Association between thymosin α1 therapy and 28-day mortality risk in critically ill patients with tuberculosis-related sepsis: a retrospective cohort study

Abstract

Purpose

Patients with tuberculosis (TB)-related sepsis often experience profound immune dysregulation and poor clinical outcomes. Evidence of immunomodulatory therapeutic efficacy in this population remains limited. We evaluated the association between thymosin α1 (Tα1) use and 28-day all-cause mortality in critically ill TB-related sepsis patients and explored whether the strength of this association varied with the baseline lymphocyte count.

Methods

This single-centre retrospective cohort study included critically ill, adult TB-related sepsis patients admitted to the intensive care unit (ICU) in 2017–2025. Exposure was any Tα1 use during ICU stay. The primary outcome was 28-day mortality. Association analyses included multivariable Cox regression, restricted mean survival time (RMST) analyses, and exploratory restricted cubic spline (RCS) modelling. Sensitivity analyses were performed.

Results

Among 327 patients, 160 received Tα1 and 167 served as controls. Tα1 use was associated with a lower hazard of 28-day mortality after adjustment for baseline covariates (adjusted hazard ratio, 0.54; 95% confidence interval, 0.36–0.79; P = 0.002); sensitivity analyses showed estimates of similar direction. The proportional hazards assumption was violated; consequent RMST analysis revealed longer adjusted restricted mean survival in the Tα1 group. Exploratory RCS modelling suggested nonlinear variability in the association between Tα1 use and 28-day mortality across baseline lymphocyte counts, which appeared more pronounced at intermediate lymphocyte counts (approximately 0.8–1.2 × 10⁹/L); however, this data-driven pattern is hypothesis-generating.

Conclusion

Among ICU patients with TB-related sepsis, Tα1 use was associated with lower observed 28-day mortality, but causality cannot be inferred. The identified variability in this association across baseline lymphocyte counts requires external validation.

Clinical trial number

Not applicable.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12879-026-14132-6.

Introduction

Tuberculosis (TB) remains a formidable global health challenge; in 2024 alone, 10.7 million incident cases and 1.23 million deaths were reported [1]. Despite therapeutic advances, a critical group of patients—particularly those with delayed diagnoses or impaired host immunity—ultimately develop life-threatening organ dysfunction. These cases, defined as TB-related sepsis, represent a distinct clinical entity characterized by profound immune failure and a disproportionately high risk of short-term mortality [2–4]. According to the Sepsis-3 consensus, sepsis is a dysregulated host response to infection [5]; in addition to acute organ injury, sepsis is classically characterized by a state of “immunoparalysis,” in which the numerical and functional depletion of immune cells compromises pathogen clearance and increases patient susceptibility to secondary insults [6–8]. The immunological landscape of TB-related sepsis is unique and complex; active TB triggers chronic T-cell exhaustion, Th1/Th2 imbalance, and the upregulation of inhibitory checkpoints such as programmed cell death protein 1 (PD-1) [9–12]. When the acute immunosuppressive environment of sepsis is superimposed on this preexisting TB-mediated impairment, a “double-hit” effect is created, potentially leaving the host in a refractory state of persistent infection and progressive organ failure. Thymosin α1 (Tα1), a pleiotropic thymus-derived peptide, has emerged as a promising immunomodulator capable of promoting T-cell maturation and reversing exhaustion-associated phenotypes [13–15]. While early evidence has suggested that this compound provides benefits for various conditions characterized by immune dysregulation or exhaustion [16–19], the recent landmark Efficacy and Safety of Thymosin α1 for Sepsis (TESTS) trial (2025) reported no significant reduction in mortality in an unselected sepsis population [20]. These neutral results underscore the critical heterogeneity of sepsis and the need to identify specific immunological endotypes for which the associations between immunomodulatory therapy and clinical outcomes may differ. We hypothesize that Tα1 use is associated with lower mortality in patients with TB-related sepsis, in whom T-cell dysfunction is a dominant pathological driver. To date, however, evidence of the efficacy of Tα1 in this high-risk population remains sparse. Given the ease of accessibility of the lymphocyte count as a bedside surrogate for adaptive immune status, we conducted this retrospective cohort study to (1) evaluate the association between Tα1 use and 28-day all-cause mortality and (2) explore whether this association varied with the baseline lymphocyte count.

Materials and methods

Study design

This single-centre retrospective cohort study was conducted at the Chongqing Public Health Medical Center, China. The institutional ethics committee approved the study protocol (Approval No. 2025-028-01-KY) and waived the requirement for informed consent because of the retrospective nature of the data (details are provided in the Ethics Statement). We screened all adult patients (aged ≥ 18 years) admitted to the intensive care unit (ICU) between January 2017 and March 2025. Patients were eligible for inclusion if they had a confirmed diagnosis of TB and developed sepsis during their stay in the ICU. Sepsis was defined according to the Sepsis-3 criteria [5]. To ensure a robust assessment of baseline variables and minimize bias, the following exclusion criteria were applied: (1) age < 18 years; (2) pregnancy or puerperium; (3) missing primary outcome or essential timing data required for evaluating mortality; (4) ICU length of stay < 24 h, as such patients lack sufficient physiological and laboratory data for baseline severity scoring; and (5) preexisting immunodeficiency-related conditions (e.g., human immunodeficiency virus (HIV) infection, solid organ transplantation, or active chemotherapy), as these conditions could fundamentally confound the host’s immune response to TB and the subsequent effect of Tα1 treatment (Fig. 1). This study is reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines [21].

Fig. 1.

Study workflow showing patient selection in the ICU of Chongqing Public Health Medical Center (2017–2025), including patient exclusion and final allocation to the control group (n = 167) and the Tα1 group (n = 160) for patients with a 28-day follow-up

Definitions

TB was diagnosed in accordance with the ATS/CDC/IDSA clinical practice guidelines and the WHO consolidated TB diagnostic guidelines [22] alongside comprehensive clinical assessments [23]. Infection was identified on the basis of infection-related diagnoses documented in the initial ICU admission notes and standardized according to the International Classification of Diseases, Tenth Revision (ICD-10) codes [24]. Sepsis was defined according to the Sepsis-3 consensus as infection accompanied by life-threatening organ dysfunction [5]. Organ dysfunction, in turn, was defined as an acute increase in the Sequential Organ Failure Assessment (SOFA) score of ≥ 2 points [25]. The baseline SOFA score was defined as the SOFA score calculated using the most extreme physiological and laboratory values recorded within the first 24 h after ICU admission. Given the retrospective nature of this study, the preinfection baseline SOFA score was assumed to be 0 for patients without documented chronic organ dysfunction, in accordance with established methodologies used in large-scale database studies [26]. Accordingly, sepsis was operationally defined as a SOFA score ≥ 2 within the first 24 h of ICU admission in the presence of an infection. TB-related sepsis was thus operationally defined as sepsis according to Sepsis-3 criteria in patients with confirmed active TB according to microbiological evidence, imaging findings, and compatible clinical symptoms or signs. Attribution to active TB, rather than to concurrent non-TB infection alone, was further supported, when longitudinal clinical information was available, by improvements in clinical manifestations, haemodynamics, respiratory mechanics, and other organ dysfunction after anti-TB treatment. Concurrent bacterial or fungal infections were recorded separately and included as prespecified covariates in the adjusted association analyses. Tα1 exposure was defined as at least one documented subcutaneous or intramuscular administration of Tα1 during ICU stay. The control group comprised eligible patients from the same ICU screening cohort and study period who did not receive Tα1 during their ICU stay. Details on characteristics of Tα1 administration, including the regimen frequency, treatment duration, timing of initiation, and concomitant immunomodulatory therapy, are summarized in Table S8. To account for potential timing-dependent effects, Tα1 initiation was categorized as early (≤ 48 h) or late (> 48 h) according to the interval between ICU admission and the first dose.

Variable extraction

The clinical and laboratory data obtained within the initial 24 h of ICU admission—defined as the baseline observation window—were extracted from the electronic medical records (EMRs). The extracted variables included demographic and anthropometric characteristics (e.g., age, sex, and body mass index [BMI]), preexisting comorbidities, clinical parameters, and laboratory results. To quantify disease severity upon ICU entry, we calculated the Sequential Organ Failure Assessment (SOFA) and Acute Physiology and Chronic Health Evaluation II (APACHE II) scores using the most extreme physiological and laboratory values recorded during the first 24-hour baseline period.

Covariates

Potential confounders were prespecified on the basis of the published literature and clinical expertise [27, 28]. These covariates were categorized into four domains: (1) Demographics and comorbidities, including age, sex, body mass index (BMI, calculated as kg/m²), and preexisting chronic conditions (e.g., coronary heart disease, diabetes mellitus, hypertension, chronic obstructive pulmonary disease (COPD), chronic kidney disease (CKD), and malignancy); (2) TB-specific clinical characteristics, including the anatomical classification of TB, treatment history, and drug resistance profile(s). Specifically, TB was classified into four mutually exclusive categories: pulmonary TB, extrapulmonary TB, pulmonary plus extrapulmonary TB, and disseminated (miliary) TB [29]. To ensure diagnostic precision, disseminated TB was strictly defined as TB displaying haematogenous spread with miliary pulmonary involvement and multisite extrapulmonary involvement [30]; this category was prioritized during classification to avoid overlap with “pulmonary plus extrapulmonary TB.” Previously treated TB was defined as a history of initial treatment failure, recurrent bacteriological positivity after standard therapy, or irregular anti-TB chemotherapy for > 1 month [31]. Drug-resistant TB was defined as the presence of phenotypic or molecular evidence of resistance to at least one first-line anti-TB agent, with each case verified through consensus between two senior TB specialists [32]. (3) Features of non-TB infection, including the presence of nontuberculous pathogens and their respective antimicrobial resistance profiles. (4) Acute ICU interventions, including life-support measures initiated within the baseline period, such as mechanical ventilation, vasopressor support, and continuous renal replacement therapy (CRRT). Demographic data were retrieved from ICU registration records, whereas comorbidities and TB-related histories were obtained through a comprehensive review of past medical histories and prior clinical diagnoses.

Missing data

Baseline covariates with missing values (BMI, 11.0%; lymphocyte count, 0.9%) were addressed using the multiple imputation by chained equations (MICE) method to minimize selection bias. Missingness in the BMI was considered unrelated to illness severity and was attributed mainly to incomplete documentation inherent to the retrospective design of the study. Under a missing-at-random assumption, five imputed datasets were generated, incorporating all prespecified covariates and outcome variables as predictors. Sensitivity analyses were subsequently performed by pooling the results from these datasets using Rubin’s rules to confirm the stability of the primary multivariable model.

Outcomes

The primary outcome was the 28-day all-cause mortality after ICU admission. Follow-up started on the day of ICU admission (day 0) and continued until death or day 28, whichever occurred first; patients who were alive on day 28 were administratively censored on day 28. Secondary outcomes included in-hospital mortality (defined as death during the index hospitalization), the incidence of new-onset infections within 28 days, the clearance rate of pathogenic microorganisms within 28 days, the use of mechanical ventilation within 28 days, the use of CRRT within 28 days, the use of vasopressors within 28 days, the number of ventilator-free days (VFDs) within 28 days, the number of ICU-free days within 28 days, and the total length of hospital stay (days). Exploratory outcomes included the last available immune/inflammatory laboratory indices obtained before hospital discharge (or during hospitalization), including lymphocyte count, procalcitonin (PCT) level, and the neutrophil-to-lymphocyte ratio (NLR). The new-onset infection outcome was defined as specified in the footnote of Table 2; cultures were clinically driven, and adjudication was performed retrospectively and not formally blinded to Tα1 exposure.

Table 2.

Primary and secondary outcomes

Outcome	Thymosin α1 group (n = 160)	Control group (n = 167)	Effect estimate (95% CI)	P value
Primary outcome
28-day all-cause mortality Secondary outcomes	72 (45.0)	92 (55.1)	HR 0.54 (0.36–0.79)	0.002
In-hospital mortality	59 (36.9)	80 (47.9)	RR 0.69 (0.50–0.93)	0.02
New-onset infections < 28 days†	105 (65.6)	82 (49.1)	RR 1.30 (1.01–1.67)	0.04
Pathogenic microorganism clearance rate <28 days‡	61 (38.1)	57 (34.1)	RR 1.25 (0.90–1.74)	0.19
Mechanical ventilation use < 28 days	91 (56.9)	81 (48.5)	RR 1.08 (0.85–1.39)	0.52
CRRT use < 28 days	11 (6.9)	18 (10.8)	RR 0.74 (0.39–1.40)	0.35
Vasopressor use < 28 days	49 (30.6)	60 (35.9)	RR 0.68 (0.48–0.97)	0.03
Ventilator-free days < 28	19.5 (0.0–28.0)	4.0 (0.0–28.0)	—	0.11
ICU-free days < 28	8.0 (0.0–18.0)	0.0 (0.0–18.5)	—	0.28
Hospital length of stay (days)	19.5 (11.0–39.25)	14.0 (7.0–25.5)	MD 5.0 (2.0–8.0)	< 0.001
Lymphocyte count (×109/L)	0.56 (0.36–0.94)	0.88 (0.54–1.27)	MD -0.25 (− 0.36 to − 0.14)	< 0.001
PCT (ng/mL)	0.45 (0.13–1.83)	0.47 (0.15–2.84)	MD -0.03 (-0.15–0.06)	0.48
NLR	9.37 (4.03–19.15)	7.51 (4.67–16.00)	MD 0.21 (-1.54–2.10)	0.80

Abbreviations: ICU, intensive care unit; CRRT, continuous renal replacement therapy; PCT, procalcitonin; NLR, neutrophil-to-lymphocyte ratio; HR, hazard ratio; RR, risk ratio; CI, confidence interval; MD, median difference

†Defined as an irrefutably positive culture from an initially unaffected site or a new organism cultured from the original infected site, combined with the declaration of a definite infection on the basis of clinical symptoms by a clinician, laboratory test results, or imaging

‡Pathogenic microorganism clearance is defined as the transition of a culture specimen from a previously definite pathogenic microorganism to a negative state, coupled with the investigator’s comprehensive assessment of clinical symptoms, laboratory test results, and imaging

The analyses for determining HRs (Cox) and RRs (robust Poisson) were adjusted for prespecified confounders. Continuous outcomes are shown as medians (IQRs); MDs (Hodges–Lehmann) are from unadjusted analyses

Laboratory variables, including lymphocyte counts, procalcitonin (PCT) levels, and NLRs, refer to the last available measurements obtained before hospital discharge

“—” indicates that the adjusted effect estimates were not calculated because the variable exhibited a highly skewed and bounded distribution

Statistical analysis

Continuous variables are presented as the medians (interquartile ranges) and were compared between groups using the Wilcoxon rank-sum test. Categorical variables are expressed as counts (percentages) and were compared between groups using the χ² test or Fisher’s exact test, as appropriate. Survival outcomes were estimated using the Kaplan–Meier method and compared with the log-rank test. The primary analysis involved a multivariable-adjusted Cox proportional hazards model that was adjusted for prespecified covariates (Model 2). Additionally, because the timing of the initiation of Tα1 treatment was categorized as a binary variable (≤ 48 vs. >48 h), reliable time-dependent Cox modelling was not feasible. Consequently, a 48-hour landmark analysis for comparing early Tα1 initiation with no Tα1 use was performed as a coprimary analysis to mitigate immortal-time biases. The proportional hazards assumption was assessed using Schoenfeld residuals; if the proportional hazards assumption was violated, restricted mean survival time (RMST) analysis up to 28 days was performed as a supplementary approach. Other supplementary analyses included the construction of the unadjusted Cox model (Model 1) and an inverse probability of treatment weighting (IPTW)-adjusted Cox model with stabilized weights (Model 3). Binary secondary outcomes were analysed using robust Poisson regression to estimate risk ratios (RRs) and 95% confidence intervals (CIs). Skewed continuous outcomes were compared using the Wilcoxon rank-sum test, and the Hodges–Lehmann estimates of median differences (MDs) reported. For bounded and highly skewed outcomes (e.g., ventilator-free days and ICU-free days), only the P values are reported. Prespecified subgroup effects were evaluated through the inclusion of treatment-by-subgroup interaction terms in the Cox model and assessed with likelihood ratio tests. Restricted cubic spline (RCS) terms with four knots at the 5th, 35th, 65th, and 95th percentiles, together with their interactions with treatment, were derived to assess potential nonlinear effect modification by the baseline lymphocyte count; both overall and nonlinear interactions were tested using likelihood ratio tests. Unless otherwise specified, the subgroup, RCS, and sensitivity analyses were adjusted using the same covariates as in the primary model (Model 2). Sensitivity analyses were performed by (1) separately excluding patients with extremely severe disease and those with CKD and malignancies; applying doubly robust estimation through additional adjustment for the CD4-positive T-cell count (CD4 count) in the IPTW-adjusted Cox model; and repeating the primary analysis using multiple imputed datasets. Further details on the propensity score estimation, weighting, and multiple imputation methods are provided in the Supplementary Methods. Finally, an exploratory dose–response analysis was conducted within the Tα1 treatment group, modeling cumulative days of Tα1 use within the 28-day risk window using RCS; the results are presented in the Supplementary Materials. All analyses were performed using R Statistical Software (Version 4.4.1; http://www.R-project.org; The R Foundation). A two-sided P < 0.05 was considered to indicate statistical significance unless otherwise specified.

Results

Basic clinical characteristics

A total of 327 critically ill patients with TB-related sepsis were included: 160 patients in the Tα1 group and 167 patients in the control group (Table 1). The two groups were comparable in terms of age, sex, and most comorbidities, as well as in the distribution of infection sites and the incidence of bloodstream infections (all P > 0.05). Differences in TB classification and microorganism category were observed between the two groups (both P < 0.001). At baseline, patients in the Tα1 group were more likely to receive mechanical ventilation (15.0% vs. 6.6%, P = 0.014) but had slightly lower SOFA scores (median 5 vs. 6, P = 0.033). Other organ support measures and the APACHE II scores did not differ significantly between the groups (all P > 0.05). With regard to laboratory parameters, compared with the control group, the Tα1 group had lower leukocyte counts, serum creatinine levels, CD4 counts, lymphocyte counts, and NLRs (all P < 0.01), whereas no significant differences in the remaining laboratory indices were observed.

Table 1.

Baseline characteristics and covariate balance before and after IPTW

Characteristic	Overall (n = 327)	Thymosin α1 group (n = 160)	Control group (n = 167)	P value	SMD before IPTW	SMD after IPTW
Demographics
Age, median (IQR), y	64 (51–70)	65 (52–73)	63 (51–70)	0.264	0.03	0.05
Sex, n (%)				0.093	0.10	0.09
Female	71 (21.7)	41 (25.6)	30 (18.0)
Male	256 (78.3)	119 (74.4)	137 (82.0)
BMI (kg/m²)	17.6 (16.0–20.2)	17.6 (16.0–19.5)	18.0 (16.0–21.0)	0.077	0.29	0.12
Comorbidities, n (%)
Coronary heart disease	23 (7.0)	15 (9.4)	8 (4.8)	0.105	0.07	0.03
COPD	94 (28.7)	43 (26.9)	51 (30.5)	0.464	0.08	0.04
Hypertension	66 (20.2)	29 (18.1)	37 (22.2)	0.364	0.06	0.04
Diabetes	85 (26.0)	37 (23.1)	48 (28.7)	0.247	0.09	0.04
Chronic kidney disease	39 (11.9)	19 (11.9)	20 (12.0)	0.978	0.03	0.03
Malignancy	10 (3.1)	6 (3.8)	4 (2.4)	0.535	0.02	0.01
TB epidemiology
Type of TB, n (%)				< 0.001	0.17	0.03
Pulmonary TB	232 (70.9)	109 (68.1)	123 (73.7)
Extrapulmonary TB	11 (3.4)	8 (5.0)	3 (1.8)
Pulmonary + extrapulmonary TB	64 (19.6)	41 (25.6)	23 (13.8)
Disseminated TB	20 (6.1)	2 (1.2)	18 (10.8)
Retreated TB	26 (8.0)	5 (3.1)	21 (12.6)	0.002	0.14	0.04
Drug-resistant TB	9 (2.8)	3 (1.9)	6 (3.6)	0.503	0.03	0.01
Infection sites and microbiology
Infection site(s), n (%)				0.196	0.05	0.06
Lungs	288 (88.1)	146 (91.2)	142 (85.0)
Urinary tract	1 (0.3)	0 (0.0)	1 (0.6)
Multiple sites	36 (11.0)	14 (8.8)	22 (13.2)
Unknown	2 (0.6)	0 (0.0)	2 (1.2)
Bloodstream infection	23 (7.0)	13 (8.1)	10 (6.0)	0.450
Microorganisms, n (%)				< 0.001	0.32	0.24
Gram-negative bacteria	50 (15.3)	28 (17.5)	22 (13.2)
Gram-positive bacteria	13 (4.0)	7 (4.4)	6 (3.6)
Fungi	32 (9.8)	10 (6.2)	22 (13.2)
Viruses	14 (4.3)	6 (3.8)	8 (4.8)
Mixed	113 (34.6)	76 (47.5)	37 (22.2)
Culture negative	105 (32.1)	33 (20.6)	72 (43.1)
Resistant pathogen (MDR/XDR)	113 (34.6)	63 (39.4)	50 (29.9)	0.073	0.11	0.04
Organ support at baseline, n (%)
Vasopressors	21 (6.4)	10 (6.2)	11 (6.6)	0.901	0.01	0.00
Mechanical ventilation	35 (10.7)	24 (15.0)	11 (6.6)	0.014	0.13	0.01
CRRT	10 (3.1)	4 (2.5)	6 (3.6)	0.751	0.02	0.00
Severity scores, median (IQR)
SOFA score	5 (3–7)	5 (3–6)	6 (4–8)	0.033	0.25	0.01
APACHE II score	19 (15–25)	20 (16–25)	18 (15–24)	0.521	0.00	0.14
Laboratory results, median (IQR)
Lactate (mmol/L)	1.60 (1.10–2.55)	1.65 (1.10–2.70)	1.60 (1.20–2.35)	0.876	0.01	0.17
Leucocyte count (×109/L)	8.73 (5.87–12.12)	7.74 (5.08–11.10)	9.88 (6.94–13.06)	< 0.001	0.37	0.13
CRP (mg/L)	99.80 (46.99–152.72)	105.60 (62.54–152.72)	94.71 (35.00–153.67)	0.123	0.01	0.03
Procalcitonin (ng/mL)	0.83 (0.21–3.27)	0.70 (0.25–3.24)	1.02 (0.19–3.28)	0.948	0.19	0.16
Creatinine (µmol/L)	57.15 (42.18–90.17)	50.40 (39.45–81.45)	63.90 (45.90–98.00)	0.001	0.24	0.16
Total bilirubin (µmol/L)	13.15 (8.93–19.50)	13.15 (8.65–18.62)	13.00 (9.47–20.88)	0.324	0.12	0.11
CD4 count (cells/mm3)	141.00 (82.00–213.00)	108.50 (63.75–146.00)	202.00 (113.00–317.00)	< 0.001	0.83	0.62
Lymphocyte count (×109/L)	0.42 (0.27–0.71)	0.32 (0.24–0.47)	0.56 (0.35–0.90)	< 0.001	0.62	0.07
NLR	16.68 (9.97–31.53)	22.32 (11.94–42.14)	14.21 (8.72–25.92)	< 0.001	0.34	0.19

Abbreviations: APACHE II, Acute Physiology and Chronic Health Evaluation II; COPD, chronic obstructive pulmonary disease; CRRT, continuous renal replacement therapy; DR-TB, drug-resistant tuberculosis; IPTW, inverse probability of treatment weighting; IQR, interquartile range; MDR/XDR, multidrug-resistant/extensively drug-resistant; NLR, neutrophil-to-lymphocyte ratio; SMD, standardized mean difference; SOFA, Sequential Organ Failure Assessment; Tα1, thymosin α1; TB, tuberculosis; BMI, body mass index; CRP, C-reactive protein; CD4 count, CD4-positive T-cell count

Values are presented as medians (IQRs) or n (%). P values were calculated from unweighted comparisons between the Tα1 and control groups. The SMDs were calculated before and after IPTW; an absolute SMD after IPTW > 0.10 was considered indicative of residual imbalance. For categorical variables with multiple levels, the SMDs are displayed on the row for the parent variable

Among patients who received Tα1, the most common regimen was 1.6 mg once daily, used in 138 of 160 patients (86.3%), whereas the other 22 patients (13.8%) received 1.6 mg twice weekly. The median treatment duration was 10 (IQR, 7–14) days, and 113 patients (70.6%) initiated Tα1 within 48 h of ICU admission (Table S8).

Primary outcome (28-day all-cause mortality)

During the follow-up period, the 28-day all-cause mortality was lower in the Tα1 group than in the control group (45.0% vs. 55.1%; Table 2). In the primary multivariable Cox model (Model 2), Tα1 use was associated with a lower hazard of 28-day mortality after adjustment for the baseline covariates (adjusted hazard ratio [HR], 0.54; 95% CI, 0.36–0.79; P = 0.002; Table 2). The findings of the prespecified 48-hour landmark analysis were consistent with those of the primary multivariable Cox analysis (HR, 0.60; 95% CI, 0.39–0.94; P = 0.026; Table 3). Given the limited sample size, the primary outcome of the late-initiation group (> 48 h; n = 47) was analysed exploratorily. To assess robustness to unmeasured confounding, for the primary multivariable estimate, the E values were 3.10 for the point estimate and 1.85 for the CI limit closest to the null (Table S7); in other words, an unmeasured confounder would have to be associated with both Tα1 use and 28-day mortality by a risk ratio of at least 1.85 to explain the lower confidence limit. Regarding model assumptions, the Schoenfeld residuals indicated a violation of the proportional hazards assumption (global test P < 0.001; Tα1 P = 0.004; Table S1). Accordingly, RMST analysis revealed a longer adjusted restricted mean survival time in the Tα1 group (adjusted difference, 5.07 days [95% CI, 2.52–7.61; P < 0.001]; Table S2). The results of the stratified Cox sensitivity analyses were consistent with the primary findings (Table S3). Analysis of the Kaplan–Meier curves and the number-at-risk tables revealed early separation between the groups, with lower cumulative mortality in the Tα1 group; this separation persisted through the 28 days of follow-up without apparent late convergence (log-rank P = 0.018; Fig. 2). To further assess robustness, the unadjusted Cox model yielded an estimate in the same direction as the multivariable-adjusted model, and a similar estimate was obtained using stabilized IPTW (Table S4). After weighting, covariate balance improved for several variables; however, residual imbalance remained in multiple baseline characteristics, most prominently baseline CD4 count (SMD = 0.62; Table S5 and Figure S1).

Table 3.

Results of 48-hour landmark analysis of 28-day mortality

Outcome	Early initiation group (n = 113)	Late initiation group (n = 47)	Control group (n = 167)	Effect estimate (95% CI)	P value
28-day all-cause mortality, early initiation ≤ 48 h vs. control	55 (48.7)	—	92 (55.1)	HR 0.60 (0.39–0.94)	0.026
28-day all-cause mortality, late initiation > 48 h vs. control	—	17 (36.2)	92 (55.1)	HR 0.38 (0.22–0.67)	< 0.001

The values are presented as n (%). HR, hazard ratio; CI, confidence interval. The 48-hour landmark analysis included patients who were alive and under observation at 48 h; for these patients, time zero was reset at 48 h. The control group included patients who did not receive thymosin α1. Early initiation was defined as thymosin α1 initiation within 48 h after admission, whereas late initiation was defined as initiation after 48 h. The adjusted Cox model included the same covariates as the main model. The late-initiation comparisons should be interpreted as exploratory because exact treatment initiation times were unavailable

Fig. 2.

Kaplan–Meier estimates of 28-day all-cause mortality following ICU admission. The follow-up data were truncated on day 28; the curves show the cumulative probability of death (1-survival). Group differences were compared using the log-rank test. The numbers at risk are shown below the plot

Secondary outcomes

With regard to secondary outcomes, the Tα1 group was associated with a lower risk of in-hospital mortality (RR 0.69, 95% CI 0.50–0.93; P = 0.02) and a lower use of vasopressors within 28 days (RR 0.68, 95% CI 0.48–0.97; P = 0.03). New-onset infection within 28 days was more common in the Tα1 group (RR 1.30, 95% CI 1.01–1.67; P = 0.04); this association persisted with the use of stricter CDC/NHSN-compatible criteria (Table S9). No significant differences were observed in pathogenic microorganism clearance, mechanical ventilation, or the use of CRRT within 28 days (all P > 0.05). The numbers of ventilator-free and ICU-free days within 28 days also did not significantly differ between the groups (P = 0.11 and 0.28, respectively); however, the hospital length of stay was longer in the Tα1 group (MD 5.0 days, 95% CI 2.0–8.0; P < 0.001). According to the last available measurement before discharge, the lymphocyte count was lower in the Tα1 group (MD − 0.25, 95% CI − 0.36 to − 0.14; P < 0.001), whereas the PCT level and the NLR did not significantly differ between the groups (P = 0.48 and 0.80, respectively).

Subgroup analyses

In the prespecified subgroup analyses (Fig. 3), Tα1 use tended to be associated with a lower risk of 28-day mortality across most subgroups. In the overall population, Tα1 use was associated with a reduced risk of 28-day mortality (HR, 0.54; 95% CI, 0.36–0.79); no significant heterogeneity in this association was detected across subgroups (all P values for interaction > 0.05; moreover, none met the Bonferroni-corrected threshold of P < 0.005 for 10 interaction tests; Fig. 3). When stratified by age and sex, the direction of the association was similar: the HRs were 0.43 (95% CI, 0.23–0.79) for patients aged < 60 years and 0.63 (95% CI, 0.39–1.01) for those aged ≥ 60 years (P for interaction = 0.298); 0.42 (95% CI, 0.19–0.92) for female patients and 0.57 (95% CI, 0.37–0.87) for male patients (P for interaction = 0.481). In the COPD, diabetes, and CKD subgroups, the point estimates suggested a numerically greater risk reduction; however, the results of the interaction tests were not statistically significant (P for interaction = 0.669, 0.156, and 0.313, respectively). For several subgroups, the results had relatively wide confidence intervals, indicating limited precision of the effect estimates.

Fig. 3.

Subgroup analyses of the association between Tα1 use and 28-day mortality. The Bonferroni-corrected significance threshold for an overall type I error of 0.05 was set at P < 0.005 (0.05/10). Hazard ratios (HRs) for the primary outcome were compared between the two groups, and the associated 95% confidence intervals (CIs) were estimated from Cox proportional hazards regression models adjusted for prespecified covariates, with the subgroup-defining variable excluded from the adjustment set. The interaction between treatment and subgroup was added as a term to the Cox regression model to assess effect modification. CI, confidence interval; COPD, chronic obstructive pulmonary disease

Nonlinear interaction between the baseline lymphocyte count and the effect of Tα1 use on 28-day all-cause mortality

The results of RCS analysis revealed evidence of a nonlinear interaction between the baseline lymphocyte count and the association of Tα1 use with 28-day mortality (P for nonlinear interaction = 0.032), although the overall interaction did not reach conventional significance (P = 0.058) (Fig. 4). The inverse association between Tα1 use and 28-day mortality appeared most pronounced at intermediate lymphocyte counts, approximately 0.8–1.2 × 10⁹/L, where the estimated 95% confidence intervals remained below 1.0. At both lower (< 0.4 × 10⁹/L) and higher (> 1.4 × 10⁹/L) baseline levels, the association appeared attenuated, accompanied by widened confidence intervals, indicating increased uncertainty in the associations for these clinical subgroups. In the RCS analysis population, the lower range contained 140 patients, among whom 77 deaths occurred within 28 days, whereas the higher range contained only 15 patients, among whom 7 deaths occurred within 28 days (Table S10).

Fig. 4.

Restricted cubic spline (RCS) analysis of baseline lymphocyte count. The curve shows the adjusted hazard ratios (HRs) for Tα1 use versus control across baseline lymphocyte counts, and the shaded area indicates the 95% confidence interval (CI). The horizontal dashed line represents an HR of 1.0, and the vertical dotted line represents a reference lymphocyte count of 0.93 × 10⁹/L, which corresponds to the lowest adjusted predicted mortality risk. The lower 95% CI of the adjusted HR did not exceed 1.0 within the displayed lymphocyte-count range; therefore, no higher-risk threshold could be identified from the current data. P values denote overall and nonlinear interactions

Sensitivity analyses

Several sensitivity analyses were conducted to assess the robustness of the primary findings (Table 4). After patients with very severe disease (SOFA ≥ 15 and APACHE II ≥ 35) were excluded, Tα1 use remained significantly associated with a reduced 28-day mortality (HR 0.63, 95% CI 0.43–0.91; P = 0.014); similar results were observed after patients with CKD and malignancies were excluded (HR 0.50, 95% CI 0.34–0.73; P < 0.001). Given the residual imbalance in the CD4 count after IPTW adjustment, a doubly robust IPTW-adjusted Cox model additionally adjusted for CD4 count was constructed, and its results revealed a significant association between Tα1 use and lower 28-day mortality (HR 0.36, 95% CI 0.24–0.55; P < 0.001). In addition, repetition of the primary analysis in the MICE-imputed datasets yielded consistent results, further supporting the robustness of the main findings (Table S6). An exploratory dose–response analysis among Tα1-treated patients was also performed, and the results are presented in the Supplementary Materials (Figure S2).

Table 4.

Sensitivity analyses for the association between Tα1 treatment and 28-day mortality

Analysis	HR	95% CI	P value
SA1: Excluding patients with very severe disease (SOFA ≥ 15 and APACHE II ≥ 35), adjusted Cox model	0.63	0.43–0.91	0.014
SA2: Excluding patients with CKD + malignancies	0.50	0.34–0.73	< 0.001
Doubly robust IPTW-adjusted Cox model (CD4 count-adjusted)	0.36	0.24–0.55	< 0.001

SA1 was performed using a multivariable, adjusted Cox model, and SA2 was performed using an IPTW-adjusted Cox model

In the doubly robust sensitivity analysis, the primary propensity score model was unchanged, while the CD4 count was additionally adjusted for in the IPTW-adjusted Cox model

Abbreviations: HR, hazard ratio; CI, confidence interval; IPTW, inverse probability of treatment weighting; CKD, chronic kidney disease

Discussion

In this single-centre retrospective cohort study of ICU patients with TB-related sepsis, the results revealed that Tα1 use was associated with a reduced 28-day all-cause mortality (adjusted HR 0.54, 95% CI 0.36–0.79). Moreover, the results of prespecified 48-hour landmark analysis and multiple sensitivity analyses were directionally consistent with those of the primary analysis. Notably, given the violation of the proportional hazards assumption, complementary RMST analysis revealed a longer adjusted restricted mean survival time in the Tα1 group. Exploratory analyses also suggested possible effect modification by the baseline lymphocyte count. However, given the nonrandomized design and potential residual confounding, these findings should be interpreted as associations rather than causal effects.

Our findings must also be contextualized within the evolving evidence base regarding the use of Tα1 in sepsis patients. The landmark TESTS trial (2025) reported neutral results in an unselected sepsis population [20], whereas recent meta-analyses have suggested potential but statistically fragile benefits [33]. We hypothesize that these discrepancies do not necessarily represent a lack of drug efficacy but rather a recurring challenge in sepsis therapeutics: biological and clinical heterogeneity. In broad, unselected cohorts, the therapeutic efficacy of immunomodulation is frequently diluted by the inclusion of patients with differing immune states. By focusing exclusively on TB-related sepsis, we targeted a specific immunological endotype with a shared aetiologic driver and a high probability of profound, preexisting immune failure. The biological plausibility of our results is anchored in the “double-hit” theory of immunoparalysis. Active TB infection is characterized by chronic T-cell exhaustion and the upregulation of the levels of inhibitory checkpoints [10–12, 34, 35]; when acute, sepsis-induced lymphocyte depletion and impairment of antigen presentation are superimposed on this preexisting TB-mediated suppressive environment, a compounded state of host defence failure is created. In this context, Tα1 may counteract this state by promoting T-cell maturation and reversing the exhaustion-associated phenotype [15–17, 36–38]. This mechanism may also help explain the observed nonproportional hazard pattern: the early separation of the cumulative mortality curves may reflect partial, early correction of the compounded immunoparalysis, whereas later mortality may be driven by irreversible organ failure, secondary infection, and other competing clinical factors. The baseline differences in our cohort groups—specifically, lower lymphocyte and CD4 counts in the Tα1 group—further reinforce the idea that individuals with TB-related sepsis represent a “high-needs” subgroup in which host immune status fundamentally influences the responsiveness to Tα1. Notably, a substantial residual imbalance in the baseline CD4 count remained even after weighting, suggesting incomplete comparability in immune status and possible preferential Tα1 use among patients with relatively high immunosuppression. If a lower CD4 count indicates mainly a poorer prognosis, the residual imbalance may have weakened the observed association between Tα1 use and lower mortality. However, because the CD4 count may also reflect differences in immune status and clinical treatment decision-making, the direction and extent of this bias remain unclear.

The results of exploratory analyses suggested possible variability in the observed association between Tα1 use and 28-day mortality across the baseline lymphocyte counts. The results of RCS analysis suggested that this association was more pronounced at a baseline lymphocyte count between approximately 0.8 and 1.2 × 10⁹/L. Although the nonlinear interaction was statistically significant, the overall interaction did not reach conventional significance; therefore, the shape of the spline curve should be interpreted cautiously. The apparent weakening of the association at very low lymphocyte counts may reflect more advanced and less reversible immune paralysis in these patients, whereas the associations for higher lymphocyte counts were estimated on the basis of a small subgroup and are therefore imprecise. This interpretation is conceptually consistent with immune plasticity, whereby a markedly dysregulated but still “salvageable” host immune system may retain some ability to respond to immunomodulation [39]. Nevertheless, this biological explanation remains speculative, and thus the spline-derived pattern should be regarded as hypothesis-generating rather than as evidence of a therapeutic window or clinically actionable threshold and requires external validation before the results can be applied to clinical practice.

With respect to safety, we observed a higher incidence of new-onset infections, a longer hospital length of stay, and a lower predischarge final lymphocyte count in the Tα1 group. A sensitivity analysis restricted to patients who experienced microbiologically confirmed events that met the CDC/NHSN-compatible criteria revealed that new-onset infections remained more frequent in the Tα1 group (Table S9). However, these findings should not be interpreted as definitive evidence of Tα1-related infectious toxicity because cultures were clinically driven and not blinded to treatment exposure. In addition, patients who received Tα1 survived longer and had longer hospital stays, resulting in greater cumulative exposure to hospital-acquired pathogens and greater opportunities for microbiological sampling. Given the inherent limitations of retrospective analyses—including indication bias, ascertainment bias, survivor bias due to longer at-risk time, and surveillance bias—these findings should not be interpreted causally. A more plausible interpretation is that instead, these findings reflect the complex immune trajectories of and in-hospital exposure risks in critically ill patients with TB-related sepsis and that single-time-point or “last” immune measures may not thoroughly capture longitudinal immune dynamics during treatment.

Finally, these findings should also be interpreted in light of several limitations. First, given the single-centre, retrospective design of the study, residual confounding could not be eliminated, while propensity score methods and sensitivity analyses could not substitute for randomization; although the results of the E value analysis suggest moderate robustness to unmeasured confounding, the high mortality observed in TB-related sepsis may be associated not only with immune dysfunction but also with TB-specific and clinical factors, including the timing of anti-TB therapy, TB classification, drug resistance, secondary infections, underlying comorbidities, and other factors. Therefore, residual TB-specific confounding cannot be completely excluded, and the present findings should be considered hypothesis-generating until prospective validation studies can be conducted. Second, because the cohort was derived from the database of a tertiary infectious disease ICU in southwest China, the findings may not be fully generalizable to regions with different TB aetiologies, drug resistance patterns, host genetic backgrounds, and ICU practices [1, 40, 41]. Third, the real-world heterogeneity in Tα1 dosing, duration, and timing may introduce indication bias. Relatedly, treating Tα1 use as a baseline exposure in the primary Cox model may have introduced immortal time bias because patients classified as treated had to be alive at treatment initiation. Thus, because initiation timing was available only as a binary variable (≤ 48 vs. >48 h), a time-dependent Cox model could not be reliably fitted; therefore, we performed a 48-hour landmark analysis. Although this approach reduced the effects of a potential immortal-time bias, conditioning on survival to 48 h and the lack of exact initiation times meant that residual selection and time-related bias, particularly for late initiation, cannot be excluded. Fourth, the use of nonuniform timing and frequency of laboratory testing increases the uncertainty in interpreting the last-measured lymphocyte count. Fifth, secondary outcomes such as new-onset infections are susceptible to definition-based and ascertainment biases. Finally, all subgroup and RCS analyses were exploratory; therefore, the observed spline pattern requires external validation before any clinical interpretation.

From a clinical perspective, the lymphocyte count is a low-cost and readily repeatable component of routine ICU testing, making it a practical surrogate for monitoring immune dynamics in critically ill patients with TB-related sepsis. However, given the complexity of this population—which included active TB, drug resistance, secondary or polymicrobial infections, organ dysfunction, and polypharmacy—a single biomarker should inform treatment decision-making rather than be the sole indicator thereof. Given these complexities, multidisciplinary assessments involving infectious disease, critical care, immunology, and clinical pharmacology are warranted, and the lymphocyte count should be cautiously and individually used to inform Tα1 treatment decisions.

Conclusion

In conclusion, Tα1 use was associated with lower 28-day mortality among critically ill patients with TB-related sepsis; however, causality cannot be inferred from the results of this observational study. The results of exploratory RCS analysis suggested possible variability in this association across baseline lymphocyte counts, but the data-driven spline pattern should not be interpreted as a validated therapeutic window or used to uniquely guide treatment selection. Prospective multicentre studies are needed for external validation of these findings.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

TB

Tuberculosis

Tα1

Thymosin α1

ICU

Intensive care unit

HR

Hazard ratio

CI

Confidence interval

RCS

Restricted cubic spline

RMST

Restricted mean survival time

TESTS

Efficacy and Safety of Thymosin α1 for Sepsis

HIV

Human immunodeficiency virus

SOFA

Sequential Organ Failure Assessment

BMI

Body mass index

COPD

Chronic obstructive pulmonary disease

CKD

Chronic kidney disease

CRRT

Continuous renal replacement therapy

EMRs

Electronic medical records

APACHE II

Acute Physiology and Chronic Health Evaluation II

MICE

Multiple imputation by chained equations

IQR

Interquartile range

RR

Risk ratio

MD

Median difference

VFDs

Ventilator-free days

PCT

Procalcitonin

NLR

Neutrophil-to-lymphocyte ratio

IPTW

Inverse probability of treatment weighting

CD4 count

CD4-positive T-cell count

DR-TB

Drug-resistant tuberculosis

MDR/XDR

Multidrug-resistant/extensively drug-resistant

Author contributions

T.W. and S.T. performed the statistical analysis and drafted the manuscript. T.W., Q.G. and S.T. were responsible for data collection and data curation. Y.D.L., P.Y. and A.Z. conceived and designed the study. P.Y. and A.Z. supervised the study. All authors critically revised the manuscript and approved the final version.

Funding

This study was supported by the Chongqing Science and Health Joint Medical Research Project (Grant Nos. 2025MSXM109 and 2025ZDXM037).

Data availability

The datasets generated and/or analyzed during the current study are not publicly available due to patient privacy and institutional restrictions but are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

This study was performed in accordance with the principles of the Declaration of Helsinki (revised in Brazil 2013). All experimental protocols were approved by the Ethics Committee of Chongqing Public Health Medical Center (Approval No. 2025-028-01-KY). The requirement for informed consent was waived by the Ethics Committee of Chongqing Public Health Medical Center because of the retrospective and noninterventional nature of this study.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.