Accurate prediction of invasive mechanical ventilation (IMV) requirement in patients with community-acquired pneumonia (CAP) is a key clinical challenge. Conventional static risk stratification scores have limited value for 72-hour risk assessment, and the incremental prognostic value of dynamic organ dysfunction changes for IMV prediction remains to be systematically verified. This study aimed to evaluate the predictive value of 72-hour changes in the Sequential Organ Failure Assessment score ( \(\Delta\) SOFA) for IMV requirement during hospitalization in CAP patients, and to develop a practical dynamic risk stratification tool designed for the 72-hour evaluation window. This was a retrospective cohort study based on the publicly available NACef database, which included 768 hospitalized CAP patients from a tertiary hospital in Colombia. After applying strict inclusion/exclusion criteria (hospital stay \(\ge\) 72 hours, complete admission/72-hour SOFA scores and IMV outcome data), 581 patients were included in the final analysis. We constructed a dual-dimensional ABCD dynamic risk stratification framework based on admission SOFA score (<2 vs. \(\ge\) 2) and 72-hour \(\Delta\) SOFA trajectory ( \(\le\) 0 vs. >0), stratifying patients into four subgroups: Low-Risk Stable (A), Low-Risk Deteriorating (B), High-Risk Stable (C), and High-Risk Deteriorating (D). Two logistic regression models (a static model with admission CURB-65, PSI and admission SOFA risk; a dynamic model adding \(\Delta\) SOFA trajectory) were developed to predict IMV requirement. The predictive performance of the models was comprehensively evaluated in terms of discrimination, calibration, reclassification improvement and clinical utility, with internal validation via 10-fold cross-validation. The four ABCD subgroups exhibited significantly different IMV rates: Group A (5.8%, 4/69), Group B (33.3%, 13/39), Group C (30.3%, 106/350), and Group D (81.3%, 100/123) ( \(\chi^{2}\) =136.90, \(P<0.001\) ). Compared with Group A (reference), Group D had a markedly elevated adjusted odds ratio (aOR) for IMV (91.81, 95% CI: 28.47–375.09, \(P<0.001\) ), followed by Group B (aOR: 12.60, 95% CI: 3.32–56.94, \(P<0.001\) ) and Group C (aOR: 5.78, 95% CI: 2.07–20.99, \(P<0.001\) ). The dynamic model achieved superior discriminative ability (AUC=0.850, 95% CI: 0.817–0.882) compared with the static model (AUC=0.741, 95% CI: 0.700–0.783), with a statistically significant improvement in AUC ( \(\Delta\) AUC=0.109, \(P<0.001\) ) . \(\Delta\) SOFA worsening ( \(\Delta\) SOFA>0) was independently associated with IMV (aOR: 13.77, 95% CI: 8.30–23.66, \(P<0.001\) ). The dynamic model also showed better calibration (Hosmer–Lemeshow P=0.443, Brier score=0.149 vs. static model: P=0.732, Brier score=0.195), significant reclassification improvement (NRI=0.4824, IDI=0.1919, both \(P<0.001\) ) and higher clinical utility (net benefit AUC=0.1287 vs. static model: 0.1063) across clinically relevant threshold probabilities (0–0.5). The optimal risk threshold for the dynamic model was 0.37, with a sensitivity of 65.5% and positive predictive value (PPV) of 77.2% for identifying high-IMV-risk patients. For hospitalized CAP patients with a hospital stay of \(\ge\) 72 hours, \(\Delta\) SOFA within 72 hours is an independent and valuable predictor of IMV requirement in CAP patients, with significant incremental prognostic value when added to conventional static risk scores. The ABCD dynamic risk stratification framework constructed based on admission SOFA and 72-hour \(\Delta\) SOFA can effectively stratify CAP patients into distinct risk subgroups with clear IMV risk gradients, and the corresponding dynamic prediction model has excellent predictive performance and clinical utility. This framework provides a simple, operable dynamic risk assessment tool for the 72-hour clinical node, supplementing incremental IMV risk information for routine disease reassessment and supporting the paradigm shift in CAP risk assessment from single static admission evaluation to comprehensive “admission + dynamic trajectory” assessment.
