for ALF-specific models such as the SIALF. Moreover, the incorporation of an online dynamic calculator enhances clinical applicability by enabling real-time bedside risk assessment, in line with recent World Journal of Gastroenterology reports, including those by Shi et al[13] and Pu et al[14], who introduced interactive web-based calculators and evaluated models using decision-curve analysis to demonstrate clinical utility.

Table 1 Discrimination performance of the predictive models for sepsis risk in patients with acute liver failure in the internal derivation, internal validation, and external validation cohorts (area under receiver operating characteristic curve).

Cohort1	SIALF	SOFA	SIRS
Derivation (MIMIC-IV, n = 357)	0.849	0.733	0.578
Internal validation (n = 153)	0.847	0.746	0.653
External validation (FMCPH, n = 228)	0.835	0.721	0.615

¹Based on the principle of maximum Youden index, the optimal sepsis in acute liver failure cutoff was chosen at the threshold maximizing sensitivity + specificity - 1. The cohort-specific optima were 0.726 (derivation), 0.717 (internal validation), and 0.719 (external validation); for clinical interpretability we adopted a unified cutoff of 0.72.

MIMIC-IV: Medical Information Mart for Intensive Care IV; FMCPH: Fifth Medical Center of Chinese PLA General Hospital; SIALF: Sepsis in acute liver failure; SOFA: Sequential organ failure assessment; SIRI: Systemic inflammatory response syndrome.

Third, the rigorous validation in both the Western Medical Information Mart for Intensive Care IV and Chinese Fifth Medical Center of the Chinese PLA General Hospital ICU cohorts highlights the strong generalizability of the model across diverse populations. This finding is consistent with previous multicenter hepatology nomograms that integrate routine clinical variables such as bilirubin levels, the international normalized ratio, and liver stiffness for perioperative and portal hypertension risk stratification[4]. Finally, risk stratification using an optimal cutoff of 0.72 effectively distinguished high-risk patients, who exhibited significantly higher 28-day (47.8% vs 28.5%, P < 0.001) and 90-day (53.5% vs 40.7%, P = 0.017) mortality. Similar nomogram-based stratifications have demonstrated significant survival differences across risk groups; for example, a study in patients with postoperative sepsis revealed marked 90-day survival disparities among low-, moderate-, and high-risk groups (P < 0.0001)[15].

Nevertheless, several important considerations remain. The authors restricted the predictors to variables available on day 1 of ICU admission. While this facilitates early risk estimation, incorporating serial measurements of dynamic laboratory and hemodynamic parameters (such as procalcitonin, presepsin, and interleukin-6) may further refine the prognostic accuracy of the nomogram and guide antimicrobial strategies[16,17]. Moreover, although the model was externally validated, both the derivation and validation cohorts were hospital-based ICU populations; therefore, generalizability to broader settings (including step-down units or resource-limited hospitals) remains uncertain, particularly given the complex gut-immune-organ cross-talk that shapes sepsis trajectories beyond the ICU[3]. Additionally, the integration of microbiological culture results with infection-related biomarkers (such as procalcitonin, C-reactive protein, and interleukin-6) could enhance early sepsis identification, improve risk stratification, and support antimicrobial stewardship[9,16]. As the original study excluded patients under 18 years of age and those with preexisting sepsis, future research could assess whether modified versions of the SIALF model extend to pediatric populations or to patients with ALF presenting with early concurrent infections. Furthermore, exploring integration with machine-learning algorithms may enhance predictive performance by accommodating complex nonlinear interactions among clinical variables. To address the limitations of using baseline-only predictors, future studies could prespecify time points for dynamic data collection (6, 12, and 24 hours for vital and laboratory markers such as procalcitonin, C-reactive protein, and interleukin-6 and daily SOFA trajectories) and could evaluate serial or landmark models (mixed-effects features and 6/12/24-hour landmarking) to quantify incremental prognostic value and timeliness. To improve the generalizability of the model beyond ICU settings, future work could pursue multicenter validation across emergency departments, step-down units, general wards, and resource-limited hospitals, with prespecified subgroup and transportability analyses (etiology, geography, and baseline severity) and pragmatic recalibration (intercept/slope, Platt, or isotonic) where needed.

In summary, Qi et al[1] have made a valuable contribution by providing a robust and user-friendly predictive tool for identifying patients with ALF who are at risk of sepsis. We anticipate that prospective, multicenter studies integrating dynamic variables, pathogen-specific data, and pragmatic endpoints (including antibiotic de-escalation, ventilator-associated complications, and nutritional strategies) will help optimize clinical utility and promote individualized management in this vulnerable population.