The relationship between hemoglobin, albumin, lymphocyte, and platelet (HALP) score and 28-day mortality in patients with sepsis: a retrospective analysis of the MIMIC-IV database

Abstract

Background

In recent years, the hemoglobin, albumin, lymphocyte, and platelet (HALP) score has emerged as a potential marker of immunological and nutritional status. This study aimed to evaluate the association between the HALP score and prognosis in patients with sepsis.

Methods

This retrospective cohort study analyzed sepsis patients using clinical data from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database. Patients were classified into Low-score and High-score groups. Confounding factors were controlled through propensity score matching (PSM) analysis. The primary outcome was 28-day mortality in individuals with sepsis. Survival probabilities between groups were compared using Kaplan-Meier curves. Multivariable Cox regression analysis and a smoothing spline fitting curve were employed to investigate the relationship between the HALP score and 28-day mortality. ROC curve analysis and subgroup analysis were performed to evaluate the predictive ability of the HALP score and its components.

Results

A total of 2,968 sepsis patients were included, with 809 (27.26%) deaths within 28 days. After PSM analysis, the High-score group had a 24% lower risk of 28-day mortality compared to the Low-score group (HR, 0.76; 95% CI, 0.64–0.91). In the unmatched cohort, the multivariable Cox regression model also indicated that the High-score group had a lower 28-day mortality risk (HR, 0.78; 95% CI, 0.67–0.91). The smoothing spline fitting curve showed a nonlinear relationship between the HALP score and 28-day mortality, with an inflection point at 24.69. When the HALP score was below 24.69, an increase of one point in the HALP score was associated with a 2% reduction in 28-day mortality (HR, 0.98; 95% CI, 0.97–0.99). The HALP score provided incremental predictive value for 28-day mortality when combined with the SOFA score. Albumin was identified as the most influential component of the HALP score.

Conclusion

Among patients with sepsis, the HALP score exhibited a nonlinear relationship with 28-day mortality. An elevated HALP score is associated with reduced 28-day, 90-day, 360-day, and in-hospital mortality among sepsis patients.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12879-025-10739-3.

Introduction

Sepsis is a serious health condition resulting from a dysregulated host response to an infectious agent, ultimately leading to organ dysfunction [1]. This definition underscores that the harm from sepsis arises not only from the pathogenic infection itself but also from the secondary damage caused by the host’s immune response to the pathogen. A global burden of disease study indicated that sepsis affected approximately 50 million individuals worldwide in 2017, resulting in11 million fatalities, accounting for nearly 20% of global deaths [2]. Although the mortality rate has declined in recent years, sepsis remains a substantial global health threat, with the World Health Assembly (WHA) recognizing it as a major threat to human health [3]. Therefore, identifying new prognostic markers for sepsis and implementing timely interventions to improve patient survival rates are crucial.

Sepsis involves complex inflammatory and immune responses. Several novel circulating inflammatory markers, such as the platelet-lymphocyte (PLR) ratio [4], the neutrophil-to-lymphocyte (NLR) ratio, neutrophil-platelet (NPR) ratio [5], neutrophil/lymphocyte to-platelet (N/LP) ratio [6], and the CRP-albumin-lymphocyte (CALLY) index [7], have shown potential prognostic value in sepsis.

The HALP score, which integrates hemoglobin, albumin, lymphocyte, and platelet levels, has emerged as a novel marker of immune-inflammatory conditions and nutritional status. Chen et al. [8] revealed that the HALP score was initially explored for its predictive value in individuals diagnosed with gastric cancer. The HALP score has also shown prognostic significance in various cancers, such as squamous cell carcinoma [9], endometrial cancer [10], and breast cancer [11]. Additionally, the HALP score has demonstrated favorable prognostic value in diseases such as myelodysplastic syndrome [12], diabetic retinopathy [13], heart failure [14], and dyslipidemia [15].

However, the prognostic role of an enhanced HALP score, which incorporates more comprehensive indicators, has not yet been explored in sepsis patients. Considering that sepsis patients frequently experience malnutrition, anemia, and thrombocytopenia [16–18], we hypothesize that the HALP score could be useful for this population. Therefore, this study aimed to assess the prognostic value of the HALP score for predicting 28-day mortality and other adverse outcomes in individuals with sepsis.

Methods

Data sources

Information for this research was drawn from the Medical Information Mart for Intensive Care IV database, version 2.2 (MIMIC-IV 2.2) [19]. The development of the MIMIC-IV was a joint project involving the Massachusetts Institute of Technology (MIT) and Beth Israel Deaconess Medical Center (BIDMC). As a free and publicly available database, it contains more than 50,000 ICU admissions from the BIDMC from 2008 to 2019. This dataset contains a wealth of information, including patients’ basic information, vital signs, laboratory measurements, diagnoses, orders, procedures, treatments, and other relevant data. The Institutional Review Board of both BIDMC and MIT reviewed and authorized the project, providing a waiver of informed consent. Huan Li, the lead author, secured access to the database by completing the Collaborative Institutional Training Initiative (CITI) program and relevant examinations (Record ID: 59871727).

Study population

We included septic patients admitted to the ICU for the first time, whose hemoglobin, albumin, lymphocyte, and platelet counts were recorded. Sepsis was diagnosed based on presumed or verified infection, along with a SOFA (sequential organ failure assessment) score of 2 points or higher [1]. The study excluded participants who met any of these criteria: (1) under 18 years of age; (2) ICU stay of fewer than 48 h; (3) pregnant or breastfeeding; and (4) no sepsis diagnosis on the initial day of ICU admission.

Variable selection

Data were extracted using PostgreSQL software (version 12.19.2) and Navicat Premium software (version 15) through structured query language (SQL). These data included baseline demographic variables, vital signs, comorbidities, disease scores, therapeutic interventions, and laboratory tests. The baseline demographic variables included gender, body mass index (BMI), age, and race. The vital signs included the mean arterial pressure (MAP), heart rate, and respiratory rate (RR). The comorbidities included cerebrovascular disease, peripheral vascular disease, renal disease, congestive heart failure, malignant cancer, diabetes, chronic pulmonary disease, and liver disease. Disease scores included the Charlson Comorbidity Index (CCI), Sequential Organ Failure Assessment (SOFA), Glasgow Coma Scale (GCS), and the Simplified Acute Physiology Score II (SAPS II). Therapeutic interventions included mechanical ventilation, vasopressor use, and renal replacement treatment (RRT). Vasopressor use was identified as any application of vasopressin, dopamine, dobutamine, norepinephrine, or epinephrine throughout the ICU stay. The laboratory tests included hemoglobin (Hb), calcium, white blood cell count (WBC), chloride, albumin, hematocrit (HCT), platelet count, absolute counts of neutrophil (Abs_Neutrophils) and lymphocyte (Abs_Lymphocytes), red blood cell count (RBC), creatinine, blood urea nitrogen (BUN), potassium, glucose, sodium, prothrombin time (PT), activated partial thromboplastin time (APTT), international normalized ratio (INR), partial pressure of oxygen (PO₂), potential of hydrogen (pH), oxygen saturation (SpO₂), partial pressure of carbon dioxide (PCO₂), and lactate. The indicators above were collected during the initial 24 h of ICU admission. For major variables (albumin, hemoglobin, platelet count, and lymphocyte absolute count), the first recorded value was used. For the remaining indicators, the worst value that was most relevant to the severity of the disease was taken when there were multiple recordings during the initial 24 h following ICU admission. In addition, we included the total fluid input and total fluid output during the first 24 h of ICU admission, as well as the lactate levels at ICU admission (lactate H0) and 24 h after ICU admission (lactate H24). Fluid balance was calculated as the total fluid input minus the total fluid output. The lactate clearance was derived using the following formula: [(lactate H0 – lactate H24) / lactate H0] × 100%. Fluid balance and lactate clearance were used to reflect the patient’s response to treatment. The formula provided was used to determine the HALP score [8]: hemoglobin (g/L) × albumin (g/L) × lymphocyte (/L) / platelet (/L). Supplementary Table S1 details the criteria for selecting values when multiple measurements were available. Variables exhibiting over 50% missing values were removed, while the missing values of the rest of the variables were imputed with means (consistent with a normal distribution) or medians (consistent with an abnormal distribution).

Outcomes

The principal outcome, 28-day mortality, was defined as the interval between ICU admission and death. Secondary outcomes included mortality rates at 90 days and 360 days, in-hospital mortality, in-ICU mortality, and the lengths of hospital and ICU stays (both measured in days).

Statistical analysis

Continuous variables are presented as mean ± standard deviation (SD) for normally distributed data and as median [interquartile range (IQR)] for non-normally distributed data. Categorical variables are presented as counts and percentages (%). To compare group differences, categorical variables were analyzed using Fisher’s exact test or chi-square test, while continuous variables were evaluated using Student’s t-test (for normally distributed data) or the Mann–Whitney U test (for non-normally distributed data).

In light of previous studies that excluded variables with more than 60% missing data [20], we set a more stringent criterion and excluded variables with over 50% missing values. The remaining variables with missing data, including lactate (16.64%), lactate H0 (16.6%), lactate H24 (29.1%), pH (11.46%), PO₂ (11.46%), PCO₂ (11.46%), APTT (1.92%), INR (1.72%), PT (1.72%), calcium (0.91%), absolute counts of neutrophil (0.44%), glucose (0.07%) and GCS (0.03%), were imputed by medians, and BMI (31.00%) was imputed by means.

The baseline characteristics between 28-day survivors and non-survivors were compared. The optimal cutoff value was determined using X-tile software, and patients were divided into Low-score (< 13.5) and High-score groups (≥ 13.5). Kaplan‒Meier survival curves and the log-rank test were used to assess survival differences among the two groups. Propensity score matching (PSM) analysis was performed to minimize baseline differences between the two groups, using one-to-one matching with a caliper width of 0.01 to ensure precise matching. The effectiveness of PSM was assessed using the standardized mean difference (SMD), with SMD values < 0.1 indicating no significant differences [21]. Baseline characteristics and primary and secondary outcomes were compared between the pre-matched and post-matched cohorts.

To assess multicollinearity among variables, the variance inflation factor (VIF) was utilized, with a value of VIF ≥ 5 signifying the existence of multicollinearity [22]. Univariate regression analysis was conducted to explore the associations between individual variables and 28-day mortality. The independent association between the HALP score and 28-day mortality in septic patients were evaluated using multivariable Cox regression models. Three models were constructed: Non-adjusted; Adjust I, which was adjusted for gender, age, race, and BMI; Adjust II, which was adjusted for gender, age, race, BMI, and additional covariates that exhibited a significant correlation with 28-day mortality (P < 0.1); or those that changed the estimate by more than 10% [23]. A two-segment linear regression model and logarithmic likelihood ratio test (LRT), which were based on smoothing spline fitting curve, were applied to evaluate the nonlinear association and threshold effect between the HALP score and 28-day mortality. To evaluate the prognostic value of the HALP score and its components (hemoglobin, albumin, lymphocytes, and platelets), we performed ROC curve analysis to calculate the area under the curve (AUC) values and compared them using the DeLong-test. We also assessed the incremental predictive value of the HALP score by combining it with the SOFA score, and similarly evaluated albumin and lymphocyte count in combination with SOFA. Additionally, we conducted a multivariable Cox regression analysis, adjusting for confounding factors, to identify the key components contributing to the HALP score’s prognostic ability. Finally, we performed a subgroup analysis in sepsis patients to compare cancer and non-cancer patients.

Statistical significance was defined as a p-value less than 0.05 in a two-tailed test. Data analyses were performed using R software (version 4.4.0) and EmpowerStats software (version 6.0) (http://www.empowerstats.com/cn/, X&Y solutions, Inc., Boston, MA).

Results

Included sepsis participants

According to the Sepsis-3.0 diagnostic criteria, 3952 first-time ICU-admitted septic patients with recorded hemoglobin, albumin, lymphocyte, and platelet counts were included. We excluded 793 patients who stayed in the ICU for less than 48 h, 17 patients who were pregnant or breastfeeding, and 174 patients who were not diagnosed with sepsis on the first day of ICU admission. Ultimately, the analysis included 2,968 septic patients, with 2,159 survivors and 809 non-survivors within 28 days (Fig. 1).

Fig. 1.

Flow chart of the study

Baseline characteristics

We compared the baseline characteristics between the 28-day survivor and non-survivor groups (Table 1). Compared to the 28-day non-survivor group, patients in the 28-day survivor group had higher level of HALP scores [15.91 (8.43–29.01) vs. 13.04 (6.90-26.08)], BMI, GCS, MAP, RBC, HCT, Hb, Abs_Lymphocytes, platelets, albumin, chloride, pH and PO₂, and lower level of age, SAPS II, SOFA, CCI, RR, WBC, Abs_Neutrophils, BUN, creatinine, glucose, potassium, INR, PT, APTT, lactate, the proportion of comorbidities (congestive heart failure, cerebrovascular disease, liver disease, renal disease, malignant cancer), and the proportion of interventions (RRT, vasopressor use).

Table 1.

Baseline demographic and clinical characteristics of total cohort, 28-day survivors, and 28-day non-survivors

Variables	Total (n = 2968)	28-day survivor(n = 2159)	28-day non-survivor (n = 809)	P-value
Demographic variables
Age, year	64.06 ± 16.62	62.33 ± 16.80	68.67 ± 15.21	< 0.001*
Gender				0.612
Male	1702 (57.35%)	1232 (57.06%)	470 (58.10%)
Female	1266 (42.65%)	927 (42.94%)	339 (41.90%)
Race				0.001*
White	1822 (61.39%)	1354 (62.71%)	468 (57.85%)
Black	244 (8.22%)	188 (8.71%)	56 (6.92%)
Others	902 (30.39%)	617 (28.58%)	285 (35.23%)
BMI, kg/m2	29.63 ± 7.05	29.84 ± 7.02	29.07 ± 7.12	0.008*
Score system
SAPSII	44.76 ± 15.07	42.09 ± 14.22	51.89 ± 14.96	< 0.001*
SOFA	7.00 (5.00–10.00)	7.00 (4.00–10.00)	9.00 (6.00–12.00)	< 0.001*
CCI	5.00 (3.00–7.00)	5.00 (3.00–7.00)	6.00 (4.00–9.00)	< 0.001*
GCS	15.00 (13.00–15.00)	15.00 (13.00–15.00)	15.00 (12.00–15.00)	< 0.001*
Comorbidities
Congestive heart failure	1011 (34.06%)	711 (32.93%)	300 (37.08%)	0.034*
Peripheral vascular disease	308 (10.38%)	214 (9.91%)	94 (11.62%)	0.174
Cerebrovascular disease	394 (13.27%)	263 (12.18%)	131 (16.19%)	0.004*
Chronic pulmonary disease	823 (27.73%)	589 (27.28%)	234 (28.92%)	0.373
Liver disease	754 (25.40%)	493 (22.83%)	261 (32.26%)	< 0.001*
Diabetes	893 (30.09%)	648 (30.01%)	245 (30.28%)	0.886
Renal disease	696 (23.45%)	471 (21.82%)	225 (27.81%)	< 0.001*
Malignant cancer	423 (14.25%)	247 (11.44%)	176 (21.76%)	< 0.001*
Vital Signs
Heart rate, b/min	111.65 ± 22.75	111.54 ± 22.52	111.95 ± 23.37	0.662
MAP, mmHg	54.88 ± 14.01	55.78 ± 13.65	52.48 ± 14.65	< 0.001*
RR, b/min	29.86 ± 6.85	29.73 ± 6.93	30.21 ± 6.62	0.012*
Laboratory tests
RBC, 109/L	3.25 ± 0.76	3.28 ± 0.74	3.16 ± 0.80	< 0.001*
HCT	29.32 ± 6.69	29.50 ± 6.63	28.87 ± 6.84	0.023*
Hb, g/dL	10.56 ± 2.36	10.67 ± 2.36	10.29 ± 2.37	< 0.001*
WBC, 109/L	14.85 (10.20–20.70)	14.50 (10.10–20.20)	15.80 (10.30–22.10)	< 0.001*
Abs_Neutrophils, 109/L	11.04 (6.99–16.47)	10.79 (6.80-15.93)	12.18 (7.69–17.76)	< 0.001*
Abs_Lymphocytes, 109/L	0.90 (0.52–1.42)	0.95 (0.56–1.48)	0.78 (0.43–1.27)	< 0.001*
Platelets, 109/L	185.00 (122.00-262.00)	190.00 (127.00-263.00)	176.00 (108.00-257.00)	0.006*
Albumin, g/dL	2.92 ± 0.64	2.97 ± 0.62	2.81 ± 0.70	< 0.001*
BUN, mg/dL	29.00 (18.00–49.00)	27.00 (17.00–43.00)	39.00 (24.00–60.00)	< 0.001*
Calcium, mg/dL	7.68 ± 0.94	7.69 ± 0.92	7.66 ± 0.98	0.453
Chloride, mEq/L	106.42 ± 7.33	106.62 ± 7.01	105.90 ± 8.11	0.018*
Creatinine, mg/dL	1.40 (0.90–2.40)	1.30 (0.90–2.25)	1.70 (1.10–2.80)	< 0.001*
Glucose, mg/dL	158.00 (124.00-219.00)	156.00 (123.00-210.50)	169.00 (127.00-238.00)	0.013*
Sodium, mEq/L	140.36 ± 5.88	140.32 ± 5.47	140.48 ± 6.85	0.516
Potassium, mEq/L	4.50 (4.10–5.10)	4.40 (4.10-5.00)	4.60 (4.20–5.30)	< 0.001*
INR	1.84 ± 1.45	1.70 ± 1.27	2.20 ± 1.80	< 0.001*
PT, s	19.95 ± 14.65	18.60 ± 12.99	23.53 ± 17.88	< 0.001*
APTT, s	49.03 ± 33.08	46.67 ± 31.22	55.33 ± 36.90	< 0.001*
Lactate, mmol/L	2.40 (1.70–3.80)	2.40 (1.60–3.40)	2.50 (1.90–5.10)	< 0.001*
pH	7.28 ± 0.12	7.29 ± 0.11	7.27 ± 0.14	< 0.001*
PO2, mmHg	57.00 (41.00–81.00)	57.00 (42.00-83.50)	55.00 (38.00–75.00)	< 0.001*
PCO2, mmHg	46.00 (40.00–54.00)	46.00 (40.00–53.00)	46.00 (39.00–56.00)	0.279
Interventions
RRT	164 (5.53%)	108 (5.00%)	56 (6.92%)	0.042*
Ventilation	2324 (78.30%)	1695 (78.51%)	629 (77.75%)	0.655
Vasopressor	1414 (47.64%)	964 (44.65%)	450 (55.62%)	< 0.001*
Lactate H0, mmol/L	1.90 (1.40–2.80)	1.90 (1.30–2.50)	2.10 (1.70–3.70)	< 0.001*
Lactate H24, mmol/L	1.60 (1.30-2.00)	1.60 (1.20–1.70)	1.70 (1.40–2.90)	< 0.001*
Lactate clearance, %	15.79 (-14.29-40.00)	15.79 (-14.29-38.55)	15.79 (-21.05-40.91)	0.087
Total input, (ml/kg/24 h)	99.96 (54.89-168.05)	99.21 (54.37-167.95)	101.86 (56.96-168.19)	0.716
Total output, (ml/kg/24 h)	23.39 (12.92–38.44)	25.14 (14.60-40.25)	17.68 (8.91–31.89)	< 0.001*
Fluid balance, (ml/kg/24 h)	73.00 (29.56-137.39)	70.76 (28.43-134.11)	79.23 (34.45-142.78)	0.009*
HALP score	15.10 (8.01–28.20)	15.91 (8.43–29.01)	13.04 (6.90-26.08)	0.001*

Data: mean ± SD (standard deviation), median (Q1–Q3) or N (%). BMI, body mass index; SAPS II, Simplified Acute Physiology Score II; SOFA, Sequential Organ Failure Assessment; CCI, Charlson Comorbidity Index; GCS, Glasgow Coma Scale; MAP, mean arterial pressure; RR, respiratory rate; RBC, red blood cell count; HCT, hematocrit; Hb, hemoglobin; WBC, white blood cell count; Abs_Neutrophils, absolute counts of neutrophil; Abs_Lymphocytes, absolute counts of lymphocyte; BUN, blood urea nitrogen; INR, international normalized ratio; PT, prothrombin time; APTT, activated partial thromboplastin time; pH, potential of hydrogen; PO₂, partial pressure of oxygen; PCO₂, partial pressure of carbon dioxide; RRT, renal replacement treatment; HOSP-mortality, in-hospital mortality; ICU-mortality, in-ICU mortality; HOSP_LOS, hospital length of stay; ICU_LOS, ICU length of stay

The optimal cutoff value (HALP score = 13.5) was determined using X-tile software (Fig. 2A, B). Patients were divided into Low-score (< 13.5) and High-score groups (≥ 13.5), and survival probabilities were assessed using Kaplan-Meier curves (Fig. 2C).

Fig. 2.

The optimal cutoff value. (A) Histograms. (B) X-tile plots. (C) Kaplan-Meier curves for 28-day survival of two groups in pre-matched cohort

After one-to-one propensity score matching, each group contained 949 patients. Table 2 presented the baseline characteristics of the Low-score and High-score groups in the pre-matched and post-matched cohorts. Compared to the pre-matched cohort, no statistically significant differences in baseline characteristics were observed after matching (all SMDs < 0.1).

Table 2.

Baseline demographic and clinical characteristics of Low-score and High-score groups in the pre-matched and post-matched cohorts

Variable	Before match				After match
	Low-Score (< 13.5) n = 1333	High-Score (≥ 13.5), n = 1635	P-value	SMD	Low-Score (< 13.5), n = 949	High-Score (≥ 13.5), n = 949	P-value	SMD
Baseline variables
Age, year	66.36 ± 15.64	62.18 ± 17.15	< 0.001	0.25	64.67 ± 16.04	64.85 ± 16.53	0.828	0.01
Gender			< 0.001	0.16			0.678	0.02
Male	706 (52.96%)	996 (60.92%)			518 (54.58%)	509 (53.64%)
Female	627 (47.04%)	639 (39.08%)			431 (45.42%)	440 (46.37%)
RACE			0.025	0.10			0.456	0.06
White	853 (63.99%)	969 (59.27%)			597 (62.91%)	602 (63.44%)
Black	107 (8.03%)	137 (8.38%)			67 (7.06%)	77 (8.11%)
Others	373 (27.98%)	529 (32.35%)			285 (30.03%)	270 (28.45%)
BMI, kg/m2	29.63 (25.56–29.72)	29.63 (26.21–31.02)	0.024	0.08	29.63 (25.68–30.74)	29.63 (25.71–30.84)	0.737	0.02
Score system
SAPS II	46.71 ± 14.73	43.17 ± 15.16	< 0.001	0.24	45.01 ± 14.34	44.60 ± 15.19	0.546	0.03
SOFA	7.00 (5.00–10.00)	7.00 (5.00–11.00)	0.001	0.12	7.00 (5.00–10.00)	7.00 (5.00–10.00)	0.834	0.01
CCI	6.00 (3.00–8.00)	5.00 (3.00–7.00)	< 0.001	0.26	5.00 (3.00–8.00)	5.00 (3.00–7.00)	0.994	0.00
GCS	15.00 (13.00–15.00)	15.00 (13.00–15.00)	0.155	0.05	15.00 (13.00–15.00)	15.00 (13.00–15.00)	0.671	0.02
Comorbidities
Congestive heart failure	490 (36.76%)	521 (31.87%)	0.005	0.10	329 (34.67%)	329 (34.67%)	1.000	0.00
Peripheral vascular disease	141 (10.58%)	167 (10.21%)	0.747	0.01	101 (10.64%)	95 (10.01%)	0.651	0.02
Cerebrovascular disease	143 (10.73%)	251 (15.35%)	< 0.001	0.14	120 (12.65%)	118 (12.43%)	0.890	0.01
Chronic pulmonary disease	409 (30.68%)	414 (25.32%)	0.001	0.12	267 (28.14%)	275 (28.98%)	0.684	0.02
Liver disease	289 (21.68%)	465 (28.44%)	< 0.001	0.16	228 (24.03%)	237 (24.97%)	0.631	0.02
Diabetes	403 (30.23%)	490 (29.97%)	0.876	0.01	294 (30.98%)	289 (30.45%)	0.804	0.01
Renal disease	349 (26.18%)	347 (21.22%)	0.002	0.12	225 (23.71%)	219 (23.08%)	0.745	0.02
Malignant cancer	260 (19.50%)	163 (9.97%)	< 0.001	0.27	139 (14.65%)	125 (13.17%)	0.353	0.04
Vital Signs
Heart rate, b/min	113.11 ± 22.75	110.47 ± 22.69	0.002	0.12	112.08 ± 22.36	111.65 ± 23.51	0.688	0.02
MAP	54.27 ± 13.06	55.38 ± 14.72	0.033	0.08	54.83 ± 13.53	54.73 ± 13.98	0.876	0.01
RR, b/min	30.14 ± 6.83	29.64 ± 6.85	0.049	0.07	30.02 ± 6.86	29.94 ± 6.81	0.792	0.01
Laboratory tests
RBC, 109/L	3.07 ± 0.68	3.39 ± 0.80	< 0.001	0.42	3.17 ± 0.69	3.20 ± 0.74	0.424	0.04
HCT	27.71 ± 5.91	30.64 ± 7.00	< 0.001	0.45	28.65 ± 5.94	28.84 ± 6.47	0.525	0.03
WBC, 109/L	14.60 (9.80–20.60)	15.00 (10.40–20.70)	0.225	0.05	14.70 (10.20–20.60)	14.50 (10.10–20.20)	0.451	0.04
Abs_Neutrophils, 109/L	11.53 (7.17–17.17)	10.82 (6.96–15.95)	0.051	0.07	11.48 (7.04–17.17)	10.78 (7.04–15.98)	0.433	0.04
BUN, mg/dL	32.00 (19.00–54.00)	27.00 (17.00–45.00)	< 0.001	0.19	30.00 (19.00–52.00)	29.00 (18.00–48.00)	0.550	0.03
Calcium, mg/dL	7.58 ± 0.93	7.77 ± 0.93	< 0.001	0.20	7.67 ± 0.95	7.66 ± 0.96	0.899	0.01
Chloride, mEq/L	106.00 ± 7.11	106.77 ± 7.49	0.005	0.10	106.30 ± 7.03	106.31 ± 7.63	0.975	0.01
Creatinine, mg/dL	1.40 (0.90–2.60)	1.40 (0.90–2.30)	0.010	0.09	1.40 (0.90–2.50)	1.30 (0.90–2.30)	0.343	0.04
Glucose, mg/dL	160.00 (125.00-218.00)	158.00 (123.00-220.00)	0.819	0.01	158.00 (125.00-220.00)	157.00 (124.00-219.00)	0.686	0.02
Sodium, mEq/L	139.86 ± 5.63	140.77 ± 6.05	< 0.001	0.16	140.23 ± 5.58	140.21 ± 6.11	0.994	0.00
Potassium, mEq/L	4.50 (4.10–5.10)	4.50 (4.10–5.10)	0.200	0.05	4.50 (4.10–5.10)	4.50 (4.10–5.10)	0.732	0.02
INR	1.40 (1.20–1.90)	1.40 (1.20–1.80)	0.805	0.01	1.40 (1.20–1.90)	1.40 (1.20–1.80)	0.851	0.01
PT, s	15.50 (13.60–20.30)	15.50 (13.40-19.85)	0.618	0.02	15.50 (13.60–20.20)	15.50 (13.50–19.30)	0.935	0.00
APTT, s	35.50 (29.60–49.50)	35.50 (29.40–53.10)	0.383	0.03	35.50 (29.30–50.10)	35.00 (29.30–51.70)	0.866	0.01
Lactate, mmol/L	2.40 (1.50–3.50)	2.40 (1.80–4.10)	0.002	0.12	2.40 (1.50–3.80)	2.40 (1.70–3.60)	0.729	0.02
pH	7.28 ± 0.12	7.28 ± 0.12	0.676	0.02	7.28 ± 0.13	7.29 ± 0.12	0.658	0.02
PO2, mmHg	57.00 (40.00–79.00)	57.00 (42.00–82.00)	0.020	0.09	57.00 (41.00–82.00)	57.00 (42.00–82.00)	0.972	0.00
PCO2, mmHg	46.00 (40.00–54.00)	46.00 (40.00-53.50)	0.970	0.00	46.00 (40.00–54.00)	46.00 (40.00–53.00)	0.851	0.01
Interventions
RRT	77 (5.78%)	87 (5.32%)	0.589	0.02	57 (6.01%)	53 (5.59%)	0.694	0.02
Ventilation	1038 (77.87%)	1286 (78.65%)	0.606	0.02	760 (80.08%)	748 (78.82%))	0.495	0.03
Vasopressor	648 (48.61%)	766 (46.85%)	0.339	0.04	446 (47.00%)	455 (47.95%)	0.679	0.02
Lactate clearance, %	15.79 (-15.38-40.00)	15.79 (-14.29-39.82)	0.488	0.03	15.79 (-14.29-41.18)	157.9 (-14.29-38.46)	0.523	0.03
Fluid balance, (ml/kg/24 h)	78.52 (29.50-147.48)	69.13 (29.84-129.36)	0.012	0.09	74.22 (26.95-142.53)	72.86 (30.75-137.42)	0.593	0.02

Outcomes of the patients with sepsis in the Low-score and High-score groups

Table 3 presented the outcomes of the Low-score and High-score groups in the pre-matched and post-matched cohorts. Before matching, compared to the Low-score group, the risk of 28-day, 90-day, 360-day, in-hospital, and in-ICU mortality decreased by 27% (HR, 0.73; 95% CI, 0.64–0.84), 29%, 30%, 28%, and 22% in High-score group. After matching, the results remained consistent. Specifically, the 28-day (22.55% vs. 29.08%), 90-day (30.45% vs. 38.15%), 360-day (39.52% vs. 47.31%), and in-hospital (21.08% vs. 25.82%) mortality rates were lower in High-score group than in Low-score group. Compared to the Low-score group, the risk of 28-day, 90-day, 360-day, and in-hospital mortality decreased by 24% (HR, 0.74; 95% CI, 0.64–0.91), 23%, 21%, and 23% in High-score group.

Table 3.

Outcomes of the patients with sepsis in the Low-score and High-score groups

Variables	Low-Score	High-Score	HR/OR/β (95% CI)	P-value
	(< 13.5)	(≥ 13.5)
Pre-matched cohort	n  = 1333	n  = 1635
28-day mortality	421 (31.58%)	388 (23.73%)	0.73 (0.64, 0.84)	< 0.001*
90-day mortality	550 (41.26%))	504 (30.83%)	0.71 (0.63, 0.80)	< 0.001*
360-day mortality	680 (51.01%)	629 (38.47%)	0.70 (0.62, 0.78)	< 0.001*
HOSP-mortality	371 (27.83%)	354 (21.65%)	0.72 (0.61, 0.85)	< 0.001*
ICU-mortality	268 (20.11%)	267 (16.33%)	0.78 (0.64, 0.94)	0.008*
HOSP_LOS	13.00 (7.00–21.00)	11.00 (7.00–19.00)	-1.16 (-2.20, -0.13)	< 0.001*
ICU_LOS	5.00 (3.00–10.00)	5.00 (3.00–9.00)	-0.25 (-0.79, 0.30)	0.066
Post-matched cohort	n  = 949	n  = 949
28-day mortality	276 (29.08%)	214 (22.55%)	0.76 (0.64, 0.91)	0.003*
90-day mortality	362 (38.15%)	289 (30.45%)	0.77 (0.66, 0.89)	< 0.001*
360-day mortality	449 (47.31%)	375 (39.52%)	0.79 (0.69, 0.90)	< 0.001*
HOSP-mortality	245 (25.82%)	200 (21.08%)	0.77 (0.62, 0.95)	0.015*
ICU-mortality	174 (18.34%)	144 (15.17%)	0.80 (0.63, 1.02)	0.066
HOSP_LOS	13.00 (7.00–21.00)	11.00 (7.00–19.00)	-0.88 (-2.18, 0.41)	0.181
ICU_LOS	5.00 (3.00–10.00)	5.00 (3.00–9.00)	-0.62 (-1.29, 0.05)	0.068

HOSP-mortality, in-hospital mortality; ICU-mortality, in-ICU mortality; HOSP_LOS, the length of hospital stay; ICU_LOS, the length of ICU stay; HR, hazard ratio; OR, Odds Ratio; CI, confidence interval

Association between HALP score and 28-day mortality in the pre-matched cohort

Univariate Cox regression analysis revealed a negative correlation between the HALP score (HR, 0.99; 95% CI, 0.99-1.00) and 28-day mortality in septic patients. Additionally, BMI, GCS, MAP, RBC count, chloride, pH (×10), and PO₂ were similarly negatively correlated with 28-day mortality. In contrast, age, race (others), RRT, vasopressor use, congestive heart failure, cerebrovascular disease, liver disease, renal disease, malignant cancer, SAPS II, SOFA, CCI, WBC count, absolute neutrophil count, BUN, creatinine, glucose, potassium, INR, APTT, lactate, PCO₂, and fluid balance were positively correlated with 28-day mortality in sepsis patients. Detailed results of the univariate analysis are presented in Fig. 3.

Fig. 3.

Univariate Cox regression analysis to assess the risk factors for 28-day mortality in patients with sepsis

HR, hazard ratio; CI, confidence interval

An independent correlation between the HALP score and 28-day mortality in sepsis patients was identified through multivariable Cox regression analysis (Table 4). We established three models for testing. When the HALP score was analyzed as a continuous variable, the HALP score showed an inverse association with 28-day mortality in the Non-adjusted model, Adjust I model, and Adjust II model (HR, 0.99; 95% CI, 0.99-1.00). And the HALP score was dichotomized into Low-score (< 13.5) and High-score (≥ 13.5) groups. Compared to Low-score group, the risk of 28-day mortality decreased by 27%, 23% and 22% in the Non-adjusted, Adjust I model and Adjust II model, respectively. Similar trends were observed when the HALP score was analyzed as a categorical variable with quartiles, with the Q3 and Q4 groups showing reduced mortality risks compared to the Q1 group. Furthermore, when the HALP score quartiles were treated as a continuous variable, the trend test P-values were all less than 0.05, further supporting the robustness of the correlation.

Table 4.

Multivariable Cox regression of the HALP score with 28-day mortality in the pre-matched cohort

Exposure	Non-adjusted HR (95% CI)	P-value	Adjust I HR (95% CI)	P-value	Adjust II HR (95% CI)	P-value
HALP score as continuous	0.99 (0.99, 1.00)	0.003*	1.00 (0.99, 1.00)	0.022*	0.99 (0.99, 1.00)	0.015*
HALP score dichotomous
Low (< 13.5)	Ref		Ref		Ref
High (≥ 13.5)	0.73 (0.64, 0.84)	< 0.001*	0.77 (0.67, 0.89)	< 0.001*	0.78 (0.67, 0.91)	0.001*
HALP score quartile
Q1 (≤ 8.01)	Ref		Ref		Ref
Q2 (8.01–15.10)	0.86 (0.71, 1.03)	0.097	0.86 (0.71, 1.03)	0.105	0.94 (0.78, 1.14)	0.535
Q3 (15.10–28.20)	0.68 (0.56, 0.83)	< 0.001*	0.72 (0.59, 0.87)	< 0.001*	0.75 (0.61, 0.93)	0.007*
Q4 (≥ 28.20)	0.73 (0.60, 0.89)	0.002*	0.79 (0.65, 0.96)	0.016*	0.78 (0.63, 0.96)	0.019*
P-value for trend	0.99 (0.99, 1.00)	0.002	0.99 (0.99, 1.00)	0.022	0.99 (0.99, 1.00)	0.013

Adjust I: age, gender, race, BMI;

Adjust II: age, gender, race, BMI, RRT, vasopressor use, congestive heart failure, cerebrovascular disease, liver disease, renal disease, malignant cancer, SAPS II, SOFA, CCI, GCS, MAP, RR, RBC, WBC, Abs_Neutrophils, BUN, lactate, chloride, creatinine, glucose, potassium, INR, APTT, lactate clearance, fluid balance, pH, PO₂, and PCO₂

HR, hazard ratio; CI, confidence interval; Ref, reference

Nonlinear relationship between the HALP score and risk of 28-day mortality

We used a smoothing spline fitting curve to visually illustrate the relationship between the HALP score and 28-day mortality in patients with sepsis (Fig. 4). Additionally, we applied a two-segment linear regression model and found the inflection point of the HALP score at 24.69 after adjusting for confounders (Supplementary Table S2). And a P-value less than 0.05 from the logarithmic likelihood ratio test (LRT) confirmed the presence of a nonlinear relationship between the HALP score and 28-day mortality. Specifically, when the HALP score was < 24.69, each one-point increase in the score was associated with a 2.0% increase in the 28-day mortality risk in sepsis patients (HR, 0.98; 95% CI, 0.97–0.99).

Fig. 4.

Smoothing spline fitting curve of the HALP score for 28-day mortality in patients with sepsis

Adjusted variables: age, gender, race, BMI, RRT, vasopressor use, congestive heart failure, cerebrovascular disease, liver disease, renal disease, malignant cancer, SAPS II, SOFA, CCI, GCS, MAP, RR, RBC, WBC, Abs_Neutrophil, BUN, lactate, chloride, creatinine, glucose, potassium, INR, APTT, lactate clearance, fluid balance, pH, PO₂, and PCO₂

Prognostic value of the HALP score and its components

The AUC values for the HALP score and its components were relatively low (all < 0.60) for predicting 28-day, 90-day, 360-day, in-hospital, and ICU mortality, with no significant differences observed (Supplementary Figure S1, Table S3). For 28-day mortality, the addition of the HALP score significantly improved the predictive performance of the SOFA score (AUC increased from 0.634 to 0.646, P = 0.005) (Supplementary Figure S2). Similarly, the combination of SOFA with albumin or lymphocyte counts also significantly improved the predictive performance compared to SOFA alone, with AUC values comparable to that of the SOFA + HALP score combination (Table 5). In the multivariable Cox regression analysis, albumin and lymphocyte count were identified as the most effective predictors of outcomes, with albumin having a greater impact (Supplementary Table S4).

Table 5.

Area under curves (AUC) for SOFA and SOFA + HALP score in predicting 28-day mortality

Variables	AUC 95%CI	P-valuea	P-valueb
SOFA	0.634 (0.611,0.656)
SOFA + HALP score	0.646 (0.624,0.669)	0.005
SOFA + Hemoglobin	0.638 (0.615,0.660)	0.201	0.054
SOFA + Albumin	0.643 (0.621,0.666)	0.025	0.103
SOFA + Lymphocyte	0.643 (0.621,0.666)	0.028	0.427
SOFA + Platelet	0.635 (0.613,0.658)	0.202	0.008

P-value^a, P-values represent the statistical significance of the differences in AUC between each combined models and the SOFA score;

P-value^b, P-values represent the statistical significance of the differences in AUC between each combined indicator and the SOFA + HALP score

Subgroup analysis of HALP score in sepsis patients with and without cancer

In sepsis patients, cancer patients had significantly lower hemoglobin, albumin, and lymphocyte levels, resulting in lower HALP scores compared to non-cancer patients (Supplementary Table S5). After adjustment, the HALP score remained significantly associated with 28-day mortality in non-cancer patients (HR, 0.99; 95% CI, 0.99–1.00) but not in cancer patients (HR, 0.99; 95% CI, 0.99–1.01). The interaction test between cancer status and the HALP score was not significant (P > 0.05), indicating no substantial difference in predictive ability.

Discussion

In this retrospective study of 2,968 participants, we assessed the prognostic value of the HALP score in predicting adverse outcomes in sepsis patients. Patients were divided into Low-score and High-score groups based on the optimal cutoff value. In both the pre-matched and post-matched cohorts, the High-score group had significantly lower 28-day, 90-day, 360-day, and in-hospital mortality rates than the Low-score group. Additionally, smoothing spline fitting curve depicted a nonlinear correlation between the HALP score and 28-day mortality, with 24.69 identified as the threshold. ROC curve analysis demonstrated that the HALP score provided incremental predictive value when combined with SOFA. Among the components of the HALP score, albumin was found to be the most influential.

The HALP score, a noninvasive and cost-effective hematological marker, has been widely used for prognostic evaluation in various diseases. In cancer research, higher HALP scores have been associated with improved outcomes [8, 24]. Similarly, in non-cancer conditions, higher HALP scores have been linked to reduced mortality in coronary heart disease patients [25], as well as a lower incidence of diabetic retinopathy [13]. However, some studies have reported inconsistent results, such as an increased risk of dyslipidemia [15] and worse prognosis in myelodysplastic syndromes [12]. Consistent with most prior research, our study demonstrated that low HALP scores were associated with a higher risk of 28-day mortality in sepsis patients.

Traditional theory suggests that sepsis is a biphasic process, with the initial phase characterized by an excessive inflammatory response. As the disease progresses, immune cells undergo apoptosis through various pathways, resulting in a period of immunosuppression that significantly affects the prognosis of sepsis [26]. Our study demonstrated that the HALP score, which incorporates the levels of albumin, hemoglobin, platelets, and lymphocytes, serves as a novel prognostic marker for the nutritional and inflammatory immune status in sepsis patients. However, changes in the HALP score may reflect alterations in one or more of its components. We found that, among the components of the HALP score, albumin appears to be the most influential driver of its prognostic ability. Albumin plays a critical role in nutritional and immune functions [27]. A retrospective study found that lower albumin levels were related to increased mortality risk in sepsis patients [28]. The increased clearance rate in sepsis leads to reduced plasma albumin levels, weakening its immunomodulatory and endothelial protective effects [29]. However, the effectiveness of using albumin as a resuscitation fluid for treating sepsis remains controversial [30]. Lymphocytes, reflecting immune capacity, are often reduced in sepsis due to immunosuppression, with lymphopenia correlating with poorer outcomes [31]. Reductions in hemoglobin levels or platelet counts are also frequently associated with disease severity and poor prognosis in sepsis [32, 33].

Furthermore, our analysis demonstrated that the HALP score provides incremental predictive value when combined with the SOFA score, consistent with a previous study in critically ill patients [34]. This finding suggests that the HALP score, as a composite marker of nutritional and immune status, can complement existing prognostic tools like SOFA, highlighting great potential for clinical application.

Given that cancer patients often exhibit distinct values for albumin, hemoglobin, lymphocytes, and platelets, we performed a subgroup analysis. Although the association between the HALP score and 28-day mortality was significant in non-cancer sepsis patients but not in those with cancer, the effect sizes were consistent (HR < 1), and the interaction test was not significant. This lack of significance may be attributed to the smaller sample size of cancer patients (n = 423), and it does not negate the potential prognostic utility of the HALP score in this subgroup. Further studies with larger cohorts are needed to validate these findings.

Overall, the HALP score made up of hemoglobin, albumin, lymphocytes, and platelets, has demonstrated prognostic value in predicting outcomes for sepsis patients.

Nonetheless, our study has several limitations. First, as a retrospective observational study using a database, we cannot establish a causal relationship. The HALP score was calculated using data from the first 24 h after ICU admission, which may not accurately reflect the true baseline value. Although we adjusted for confounding factors, some potential confounders may remain unaccounted for, and the exclusion of patients with unavailable data could introduce bias. Additionally, the lack of a standardized cutoff value for the HALP score may contribute to heterogeneity across studies. Furthermore, due to missing data, we were unable to include lipid levels, which may be relevant to sepsis prognosis. Although sensitivity analyses using multiple imputation and random forest methods showed consistent results (Supplementary Table S6, Table S7, Table S8), imputation methods may still introduce potential bias. Future prospective studies with larger cohorts are needed to validate the prognostic value of the HALP score in sepsis patients.

Conclusion

Our study found that the HALP score is associated with short-term and long-term adverse outcomes in sepsis patients. Therefore, assessing patient status with the HALP score and providing timely interventions may contribute to improving the prognosis of sepsis patients.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Abbreviations

Abs_Lymphocytes

Absolute counts of lymphocyte;

Abs_Neutrophils

Absolute counts of neutrophil;

APTT

Activated partial thromboplastin time;

AUC

Area Under the Curve;

BIDMC

Beth Israel Deaconess Medical Center;

BMI

Body mass index;

BUN

Blood urea nitrogen;

CCI

Charlson Comorbidity Index;

CI

Confidence interval

GCS

Glasgow Coma Scale;

Hb

Hemoglobin;

HCT

Hematocrit;

HOSP-mortality

In-hospital mortality;

HOSP_LOS

The length of hospital stay;

HR

Hazard ratio;

ICU-mortality

In-ICU mortality;

ICU

Intensive care unit;

ICU_LOS

The length of ICU stay;

INR

International normalized ratio;

IQR

Interquartile range;

LRT

Logarithmic likelihood ratio test;

MAP

Mean arterial pressure;

MIMIC-IV

Medical Information Mart for Intensive Care-IV

MIT

Massachusetts Institute of Technology;

PCO2

Partial pressure of carbon dioxide;

pH

Potential of hydrogen;

PO2

Partial pressure of oxygen;

PSM

Propensity score matching;

PT

Prothrombin time;

RBC

Red blood cell count;

ROC

Receiver Operating Characteristic;

RR

Respiratory rate;

RRT

Renal replacement treatment;

SAPS

II: Simplified Acute Physiology Score II;

SD

Standard deviation;

SMD

Standardized mean difference;

SOFA

Sequential Organ Failure Assessment;

SQL

Structured Query Language;

VIF

Variance inflation factor;

WBC

White blood cell count;

Author contributions

Huan Li collected and assembled the data; Yiran Zhou and Xinying Zhang handled the software operation and conducted the statistical analysis; Run Yao was responsible for data analysis; Huan Li drafted the original manuscript; and Ning Li secured funding. All the authors endorsed the final manuscript.

Funding

This research was funded by the National Natural Science Foundation of China (No. 82270633), Hunan Provincial Natural Science Foundation of China (No.2022JJ70076), and Clinical Research Center Foundation of Xiangya Hospital (No. LN2021XYSX).

Data availability

The data used in this study were extracted from the publicly available MIMIC-IV 2.2 database (https://physionet.org/content/mimiciv/2.2/). Access to the database requires registration and successful completion of the Collaborative Institutional Training Initiative (CITI) program for data use certification. The data extraction codes are available in the MIMIC Code Repository (https://github.com/MIT-LCP/mimic-code/).

Declarations

Ethics approval and consent to participate

The dataset in this study was obtained from MIMIC-IV 2.2. The Institutional Review Board of both Beth Israel Deaconess Medical Center (BIDMC) and Massachusetts Institute of Technology (MIT) reviewed and authorized the project, providing a waiver of informed consent. As the data is publicly available and does not include identifiable patient information, informed patient consent was not required.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.