A total of 275 patients fulfilled the inclusion criteria and were included in the study (Fig. S1 in the Supplement summarizes details of study recruitment). A total of 73 (26.55%) of the patients had SAP. The demographic and clinical characteristics of the study population and the differences between SAP and non-SAP patients are shown in Table 1. The groups significantly differed in age, and SAP patients were older than the non-SAP patients (70.74 ± 11.89 vs. 63.26 ± 12.80 years, t = -4.36, P = 1.84×10− 5). SAP was associated with a higher chance of post stroke dysphagia (86.30% vs. 24.75%, χ2 = 83.92, P = 5.15×10− 20) and nasogastric tube (NGT) placement (72.60% vs. 10.40%, χ2 = 105.50, P = 9.49×10− 25). Patients with a history of atrial fibrillation (AF) (31.51% vs. 8.42%, χ2 = 23.00, P = 1.62×10− 6), AF detected by Electrocardiography (ECG) (38.36% vs. 10.40%, χ2 = 28.63, P = 8.76×10− 8), history of stroke (26.03% vs. 12.38%, χ2 = 7.44, P = 6.38×10− 3), and higher fasting blood glucose (FBG) (6.98 ± 3.08 vs. 5.61 ± 1.84 µmol/l, t = -3.57, P = 4.22×10− 4) were more likely to suffer from dysphagia. SAP patients showed higher scores in both admission NIHSS (11.99 ± 6.66 vs. 4.25 ± 4.53, t = -9.19, P = 1.03×10− 17) and discharge NIHSS (9.63 ± 6.43 vs. 3.07 ± 3.47, t = -8.29, P = 5.19×10− 15). Patients with progressive stroke had a higher chance of exhibiting SAP (27.40% vs. 5.94%, χ2 = 24.01, P = 9.58×10− 7. A strong association was found between SAP and stroke using TOAST classifications (χ2 = 49.51, P = 4.57×10− 10). We investigated the associations of SAP and stroke using the TOAST classification as shown in Fig. S2. Patients with LAA and CE suffered SAP more often (LAA: OR = 2.20, P = 4.40×10− 3; CE: OR = 4.61, P = 7.39×10− 7), and patients with SVO suffered less often (OR = 0.15, P = 1.99×10− 7). Patients of SOE and SUE did not show significant differences. Patients with or without SAP showed significant associations between distinct lesion locations (z = 2.88, P = 3.60×10− 3). The associations between SAP and lesion locations are shown in Fig. S3. Patients with right cerebral hemisphere lesions suffered SAP more often (OR = 2.46, P = 1.22×10− 3), and patients with left cerebral hemisphere lesions suffered less often (OR = 0.48, P = 1.54×10− 2).
Table 1
Demographic and Clinical Data of Patients with Stroke Associated Pneumonia and Controls
Demographic and Clinical Data | Controls (n = 202) | Pneumonia (n = 73) | t/χ2/z | P value |
Age (years) | 63.26 ± 12.80 | 70.74 ± 11.89 | t = -4.36 | 1.84×10− 5* |
Gender female/male | 134/68 | 48/25 | χ2 = 0.01 | 0.93 |
Post stroke dysphagia yes/no | 50/152 | 63/10 | χ2 = 83.92 | 5.15×10− 20** |
Nasogastric tube placement yes/no | 21/181 | 53/20 | χ2 = 105.50 | 9.49×10− 25** |
Systolic Blood Pressure (mmHg) | 144.30 ± 22.05 | 142.50 ± 18.49 | t = 0.63 | 0.53 |
Diastolic Blood Pressure (mmHg) | 81.27 ± 13.20 | 79.88 ± 12.30 | t = 0.78 | 0.43 |
History of Hypertension yes/no | 143/59 | 58/15 | χ2 = 2.04 | 0.15 |
History of Diabetes yes/no | 39/163 | 18/55 | χ2 = 0.93 | 0.33 |
Smoking yes/quit/no | 52/3/147 | 17/5/51 | χ2 = 5.50 | 0.06 |
History of AF† yes/no | 17/185 | 23/50 | χ2 = 23.00 | 1.62×10− 6* |
Other Heart Diseases yes/no | 7/195 | 4/69 | χ2 = 0.16 | 0.69 |
Previous Stroke yes/no | 25/177 | 19/54 | χ2 = 7.44 | 6.38×10− 3* |
Triglyceride (mmol/L) | 1.44 ± 0.89 | 1.23 ± 0.51 | t = 2.44 | 0.02 |
Total Cholesterol (mmol/L) | 4.27 ± 1.06 | 4.09 ± 1.02 | t = 1.24 | 0.22 |
LDLC (mmol/L) | 2.43 ± 0.83 | 2.32 ± 0.77 | t = 0.96 | 0.34 |
Creatinine (µmol/L) | 69.30 ± 18.31 | 70.84 ± 18.58 | t = -0.61 | 0.54 |
Uric Acid (µmol/L) | 294.10 ± 84.21 | 289.30 ± 110.20 | t = 0.34 | 0.73 |
Fasting Blood Glucose (µmol/L) | 5.61 ± 1.84 | 6.98 ± 3.08 | t = -3.57 | 4.22×10− 4* |
Homocysteine (µmol/L) | 12.30 ± 9.25 | 11.23 ± 3.16 | t = 1.43 | 0.15 |
Hemoglobin A1c (%) | 6.76 ± 1.46 | 6.89 ± 1.45 | t = -0.68 | 0.50 |
ECG† (AF) yes/no | 21/181 | 28/45 | χ2 = 28.63 | 8.76×10− 8** |
Admission NIH Stroke Scale | 4.25 ± 4.53 | 11.99 ± 6.66 | t = -9.19 | 1.03×10− 17** |
Discharge NIH Stroke Scale | 3.07 ± 3.47 | 9.63 ± 6.43 | t = -8.29 | 5.19×10− 15** |
TOAST Classification† | 62/24/91/11/14 | 36/28/8/1/0 | χ2 = 49.51 | 4.57×10− 10** |
Progressive Stroke yes/no | 12/192 | 20/53 | χ2 = 24.01 | 9.58×10− 7** |
Brain lesion location† | 85/55/22/35/0/3/2 | 19/35/9/6/1/1/2 | z = 2.88 | 3.60×10− 3* |
Special Treatment Thrombolytic/ endovascular/no | 25/12/165 | 12/3/58 | z = 0.74 | 0.46 |
Epilepsy yes/no | 5/197 | 2/71 | χ2 < 0.01 | 1.00 |
Progessive stroke yes/no | 22/180 | 6/67 | χ2 = 0.42 | 0.52 |
Length of hospital stay | 11.62 ± 6.54 | 10.85 ± 5.01 | t = 1.03 | 0.30 |
* Continuous data are shown as mean ± SD, values in patients with stroke associated pneumonia and controls with statistical significance based on two sample T-test. Categorical data differences in patients and controls are represented with statistical significance based on Chi-square test (χ2 & P) or Fisher’s exact test (z & P). *: P < 0.005, **: P < 1.00×10− 6. |
† AF refers to atrial fibrillation, LDLC refers to low density lipoprotein cholesterol, ECG refers to Electrocardiogram. TOAST refers to five classifications: (1) large-artery atherosclerosis, (2) cardioembolism, (3) small-vessel occlusion, (4) stroke of other determined etiology and (5) stroke of undetermined etiology. Brain lesion locations refer to seven classifications: (1) left cerebral hemisphere lesion, (2) right cerebral hemisphere lesion, (3) bilateral cerebral hemisphere lesions, (4) brainstem lesion, (5) left cerebral hemisphere and brainstem lesions, (6) right cerebral hemisphere and brainstem lesions, (7) bilateral cerebral hemisphere and brainstem lesions. |
Risk factors and prediction scale of SAP
Variables with significant associations were combined with gender and used to construct the SAP prediction scale using a multivariate logistic model. As shown in Table 2, age (OR = 1.05, χ2 = 4.00, P = 4.55×10− 2), NGT (OR = 4.03, χ2 = 5.87, P = 0.02) and right cerebral hemisphere lesions (OR = 3.77, χ2 = 6.75, P = 0.01) showed the most significant associations with SAP in multivariate statistical analysis, and the other variables, including gender, swallowing function, history of AF, history of stroke, FBG, NIHSS score, TOAST classifications and progressive stroke, showed no association.
Table 2
Multivariable Logistic Regression Model for Predicting Patients with Pneumonia
Variables | Odds Ratio | 95% CI | χ2 | P value |
Age | 1.05 | 1.00, 1.10 | 4.00 | 4.55×10− 2* |
Gender | 2.34 | 0.83, 6.60 | 2.58 | 0.12 |
Post stroke dysphagia yes/no | 2.29 | 0.71, 7.39 | 1.91 | 0.17 |
Nasogastric tube placement yes/no | 4.03 | 1.30, 12.43 | 5.87 | 0.02* |
History of AF† yes/no | 1.78 | 0.35, 9.13 | 0.48 | 0.49 |
History of Stroke yes/no | 1.94 | 0.71, 5.31 | 1.66 | 0.20 |
Fasting Blood Glucose (µmol/L) | 1.12 | 0.93, 1.35 | 1.39 | 0.24 |
ECG† (AF) yes/no | 5.22 | 0.20, 137.44 | 0.98 | 0.32 |
Admission NIH Stroke Scale | 1.06 | 0.93, 1.22 | 0.78 | 0.38 |
Discharge NIH Stroke Scale | 1.11 | 0.92, 1.33 | 1.20 | 0.27 |
Large-artery Atherosclerosis | 1.00 | | | |
Cardioembolism | 0.23 | 0.01, 6.09 | 0.78 | 0.38 |
Small-vessel Occlusion | 0.57 | 0.18, 1.78 | 0.94 | 0.33 |
stroke of other determined etiology | 1.88 | 0.15, 23.72 | 0.24 | 0.63 |
stroke of undetermined etiology | <0.01 | <0.01, >999.99 | < 0.01 | 0.97 |
Progressive Stroke | 2.10 | 0.64, 6.92 | 1.48 | 0.22 |
Lesioned hemi left | 1.00 | | | |
Lesioned hemi right | 3.77 | 1.39, 10.27 | 6.75 | 0.01* |
Lesioned hemi both | 3.51 | 0.75, 16.36 | 2.56 | 0.11 |
Lesioned brainstem | 1.19 | 0.25, 5.60 | 0.05 | 0.83 |
Lesioned hemi left and brainstem | >999.99 | <0.01, >999.99 | < 0.01 | 0.99 |
Lesioned hemi right and brainstem | 2.14 | 0.02, 209.00 | 0.11 | 0.74 |
Lesioned hemi both and brainstem | 2.20 | 0.19, 25.10 | 0.40 | 0.53 |
*: P < 0.05. |
A ROC analysis was performed to examine the accuracy of the SAP prediction scale. As shown in Fig. 1, age, nasal feeding diet and right cerebral hemisphere lesions together showed a significantly high AUC (area under the ROC curve) of 0.93, with P < 0.05. Age and gender effects were included in the multivariate logistic model to construct prediction scales.
The calculated Youden index of AIS-APS score prediction system was 0.32. According to the ROC analysis of our SAP prediction model established in this study, the best critical point was obtained, and the calculated Youden index was 0.77, which was better than that of AIS-APS score prediction system. The results of McNemar's test showed that the sensitivity of our SAP prediction model was higher (χ2 = 39.00, P < 0.05), and the specificity of AIS-APS score was higher (χ2 = 15.21, P < 0.05).
Based on the 13 variables in our SAP prediction model, statistical analysis was performed on patients with 2 or more risk factors at the same time. The 50 most common combinations are shown and were compared between SAP patients and non-SAP patients in Fig. 2. Continuous variables were divided into 2 categories. Age was divided into two groups according to the median: ≤ 67 years and > 67 years. FBG was divided into ≤ 6.1 mmol/l and > 6.1 mmol/l. Admission NIHSS and discharge NIHSS were divided into ≤ 3 and > 3 score. To avoid repetition, only LAA type was retained in the TOAST classification. Lesions were divided into two variables: left and right brain lesions, and patients with both left and right brain lesions were regarded as having bilateral lesions. Of the 50 combinations, the distribution of 44 combinations was significantly different between SAP patients and non-SAP patients (P < 0.05). After adjustment for other risk factors that were not included in the combinations, 24 combinations remained statistically significant (P < 0.05). OR value of each combination was calculated, and 5 combinations with the highest OR value were abnormal swallowing function, NGT, admission NIHSS > 3 score and discharge NIHSS > 3 score.
Once the predictive variables were determined, SHAP visualization of the selected predictors was modelled for all SAP patients as shown in Fig. 3.
Risk factors and prediction scale of SAP severity
Among 73 SAP patients, the univariate analysis of continuous variables for CURB-65 is shown in Table S1. Age (r = 0.31, P = 0.01), FBG (r = 0.35, P = 2.66×10− 3), admission NIHSS (r = 0.43, P = 1.33×10− 4) and discharge NIHSS (r = 0.47, P = 2.74×10− 5) showed the most significant associations with CURB-65. The univariate analysis of categorical data for CURB-65 is shown in Table S2. CURB-65 was associated with post stroke dysphagia (t = -3.67, P = 4.66×10− 4) and NGT placement (t = -4.09, P = 1.12×10− 4). Patients with a history of AF (t = -3.10, P = 2.77×10− 3), AF detected by ECG (t = -3.35, P = 1.30×10− 3) and progressive stroke (t = -2.10, P = 0.04) were more likely to suffer from higher CURB-65. A strong association was found between the CURB-65 and TOAST classifications (F = 5.55, P = 1.80×10− 3).
Variables with significant associations were combined with gender and were used to construct the SAP severity prediction scale measured by CURB-65 using a multivariate logistic model. As shown in Table S3, age (t = 3.25, P = 1.90×10− 3), swallowing function (t = 2.44, P = 0.02), FBG (t = 2.31, P = 0.02) and admission NIHSS (t = 2.41, P = 0.02) showed the most significant associations with SAP severity in multivariate statistical analysis by CURB-65, and the other variables, including gender, NGT placement, history of AF, AF detected by ECG, discharge NIHSS score, TOAST classifications and progressive stroke, showed no association. The relationship between the predicted value and the actual value of the CURB-65 score is shown in Fig. 4 (A). Pearson correlation analysis was performed and the results showed a significant correlation (r = 0.75, P < 0.05).
Among 73 SAP patients, univariate analysis of continuous variables for PSI is shown in Table S4. Age (r = 0.41, P = 2.79×10− 4) and discharge NIHSS (r = 0.36, P = 1.92×10− 3) showed the most significant associations with PSI. The univariate analysis of categorical data for PSI is shown in Table S5. The PSI was associated with NGT placement (t = -2.64, P = 0.01), history of hypertension (t = -3.64, P = 5.14×10− 4) and progressive stroke (t = -2.04, P = 4.51×10− 2).
Variables with significant associations were combined with gender and used to construct the SAP severity prediction scale measured by the PSI using a multivariate logistic model. As shown in Table S6, age (t = 3.15, P = 2.40×10− 3), history of hypertension (t = 2.62, P = 0.01) and discharge NIHSS (t = 2.14, P = 0.04) showed the most significant associations with SAP severity in the multivariate statistical analysis by PSI, and the other variables showed no association. The relationship between the predicted value and the actual value of the PSI score is shown in Fig. 4 (B). Pearson correlation analysis was performed and the results showed a significant correlation (r = 0.62, P < 0.05).
Prediction of poor outcomes in SAP patients
One of the 73 SAP patients had a missing outcome and 3 of them had an adverse outcome (transfer to the intensive care unit or death). The univariate analysis is shown in Table S7 in Supplemental Material. FBG (t = 4.96, P = 4.64×10− 6) and thrombolytic/endovascular treatment, (z = 2.04, P = 0.04) were significantly associated with poor outcomes. The multivariate analysis result is shown in Table S8 in Supplemental Material, AUC = 0.92, with P < 0.05 (Fig. 4(B)).