Using Deep Transfer Learning to Detect Hyperkalemia From Ambulatory Electrocardiogram Monitors in Intensive Care Units: Personalized Medicine Approach.

BACKGROUND: Hyperkalemia is a critical condition, especially in intensive care units. So far, there have been no accurate and noninvasive methods for recognizing hyperkalemia events on ambulatory electrocardiogram monitors. OBJECTIVE: This study aimed to improve the accuracy of hyperkalemia predictions from ambulatory electrocardiogram (ECG) monitors using a personalized transfer learning method; this would be done by training a generic model and refining it with personal data. METHODS: This retrospective cohort study used open source data from the Waveform Database Matched Subset of the Medical Information Mart From Intensive Care III (MIMIC-III). We included patients with multiple serum potassium test results and matched ECG data from the MIMIC-III database. A 1D convolutional neural network-based deep learning model was first developed to predict hyperkalemia in a generic population. Once the model achieved a state-of-the-art performance, it was used in an active transfer learning process to perform patient-adaptive heartbeat classification tasks. RESULTS: The results show that by acquiring data from each new patient, the personalized model can improve the accuracy of hyperkalemia detection significantly, from an average of 0.604 (SD 0.211) to 0.980 (SD 0.078), when compared with the generic model. Moreover, the area under the receiver operating characteristic curve level improved from 0.729 (SD 0.240) to 0.945 (SD 0.094). CONCLUSIONS: By using the deep transfer learning method, we were able to build a clinical standard model for hyperkalemia detection using ambulatory ECG monitors. These findings could potentially be extended to applications that continuously monitor one's ECGs for early alerts of hyperkalemia and help avoid unnecessary blood tests.

C-Reactive Protein Concentration Can Help to Identify Bacteremia in Children Visiting the Emergency Department: A Single Medical Center Experience.

BACKGROUND: For febrile children who are evaluated in a pediatric emergency department (PED), blood culture can be considered the laboratory criterion standard to detect bacteremia. However, high rates of negative, false-positive, or contaminated blood cultures in children often result in this testing being noncontributory. This study determined the factors associated with true-positive blood cultures in children. METHODS: This retrospective study was conducted at a tertiary medical center's PED. The blood culture use reports were prepared by an infectious disease specialist and were classified as bacteremia, nonbacteremia, and contamination. RESULTS: We registered a total of 239,459 PED visits during the 8-year period, and 21,841 blood culture samples were taken. Of the laboratory test studies, higher C-reactive protein (CRP) levels and lower hemoglobin levels were observed in the bacteremia group compared with other groups (all P < 0.001). The cut-off value calculated for each age group was adjusted for better clinical usage and significantly improved the blood culture clinical utility documented in the following age groups: 0 to 1 years (CRP level = 30 mg/L, odds ratio [OR] = 5.4, P < 0.001), 1 to 3 years (CRP level = 45 mg/L, OR = 3.7, P < 0.001), and 12 to 18 years (CRP level = 50 mg/L, OR = 6.3, P = 0.006). Using the CRP cut-off value established in this study, we could reduce the blood culture samples in the PED by 14,108 (64.6%). CONCLUSIONS: This study provides new evidence that CRP may be a useful indicator for blood culture sampling in certain age groups and may help improve the efficiency of blood culture in the PED.

ICU admission following an unscheduled return visit to the pediatric emergency department within 72 hours.

INTRODUCTION: The purpose of this study was to describe the demographic characteristics and prognosis of children admitted to the intensive care unit (ICU) after a pediatric emergency department (PED) return visit within 72 h. METHOD: We conducted this retrospective study from 2010 to 2016 in the PED of a tertiary medical center in Taiwan and included patients under the age of 18 years old admitted to the ICU after a PED return visit within 72 h. Clinical characteristics were collected to perform demographic analysis. Pediatric patients who were admitted to the ICU on an initial visit were also enrolled as a comparison group for outcome analysis, including mortality, ventilator use, and length of hospital stay. RESULTS: We included a total of 136 patients in this study. Their mean age was 3.3 years old, 65.4% were male, and 36.0% had Chronic Health Condition (CHC). Disease-related return (73.5%) was by far the most common reason for return. Compared to those admitted on an initial PED visit, clinical characteristics, including vital signs at triage and laboratory tests on return visit with ICU admission, demonstrated no significant differences. Regarding prognosis, ICU admission on return visit has a higher likelihood of ventilator use (aOR:2.117, 95%CI 1.021~4.387), but was not associated with increased mortality (aOR:0.658, 95%CI 0.150~2.882) or LOHS (OR:-1.853, 95%CI -4.045~0.339). CONCLUSION: Patients who were admitted to the ICU on return PED visits were associated with an increased risk of ventilator use but not mortality or LOHS compared to those admitted on an initial visit.

The influence of crowding on clinical practice in the emergency department.

BACKGROUND: This study aimed to clarify the association between the crowding and clinical practice in the emergency department (ED). METHODS: This 1-year retrospective cohort study conducted in two EDs in Taiwan included 70,222 adult non-trauma visits during the day shift between July 1, 2011, and June 30, 2012. The ED occupancy status, determined by the number of patients staying during their time of visit, was used to measure crowding, grouped into four quartiles, and analyzed in reference to the clinical practice. The clinical practices included decision-making time, patient length of stay, patient disposition, and use of laboratory examinations and computed tomography (CT). RESULT: The four quartiles of occupancy statuses determined by the number of patients staying during their time of visit were <24, 24-39, 39-62, and >62. Comparing >62 and <24 ED occupancy statuses, the physicians' decision-making time and patients' length of stay increased by 0.3h and 1.1h, respectively. The percentage of patients discharged from the ED decreased by 15.5% as the ED observation, general ward, and intensive care unit admissions increased by 10.9%, 4%, and 0.7%, respectively. CT and laboratory examination slightly increased in the fourth quartile of ED occupancy. CONCLUSION: Overcrowding in the ED might increase physicians' decision-making time and patients' length of stay, and more patients could be admitted to observation units or an inpatient department. The use of CT and laboratory examinations would also increase. All of these could lead more patients to stay in the ED.

The impact of prolonged waiting time for coronary care unit admission on patients with non ST-elevation acute coronary syndrome.

BACKGROUND: The boarding of patients in the emergency department consumes nursing and physician resources, and may delay the evaluation of new patients. It may also contribute to poor cardiovascular outcomes in patients with acute coronary syndrome (ACS). This study analyzed the relationship between the delay in coronary care unit (CCU) admission and the clinical outcomes of patients with ACS with non-ST-segment elevation (NSTE-ACS). METHODS: Patients were divided into 2 groups according to the CCU waiting time (<12h and >12h). Outcome variables including in-hospital mortality, gastrointestinal bleeding and stroke during hospitalization, and duration of hospital stay were compared between the 2 study groups. We used the GRACE risk scores to classify disease severity of the study patients for stratifying analysis. RESULT: A difference was found in the outcome of gastrointestinal bleeding. Among those with GRACE risk scores of <3 (low mortality risk) and 3 (high mortality risk), 5% and 3.1% of patients developed gastrointestinal bleeding, respectively, with CCU waiting time of >12h compared to CCU waiting time of <12h. However, there was no significant statistical difference (P=0.065 and 0.547). In addition, there were no significant differences in the in-hospital mortality rate, incidence of stoke, and duration of hospital stay between the 2 groups. CONCLUSION: There was no significant difference in the clinical outcomes of NSTE-ACS patients without profound shock between those with CCU waiting times of <12 and >12h. If necessary, CCU admission should be prioritized for patients whose hemodynamic instability or respiratory failure.

Machine learning-based prediction of coronary care unit readmission: A multihospital validation study.

Objective: Readmission to the coronary care unit (CCU) has significant implications for patient outcomes and healthcare expenditure, emphasizing the urgency to accurately identify patients at high readmission risk. This study aims to construct and externally validate a predictive model for CCU readmission using machine learning (ML) algorithms across multiple hospitals. Methods: Patient information, including demographics, medical history, and laboratory test results were collected from electronic health record system and contributed to a total of 40 features. Five ML models: logistic regression, random forest, support vector machine, gradient boosting, and multilayer perceptron were employed to estimate the readmission risk. Results: The gradient boosting model was selected demonstrated superior performance with an area under the receiver operating characteristic curve (AUC) of 0.887 in the internal validation set. Further external validation in hold-out test set and three other medical centers upheld the model's robustness with consistent high AUCs, ranging from 0.852 to 0.879. Conclusion: The results endorse the integration of ML algorithms in healthcare to enhance patient risk stratification, potentially optimizing clinical interventions, and diminishing the burden of CCU readmissions.

Explainable Deep Learning Model for Predicting Serious Adverse Events in Hospitalized Geriatric Patients Within 72 Hours.

Background: The global aging population presents a significant challenge, with older adults experiencing declining physical and cognitive abilities and increased vulnerability to chronic diseases and adverse health outcomes. This study aims to develop an interpretable deep learning (DL) model to predict adverse events in geriatric patients within 72 hours of hospitalization. Methods: The study used retrospective data (2017-2020) from a major medical center in Taiwan. It included non-trauma geriatric patients who visited the emergency department and were admitted to the general ward. Data preprocessing involved collecting prognostic factors like vital signs, lab results, medical history, and clinical management. A deep feedforward neural network was developed, and performance was evaluated using accuracy, sensitivity, specificity, positive predictive value (PPV), and area under the receiver operating characteristic curve (AUC). Model interpretation utilized the Shapley Additive Explanation (SHAP) technique. Results: The analysis included 127,268 patients, with 2.6% experiencing imminent intensive care unit transfer, respiratory failure, or death during hospitalization. The DL model achieved AUCs of 0.86 and 0.84 in the validation and test sets, respectively, outperforming the Sequential Organ Failure Assessment (SOFA) score. Sensitivity and specificity values ranged from 0.79 to 0.81. The SHAP technique provided insights into feature importance and interactions. Conclusion: The developed DL model demonstrated high accuracy in predicting serious adverse events in geriatric patients within 72 hours of hospitalization. It outperformed the SOFA score and provided valuable insights into the model's decision-making process.

Prospective clinical evaluation of deep learning for ultrasonographic screening of abdominal aortic aneurysms.

Abdominal aortic aneurysm (AAA) often remains undetected until rupture due to limited access to diagnostic ultrasound. This trial evaluated a deep learning (DL) algorithm to guide AAA screening by novice nurses with no prior ultrasonography experience. Ten nurses performed 15 scans each on patients over 65, assisted by a DL object detection algorithm, and compared against physician-performed scans. Ultrasound scan quality, assessed by three blinded expert physicians, was the primary outcome. Among 184 patients, DL-guided novices achieved adequate scan quality in 87.5% of cases, comparable to the 91.3% by physicians (p = 0.310). The DL model predicted AAA with an AUC of 0.975, 100% sensitivity, and 97.8% specificity, with a mean absolute error of 2.8 mm in predicting aortic width compared to physicians. This study demonstrates that DL-guided POCUS has the potential to democratize AAA screening, offering performance comparable to experienced physicians and improving early detection.

Artificial Intelligence Application in Skull Bone Fracture with Segmentation Approach.

This study aims to evaluate an AI model designed to automatically classify skull fractures and visualize segmentation on emergent CT scans. The model's goal is to boost diagnostic accuracy, alleviate radiologists' workload, and hasten diagnosis, thereby enhancing patient outcomes. Unique to this research, both pediatric and post-operative patients were not excluded, and diagnostic durations were analyzed. Our testing dataset for the observer studies involved 671 patients, with a mean age of 58.88 years and fairly balanced gender representation. Model 1 of our AI algorithm, trained with 1499 fracture-positive cases, showed a sensitivity of 0.94 and specificity of 0.87, with a DICE score of 0.65. Implementing post-processing rules (specifically Rule B) improved the model's performance, resulting in a sensitivity of 0.94, specificity of 0.99, and a DICE score of 0.63. AI-assisted diagnosis resulted in significantly enhanced performance for all participants, with sensitivity almost doubling for junior radiology residents and other specialists. Additionally, diagnostic durations were significantly reduced (p < 0.01) with AI assistance across all participant categories. Our skull fracture detection model, employing a segmentation approach, demonstrated high performance, enhancing diagnostic accuracy and efficiency for radiologists and clinical physicians. This underlines the potential of AI integration in medical imaging analysis to improve patient care.

RAPID-ED: A predictive model for risk assessment of patient's early in-hospital deterioration from emergency department.

Introduction: The objective of this multi-center retrospective cohort study was to devise a predictive tool known as RAPID-ED. This model identifies non-traumatic adult patients at significant risk for cardiac arrest within 48 hours post-admission from the emergency department. Methods: Data from 224,413 patients admitted through the emergency department (2016-2020) was analyzed, incorporating vital signs, lab tests, and administered therapies. A multivariable regression model was devised to anticipate early cardiac arrest. The efficacy of the RAPID-ED model was evaluated against traditional scoring systems like National Early Warning Score (NEWS) and Modified Early Warning Score (MEWS) and its predictive ability was gauged via the area under the receiver operating characteristic curve (AUC) in both hold-out validation set and external validation set. Results: RAPID-ED outperformed traditional models in predicting cardiac arrest with an AUC of 0.819 in the hold-out validation set and 0.807 in the external validation set. In this critical care update, RAPID-ED offers an innovative approach to assessing patient risk, aiding emergency physicians in post-discharge care decisions from the emergency department. High-risk score patients (≥13) may benefit from early ICU admission for intensive monitoring. Conclusion: As we progress with advancements in critical care, tools like RAPID-ED will prove instrumental in refining care strategies for critically ill patients, fostering an improved prognosis and potentially mitigating mortality rates.

Serum Potassium Monitoring Using AI-Enabled Smartwatch Electrocardiograms.

BACKGROUND: Hyperkalemia, characterized by elevated serum potassium levels, heightens the risk of sudden cardiac death, particularly increasing risk for individuals with chronic kidney disease and end-stage renal disease (ESRD). Traditional laboratory test monitoring is resource-heavy, invasive, and unable to provide continuous tracking. Wearable technologies like smartwatches with electrocardiogram (ECG) capabilities are emerging as valuable tools for remote monitoring, potentially allowing for personalized monitoring with artificial intelligence (AI)-ECG interpretation. OBJECTIVES: The purpose of this study was to develop an AI-ECG algorithm to predict serum potassium level in ESRD patients with smartwatch-generated ECG waveforms. METHODS: A cohort of 152,508 patients with 293,557 ECGs paired serum potassium levels obtained within 1 hour at Cedars Sinai Medical Center was used to train an AI-ECG model ("Kardio-Net") to predict serum potassium level. The model was further fine-tuned on 4,337 ECGs from 1,463 patients with ESRD using inputs from 12- and single-lead ECGs. Kardio-Net was evaluated in held-out test cohorts from Cedars Sinai Medical Center and Stanford Healthcare (SHC) as well as a prospective international cohort of 40 ESRD patients with smartwatch ECGs at Chang Gung Memorial Hospital. RESULTS: The Kardio-Net, when applied to 12-lead ECGs, identified severe hyperkalemia (>6.5 mEq/L) with an AUC of 0.852 (95% CI: 0.745-0.956) and a mean absolute error (MAE) of 0.527 mEq/L. In external validation at SHC, the model achieved an AUC of 0.849 (95% CI: 0.823-0.875) and an MAE of 0.599 mEq/L. For single-lead ECGs, Kardio-Net detected severe hyperkalemia with an AUC of 0.876 (95% CI: 0.765-0.987) in the primary cohort and had an MAE of 0.575 mEq/L. In the external SHC validation, the AUC was 0.807 (95% CI: 0.778-0.835) with an MAE of 0.740 mEq/L. Using prospectively obtained smartwatch data, the AUC was 0.831 (95% CI: 0.693-0.975), with an MAE of 0.580 mEq/L. CONCLUSIONS: We validate a deep learning model to predict serum potassium levels from both 12-lead ECGs and single-lead smartwatch data, demonstrating its utility for remote monitoring of hyperkalemia.

Explainable deep learning model to predict invasive bacterial infection in febrile young infants: A retrospective study.

BACKGROUND: Machine learning models have demonstrated superior performance in predicting invasive bacterial infection (IBI) in febrile infants compared to commonly used risk stratification criteria in recent studies. However, the black-box nature of these models can make them difficult to apply in clinical practice. In this study, we developed and validated an explainable deep learning model that can predict IBI in febrile infants ≤ 60 days of age visiting the emergency department. METHODS: We conducted a retrospective study of febrile infants aged ≤ 60 days who presented to the pediatric emergency department of a medical center in Taiwan between January 1, 2011 and December 31, 2019. Patients with uncertain test results and complex chronic health conditions were excluded. IBI was defined as the growth of a pathogen in the blood or cerebrospinal fluid. We used a deep neural network to develop a predictive model for IBI and compared its performance to the IBI score and step-by-step approach. The SHapley Additive Explanations (SHAP) technique was used to explain the model's predictions at different levels. RESULTS: Our study included 1847 patients, 53 (2.7%) of whom had IBI. The deep learning model performed similarly to the IBI score and step-by-step approach in terms of sensitivity and negative predictive value, but provided better specificity (54%), positive predictive value (5%), and area under the receiver-operating characteristic curve (0.87). SHapley Additive exPlanations identified five influential predictive variables (absolute neutrophil count, body temperature, heart rate, age, and C-reactive protein). CONCLUSION: We have developed an explainable deep learning model that can predict IBI in febrile infants aged 0-60 days. The model not only performs better than previous scoring systems, but also provides insight into how it arrives at its predictions through individual features and cases.

Development and validation of a deep learning pipeline to measure pericardial effusion in echocardiography.

Objectives: The aim of this study was to develop a deep-learning pipeline for the measurement of pericardial effusion (PE) based on raw echocardiography clips, as current methods for PE measurement can be operator-dependent and present challenges in certain situations. Methods: The proposed pipeline consisted of three distinct steps: moving window view selection (MWVS), automated segmentation, and width calculation from a segmented mask. The MWVS model utilized the ResNet architecture to classify each frame of the extracted raw echocardiography files into selected view types. The automated segmentation step then generated a mask for the PE area from the extracted echocardiography clip, and a computer vision technique was used to calculate the largest width of the PE from the segmented mask. The pipeline was applied to a total of 995 echocardiographic examinations. Results: The proposed deep-learning pipeline exhibited high performance, as evidenced by intraclass correlation coefficient (ICC) values of 0.867 for internal validation and 0.801 for external validation. The pipeline demonstrated a high level of accuracy in detecting PE, with an area under the receiving operating characteristic curve (AUC) of 0.926 (95% CI: 0.902-0.951) for internal validation and 0.842 (95% CI: 0.794-0.889) for external validation. Conclusion: The machine-learning pipeline developed in this study can automatically calculate the width of PE from raw ultrasound clips. The novel concepts of moving window view selection for image quality control and computer vision techniques for maximal PE width calculation seem useful in the field of ultrasound. This pipeline could potentially provide a standardized and objective approach to the measurement of PE, reducing operator-dependency and improving accuracy.

Improving detection of impacted animal bones on lateral neck radiograph using a deep learning artificial intelligence algorithm.

OBJECTIVE: We aimed to develop a deep learning artificial intelligence (AI) algorithm to detect impacted animal bones on lateral neck radiographs and to assess its effectiveness for improving the interpretation of lateral neck radiographs. METHODS: Lateral neck radiographs were retrospectively collected for patients with animal bone impaction between January 2010 and March 2020. Radiographs were then separated into training, validation, and testing sets. A total of 1733 lateral neck radiographs were used to develop the deep learning algorithm. The testing set was assessed for the stand-alone deep learning AI algorithm and for human readers (radiologists, radiology residents, emergency physicians, ENT physicians) with and without the aid of the AI algorithm. Another radiograph cohort, collected from April 1, 2020, to June 30, 2020, was analyzed to simulate clinical application by comparing the deep learning AI algorithm with radiologists' reports. RESULTS: In the testing set, the sensitivity, specificity, and accuracy of the AI model were 96%, 90%, and 93% respectively. Among the human readers, all physicians of different subspecialties achieved a higher accuracy with AI-assisted reading than without. In the simulation set, among the 20 cases positive for animal bones, the AI model accurately identified 3 more cases than the radiologists' reports. CONCLUSION: Our deep learning AI model demonstrated a higher sensitivity for detection of animal bone impaction on lateral neck radiographs without an increased false positive rate. The application of this model in a clinical setting may effectively reduce time to diagnosis, accelerate workflow, and decrease the use of CT.

Correction: Yau et al. Risk Factors for Early Return Visits to the Emergency Department in Patients Presenting with Nonspecific Abdominal Pain and the Use of Computed Tomography Scan. Healthcare 2021, 9, 1470.

The authors would like to make corrections to their published paper [...].

Utilization of Personalized Machine-Learning to Screen for Dysglycemia from Ambulatory ECG, toward Noninvasive Blood Glucose Monitoring.

Blood glucose (BG) monitoring is important for critically ill patients, as poor sugar control has been associated with increased mortality in hospitalized patients. However, constant BG monitoring can be resource-intensive and pose a healthcare burden in clinical practice. In this study, we aimed to develop a personalized machine-learning model to predict dysglycemia from electrocardiogram (ECG) data. We used the Medical Information Mart for Intensive Care III database as our source of data and obtained more than 20 ECG records from each included patient during a single hospital admission. We focused on lead II recordings, along with corresponding blood sugar data. We processed the data and used ECG features from each heartbeat as inputs to develop a one-class support vector machine algorithm to predict dysglycemia. The model was able to predict dysglycemia using a single heartbeat with an AUC of 0.92 ± 0.09, a sensitivity of 0.92 ± 0.10, and specificity of 0.84 ± 0.04. After applying 10 s majority voting, the AUC of the model's dysglycemia prediction increased to 0.97 ± 0.06. This study showed that a personalized machine-learning algorithm can accurately detect dysglycemia from a single-lead ECG.

Clinical Validation of the Shock Index, Modified Shock Index, Delta Shock Index, and Shock Index-C for Emergency Department ST-Segment Elevation Myocardial Infarction.

Background: ST-segment elevation myocardial infarction (STEMI) is a leading cause of death worldwide. A shock index (SI), modified SI (MSI), delta-SI, and shock index-C (SIC) are known predictors of STEMI. This retrospective cohort study was designed to compare the predictive value of the SI, MSI, delta-SI, and SIC with thrombolysis in myocardial infarction (TIMI) risk scales. Method: Patients > 20 years old with STEMI who underwent percutaneous coronary intervention (PCI) were included. Receiver operating characteristic (ROC) curve analysis with the Youden index was performed to calculate the optimal cutoff values for these predictors. Results: Overall, 1552 adult STEMI cases were analyzed. The thresholds for the emergency department (ED) SI, MSI, SIC, and TIMI risk scales for in-hospital mortality were 0.75, 0.97, 21.00, and 5.5, respectively. Accordingly, ED SIC had better predictive power than the ED SI and ED MSI. The predictive power was relatively higher than TIMI risk scales, but the difference did not achieve statistical significance. After adjusting for confounding factors, the ED SI > 0.75, MSI > 0.97, SIC > 21.0, and TIMI risk scales > 5.5 were statistically and significantly associated with in-hospital mortality of STEMI. Compared with the ED SI and MSI, SIC (>21.0) had better sensitivity (67.2%, 95% CI, 58.6−75.9%), specificity (83.5%, 95% CI, 81.6−85.4%), PPV (24.8%, 95% CI, 20.2−29.6%), and NPV (96.9%, 95% CI, 96.0−97.9%) for in-hospital mortality of STEMI. Conclusions: SIC had better discrimination ability than the SI, MSI, and delta-SI. Compared with the TIMI risk scales, the ACU value of SIC was still higher. Therefore, SIC might be a convenient and rapid tool for predicting the outcome of STEMI.

Ambient Air Pollution and Risk for Stroke Hospitalization: Impact on Susceptible Groups.

Stroke is a leading cause of death, and air pollution is associated with stroke hospitalization. However, the susceptibility factors are unclear. Retrospective studies from 2014 to 2018 in Kaohsiung, Taiwan, were analyzed. Adult patients (>17 years) admitted to a medical center with stroke diagnosis were enrolled and patient characteristics and comorbidities were recorded. Air pollutant measurements, including those of particulate matter (PM) with aerodynamic diameters < 10 µm (PM10) and < 2.5 µm (PM2.5), nitrogen dioxide (NO2), and ozone (O3), were collected from air quality monitoring stations. During the study period, interquartile range (IQR) increments in PM2.5 on lag3 and lag4 were 12.3% (95% CI, 1.1−24.7%) and 11.5% (95% CI, 0.3−23.9%) concerning the risk of stroke hospitalization, respectively. Subgroup analysis revealed that the risk of stroke hospitalization after exposure to PM2.5 was greater for those with advanced age (≥80 years, interaction p = 0.045) and hypertension (interaction p = 0.034), after adjusting for temperature and humidity. A dose-dependent effect of PM2.5 on stroke hospitalization was evident. This is one of few studies focusing on the health effects of PM2.5 for patients with risk factors of stroke. We found that patients with risk factors, such as advanced age and hypertension, are more susceptible to PM2.5 impacts on stroke hospitalization.