Machine learning models show high AUC for predicting stroke-related dysphagia and aspiration risksMachine Learning Models Show Promise for Predicting Stroke Swallowing Risks

medRxiv Published August 30, 2026 Study authors: Mohammadi Yazdi, S.; Motevaselian, M.; Khatami, S.; Radfar, N.; jourahmad, z.; Perez, H. A. DOI ↗ Editorial oversight: Dr. Ji-eun Park, MD · Brain, Mind & Pain

AI-generated summary of the cited source, checked by automated accuracy review. How we work

Key Takeaway

Note that while machine learning models show high AUC for dysphagia prediction, they lack the validation for clinical use.

This meta-analysis evaluates the discrimination, validity, and readiness of machine learning and data-driven prediction models for outcomes related to post-stroke dysphagia (PSD). The analysis focused on identifying early or incident PSD, aspiration, penetration-aspiration, and severe dysphagia in adult stroke patients.

The meta-analysis reported high pooled AUC values across several outcomes. Specifically, the pooled AUC for early or incident PSD was 0.94 (95% CI 0.60-0.99; I2 = 95.6%). For aspiration or penetration-aspiration, the pooled AUC was 0.84 (95% CI 0.71-0.92). The pooled AUC for severe dysphagia was 0.89 (95% CI 0.24-1.00). An exploratory analysis of risk-prediction models yielded an AUC of 0.90 (95% CI 0.80-0.95; I2 = 90.3%; prediction interval 0.51-0.99).

Several limitations were noted, including a high risk of bias due to analysis-domain concerns. The authors noted that calibration and external validation were uncommon in the included studies. Consequently, the current evidence does not establish reliable performance in clinical care. Independent validation, calibration, complete model reporting, and clinical-impact studies are required before these models can guide post-stroke swallowing care.

How this fits prior evidence

This meta-analysis addresses a gap in the technological management of post-stroke dysphagia. While prior evidence highlights the efficacy of video game-based interventions to improve swallowing function and quality of life in post-stroke dysphagia, this study evaluates the diagnostic and predictive accuracy of machine learning models for identifying risks such as aspiration and severe dysphagia.

Researchers analyzed how machine learning and data-driven models perform at predicting complications after a stroke. These complications include dysphagia, which is difficulty swallowing, and the risk of aspiration, where food or liquid enters the lungs. The study looked at several specific outcomes, including early onset of swallowing issues and severe cases of the condition.

The analysis showed that these models had high scores for identifying several risks. Specifically, the models showed an AUC of 0.94 for early dysphagia and 0.89 for severe cases. They also showed an AUC of 0.84 for identifying aspiration or penetration-aspiration. These numbers suggest the models are good at distinguishing between different levels of risk in a research setting.

While the results are promising, there are important reasons to be cautious. The study noted a high risk of bias and a lack of external validation for many models. Because these tools have not been fully tested in everyday clinical settings, they are not yet ready to guide patient care. More research is needed to ensure these tools work reliably for patients in a hospital or clinic.

What this means for you:

Machine learning models show promise in identifying stroke-related swallowing risks but need more testing before clinical use.

Common questions

What specific risks can these machine learning models identify?

The models were tested on several outcomes related to swallowing after a stroke. They showed high accuracy for identifying early or incident dysphagia, severe dysphagia, and the risk of aspiration or penetration-aspiration. These results help researchers understand how well technology can spot potential complications for patients.

Are these models ready to be used in hospitals today?

Not yet. While the models showed high accuracy in this analysis, the study notes that they lack enough independent validation and calibration. Because of these limitations, the evidence does not yet establish that these models perform reliably enough to guide actual patient care in a clinical setting.

What are the specific accuracy scores for these models?

The models showed a pooled AUC of 0.94 for early dysphagia and 0.89 for severe dysphagia. For the risk of aspiration or penetration-aspiration, the models showed a pooled AUC of 0.84. These scores indicate how well the models could distinguish between different levels of risk.

Study Details

Study typeMeta analysis

EvidenceLevel 1

PublishedJul 2026

View Original Abstract ↓

Background: Post-stroke dysphagia (PSD) contributes to aspiration, pneumonia, malnutrition, prolonged hospitalization and mortality. We evaluated the discrimination, validity and readiness of machine learning and data-driven prediction models for PSD-related outcomes. Methods: Following a prospectively registered protocol (PROSPERO CRD420261419259), we searched PubMed/MEDLINE, Embase, Web of Science Core Collection, CINAHL and CENTRAL from inception through June 7, 2026. Eligible studies developed or validated multivariable prediction models for PSD-related outcomes in adults with stroke. We used PROBAST and PROBAST+AI to assess risk of bias and applicability and TRIPOD+AI to evaluate reporting. Area under the curve (AUC) estimates were pooled on the logit scale with random-effects models. Results: Twenty-four studies were included and ten contributed to meta-analysis. Four studies predicting early or incident PSD yielded a pooled AUC of 0.94 (95% CI 0.60-0.99; I2 = 95.6%). Pooled AUCs were 0.84 (95% CI 0.71-0.92) for aspiration or penetration-aspiration and 0.89 (95% CI 0.24-1.00) for severe dysphagia. The exploratory analysis of all ten risk-prediction models produced an AUC of 0.90 (95% CI 0.80-0.95), but heterogeneity was substantial (I2 = 90.3%) and the prediction interval was 0.51-0.99. Every study had high risk of bias because of analysis-domain concerns; calibration and external validation were uncommon. Conclusions: Reported discrimination was often high, but the evidence does not establish reliable performance in care. Independent validation, calibration, complete model reporting and clinical-impact studies are needed before these models guide post-stroke swallowing care. Keywords: Post-stroke dysphagia; Stroke; Deglutition disorders; Machine learning; Clinical prediction model; Area under the curve; Meta-analysis

Machine learning models show high AUC for predicting stroke-related dysphagia and aspiration risksMachine Learning Models Show Promise for Predicting Stroke Swallowing Risks

How this fits prior evidence

Common questions

Study Details

Acupuncture and related therapies improve Pittsburgh Sleep Quality Index scores in post-stroke insomnia

Acupuncture therapies show promise for insomnia after a stroke

Clinical research that matters. Delivered to your inbox.

Machine learning models show high AUC for predicting stroke-related dysphagia and aspiration risksMachine Learning Models Show Promise for Predicting Stroke Swallowing Risks

How this fits prior evidence

Common questions

More on Stroke

Study Details

Acupuncture and related therapies improve Pittsburgh Sleep Quality Index scores in post-stroke insomnia

Acupuncture therapies show promise for insomnia after a stroke

Clinical research that matters. Delivered to your inbox.

Related in Neurology

From Other Specialties