Bold claim: a new algorithm predicts post-stroke pneumonia with 93% accuracy, highlighting tracheostomy status, aspiration, cough frequency, malnutrition, and cognitive impairment as key risk factors.
Overview
A forward-looking study tested a novel predictive tool for pneumonia after stroke. The researchers demonstrated 93% accuracy in forecasting pneumonia risk in a cohort of more than 300 stroke patients treated at a tertiary hospital in South Korea between 2019 and 2024. All participants had confirmed ischemic or hemorrhagic stroke and signs of swallowing difficulties.
What was done
Within 24 hours of admission, each patient underwent a battery of assessments: videofluoroscopic swallowing study (VFSS) to evaluate swallowing safety, a modified cough reflex test, cognitive assessment using the Mini-Mental State Examination (MMSE), and serum albumin measurements to gauge nutritional status. Patients were monitored for four weeks. Pneumonia diagnosis followed Mann criteria and risk was categorized as none (<2%), low (2–25%), or high (>25%). The study aimed to identify predictors of pneumonia among stroke patients.
Key findings
Overall, pneumonia occurred in 8.5% of patients within four weeks. Those who developed pneumonia had higher rates of:
- tracheostomy (54% vs 11%),
- VFSS-confirmed aspiration (71% vs 23.5%), and
- bilateral hemispheric brain lesions (64% vs 23%).
They also showed lower cognitive scores and poorer nutritional status (mean MMSE 8.2 vs 18.4; serum albumin 3.6 vs 4.0 g/dL; all comparisons P < .01).
Predictors of pneumonia included:
- tracheostomy status (odds ratio [OR] 9.3),
- VFSS-confirmed aspiration (OR 8.2),
- bilateral stroke lesions (OR 5.9). Additional associations were found with MMSE scores, cough frequency, and albumin levels (P < .005 for all).
Clinical implications
Among patients who had a tracheostomy, 31% developed pneumonia compared with 5% who did not. In those with a tracheostomy who also had a cough frequency under three, risk was particularly high, especially if MMSE was below 6 (50% vs 19% if MMSE ≥ 6). For patients without tracheostomy, pneumonia risk rose to 62.5% when VFSS-confirmed aspiration occurred alongside serum albumin below 3.5 g/dL.
Algorithm performance
The predictive model achieved 93% accuracy with an area under the receiver operating characteristic curve (AUC) of 0.89. In patients deemed no-risk, overall accuracy reached 99%.
Practical takeaway
The authors describe the tool as providing a comprehensive framework for screening post-stroke pneumonia and enabling early preventive measures for those at elevated risk. They also caution that broader clinical adoption requires validation in larger, more diverse populations, including higher-risk groups, and external multicenter testing.
Study details
The research was led by Jong Weon Lee of Yonsei University College of Medicine, Seoul, South Korea, and was published online on November 19 in Frontiers in Neurology.
Limitations and caveats
Limitations include reliance on VFSS for detecting aspiration, which may be limited in availability or feasibility in some settings. The relatively small sample and the exclusion of higher-risk patients (e.g., those needing supplemental oxygen or with active pneumonia) limit generalizability. Dysphagia assessment did not employ standardized scales, potentially affecting reproducibility. The study did not include a head-to-head comparison with existing predictive models, and it did not assess certain potential predictors such as oral hygiene due to tool availability.
Disclosures and process
The investigators reported no relevant conflicts of interest. The article notes that editorial tools, including AI, were used in its preparation, with human editors reviewing the final manuscript.
Discussion prompt
Given these findings, how would you balance the benefits of early pneumonia prediction with the practical constraints of implementing VFSS-based assessments in diverse clinical settings? Do you think the emphasis on tracheostomy as a strong predictor could influence decision-making about airway management after stroke? Would you like a version that highlights practical steps for hospitals to start integrating such a tool with existing workflows?
