What is the overall accuracy of models that predict dysphagia risk in older adults?

Several predictive models have been developed to estimate dysphagia risk in older adults, but their accuracy varies widely. A systematic review of 17 studies reported that the area under the curve (AUC) — a measure of how well a model distinguishes between those who will and will not develop dysphagia — ranged from 0.682 to 0.926, with a pooled average of 0.82 ⁶. However, most models have methodological limitations, and only two studies performed external validation, meaning the models may not perform as well in new patient groups ⁶. Other models specific to post-stroke dysphagia have shown AUCs from 0.756 to 0.955, but all had a high risk of bias ².

What the research says

A 2025 systematic review of predictive models for dysphagia in older adults found that the pooled AUC from a meta-analysis of 17 studies was 0.82 (95% CI: 0.77–0.88), indicating good overall discrimination ⁶. However, all included studies had a high risk of bias, mainly due to retrospective designs, small sample sizes, and lack of external validation ⁶. Common predictors included advanced age, smoking history, reduced self-care ability, polypharmacy, frailty, malnutrition, cognitive impairment, and poor oral health ⁶.

For post-stroke dysphagia specifically, a separate systematic review of 18 model development studies reported AUCs ranging from 0.756 to 0.955, with 16 studies showing good applicability ². Again, all studies had a high risk of bias, and only a few performed external validation ². Key predictors included age, NIHSS score, Kubota Water Swallowing Test, history of aspiration, and Glasgow Coma Scale score ².

Individual studies have also developed models with promising accuracy. For example, a nomogram for predicting aspiration in post-stroke dysphagia patients had an AUC of 0.834 in the training cohort and 0.882 in an external validation cohort ⁹. Another model for post-stroke dysphagia achieved an AUC of 0.915 in the development cohort ¹⁰. A model for dysphagia in older hospitalized Chinese patients had high predictive accuracy, though the exact AUC was not reported in the abstract ⁸.

Despite these promising numbers, the overall evidence is limited by methodological flaws. Most models have not been tested in diverse populations or real-world clinical settings, so their generalizability is uncertain ⁶. The FEES Dysphagia Index (FDI), while not a risk prediction model, showed good accuracy for predicting outcomes like hospital-acquired pneumonia (AUC 0.70) and mortality (AUC 0.71) in neurological patients ¹.

What to ask your doctor

• What screening tools or risk prediction models do you use to assess dysphagia risk in older adults?
• How accurate are these models for patients like me or my family member?
• Are there any validated models that have been tested in a similar population (e.g., age, comorbidities)?
• What are the key risk factors for dysphagia that I should be aware of?
• If a model suggests low risk, should we still monitor for swallowing difficulties?