A dynamic nomogram for predicting 28-day mortality in septic shock: a Chinese retrospective cohort study

Study design and population

This retrospective cohort study enrolled patients from the Intensive Care Unit (ICU) of the Second Affiliated Hospital, School of Medicine, Zhejiang University between December 2020 and September 2021. Inclusion criteria: (1) patients diagnosed with septic shock based on The Third International Consensus Definitions for Sepsis and Septic Shock (Sepsis-3); (2) patients aged ≥18 years; (3) patients surviving more than 48 h in the ICU; and (4) patients with complete clinical data. Exclusion criteria: (1) patients with autoimmune diseases; (2) patients undergoing organ transplantation or immunosuppressive therapy; (3) patients with advanced malignant tumor; (4) pregnant women; or (5) patients referred or transferred to other hospitals. This study has been approved by the Institutional Review Board of Second Affiliated Hospital, School of Medicine, Zhejiang University (No. 2021(0955)). The need for written informed consent was waived due to retrospective nature of the study.

Outcome variables and follow-up

The primary outcome was 28-day mortality, which was measured from ICU admission to 28 days after the admission. The included patients were followed up by inspection during ICU stay. The follow-up was terminated in October 2021, and the follow-up duration was 28 days.

Data collection

Patient data were extracted from electronic and paper medical records (Seo et al., 2016; Vallabhajosyula et al., 2018): (1) demographic data and physical condition data on admission: sex, age (years), body mass index (BMI, kg/m²), past medical history (hypertension, chronic lung disease, diabetes, chronic kidney disease, heart disease, chronic liver disease, cardiovascular disease, chronic bronchitis), comorbidities (malignancy, coronary artery disease, dyslipidemia, atrial fibrillation, chronic liver disease, chronic obstructive pulmonary disease (COPD), cerebrovascular disease, peripheral vascular disease), infection site (the urinary system, bloodstream, gastrointestinal tract, gallbladder, and others), coma degree (normal state, mild coma, moderate coma, severe coma), systolic blood pressure (SBP, mmHg), diastolic blood pressure (DBP, mmHg), heart rate (times/min), respiratory rate (times/min), body temperature (°C), urine volume (mL); (2) laboratory data on admission: neutrophil to lymphocyte ratio (NLR), systemic immune-inflammation index (SII), blood routine (hemoglobin (Hb, g/L), red blood cell (RBC, 10¹²/L), platelet (PLT, 10⁹/L), hematocrit (HCT, L/L)), blood biochemistry (liver function: alanine aminotransferase (ALT, U/L), aspartate aminotransferase (AST, U/L), total bilirubin (TBIL, µmol/L), albumin (ALB, g/L); renal function: blood urea nitrogen (BUN, mmol/L), creatinine (Cr, µmol/L), uric acid (UA, µmol/L); ion: potassium (mmol/L), sodium (mmol/L), chlorine (mmol/L), calcium (mmol/L); blood glucose (GLU, mmol/L)), inflammatory marker (C-reactive protein (CRP, mg/L), procalcitonin (PCT, µg/L), white blood cell (WBC, 10⁹/L)), blood gas analysis (lactic acid (LAC, mmol/L), blood pH, oxygenation index (PaO₂/FiO₂), blood oxygen saturation (SaO₂, %)), coagulation function (activated partial thromboplastin time (APTT, s), thrombin time (TT, s), prothrombin time (PT, s), fibrinogen (FIB, g/L)), immunological indicator (immunoglobulin G (IgG, mg/dL), complement C3 (g/L), complement C4 (mg/L)); (3) treatment method: fluid resuscitation, vasoactive drugs, glucocorticoids, mechanical ventilation, hemodialysis. Data on length of ICU stay (h) were also collected. Coma degree was assessed by the Glasgow Coma Scale (GCS) from three aspects: eye opening (maximum 4 points), verbal response (maximum 5 points) and motor response (maximum 6 points) (Mehta & Chinthapalli, 2019). The lower the total score, the more serious the coma: 15 points indicated normal state, 13–14 points indicated mild coma, 9–12 points indicated moderate coma, and 3–8 points indicated severe coma.

Construction and validation of prediction nomogram

The included patients were divided into training and testing sets in a ratio of 7:3 using completely random sampling. The prediction model was developed in the training set, and subsequently validated using the testing set. Univariate logistic regression was adopted to screen for potential predictors, and stepwise regression was further used to screen for predictors in the training set. Based on these predictors, a prediction model was constructed for predicting 28-day mortality in patients with septic shock. The prediction model was visualized by constructing a nomogram to estimate the probability of 28-day all-cause mortality in patients diagnosed with septic shock. The receiver operator characteristic curve (ROC) was used to assess the predicting performance of constructed prediction model. Additionally, the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated for the model in both the training and testing sets, which were then compared with those for the SOFA and APACHE II systems.

Statistical analysis

Measurement data with normal distribution were described as mean ± standard deviation (Mean ± SD), and the t test was used for inter-group comparisons; measurement data with skewed distribution were shown as median and quartiles (M (Q₁, Q₃)), and the Wilcoxon rank sum test was applied for comparisons between the two groups. Enumeration data were described using the number of cases and constituent ratio (n (%)), and comparisons between groups were performed using the chi-square test or Fisher’s exact test. Missing values were filled with medians. For between-group difference analysis, patients were divided into survivor and non-survivor groups. As for the association between variables and 28-day all-cause mortality, the odds ratios (ORs) and 95% confidence intervals (CIs) were calculated. All statistical tests were two-sided, and P < 0.05 was deemed statistically significant. SAS 9.4 (SAS Institute Inc., Cary, NC, USA) was employed for statistical analysis.

Patient characteristics

After excluding 53 patients undergoing lung transplantation and transferred to other hospitals, 304 patients with septic shock were included in this study, with 226 (74.34%) in the survivor group and 78 (25.66%) in the non-survivor group. Figure 1 presents the flow chart of patient selection. The average age and BMI of the included patients were 59.70 ± 15.94 years and 23.01 ± 3.99 kg/m², respectively. There were 207 (68.09%) male patients, and 97 (31.91%) female patients. These included patients had an average SBP and DBP of 127.10 ± 28.58 and 70.64 ± 16.54 mmHg, respectively. The characteristics of these patients are shown in Table 1. Significant differences were observed in age, chronic lung disease, SBP, DBP, heart rate, atrial fibrillation, cerebrovascular disease, NLR, AST, ALB, BUN, Cr, Cl, CRP, PCT, LAC, oxygenation index (PaO₂/FiO₂), SaO₂, PT, C3, glucocorticoids, mechanical ventilation, and hemodialysis between survivor and non-survivor groups.

Predictors for 28-day mortality in septic shock

According to univariate logistic regression, age, chronic lung disease, SBP, DBP, heart rate, atrial fibrillation, cerebrovascular disease, coma degree, TBIL, ALB, Na, Cl, CRP, LAC, oxygenation index (PaO₂/FiO₂), SaO₂, PT, C3, glucocorticoids, and hemodialysis were associated with the risk of 28-day mortality among patients with septic shock (all P <0.05) (Table 2). Further, stepwise regression in the training set identified SBP, cerebrovascular disease, Na, oxygenation index (PaO₂/FiO₂), PT, glucocorticoids, and hemodialysis as the predictors for 28-day mortality of septic shock patients.

Construction and validation of dynamic nomogram

A prediction model integrating SBP, cerebrovascular disease, Na, oxygenation index (PaO₂/FiO₂), PT, glucocorticoids, and hemodialysis was established to prediction the probability of 28-day all-cause mortality in patients diagnosed with septic shock. As shown in Fig. 2, we plotted a nomogram for visualizing the prediction model. The predicted probability of 28-day mortality was easy to obtain by entering the corresponding values for each predictor. For example, a 60-year-old man had a SBP of 89 mmHg, a Na level of 133.00 mmol/L, an oxygenation index of 2.06, and PT of 19.8 s. He did not have cerebrovascular disease, and was treated with glucocorticoids and hemodialysis. Based on the dynamic nomogram application, the predicted probability of 28-day mortality in this patient was 0.387 (Fig. 3). An online dynamic nomogram is presented for convenient clinical application: https://zhijunxu.shinyapps.io/DynNomapp/.

The ROC curves were utilized in this study to evaluate the predictive ability of the developed prediction model, as demonstrated in Table 3. In the training set, the area under the curve (AUC) of ROC curves for this developed model was 0.829 (95% CI [0.768–0.889]), which were higher compared with SOFA (AUC =0.732, 95% CI [0.649–0.814]) and APACHE II (AUC =0.713, 95% CI [0.634–0.791]), respectively. Similarly, in the testing set, the AUC of the developed model was 0.825 (95% CI [0.728–0.923]), which were also higher than SOFA (AUC =0.707, 95% CI [0.584–0.830]) and APACHE II (AUC =0.718, 95% CI [0.601–0.835]) (Table 3, Fig. 4). In addition, the accuracy, sensitivity, specificity, PPV, and NPV of this model 0.773 (95% CI [0.710–0.827]), 0.783 (95% CI [0.719–0.846]), 0.740 (95% CI [0.618–0.862]), 0.514 (95% CI [0.398–0.629]), 0.906 (95% CI [0.858–0.955]) in the training set, and the accuracy, sensitivity, specificity, PPV, and NPV of this model were 0.747 (95% CI [0.645–0.833]), 0.677 (95% CI [0.563–0.791]), 0.923 (95% CI [0.821–1.000]), 0.533 (95% CI [0.388–0.679]), and 0.957 (95% CI [0.898–1.000]), respectively in the testing set.