Development and validation of a clinical severity prediction model for hospitalized children with pertussis Authors: Dongmei Zhang, Yue Zhang, Xiaoxia Lu, Dan Sun Published: 2026 ABSTRACT AbstractBackgroundTo explore the risk factors for severe pertussis and to develop a classification model for illness severity in hospitalized children with pertussis.MethodsA regression analysis was carried out on the cases diagnosed with pertussis in Wuhan Children’s Hospital from January 2022 to October 2024. With April 2024 as the cut-off point, the patients were divided into a training group and a verification group. SPSS and R language were used to analyze the risk factors of severe pertussis in the training group and build a classification model for illness severity in hospitalized children with pertussis.ResultsAmong the relevant predictors, HR, RR, Moist rales, Hb, Alb and IgG are the independent risk factors for severe pertussis. In this study, a nomogram model for illness severity in hospitalized children with pertussis was constructed. Firstly, through training sets, it was confirmed that the prediction model had good discrimination (AUC = 0.935) and calibration (Mean squared error = 0.00062). Finally, the predictive value of the model was further confirmed in the internal validation sets as well (AUC = 0.93, Mean squared error = 0.00089).ConclusionThe nomogram constructed in this study provides a clinical tool for assessing severity stratification in hospitalized children with pertussis, guiding the early identification and clinical management of severe cases.Clinical trial numberNot applicable.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12879-026-13470-9.Keywords: Severe pertussis, Risk factor, Severity stratification model Introduction Bordetella pertussis, the causative agent of pertussis, remains a significant public health concern due to its high transmissibility [1, 2]. Recent epidemiological data reveal a global resurgence of pertussis, with incidence rates escalating annually despite widespread vaccination programs [3]. Notably, the COVID-19 pandemic has disrupted traditional seasonal patterns; previously peaking in August, pertussis now exhibits irregular fluctuations and unexpected winter surges, potentially linked to altered social behaviors and immunity landscapes [4]. Research on severity stratification models for pediatric pertussis remains scarce, including the recent model by Zhang Shiying [5] et al. on clinical and immune predictors of severe pertussis. Most studies tend to focus on the epidemiology of pertussis and the analysis of risk factors for critical pertussis based on small-scale data. The re-emergence of pertussis epidemics causes severe damage to the health of affected children. To address this critical issue, We conducted a single-center cohort study aimed at developing and validating a model for severity stratification in hospitalized children with pertussis. By integrating multiple risk factors through scientific statistical modeling, this tool is designed to enable early identification of severe pertussis. Methods Patient selection Participants This study constructed a model based on a training set of 820 pertussis children (including 134 severe cases and 686 mild cases) from Wuhan Children’s Hospital from January 2022 to April 2024, and validated it using data from 349 pertussis children collected from May to October 2024 (Fig. 1). Flowchart of enrolled patients in training set and validation set description Diagnostic criteria for pertussis The diagnosis of pertussis was established according to the “Pertussis Diagnosis and Treatment Protocol (2024 Edition)” by the National Health Commission [6], confirmed by a positive nucleic acid test for Bordetella pertussis on nasopharyngeal swabs using a commercial IS481-targeted PCR kit. Pertussis classification criteria Criteria for Severe Pertussis: The presence of manifestations such as hypoxemia or acute respiratory failure, pertussis encephalopathy, pulmonary artery hypertension, hypotension or cardiogenic shock, or failure of other organs. Criteria for Severe Pertussis: The presence of manifestations such as hypoxemia or acute respiratory failure, pertussis encephalopathy, pulmonary artery hypertension, hypotension or cardiogenic shock, or failure of other organs. Criteria for Mild Pertussis: Mild pertussis was defined as laboratory-confirmed pertussis without any severe manifestations. Criteria for Mild Pertussis: Mild pertussis was defined as laboratory-confirmed pertussis without any severe manifestations. Exclusion criteria Patients were excluded if they had pre-existing comorbidities (e.g., congenital heart disease, chronic lung disease), left against medical advice, or lacked essential clinical documentation. Data collection The collected clinical data encompassed the following aspects: Baseline Characteristics: Gender and age. Baseline Characteristics: Gender and age. Relevant Histories: History of cough contact prior to illness.Pre-admission macrolide use. Relevant Histories: History of cough contact prior to illness.Pre-admission macrolide use. History of cough contact prior to illness. History of cough contact prior to illness. Pre-admission macrolide use. Pre-admission macrolide use. Pre-admission symptoms: Presence of fever, and cough-related symptoms including facial flushing, breath-holding, and convulsions. Pre-admission symptoms: Presence of fever, and cough-related symptoms including facial flushing, breath-holding, and convulsions. Physical Signs on Admission:​​ Vital signs: Respiratory rate (RR) and heart rate (HR).Lung auscultation findings: moist rales, rhonchi, coarse rales, and wheezing rales. Physical Signs on Admission:​​ Vital signs: Respiratory rate (RR) and heart rate (HR).Lung auscultation findings: moist rales, rhonchi, coarse rales, and wheezing rales. Vital signs: Respiratory rate (RR) and heart rate (HR). Vital signs: Respiratory rate (RR) and heart rate (HR). Lung auscultation findings: moist rales, rhonchi, coarse rales, and wheezing rales. Lung auscultation findings: moist rales, rhonchi, coarse rales, and wheezing rales. Laboratory Investigations on Admission:​​ Complete Blood Count (CBC): White blood cells (WBC) with differentials (neutrophils-NEU, lymphocytes-LYM, monocytes-MONO), red blood cells (RBC), hemoglobin (Hb), and platelets (PLT).Inflammatory Markers: Procalcitonin (PCT) and high-sensitivity C-reactive protein (hs-CRP).Organ Function Panels: Liver function tests, myocardial enzyme profile.Humoral Immunity Panel: Immunoglobulin levels (IgG, IgM, IgA, IgE). Laboratory Investigations on Admission:​​ Complete Blood Count (CBC): White blood cells (WBC) with differentials (neutrophils-NEU, lymphocytes-LYM, monocytes-MONO), red blood cells (RBC), hemoglobin (Hb), and platelets (PLT).Inflammatory Markers: Procalcitonin (PCT) and high-sensitivity C-reactive protein (hs-CRP).Organ Function Panels: Liver function tests, myocardial enzyme profile.Humoral Immunity Panel: Immunoglobulin levels (IgG, IgM, IgA, IgE). Complete Blood Count (CBC): White blood cells (WBC) with differentials (neutrophils-NEU, lymphocytes-LYM, monocytes-MONO), red blood cells (RBC), hemoglobin (Hb), and platelets (PLT). Complete Blood Count (CBC): White blood cells (WBC) with differentials (neutrophils-NEU, lymphocytes-LYM, monocytes-MONO), red blood cells (RBC), hemoglobin (Hb), and platelets (PLT). Inflammatory Markers: Procalcitonin (PCT) and high-sensitivity C-reactive protein (hs-CRP). Inflammatory Markers: Procalcitonin (PCT) and high-sensitivity C-reactive protein (hs-CRP). Organ Function Panels: Liver function tests, myocardial enzyme profile. Organ Function Panels: Liver function tests, myocardial enzyme profile. Humoral Immunity Panel: Immunoglobulin levels (IgG, IgM, IgA, IgE). Humoral Immunity Panel: Immunoglobulin levels (IgG, IgM, IgA, IgE). No components of the outcome (organ failure) are used as predictors. Ethics This study complied with the standards of medical ethics, and the current study followed the principles of the Declaration of Helsinki, which was approved by the Medical Ethics Committee of Wuhan Children’s Hospital (Approval No.2024R044-E01) without the need for written informed consent. Statistical analysis All analyses were performed using SPSS (v23.0; IBM Corp) and R Statistical Software (v4.3.2; R Foundation). The missing data were mainly supplemented using multiple imputation, and the sensitivity analysis showed no significant impact. Variables were selected based on admission values to ensure early prediction. Continuous variables were assessed for normality via Shapiro-Wilk test. Non-normally distributed data were expressed as median with interquartile range (IQR) and analyzed using Mann-Whitney U test, while categorical variables were presented as frequencies (%) and compared through χ²test [7]. Multivariable logistic regression (backward stepwise selection, entry criteria P < 0.10) identified independent predictors of severe pertussis. Model building and validation followed TRIPOD guidelines [8]: Model development: Final predictors were selected based on clinical relevance and statistical significance (P < 0.05). Model development: Final predictors were selected based on clinical relevance and statistical significance (P < 0.05). Performance evaluation: Discrimination was quantified by area under receiver operating characteristic curve (AUC-ROC) with 95% CIs using the pROC package [9]. MSE measures calibration error; lower values indicate better fit. Bias correction was applied via bootstrap resampling. Calibration was assessed via calibration plots (1,000 bootstrap resamples) [10]. Performance evaluation: Discrimination was quantified by area under receiver operating characteristic curve (AUC-ROC) with 95% CIs using the pROC package [9]. MSE measures calibration error; lower values indicate better fit. Bias correction was applied via bootstrap resampling. Calibration was assessed via calibration plots (1,000 bootstrap resamples) [10]. Clinical utility: Decision curve analysis (DCA) implemented through rmda package evaluated net benefit across probability thresholds [11]. Clinical utility: Decision curve analysis (DCA) implemented through rmda package evaluated net benefit across probability thresholds [11]. All statistical tests were two-tailed with P < 0.05 indicating significance. Results General information and examination results of the total cohort, training group and validation group From January 2022 to October 2024, a total of 1,387 pertussis cases were confirmed at Wuhan Children’s Hospital. The annual confirmed cases were 104 (7.5%) in 2022, with an incidence rate of 1.35% (compared to 7,689 bronchopneumonia cases diagnosed that year); 344 cases (24.8%) in 2023, with an incidence rate of 2.99% (compared to 11,501 bronchopneumonia cases); and 939 cases (67.7%) in 2024, with an incidence rate of 7.52% (compared to 12,490 bronchopneumonia cases). From January 2022 to April 2024, pertussis cases were designated as the training group, with 167 children excluded from the study out of 987 cases; cases from May to October 2024 served as the validation group, with 51 children excluded from the study out of 400 cases (Fig. 1). The demographic data, clinical characteristics and other relevant factors of the total cohort, training group and validation group are detailed in Table 1. Table 1Patients and treatment characteristicsVariablesTotal cohort (n = 1169)Training set (n = 820)Internal validation set (n = 349)Female, n (%)485(41.4)336(40.9)149(42.7)Age, median (IQR), m10(3, 79)8(3, 78.75)15(3, 79)Cough contact history, n, (%)280(24)192(23.4)88(25.2)Pre-admission macrolide use, n,(%)546(46.7)409(49.9)137(39.3)Fever, n,(%)305(26.1)225(27.4)80(22.9)facial flushing, n,(%)816(69.8)567(69.1)249(71.3)breath-holding, n,(%)213(18.2)165(20.1)48(13.8)convulsion, n,(%)16(1.4)14(1.7)2(0.6)HR, median (IQR), bpm120(101, 129.5)120(102, 130)116(100, 126)RR, median (IQR), bpm28(24, 35)28(24, 35)26(26, 32)Moist rales, n,(%)251(21.5)199(24.3)52(14.9)rhonchi, n,(%)20(1.7)16(2)4(1.1)Coarse rales, n, (%)300(25.7)207(25.2)93(26.6)wheezing ralas, n,(%)112(9.6)82(10)30(8.6)WBC, median (IQR),10^9/L12.43(8.85, 18.05)12.83(9.09, 18.97)11.61(8.59, 16.51)NEU, median (IQR), 10^9/L3.92(2.73, 6.22)4.14(2.78, 6.64)3.67(2.61, 5.34)LYM, median (IQR), 10^9/L6.48(3, 83, 11.25)6.74(3.82, 11.67)6.07(3.84, 10.07)RBC, median (IQR), 10^12/L4.40(4.01, 4.72)4.41(4.01, 4.77)4.39(4.01, 4.66)Hb, median (IQR), g/l120(111, 129)120(111, 129)120(112, 127)PLT, median (IQR),10^9/L398(317, 516)407(318, 516)381(310, 513)PCT, median (IQR), ng/ml0.06(0.05, 0.09)0.07(0.05, 0.1)0.06(0.04, 0.08)hs-CRP, median (IQR), mg/l1.00(0.80, 2.28)1(0.8, 3)1(0.78, 1.26)ALB, median (IQR), g/l44.1(42.1, 46.3)43.9(41.8, 46)44.8(42.85, 46.5)LDH, median (IQR), U/L302(251, 364)302(251, 368)303(250.5, 354.5)CK, median (IQR), U/L117(84, 167)117(84, 165)118(84, 167.5)CK-MB, median (IQR), U/L35(26, 50)35(26, 52)34(25, 46.9)IgG, median (IQR), g/l6.24(4.21, 8.9)6.24(4.21, 8.9475)6.24(4.195, 8.835)IgA, median (IQR), g/l0.41(0.26, 1.09)0.41(0.26, 1.08)0.41(0.27, 1.13)IgM, median (IQR), g/l0.73(0.47, 1.075)0.73(0.46, 1.05)0.73(0.52, 1.135)IgE, median (IQR), IU/ml19.1(6.49, 76.8)19.1(5.9725, 80.025)19.1(8.025, 71.75) Patients and treatment characteristics Baseline data of the training group A total of 820 pertussis patients meeting the inclusion criteria were enrolled in this study. Among them, 134 (16.3%) were assigned to the severe group, with 35.8% females and a median (IQR) age of 2 (1, 4). The remaining 686 (83.7%) were assigned to the mild group, consisting of 42% females with a median (IQR) age of 21 (4, 82). Compared with the mild group, severe patients were younger and had significantly higher heart and respiratory rates, more frequent use of mechanical ventilation, higher incidence of pertussis encephalopathy, history of contact with coughing individuals, prior macrolide use, and symptoms such as flushing, breath-holding cough, decreased heart rate, and convulsions during severe coughing. Complications such as respiratory and heart failure could occur in the severe group, and lung auscultation often revealed moist rales and wheezing. All these differences were statistically significant (P < 0.05). A comparative analysis of initial hematological, biochemical, and immunological profiles was performed between the two groups upon admission. The results revealed that the levels of WBC, NEU, LYM, PLT, PCT, AST, ALT, TBIL, LDH and CK-MB were significantly higher in the severe group than in the mild group (P < 0.05), while the RBC, Hb, ALB, CK, BUN, CR, IgG, IgA, IgM and IgE were significantly lower in the severe group compared to the mild group (P < 0.05)(Table 2). Table 2Baseline data of the training groupVariablesSevere group (n = 134)Mild group (n = 686)P valueFemale, n,(%)48(35.8)288(42)0.185Age< 0.0011–6 m113(84.3)279(40.6)7–12 m9(6.7)40(5.83)13–36 m6(4.47)37(5.39)37–60 m1(0.74)43(6.26)61–120 m5(3.73)259(37.7)>120 m028(4.08)Cough contact history, n, (%)47(31.3)150 (21.8)< 0.001Pre-admission macrolide use, n,(%)46(34.3)363(53)< 0.001Fever, n,(%)39 (29.1)186(27.1)0.637facial flushing, n,(%)107(79.9)460(67.1)0.003breath-holding, n,(%)63(47)102(14.9)< 0.001convulsion, n,(%)13 (9.7)1 (0.1)< 0.001HR, median (IQR), bpm146.5 (130, 167)112.5 (98, 125)< 0.001RR, median (IQR), bpm39 (33, 45)26 (23, 32)< 0.001Moist rales, n,(%)67(50)132 (19.2)< 0.001rhonchi, n,(%)2 (1.5)14(2)0.675Coarse rales, n, (%)34(25)173(25.2)0.97wheezing ralas, n,(%)26(19.7)56 (8.2)< 0.001WBC, median (IQR),10^9/L17.14(11.33,25.63)12.33(8.57,17.75)< 0.001NEU, median (IQR), 10^9/L5.18 (3.07, 9.3)4 (2.76, 6.34)0.002LYM, median (IQR), 10^9/L9.84 (6.1, 15.83)6.17 (3.55, 11)< 0.001RBC, median (IQR), 10^12/L3.73 (3.32, 4.27)4.5 (4.16, 4.83)< 0.001Hb, median (IQR), g/l108.5 (97, 117)122 (114, 130)< 0.001PLT, median (IQR),10^9/L475 (389, 616)392 (312, 491)< 0.001PCT, median (IQR), ng/ml0.10 (0.07, 0.16)0.06 (0.04, 0.09)< 0.001hs-CRP, median (IQR), mg/l1(0.8,3.67)1 (0.8, 2.85)0.979ALB, median (IQR), g/l40.95 (38.1, 43.5)44.1 (42.4, 46.4)< 0.001LDH, median (IQR), U/L329 (275, 385)300 (247, 361)< 0.001CK, median (IQR), U/L92 (68, 147)118 (88, 167)< 0.001CK-MB, median (IQR), U/L38 (30, 52)35 (25, 52)0.015IgG, median (IQR), g/l4.96 (3.67,6.68)6.4 (4.26, 9.14)< 0.001IgA, median (IQR), g/l0.27 (0.26, 0.3)0.47 (0.27, 1.22)< 0.001IgM, median (IQR), g/l0.43 (0.29, 0.73)0.74 (0.52, 1.09)< 0.001IgE, median (IQR), IU/ml19.1 (8.43, 92.9)5.94 (2.21, 19.1)< 0.001 Baseline data of the training group Multivariate logistic regression analyses of severe pertussis The clinical data of the two groups of children were compared, Variables with statistically significant differences (P < 0.05) in the baseline tables (Table 2) were subjected to univariate logistic regression analysis, and the meaningful indicators (P < 0.05) were then included in the multivariate Logistic regression analysis to identify independent risk factors. The results of the multivariate analysis revealed that: HR (OR = 1.072, 95%CI: 1.051–1.093, P < 0.001), RR (OR = 1.086, 95%CI:1.043–1.132,P < 0.001), Moist rales (OR = 2.897, 95%CI: 1.629–5.153, P < 0.001), IgG (OR = 1.13, 95%CI:1.022–1.25, P = 0.017) are the independent risk factors for severe pertussis. Hb (OR = 0.954, 95%CI: 0.932–0.967,P < 0.001), Alb(OR = 0.844, 95%CI:0.779–0.914, P<0.001) are the protective factors for severe pertussis (Table 3). Table 3Multivariate logistic regression analysis of severe pertussisVariablesβOR95%CI P ToleranceVIFHR0.0691.0721.051 ~ 1.093< 0.0010.4742.11RR0.0831.0861.043 ~ 1.132< 0.0010.4752.105Moist rales1.0642.8971.629 ~ 5.153< 0.0010.9331.072Hb-0.0470.9540.932 ~ 0.976< 0.0010.7131.402Alb-0.170.8440.779 ~ 0.914< 0.0010.8601.163IgG0.1221.131.022 ~ 1.250.0170.7341.363 Multivariate logistic regression analysis of severe pertussis Construction of a nomogram prediction model Based on the independent risk factors identified by multivariate logistic regression analysis (using R), a nomogram model was developed to predict the disease severity in hospitalized children with pertussis. A total of 6 indicators were included in the model, namely HR, RR, Moist rales, Hb, ALB and IgG. The prediction nomogram for severe pertussis includes eight variable axes. (HR, RR, Moist rales, Hb, ALB, IgG), One point scale axis (Point), One total points axis (Total Points) and the corresponding risk axis (SC risk) (Fig. 2). The nomogram score is derived by summing points for each variable; variables with OR < 1 (e.g., Hb) contribute negatively to the risk score, indicating protective effects. A model diagram for predicting the risk of severe illness in children with pertussis. * "Bpm" refers to "beats per minute" Internal validation of the nomogram prediction model To evaluate the discrimination and calibration of the nomogram model, this study plotted the ROC (Receiver Operating Characteristic) curve and calculated the AUC (Area Under the Curve) value to assess the discrimination of the model. Then, 1000 internal resamplings were conducted through the Bootstrap resampling method, and the calibration curve was drawn and the MSE (Mean Squared Error) value was calculated to evaluate the calibration of the model. The model demonstrated good discriminatory performance, as evidenced by an AUC of 0.935 (95% CI: 0.913–0.9586), a sensitivity of 88.8%, and a specificity of 85.4%. Moreover, the calibration curve indicated that the model had a high level of calibration, with an MSE of 0.00062 (Fig. 3a and b). To verify the clinical effectiveness of the nomogram model, this study further carried out the Decision Curve Analysis (DCA). In the DCA, the abscissa represents the threshold probability, and the ordinate represents the net benefit rate. The farther the curve is from the extreme values (the intersection point of the All line and the None line), the higher the clinical effectiveness of the model. The results demonstrated that the model had good clinical application value (Fig. 3c). Decision curve analysis showed that the prediction model achieved positive net benefit at threshold probabilities ranging from 5% to 16%. The net benefit was 0.0237 at 5%, 0.0582 at 10%, 0.1044 at 15%, and 0.1126 at 16%, indicating the clinical utility of the model.The model’s calibration curve demonstrated acceptable agreement with a notable calibration slope and intercept, with a Brier score of 0.063, a calibration slope close to 1, and a calibration intercept close to 0. Internal validation of the nomogram prediction model. (a) Receiver operating characteristics (ROC) curves reflecting the discriminative power of the nomogram in the training group; (b) Calibration curve of the nomogram reflecting the accuracy of the nomogram in the training group; (c) Decision curve of the nomogram model for predicting severe pertussis in the training group Discriminative power and accuracy of the nomogram model in internal cohort validation The predictive power of the nomogram was internally validated through a retrospective study, which enrolled 349 pertussis patients, who were admitted to the Wuhan Children’s Hospital, from May 2024 to Oct 2024.In the internal validation cohort, the prediction nomogram for severe cases yielded an AUC of 0.93 (95% CI: 0.893–0.967), demonstrating high sensitivity (90.9%) and specificity (80.3%). The calibration curve also had good consistency between observation and prediction, in terms of the risk of severe cases, with an MSE of 0. 00089, which indicates that the deviation of the predicted value from the observed value was very small (Fig. 4A and B). It was further demonstrated through internal cohort validation that this model has good clinical application value (Fig. 4c). The prediction model also maintained a positive net benefit within the threshold range of 5–16%, supporting its clinical utility. The calibration curve of the prediction model showed some deviation from ideal calibration: a Brier score of 0.057, a calibration slope close to 1, and a calibration intercept close to 0. Discriminative power and accuracy of the nomogram model in internal cohort validation. (a) Receiver operating characteristics (ROC) curves reflecting the discriminative power of the nomogram in the internal cohort validation group; (b) Calibration curve of the nomogram reflecting the accuracy of the nomogram in the internal cohort validation group; (c) Decision curve of the nomogram model for predicting severe pertussis in the internal cohort validation group Discussion The resurgence of pertussis poses a critical challenge globally, with China experiencing an unprecedented surge since early 2023. Surveillance data from Chinese health authorities indicate a dramatic escalation, with monthly cases rising from 821 to over 90,000 within a year, accompanied by 20 fatalities in 2024 alone [4, 12]. The number of cases in this study began to increase in early 2023 and rose sharply towards the end of the year, which is consistent with national surveillance data (Supplementary Fig. 1).This alarming trend underscores the urgency to identify high-risk populations and optimize clinical management. In this study, severe pertussis accounted for one-sixth of all cases. The investigation aimed to identify high-risk factors for severe pertussis to guide correction efforts during treatment and prevent disease progression. Among the 1,169 included cases, infants aged 1–6 months constituted 45.7%, those aged 7–12 months 6.16%, and children aged 13–36 months 5.99%. In the severe group (n = 178), the proportions were 85.3%, 6.18%, and 0.59%, respectively, while in the mild group (n = 991) they were 38.6%, 6.15%, and 6.35% (Supplementary Fig. 2). Thus, infants under 6 months old represented a notably larger proportion in the severe group. This study developed a prediction model for severe pertussis using the independent risk factors and statistical parameters screened by the Logistic regression model. The model ultimately included six indicators: heart rate, respiration, moist rales, hemoglobin, albumin, and IgG, all of which are independent risk factors for severe pertussis. In the nomogram model, distinct scores are assigned to each indicator. During clinical work, clinicians can assess whether a case of pertussis aligns with the aforementioned six indicators and subsequently compute the total score, which allows for an estimation of the risk of developing severe symptoms. Essentially, the higher the score, the greater the likelihood of a severe outcome. This nomogram model is not only intuitive but also user-friendly. Furthermore, the findings of this study demonstrate that it possesses excellent predictive capabilities, as evidenced by an area under the curve (AUC) value of 0.935. The calibration curve further validates its high level of calibration and minimal error, with a mean squared error (MSE) of 0.00062. Additionally, decision curve analysis was employed in this study to further corroborate the model’s clinical utility. Ultimately, in order to further verify the predictive ability and universality of the model, the data from other time periods within our institution were utilized as a validation cohort for verification and evaluation. The results demonstrated that the model still maintained a favorable predictive value within the validation cohort. Besides, the calibration curve confirmed that the model had a high level of calibration with minimal errors (with the area under the curve, AUC = 0.93, and the mean squared error, MSE = 0.00089). Subsequently, the decision curve analysis was conducted once again by using the validation cohort to further validate the clinical effectiveness of the model. This dual-validation process demonstrated that the nomogram is robust for stratifying disease severity in hospitalized children with pertussis, ensuring reliable predictions. In children with pertussis, cough-induced suffocation leads to hypoxia, which subsequently triggers vascular endothelial damage and enhanced permeability. When the albumin level in the blood declines, it is likely to prompt an augmentation of fluid in the third space, thereby giving rise to pulmonary interstitial edema as well as edema in the interstitial spaces of organs and skin tissues. This exacerbates the dysfunction of both lung and other organ systems. Auscultation of the lungs at this stage typically reveals the presence of moist rales. Hemoglobin, serving as a natural oxygen carrier, experiences a diminished oxygen-carrying capacity in the context of reduced content, which in turn leads to a further decrease in arterial blood oxygen content and a consequent aggravation of hypoxic symptoms. Hence, the reduction in albumin and hemoglobin levels, along with the emergence of moist rales in the lungs, can serve as predictors for the onset of severe pertussis. Children afflicted with severe pertussis are inclined to suffer from cardio-pulmonary dysfunction and failure of other visceral organs. Prior to the manifestation of multiple organ failure, there might exist certain latent factors. By ameliorating these underlying elements, it is possible to reverse the progression of the disease course. This article points out that pertussis patients with a rapid heart rate and relatively fast breathing are more likely to develop severe conditions. The accelerated heart rate and breathing in pertussis patients may be related to inflammatory reactions, hypoxia and myocardial injury. This is consistent with the persistent tachycardia and dyspnea accompanied by early respiratory failure, which are the main clinical features of malignant pertussis mentioned in the articles by Kaihu Yao [13] and Kundrat [14], Long Xiang [15] and Murray E [16] mentioned that severe pertussis cases admitted to the intensive care unit were all accompanied by an accelerated heart rate and breathing under the condition of leukocytosis, which is consistent with what is described in this article. Pertussis toxin can also have an impact on the heart rate control of experimental animals by altering the vagus nerve [17]. The view that children with severe pertussis are more likely to be complicated with cardiac insufficiency is the same as that in the articles by Xiang Long [15], Mengzhao Li [18]and Yang Hu [19]. Both Yang Hu [19] and Shuai Guo [20] have pointed out that children with severe pertussis have severe leukocytosis. A large number of white blood cells gather in the pulmonary arterioles, veins and lymphatic vessels, resulting in a significant increase in pulmonary artery pressure, a reduction in pulmonary blood flow, an exacerbation of hypoxemia, and ultimately leading to heart failure. Pertussis toxin is a major virulence factor produced by Bordetella pertussis. It inhibits the activation of G proteins by modifying Gi/o class G proteins in mammals. This type of G protein has a cardioprotective effect, and cardiomyocyte apoptosis, myocardial ischemia and heart failure are all associated with the loss of G protein activity. Such inhibition can lead to cardiopulmonary dysfunction [21]. In Shuai Guo’s [20] article, it was proposed that the virulence factors of Bordetella pertussis invade and newly infect, damaging the bronchial and alveolar epithelium. This may also lead to an increase in secondary pulmonary vascular resistance. In younger infants, there are still prominent muscular components in the arteries, and these blood vessels are very sensitive to constrictive stimuli, thus causing vasospasm. Hypoxemia or co - existing acidosis can trigger pulmonary vasoconstriction, resulting in an increase in pulmonary vascular resistance and then cardiopulmonary failure. Jingjuan Wang [21] pointed out that the inhibition of Gi protein can lead to the imbalance of calcium ion in cardiomyocytes, resulting in cardiomyocyte apoptosis, cardiopulmonary failure and even death. This study exhibits certain inherent limitations. Primarily, comprehensive data collection pertaining to the vaccination history and the precise genetic profiles of pertussis pathogen strains was lacking. In subsequent phases, concerted efforts will be directed towards meticulously augmenting and refining the case datasets to facilitate more profound and exhaustive research. Secondly, the current study did not integrate external data sources. Future research endeavors will necessitate the incorporation of multi-center external datasets for the purposes of rigorous validation and corroboration, ensuring enhanced robustness and generalizability of the findings. Conclusions The early clinical manifestations and laboratory indices of severe pertussis in children are characterized by a lack of specificity. In clinical practice, particular attention should be paid to children with low levels of hemoglobin and albumin, as they may potentially be afflicted with severe pertussis. Younger pertussis patients, especially those manifesting accelerated heart rates and respiratory frequencies, are more susceptible to the development of heart failure, invariably resulting in extended hospital stays and substantial financial expenditures. Clinicians are thus advised to employ scoring systems and predictive models to conduct comprehensive risk assessments on patients, with the aim of identifying high-risk individuals prone to severe pertussis for the implementation of early preventive measures and personalized interventions, ultimately improving the prognosis of affected children. The model identifies high-risk patients based on admission parameters (e.g., elevated HR/RR), allowing clinicians to initiate monitoring or prophylaxis before severe development. The nomogram developed in this study has demonstrated a relatively reliable capacity to predict the risk of severe pertussis and holds significant guiding value for clinical decision-making in pertussis cases. Looking ahead, it is anticipated that more large-scale, prospective, and multi-center studies will be forthcoming to further validate and explore relevant aspects. Electronic Supplementary Material Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 2 Abbreviations Heart rate Respiratory rate Heart failure Respiratory failure Hemoglobin Albumin Immunoglobulin E Area Under the Curve Mean squared error Invasive mechanical ventilation Non-invasive ventilation Complete Blood Count White blood cells Neutrophils Lymphocytes Monocytes Red blood cells Platelets PROCALCITONIN High-sensitivity C-reactive protein Immunoglobulin G Immunoglobulin M Immunoglobulin A Interquartile range Decision curve analysis Lactate Dehydrogenase Creatine Kinase Creatine Kinase-MB isoenzyme beats per minute Receiver Operating Characteristic Outer membrane protein Hypoxemia Author contributions Data curation: Dongmei Zhang; Methodology: Yue Zhang; Software: Dongmei Zhang; Writing-original draft: Dongmei Zhang; Writing-review & editing: Xiaoxia Lu and Dan Sun. Funding This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. Data availability The datasets used and/or analyzed during the current study are available from the corresponding author (bloveriver@163.com) upon reasonable request. Declarations Ethics approval and consent to participate This study was conducted in accordance with the Declaration of Helsinki, approved by the ethics committee of Wuhan Children’s Hospital (2024R044-E01). Due to the retrospective nature of the study, the Medical Ethics Committee of Wuhan Children’s Hospital waived the need of obtaining informed consent. 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. Dongmei Zhang and Yue Zhang contributed equally to this work and share first authorship. Contributor Information Xiaoxia Lu, Email: lusi74@163.com. Dan Sun, Email: bloveriver@163.com. References Associated Data Supplementary Materials Supplementary Material 1 Supplementary Material 2 Data Availability Statement The datasets used and/or analyzed during the current study are available from the corresponding author (bloveriver@163.com) upon reasonable request.