Pertussis infection in critically ill infants: meta-analysis and validation of a mortality score

Abstract

Background

Despite widespread vaccination programs, pertussis continues circulating within populations and remains a life-threatening infection in infants. While several mortality risk factors have been described, a comprehensive synthesis is lacking. We conducted a meta-analysis of studies investigating mortality risk factors in Pertussis infections and validated those factors in a large cohort.

Methods

Observational studies published in English were systematically searched in PubMed, EMBASE, and LiSSa databases from 01/2000 to 06/2024. The search yielded 816 unique citations. The primary outcome was mortality before discharge from the Pediatric Intensive Care Unit (PICU). Two independent reviewers assessed the risk of bias and extracted data. A REML-random effect model was used to calculate pooled prevalence and conduct the analysis. The identified risk factors were subsequently evaluated in a monocentric cohort of patients admitted to a tertiary hospital’s PICU for severe pertussis between January 1996 and December 2020. Data analysis was conducted between June and August 2024.

Results

Seventeen studies, including 2,725 patients, met the inclusion criteria. The pooled prevalence of mechanical ventilation, continuous renal replacement therapy, and Extracorporeal Membrane Oxygenation support were 55% (95% CI: 40–70; I² = 98), 15% (95% CI: 3–27; I² = 95), and 8% (95% CI: 3–12; I² = 93), respectively. The pooled mortality incidence was 19% (95% CI:12–26; I² = 96). Identified mortality risk factors included elevated heart rate, presence of pulmonary hypertension, presence of seizures, and elevated white blood cell (WBC) count. Validation in an 83-patient cohort (median age: 45 days, IQR: 30–55) revealed a mortality rate of 12%. Risk factors identified in the meta-analysis were significantly associated with non-survival in the cohort. A mortality prediction score was developed incorporating age < 30 days, heart rate > 200/min, and WBC > 30 G/l, achieving an area under the curve of 0.92 (95% CI: 0.86–0.99).

Conclusion

This meta-analysis identified a simple yet effective score to assess the severity of pertussis infection in infants admitted to PICU. Accurate risk stratification may enable timely treatment of critically ill patients, potentially improving outcomes.

Trial registration: The study protocol was registered on PROSPERO: CRD42024582057.

Graphic abstract

Supplementary Information

The online version contains supplementary material available at 10.1186/s13054-025-05300-2.

Background

Bordetella pertussis, a Gram-negative bacterium, causes whooping cough, a highly contagious respiratory infection [1]. Despite the availability of effective immunization, pertussis continues to circulate, leading to outbreaks and potentially fatal cases among vulnerable populations [2, 3]. Pertussis remains a significant cause of infant mortality, affecting both high-income and low-to-middle-income countries [3–5]. Additionally, several European countries have reported a resurgence of pertussis cases in recent years [6].

Numerous studies have identified risk factors associated with mortality in severe pertussis infections, highlighting hyperleukocytosis as a critical marker of disease severity and mortality [7]. Other significant risk factors include younger age at presentation and pulmonary hypertension, observations initially reported in the description of “malignant” pertussis nearly 60 years ago [8–10]. However, a comprehensive synthesis of these risk factors is lacking. Studies vary in their definitions of age-related risk thresholds for leukocyte counts and often involve a limited number of patients. Additionally, outcome measures differ, ranging from hospitalization to mortality. A unified, synthesized mortality predictor for pertussis is needed to identify high-risk patients promptly. Such tools would support decisions regarding early PICU admission and intervention, such as leukoreduction therapy, which has been suggested to improve patient outcomes [11, 12]. A clinical score reflecting mortality risk could aid physicians in determining which patients require escalation of care, including advanced medical support.

In this study, we aimed to identify mortality risk factors associated with severe pertussis infections in children through a meta-analysis. Subsequently, we evaluated these risk factors in a single-center cohort to create a bedside mortality prediction model, providing a practical tool for clinical use.

Methods

Meta-analysis: search strategies

The clinical research question addressed was: What are the clinical criteria associated with mortality in critically ill infants with pertussis infection? Pubmed, LiSSa, and EMBASE databases were systematically searched between June and July 2024 for relevant peer-reviewed articles published in English from January 2000 to June 2024, following PRISMA guidelines [13]. A combination of MeSH terms, including “Bordetella pertussis,” “critical care,” and “mortality,” was employed (details provided in Supplementary materials and methods). References of selected publications were also screened to identify additional relevant articles potentially missed in the initial search.

Two reviewers (VLC and CC) independently screened titles and abstracts to determine eligibility. The inclusion criteria were studies reporting mortality risk factors for Bordetella pertussis infection in children. Reviews and case reports were excluded if the case series included fewer than five patients. Studies focusing solely on risk factors for severe infection without addressing mortality were also excluded. Mortality risk factors reported in at least two studies were included in the meta-analysis. The study was registered in the PROSPERO database (CRD42024582057).

Meta-analysis: risk factors selection and statistics

Data extraction was conducted independently by two authors (VLC and CC). Baseline data extracted included year of publications, number of patients included, age of the population, and mortality rate. Population severity was assessed by reporting mechanical ventilation, continuous renal replacement therapy (CRRT), extracorporeal membrane oxygenation (ECMO), and blood exchange therapy. Pulmonary hypertension (PAH) presence was also recorded. Duration of mechanical ventilation and PICU stay were also noted when available. Mortality-related factors reported in more than two studies were extracted for consistent reporting. Crude descriptive statistics from reviewed articles were included in the pooled analysis. For studies reporting medians and interquartile ranges (IQR) instead of means and standard deviations (SD), the algorithm published by Wan et al. was used to estimate the mean and SD [14].

Statistical analysis was performed using Stata v18 (StataCorp, College Station, TX, USA). The meta-analysis software pack was used to calculate pooled prevalence and perform the analyses using the REML random effect model. Forest plots were generated for data visualization. Heterogeneity was assessed using the I² statistics, with I² values higher than 50% indicating high heterogeneity.

Cohort study: patient population

A single-center retrospective study was conducted in the PICU of the Bicêtre Hospital (Bicêtre Hospital AP-HP, Paris-Saclay University, Le Kremlin-Bicêtre, France) from January 1996 to December 2020. Patients under 90 days of age admitted for a severe pertussis infection were screened, and PICU reports were reviewed. The study was approved by the ethical committee from the French Society of Intensive Care (CE SRLF 23–074). The study has been registered at the “Commission Nationale de l’Informatique et des Libertés” corresponding to the reference methodology (MR-004). Data processing adhered to European legislation, and families were informed via a welcome booklet about the potential use of clinical data for research and their right to opt-out.

Cohort study: data collection

Baseline demographic data at admission were collected from the individual medical records alongside clinical, biochemical, and hematological parameters during the PICU stay. The use of vasopressor drugs, mechanical ventilation, CRRT, ECMO, and exchange transfusion were recorded. Cardiac failure was defined by a low cardiac index (< 2.2 l/min/m²) or the use of dobutamine or adrenaline after echocardiographic evaluation. Pulmonary hypertension was assessed by trained echocardiographists using tricuspid and/or pulmonary regurgitation velocity. The highest leucocyte count (G/L) during the PICU stay was documented.

Cohort study: statistics

Categorical variables were expressed as percentages (%), and continuous variables as the medians with IQR. When appropriate, categorical variables were analyzed using a chi-square or Fisher’s exact tests, while continuous variables were analyzed using Wilcoxon tests.

The primary outcome was mortality in the PICU. Bivariate analysis was used to test the association between demographic, clinical, laboratory, and management characteristics and mortality. Univariate logistic regression provided odds ratios (OR) and 95% confidence intervals (CI) for mortality predictors.

A bedside clinical score for severe pertussis, developed via the meta-analysis, was validated using forward multivariate logistic regression in the cohort. Model fit was assessed with the Hosmer–Lemeshow test (p-value > 0.05 indicating good fit), and the Receiver Operating Curve (ROC) was used to evaluate predictive accuracy. Variables retained for the mortality score were weighted based on their relative β parameters. A p-value < 0.05 was considered statistically significant. All analyses were conducted using STATA v18 software (StataCorp, College Station, TX, USA).

Results

Review and population

Using the outlined inclusion and exclusion criteria, 816 articles were identified. After screening, 17 studies were included (Fig. 1), with bias assessment detailed in supplementary Table 1 [5, 11, 15–28]. These studies involved 2,725 patients; their main characteristics are presented in Table 1. Of these, 12 studies focused exclusively on PICU-admitted patients [11, 15–19, 22–27], while five included all patients admitted with pertussis infections [5, 20, 21, 28, 29]. Two studies employed a case–control design [20, 28].

Fig. 1.

Preferred reporting items for systematic reviews and meta-analysis flow diagram

Table 1.

Description of included studies in the meta-analysis

Study ID	Hosp	Country	Number	Age in days	Immunization	MV	CRRT	ECMO	MV dur	PICU LOS
			Patients	(mean (SD) [range])	(%)	(%)	(%)	(%)	(mean (SD)	(mean (SD)
Smith et al. [25]	PICU	UK	9	md	0	100	md	11,1	6.4 (9.2)	md
Pierce et al. [22]	PICU	UK	13	34 (20.3) [14–90]	md	84,6	md	md	7.6 (5)	15.3 (14.1)
Mikelova et al. [20]	Mixed	Canada	48	[md]	10.4	50	md	10,4	md	md
Rocha et al. [23]	PICU	Portugal	18	28 (14.3) [6–58]	0	33,3	md	5,5	md	14.2 (11.8)
Berger et al. [15]	PICU	USA	127	51,3 (32.2) [md]	26	43,3	15,7	9,4	10.3 (9.7)	md
Borgi et al. [16]	PICU	Tunis	17	236,7 (489.8) [24–630]	29	100	md	md	md	md
Winter et al. [28]	Mixed	USA	236	[md]	12	26,7	md	7,6	md	md
Straney et al. [26]	PICU	Australia & NZ	416	61 (52.5) [md]	md	31	md	1,6	3.9 (5)	4.5 (4.5)
Palvo et al. [21]	Mixed	Brazil	55	504,5 (409.6) [17–1885]	36	21,8	md	md	15.7 (10.3)	md
Kazantzi et al. [18]	PICU	Greece	31	61,7 (42.7) [md]	19	41,9	md	md	md	10.3 (13.9)
Abu-Raya et al. [29]	Mixed	Canada	1402	77 (57.13) [7–6202]	48.5*	0	md	md	md	22 (13.3)
Sik et al. [24]	PICU	Turkey	18	41,4 (2.6) [5, 38–47]	0	38,9	md	md	10.3 (4.7)	md
Kavitha et al. [17]	PICU	India	36	28 (29.7) [md]	md	30,6	md	md	md	md
Liu et al. [19]	PICU	China	59	[md]	3	86,4	1,7	1,6	md	md
Shi et al. [5]	Mixed	China	144	70 (67.4) [3–1460]	29	20,1	4,2	1,3	md	9.5 (11.9)
Tuan et al. [27]	PICU	Vietnam	73	139,75 (9.3) [1–446]	7	100	34,2	26,0	9.3 (5.3)	16 (12.1)
Coquaz-Garoudet et al. [11]	PICU	France	23	46 (32.3) [md]	8	78,2	26,1	30,4	md	md

CRRT continuous renal replacement therapy, ECMO extracorporeal membrane oxygenation, MV mechanical ventilation, MV dur mechanical ventilation duration, PICU LOS paeditric intensive care unit length of stay, md missing data, SD standard deviation

Pulmonary hypertension was reported in 14 of the 17 studies, with a pooled prevalence of 21% (95% CI: 12–30%; I² = 95) (Supplementary Fig. 1A). The use of organ support varied across studies. The pooled prevalence of mechanical ventilation requirement was 55% (95% CI: 40–70%; I² = 98) (studies reporting 16/17), CRRT requirement 15% (95% CI: 3–27%; I² = 95) (studies reporting CRRT 5/17) and ECMO requirement 8% (95% CI: 3–12%; I² = 93) (studies reporting ECMO 11/17) (Fig. 2 A–C). The use of exchange transfusion was reported in 12/17 studies with a pooled prevalence of 12% (95% CI: 6–18%) (Supplementary Fig. 1B). Mortality rates varied significantly (I² = 96%), reflecting population and study setting differences. The pooled mortality rate was 19% (95% CI: 12–26%) (15/17 studies included) (Fig. 2D).

Fig. 2.

Forrest plot of included studies reporting the severity of severe Bordetella pertussis infection. A Mechanical ventilation pooled prevalence. 16/17 studies reported mechanical ventilation. B Continuous renal replacement therapy (CCRT) pooled prevalence. 5/17 studies reported CRRT. C Extracorporeal membrane oxygenation (ECMO) pooled prevalence and 11/17 studies reported ECMO. D Mortality pooled prevalence, 15/17 reported mortality. CI, confidence interval

Review of mortality risk factors

Mortality risk factors were classified into clinical and biological variables. Reporting of these factors varied, with some studies presenting age and WBC as continuous or categorical.

Clinical variables associated with mortality included tachycardia (mean difference 18.2 bpm; 95% CI: 7.2–29; p < 0.001; I² = 81) (5/17 studies), seizures (OR 4.06; 95% CI: 1.7–9.3; p < 0.001; I² = 48) (9/17 studies) and pulmonary hypertension (OR 43.1; 95% CI: 13.5–137.8; p < 0.001; I² = 74) (10/17 studies). In contrast, apnea was not associated with mortality (OR 0.97; 95% CI: 0.3–3.5; p = 0.66; I² = 70) (8/17 studies) (Fig. 3 A–D). Age was not associated with mortality as a continuous variable (mean difference -3.5 days; 95% CI: -11.7–4.5; p = 0.39; I² = 0) (9/17 studies) (Supplementary Fig. 2 A–D). Dichotomic age cut-offs of 2 months (6/17 studies) and 3 months (5/17 studies) were also not associated with mortality.

Fig. 3.

Forrest plot of included studies reporting clinical risk factors for non-survival. A Mean difference for heart rate, favoring non-survival. 5/17 studies included. B Pooled odds ratio of the presence of pulmonary arterial hypertension, favoring non-survival. 10/17 studies included. C Pooled odds ratio of the presence of seizure, favoring non-survival. 9/17 studies included. D Pooled odds ratio of the presence of apnea. 8/17 studies included. CI, confidence interval

For biological variables, WBC counts were consistently reported. Elevated WBCs at admission (32.5 G/L; 95% CI: 20.7–44.3; p < 0.001; I² = 72) (8/17 studies) and peak WBC counts (43.4 G/L; 95% CI: 30–56.8; p < 0.001; I² = 84) (8/17 studies) were strongly associated with mortality (Fig. 4). Similar associations were observed for higher admission lymphocyte levels (7/17 studies), neutrophil counts (4/17 studies), and peak lymphocyte counts (6/17 studies) (Supplementary Fig. 3 A–C). Studies reporting WBCs as categorical variables with thresholds of 30 G/l (5/17 studies) or 50 G/l (5/17 studies) confirmed a strong association with mortality (Supplementary Fig. 3 D and E).

Fig. 4.

Forrest plot of included studies reporting biological risk factors for non-survival. A Mean difference for admission total white blood cell count, favoring non-survival. 8/17 studies included. B Mean difference for peak total white blood cell count, favoring non-survival. 8/17 studies included. CI, confidence interval

Validation cohort characteristics

The validation cohort comprised 83 patients with a median age at admission of 45 days (IQR: 30–55 days). The diagnosis was confirmed with PCR in 73 patients, serology in 2 patients, and clinical criteria in 8 patients. Pulmonary hypertension was diagnosed in 12/83 (14,5%) patients. Mechanical ventilation was required in 27/83 (32,5%) patients, with a median ventilator support duration of 6 days (IQR: 3–11). CRRT was used in 5/83 (5%) patients and ECMO in 4/83 (2/4 surviving). Complete demographic and organ failure data are provided in Supplementary Table 2. Median PICU stay was 8 days (IQR: 4–12), and cohort mortality was 12% (10/83).

Factors associated with mortality

Variables identified in the meta-analysis and the cohort were tested in a bivariate analysis (Table 2). Three factors—admission WBC > 30 G/l, age < 30 days, and tachycardia > 200/min—were incorporated into a multivariate model. Optimal cut-off values were selected to maximize the predictive accuracy.

Table 2.

Univariate and multivariate analysis of factors associated with in-PICU survival

Variables	OR (95% CI)	p value	Missing data
Univariate analysis
Age (days)	0.95 (95% CI 0.9–0.99)	0.048	0
PRISM 2 score	1.15 (95% CI 1.04–1.27)	0.007	13
PELOD score	1.12 (95% CI 1.01–1.25)	0.022	13
Leucocytes (G/l) on admission	1.08 (95% CI 1.04–1.1)	< 0.001	4
Peak leucocytes (G/l)	1.06 (95% CI 1.03–1.1)	< 0.001	5
Heart rate (/min) on admission	1.06 (95% CI 1.01–1.1)	0.004	2
Pulmonary hypertension	210 (95% CI 19.6–2240.2)	< 0.001	0
Mechanical ventilation	NA predict perfectly	NA	0
CRRT	48 (95% CI 4,6–500,4)	0.001	0
Blood exchange	31.7 (95% CI 6,2–161,8)	< 0.001	0
Multivariate model
Admission WBC > 30 G/l	10.8 (95% CI 1.5–77.9)	0.018
Age < 30 days	23 (95% CI 2.2–236.7)	0.008
Heart rate > 200/min	44.8 (95% CI 2.1–964.6)	0.015

Forward multivariate model Hosmer–Lemeshow test p-value = 0.56, R² 0.44

NA, none applicable; OR, odds ratio (95% confidence intervals); PRISM, paediatric risk of mortality score; PELOD, paediatric logistic organ dysfunction; CRRT, Continuous renal replacement therapy; WBC, white blood cells

The resulting model demonstrated a substantial predictive value (AUC 0.92; 95% CI: 0.85–0.99) with an R² of 0.44 (Table 2). A bedside score was developed, assigning 1 point for WBC > 30 G/l, 2 points for age < 30 days, and 4 points for heart rate > 200/min. The composite score showed excellent predictive ability (AUC 0.92; 95%CI: 0.86–0.99). No patients with a score of 0 died, while mortality rates increased significantly with a score of 3 or above. The complete scoring system is detailed in Table 3.

Table 3.

Mortality rate according to the pertussis mortality score

Score items	Point allocated
WBC > 30 G/L	1
Age < 30 days	2
Heart rate > 200/min	4

Pertussis mortality score*	Survivors (N(%))	Non-survivors (N(%))
0	46 (100)	0 (0)
1	11 (92)	1 (8)
2	12 (85)	2 (15)
3	3 (43)	4 (57)
5	1 (25)	3 (75)

Simplified pertussis score**
0	46 (100)	0 (0)
≥ 1	27 (59)	10 (100)

% for row numbers

WBC, total white blood cell

^*Complete pertussis mortality score: AUC 0.92 (95% CI0.86–0.99)

^**Simplified pertussis score sensitivity is 100%, and specificity is 63%. AUC 0.81 (95% CI 0.71–0.89)

Discussion

Studies examining PICU-admitted pertussis infections highlight significant mortality rates and a high requirement for organ support. Through meta-analysis and cohort validation, we identified three critical mortality risk factors: tachycardia > 200 /min, young age (< 30 days), and leucocytes > 30 G/l. These factors delineate cases with very high mortality risk.

The recent resurgence of pertussis infection in Europe underscores the importance of ongoing prevention efforts and early identification of severe infections [6, 30–32]. Despite advancements in organ support within PICUs and robust public vaccination strategies, PICU admission and mortality rates have remained stable over the past two decades [29, 33, 34]. While not fully understood, this stability highlights the need for further progress in managing pertussis infections, as emphasized by the PERISCOPE consortium [10]. Enhancing maternal immunization, vaccinating caregivers and close family members, and increasing public awareness are pivotal strategies to reduce pertussis-related mortality [2, 29, 35–37]. Early testing and diagnosis of pertussis, especially in high-risk populations during epidemic periods, are crucial. Prompt detection facilitates timely treatment initiation and referral to PICU when necessary, potentially improving outcomes [1, 11, 38].

This meta-analysis highlighted the severity of pertussis infection in infants admitted to the PICU, where many patients require organ support [39–41]. While less frequently employed, ECMO remains a topic of debate due to its low survival rates compared to other conditions, such as meconium aspiration syndrome or respiratory syncytial virus pneumonia [42]. However, ECMO support may still have a role in selected cases, particularly when combined with leukoreduction therapy, as suggested by Domenico et al. [43].

Besides antibiotics treatment and organ failure support, leukoreduction therapies, including leukapheresis or whole blood exchange, remain the cornerstone for managing severe pertussis infections associated with hyperleukocytosis [44]. Hyperleukocytosis has been repeatedly linked to organ failure and mortality, making it a critical therapeutic target [12, 45, 46]. Manual blood exchange, readily available in most centers, was the most reported leukoreductive therapy, with a pooled prevalence in the meta-analysis of 12%. Early leukoreduction has been associated with improved outcomes, as suggested by Rowlands et al. and corroborated by Coquaz-Garoudet et al. [11, 12].

Early recognition of severe pertussis infection is vital to enable timely PICU referral, organ support initiation, and leukoreduction, all of which can improve patient outcomes [11, 12]. Herein, we propose a simple bedside mortality risk score incorporating WBC > 30 G/L, age < 30 days, and tachycardia > 200/min. These variables reflect critical aspects of pertussis pathophysiology. Elevated WBC may contribute to hyperviscosity and microvascular occlusion or significantly increase bacterial toxin load [1, 7, 9].

Tachycardia, referred to as “tachycardia sine materia”, is indicative of systemic inflammation and potential cardio-pulmonary failure and is a hallmark of severe pertussis infection [8]. Pulmonary hypertension, a well-documented feature of severe pertussis, may result from hyperleukocytosis as well as a direct effect of pertussis toxin on the endothelium but is less practical to access in resource-limited settings, making heart rate a more accessible proxy [1, 7]. Finally, an age of less than 30 days, although not identified in the meta-analysis, was a decisive risk factor in our cohort and other studies [34, 40, 47–49]. It likely reflects the vulnerability of unvaccinated infants with potentially immature immune systems that may lead to a higher Bordetella inoculum, as suggested by Nakamura et al. [2, 50, 51].

Our study has several limitations. Despite a comprehensive literature review, some relevant studies may have been missed. Restricting the review to English-language articles may have introduced a bias, though evidence suggests this limitation has minimal impact on systematic review conclusions [52]. Data transformation from original articles and the heterogeneity in the study settings, spanning 24 years and multiple countries, may affect the generalizability of pooled findings. Additionally, our cohort study was monocentric and retrospective, warranting caution in extrapolating results to broader populations. Nonetheless, the robust methodological approach and alignment with clinical understanding of severe pertussis infection strengthen the validity of our findings.

Conclusion

This meta-analysis and cohort study highlight the severity of severe pertussis infection in infants and identify key risk factors for mortality. We propose a simple bedside-based mortality risk score based on WBC > 30 G/L, age < 30 days, and tachycardia > 200/min. While promising, this score requires further external validation and real-world testing.

Author contributions

"V.L.C., C.C., J.V.,and P.T. designed the study, collected the data and wrote the main manuscript, V.L.C., C.C., J.V. R.J, M.B., C.M. and P.T. analyzed the data, critically reviewed and approved the manuscript. "

Funding

None.

Availability of data and materials

Data will be available to the contact the corresponding author upon reasonable request at pierre.tissieres@aphp.fr.

Declarations

Ethics approval and consent to participate

The study was approved by the ethical committee from the French Society of Intensive Care (CE SRLF 23–074). The study has been registered at the “Commission Nationale de l’Informatique et des Libertés” corresponding to the reference methodology (MR-004). The data was processed following European legislation. A note informed patients and their families in the welcome booklet that their child’s clinical data could be collected for research purposes and of their right to decline such research.

Competing interests

The authors declare no competing interests.