Abstract
At present, there is insufficient evidence to evaluate the prognosis of patients with sepsis. This study anazed the clinical data of 822 sepsis patients in the ICU of a tertiary Grade A hospital to construct and validate a nomogram model for predicting the 28-day mortality risk in sepsis patients. The model was constructed using multivariate logistic regression analysis to screen for independent risk factors affecting prognosis, and a mortality risk prediction model was built based on these independent risk factors. The performance of the model was evaluated using the Hosmer–Lemeshow test, receiver operating characteristic curve (ROC), calibration plot, and decision curve analysis (DCA). Multivariate logistic regression identified five independent risk factors for 28-day mortality in sepsis patients: Age, SOFA score, CRP, Mechanical ventilation, and the use of Vasoactive drugs. The odds ratios (OR) and 95% confidence intervals (95% CI) for these factors were 1.037 (1.024–1.050), 1.093 (1.044–1.145), 1.034 (1.026–1.042), 1.967 (1.176–3.328), and 2.515 (1.611–3.941), respectively, with all P-values < 0.05. Based on these five independent risk factors, a nomogram model was constructed, with the area under the ROC curve (AUC) in the training set and external validation set being 0.849 (95% CI 0.818–0.880) and 0.837 (95% CI 0.887–0.886), respectively. Both the DCA curve and calibration plot confirmed that the model has good clinical efficacy. The nomogram prediction model established in this study has excellent predictive ability, which can help clinicians identify high-risk patients early and provide guidance for clinical decision-making.
Keywords: Sepsis, Mortality risk factors, Nomogram model, Prognostic outcome, Area under the receiver operating characteristic curve, Intensive care unit
Subject terms: Diseases, Risk factors
Introduction
Sepsis is a life-threatening organ dysfunction caused by infection, posing a severe threat not only to human life and health but also imposing a heavy medical burden on society and families1–3. According to the 2020 study published in The Lancet, in 2017, sepsis and sepsis-related deaths accounted for 19.7% of global mortality4, Although there is currently a lack of specific treatment protocols for sepsis, implementing preventive and therapeutic strategies early for patients with poor prognoses is crucial for improving patient survival rates5.
In previous research on prognostic nomogram models for septic patients, certain achievements have been made, but there were limitations regarding the study subjects and sample size6,7.With the continuous development of critical care public databases and advancements in research algorithms, more and more prediction model development algorithms have been applied by scholars to explore the occurrence and prognosis of diseases8,9.Additionally, recent studies have successfully applied machine learning-related research based on cutting-edge technologies in medical diagnostics10,11,yet there is still a lack of evidence in the prognosis research for septic patients12,13. On the other hand, most prognosis studies for septic patients based on critical care public databases focus on foreign public databases. Whether these studies are applicable to local clinical practice requires in-depth exploration based on the local population. In summary, this provides a theoretical basis for further improvement needs.
Therefore, the first major contribution of this study is based on the Sepsis-3.0 definition and criteria, to collect and analyze clinical data of sepsis patients in this region, identify the risk factors for short-term mortality in sepsis patients, and construct a nomogram model. This study proposes a more simple and effective nomogram model that can be widely applied in resource-limited environments, aiming to more quickly and accurately identify high-risk patients and thereby mitigate the threat posed by sepsis. Secondly, by integrating commonly used multi-source clinical data resources, including basic patient information, clinical indicators, and laboratory test results, the predictive accuracy of the model is enhanced. This not only better reflects the overall health status of patients but also provides more reliable predictive results in various levels of healthcare environments. Thirdly, through comprehensive data analysis, the study explores independent prognostic risk factors for sepsis patients and utilizes these factors to construct a predictive model, thereby enhancing the model’s clinical decision support capabilities.
Finally, to better reflect the logical and structural aspects of this study, the following outline will be presented: The first part will review the current state of research and its limitations; the second part will describe the design and methodology of this study, including detailed steps for data collection, processing, and model construction; the third part will present and analyze the predictive performance and clinical value of the model; the fourth part will discuss the findings and propose directions and recommendations for future research; and finally, the conclusion section will summarize the main contributions of this study and its potential impact on the management of sepsis patients.
Materials and methods
Study subjects
The clinical dataset of sepsis patients treated in the ICU of a tertiary hospital in Xinjiang Province from January 2018 to June 2022 was retrospectively collected as the training set. Additionally, the clinical dataset of sepsis patients treated in the ICU of a prefecture-level tertiary hospital in Xinjiang from January 2023 to June 2024 was prospectively included as the validation set. All research methods were performed in accordance with the guidelines and regulations of the Nature Portfolio journal.
Inclusion and exclusion criteria
Inclusion criteria were: (1) patients meeting the definition and diagnostic criteria of Sepsis 3.01; (2) age ≥ 18 years. Exclusion criteria were: (1) patients during pregnancy and lactation; (2) patients with ICU stay < 24 h; (3) patients with advanced malignant tumors or other terminal illnesses; (4) patients with incomplete clinical data collection. This study has been approved by the Hospital Medical Ethics Committee (Approval No.: ZZQ202301160007). In addition, due to the nature of the retrospective cohort, the Medical Ethics Committee of Changji State People’s Hospital has waived the requirement for informed consent. However, in the prospective validation cohort, informed consent was obtained from all study subjects or their legal guardians.
Main clinical features included
Through literature review and expert consultation, a general information form was designed. The data collection was conducted by four trained and assessed graduate students using the hospital’s electronic medical record system. The collected data included: (1) general information (age, gender, site of infection, comorbidities); (2) relevant laboratory indicators (blood gas analysis, vital signs, blood routine, liver and kidney function, coagulation function, etc.); (3) treatment measures (whether mechanical ventilation, vasoactive drugs, or surgery were used during the ICU stay); (4) organ function status scores, among 58 routine clinical features. The primary outcome variable in this study was a binary variable, indicating whether sepsis patients died within 28 days. All data were cross-checked by two individuals, and missing values were imputed using multiple imputation methods.
Statistical methods
SPSS 25.0 statistical software was used for data statistical analysis, count data in general information were represented as frequency and percentage, and comparison between groups was performed using χ2 test or Fisher exact probability method; measurement data conforming to normal distribution were represented as mean ± standard deviation, comparison between two groups was performed using independent samples t-test; The measurement data that did not conform to the normal distribution were described by the median (quartile) [M (P25, P75)], and the Mann–Whitney U test was used to compare the groups. Statistically signifcant variables were included in the multivariate analyses. Multivariate logistic regression was used to screen out the independent prognostic risk factors to construct the prediction model, and the visualized analysis was carried out with the nomogram model. The performance of the model was evaluated by drawing receiver operating characteristic (ROC) and calibration curve.
Results
Basic characteristics of sepsis patients in different groups
Based on the inclusion and exclusion criteria, this study enrolled a total of 822 sepsis patients, with 580 in the training set and 242 in the validation set. The patients were divided into survival and death groups according to whether they died within 28 days. The baseline data of the two groups were compared to identify differences. Significant differences were observed in the following characteristics between the groups: Age, Cardiovascular, Mechanical Ventilation, Vasoactive Drugs, CRRT, Diastolic Blood Pressure, PaO2/FiO2, PH, Lac, K+, CRP, NEUT, WBC, HB, INR, APTT, BNP, BUN, SCr, AST/ALT, CHE, BA, APACHEII, SOFA, and GCS. Tables 1 and 2 provide the clinical characteristics of all enrolled study subjects.
Table 1.
Comparison of baseline characteristics of patients in the training set.
| Variables | Survival group (n = 324) | Death group (n = 256) | Statistic t/Z/χ2 | p-Value |
|---|---|---|---|---|
| Demographic characteristics | ||||
| Age(years) | 58.12 ± 17.54 | 67.79 ± 16.52 | − 6.76 | < 0.001 |
| Male cases [(%)] | 208 (64.20) | 176 (68.75) | 1.32 | 0.250 |
| Comorbid disease cases [(%)] | ||||
| Cerebrovascular | 13 (4.01) | 8 (3.12) | 0.32 | 0.570 |
| Cardiovascular | 41 (12.65) | 52 (20.31) | 6.23 | 0.013 |
| Hypertension | 113 (34.88) | 108 (42.19) | 3.24 | 0.072 |
| Diabetes | 85 (26.23) | 62 (24.22) | 0.31 | 0.579 |
| Chronic kidney | 22 (6.79) | 22 (8.59) | 0.66 | 0.415 |
| Cancer | 18 (5.56) | 23 (8.98) | 2.56 | 0.110 |
| Site of infection cases [(%)] | ||||
| Pulmonary | 170 (52.47) | 134 (52.34) | 0.00 | 0.976 |
| Abdomen | 55 (16.98) | 54 (21.09) | 1.59 | 0.207 |
| Horacic cavity | 14 (4.32) | 9 (3.52) | 0.24 | 0.622 |
| Urinary system | 44 (13.58) | 34 (13.28) | 0.01 | 0.917 |
| Bloodborne | 14 (4.32) | 7 (2.73) | 1.03 | 0.310 |
| Skin and soft tissue | 29 (8.95) | 18 (7.03) | 0.71 | 0.400 |
| Treatment measures cases [(%)] | ||||
| Mechanical ventilation | 205 (63.27) | 220 (85.94) | 37.52 | < 0.001 |
| Vasoactive drugs | 108 (33.33) | 179 (69.92) | 76.59 | < 0.001 |
| CRRT | 53 (16.36) | 67 (26.17) | 8.39 | 0.004 |
| Surgical treatment | 44 (13.58) | 35 (13.67) | 0.00 | 0.975 |
| Vital signs | ||||
| Temperature (°C) | 37.12 ± 0.79 | 37.14 ± 0.87 | − 0.32 | 0.751 |
| Prognostic markers (day) | ||||
| ICU length of stay | 11.00 (7.00, 20.00) | 11.00 (7.00, 19.00) | − 0.32 | 0.750 |
| Heart rate | 91.00 (78.00, 110.00) | 90.00 (78.00, 109.00) | − 0.23 | 0.814 |
| Respiration | 20.00 (19.00, 22.00) | 20.00 (20.00, 22.00) | − 1.4 | 0.160 |
| Systolic blood pressure | 128.00 (117.00, 140.25) | 123.50 (111.00, 140.00) | − 1.05 | 0.293 |
| Diastolic blood pressure | 78.00 (67.00, 80.00) | 75.00 (63.00, 80.00) | − 2.03 | 0.042 |
| Aboratory markers | ||||
| PaO2(mmHg) | 104.50 (83.00, 148.25) | 112.00 (83.00, 143.00) | − 0.16 | 0.871 |
| FiO2(%) | 45.00 (37.00, 50.00) | 45.00 (37.00, 60.00) | − 0.74 | 0.460 |
| PaCO2(mmHg) | 32.00 (27.75, 37.25) | 33.00 (28.00, 40.00) | − 1.17 | 0.241 |
| PaO2/FiO2 (mmHg) | 257.50 (198.00, 333.25) | 242.00 (167.50, 321.25) | − 2 | 0.046 |
| PH | 7.42 (7.38, 7.46) | 7.40 (7.34, 7.45) | − 2.55 | 0.011 |
| Lac(mmol/L) | 1.60 (1.20, 2.40) | 1.80 (1.40, 2.52) | − 3.49 | < 0.001 |
| Na+ (mmol/L) | 135.00 (130.00, 139.00) | 136.00 (131.00, 140.00) | − 1.36 | 0.174 |
| K+ (mmol/L) | 3.90 (3.50, 4.40) | 4.00 (3.60, 4.50) | − 2.57 | 0.010 |
| MONO(109/L) | 0.68 (0.40, 0.92) | 0.66 (0.36, 0.92) | − 0.22 | 0.827 |
| CRP(mg/L) | 34.64 (22.78, 70.93) | 84.00 (54.17, 90.00) | − 11.1 | < 0.001 |
| IL-6(pg/ml) | 88.44 (35.80, 284.90) | 103.58 (47.00, 328.23) | − 1.36 | 0.175 |
| PCT(ng/mL) | 2.04 (0.34, 16.20) | 1.79 (0.44, 16.24) | − 0.29 | 0.769 |
| WBC(109/L) | 11.85 (8.52, 16.68) | 12.70 (8.67, 16.83) | − 0.82 | 0.410 |
| NEUT(109/L) | 9.17 (6.50, 14.54) | 12.38 (8.58, 18.67) | − 4.83 | < 0.001 |
| LYMPH(109/L) | 0.81 (0.48, 1.29) | 0.79 (0.49, 1.31) | − 0.3 | 0.763 |
| RBC(109/L) | 3.49 (2.73, 4.06) | 3.22 (2.50, 3.92) | − 3.03 | 0.002 |
| HB(g/L) | 103.00 (82.00, 121.25) | 96.00 (74.00, 116.25) | − 2.27 | 0.023 |
| PLT(109/L) | 160.50 (101.00, 241.25) | 150.00 (93.00, 213.00) | − 1.51 | 0.131 |
| PT(s) | 14.65 (12.80, 18.12) | 14.40 (12.78, 17.52) | − 0.48 | 0.629 |
| INR | 1.19 (1.09, 1.37) | 1.27 (1.11, 1.52) | − 2.54 | 0.011 |
| FIB(g/L) | 4.10 (3.17, 4.78) | 4.02 (3.32, 4.74) | − 0.22 | 0.824 |
| APTT(s) | 32.80 (28.80, 36.00) | 33.85 (29.82, 38.62) | − 2.34 | 0.019 |
| BNP(ng/mL) | 1350.00 (311.00, 6185.00) | 2795.00 (514.00, 9280.00) | − 3.21 | 0.001 |
| BUN(mmol/L) | 9.62 (6.01, 17.89) | 12.71 (7.98, 20.36) | − 3.6 | < 0.001 |
| SCr(μmol/L) | 85.71 (54.60, 178.89) | 102.10 (62.77, 221.88) | − 2.07 | 0.039 |
| TBIL(μmol/L) | 21.88 (14.31, 34.14) | 22.84 (14.38, 34.14) | − 0.21 | 0.834 |
| DBIL(μmol/L) | 0.30 (0.30, 3.45) | 0.30 (0.30, 4.79) | − 0.74 | 0.460 |
| IBIL(μmol/L) | 8.64 (5.55, 15.95) | 8.29 (5.23, 15.27) | − 0.75 | 0.456 |
| TB(g/L) | 58.05 (52.46, 64.00) | 58.53 (53.02, 64.94) | − 0.22 | 0.827 |
| ALB(g/L) | 29.42 (26.13, 33.26) | 29.26 (25.58, 32.24) | − 1.22 | 0.221 |
| GLB(g/L) | 28.72 (25.28, 32.04) | 29.11 (25.49, 33.13) | − 1.25 | 0.211 |
| AST(u/L) | 52.52 (32.12, 125.45) | 52.39 (32.58, 108.74) | − 0.01 | 0.992 |
| AST/ALT(u/L) | 1.67 (1.25, 2.29) | 1.90 (1.35, 2.60) | − 2.43 | 0.015 |
| GGT(u/L) | 42.26 (24.23, 78.85) | 45.70 (25.00, 84.85) | − 0.59 | 0.554 |
| ALP(Iu/L) | 83.39 (58.12, 119.34) | 85.74 (60.92, 124.82) | − 0.83 | 0.406 |
| CHE(u/L) | 3414.93 (2305.59, 4759.54) | 2807.20 (1969.79, 3913.73) | − 3.82 | < 0.001 |
| BA(μmol/L) | 27.34 (11.77, 42.74) | 33.28 (18.91, 48.77) | − 2.75 | 0.006 |
| Scoring system (score) | ||||
| APACHEII | 14.00 (10.00, 20.00) | 17.00 (13.00, 22.00) | − 4.86 | < 0.001 |
| SOFA | 6.00 (4.00, 9.00) | 9.00 (6.00, 14.00) | − 7.19 | < 0.001 |
| GCS | 14.00 (9.00, 15.00) | 12.00 (8.00, 14.00) | − 2.81 | 0.005 |
t, t-test; Z, Mann–Whitney test; χ2, Chi-square test; SD, standard deviation; M, median; Q1, 1st quartile; Q3, 3st quartile.
Table 2.
Comparison of baseline characteristics of patients in the validation set.
| Variables | Survival group (n = 127) | Death group (n = 115) | Statistic t/Z/χ2 | P |
|---|---|---|---|---|
| Demographic characteristics | ||||
| Age(years) | 55.53 ± 16.63 | 66.88 ± 13.66 | − 5.82 | < 0.001 |
| Male cases [(%)] | 72 (56.69) | 77 (66.96) | 2.69 | 0.101 |
| Comorbid disease cases [(%)] | ||||
| Cerebrovascular | 9 (7.09) | 6 (5.22) | 0.36 | 0.547 |
| Cardiovascular | 15 (11.81) | 14 (12.17) | 0.01 | 0.931 |
| Hypertension | 43 (33.86) | 44 (38.26) | 0.51 | 0.476 |
| Diabetes | 31 (24.41) | 31 (26.96) | 0.21 | 0.650 |
| Chronic kidney | 4 (3.15) | 3 (2.61) | 0.00 | 1.000 |
| Cancer | 13 (10.24) | 18 (15.65) | 1.58 | 0.208 |
| Site of infection cases [(%)] | ||||
| Pulmonary | 60 (47.24) | 53 (46.09) | 0.03 | 0.857 |
| Abdomen | 36 (28.35) | 43 (37.39) | 2.25 | 0.134 |
| Horacic cavity | 4 (3.15) | 1 (0.87) | 0.63 | 0.428 |
| Urinary system | 6 (4.72) | 5 (4.35) | 0.02 | 0.888 |
| Bloodborne | 1 (0.79) | 4 (3.48) | 1.03 | 0.309 |
| Skin and Soft Tissue | 17 (13.39) | 12 (10.43) | 0.50 | 0.480 |
| Treatment measures cases [(%)] | ||||
| Mechanical ventilation | 95 (74.80) | 105 (91.30) | 11.46 | < 0.001 |
| Vasoactive drugs | 55 (43.31) | 93 (80.87) | 35.85 | < 0.001 |
| CRRT | 20 (15.75) | 40 (34.78) | 11.73 | < 0.001 |
| Surgical treatment | 16 (12.60) | 13 (11.30) | 0.10 | 0.757 |
| Vital signs | ||||
| Temperature (°C) | 37.04 ± 0.82 | 37.18 ± 0.99 | − 1.16 | 0.247 |
| Heart rate | 91.00 (77.00, 105.00) | 86.00 (77.00, 105.00) | − 0.59 | 0.558 |
| Respiration | 19.00 (19.00, 20.50) | 19.00 (19.00, 22.00) | − 0.91 | 0.364 |
| Systolic blood pressure | 130.00 (120.00, 140.00) | 121.00 (110.00, 139.50) | − 2.10 | 0.035 |
| Diastolic blood pressure | 78.00 (68.50, 83.50) | 75.00 (60.50, 80.00) | − 2.16 | 0.031 |
| Aboratory markers | ||||
| PaO2(mmHg) | 104.00 (79.50, 149.50) | 118.00 (86.50, 142.50) | − 0.90 | 0.366 |
| FiO2(%) | 41.00 (37.00, 50.00) | 44.00 (37.00, 60.00) | − 1.38 | 0.167 |
| PaCO2(mmHg) | 32.00 (27.00, 37.00) | 33.00 (28.00, 39.00) | − 1.04 | 0.297 |
| PaO2/FiO2 (mmHg) | 260.00 (201.00, 344.00) | 257.00 (180.00, 342.00) | − 0.53 | 0.594 |
| PH | 7.41 (7.37, 7.45) | 7.40 (7.34, 7.46) | − 0.69 | 0.493 |
| Lac(mmol/L) | 1.70 (1.40, 2.80) | 2.10 (1.60, 3.35) | − 1.96 | 0.050 |
| Na+ (mmol/L) | 135.00 (130.00, 138.00) | 136.00 (132.00, 141.53) | − 1.98 | 0.048 |
| K+ (mmol/L) | 3.90 (3.52, 4.40) | 4.10 (3.60, 4.40) | − 0.73 | 0.463 |
| MONO(109/L ) | 0.67 (0.39, 0.97) | 0.56 (0.31, 0.89) | − 1.53 | 0.127 |
| CRP(mg/L) | 58.30 (32.05, 73.15) | 82.00 (63.00, 90.00) | − 6.25 | < 0.001 |
| IL-6(pg/ml) | 66.55 (31.08, 200.05) | 137.20 (54.10, 447.46) | − 3.42 | < 0.001 |
| PCT(ng/mL) | 0.50 (0.13, 4.54) | 3.17 (0.43, 16.73) | − 3.67 | < 0.001 |
| WBC(109/L) | 12.23 (8.70, 17.34) | 12.06 (8.16, 16.31) | − 1.22 | 0.224 |
| NEUT(109/L) | 10.64 (7.36, 15.31) | 10.98 (6.42, 14.68) | − 0.77 | 0.442 |
| LYMPH(109/L) | 0.75 (0.49, 1.21) | 0.79 (0.48, 1.35) | − 0.14 | 0.888 |
| RBC(109/L) | 3.51 (2.95, 4.29) | 3.27 (2.58, 4.04) | − 2.25 | 0.024 |
| HB(g/L) | 109.00 (86.50, 121.00) | 96.00 (73.50, 115.00) | − 2.44 | 0.015 |
| PLT(109/L ) | 183.00 (129.00, 263.00) | 137.00 (72.00, 209.50) | − 3.59 | < 0.001 |
| PT(s) | 13.50 (12.35, 15.25) | 14.90 (12.95, 17.20) | − 2.71 | 0.007 |
| INR | 1.17 (1.09, 1.31) | 1.30 (1.15, 1.52) | − 3.46 | < 0.001 |
| FIB(g/L) | 3.68 (2.80, 4.67) | 3.71 (2.77, 4.67) | − 0.65 | 0.513 |
| APTT(s) | 31.70 (28.25, 38.35) | 33.80 (29.40, 39.35) | − 1.78 | 0.075 |
| BNP(ng/mL) | 939.00 (253.50, 4445.00) | 2240.00 (574.50, 7010.00) | − 3.15 | 0.002 |
| BUN(mmol/L) | 8.67 (5.35, 13.93) | 13.94 (7.84, 21.80) | − 4.67 | < 0.001 |
| SCr(μmol/L) | 73.25 (49.97, 157.44) | 136.59 (66.41, 241.69) | − 3.37 | < 0.001 |
| TBIL(μmol/L) | 18.14 (12.96, 28.48) | 26.10 (15.20, 43.97) | − 3.22 | 0.001 |
| DBIL(μmol/L) | 0.30 (0.30, 0.38) | 0.30 (0.30, 7.83) | − 2.38 | 0.017 |
| IBIL(μmol/L) | 8.21 (5.55, 15.64) | 9.70 (5.46, 18.34) | − 1.17 | 0.242 |
| TB(g/L) | 57.30 (52.22, 64.03) | 57.24 (50.27, 62.75) | − 1.09 | 0.277 |
| ALB(g/L) | 29.67 (25.45, 33.89) | 28.30 (23.91, 32.05) | − 1.65 | 0.099 |
| GLB(g/L) | 28.40 (24.97, 32.09) | 28.26 (24.99, 32.58) | − 0.04 | 0.967 |
| AST(u/L) | 45.49 (31.76, 90.52) | 52.53 (37.10, 106.46) | − 1.70 | 0.090 |
| AST/ALT(u/L) | 1.65 (1.11, 2.48) | 1.90 (1.32, 2.61) | − 1.42 | 0.155 |
| GGT(u/L) | 38.56 (24.19, 86.88) | 41.13 (23.45, 79.91) | − 0.19 | 0.848 |
| ALP(Iu/L) | 77.70 (59.95, 115.68) | 82.17 (55.83, 112.12) | − 0.60 | 0.549 |
| CHE(u/L) | 4065.95 (2652.43, 5192.73) | 2654.78 (1809.35, 3928.22) | − 4.85 | < 0.001 |
| BA(μmol/L) | 28.87 (15.49, 55.31) | 35.28 (18.31, 57.58) | − 0.96 | 0.338 |
| Scoring system (score) | ||||
| APACHEII | 13.00 (10.00, 17.00) | 17.00 (14.00, 23.00) | − 4.74 | < 0.001 |
| SOFA | 8.00 (6.00, 11.00) | 13.00 (9.00, 17.00) | − 7.04 | < 0.001 |
| GCS | 14.00 (10.00, 15.00) | 12.00 (9.00, 14.00) | − 3.01 | 0.003 |
| Prognostic markers (day) | ||||
| ICU length of stay | 11.00 (7.00, 21.00) | 11.00 (7.00, 19.00) | Z = − 0.06 | 0.955 |
T,t-test; Z, Mann–Whitney test; χ2, Chi-square test; SD, standard deviation; M, median; Q1, 1st quartile; Q3, 3st quartile.
Multivariate logistic regression analysis for identifying independent risk factors and developing a risk prediction model
All 25 variables with significant differences from the training set were included in the multivariate logistic regression analysis. The results indicated that age, SOFA score, CRP, mechanical ventilation, and vasoactive drugs were independent risk factors for 28-day mortality in sepsis patients (all P < 0.05), See Table 3.The logistic regression equation is as follows: LogisticP = − 6.200 + 0.036(Age) + 0.089(SOFA) + 0.033(CRP) + 0.676(MV) + 0.922(Vasoactive Drugs), This model was represented as a nomogram. Figure 1 illustrates the nomogram, where the scores for each patient’s indicators are determined by projecting a vertical line from each variable to the score axis. By summing all the scores, a total score is obtained, which corresponds to the probability of mortality within 28 days for sepsis patients. This model provides a valuable tool to assist clinical decision-making in the early stages of patient management. Figure 2 presents the SHAP summary plot for risk factors associated with 28-day mortality in sepsis patients. The features are ranked based on the impact of their SHAP values on the outcome, with higher combined values indicating a higher risk of death within 28 days. This plot provides a more intuitive understanding of the relationship between the model and the variables. Red signifies high feature values, purple represents values close to the overall mean, and blue indicates low feature values.
Table 3.
Multivariate logistic regression analysis of risk factors for mortality in sepsis patients in the training set.
| Predictor | Estimate | SE | Z | p | Odds ratio | Lower | Upper |
|---|---|---|---|---|---|---|---|
| (Intercept) | − 6.200 | 0.577 | − 10.749 | 0.001 | 0.002 | 0.001 | 0.006 |
| Age | 0.036 | 0.006 | 5.787 | 0.001 | 1.037 | 1.024 | 1.050 |
| SOFA | 0.089 | 0.024 | 3.750 | 0.001 | 1.093 | 1.044 | 1.145 |
| CRP | 0.033 | 0.004 | 8.545 | 0.001 | 1.034 | 1.026 | 1.042 |
| MV | 0.676 | 0.265 | 2.554 | 0.011 | 1.967 | 1.176 | 3.328 |
| Vasoactive Drugs | 0.922 | 0.228 | 4.049 | 0.001 | 2.515 | 1.611 | 3.941 |
Fig. 1.
Nomogram for predicting mortality in sepsis patients.
Fig. 2.
SHAP value summary plot of the prediction model.
Internal and external validation and evaluation of the prediction model
The ROC- curves for both the training and validation sets indicated that the area under the curve (AUC) for the prediction model is 0.849[95% confidence interval (95% CI 0.818–0.880)] and 0.837 (95% CI 0.787–0.886), respectively (Fig. 3). The calibration curves demonstrate consistency between the model’s predicted outcomes and the observed results (Fig. 4). The Brier Score for the calibration curve in the training set is 0.157, while in the validation set it is 0.163; The Hosmer–Lemeshow test yielded P values of 0.057 and 0.260, respectively, indicating no significant differences;The decision curve analysis (DCA) suggests that the prediction model provides greater net benefits compared to both “treat all” and “treat none” strategies (Fig. 5). The results of the CIC curve showed (Fig. 6) that the number of high-risk patients (the number of deaths predicted by the model) highly matched the number of high-risk patients (the number of true positive deaths).
Fig. 3.
ROC curves for the risk prediction model of mortality in sepsis patients in the training set (A) and validation set (B).
Fig. 4.
Calibration curves for the risk prediction model of mortality in sepsis patients in the training set (A) and validation set (B).
Fig. 5.
Decision curve analysis (DCA) for the risk prediction model of mortality in sepsis patients in the training set (A) and validation set (B).
Fig. 6.
Cumulative incidence curves (CIC) for the risk prediction model of mortality in sepsis patients in the training set (A) and validation set (B).
Discussion
Sepsis can occur as a result of infections in any part of the body, characterized by rapid progression, high incidence, and mortality rates, which pose a significant threat to human health and increase the burden on healthcare systems. Currently, the management of sepsis primarily focuses on early identification and dynamic assessment of the severity of the disease to provide appropriate treatment, thereby improving the unfavorable outcomes of patients14,15. This study comprehensively collected routine clinical characteristics of sepsis patients, used multivariate Logistic regression to screen out five independent risk factors for short-term outcomes in sepsis patients, constructed a nomogram model for the risk of death within 28 days in sepsis patients, and provided a SHAP value summary plot that illustrates the feature values influencing the model’s output. This work provides a basis for early identification of patients with unfavorable outcomes and for guiding clinical decision-making.
For the development and application of predictive models, a critical issue that needs to be addressed is that most scholars’ research is limited to theoretical aspects. However, in clinical practice, there is a greater need for the direct application of predictive models to analyze patient conditions. Unfortunately, the research and deployment of these models have not fully met clinical expectations. Logistic regression models offer mature advantages such as simplicity, high efficiency, and low cost16. Nomogram models can transform complex statistical models into intuitive graphics and can integrate multiple clinical features, allowing for individualized risk assessment and decision support based on the patient’s specific circumstances. This enhances the credibility of the model17,18. Previous studies have shown that the prediction model for sepsis combined with pulmonary infection, constructed based on a nomogram, has an area under the ROC curve (AUC) of 0.743, which is significantly superior to the AUC of 0.647 for the traditional SOFA score prognosis evaluation tool19.Therefore, in this study, we used multivariate logistic regression to identify independent risk factors affecting prognosis and constructed a mortality risk nomogram model. The area under the ROC curve for the training set and the external validation set was 0.849 and 0.837, respectively. Both the DCA curve and the CIC curve also demonstrated that the model has good clinical utility, indicating that the developed nomogram model is reliable and can assist clinicians in assessing disease progression and prognosis in sepsis patients. Previous studies have achieved certain results in evaluating disease progression and prognosis in sepsis patients20,21, but these two studies focused only on sepsis patients who developed acute respiratory distress syndrome and related acute kidney injury, and the study data were sourced from critical care public databases. The study populations may not fully align with the latest definitions and standards of sepsis as per the current guidelines. Unlike the above two studies, our research is based on the sepsis 3.0 definition and diagnostic criteria. The clinical features included in our study are derived from clinical medical records, which are simple and easy to obtain, facilitating broader application in clinical practice. Additionally, the data for both the training set and the validation set were sourced from different centers, enhancing the model’s generalization ability. Therefore, the results may be more applicable to the prognosis assessment of sepsis patients in this region, but further validation through large-scale studies is still needed to confirm the model’s stability and generalization ability.
The model constructed in this study incorporates five easily obtainable and critically important indicators in clinical practice: Age, SOFA score, CRP, Mechanical ventilation, and Vasoactive drugs. These features support previous studies on sepsis prognosis, indicating the relevance and reliability of the included characteristics. Age is a crucial factor influencing sepsis prognosis22,23, As individuals age, their immune function gradually declines, increasing the risk of infections and making the infection process more prolonged and severe in older patients. Additionally, cognitive decline, malnutrition, and mobility issues further increase the mortality risk in older patients with sepsis24.The SOFA score is a widely recognized tool for assessing the degree of organ dysfunction and has demonstrated good predictive performance in predicting patient outcomes25, Our findings align with previous studies, indicating that the SOFA score has the ability to predict sepsis patient outcomes26.CRP is an acute-phase protein synthesized and secreted by the liver. Upon inflammatory stimulation, it is rapidly synthesized and secreted into the bloodstream, reflecting the severity of early sepsis infection27. In the early stages of sepsis, if local infection is not effectively controlled, innate immune cells are overactivated, leading to a cytokine storm. In later stages, due to the excessive activation of immune cells in the early phase, leading to injury or death, a state of prolonged immune suppression and reduced immune function develops, eventually resulting in severe infections or recurrent sepsis28, Numerous studies have confirmed that CRP is a significant risk factor for sepsis29,30, This suggests that dynamic assessment of the SOFA score combined with CRP regulation could help mitigate adverse outcomes in sepsis.Mechanical ventilation is one of the primary treatment techniques for ICU patients. However, it can cause lung injury, including direct mechanical damage and indirect systemic inflammatory responses31. Research indicates that mechanical ventilation, along with the age of critically ill patients, is closely associated with mortality rates32。Sepsis-3 defines septic shock as a subtype of sepsis, occurring when initial, aggressive goal-directed fluid resuscitation fails to maintain optimal circulation, necessitating the use of vasoactive drugs to ensure adequate tissue perfusion pressure. These observations suggest that the use of mechanical ventilation and vasoactive drugs can serve as important indicators for sepsis prognosis. This study further confirms that patients in the mortality group who used mechanical ventilation and vasoactive drugs were more prevalent than those in the survival group. Tables 1 and 2 summarize the comorbidities of the study subjects, including cerebrovascular disease, cardiovascular disease, hypertension, diabetes, chronic kidney disease, and cancer. All comorbidities were diagnosed by the attending physician according to strict guideline standards and are documented in the electronic medical records. The investigators collected and verified this information.Notably, in Table 2, the mortality rate in the group of sepsis patients with hypertension who died is higher than in the survival group. This may be related to poor control of baseline blood pressure and the potential damage to target organs caused by long-term hypertension. When sepsis occurs, the adverse impact on the emergency protective response may be a factor contributing to the slightly higher mortality rate in patients with this comorbidity in the death group. However, this did not reach statistical significance in this study, and we will explore this further in future research.Furthermore, as shown in Fig. 1, each feature corresponds to a specific individual score at different values. Based on the total score, personalized predictions can be made regarding the risk of death within 28 days for sepsis patients. This allows for early identification of patients with poor prognosis, enabling timely implementation of effective clinical decisions.
Limitations analysis
This study has certain limitations. The training set was based on retrospective research, which may introduce some confounding factors. Additionally, the data collection spanned a considerable time period, potentially influenced by advancements in the clinical treatment of sepsis. Therefore, it is essential to conduct multicenter, prospective cohort studies in the future to further refine the model’s ability to assess sepsis patients’ conditions and predict mortality.
Conclusion
In summary, this study demonstrates that age, SOFA score, CRP, mechanical ventilation, and vasoactive drugs are independent risk factors for mortality in sepsis patients. The nomogram model developed based on risk factors demonstrates high accuracy in both the training set and external validation set. It offers advantages in early warning for poor prognosis in sepsis patients and in assessing their risk of death. Additionally, improving treatment strategies for risk factors in the predictive model has a positive impact on the prognosis of such patients.
Supplementary Information
Author contributions
Conceptualization: Yanjie YANG, Ge LIN, Huiling ZHAO, Shupeng Liu, Sun yue. Methodology: Ge LIN, Huiling ZHAO. Writing—original draft: Yanjie YANG, li zhang,Huiling ZHAO, Writing—review and editing: Yanjie YANG, Xin Gu, Hu Peng, Shupeng Liu, Sun yue. Funding acquisition: li zhang All authors reviewed the manuscript.
Funding
Tis study was funded by the 2022 Critical Care Medicine Runze Fund of the Wu Jieping Medical Foundation (320.6750.2023–02-3). Training Program for Outstanding Talents and Innovative Teams at the First Affiliated Hospital of Xinjiang Medical University (cxtd202414).
Data availability
All of our data can be obtained by uploading a supporting message or through the corresponding author.
Declarations
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.
Yanjie Yang and Xin Gu contributed equally to this work.
Supplementary Information
The online version contains supplementary material available at 10.1038/s41598-025-89442-x.
References
- 1.Singer, M. et al. The third international consensus definitions for sepsis and septic shock (Sepsis-3). JAMA315(8), 801–810 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Rivers, E. et al. Early goal-directed therapy in the treatment of severe sepsis and septic shock. N. Engl. J. Med.345(19), 1368–1377 (2001). [DOI] [PubMed] [Google Scholar]
- 3.Yealy, D. M. et al. A randomized trial of protocol-based care for early septic shock. N. Engl. J. Med.370(18), 1683–1693 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Rudd, K. E. et al. Global, regional, and national sepsis incidence and mortality, 1990–2017: Analysis for the Global Burden of Disease Study. Lancet395(10219), 200–211 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Xu, X. X. & Po, J. Y. Research progress on early antibiotic treatment in sepsis patients. J. Pract. Med.39(10), 1311–1315 (2023). [Google Scholar]
- 6.Peng, M., Deng, F. & Qi, D. Development of a nomogram model for the early prediction of sepsis-associated acute kidney injury in critically ill patients. Sci. Rep.13(1), 15200 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Wang, X. et al. Development and validation of a nomogram model for predicting 28-day mortality in patients with sepsis. Heliyon10(16), e35641 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Dong, Z. et al. Prediction of 3-year risk of diabetic kidney disease using machine learning based on electronic medical records. J. Transl. Med.20(1), 143 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Luo, X. Q. et al. Machine learning for early discrimination between transient and persistent acute kidney injury in critically ill patients with sepsis. Sci. Rep.11(1), 20269 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Minic, A. et al. Applying recurrent neural networks for anomaly detection in electrocardiogram sensor data. Sensors (Basel)23(24), 9878 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Rao, G. M. et al. AttGRU-HMSI: Enhancing heart disease diagnosis using hybrid deep learning approach. Sci. Rep.14(1), 7833 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Yao, R. Q. et al. A machine learning-based prediction of hospital mortality in patients with postoperative sepsis. Front. Med. (Lausanne)7, 445 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Bao, C., Deng, F. & Zhao, S. Machine-learning models for prediction of sepsis patients mortality. Med. Intensiva (Engl. Ed.)47(6), 315–325 (2023). [DOI] [PubMed] [Google Scholar]
- 14.Bloch, E. et al. Machine learning models for analysis of vital signs dynamics: A case for sepsis onset prediction. J. Healthc. Eng.2019, 5930379 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Yang Z. W., Shan Y. L., du Xu, Z. et al. Comparative study of sepsis mortality prediction models based on machine learning. Chin. Med. Sci. J.37(3), 201–209, front insert 204. (2022). [DOI] [PubMed]
- 16.Zhang, X. et al. Structured sparse logistic regression with application to lung cancer prediction using breath volatile biomarkers. Stat. Med.39(7), 955–967 (2020). [DOI] [PubMed] [Google Scholar]
- 17.Ren, Y. et al. Development and validation of a nomogram for predicting Mycoplasma pneumoniae pneumonia in adults. Sci. Rep.12(1), 21859 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Shao, C. et al. A nomogram for predicting pancreatic mucinous cystic neoplasm and serous cystic neoplasm. Abdom. Radiol. (NY)46(8), 3963–3973 (2021). [DOI] [PubMed] [Google Scholar]
- 19.Ren, Y. et al. Risk factor analysis and nomogram for predicting in-hospital mortality in ICU patients with sepsis and lung infection. BMC Pulm. Med.22(1), 17 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Xu, C. et al. A prediction model for predicting the risk of acute respiratory distress syndrome in sepsis patients: A retrospective cohort study. BMC Pulm. Med.23(1), 78 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Hu, H. et al. A prediction model for assessing prognosis in critically Ill patients with sepsis-associated acute kidney injury. Shock56(4), 564–572 (2021). [DOI] [PubMed] [Google Scholar]
- 22.Tian, H. C. et al. Epidemiology of Sepsis-3 in a sub-district of Beijing: Secondary analysis of a population-based database. Chin. Med. J. (Engl.)132(17), 2039–2045 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Yue, P. Z. et al. Research progress on risk factors for mortality in sepsis patients. Chin. J. Integr. Med. Emerg.27(4), 505–509 (2020). [Google Scholar]
- 24.Rowe, T. A. & Mckoy, J. M. Sepsis in older adults. Infect. Dis. Clin. N. Am.31(4), 731–742 (2017). [DOI] [PubMed] [Google Scholar]
- 25.Koch, C. et al. Comparison of qSOFA score, SOFA score, and SIRS criteria for the prediction of infection and mortality among surgical intermediate and intensive care patients. World J. Emerg. Surg.15(1), 63 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Kilinc Toker, A., Kose, S. & Turken, M. Comparison of SOFA Score, SIRS, qSOFA, and qSOFA + L criteria in the diagnosis and prognosis of sepsis. Eurasian J. Med.53(1), 40–47 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Nam, K. W. et al. Monocyte to high-density lipoprotein cholesterol ratio is associated with cerebral small vessel diseases. BMC Neurol.24(1), 18 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Qi, X. et al. Identification and characterization of neutrophil heterogeneity in sepsis. Crit. Care25(1), 50 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Liang, P. & Yu, F. Value of CRP, PCT, and NLR in prediction of severity and prognosis of patients with bloodstream infections and sepsis. Front. Surg.9, 857218 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Sui, Y. D., Xin, W. N. & Feng, L. L. Comparison of the clinical application values of PCT, hs-CRP and SAA detection in the early diagnosis of sepsis. Pak. J. Med. Sci.36(7), 1683–1687 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Zampieri, F. G. & Mazza, B. Mechanical ventilation in sepsis: A reappraisal. Shock47(1S Suppl 1), 41–46 (2017). [DOI] [PubMed] [Google Scholar]
- 32.Feng, Y. et al. Age, duration of mechanical ventilation, and outcomes of patients who are critically ill. Chest136(3), 759–764 (2009). [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
All of our data can be obtained by uploading a supporting message or through the corresponding author.