Association between Lok index and all-cause mortality among sepsis patients in critical care: insights from the MIMIC-IV database

Abstract

Background

Sepsis is a multiple organ dysfunction syndrome, and the liver plays a significant role in the inflammatory response of sepsis. The Lok index is a common indicator of liver health, but relationship between the Lok index and sepsis prognosis is not yet clear. This study aimed to explore the association between the Lok index and clinical outcomes in patients with severe sepsis.

Method

We extracted data on patients with sepsis from the Medical Information Database for Intensive Care IV (MIMIC-IV). Cox proportional hazards models were constructed to evaluate the relationship between Lok index and all-cause mortality at 30 days, 90 days, and 1 year. Kaplan-Meier survival curves showed the differences in mortality outcomes between different Lok index groups. Restricted cubic splines (RCS) were used to examine the dose-response relationship between Lok index and all-cause mortality risk.

Results

A total of 6,404 patients were included in this study and the median age of participants is 72 years, including 3,684 males (57.52%). All-cause mortality at 30, 90, and 365 days was the primary outcome, and patients were stratified into quartiles based on the Lok index. Kaplan-Meier survival curves show that a higher Lok index is associated with a higher risk of all-cause mortality at 30, 90, and 365 days (log-rank P < 0.010). Cox proportional hazards regression analysis shows that the highest quartile of Lok index had a significantly higher risk of death (HR for the fully adjusted model [HR for 30 days: 1.011 (95% CI 1.005–1.017), P < 0.001; for 90 days: 1.011 (95% CI 1.006–1.016), P < 0.001; for 1 year: 1.011 (95% CI 1.007–1.016), P < 0.001). RCS analysis reveals a significant nonlinear correlation between Lok index and all-cause mortality (P nonlinear = 0.003). In the subgroup analysis, the hazard ratio (HR) of 30-day all-cause mortality was significantly different in the atrial fibrillation (AF) and hypertension subgroup.

Conclusion

Among patients with severe sepsis, elevated Lok index values predicted an increased likelihood of death at 30, 90, and 365 days. These findings suggest that the Lok index may serve as a novel and readily available prognostic tool for risk stratification in sepsis.

Clinical trial number

Not applicable.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12879-025-12469-y.

Background

Sepsis is a critical illness characterized by an abnormal immune reaction to infection, resulting in organ dysfunction [1]. Globally, sepsis continues to pose a significant health challenge, contributing to around 50 million cases and 10 million deaths each year [2, 3]. In the United States, about 2% of hospitalized patients experience severe sepsis, with half requiring intensive care, accounting for 10% of intensive care unit (ICU) admissions [4]. The mortality rate is 28.6%, with age-stratified rates ranging from 10% in children to 38.4% in the elderly population aged 85 and over [5]. Currently, there is a lack of accurate indicators to predict outcomes in severe sepsis patients [6, 7], leading to poor prognosis. Given the high mortality rate, early identification of high-risk patients is crucial. Since hepatic dysfunction is a key factor in sepsis pathophysiology [8], liver-related biomarkers may serve as potential prognostic tools.

Hepatic involvement is essential in sepsis through mechanisms such as bacterial clearance and metabolic adaptation to inflammation [9]. In liver disease, portal hypertension [10] reduces intestinal mucosal barrier function and increases permeability [11]. Bacteria from the intestinal lumen can enter the bloodstream via the lymphatic system or portal vein. Patients with liver disease are more susceptible to infections and sepsis due to impaired immune function and hyperdynamic circulation [12]. The severity of sepsis is closely associated with all-cause mortality [13].

Given the liver’s importance in sepsis, biomarkers related to liver function have emerged as potential prognostic tools. Emerging evidence shows that specific liver-related indicators can assess fibrosis severity and reflect the systemic inflammation and organ failure commonly seen in sepsis [14]. Several indices assess liver function, including the Fibrosis-4 Index (FIB-4), Lok index, Forns index, glutamic oxaloacetic transaminase (AST), aspartate aminotransferase-to-platelet ratio (APRI), Child-Turcotte-Pugh score, and Model for End-Stage Liver Disease (MELD) [15–17]. Among these, FIB-4 is a commonly adopted non-invasive tool for liver fibrosis risk stratification [18]. The Lok index, also non-invasive, was developed to predict cirrhosis in chronic hepatitis C patients [19] and demonstrates high accuracy in detecting significant liver fibrosis and cirrhosis. Unlike FIB-4, the Lok index incorporates the international normalized ratio (INR), which reflects hepatic synthetic capacity and coagulation function. Sepsis invariably involves both liver dysfunction [9] and coagulopathy [20], which are key determinants of patient outcomes. Therefore, the Lok index may provide a more holistic assessment of pathophysiology. This could enhance its prognostic capability compared to indices based solely on fibrosis markers. The FIB-4 and APRI indices primarily depend on platelet count, AST, and glutamic-pyruvic transaminase (ALT) to assess liver fibrosis risk. However, these indices lack direct evaluation of coagulation function. Therefore, the Lok index, by integrating INR, can more comprehensively capture liver dysfunction and coagulation abnormalities related to sepsis, thereby providing more accurate prognostic information. Moreover, the Lok index has shown high accuracy in chronic liver disease research. Its components (platelets, the ratio of AST/ALT, INR) are routinely tested in the ICU and are easy to calculate, making it suitable for rapid bedside risk assessment. While previous studies have suggested an association between FIB-4 and sepsis outcomes [21], no research has investigated whether the Lok index correlates with prognosis in septic patients. Evidence suggests the clinical utility of the Lok index may exceed that of FIB-4 [22]. Using the Medical Information Mart for Intensive Care IV (MIMIC-IV) database, this study assessed whether the Lok index is associated with all-cause mortality in critically ill patients diagnosed with sepsis.

Methods

Data source

Data were sourced from the publicly accessible MIMIC-IV v3.1 database. Access was granted to one of the authors (Qin Qin) after completing the required training (certification number: 65991171).

Study population

Patients were enrolled based on a sepsis diagnosis meeting International Classification of Diseases (ICD)-9 or ICD-10 criteria as defined by the International Classification of Diseases. The exclusion criteria were: patients without sepsis diagnosis, those with missing key laboratory parameters (ALT, AST, platelet count, or INR), those with ICU stay shorter than 24 h, and those with erroneous data entries.

Data extraction

Variables were initially identified based on their clinical relevance and availability within the MIMIC-IV database. The selection process was guided by a comprehensive review of existing literature and clinical expertise to ensure that variables with potential prognostic value were included. Variables were extracted using a Structured Query Language (SQL) with pgAdmin 4 v8.12 and R project for statistical computing version 4.4.2. MIMIC-IV derived tables based on the centralized MIMIC code repository were used to extract relevant variables for study purposes. Data includes patients for demographic details, ICU admission information, laboratory and clinical measurements, and comorbidity tables derived from standardized coding systems (e.g., ICD-9 and ICD-10) for comorbidities. To ensure consistency and reproducibility, all derived variables were defined using publicly available code from the MIMIC Code Repository. Variables with significant measurement variability or those not routinely collected in clinical practice were excluded. For laboratory variables, the first recorded value within the initial 24 h of ICU admission was selected to represent baseline status. For patients who have been hospitalized multiple times, we included the records of their first hospitalization in the analysis. After excluding patients with missing values in the four key variables (ALT, AST, platelets, and INR), the number of samples included in the final analysis decreased from the initial 28,032 to 6,418. A flowchart of patient selection is presented in Fig. 1. Variables were categorized into four domains: demographic data (age, sex, weight, height, and body mass index [BMI]), comorbidities (atrial fibrillation [AF], heart failure, diabetes mellitus, and kidney disease), laboratory findings (red blood cell count [RBC], white blood cell count[WBC], hemoglobin, platelet, serum sodium, serum creatinine, fasting blood glucose, and triglycerides), and illness severity scores (Acute Physiology and Chronic Health Evaluation III [APACHE III], Simplified Acute Physiology Score II [SAPS II], Oxford Acute Disease Severity Score [OASIS], and Sequential Organ Failure Assessment [SOFA]).

Fig. 1.

Patient selection process and applied exclusion criteria

In multivariate analysis, variables with excessive missing data can introduce significant bias and compromise model stability. A missing rate exceeding 20%-30% is generally considered problematic. This study adopted a conservative 20% threshold to balance information retention with model reliability. Variables exceeding this threshold were excluded from analysis to minimize bias. Among the collected data, more than 20% of the data for BMI(40.0%), height(41.2%), weight(35.2%) and triglycerides(83.9%) were missing. Therefore, we excluded these data. The patterns of missingness for key variables are summarized in Supplementary Table 1. Multiple imputation by chained equations (MICE) was applied to continuous variables with missing data below the 20% threshold to create 5 complete datasets. Interpolation involves the missing values of the following variables: WBC − 7 values, RBC − 2 values, sodium (sodium) − 1 value, creatinine (creatinine) − 1 value, fasting blood glucose (FBG) − 3 values, and Glasgow Coma Scale (GCS) − 38 values. The MICE procedure incorporated predictive mean matching and was run for 10 iterations to achieve convergence. Analyses were performed on each dataset and results were pooled using Rubin’s rules.

The Lok index was derived using the following formula: Lok index = − 5.56 − 0.0089 × platelet count (10⁹/L) + 1.26 × AST: ALT ratio + 5.27 × INR. Patients were stratified into four strata using Lok index ranges as the grouping standard (Fig. 1) to ensure balanced group sizes for comparative analysis. This data-driven approach avoids arbitrary cut-off selection and allows for the examination of potential nonlinear trends in mortality risk across the spectrum of the Lok index. This method is statistically robust for initial exploration of associations and helps identify whether risk increases incrementally or exhibits threshold effects at higher levels.

Clinical outcome

The primary outcome was all-cause mortality assessed at 30, 90, and 365 days.

Statistical methodology

The Kolmogorov–Smirnov test was employed to examine the distribution of continuous variables. Since none followed a normal distribution, data were summarized using interquartile ranges. Frequencies and percentages were used to summarize categorical data. Associations between categorical variables were analyzed using the Chi-square test, or Fisher’s exact test when sample sizes were small. The Kruskal–Wallis test was applied to compare continuous data. Kaplan–Meier (K-M) survival analysis evaluated endpoint event incidence across Lok index groups, with log-rank tests assessing between-group differences. To address collinearity, variance inflation factors were calculated and variables exceeding a threshold of 5 were removed from analysis. The variable we excluded was hemoglobin (variance inflation factor [VIF] = 5.543). VIF values for other key variables are reported in Supplementary Tables 2–3, no multicollinearity exist in the remaining variables. The proportional hazards assumption for the Cox models was verified using Schoenfeld residuals, and no significant violations were found.

Hazard ratios (HRs) with 95% confidence intervals (95% CIs) were estimated for 30-, 90-, and 365-day mortality using Cox proportional hazards models. A stepwise Cox modeling approach was applied: Model 1 was univariate; Model 2 controlled for demographic factors (age, gender); Model 3 further included age, gender, race, admission age, malignancy, cirrhosis, renal dysfunction, atrial fibrillation, diabetes, hypertension, RBC, sodium, creatinine, FBG, WBC, lactic acid, SOFA score, APSIII score, SAPS II score, OASIS score, GCS score. To examine nonlinear dose-response patterns, RCSs were utilized to model the association between Lok index and mortality, and the RCSs were performed using Model 3. Sensitivity analyses excluded patients with cirrhosis, cancer, and end-stage renal disease. Subgroup analyses by age and sex verified result robustness. All analyses used R v4.4.2; two-sided P < 0.05 indicated significance.

Results

Baseline characteristics

Table 1 presents baseline characteristics of 6,404 sepsis patients admitted to the ICU. Median age was 72 years (interquartile range: 63–81), and 3,684 (58%) were male.

Table 1.

Clinical characteristics stratified by Lok index quartiles

Characteristic	N 1	Overall (n = 6,404)	Q1 (Lok < 0.568, n = 1,601)	Q2 (0.568 ≤ Lok < 2.213, n = 1,601)	Q3 (2.213 ≤ Lok < 4.881, n = 1,601)	Q4 (4.881 ≤ Lok, n = 1,601)	p-value3
		N = 6,4042	N = 1,6012	N = 1,6012	N = 1,6012	N = 1,6012
Gender	6,404						< 0.001
Female		2,720(42%)	790(49%)	696(43%)	603(38%)	631(39%)
Male		3,684(58%)	811(51%)	905(57%)	998(62%)	970(61%)
Race	6,404						0.704
Asian		156(2.4%)	43(2.7%)	40(2.5%)	33(2.1%)	40(2.5%)
Black		751(12%)	206(13%)	191(12%)	171(11%)	183(11%)
White		4,154(65%)	1,024(64%)	1,028(64%)	1,066(67%)	1,036(65%)
Others		1,343(21%)	328(20%)	342(21%)	331(21%)	342(21%)
Age	6,404	72(63,81)	70(60,79)	73(64,82)	74(64,82)	73(62,82)	< 0.001
Malignancy	6,404						0.015
no		5,069(79%)	1,295(81%)	1,249(78%)	1,234(77%)	1,291(81%)
yes		1,335(21%)	306(19%)	352(22%)	367(23%)	310(19%)
cirrhosis	6,404						< 0.001
no		5,651(88%)	1,518(95%)	1,458(91%)	1,363(85%)	1,312(82%)
yes		753(12%)	83(5.2%)	143(8.9%)	238(15%)	289(18%)
Renal dysfunction	6,404						0.029
no		5,610(88%)	1,386(87%)	1,409(88%)	1,432(89%)	1,383(86%)
yes		794(12%)	215(13%)	192(12%)	169(11%)	218(14%)
Atrial fibrillation	6,404						< 0.001
no		2,592(40%)	816(51%)	720(45%)	601(38%)	455(28%)
yes		3,812(60%)	785(49%)	881(55%)	1,000(62%)	1,146(72%)
Diabetes	6,404						< 0.001
no		3,234(50%)	724(45%)	821(51%)	873(55%)	816(51%)
yes		3,170(50%)	877(55%)	780(49%)	728(45%)	785(49%)
Hypertension	6,404						< 0.001
no		3,144(49%)	718(45%)	746(47%)	814(51%)	866(54%)
yes		3,260(51%)	883(55%)	855(53%)	787(49%)	735(46%)
RBC, m/uL	6,404	3.48(2.97,4.05)	3.64(3.15,4.20)	3.55(3.05,4.10)	3.38(2.87,3.91)	3.35(2.79,3.95)	< 0.001
Platelet, K/uL	6,404	187(133,255)	251(198,327)	184(140,243)	157(113,217)	155(105,221)	< 0.001
Hemoglobin, g/dL	6,404	10.30(8.80,11.90)	10.70(9.20,12.30)	10.50(9.00,12.10)	10.10(8.50,11.70)	10.00(8.40,11.60)	< 0.001
Sodium, mEq/L	6,404	138(135,141)	138(135,141)	138(135,141)	138(135,141)	138(134,141)	0.009
Creatinine	6,404	1.30(0.90,2.20)	1.20(0.80,2.00)	1.30(0.90,2.10)	1.30(0.90,2.00)	1.60(1.10,2.50)	< 0.001
FBG, mg/dL	6,404	134(107,181)	141(110,192)	134(108,180)	131(105,174)	132(106,176)	< 0.001
ALT, U/L	6,404	27(16,64)	29(17,63)	25(15,55)	26(15,62)	29(16,79)	< 0.001
WBC, K/uL	6,404	11(8,16)	12(9,16)	11(8,16)	12(8,16)	12(8,17)	0.011
AST, U/L	6,404	42(24,97)	30(20,56)	37(22,76)	47(28,121)	61(32,183)	< 0.001
INR	6,404	1.40(1.20,1.70)	1.10(1.10,1.20)	1.30(1.20,1.40)	1.50(1.30,1.60)	2.30(1.80,3.20)	< 0.001
Lactic Acid, mmol/L	6,404	1.80(1.20,2.70)	1.60(1.10,2.20)	1.70(1.20,2.50)	1.90(1.30,2.90)	2.10(1.40,3.60)	< 0.001
SOFA score	6,404	6(4,9)	5(3,7)	6(3,8)	6(4,9)	7(5,10)	< 0.001
APSIII score	6,404	51(40,65)	49(38,61)	49(38,63)	52(41,65)	56(45,71)	< 0.001
SAPS II score	6,404	40(32,50)	37(30,45)	39(32,49)	41(33,50)	44(35,54)	< 0.001
OASIS score	6,404	34(28,40)	34(27,40)	34(28,40)	34(28,41)	34(28,41)	0.002
GCS score	6,404	15(13,15)	15(13,15)	15(13,15)	15(14,15)	15(13,15)	0.005
Status 30d	6,404	1,804(28%)	346(22%)	412(26%)	473(30%)	573(36%)	< 0.001
Time 30d	6,404	30(21,30)	30(30,30)	30(27,30)	30(20,30)	30(12,30)	< 0.001
Status 90d	6,404	2,368(37%)	477(30%)	547(34%)	617(39%)	726(45%)	< 0.001
Time 90d	6,404	90(21,90)	90(52,90)	90(27,90)	90(20,90)	90(12,90)	< 0.001
Status 1year	6,404	3,123(49%)	669(42%)	734(46%)	800(50%)	920(57%)	< 0.001
Time 1year	6,404	365(21,365)	365(52,365)	365(27,365)	365(20,365)	154(12,365)	< 0.001

¹ N Non-missing

² n(%); Median(Q1,Q3)

³ Pearson’s Chi-squared test; Kruskal-Wallis rank sum test

Abbreviations: RBC, red blood cell count; FBG, fasting blood glucose; ALT, glutamic-pyruvic transaminase; WBC, white blood cell count; AST, glutamic oxaloacetic transaminase; INR, International Normalized Ratio; ‌SOFA, Sequential Organ Failure Assessment; APSIII, Acute Physiology and Chronic Health Evaluation III; SAPS II, Simplified Acute Physiology Score II; OASIS, Oxford Acute Disease Severity Score; GCS, Glasgow Coma Scale

Patients were stratified by Lok index quartiles: Q1 (≤ 0.568), Q2 (0.568–2.213), Q3 (2.213–4.881), and Q4 (≥ 4.881). Patients in Q3 and Q4 had a greater proportion of males (P < 0.001), higher prevalence of cirrhosis and AF, lower RBC and platelet counts, and elevated creatinine, AST, and lactate levels. These patients showed higher SOFA, APACHE III, and SAPS II scores, indicating greater disease severity, and increased all-cause mortality at all time points.

K-M survival analysis

Survival probability declined sharply in the initial 0–5 days, then gradually decreased throughout the follow-up periods extended from 0 to 30 days. The decline was particularly pronounced during the first 10 days before gradually stabilizing. The similar trends observed across all periods but more gradual decline over the year-long observation (Fig. 2). Throughout the 30-day period, patients in the Q1 group consistently exhibited the highest survival probability at each time interval, while the Q4 group showed the lowest survival probability. The 30-day mortality rates across quartiles were 22% for Q1, 26% for Q2, 30% for Q3, and 36% for Q4. Pairwise comparisons demonstrated significant differences between all group pairs except Q2 and Q3.

Fig. 2.

K-M survival curves at 30 (A), 90 (B), and 365 days (C) (From top to bottom, Q1, Q2, Q3, Q4)

This mortality pattern persisted during extended follow-up periods. At 90 days, mortality rates increased to 30%, 34%, 39%, and 45% for quartiles Q1 through Q4, respectively. The pairwise comparison results remained consistent with the 30-day findings. By 365 days, mortality rates had further increased to 42%, 46%, 50%, and 57% across Q1 to Q4. Notably, significant differences between Q4 and all other groups, as well as between Q3 and Q1, remained consistent across all three time points.

Overall, pronounced differences in survival probabilities existed between the Q4 group and all other groups. The Q4 group had an obviously higher mortality rate, indicating that the Lok index significantly influences survival outcomes. The sustained group differences in survival probabilities throughout the entire follow-up period suggest that the impact of Lok index grouping on survival is enduring.

Cox proportional hazards analysis

Three Cox proportional hazards models assessed the independent association between Lok index and all-cause mortality (Tables 2, 3 and 4).

Table 2.

Cox proportional HR for 30-day all-cause mortality

	Model 1			Model 2			Model 3
	HR	95%CI	P	HR	95%CI	P	HR	95%CI	P
Lok	1.021	1.016,1.026	<0.001	1.202	1.015,1.026	<0.001	1.011	1.005, 1.017	< 0.001
Q1	Ref	Ref		Ref	Ref		Ref	Ref
Q2	1.236	1.072,1.426	0.004	1.177	1.020,1.358	0.026	1.085	0.939, 1.253	0.271
Q3	1.449	1.261,1.664	<0.001	1.367	1.189,1.572	<0.001	1.138	0.986, 1.314	0.077
Q4	1.846	1.616,2.110	<0.001	1.768	1.546,2.021	<0.001	1.312	1.136, 1.516	< 0.001
P for trend	1.223	1.172,1.275	<0.001	1.208	1.158,1.261	<0.001	1.092	1.043, 1.144	< 0.001

HR, Hazard Ratio; CI, confidence interval

Model 1 was crude model;

Model 2 was adjusted for age, gender, admission age;

Model 3 was adjusted for age, gender, race, admission age, malignancy, cirrhosis, renal dysfunction, atrial fibrillation, diabetes, hypertension, RBC, sodium, creatinine, FBG, WBC, lactic acid, SOFA score, APSIII score, SAPS II score, OASIS score, GCS score

Abbreviations: RBC, red blood cell count; FBG, fasting blood glucose; WBC, white blood cell count; ‌SOFA, Sequential Organ Failure Assessment; APSIII, Acute Physiology and Chronic Health Evaluation III; SAPS II, Simplified Acute Physiology Score II; OASIS, Oxford Acute Disease Severity Score; GCS, Glasgow Coma Scale

Table 3.

Cox proportional HR for 90-day all-cause mortality

	Model 1			Model 2			Model 3
	HR	95%CI	P	HR	95%CI	P	HR	95%CI	P
Lok	1.020	1.015, 1.024	< 0.001	1.019	1.014, 1.024	< 0.001	1.011	1.006, 1.016	< 0.001
Q1	Ref	Ref		Ref	Ref		Ref	Ref
Q2	1.198	1.060, 1.355	0.004	1.137	1.005, 1.285	0.041	1.057	0.933, 1.196	0.384
Q3	1.392	1.235, 1.569	< 0.001	1.306	1.158, 1.473	< 0.001	1.116	0.985, 1.263	0.084
Q4	1.749	1.558, 1.963	< 0.001	1.670	1.487, 1.875	< 0.001	1.305	1.152, 1.478	< 0.001
P for trend	1.202	1.159, 1.247	< 0.001	1.187	1.144, 1.231	< 0.001	1.092	1.049, 1.136	< 0.001

HR, Hazard Ratio; CI, confidence interval

Model 1 was crude model;

Model 2 was adjusted for age, gender, admission age;

Model 3 was adjusted for age, gender, race, admission age, malignancy, cirrhosis, renal dysfunction, atrial fibrillation, diabetes, hypertension, RBC, sodium, creatinine, FBG, WBC, lactic acid, SOFA score, APSIII score, SAPS II score, OASIS score, GCS score

Abbreviations: RBC, red blood cell count; FBG, fasting blood glucose; WBC, white blood cell count; ‌SOFA, Sequential Organ Failure Assessment; APSIII, Acute Physiology and Chronic Health Evaluation III; SAPS II, Simplified Acute Physiology Score II; OASIS, Oxford Acute Disease Severity Score; GCS, Glasgow Coma Scale

Table 4.

Cox proportional HR for 1-year all-cause mortality

	Model 1			Model 2			Model 3
	HR	95%CI	P	HR	95%CI	P	HR	95%CI	P
Lok	1.019	1.015, 1.023	< 0.001	1.018	1.014, 1.022	< 0.001	1.011	1.007, 1.016	< 0.001
Q1	Ref	Ref		Ref	Ref		Ref	Ref
Q2	1.150	1.036, 1.277	0.009	1.094	0.985, 1.215	0.095	1.021	0.918, 1.135	0.707
Q3	1.307	1.179, 1.448	< 0.001	1.234	1.113, 1.369	< 0.001	1.070	0.961, 1.190	0.217
Q4	1.634	1.479, 1.805	< 0.001	1.567	1.418, 1.732	< 0.001	1.269	1.140, 1.413	< 0.001
P for trend	1.176	1.139, 1.213	< 0.001	1.162	1.126, 1.200	< 0.001	1.083	1.046, 1.121	< 0.001

HR, Hazard Ratio; CI, confidence interval

Model 1 was crude model;

Model 2 was adjusted for age, gender, admission age;

Model 3 was adjusted for age, gender, race, admission age, malignancy, cirrhosis, renal dysfunction, atrial fibrillation, diabetes, hypertension, RBC, sodium, creatinine, FBG, WBC, lactic acid, SOFA score, APSIII score, SAPS II score, OASIS score, GCS score

In the unadjusted model, the Lok index was significantly associated with mortality at all time points: 30-day (HR 1.021, 95% CI 1.016–1.026, P < 0.001), 90-day (HR 1.020, 95% CI 1.015–1.024, P < 0.001), and at 1-year (HR 1.019, 95% CI 1.015–1.023, P < 0.001). After adjusting for demographic variables, statistical significance persisted for mortality at 30-day (HR 1.202; 95% CI, 1.015–1.026), 90-day (HR 1.019; 95% CI, 1.014–1.024), and 1-year (HR 1.018; 95% CI, 1.014–1.022), all with P < 0.001. In the fully adjusted model controlling for comorbidities, laboratory values, and severity scores, the HRs remained consistent across 30-, 90-, and 365-day mortality at 1.011, with corresponding 95% confidence intervals of 1.005–1.017, 1.006–1.016, and 1.007–1.016, respectively (all P < 0.001).

Mortality risk was significantly higher in Q4 compared to Q1 in quartile-stratified analysis. In the unadjusted model, the results were 30-day (HR 1.846, 95% CI 1.616–2.110), 90-day (HR 1.749, 95% CI 1.558–1.963), and 1-year (HR 1.634, 95% CI 1.479–1.805) mortality; all statistically significant (P < 0.001). After accounting for partial confounders, the association remained robust: HRs were 1.768 (95% CI, 1.546–2.021) at 30 days, 1.670 (95% CI, 1.487–1.875) at 90 days, and 1.567 (95% CI, 1.418–1.732) at 1 year (all P < 0.001). This association persisted in the fully adjusted model: 30-day HR was 1.312 (95% CI, 1.136–1.516), 90-day HR was 1.305 (95% CI, 1.152–1.478), and 1-year HR was 1.269 (95% CI, 1.140–1.413); P < 0.001 for all. ⇒ In the fully adjusted model, the HR for each unit increase in the Lok index was small (HR = 1.011). However, this association was highly statistically significant (P < 0.001), indicating it was unlikely to be due to chance. From a clinical perspective, this “per-unit” increase in risk needs to be interpreted in conjunction with the actual clinical variation range of this index. For instance, suppose the Lok index score of a patient differs by 10 units (which could occur in clinical practice), then the risk of death will increase by approximately 11.6% accordingly (calculation method: (1.011^10 − 1) * 100% ≈ 11.6%). The consistency of the risk ratio at the three time points (30 days, 90 days, and 1 year) is high, further supporting that the Lok index is an independent and stable predictor of mortality. Furthermore, the results of the quartile stratified analysis provided stronger clinical evidence for this association. Compared with patients in the lowest quartile (Q1), patients in the highest quartile (Q4) still had a significantly higher 30-day mortality risk of 31.2% (HR 1.312) after complete adjustment. This increased risk persisted at 90 days (HR 1.305) and 1 year (HR 1.269). This indicates that the patient group in the high end of the Lok index distribution has a clinically significant increase in mortality risk. These findings indicate that higher Lok index values (≥ 4.881) are independently correlated with higher risk of death across multiple time points, with risk escalating progressively across increasing Lok index levels.

RCS analysis

Figure 3 displays RCS curves illustrating the dose-response relationship between Lok index and all-cause mortality at 30, 90, and 365 days. The curves indicate that mortality risk increases with higher Lok index values. A significant nonlinear association was observed at 30 days (P = 0.028). Similar upward trends were observed at 90 days and 1 year, indicating consistent association between increasing Lok index and higher mortality risk.

Fig. 3.

RCS curves at 30 (A), 90 (B), and 365 days (C)

Subgroup analysis

Patients were stratified by sex, age, race, atrial fibrillation, diabetes, and hypertension for subgroup analyses (Fig. 4). In most subgroups, individuals in Q4 had significantly higher all-cause mortality at all time points compared with the lowest quartile, regardless of sex or age.

Fig. 4.

Forest plots of HRs for ICU mortality across multiple clinical subgroups at 30 (A), 90 (B), and 365 days(C)

However, no significant associations were observed for mortality among patients without hypertension at all time points, and for 30-day mortality among those with AF. Interaction testing revealed significant differences in 90-day mortality between patients aged ≥ 70 versus < 70 (P < 0.05), in all-timepoint mortality between those with and without AF, and in 30- and 90-day mortality between hypertensive and non-hypertensive patients (P < 0.05).

Sensitivity analysis

Tables 5, 6 and 7 present sensitivity analyses for 30-day, 90-day, and 1-year mortality. The association between Lok index and all-cause mortality remained significant after excluding patients with renal failure and cirrhosis. The trend test demonstrated a linear increase in mortality risk across Lok index quartiles (Q1 to Q4), and this trend remained statistically significant across all models and time points.

Table 5.

Sensitivity analysis for 30-day all-cause mortality

	Model 1			Model 2			Model 3
	HR	95%CI	P	HR	95%CI	P	HR	95%CI	P
Lok	1.020	1.015, 1.026	< 0.001	1.020	1.013, 1.026	< 0.001	1.010	1.003, 1.017	0.005
Q1	Ref	Ref		Ref	Ref		Ref	Ref
Q2	1.197	1.019, 1.407	0.029	1.143	0.972, 1.343	0.106	1.091	0.927, 1.283	0.296
Q3	1.459	1.249, 1.704	< 0.001	1.362	1.165, 1.593	< 0.001	1.145	0.975, 1.344	0.099
Q4	1.730	1.487, 2.012	< 0.001	1.614	1.385, 1.880	< 0.001	1.260	1.071, 1.482	0.005
P for trend	1.202	1.146, 1.260	< 0.001	1.176	1.121, 1.234	< 0.001	1.077	1.023, 1.134	0.005

HR, Hazard Ratio; CI, confidence interval

Model 1 was crude model;

Model 2 was adjusted for age, gender, admission age;

Model 3 was adjusted for age, gender, race, admission age, malignancy, cirrhosis, renal dysfunction, atrial fibrillation, diabetes, hypertension, RBC, sodium, creatinine, FBG, WBC, lactic acid, SOFA score, APSIII score, SAPS II score, OASIS score, GCS score

Table 6.

Sensitivity analysis for 90-day all-cause mortality

	Model 1			Model 2			Model 3
	HR	95%CI	P	HR	95%CI	P	HR	95%CI	P
Lok	1.019	1.014, 1.024	< 0.001	1.017	1.012, 1.023	< 0.001	1.009	1.003, 1.015	0.004
Q1	Ref	Ref		Ref	Ref		Ref	Ref
Q2	1.166	1.014, 1.342	0.031	1.104	0.960, 1.271	0.166	1.054	0.915, 1.214	0.464
Q3	1.437	1.256, 1.645	< 0.001	1.326	1.158, 1.519	< 0.001	1.139	0.990, 1.309	0.069
Q4	1.642	1.439, 1.875	< 0.001	1.516	1.327, 1.733	< 0.001	1.239	1.075, 1.427	0.003
P for trend	1.184	1.136, 1.234	< 0.001	1.155	1.107, 1.205	< 0.001	1.075	1.028, 1.125	0.002

HR, Hazard Ratio; CI, confidence interval

Model 1 was crude model;

Model 2 was adjusted for age, gender, admission age;

Model 3 was adjusted for age, gender, race, admission age, malignancy, cirrhosis, renal dysfunction, atrial fibrillation, diabetes, hypertension, RBC, sodium, creatinine, FBG, WBC, lactic acid, SOFA score, APSIII score, SAPS II score, OASIS score, GCS score

Table 7.

Sensitivity analysis for 1-year all-cause mortality

	Model 1			Model 2			Model 3
	HR	95%CI	P	HR	95%CI	P	HR	95%CI	P
Lok	1.018	1.013, 1.023	< 0.001	1.017	1.012, 1.022	< 0.001	1.009	1.004, 1.015	< 0.001
Q1	Ref	Ref		Ref	Ref		Ref	Ref
Q2	1.132	1.005, 1.277	0.042	1.073	0.951, 1.210	0.253	1.025	0.908, 1.158	0.684
Q3	1.335	1.188, 1.500	< 0.001	1.237	1.100, 1.391	< 0.001	1.076	0.953, 1.215	0.238
Q4	1.510	1.336, 1.683	< 0.001	1.392	1.239, 1.563	< 0.001	1.169	1.034, 1.322	0.013
P for trend	1.148	1.107, 1.190	< 0.001	1.121	1.081, 1.163	< 0.001	1.054	1.013, 1.096	0.009

HR, Hazard Ratio; CI, confidence interval

Model 1 was crude model;

Model 2 was adjusted for age, gender, admission age;

Model 3 was adjusted for age, gender, race, admission age, malignancy, cirrhosis, renal dysfunction, atrial fibrillation, diabetes, hypertension, RBC, sodium, creatinine, FBG, WBC, lactic acid, SOFA score, APSIII score, SAPS II score, OASIS score, GCS score

Discussion

This analysis assessed the association between the Lok index and all-cause mortality among ICU-treated sepsis patients. Elevated Lok index levels were found to be consistently linked with greater mortality across all measured intervals. The RCS curves revealed a nonlinear relationship, with HRs increasing approximately linearly with the Lok index. Once the index surpasses 10, the HR remains consistently high (> 1.2) with modest increases. These associations remained significant after controlling for confounders, indicating that the Lok index functions as an independent risk predictor with potential clinical utility for guiding patient management. The strong association between elevated Lok index values and adverse outcomes suggests significant prognostic value for risk stratification in ICU-treated sepsis cases.

Multiple indicators can predict sepsis prognosis. Researchers have found that indices such as qSOFA, Systemic Inflammatory Response Syndrome (SIRS), National Early Warning Score (NEWS) scoring systems [23], Plasminogen Activator Inhibitor-1 (PAI-1) [24], and circulating biologically active adrenomedullin (ADM) [25] can predict sepsis mortality. SOFA has been noted to offer moderate sensitivity and specificity, SIRS to have higher sensitivity, and Modified Early Warning Score (MEWS) to demonstrate greater specificity in mortality prediction [6]. PAI-1 increases in sepsis patients and may exacerbate thrombophilia [26], while elevated ADM reflects endothelial injury [27]. These indicators predict sepsis mortality from perspectives of inflammation, organ dysfunction, coagulation function, and endothelial injury. Compared with these indicators, the Lok index has advantages. First, it includes platelet count, AST, ALT, and INR, characterizing liver function and coagulation status in sepsis patients; second, these values are routinely available, making calculation simpler; third, given the liver’s crucial role in sepsis, incorporating liver function indicators may enhance mortality prediction accuracy.

Other liver indices like FIB-4 and APRI also show associations with sepsis. Studies found that FIB-4 can predict poor sepsis prognosis [21], while elevated liver fibrosis scores are associated with higher in-hospital mortality in patients with sepsis-induced coagulopathy [28]. APRI is also effective for predicting sepsis outcomes [29]. However, compared with these indices, the Lok index includes INR values. When INR exceeds 1.5, it indicates coagulopathy and can be used to diagnose liver dysfunction during sepsis [30]. Earlier studies demonstrated that PT/INR is a relevant marker for estimating 30-day all-cause mortality in inpatients diagnosed with sepsis [31]. Therefore, the Lok index could serve as a more accurate predictor of mortality in patients with sepsis. From a biological perspective, an increase in INR is closely associated with adverse outcomes of sepsis. INR reflects the function of the extrinsic coagulation pathway, and its elevation indicates a decline in the liver’s ability to synthesize coagulation factors and vitamin K metabolism disorders. In sepsis, the systemic inflammatory response can lead to endothelial damage and activation of the coagulation system, subsequently causing a consumptive decrease in coagulation factors and impairment of the anticoagulation mechanism. This coagulation dysfunction promotes the formation of microthrombi and the progression of disseminated intravascular coagulation (DIC), causing disorders in the microcirculation of organs and ischemic damage, ultimately increasing the risk of multiple organ failure and death [32, 33]. This suggests that the Lok index may have greater clinical significance, but further prospective studies are needed to confirm its clinical value.

Subgroup analyses revealed significant interactions between the Lok index and AF and hypertension. The interaction between AF and Lok index was significant at 1 year (P < 0.05), with a stronger association in those without AF. New-onset AF patients often receive anticoagulants (warfarin or Direct Oral Anticoagulants) during hospitalization [34]. These drugs suppress microthrombi, reduces sepsis-related DIC [33], and improve microcirculation perfusion. However, AF itself often predicts worse sepsis outcomes [35, 36]. The opposing effects create a complex AF and Lok interaction, which needs prospective dissection. The interaction between hypertension and Lok index was significant at 90 days. This indicates that hypertension enhances the association between Lok index and mortality. Subgroup analyses revealed particularly intriguing interactions regarding AF and hypertension.

The absence of a significant association between the Lok index and 30-day mortality among patients with AF requires further examination. Patients with new-onset AF during sepsis often receive therapeutic anticoagulation. These anticoagulants mitigate microthrombus formation and improve microcirculatory perfusion. Such therapeutic effects may ameliorate sepsis-induced coagulopathy (SIC) and DIC—conditions partially reflected by the INR component of the Lok index. Consequently, this therapeutic modification could potentially attenuate the prognostic value of an elevated Lok index in the short term within the AF subgroup.

AF itself represents an established marker of disease severity and poor prognosis in sepsis. The inherent risk associated with AF contrasts with the potential benefits of anticoagulation therapy. This confluence of opposing effects likely underlies the complex interaction observed in the data. Future prospective studies that carefully account for anticoagulation status are essential to clarify this relationship.

The absence of significant associations in non-hypertensive patients across all time points, combined with significant interactions with hypertension at 30 and 90 days, suggests that hypertension functions as an effect modifier. Hypertension involves chronic endothelial dysfunction and persistent activation of the renin-angiotensin-aldosterone system (RAAS). This pathophysiological state may create conditions that synergistically amplify the risk indicated by the Lok index. In hypertensive patients, pre-existing vascular injury and dysregulated neurohormonal axes could be further compromised by the hepatic dysfunction and coagulopathy signaled by an elevated Lok index. This combination potentially leads to a multiplicative increase in mortality risk. Non-hypertensive patients lack this chronic “priming,” which may explain the weaker association between the Lok index and outcomes in this subgroup.

These subgroup findings demonstrate that the prognostic utility of the Lok index varies across different sepsis populations and depends on specific comorbidities and their treatments. Hypertension may synergistically amplify mortality risk through RAAS activation [37] or endothelial damage [38], creating a multiplicative effect with the Lok index.

Despite utilizing robust data from the MIMIC-IV cohort, the analysis is subject to several limitations. First, this retrospective study limits causal inference; prospective studies are needed for validation. Second, the MIMIC-IV database primarily contains data from a single medical center in the United States. Patient demographics, healthcare resources, and specific sepsis management protocols in this setting may differ substantially from those in other healthcare systems and geographical regions. These protocols include antibiotic stewardship, fluid resuscitation strategies, and organ support practices. Such differences could influence the performance of the Lok index for prognostic prediction in sepsis. Therefore, the generalizability of these findings to other populations and clinical settings requires careful consideration. Validation through future multi-center, multi-regional prospective studies is essential. Third, missing data for some variables like C-reactive protein may affect result accuracy. Fourth, a key limitation is the selection bias due to the exclusion of patients who lacked the Lok index. The included patients were in a more severe condition and had a higher mortality rate. This means that our findings regarding the prognostic value of the Lok index are mainly applicable to this high-risk population. Whether the Lok index predicts outcome in milder cases remains untested. Although multiple confounders were adjusted in Cox regression analysis, unmeasured confounders may still bias results. Consistent exclusion criteria were applied to avoid selective adjustment, but these unmeasured factors may nevertheless have important impacts on sepsis patient prognosis. Additionally, while this study assessed mortality at 30, 90, and 365 days, long-term follow-up data after discharge were limited. Future studies should evaluate long-term outcomes including quality of life, recurrence, and complications to more comprehensively assess Lok index prognostic value. In addition, several potential biases inherent to the MIMIC-IV database should be considered. First, as a single-center database despite its large sample size, the generalizability of our findings may be limited to institutions with similar patient demographics and clinical practices. Second, misclassification or documentation errors in electronic health records could affect the accuracy of variables such as comorbidities, laboratory values, or medication records. Although we used structured data fields and applied standardized definitions to minimize such errors, residual bias may remain. Finally, unmeasured confounding due to variables not routinely captured in MIMIC-IV—such as detailed treatment protocols and socioeconomic factors—may also influence the observed associations. Besides, our study period spans the introduction of the Sepsis-3 criteria in 2016. As our cohort is defined using ICD-9 and ICD-10 codes rather than clinical SOFA scores, there is a potential for misclassification bias, particularly for patients admitted prior to 2016. Although the MIMIC-IV database applies consistent coding practices, evolving diagnostic trends and treatment standards over time could theoretically influence our findings. However, the stability of our effect estimates across multiple sensitivity analyses and the inclusion of detailed severity-of-illness scores (SOFA, APACHE III, etc.) in our models, which capture the core physiologic elements of sepsis, provide some reassurance that the identified association between the Lok index and mortality is robust.

Although existing studies show high accuracy of Lok index in predicting cirrhosis complications, its application in predicting all-cause mortality needs further validation. Future prospective studies should verify Lok index effectiveness across diverse populations to better evaluate its clinical reliability. Strategic combination of Lok index with complementary biomarkers may significantly improve prediction accuracy and sensitivity.

Conclusion

In conclusion, this retrospective observational study suggests that the Lok index may serve as an independent prognostic indicator for both short- and long-term mortality in critically ill patients with sepsis, exhibiting a nonlinear relationship with mortality risk. These results support further investigation into its potential role in identifying high-risk patients and informing clinical assessment. However, future prospective and multicenter studies are essential to validate these findings and determine the generalizability of the Lok index before any clinical application can be considered.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

FIB-4

Fibrosis-4 Index

INR

International Normalized Ratio

MIMIC-IV

Medical Information Database for Intensive Care IV

RCS

Restricted cubic splines

AF

Atrial fibrillation

HR

Hazard ratio

SQL

Structured Query Language

MICE

Multiple imputation by chained equations

WBC

White blood cell count

RBC

Red blood cell count

FBG

Fasting blood glucose

GCS

Glasgow Coma Scale

K-M

Kaplan–Meier

PAI-1

Plasminogen Activator Inhibitor-1

ADM

Active adrenomedullin

RAAS

Renin-angiotensin-aldosterone system

Author contributions

All authors contributed to the study conception and design. Writing - original draft preparation: [Qin Qin]; Writing - review and editing: [Qin Qin]; Conceptualization: [Qin Qin]; Methodology: [Wentao Ye]; Formal analysis and investigation: [Tianyang He]; Resources: [Lisha Xiang]; Supervision: [Qin Qin], and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript.

Funding

The authors declare that they did not receive any funding from any source.

Data availability

The datasets analysed during the current study are available from the MIMIC-IV v3.1 database.

Declarations

Ethics approval and consent to participate

The MlMIC-IV database was approved by the Massachusetts Institute of Technology (Cambridge, MA) and Beth Israel Deaconess Medical Center (Boston, MA), and consent was obtained for the collection of original data. All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and national research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards. Therefore, ethical approval and the need for informed consent were waived for this study.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.