Innovative nomogram for predictive risk stratification of aspiration pneumonia in post-stroke dysphagia patients

Abstract

Background: Post-stroke dysphagia (PSD) affects up to 76% of stroke patients and increases aspiration pneumonia (AP) risk, leading to higher mortality among older survivors. Current risk assessment tools for AP in PSD patients lack precision.

Methods: We conducted a retrospective study of 7,134 stroke patients admitted to Jinshan Hospital from 2019 to 2023. We used multivariable logistic regression to identify AP predictors and constructed a nomogram model using these predictors. Model performance was evaluated using bootstrap resampling, calibration, and decision curve analysis. Internal validation was conducted on 30% of cases, and external validation was performed on 500 PSD patients from community health centers.

Results: Among 2,663 PSD patients, 578 (21.7%) developed AP. Independent predictors included age, stroke severity, hyperlipidemia, hyperhomocysteinemia, heart failure, CRP, WBC, neutrophil ratio, Hb, FBG, prealbumin, BNP, and serum sodium. The nomogram model showed excellent discrimination (C-index: 0.885) and good agreement between predicted and observed AP probabilities. It provided net benefit across various threshold probabilities.

Conclusion: Our study developed the first dedicated nomogram for AP risk prediction in PSD patients, incorporating novel predictor combinations and demonstrating robust validation across multi-center cohorts. This fills an important clinical need under community conditions by enabling early identification of high-risk PSD patients using routinely available clinical variables.

Keywords: aspiration pneumonia; nomogram model; post-stroke dysphagia; risk prediction; stroke outcomes.

Figure 1

Flow chart for collecting patients' clinical case data.

Figure 2

Forest plot of OR values for various influencing factors of aspiration pneumonia in stroke patients. OR, odds ratio; NIHSS score, national institute of health stroke scale score.

Figure 3

The distribution of 13 independent influencing factors in the AP and Non-AP groups of the overall cohort. AP, aspiration pneumonia; NIHSS score, national institute of health stroke scale score; CRP, c-reactive protein; WBC, white blood cells; NE, neutrophil; Hb, hemoglobin; FBG, fasting blood glucose; PA, prealbumin; BNP, brain natriuretic peptide. *P < 0.05; **P < 0.01; ***P < 0.001.

Figure 4

(A) The tornado charts display the sensitivity analysis of all variables, ranked by the absolute values of their standardized beta (std b) coefficients (horizontal axis). Variables are ordered vertically by their contribution magnitude (|std b coefficients|), with longer bars indicating stronger associations with the outcome (aspiration pneumonia); (B) A nomogram prediction model for the risk of AP in PSD patients. Example: to predict the occurance risk of aspiration pneumonia for a male stroke patient of 40 with only hyperhomocysteinemia history. His NIHSS score is 3 and his laboratory test results are as follow: CRP: 69.49 mg/L, WBC: 12.2 × 10⁹/L, NE%: 49.9%, Hb: 133 g/L, FBG: 6.52 mmol/L, PA: 311 mg/L, BNP: 145.78 ng/L, Na⁺: 144 mmol/L. (1) Locate each variable's value on the corresponding axis and draw a vertical line to the 'Points' axis (Age: 0, NIHSS score: 0, Hyperlipidemia: 37.5, Hyperhomocysteinemia: 90, Cardiac insufficiency: 0, CRP: 95, WBC: 70, NE%: 0, Hb: 0, FBG: 0, PA: 0, BNP: 20, Na⁺: 0); (2) Sum all points (37.5 + 90 + 95 + 70 + 20 = 312.5); (3) Locate the total points on the “Total Points” axis and draw a vertical line to the bottom axis to read the predicted probability (312.5 total points ≈ 27% risk). std b, standardized beta; NIHSS score, national institute of health stroke scale score; CRP, c-reactive protein; NE, neutrophil; WBC, white blood cells; PA, prealbumin; Hb, hemoglobin; BNP, brain natriuretic peptide; Hct, hematocrit; Alb, albumin; FBG, fasting blood glucose; Plt, platelet; Scr, serum creatinine; TP, total protein; ALT, alanine aminotransferase; GFR, glomerular filtration rate; LDL, low-density lipoprotein; TG, triglyceride; TC, total cholesterol.

Figure 5

The calibration curve (A), clinical impact curve (“Number high risk” = total number of patients identified as high-risk by nomogram model; “Number high risk with event” = subset of high-risk patients who actually developed AP) (B), clinical decision curve [Models compared: nomogram model (age, NIHSS score, hyperlipidemia, hyperhomocysteinemia, cardiac insufficiency, CRP, WBC, NE%, Hb, FBG, PA, BNP, and Na⁺), AIS-APS score (age, atrial fibrillation, cardiac insufficiency, COPD, smoke, stoke history, Glasgow score, dysphagia, stroke type and FGB), and All model (assuming all patients receive treatment).] (C), and ROC curve (D) in the training queue, internal validation queue, and overall cohort of nomogram. AP, aspiration pneumonia; AIS-APS score, acute ischemic stroke-associated pneumonia score.

Figure 6

The calibration curve (A), clinical impact curve (B), clinical decision curve (C), and ROC curve (D) in the external validation cohort of the nomogram. The basic process of constructing and validating the nomogram prediction model (E). AP, aspiration pneumonia; AIS-APS score, acute ischemic stroke-associated pneumonia score.