Immunosuppression, resistance burden, and qSOFA on short-term prognosis and difficult clearance in hospitalized patients with Salmonella infection: a single-center retrospective cohort study Authors: Xuewei Wang, Shuqi Yang, Yijie Lin, Chunlin Chen, Liting Zhan, Zimei Wan, Wenjin Yuan, Zhangyan Weng, Huatang Zhang, Miaona Xie, Xiuhua He, Mengling Guo, Zhijun Su, Yijuan Zheng, Qingqing Chen, Xueping Yu Published: 2026 ABSTRACT AbstractBackgroundImmunosuppressed patients who develop Salmonella infection are more likely to experience bacteremia, sepsis, and treatment failure; however, systematic evidence remains limited regarding how different sources of immunosuppression and resistance burden jointly affect early outcomes. Using a source-based classification of immunosuppression, we evaluated the relationships of host severity (qSOFA), pathogen resistance burden, 30-day composite adverse outcomes, and difficult clearance.MethodsWe retrospectively enrolled 232 hospitalized patients with Salmonella infection between October 2014 and January 2026. Patients were stratified as typhoidal/paratyphoidal Salmonella (TS) or nontyphoidal Salmonella (NTS) according to serotype and grouped by the source of immunosuppression. Infection phenotypes were classified as enteric, bloodstream/sepsis, mixed/multisite, or other focal phenotypes. Independent associated factors were identified using multivariable logistic regression.ResultsThe incidence of the 30-day composite adverse outcome was 10.8% (25/232), and the incidence of difficult clearance was 22.0% (51/232). Immunosuppressed patients were more likely to present with bloodstream or disseminated infection phenotypes, but immunosuppression itself was not an independent determinant of 30-day adverse outcomes. qSOFA ≥ 2 was independently associated with the 30-day composite adverse outcome among patients with available qSOFA data (aOR 5.36, 95% CI 1.56–15.95, P = 0.006). Typhoidal Salmonella was significantly associated with difficult clearance in the primary model. Higher phenotypic resistance burden showed an association with difficult clearance in the primary analysis, but this association was not robust after the score was reconstructed using only high-coverage antimicrobial agents.ConclusionsIn hospitalized patients with Salmonella infection, infection-phenotype distributions varied across sources of immunosuppression, although this observation was exploratory because of small subgroup sizes. The qSOFA-related prognostic findings primarily apply to adult or adult-skewed patients with available qSOFA data. Phenotypic resistance burden may reflect treatment complexity, but its association with difficult clearance should be interpreted cautiously because it lacked robustness in the high-coverage sensitivity analysis.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1186/s12879-026-13707-7.Keywords: Salmonella, Source-based classification of immunosuppression, qSOFA, Antimicrobial resistance, Difficult clearance, Prognosis Introduction With the growing number of patients receiving cancer chemotherapy, organ transplantation, biologics/targeted therapies, and long-term glucocorticoid treatment, the risk of invasive nontyphoidal Salmonella (NTS) infection in immunosuppressed hosts continues to increase. Clinically, infection may rapidly progress from localized intestinal disease to bacteremia, septicemia, and even sepsis, and is associated with higher risks of short-term mortality and complications [1–5]. Meanwhile, the spread of reduced susceptibility/resistance to key therapeutic agents such as fluoroquinolones and third-generation cephalosporins further increases the likelihood of inadequate empirical coverage and treatment failure [6–8]. In some patients, even after antimicrobial treatment is initiated, persistent bacteremia, delayed infection control, or escalation of therapy may still occur, constituting a phenomenon of “difficult clearance” [9]. Such patients often have longer hospital stays and more complications, and may portend worse short-term clinical outcomes. However, three key gaps remain in the current evidence. First, most studies simplify immunosuppression as a binary exposure (present/absent), overlooking the heterogeneity of glucocorticoids, chemotherapy, transplantation, and biologics in terms of mechanisms of immune injury and infection spectrum [10, 11]. Second, direct comparative data are limited regarding whether different sources of immunosuppression drive distinct infection phenotypes. Third, the relative contributions of early host organ dysfunction (qSOFA) and pathogen resistance burden to short-term outcomes and infection clearance have not been integrated into a single risk framework [5, 12, 13]. In particular, “difficult clearance” as a process-oriented endpoint may reflect the overall imbalance among host, pathogen, and treatment earlier than mortality or ICU admission alone. In this single-center hospitalized cohort, we simultaneously incorporated source-based classification of immunosuppression, infection phenotypes, the burden of nonsusceptibility to seven commonly used antimicrobial agents, and bedside qSOFA scores, while setting two outcomes: a 30-day composite adverse outcome and difficult clearance. The core innovation of this study was not merely to describe that immunosuppressed patients are more severely ill, but to address three clinical questions: which hosts are more likely to develop disseminated infection, which factors truly determine short-term prognosis, and which patients are more likely to experience difficult clearance during treatment. In doing so, we sought to construct a more actionable framework for early risk stratification. Methods Study design and data sources This was a single-center retrospective cohort study conducted at Fujian Medical University Affiliated First Quanzhou hospital. Data were obtained from the electronic medical record system, microbiology laboratory system, and antimicrobial susceptibility database of Quanzhou First Hospital. The study period extended from October 24, 2014, to January 8, 2026. The study followed the STROBE reporting guideline. The study complied with the Declaration of Helsinki and the ICMJE Recommendations and was approved by the institutional ethics committee ([2026] K080; 19 March 2026). The requirement for informed consent was waived because of the anonymised retrospective design. All data were de-identified before analysis. The index time was defined as the time of first symptom onset related to Salmonella infection: for community-acquired infections, the date of admission was used; for nosocomial infections, the date when infection-related symptoms were first documented in the medical record was used. Synchronous positive cultures from different sites during the same hospitalization were considered a single infectious episode. If the same patient had multiple hospitalizations with Salmonella infection, only the first episode was included to avoid duplicate counting. Study population Patients were included if at least one clinical specimen obtained during hospitalization (blood, stool, urine, or another sterile-site specimen) was culture- or PCR-positive for Salmonella and if data on immunosuppression status, infection phenotype, and outcomes were available for classification. Inclusion and exclusion criteria The inclusion criteria were as follows: (1) at least one clinical specimen obtained during hospitalization (blood, stool, urine, or another sterile-site specimen) was culture- or PCR-positive for Salmonella; and (2) data on immunosuppression status, infection phenotype, and outcomes were available for classification. The exclusion criteria were as follows: (1) non-Salmonella infection; (2) missing core variables that precluded classification or outcome adjudication; and (3) repeated hospitalizations for infection in the same patient, in which case only the first event was retained. All information related to immunosuppression was cross-checked using both the electronic medical record and medication orders. Study outcomes The 30-day composite adverse outcome was defined as the occurrence of any of the following events within 30 days after the index time: all-cause death, ICU admission, mechanical ventilation, or infection-related shock. Patients discharged early without available 30-day follow-up outcomes were considered to have missing outcome data and were not included in the primary multivariable model. Difficult clearance was defined as the occurrence of any of the following during hospitalization: persistent bacteremia for 48–72 h, failure of clinical improvement at 72 h, escalation of antimicrobial therapy, or treatment failure. This composite endpoint was designed to capture process-oriented risks reflecting persistent pathogen presence or insufficient treatment response, rather than being limited to late hard endpoints such as death. Main exposure: source-based classification of immunosuppression The main exposure of interest was source-based classification of immunosuppression. According to the source of immunosuppression, patients were categorized into five groups: (1) high-dose glucocorticoids (e.g., oral or intravenous administration); (2) chemotherapy/bone marrow suppression (related to chemotherapeutic agents or bone marrow suppressants); (3) organ transplantation/immunosuppressants (e.g., kidney transplantation, liver transplantation, or use of immunosuppressive drugs); (4) biologics (e.g., anti-TNF, anti-IL−6, etc.); and (5) no immunosuppression (reference group). The operational definition of high-dose glucocorticoids followed commonly used thresholds (e.g., prednisone ≥ 20 mg/day or equivalent for ≥ 14 days) [14, 15]. If a patient had multiple sources of immunosuppression, a prespecified hierarchy was used to assign the patient to the category with higher immunosuppressive intensity and greater temporal proximity to the index time. The prespecified priority order was organ transplantation/immunosuppressants > chemotherapy/bone marrow suppression > biologics/targeted immunomodulation > high-dose glucocorticoids [15]. Each classification was carefully verified against medical records and treatment documentation to ensure high consistency. Microbiology and stratification: typhoidal versus nontyphoidal Salmonella Based on Salmonella serotypes and pathogenic characteristics, infections were classified as typhoidal Salmonella (TS) or nontyphoidal Salmonella (NTS). Specifically, TS included Salmonella enterica serovars Typhi and Paratyphi, whereas all other serotypes (including Salmonella Typhimurium) were classified as NTS [16]. Classification was determined according to the literature and microbiological results, and patients were stratified accordingly. Typhoidal cases were included as a covariate in the models to reduce their potential impact on analyses focused on the NTS population. Pathogen identification was performed using the MALDI-TOF MS system. Antimicrobial susceptibility testing was conducted using the Phoenix 100 automated system (BD) or the Kirby-Bauer disk diffusion method. Susceptibility results were interpreted according to CLSI M100 (35th edition, 2025) [17]. If repeated isolates were obtained from the same infectious episode, the susceptibility result closest to the index time and with the most complete information was used. Infection phenotypes and severity assessment According to clinical manifestations and the site of infection, infection was classified into four phenotypes: (1) enteric phenotype, characterized by intestinal infection only (e.g., diarrhea, vomiting, or abdominal pain), with Salmonella detected in stool specimens or by gastrointestinal endoscopy; (2) sepsis/bloodstream phenotype, including all cases with positive blood cultures and patients meeting a sepsis diagnostic framework [11]; (3) mixed/multisite phenotype, indicating infection at multiple sites (e.g., concurrent intestinal and bloodstream infection); and (4) other focal phenotype, including single localized infections such as pneumonia or urinary tract infection. These categories were assigned on the basis of patients’ clinical manifestations and laboratory diagnoses and were used in subsequent analyses. qSOFA and phenotype adjudication: qSOFA comprises three items—respiratory rate, systolic blood pressure, and mental status. The worst values within 24 h after the index time were used for calculation. In addition to blood culture positivity, the bloodstream/sepsis phenotype also included cases meeting the Sepsis−3 framework and judged clinically to represent sepsis [11, 16]. Phenotype adjudication was independently reviewed by two investigators, with disagreements resolved by a third reviewer. Definition of resistance profile Seven commonly used antimicrobial agents were evaluated: fluoroquinolones (ciprofloxacin [CIP] and levofloxacin [LVX]), azithromycin (AZM), trimethoprim-sulfamethoxazole (TMP-SMX), cefotaxime (CTX), meropenem (MEM), and cefoperazone-sulbactam (SCF). Susceptibility results were based on the microbiology laboratory report using the S/I/R categories, and the nonsusceptibility rate was defined as I + R. According to the number of nonsusceptible drugs, patients were categorized into the following resistance strata: susceptible group (0 nonsusceptible drugs), standard-resistance group (1–2 nonsusceptible drugs), multidrug-resistance group (3–4 nonsusceptible drugs), and extensive-resistance group (≥ 5 nonsusceptible drugs). In the resistance-burden stratification, testing coverage varied among the seven antimicrobial agents. Untested drugs were coded as 0 only to construct a conservative operational resistance-burden score and should not be interpreted as confirmed susceptibility. Coverage and non-susceptibility rates for each agent are summarized in Supplementary Table S2. Agents with testing coverage < 80% were excluded in sensitivity analyses of the resistance-burden score. This stratification is better suited to risk stratification and should not be interpreted as an exact surrogate for the true resistance profile. Because very few cases had ≥ 5 nonsusceptible drugs, cases with ≥ 3 nonsusceptible drugs were combined into one group in multivariable models to improve estimation stability. The resistance-burden score was therefore a phenotypic susceptibility-based operational score and should not be interpreted as a genomic resistome profile. Statistical analysis Descriptive statistics were first performed. Categorical variables were compared using the χ² test or Fisher’s exact test, and continuous variables were compared using the t test or Mann-Whitney U test according to distributional characteristics. Multivariable logistic regression models were then constructed separately with the 30-day composite adverse outcome and difficult clearance as dependent variables. Covariates were prespecified clinically relevant variables, including age, sex, immunosuppression status, TS/NTS status, qSOFA, and resistance burden. Because qSOFA components were missing in a subset of patients, the primary multivariable analyses were conducted as complete-case analyses restricted to patients with available qSOFA scores (n = 150). As patients excluded from the qSOFA-complete model were predominantly paediatric, the qSOFA-related prognostic findings should be interpreted mainly in the qSOFA-complete, adult-skewed analytic population. Given the retrospective nature of data collection and the limited number of outcome events, no multiple imputation was performed. Patients with missing outcome data were excluded from the corresponding analyses. Given the limited number of events, a large number of comorbidity variables were not further included to avoid overfitting. Results are reported uniformly as adjusted odds ratios (aORs), 95% confidence intervals (CIs), and two-sided P values. To assess robustness, sensitivity analyses were conducted by (1) redefining the resistance-burden score using only agents with high testing coverage—ciprofloxacin, trimethoprim–sulfamethoxazole, and cefotaxime—and (2) restricting the cohort to non-typhoidal Salmonella cases. Results of these analyses are presented in Supplementary Table S3. All statistical tests were two-sided, and P < 0.05 was considered statistically significant. Multicollinearity was assessed using variance inflation factors, and calibration was evaluated using the Hosmer–Lemeshow goodness-of-fit test. These diagnostics revealed no substantial multicollinearity and acceptable model fit. Results Study population and overall characteristics A total of 232 hospitalized patients with Salmonella infection were included, comprising 9 TS cases (3.9%) and 223 NTS cases (96.1%). The overall mean age was 28.6 ± 29.6 years, and the median age was 13.5 years (IQR 1.0–57.0); 63.4% were male. The study flowchart is shown in Fig. 1, and detailed baseline characteristics are presented in Table 1. Flowchart of study population screening Table 1Baseline characteristicsVariableHigh-dose glucocorticoids (n = 14)Chemotherapy/bone marrow suppression (n = 8)Organ transplantation/immunosuppressants (n = 1)Biologics/targeted therapy (n = 3)No immunosuppression (n = 206)Overall (n = 232)Age (mean ± SD)56.6 ± 15.555.4 ± 19.763.053.0 ± 4.425.1 ± 29.128.6 ± 29.6Male, n (%)10 (71.4)6 (75.0)0 (0.0)1 (33.3)130 (63.1)147 (63.4)qSOFA, median (IQR)1.0 (0.0–1.0)1.0 (0.0-1.2)0.0 (0.0–0.0)0.0 (0.0-0.5)1.0 (0.0–2.0)1.0 (0.0–2.0)Diabetes mellitus, n (%)4 (28.6)1 (12.5)0 (0.0)0 (0.0)21 (10.2)25 (10.8)Solid tumor, n (%)5 (35.7)1 (12.5)0 (0.0)1 (33.3)6 (2.9)14 (6.0)Enteric phenotype, n (%)7 (50.0)6 (75.0)1 (100.0)0 (0.0)136 (66.0)150 (64.7)Bloodstream/sepsis phenotype, n (%)4 (28.6)0 (0.0)0 (0.0)2 (66.7)45 (21.8)51 (22.0)30-day adverse outcome, n (%)3 (21.4)0 (0.0)0 (0.0)0 (0.0)22 (10.7)25 (10.8) Baseline characteristics Table 2Major antimicrobial nonsusceptibility strata and adverse outcomesResistance stratumTotal nAdverse outcome, n (%)No adverse outcome, n (%)Susceptible (0 nonsusceptible drugs)793 (3.8)76 (96.2)Standard resistance (1–2 nonsusceptible drugs)13220 (15.2)112 (84.8)Multidrug resistance (3–4 nonsusceptible drugs)192 (10.5)17 (89.5)Extensive resistance (≥ 5 nonsusceptible drugs)20 (0.0)2 (100.0)Footnotes: The trend shown in Table 2 should be interpreted descriptively. In the multivariable model, resistance burden did not reach statistical significance for the 30-day composite adverse outcome. Because partially untested drugs were coded as 0, this analysis should be regarded as a conservative operational estimate rather than a precise measure of true resistance prevalence Major antimicrobial nonsusceptibility strata and adverse outcomes Footnotes: The trend shown in Table 2 should be interpreted descriptively. In the multivariable model, resistance burden did not reach statistical significance for the 30-day composite adverse outcome. Because partially untested drugs were coded as 0, this analysis should be regarded as a conservative operational estimate rather than a precise measure of true resistance prevalence A supplementary analysis compared the 150 patients included in the qSOFA-complete multivariable model with the 82 patients excluded because of missing qSOFA data. Patients excluded for missing qSOFA were substantially younger (median age 1 year versus 50 years) and were predominantly paediatric, sex and outcome rates were broadly comparable between groups, whereas other baseline characteristics should be interpreted in light of the marked age imbalance (Supplementary Table S1). This age imbalance indicates that the qSOFA-complete model was largely adult-focused; therefore, qSOFA-related prognostic results should be interpreted primarily in hospitalized adults or adult-skewed patients with available qSOFA components. The distribution of immunosuppression sources was highly unbalanced: 206 patients (88.8%) had no definite immunosuppression, 14 (6.0%) received high-dose glucocorticoids, 8 (3.4%) had chemotherapy/bone marrow suppression, 3 (1.3%) received biologics/targeted therapy, and 1 (0.4%) had organ transplantation/immunosuppressant exposure. Patients in the immunosuppressed subgroups were generally older, suggesting that baseline host frailty may be an important background factor underlying subsequent differences in phenotype and prognosis. Distribution of infection phenotype Overall, the enteric phenotype was the most common (64.7%), followed by the bloodstream/sepsis phenotype (22.0%) and the mixed/multisite phenotype (12.5%) (Fig. 2). Along the main line of “host type-infection spectrum,” the high-dose glucocorticoid group and the biologics group were more likely to present with bloodstream or disseminated infection, whereas the chemotherapy group still predominantly showed the enteric phenotype, suggesting that different sources of immunosuppression may influence the trajectory of infection spread through different mechanisms. Migration map of infection phenotypes TS and NTS differed significantly in the distribution of infection phenotypes: TS was more likely to present as bloodstream/sepsis phenotype, whereas NTS remained predominantly enteric (P = 0.0038). However, the incidence of the 30-day composite adverse outcome was similar between the two groups, suggesting that pathogen type affects phenotype but does not necessarily directly determine short-term clinical outcomes. Resistance profiles and resistance strata The major antimicrobial nonsusceptibility profile showed the highest burden for fluoroquinolones, especially CIP, followed by TMP-SMX and CTX, whereas MEM showed almost no nonsusceptibility. Because not all isolates were tested for every agent, including LVX, nonsusceptibility rates for individual drugs should be interpreted together with testing coverage. The resistance profile is shown in Fig. 3. Antimicrobial resistance spectrum of Salmonella isolates According to overall resistance burden, the susceptible group, standard-resistance group, multidrug-resistance group, and extensive-resistance group accounted for 34.1%, 56.9%, 8.2%, and 0.9% of the cohort, respectively. The rate of the 30-day composite adverse outcome generally increased with higher resistance strata, but the estimate for the most severe stratum was unstable because of the very small sample size. Testing coverage varied across the seven antimicrobial agents. Ciprofloxacin, trimethoprim–sulfamethoxazole, and cefotaxime each achieved > 90% testing coverage, whereas levofloxacin, azithromycin, meropenem and cefoperazone–sulbactam were tested in < 80% of isolates (Supplementary Table S2). Consequently, the high-coverage agents were used for sensitivity analyses of the resistance-burden score. Thirty-day composite adverse outcomes The overall incidence of the 30-day composite adverse outcome was 10.8% (25/232). The high-dose glucocorticoid group had the highest proportion of adverse outcomes (21.4%, 3/14), whereas no events were observed in the other immunosuppressed subgroups. However, because of the limited number of events, the between-group difference did not reach statistical significance. This suggests that the source of immunosuppression may influence the direction of risk, but was insufficient to become a stable independent predictor in this cohort. Difficult clearance The composite endpoint of difficult clearance occurred in 22.0% (51/232) of patients. The most common component was failure of clinical improvement at 72 h (19.4%), followed by escalation of antimicrobial therapy (6.0%), persistent bacteremia (3.9%), and treatment failure (3.9%). This finding suggests that focusing only on death or ICU admission would underestimate the clinical burden of persistent non-resolution during treatment. When stratified by immunosuppression status, the difficult-clearance rate was 21.4% (44/206) in patients without definite immunosuppression and 26.9% (7/26) in patients with immunosuppression; the difference was not statistically significant (P = 0.279). Multivariable logistic regression analysis A multivariable logistic regression model was constructed in 150 patients with available qSOFA scores, adjusting for age, sex, immunosuppression status, typhoidal status, and resistance burden. Because the number of extensive-resistance cases was small and events were sparse, cases with ≥ 3 nonsusceptible drugs were merged into one group (≥ 3 vs. 0) in the multivariable model to improve estimation stability. In the model for the 30-day composite adverse outcome, qSOFA ≥ 2 was independently associated with increased risk (aOR 5.36, 95% CI 1.56–15.95, P = 0.006). Resistance burden showed an increasing but statistically non-significant trend. The standard-resistance group had an aOR of 4.28 (P = 0.070), and the ≥ 3 nonsusceptible-drug group had an aOR of 5.99 (P = 0.108). Therefore, resistance burden was not interpreted as an independent predictor of the 30-day composite adverse outcome in this cohort. Immunosuppression status, typhoidal status, age, and sex were not statistically significant. Detailed results are shown in Fig. 4; Table 3. Forest plot of multivariable logistic regression for the 30-day adverse outcome Table 3Multivariable logistic regression for 30-day adverse outcomesVariableAdjusted OR (95% CI)P valueqSOFA ≥ 2 (vs. < 2)5.36 (1.56–15.95)0.006Standard resistance (1–2 nonsusceptible drugs) vs. 04.28 (0.89–20.58)0.070≥ 3 nonsusceptible drugs vs. 05.99 (0.68–53.07)0.108Immunosuppression (yes vs. no)0.68 (0.17–2.76)0.588Typhoidal status (yes vs. no)2.05 (0.21–20.05)0.538Age (per 1-year increase)1.02 (0.99–1.04)0.140Sex (male vs. female)0.56 (0.20–1.56)0.268 Multivariable logistic regression for 30-day adverse outcomes Table 3 presents adjusted ORs and 95% CIs for the primary multivariable logistic regression model. Patients with missing qSOFA data were excluded from this model. To avoid unstable estimates caused by sparse events in the extensive-resistance category, cases with ≥ 3 nonsusceptible drugs were combined in the regression analysis. A separate multivariable logistic regression model was then constructed with difficult clearance as the dependent variable, adjusting for age, sex, immunosuppression status, qSOFA score, typhoidal status, and resistance burden. In the primary model, typhoidal status was significantly associated with difficult clearance (aOR 23.41, 95% CI 3.98–37.85, P < 0.001), and the ≥ 3 nonsusceptible-drug group was significantly associated with difficult clearance (aOR 6.27, 95% CI 1.23–31.95, P = 0.027). However, when the resistance-burden score was reconstructed using only high-coverage agents, this association was no longer statistically significant (aOR 2.59, P = 0.440). Therefore, the relationship between resistance burden and difficult clearance should be interpreted as an exploratory phenotypic signal rather than a robust independent association. The primary difficult-clearance model results are shown in Fig. 5; Table 4, and the sensitivity analysis is presented in Supplementary Table S3. Forest plot of multivariable logistic regression for difficult clearance Table 4Multivariable logistic regression for difficult clearanceVariableAdjusted OR (95% CI)P valueqSOFA ≥ 2 (vs. < 2)3.72 (1.33–10.45)0.013Standard resistance (1–2 nonsusceptible drugs) vs. 02.45 (0.80–7.49)0.115≥ 3 nonsusceptible drugs vs. 06.27 (1.23–31.95)0.027Immunosuppression (yes vs. no)1.25 (0.42–3.74)0.689Typhoid status (yes vs. no)23.41 (3.98–37.85)< 0.001Age (per 1-year increase)1.01 (0.99–1.02)0.522Sex (male vs. female)1.12 (0.46–2.69)0.803 Multivariable logistic regression for difficult clearance Sensitivity analysis in the NTS subset Among the 142 patients with NTS only and available qSOFA, the predictive effect of qSOFA ≥ 2 on the 30-day adverse outcome remained stable (aOR 5.39, 95% CI 1.62–17.89, P = 0.006), indicating that the core conclusion of the primary analysis was not driven by the small number of TS cases. The detailed results are shown in Table 5. Table 5Multivariable logistic regression for 30-day adverse outcomes in the NTS subsetVariableAdjusted OR (95% CI)P valueqSOFA ≥ 2 (vs. < 2)5.39 (1.62–17.89)0.006Immunosuppression (yes vs. no)0.76 (0.18–3.15)0.701Standard resistance (1–2 nonsusceptible drugs) vs. 03.81 (0.77–18.76)0.100≥ 3 nonsusceptible drugs vs. 05.48 (0.62–48.62)0.127Age (per 1-year increase)1.02 (0.99–1.04)0.150Sex (male vs. female)0.53 (0.19–1.48)0.224 Multivariable logistic regression for 30-day adverse outcomes in the NTS subset In the NTS-only sensitivity analysis with difficult clearance as the outcome, the ≥ 3 nonsusceptible-drug group remained statistically significant (aOR 6.43, P = 0.029). However, this finding should be interpreted together with the high-coverage-agent sensitivity analysis, in which the association between resistance burden and difficult clearance was no longer statistically significant. Taken together, these analyses suggest that resistance burden may represent a clinically relevant but methodologically sensitive signal. The results are shown in Table 6. Table 6Multivariable logistic regression for difficult clearance in the NTS subsetVariableAdjusted OR (95% CI)P valueqSOFA ≥ 2 (vs. < 2)3.62 (1.28–10.21)0.015Immunosuppression (yes vs. no)1.49 (0.48–4.61)0.489Standard resistance (1–2 nonsusceptible drugs) vs. 02.46 (0.74–8.16)0.141≥ 3 nonsusceptible drugs vs. 06.43 (1.21–34.17)0.029Age (per 1-year increase)1.01 (0.99–1.02)0.563Sex (male vs. female)1.26 (0.51–3.15)0.617 Multivariable logistic regression for difficult clearance in the NTS subset Discussion The key value of this study lies not in re-demonstrating that immunosuppressed patients are more severely ill, but in integrating source-based classification of immunosuppression, resistance burden, qSOFA, and the endpoint of difficult clearance into a single analytical framework. In this cohort, infection-phenotype distributions varied across sources of immunosuppression, although these subgroup findings should be considered exploratory because some categories were very small. Among patients with available qSOFA data, qSOFA ≥ 2 was the most consistent factor associated with 30-day adverse outcomes. However, because patients excluded for missing qSOFA data were predominantly paediatric and the qSOFA-complete cohort was adult-skewed, the qSOFA-related prognostic findings should be interpreted primarily as applying to hospitalized adults or adult-skewed patients with available qSOFA components. Previous studies have shown that invasive NTS infection is more common in immunosuppressed patients or those with chronic diseases and is often associated with higher risks of complications and death [1–3]. At the same time, the spread of resistant strains has been regarded as an important contributor to treatment failure and disease worsening [6–8]. Building on this background, the present study classified immunosuppression by source and found that patients receiving high-dose glucocorticoids or biologic therapy were more likely to develop bloodstream or multisite disseminated infection. This suggests that immunosuppression not only affects the risk of infection occurrence, but may also weaken local host immune defenses, making it easier for infection to breach the intestinal barrier and enter the bloodstream, thereby reshaping the infection phenotype structure. In the multivariable model for 30-day adverse outcomes, early organ dysfunction reflected by qSOFA was the most consistent prognostic factor, whereas resistance burden showed only a non-significant increasing trend. Compared with the traditional binary classification of immunosuppression (yes/no), source-based stratification helps better explain the heterogeneity between host immune status and infection phenotypes. Beyond short-term adverse outcomes, this study further incorporated difficult clearance as a process-oriented endpoint. The primary model suggested that a higher phenotypic resistance burden may be associated with difficult clearance. However, this association was not robust when the resistance-burden score was reconstructed using only antimicrobial agents with high testing coverage. Therefore, the resistance-burden finding should be interpreted as an exploratory signal reflecting possible treatment complexity, rather than as a stable independent predictor. This cautious interpretation is important because incomplete phenotypic susceptibility testing may have influenced the construction of the resistance-burden score. Although some comparisons did not reach statistical significance, this directional pattern is consistent with prior research suggesting that persistent Salmonella infection and antibiotic-tolerant subpopulations (persister cells) may lead to treatment failure [18–19]. Resistant strains may prolong infection duration and increase treatment complexity by forming tolerant subpopulations or surviving persistently within the host. In addition, this study observed a higher risk of difficult clearance in TS infection, which may be related to the stronger capacity of typhoidal Salmonella for systemic dissemination and long-term survival within the host, indicating that pathogen-specific characteristics may also influence the clearance process. This study has three direct implications for clinical management. First, in immunosuppressed hosts, blood cultures and multisite microbiological sampling should be obtained as early as possible at admission or at the infection index time, together with imaging evaluation, to promptly identify bloodstream or disseminated infection. Second, among adult or adult-skewed patients with available qSOFA components, early qSOFA assessment may support rapid recognition of patients at increased risk of short-term adverse outcomes. Third, phenotypic antimicrobial susceptibility results remain clinically important for guiding treatment adjustment, but the resistance-burden score in this study should be viewed as an exploratory risk-stratification signal rather than a definitive predictor of difficult clearance. When patients fail to improve within 72 h or have apparent high resistance burden, source control and antimicrobial regimens should be reassessed promptly. Our assessment of missing qSOFA data showed that excluded patients were predominantly paediatric, whereas the qSOFA-complete analytic cohort was skewed toward adults. Although outcome rates were comparable between included and excluded patients, the age imbalance indicates that the qSOFA-related prognostic findings mainly apply to hospitalized adults or adult-skewed patients with available qSOFA components. These findings should not be directly generalized to paediatric Salmonella infections, for which paediatric-specific severity assessment tools may be more appropriate. Only three antimicrobial agents—ciprofloxacin, trimethoprim–sulfamethoxazole and cefotaxime—were tested in the vast majority of isolates. The resistance-burden score should therefore be viewed as a phenotypic susceptibility-based operational measure rather than a precise estimate of true resistance prevalence. Importantly, the association between higher resistance burden and difficult clearance was not robust in the high-coverage sensitivity model. This finding indicates that the observed association in the primary model may have been influenced by the inclusion of lower-coverage antimicrobial agents and should therefore be interpreted cautiously. Differences in infection-phenotype distributions across sources of immunosuppression should be interpreted with caution. Some immunosuppression subgroups, such as organ transplantation and biologic therapy, were represented by very few patients; thus these observations represent exploratory trends rather than definitive source-specific effects. Early organ dysfunction, reflected by the qSOFA score, emerged as the most consistent predictor of short-term adverse outcomes. In contrast, antimicrobial resistance exerted its principal impact on treatment dynamics, as demonstrated by its association with difficult clearance but not with the 30-day composite outcome. Several limitations should be acknowledged. First, this was a single-center retrospective study, and the numbers of patients in the organ transplantation and biologics subgroups were very small, limiting fine-grained comparisons across different sources of immunosuppression. Second, missing qSOFA data required complete-case analysis and the excluded patients were predominantly paediatric; therefore, qSOFA-related findings mainly apply to hospitalized adults or adult-skewed patients with available qSOFA components. Third, incomplete antimicrobial susceptibility testing limited the robustness of the resistance-burden analysis, as the association between higher resistance burden and difficult clearance was not confirmed in the high-coverage sensitivity model. Finally, the resistance-burden score was based on phenotypic susceptibility testing and should not be interpreted as a genomic resistome profile. Future multicenter studies integrating complete susceptibility testing, treatment-response data, and whole-genome sequencing, including resistance gene and plasmid analyses, are needed to validate these findings and better characterize Salmonella resistance determinants. In conclusion, in this retrospective cohort of hospitalized patients with Salmonella infection, infection‑phenotype distributions varied across different sources of immunosuppression, although subgroup sizes were small and these findings are exploratory. Among adult or adult-skewed patients with available qSOFA data, early organ dysfunction reflected by qSOFA ≥ 2 was the most consistent factor associated with short-term adverse outcomes. Phenotypic resistance burden may reflect treatment complexity and possible risk of difficult clearance, but this association lacked robustness in the high-coverage sensitivity analysis. These findings support integrating host severity assessment and antimicrobial susceptibility information for risk stratification, while emphasizing the need for validation in larger multicenter cohorts with more complete susceptibility testing and genomic resistance profiling. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Acknowledgements We thank the clinical, laboratory, and data‑management teams at Fujian Medical University Affiliated First Quanzhou Hospital for their support in data collection and case review. Abbreviations Adjusted odds ratio Azithromycin Confidence interval Ciprofloxacin Clinical and Laboratory Standards Institute Cefotaxime Intensive care unit Interleukin Interquartile range Levofloxacin Matrix-assisted laser desorption/ionization time-of-flight mass spectrometry Meropenem Nontyphoidal Salmonella Polymerase chain reaction Quick Sequential Organ Failure Assessment Cefoperazone-sulbactam Strengthening the Reporting of Observational Studies in Epidemiology Trimethoprim-sulfamethoxazole Tumor necrosis factor Typhoidal Salmonella Author contributions XWW, SQY, and YJL contributed equally to this work. XPY, YJZ, and QQC conceived and supervised the study. XWW, SQY, YJL, CLC, LTZ, ZMW, WJY, ZYW, HTZ, MNX, XHH, MLG, and ZJS contributed to data collection and curation. XWW, SQY, YJL, QQC, and YJZ contributed to data analysis and interpretation. XWW, SQY, and YJL drafted the manuscript. HTZ, QQC, and XPY critically revised the manuscript. All authors read and approved the final manuscript. Funding This study was supported by the National Natural Science Foundation of China (82570707, 82370604), the Major Science and Technology Innovation Project of Fujian Province (2023Y9269), the Natural Science Foundation of Fujian Province (2023J01239), the Medical Innovation Project of Fujian Provincial Health Commission (2015CXA058), and the Climbing Project of the Medical-Education Integration Fund of Fujian Normal University (2025YJRHPD001). The funders had no role in study design, data collection, data analysis, interpretation, or manuscript preparation. Data availability The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request. Declarations Ethics approval and consent to participate The study complied with the Declaration of Helsinki and the ICMJE Recommendations and was approved by the institutional ethics committee of Fujian Medical University Affiliated First Quanzhou Hospital ([2026] K080; 19 March 2026). The requirement for informed consent was waived because of the anonymised retrospective design. All data were de-identified before analysis. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Xuewei Wang, Shuqi Yang and Yijie Lin contributed equally to this work. Contributor Information Yijuan Zheng, Email: zhengxiaoyi199011@163.com. Qingqing Chen, Email: 29100129@qq.com. Xueping Yu, Email: xpyu15@fudan.edu.cn. References Associated Data Supplementary Materials Supplementary Material 1 Data Availability Statement The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.