Answers are grounded in this paper and cite their sources.
Background Severe Mycoplasma pneumoniae pneumonia (SMPP) in children may progress rapidly during the early stage of hospitalization. Current clinical management lacks reliable biomarker-driven risk stratification tools, which may contribute to delayed recognition of deterioration. Early within-patient hematological and inflammatory trajectories and quantitative HRCT imaging capture complementary dimensions of disease activity, yet their incremental and complementary value at a fixed 24-h reassessment point has not been temporally validated. Objective To develop and temporally validate a multimodal model incorporating dynamic hematological and inflammatory trajectory scores with structured HRCT assessment and radiomics for predicting objective clinical deterioration during 24–72 h among children with SMPP who remained event-free at the 24-h landmark. The model was designed to estimate deterioration risk, not treatment response. Methods A retrospective cohort design was used. The development cohort included 550 children with SMPP who remained free of the primary outcome at the 24-h landmark. A further 300 children meeting the same landmark eligibility criterion and admitted to the same hospital during a later, nonoverlapping period comprised the temporal validation cohort. Candidate predictors included nine clinical variables, a structured HRCT severity score reflecting pulmonary pathological burden (consolidation, atelectasis, pleural effusion, bronchial wall thickening, necrosis), a radiomics-derived Rad-score, and two dynamic trajectory scores: a platelet trajectory score summarizing changes in PLT, MPV, PDW, and P-LCR, and an inflammation trajectory score summarizing changes in CRP, LDH, and D-dimer. The primary outcome was clinical deterioration during 24–72 h, defined by escalation to high-flow nasal cannula, noninvasive or invasive ventilation, unplanned PICU transfer, persistent hypoxemia requiring oxygen escalation, vasoactive support, or new severe extrapulmonary organ dysfunction. A penalized regression model was constructed to quantify risk, with performance assessed via discrimination, calibration, and decision curve analysis. Pre-landmark treatment exposures were recorded and evaluated separately in sensitivity analyses; their associations with subsequent deterioration were interpreted as noncausal because of confounding by indication. Results The primary outcome occurred in 140 of 550 children (25.5%) in the development cohort and in 73 of 300 children (24.3%) in the validation cohort. During the first 24 h, the median platelet count decreased by 12.2 × 10 9 /L among children who deteriorated, whereas it increased by 2.2 × 10 9 /L among those who did not deteriorate (P < 0.001). Model performance improved hierarchically across the prespecified models, with a bootstrap-corrected AUC of 0.826 for the Full model. In temporal validation, the locked Full model yielded an AUC of 0.832 (95% CI 0.772–0.884), a Brier score of 0.126, and a calibration slope of 0.853; the difference in AUC from the Clinical model was 0.163 (95% CI 0.100–0.231). The increment in AUC over the prespecified model without radiomics was 0.035 (95% CI 0.001–0.069; P = 0.042). At the prespecified low-risk threshold (8%), sensitivity was 0.932 and the negative predictive value was 0.949, identifying a lower-predicted-risk subgroup that warrants prospective evaluation as a potential rule-out group. Conclusion Early dynamic hematological and inflammatory trajectories provided complementary information to quantitative HRCT imaging for predicting post-landmark deterioration among children with SMPP who remained event-free at the 24-h landmark. The resulting model may identify a lower-predicted-risk group at the 24-h reassessment. These biomarker trajectories may support future risk-enriched prospective studies by helping define clinically relevant risk groups; however, treatment effects were not evaluated in the present study.
No takes yet. Share an insight, caveat, or question.
Zhang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: