Assessment of predictive efficacy for prognosis of pediatric severe Mycoplasma pneumoniae pneumonia: a comprehensive study based on high-resolution computed tomography imaging and platelet parameters

Abstract

Background

Severe Mycoplasma pneumoniae pneumonia (SMPP) in children can be associated with persistent pulmonary abnormalities and delayed recovery, making early identification of patients at risk of an unfavorable post-discharge physiological outcome clinically important for follow-up planning. This study aimed to evaluate the associations of early high-resolution computed tomography (HRCT) findings and hematologic parameters with a 3-month Pediatric Critical Illness Score (PCIS)-defined physiological outcome in children with SMPP and to explore the internal performance of a parsimonious model combining complementary imaging and hematologic information.

Methods

This secondary analysis used a publicly available, de-identified dataset containing 153 children with SMPP, of whom 39 had PCIS <80 three months after discharge. Distribution-appropriate comparative methods were used. Derived platelet ratios were excluded from primary model development to avoid redundancy with their component variables. A parsimonious logistic model incorporated lobar pneumonia, lung consolidation, and platelet count (PLT); PLT was modeled using a 3-knot restricted cubic spline because of evidence of nonlinearity. Internal validation used 1,000 bootstrap resamples. Discrimination, calibration, Brier score, and decision-curve analysis were evaluated.

Results

Lobar pneumonia and lung consolidation were more frequent and PLTs were lower among children with PCIS <80. PLT showed a significant nonlinear association with the outcome. The model had an apparent area under the receiver operating characteristic curve (AUC) of 0.890 [95% bootstrap confidence interval (CI): 0.832–0.940] and an optimism-corrected AUC of 0.877. The apparent and optimism-corrected Brier scores were 0.117 and 0.129, respectively; the optimism-corrected calibration intercept was −0.041 and the calibration slope was 0.890.

Conclusions

Early HRCT abnormalities and PLT provided complementary information regarding a 3-month PCIS-defined physiological outcome in this severe-disease cohort. Because of the limited number of outcome events and the absence of external validation, the model should be considered exploratory and should not yet be used as a clinical decision rule.

Highlight box.

Key findings

• In children with severe Mycoplasma pneumoniae pneumonia (MPP), lobar pneumonia and lung consolidation on early high-resolution computed tomography (HRCT) were independently associated with an adverse 3-month Pediatric Critical Illness Score (PCIS)-defined physiological outcome. Platelet count (PLT) showed a significant nonlinear association with the outcome. The combined exploratory model achieved an optimism-corrected area under the receiver operating characteristic curve of 0.877.

What is known and what is new?

• HRCT abnormalities and hematologic inflammatory or coagulation-related markers have individually been associated with disease severity and complications in pediatric MPP.

• This study jointly evaluated early HRCT findings and directly measured hematologic information for a 3-month PCIS-defined physiological outcome. A parsimonious model combining lobar pneumonia, lung consolidation, and a nonlinear platelet-count effect provided complementary imaging and systemic hematologic information.

What is the implication, and what should change now?

• Early HRCT findings combined with PLT may help identify children who could benefit from closer post-discharge surveillance. However, the model is exploratory and requires independent multicenter external validation before its risk estimates or probability thresholds are used for clinical decision-making.

Introduction

Mycoplasma pneumoniae (M. pneumoniae) is an important cause of community-acquired lower respiratory tract infection, particularly pneumonia in school-aged children and adolescents. In a large multicenter study of children hospitalized with radiographically confirmed community-acquired pneumonia, Mycoplasma pneumoniae was detected in approximately 8% of children overall and in approximately 19% of those aged 5–17 years (1). This clinical diversity is difficult to assess at a young age in routine pediatric practice. Children with similar clinical presentations at admission may subsequently follow divergent inpatient and post-discharge courses, particularly when pulmonary lesions persist despite appropriate antimicrobial treatment or when systemic inflammation remains active. Recent evidence suggests that host immune responses, airway injury, inflammatory mediators, and coagulation-related parameters contribute to disease severity in addition to pathogen burden (2-4). Thus, a pragmatic strategy would combine local pulmonary data with systemic markers that would reflect inflammatory and immune status in the child. Severe Mycoplasma pneumoniae pneumonia (SMPP) is clinically characterized by substantial respiratory involvement and/or rapidly progressive pulmonary or extrapulmonary disease, including persistent high fever, hypoxemia or respiratory compromise, extensive or progressive pulmonary infiltrates, consolidation or atelectasis, pleural involvement, plastic bronchitis, necrotizing pneumonia, and clinically significant extrapulmonary complications (5-8). Reported frequencies of severe disease vary across cohorts according to severity definitions, referral setting, age distribution, epidemic period, and regional disease characteristics.

Although estimates vary according to population and severity definition, one pediatric cohort reported that approximately 12% of Mycoplasma pneumoniae pneumonia (MPP) cases progressed to severe disease (9), whereas a computed tomography (CT)-based cohort reported severe MPP in 16.2% (122/752) of children (10), illustrating substantial between-cohort heterogeneity. Children may also suffer from delayed recovery, prolonged hospital stay, and a tendency to have persistent airway and/or parenchymal changes. Previous studies have linked atopic predisposition with greater disease severity; serum inflammatory markers and systemic immune indices with more severe clinical courses; D-dimer, platelet-related measures, and soluble P-selectin with inflammatory and coagulation burden; and CT findings such as lobar involvement, consolidation, pleural effusion, and necrotizing change with complicated disease (2,6,11-17). These previous findings indicate that several biological and imaging domains may carry clinically relevant information, but they do not establish that routinely collected demographic or clinical variables are intrinsically uninformative. The present study therefore evaluated clinical, high-resolution CT (HRCT), and hematologic information within the same cohort and examined their associations with the prespecified 3-month outcome.

HRCT is a technique that can visualize the structure of the lungs with high resolution, and is useful especially in cases of severe, persistent, and complicated pneumonia in children. HRCT can better reveal the distribution of lesions in segments or lobes, interstitial change, pleural effusion, bronchial wall thickening, atelectasis, and lung consolidation than conventional radiography. CT findings have been associated with disease severity, treatment response, and complications in pediatric MPP (6,13,18). In children with severe disease, lobar involvement and consolidation are not merely descriptive imaging signs. They may indicate dense inflammatory exudation, impaired ventilation, mucus obstruction, alveolar filling, and regional gas-exchange dysfunction. These processes can explain why certain HRCT features are clinically meaningful for prognosis rather than only for diagnosis. Cross-pathogen pediatric evidence also indicates that CT lesion patterns can provide clinically relevant severity information in severe pneumonia caused by other respiratory pathogens, including human metapneumovirus-associated disease (19). However, directly comparable post-discharge pediatric models jointly integrating CT findings and hematologic inflammatory markers remain limited.

Concurrently, the role of platelet parameters has gained increasing attention in pediatric pneumonia. Platelets are involved in hemostasis, endothelial interactions, immune modulation, cytokine release, recruitment of leukocytes and in the coagulation-inflammation network. Platelet count (PLT) and platelet indices can be altered in severe infection due to inflammatory consumption, marrow response, endothelial activation, immune-mediated mechanisms, and/or microthrombotic processes. Lymphocytes are also of importance because M. pneumoniae infection may also affect the cellular immunity, and both reduction and augmentation of lymphocytes may be seen in different phases of the disease and in different populations. Platelet- and lymphocyte-related measures, including platelet-to-lymphocyte ratio (PLR) and mean platelet volume-to-lymphocyte ratio (MPVLR), have been investigated as accessible indicators of inflammatory, immune, and coagulation-related status in pediatric MPP (14-17,20,21).

Despite this growing interest, a disconnection remains between imaging-based assessment and blood-based inflammatory assessment in the available literature. Imaging features were used in many previous studies and laboratory indicators were few, while only a few studies tried to combine HRCT features with platelet parameters in a single prognostic framework. This is relevant because the HRCT mainly reflects local pulmonary injury and the indices of platelets and lymphocytes, systemic immune and coagulation response. A combination of both might thus afford greater biological coverage than either one alone. It can also be implemented in routine care because HRCT and complete blood count parameters are commonly available during acute hospitalization for severe pneumonia.

The present study was designed to address this clinical need by evaluating the prognostic value of HRCT imaging features and platelet parameters in children with SMPP. Using the publicly available de-identified cohort of 153 children, the present secondary analysis compared clinical characteristics, early HRCT features, and hematologic parameters between the two source-defined 3-month Pediatric Critical Illness Score (PCIS) outcome groups and constructed a parsimonious exploratory multivariable model using non-redundant early predictors. The model was internally evaluated for discrimination, calibration, overall prediction error, and potential clinical net benefit. The aim was to construct and internally evaluate an exploratory model combining non-redundant early HRCT and hematologic information; the model was not regarded as ready for clinical implementation without independent external validation. This study differs from previous prediction studies in three ways. Guan et al. developed and evaluated a nomogram for pulmonary embolism in children with SMPP, demonstrating the feasibility of multivariable risk tools in this population but addressing a complication-specific endpoint (22). Li et al. focused on acute severity prediction (23), whereas Xie et al. developed an interpretable machine-learning model for early identification of severe disease (24). In contrast, the present study evaluates a 3-month PCIS-defined physiological-status outcome and combines early HRCT findings with directly measured hematologic information. It also integrates HRCT lesion features with platelet-related inflammatory parameters, thereby combining local pulmonary injury and systemic inflammatory/coagulation information in a single clinically interpretable model. We present this article in accordance with the TRIPOD+AI reporting checklist (25) (available at https://tp.amegroups.com/article/view/10.21037/tp-2026-0748/rc).

Methods

Study design and setting

This study was a secondary analysis of a publicly available, de-identified dataset originally derived from pediatric patients with SMPP treated at The 921st Hospital of the Joint Logistics Support Force, Changsha, China, between June 2024 and January 2025. The present analysis involved no new participant recruitment, intervention, or access to direct identifiers. Because the analytic cohort was fixed by the public source dataset rather than assembled through new hospital recruitment, current hospital bed capacity and annual pediatric pneumonia volume were not included in the public data and could not be independently verified; these institutional statistics were therefore not used to characterize the analytic sample.

This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The source study reports approval by the Ethics Committee of The 921st Hospital of the Joint Logistics Support Force (Approval No. 92120250711RB01). The present work was a secondary analysis of publicly available, de-identified data; no new participants were recruited, no identifiable information was accessed, and no additional institutional ethics approval or participant consent was required for the present secondary analysis.

According to the source-study documentation, eligible cases were identified from the hospital electronic medical record system and independently reviewed by two pediatricians using predetermined eligibility criteria; disagreements were adjudicated by a third senior physician. These procedures were performed during creation of the source cohort. The present secondary analysis did not access the hospital electronic medical record system and analyzed only the publicly available, de-identified analytic dataset.

Participants and diagnostic criteria

The publicly available analytic dataset contained 153 children who had met the source-study eligibility criteria, and all 153 records were included in the present secondary analysis. The inclusion criteria were: complete clinical data, 3-month follow-up information, diagnosis of SMPP, confirmation of M. pneumoniae in the serum by positive serum M. pneumoniae immunoglobulin M (IgM) antibody and/or positive M. pneumoniae quantitative polymerase chain reaction (qPCR) from nasopharyngeal aspirate (NPA), and clinical assessment and follow-up were satisfactory and the patient was compliant with them. The public analytic dataset retained the source-study eligibility criterion of positive serum M. pneumoniae IgM and/or positive nasopharyngeal-aspirate qPCR but did not contain patient-level indicators identifying which diagnostic modality was positive in each child. Consequently, IgM-only, qPCR-only, and dual-positive proportions could not be reconstructed from the data available for this secondary analysis. The clinical signs of severe disease were defined as the presence of one or more of the following: requirement for oxygen; limitation of respiratory function; severe pulmonary lesions (consolidation, atelectasis, pleural effusion); extrapulmonary complications (myocarditis, encephalitis or damage to the hematologic system); rapid progression; and failure to respond to standard therapy. These eligibility criteria were inherited from the source study, which applied American Thoracic Society-based severe-disease criteria together with one or more severe clinical features. The public analytic dataset did not retain the patient-level oxygenation thresholds, temporal radiographic-progression criteria, or exact drug regimen and duration used to operationalize “respiratory dysfunction”, “rapid progression”, or “poor response to conventional treatment”. These source-study criteria therefore could not be independently reconstructed or redefined in the present secondary analysis.

Exclusion criteria were established to minimise the influence of confounding factors which may be associated with imaging results or platelet parameters. Exclusion criteria for children included the use of hormonal drugs prior to admission, and congenital bronchopulmonary dysplasia, underlying pulmonary and respiratory diseases including pulmonary fibrosis, asthma and pulmonary tuberculosis, as well as hematologic diseases, coagulation dysfunction and other infections by other pathogens. The exclusions were needed because chronic pulmonary diseases can alter the appearance of HRCT and hematologic/coagulation disorders can alter the HRCT appearance without being related to the severity of the pneumonia. The flow of records through the present secondary analysis is shown in Figure 1.

Figure 1.

Flow of records through the present secondary analysis. Source-study screening and pre-publication exclusion counts were not available in the public analytic dataset and therefore could not be reconstructed. PCIS, Pediatric Critical Illness Score.

Clinical data collection

The public analytic dataset contained sex, age, body mass index, place of residence, fever duration, disease-course duration, wheezing, cough type, antibiotic-treatment duration, hospitalization duration, pulmonary atelectasis, lobar pneumonia, pleural effusion, lung consolidation, PLT, platelet distribution width (PDW), mean platelet volume, platelet-large cell ratio (P-LCR), and lymphocyte count (LY). PLR and MPVLR were recalculated from the patient-level component variables for descriptive and sensitivity analyses. Demographic and presenting clinical variables originated from the admission assessment; HRCT features were based on imaging obtained within 24 hours of admission; and hematologic variables were derived from the first fasting venous blood sample after admission. Antibiotic-treatment duration and hospitalization duration accrued after admission and were considered descriptive in-hospital variables rather than early model predictors. The 3-month PCIS-defined outcome grouping was supplied by the source dataset.

The duration of fever was documented in days and the number of days from onset of symptoms to admission or recorded clinical assessment was used to determine disease course. Cough was categorized as moist cough or dry cough according to medical record documentation. All variables were checked for completeness before statistical analysis.

Prognostic grouping and PCIS assessment

The 3-month outcome classification was inherited from the source dataset. In the source study, PCIS ≥80 defined the comparison group and PCIS <80 defined the adverse physiological-status group. This cutoff was prespecified in the source data and was not selected, optimized, or recalibrated using the present analysis. The 80-point boundary was retained solely to reproduce the source-study endpoint and should not be interpreted as a validated pneumonia-specific prognostic threshold. PCIS is a general pediatric critical-illness score rather than a pneumonia-specific pulmonary outcome; therefore, the endpoint is interpreted here as a source-defined 3-month physiological-status outcome. The present secondary analysis did not independently remeasure PCIS or validate the score for post-discharge pneumonia recovery. Mortality was not an outcome of the present analysis.

The source-defined follow-up time was three months after discharge. The public analytic dataset contained the resulting PCIS-based outcome group but did not contain the individual follow-up physiological components, serial imaging findings, pulmonary-function measurements, quality-of-life scores, or mortality outcomes. The present secondary analysis therefore used only the supplied 3-month PCIS-defined physiological-status grouping.

HRCT acquisition and imaging evaluation

A 64-row 128-slice chest CT scanner (Philips Brilliance) was used to obtain HRCT scans of all children within 24 hours of admission. Children were scanned in the supine position, consistent with the standardized HRCT protocol reported in the source study. The scan was from the apices of the lungs to the area under the diaphragm. The scanning parameters were set as follows: tube voltage of 120 kV, tube current of 80–160 mA, layer thickness of 5 mm, layer spacing of 5 mm, and pitch of 1.2 mm. All children in the study had a standardized CT protocol for severe pneumonia that was used routinely by the hospital, and the same equipment, scanning position and fixed parameters were applied to all children in the study. The standardization of the imaging protocol allowed for decreased technical variability and comparability of patients.

Pulmonary atelectasis, lobar pneumonia, pleural effusion, and lung consolidation were recorded as present or absent according to the source-study imaging assessment. These features represent clinically relevant structural manifestations of severe pneumonia, including reduced ventilation, parenchymal inflammatory involvement, and pleural reaction.

According to the source-study documentation, HRCT images were independently and blindly evaluated by two deputy chief physicians with more than 10 years of experience in pediatric radiologic diagnosis, and disagreements were adjudicated by a third senior physician. Routine clinical radiology-report text was not available in the public analytic dataset. The present secondary analysis therefore used the source-provided adjudicated binary imaging variables and did not independently reinterpret the CT images or routine radiology reports.

Platelet and lymphocyte-related parameters

Fasting antecubital venous blood samples were collected after admission. Blood cells were analyzed using a Sysmex XN-3000 automated hematology analyzer (Sysmex, Japan) and PLT, PDW, mean platelet volume (MPV), P-LCR, and LY were obtained. MPVLR and PLR were recalculated from the raw complete blood count values before final analysis. For each child, MPVLR was calculated as MPV/LY and PLR was calculated as PLT/LY at the individual-patient level. Group-level summaries were then calculated from these individual-patient ratio values using distribution-appropriate descriptive statistics. During data verification, the spreadsheet-exported PLR and MPVLR columns were not used directly because they were not mathematically identical to MPV/LY and PLT/LY in individual records. These parameters have been chosen because they are regularly available, affordable and may provide relevant information about inflammatory activation, platelet consumption, coagulation response and immune status.

Platelet consumption may lead to a reduced PLT, while platelet activation parameters such as MPV and P-LCR may reflect changes in platelet size and reactivity. LY reflects the host immune response and may increase or decrease according to inflammatory burden, immune activation, and disease stage. PLR and MPVLR provide patient-level summaries of platelet-lymphocyte balance, but they should be interpreted cautiously because ratio values may be sensitive to calculation method and within-patient variation.

Model specification and internal validation

Because the cohort contained only 39 adverse outcomes and the derived platelet ratios introduced deterministic redundancy with their component variables, the primary analysis used a parsimonious exploratory logistic model containing lobar pneumonia, lung consolidation, and PLT. PLR and MPVLR were retained only for descriptive and sensitivity analyses and were not entered simultaneously with their component variables in primary model development. In the original six-variable formulation, simultaneous inclusion of component and ratio variables produced marked multicollinearity, with variance inflation factors of approximately 65.5 for PLT, 25.6 for LY, 11.9 for MPVLR, and 28.7 for PLR. The original formulation was therefore discontinued. PLT was modeled using a restricted cubic spline because the linearity assessment demonstrated evidence of nonlinearity. Three knots were placed at 192.26×10⁹/L, 233.60×10⁹/L, and 271.94×10⁹/L, corresponding to the 10th, 50th, and 90th percentiles of PLT.

Before model reduction, 17 early candidate predictor parameters were available: eight demographic or presenting clinical variables, four HRCT variables, and five directly measured hematologic variables. Antibiotic-treatment duration and total hospitalization duration were not treated as early candidate predictors because they accrued after admission, while PLR and MPVLR were excluded from primary modeling because they are derived from component variables. With 39 adverse outcomes, this corresponds to approximately 2.29 events per initial candidate parameter. A formal sample-size assessment was performed according to Riley et al. (26). Using an anticipated Cox-Snell R² of approximately 0.102 and a target global shrinkage factor of 0.90, the shrinkage criterion indicated a required sample size of approximately 1,416 participants. The available sample of 153 participants was therefore substantially below the calculated requirement; the model was consequently treated as exploratory rather than as a definitive clinical prediction model. Internal validation used 1,000 bootstrap resamples, with model fitting, spline specification, coefficient estimation, and performance assessment repeated within the bootstrap procedure. Because the revised primary analysis used a fixed parsimonious model specification rather than an automated variable-selection algorithm, no variable-selection step was repeated within the bootstrap resamples. Each bootstrap sample re-estimated the spline knots, model coefficients, and performance measures. Residual optimism arising from the data-informed revised model specification cannot be completely removed by this procedure and is therefore acknowledged as a limitation; independent external validation remains necessary. Model performance was evaluated using discrimination, calibration, and overall prediction error. Discrimination was quantified using the area under the receiver operating characteristic curve (AUC) with a 95% bootstrap confidence interval (CI). Calibration was assessed using the calibration intercept, calibration slope, Brier score, and a flexible observed-versus-predicted calibration curve with bootstrap uncertainty intervals; apparent and optimism-corrected estimates were reported separately in accordance with contemporary calibration guidance (27).

No probability threshold was prespecified for clinical decision-making because the relative consequences of false-positive and false-negative classifications were not established. Threshold-free measures were therefore emphasized. For exploratory purposes only, the Youden threshold was re-estimated within bootstrap resamples to quantify threshold instability; no single cutoff was regarded as clinically validated (28). Decision-curve analysis was used to evaluate potential clinical utility over threshold probabilities of 0.10–0.30. The modeled action was enhanced post-discharge surveillance rather than automatic treatment escalation. Model net benefit was compared with treat-all and treat-none strategies, and bootstrap resampling was used to quantify optimism and uncertainty. Potential utility was inferred only where model net benefit exceeded both comparator strategies (29).

Statistical analysis

Statistical analyses were performed using SPSS 25.0 and R. Categorical variables were summarized as n (%) and compared using the Chi-squared test or Fisher’s exact test, as appropriate. Continuous variables were evaluated separately within the two outcome groups using histograms, normal Q–Q plots, and the Shapiro-Wilk test. Variables with distributions reasonably compatible with normality were summarized as mean ± standard deviation (SD) and compared using independent-samples t-tests. Equality of variances was assessed using Levene’s test; Welch’s t-test was used when the assumption of equal variances was not supported. Variables showing material departure from normality were summarized as median and interquartile range (IQR) and compared using the Mann-Whitney U test. Distributional diagnostics are provided in Figure S1 and Table S1. Linearity in the logit was examined for continuous candidate predictors using 3-knot restricted cubic splines. Likelihood-ratio tests comparing spline and linear specifications were used to assess evidence of nonlinearity. False-discovery-rate (FDR) correction using the Benjamini-Hochberg method was performed as a sensitivity analysis for the univariate clinical, imaging, and laboratory comparisons.

Results

Baseline characteristics of the study population

The public analytic dataset contained 153 children with SMPP. At the source-defined 3-month assessment, 114 children (74.51%) had PCIS ≥80 and 39 (25.49%) had PCIS <80. The PCIS <80 group is interpreted here as an adverse physiological-status group rather than as a mortality or survival group.

Clinical and in-hospital characteristics are summarized in Table 1. After distribution-appropriate testing, antibiotic-treatment duration showed a nominal between-group difference (P=0.047), whereas the other listed clinical characteristics did not show statistically significant unadjusted differences. The antibiotic-duration association did not remain statistically significant after FDR correction. These descriptive comparisons do not establish that clinical variables are inferior to imaging or hematologic variables for prediction.

Table 1. Clinical and in-hospital characteristics according to 3-month PCIS-defined outcome.

Clinical characteristic	PCIS <80 (n=39)	PCIS ≥80 (n=114)	Test statistic	P value
Sex			χ2=0.351	0.55
Male	23	61
Female	16	53
Age (years)	9.0 (7.5–9.5)	8.0 (7.0–9.0)	U=2,557.0	0.15
BMI (kg/m2)	17.47 ± 1.63	18.03 ± 1.85	t=−1.705	0.09
Place of residence			χ2=1.790	0.18
Town	25	59
Countryside	14	55
Fever duration (days)	7.0 (7.0–8.0)	8.0 (7.0–9.0)	U=1,802.0	0.07
Disease course (days)	8.0 (7.0–11.0)	9.0 (8.0–10.0)	U=2,210.5	0.96
Wheezing			χ2=0.277	0.60
Yes	20	64
No	19	50
Cough nature			χ2=1.152	0.28
Moist cough	28	71
Dry cough	11	43
Antibiotic-treatment duration (days)	5.0 (4.0–7.0)	6.0 (4.0–7.0)	U=1,754.5	0.047
Hospitalization duration (days)	11.0 (9.0–12.0)	11.0 (10.0–13.0)	U=1,903.0	0.18

Data are presented as n (%), mean ± SD, or median (IQR), as appropriate. Distributional assessment determined the selected comparison method. Antibiotic-treatment duration and hospitalization duration accrued after admission and were considered descriptive in-hospital variables rather than early model predictors. BMI, body mass index; IQR, interquartile range; PCIS, Pediatric Critical Illness Score; SD, standard deviation.

HRCT imaging features according to 3-month PCIS-defined outcome

The prevalence of several HRCT findings differed between the two 3-month PCIS-defined outcome groups (Table 2). The PCIS <80 group had higher rates of pulmonary atelectasis (43.59% vs. 24.56%, P=0.02), lobar pneumonia (61.54% vs. 35.09%, P=0.004), pleural effusion (48.72% vs. 29.82%, P=0.03), and lung consolidation (66.67% vs. 27.19%, P<0.001).

Table 2. HRCT imaging features according to 3-month PCIS-defined outcome.

Imaging feature	PCIS <80 (n=39)	PCIS ≥80 (n=114)	Chi-squared	P
Pulmonary atelectasis	17 (43.59)	28 (24.56)	5.068	0.02
Lobar pneumonia	24 (61.54)	40 (35.09)	8.355	0.004
Pleural effusion	19 (48.72)	34 (29.82)	4.581	0.03
Lung consolidation	26 (66.67)	31 (27.19)	19.370	<0.001

Data are presented as n (%). HRCT, high-resolution computed tomography; PCIS, Pediatric Critical Illness Score.

Among the available HRCT variables, lobar pneumonia and lung consolidation showed the smallest unadjusted P values and remained statistically significant after FDR correction, whereas pulmonary atelectasis and pleural effusion were significant only before FDR correction. Lung consolidation was present in 66.67% of the PCIS <80 group and 27.19% of the PCIS ≥80 group (P<0.001).

Platelet and lymphocyte-related parameters

Platelet and lymphocyte-related parameters are summarized in Table 3. Compared with the PCIS ≥80 group, the PCIS <80 group had a lower PLT [(207.42±22.35 vs. 238.51±29.67) ×10⁹/L, P<0.001], a higher LY [(3.82±1.26 vs. 3.31±1.09) ×10⁹/L, P=0.02], and a lower recalculated PLR [52.71 (IQR, 46.11–69.99) vs. 73.56 (57.77–95.35), P<0.001]. Recalculated MPVLR did not differ significantly between groups [2.48 (1.81–3.40) vs. 2.87 (2.11–3.79), P=0.19]. PDW, MPV, and P-LCR also showed no statistically significant between-group differences.

Table 3. Platelet and lymphocyte-related parameters according to 3-month PCIS-defined outcome.

Parameter	PCIS <80 (n=39)	PCIS ≥80 (n=114)	Test statistic	P value
PLT (×109/L)	207.42±22.35	238.51±29.67	t=−5.983	<0.001
PDW (%)	13.50 (10.50–15.50)	11.90 (9.23–14.45)	U=2,459.5	0.32
MPV (fL)	9.51±3.16	9.26±3.07	t=0.450	0.65
P-LCR (%)	21.30±3.71	20.84±3.95	t=0.640	0.52
LY (×109/L)	3.82±1.26	3.31±1.09	t=2.409	0.02
MPVLR	2.48 (1.81–3.40)	2.87 (2.11–3.79)	U=1,909.0	0.19
PLR	52.71 (46.11–69.99)	73.56 (57.77–95.35)	U=1,286.0	<0.001

Data are presented as mean ± SD or median (IQR). MPVLR and PLR were recalculated at the individual-patient level using raw MPV, PLT, and LY values before group summarization. IQR, interquartile range; LY, lymphocyte count; MPV, mean platelet volume; MPVLR, mean platelet volume-to-lymphocyte ratio; P-LCR, platelet-large cell ratio; PCIS, Pediatric Critical Illness Score; PDW, platelet distribution width; PLR, platelet-to-lymphocyte ratio; PLT, platelet count; SD, standard deviation.

PLR and MPVLR were recalculated from their patient-level component variables to ensure internal mathematical consistency. The resulting PLR distribution was directionally compatible with the combination of lower PLT and higher LY in the PCIS <80 group. Because the source spreadsheet ratio columns were not mathematically identical to ratios recalculated from the component variables, only the recalculated values were used for the present descriptive and sensitivity analyses.

After FDR correction across the 21 clinical, HRCT, and laboratory comparisons, lung consolidation, lobar pneumonia, PLT, and recalculated PLR remained statistically significant. Pulmonary atelectasis, pleural effusion, LY, and antibiotic-treatment duration were nominally significant in the unadjusted analyses but did not remain significant after FDR correction, whereas MPVLR was not significant in the unadjusted analysis (P=0.19) (Table 4). These findings are reported as univariate associations and should not be interpreted as evidence of independent predictive ability.

Table 4. FDR sensitivity analysis for candidate findings.

Variable	Unadjusted P value	FDR-adjusted P value	Remained significant after FDR?
PLT	<0.001	<0.001	Yes
Lung consolidation	<0.001	<0.001	Yes
PLR	<0.001	<0.001	Yes
Lobar pneumonia	0.004	0.02	Yes
LY	0.02	0.07	No
Pulmonary atelectasis	0.02	0.09	No
Pleural effusion	0.03	0.10	No
Antibiotic-treatment duration	0.047	0.12	No
MPVLR	0.19	0.28	No

FDR correction was performed across the 21 clinical, HRCT, and laboratory comparisons using the Benjamini-Hochberg method. The table displays variables with unadjusted P<0.05 together with MPVLR, which is shown for comparison with PLR. FDR, false discovery rate; HRCT, high-resolution computed tomography; LY, lymphocyte count; MPVLR, mean platelet volume-to-lymphocyte ratio; PLR, platelet-to-lymphocyte ratio; PLT, platelet count.

Parsimonious model specification and nonlinear effects

Because of the limited number of outcome events and the substantial redundancy introduced by derived platelet ratios, the primary exploratory model was restricted to lobar pneumonia, lung consolidation, and PLT. Restricted cubic spline analysis demonstrated a nonlinear association between PLT and the study outcome after adjustment for the two HRCT variables (P for nonlinearity =0.002). In the predictor-wise spline assessments, evidence of nonlinearity was observed for fever duration (P=0.002), disease-course duration (P=0.04), and PLT (P=0.002). Age, body mass index (BMI), antibiotic-treatment duration, hospitalization duration, PDW, MPV, P-LCR, LY, MPVLR, and PLR did not show statistically significant evidence of nonlinearity (all P>0.05). Complete likelihood-ratio test results are provided in Table S2.

Logistic regression model

The parsimonious model contained lobar pneumonia, lung consolidation, and the nonlinear platelet-count effect. Lobar pneumonia was independently associated with the PCIS-defined adverse outcome [odds ratio (OR) =2.72, 95% CI: 1.07–6.94, P=0.04], and lung consolidation showed a stronger adjusted association (OR =6.18, 95% CI: 2.41–15.85, P<0.001). PLT showed a significant overall association with outcome risk [likelihood-ratio χ²=37.62, 2 degrees of freedom (df), P<0.001], with evidence of nonlinearity (P for nonlinearity =0.002) (Table 5). These coefficients quantify adjusted associations and should not be interpreted as measures of predictive performance; discrimination, calibration, overall prediction error, and clinical utility were evaluated for the complete model.

Table 5. Parsimonious exploratory logistic model for the 3-month PCIS-defined adverse physiological outcome.

Model term	Regression coefficient	Standard error	Z value	P value	Effect estimate/interpretation
Intercept	−1.0637	3.1189	−0.341	0.73	Model intercept
Lobar pneumonia, present vs. absent	1.0011	0.4776	2.096	0.04	OR 2.72 (95% CI: 1.07–6.94)
Lung consolidation, present vs. absent	1.8218	0.4802	3.794	<0.001	OR 6.18 (95% CI: 2.41–15.85)
PLT linear spline component	−0.000812	0.015416	−0.053	0.96	Spline component; not interpreted separately
PLT nonlinear spline component	−0.124361	0.049685	−2.503	0.01	Spline component; not interpreted separately

PLT was modeled with a restricted cubic spline using knots at 192.26×10⁹/L, 233.60×10⁹/L, and 271.94×10⁹/L. The overall association of PLT with the outcome was significant (likelihood-ratio χ²=37.62, 2 df, P<0.001), with evidence of nonlinearity (P for nonlinearity =0.002). Individual spline coefficients should not be interpreted as constant per-unit ORs. CI, confidence interval; df, degree of freedom; OR, odds ratio; PCIS, Pediatric Critical Illness Score; PLT, platelet count.

These findings support the value of integrating HRCT and hematologic information. Pulmonary atelectasis, lobar pneumonia, pleural effusion, and lung consolidation were assessed from HRCT images. The two domains therefore provided complementary information within the complete model.

The complete model equation was: Logit (p) = −1.063705 + 1.001131 (lobar pneumonia) + 1.821780 (lung consolidation) − 0.0008115 (PLT) − 0.124361 H (PLT), where lobar pneumonia and lung consolidation were coded 1 when present and 0 when absent, PLT was expressed in ×10⁹/L, and H (PLT) represents the restricted-cubic-spline nonlinear basis using knots at 192.26×10⁹/L, 233.60×10⁹/L, and 271.94×10⁹/L. The predicted probability was calculated as p=1/(1 + exp(−logit (p))).

For reproducibility, the nonlinear spline basis was defined as H (x) = {(x − k₁)_+^3 − [(k₃ − k₁)/(k₃ − k₂)](x − k₂)_+^3 + [(k₂ − k₁)/(k₃ − k₂)](x − k₃)_+^3}/(k₃ − k₁)^2, where k₁=192.26, k₂=233.60, k₃=271.94, and (z)_+ = max (z, 0).

Nomogram construction and internal calibration

An illustrative nomogram was constructed from the parsimonious exploratory model using lobar pneumonia, lung consolidation, and the nonlinear platelet-count effect (Figure 2). The nomogram translates the fitted model terms into an estimated probability of the 3-month PCIS-defined adverse physiological outcome. Because all model development and validation were internal, the nomogram is presented as an interpretable representation of the fitted model rather than as a clinically validated bedside decision rule.

Figure 2.

Nomogram derived from the parsimonious exploratory model combining lobar pneumonia, lung consolidation, and the nonlinear platelet-count effect. PLT, platelet count.

The apparent AUC of the parsimonious exploratory model was 0.890 (95% bootstrap CI: 0.832–0.940), whereas the optimism-corrected AUC was 0.877 (Figure 3). The apparent Brier score was 0.117 and the optimism-corrected Brier score was 0.129. Apparent calibration was, as expected in the development data, close to an intercept of 0 and slope of 1. After bootstrap optimism correction, the calibration intercept was −0.041 and the calibration slope was 0.890, indicating mild overfitting (Figure 4). The 95% bootstrap CIs were approximately −0.543 to 0.508 for the calibration intercept and 0.563 to 1.283 for the calibration slope. These estimates represent internal validation and should not be interpreted as evidence of external calibration.

Figure 3.

Receiver operating characteristic curve of the parsimonious exploratory HRCT-platelet model for the 3-month PCIS-defined adverse physiological outcome. The apparent AUC was 0.890 (95% bootstrap CI: 0.832–0.940). AUC, area under the receiver operating characteristic curve; CI, confidence interval; HRCT, high-resolution computed tomography; PCIS, Pediatric Critical Illness Score.

Figure 4.

Apparent and bootstrap optimism-corrected calibration of the exploratory HRCT-platelet model. The diagonal line indicates ideal calibration; flexible calibration smoothers show observed versus predicted outcome probability, and shading represents the 95% bootstrap uncertainty interval. HRCT, high-resolution computed tomography.

Exploratory bootstrap threshold analysis demonstrated substantial instability. The median internally re-estimated Youden threshold was 0.247, with a 95% bootstrap CI of approximately 0.131–0.486. The corresponding median sensitivity was 84.6% (95% bootstrap CI: 64.1–97.4%), specificity was 79.0% (63.2–90.4%), and accuracy was 80.4% (71.2–83.7%). Because the threshold and its operating characteristics were derived internally and showed considerable uncertainty, no probability cutoff is recommended for clinical use (Table 6).

Table 6. Apparent and bootstrap-validated performance of the parsimonious exploratory model.

Metric	Apparent estimate	Bootstrap/optimism-corrected estimate
AUC	0.890	0.877
AUC 95% bootstrap CI	0.832–0.940	–
Brier score	0.117	0.129
Calibration intercept	~0.000	−0.041 (95% bootstrap CI: −0.543 to 0.508)
Calibration slope	~1.000	0.890 (95% bootstrap CI: 0.563–1.283)
Exploratory resampled Youden threshold	–	Median 0.247 (95% CI: 0.131–0.486)
Sensitivity at resampled threshold	–	84.6% (64.1–97.4%)
Specificity at resampled threshold	–	79.0% (63.2–90.4%)
Accuracy at resampled threshold	–	80.4% (71.2–83.7%)
DCA model net benefit at threshold 0.10	–	0.203 (0.129–0.272)
DCA model net benefit at threshold 0.20	–	0.174 (0.098–0.242)
DCA model net benefit at threshold 0.30	–	0.130 (0.053–0.200)
Bootstrap resamples	–	1,000
External validation	Not performed	Required before clinical implementation

Threshold-specific operating characteristics were exploratory and were obtained from bootstrap re-estimation of the Youden threshold; no probability cutoff is proposed for clinical use. AUC, area under the receiver operating characteristic curve; CI, confidence interval; DCA, decision curve analysis.

Decision-curve analysis was interpreted over threshold probabilities of 0.10–0.30. The modeled clinical action was enhanced post-discharge surveillance, such as earlier outpatient reassessment or repeat laboratory evaluation, rather than automatic treatment escalation. At threshold probabilities of 0.10, 0.20, and 0.30, the optimism-corrected model net benefits were 0.203, 0.174, and 0.130, respectively (Figure 5). The corresponding 95% bootstrap CIs were 0.129–0.272, 0.098–0.242, and 0.053–0.200. Treat-all net benefits at the same thresholds were 0.172, 0.069, and −0.064, respectively. The model was considered potentially useful only where its net benefit exceeded both treat-all and treat-none strategies.

Figure 5.

Bootstrap optimism-corrected decision-curve analysis of the exploratory HRCT-platelet model over threshold probabilities of 0.10–0.30. Shading represents the 95% bootstrap uncertainty interval. The modeled clinical action is enhanced post-discharge surveillance rather than automatic treatment escalation. HRCT, high-resolution computed tomography.

Discussion

This secondary analysis evaluated associations of early HRCT and hematologic parameters with a source-defined 3-month PCIS-based physiological-status outcome in 153 children with SMPP. The principal findings were that lobar pneumonia and lung consolidation were independently associated with the PCIS <80 outcome, PLT showed a significant nonlinear association with outcome risk, and the parsimonious combined model showed good apparent discrimination but measurable optimism after internal bootstrap validation. These findings should be interpreted as exploratory because all model development and validation were performed within the same public analytic cohort.

Thirty-nine of 153 children (25.49%) belonged to the source-defined PCIS <80 group at three months. Published pediatric cohorts have reported severe-disease proportions of approximately 12% and 16.2%, respectively (9,10), while severe-disease characteristics and frequency vary across referral settings and study populations (5,8). Direct numerical comparison of those estimates with the present 25.49% post-discharge PCIS-defined outcome would therefore be inappropriate.

If independent external validation confirms adequate performance, the model could provide a common quantitative framework for multidisciplinary follow-up planning. For example, radiologists could explicitly communicate the presence of lobar involvement and consolidation; pediatricians could integrate these imaging findings with the platelet result when determining the timing of reassessment; nursing staff could coordinate earlier outpatient review for higher-risk children; and caregivers could receive clearer return precautions regarding persistent fever, respiratory difficulty, or delayed recovery. The model should not independently determine antimicrobial, corticosteroid, respiratory-support, or other treatment-escalation decisions.

Several HRCT findings were more frequent in the PCIS <80 group than in the PCIS ≥80 group, including pulmonary atelectasis, lobar pneumonia, pleural effusion, and lung consolidation. These between-group associations are biologically plausible but should not be interpreted as causal effects or as individual measures of predictive performance. Obstruction of the bronchi, plugging with mucus, inflammatory edema, or external compression are some of the causes of atelectasis. After the onset of atelectasis, alveolar ventilation will decrease, and local secretion retention can also increase, leading to a cycle of inflammation, obstruction, and impaired gas exchange. Pleural effusion may be a sign of profound local inflammation, and may help to limit respiration. Lobar pneumonia and lung consolidation indicate a large amount of involvement of the lung parenchyma, and are closely linked to alveolar exudation, inflammatory infiltration and decreased oxygen diffusion.

In the parsimonious model, lobar pneumonia and lung consolidation were independently associated with the source-defined PCIS <80 outcome. Lobar pneumonia was associated with 2.72-fold higher odds (95% CI: 1.07–6.94), whereas lung consolidation was associated with 6.18-fold higher odds (95% CI: 2.41–15.85). PLT showed a significant nonlinear association with outcome risk rather than a constant per-unit effect. These adjusted associations should not be interpreted as individual measures of predictive ability.

PLT also differed between the two outcome groups, with lower values in the PCIS <80 group. In the adjusted model, the association between PLT and the outcome was nonlinear; therefore, the effect should not be interpreted as a constant inverse change in odds per unit increase in PLT. Platelets are now known to play more roles than just hemostatic ones as immune-active cells. In severe infection and inflammatory lung injury platelets can adhere to injured endothelial cells, interact with leukocytes, and secrete inflammatory mediators, as well as engage in microvascular thrombosis. Low PLT could be due to consumption or redistribution in the inflamed pulmonary and vascular compartments. Previous studies have reported associations between M. pneumoniae disease severity and platelet-related or coagulation-related markers, including PLT, D-dimer, and soluble P-selectin (14-16). The present analysis supports a nonlinear association between PLT and the 3-month PCIS-defined outcome and therefore does not establish a constant protective or harmful effect for each unit change in PLT.

The PCIS <80 group had a higher LY in the univariate comparison (P=0.02); however, this association did not remain statistically significant after FDR correction. LY was not retained in the parsimonious primary model because the final specification prioritized a limited set of non-redundant predictors. Accordingly, the present analysis does not support an independent adjusted effect of LY.

After recalculation from patient-level component values, PLR was lower in the PCIS <80 group whereas MPVLR showed no statistically significant between-group difference. Because PLR and MPVLR are mathematically derived from component variables, neither ratio was entered simultaneously with its components in primary model development. Their findings are therefore interpreted as descriptive or sensitivity results rather than independent multivariable effects. Previous studies have reported varying associations of systemic immune-inflammatory parameters with SMPP, with effect estimates depending on cohort age, disease stage, severity definition, and measurement timing (12,17,20,24,30). The study did not perform a prespecified head-to-head comparison of a clinical-only model, an imaging-only model, and a laboratory-only model. Accordingly, the findings should not be interpreted as demonstrating superiority of one predictor domain over another. Rather, the complete exploratory model combines HRCT and platelet information that represent different biological domains and should be evaluated as a single internally validated model. The platelet and lymphocyte related parameters indicate systemic inflammatory, immune and coagulation changes. Local lung injury and systemic immune response are the major causes of severe pneumonia caused by M. pneumoniae infection. M. pneumoniae infection can involve immune dysregulation, cytokine release, airway epithelial injury, inflammatory exudation, and systemic immune changes that may connect local pulmonary abnormalities with hematologic findings (3,4,8,31,32). Combining these two domains provides a framework for jointly representing local pulmonary abnormalities and systemic hematologic information without implying that either predictor domain is superior.

The nomogram provides an interpretable representation of the parsimonious logistic model. It displays the contributions of lobar pneumonia, lung consolidation, and the nonlinear platelet-count effect to the estimated outcome probability. The apparent AUC was 0.890 and decreased to 0.877 after bootstrap optimism correction, while the optimism-corrected calibration slope was 0.890, indicating measurable but limited overfitting. Because these estimates were derived entirely within the development dataset, the nomogram should be considered exploratory until externally validated.

Limitations and responsible clinical interpretation

This study has several important limitations. First, only 39 adverse outcomes were observed. Model-development adequacy depends on the total number of candidate predictor parameters considered before reduction rather than only on the number retained in the final model. Seventeen early candidate parameters were available after exclusion of post-baseline duration variables and derived ratios, corresponding to approximately 2.29 events per candidate parameter. A Riley et al.-based sample-size assessment indicated that approximately 1,416 participants would be required under the stated assumptions for the target shrinkage criterion, substantially exceeding the available sample. The model should therefore be regarded as exploratory and at meaningful risk of overfitting despite bootstrap correction. Second, all validation was internal; independent multicenter external validation is required before the model or any probability threshold can be used clinically. Third, PCIS is not a pneumonia-specific pulmonary outcome and does not directly assess radiographic resolution, pulmonary function, airway obstruction, exercise tolerance, quality of life, or survival. Fourth, quantitative HRCT lesion burden, serial platelet trends, cytokine profiles, pulmonary-function testing, and standardized follow-up radiographic outcomes were unavailable for all participants.

The nomogram should be used only as an adjunct to clinical assessment. A high predicted risk may support closer monitoring, repeated assessment of inflammatory markers, follow-up imaging when clinically indicated, pulmonary function testing in age-appropriate children, and structured post-discharge review. However, treatment escalation should remain based on the child’s respiratory status, fever pattern, oxygen requirement, imaging progression, laboratory trends, complications, and contraindications to therapy. Prospective multicenter validation is required before the model can be used as a routine clinical decision rule.

Conclusions

In this secondary analysis of 153 children with SMPP, early HRCT findings and PLT were associated with a source-defined 3-month PCIS-based physiological outcome. A parsimonious exploratory model combining lobar pneumonia, lung consolidation, and the nonlinear platelet-count effect showed good apparent discrimination but measurable optimism after bootstrap correction. The principal contribution of this analysis is the joint evaluation of local pulmonary imaging and systemic hematologic information for a post-discharge outcome rather than acute severity alone. Because only 39 adverse outcomes were available and all validation was internal, independent external validation is required before the model, its risk estimates, or any probability cutoff can be used for clinical decision-making.

Supplementary

The article’s supplementary files as

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.