Thermal imaging and deep learning-based fit-checking for respiratory protection.

This study develops an artificial intelligence model to quickly and easily determine correct mask-wearing in real time using thermal videos that ascertained temperature changes caused by air trapped inside the mask. Five types of masks approved by the Korean Ministry of Food and Drug Safety were worn in four different ways across 50 participants, generating 5000 videos. The results showed that 3DCNN outperformed ConvLSTM in both binary and multi-classification for mask wearing methods, with the highest AUROC of 0.986 for multi-classification. Each mask type scored AUROC values > 0.9, with KF-AD being the best classified. This improved use of thermal imaging and deep learning for mask fit-checking could be useful in high-risk environments. It can be applied to various mask types, which enables easy generalizability and advantages in public and occupational health and healthcare system.

Comparison of AI with and without hand-crafted features to classify Alzheimer's disease in different languages.

BACKGROUND: Detecting and analyzing Alzheimer's disease (AD) in its early stages is a crucial and significant challenge. Speech data from AD patients can aid in diagnosing AD since the speech features have common patterns independent of race and spoken language. However, previous models for diagnosing AD from speech data have often focused on the characteristics of a single language, with no guarantee of scalability to other languages. In this study, we used the same method to extract acoustic features from two language datasets to diagnose AD. METHODS: Using the Korean and English speech datasets, we used ten models capable of real-time AD and healthy control classification, regardless of language type. Four machine learning models were based on hand-crafted features, while the remaining six deep learning models utilized non-explainable features. RESULTS: The highest accuracy achieved by the machine learning models was 0.73 and 0.69 for the Korean and English speech datasets, respectively. The deep learning models' maximum achievable accuracy reached 0.75 and 0.78, with their minimum classification time of 0.01s and 0.02s. These findings reveal the models' robustness regardless of Korean and English and real-time diagnosis of AD through a 30-s voice sample. CONCLUSION: Non-explainable deep learning models that directly acquire voice representations surpassed machine learning models utilizing hand-crafted features in AD diagnosis. In addition, these AI models could confirm the possibility of extending to a language-agnostic AD diagnosis.

Assessing the Proficiency of LLMs with Various Tasks and Evaluators.

Previous studies have been limited to giving one or two tasks to Large Language Models (LLMs) and involved a small number of evaluators within a single domain to evaluate the LLM's answer. We assessed the proficiency of four LLMs by applying eight tasks and evaluating 32 results with 17 evaluators from diverse domains, demonstrating the significance of various tasks and evaluators on LLMs.

Real-world treatment intensities and pathways of macular edema following retinal vein occlusion in Korea from Common Data Model in ophthalmology.

Despite many studies, optimal treatment sequences or intervals are still questionable in retinal vein occlusion (RVO) macular edema. The aim of this study was to examine the real-world treatment patterns of RVO macular edema. A retrospective analysis of the Observational Medical Outcomes Partnership Common Data Model, a distributed research network, of four large tertiary referral centers (n = 9,202,032) identified 3286 eligible. We visualized treatment pathways (prescription volume and treatment sequence) with sunburst and Sankey diagrams. We calculated the average number of intravitreal injections per patient in the first and second years to evaluate the treatment intensities. Bevacizumab was the most popular first-line drug (80.9%), followed by triamcinolone (15.1%) and dexamethasone (2.28%). Triamcinolone was the most popular drug (8.88%), followed by dexamethasone (6.08%) in patients who began treatment with anti-vascular endothelial growth factor (VEGF) agents. The average number of all intravitreal injections per person decreased in the second year compared with the first year. The average number of injections per person in the first year increased throughout the study. Bevacizumab was the most popular first-line drug and steroids were considered the most common as second-line drugs in patients first treated with anti-VEGF agents. Intensive treatment patterns may cause an increase in intravitreal injections.

Semi-AI and Full-AI digitizer: The ways to digitalize visual field big data.

BACKGROUND AND OBJECTIVE: Glaucoma is one of the major diseases that cause blindness, which is incurable and irreversible, and it is essential to detect glaucoma vision deficits in treatment and check the progression of vision disorders in advance. In order to minimize the risk of glaucoma, it is necessary not only to diagnose and observe glaucoma but also to predict prognosis via indicators from Visual Field (VF) tests. However, information from the VF test cannot be directly used in clinical studies because most medical institutions store VF test sheets in Portable Document Format (PDF) or image files in different standards. METHODS: We developed AI-based real-time VF big data digitizing systems that digitalize VF test images in real-time in two ways; Semi-AI and Full-AI digitizer. The Semi-AI digitizer detects the VF text area with actual coordinates derived from mouse handler system. Full-AI digitizer detects the VF text area with Faster Region Based Convolutional Neural Networks (RCNN). After detecting the text area, both systems extract texts with Recurrent Neural Network based Optical Character Recognition. Semi-AI and Full-AI digitizer post-processes the extracted text results with in-system algorithm and out-of-system algorithm, respectively. RESULTS: Both systems used 325,310 VF test sheets from a tertiary hospital and extracted a total of 5,530,270 texts. From the 100 randomly selected VF sheets, 3,400 texts were used for the validation. Semi-AI and Full-AI digitizer showed 0.993 and 0.983 of accuracy, respectively. CONCLUSION: This study demonstrates the effectiveness of AI applications in detecting text areas and the different implementation methodologies of the post-processing process. In detecting text area, Semi-AI may be better than Full-AI digitizer in terms of system speed and human labor labeling if the number of types to be classified is small. However, Full-AI digitizer is recommended because it allows detecting text area regardless of resolution and size of the VF sheets, as the types of real-world VF test sheets cannot be predicted, and the types become more unpredictable when extended to multi-hospital studies. For Post-preprocessing, Semi-AI methodology is recommended because Semi-AI produced higher results with less effort and considered the convenience of researchers by implementing them as in-system.

Characteristics of Hypoglycemic Diabetic Patients Visiting the Emergency Room.

INTRODUCTION: Severe hypoglycemia can be life-threatening; therefore, it is important to identify the characteristics of the hypoglycemic patients. The aim of this study is to analyze the type and characteristics of diabetic patients with hypoglycemia who visited an emergency room. METHODS: We included diabetic patients with hypoglycemia who visited the emergency room of St. Mary's Hospital in Seoul from January 2009 to August 2018 in the study. Hypo_S group patients visited the emergency room once whereas Hypo_M group patients visited twice or more. We also compared the incidence of cardiovascular disease between the groups within 5 years after hypoglycemia. RESULTS: A total of 843 patients were included in this study, with a mean age of 71 ± 14 years and average glycated hemoglobin (HbA1c) level of 6.7 ± 1.4%. For patients with hypoglycemia, lower body mass index, lower HbA1c, shorter diabetes duration, and lower glomerular filtration rate have a statistically significant relationship with patient characteristics in the emergency room group (all p < 0.001). Hypoglycemia symptoms were most frequently observed between 6:00 and 12:00 am (p < 0.001). Cardiovascular diseases within 5 years after discharge were more frequent in the Hypo_S group than in the Hypo_M group; however, there was no statistical significance. The frequency of aneurysms was significantly higher in patients with hypoglycemia than in other patients in the emergency room (p < 0.05). CONCLUSION: Relatively thin older patients with a diabetes duration shorter than 10 years and good blood sugar control showed higher frequency of visits to the emergency room due to hypoglycemia. For these patients, medical staff should always be mindful of their susceptibility to hypoglycemia when prescribing insulin or OHA and educate them on the prevention of hypoglycemia.

Differences in Clinical Outcomes between Patients with and without Hypoglycemia during Hospitalization: A Retrospective Study Using Real-World Evidence.

BACKGROUND: Some patients admitted to hospitals for glycemic control experience hypoglycemia despite regular meals and despite adhering to standard blood glucose control protocols. Different factors can have a negative impact on blood glucose control and prognosis after discharge. This study investigated risk factors for hypoglycemia and its effects on glycemic control during the hospitalization of patients in the general ward. METHODS: This retrospective study included patients who were admitted between 2009 and 2018. Patients were provided regular meals at fixed times according to ideal body weights during hospitalization. We categorized the patients into two groups: those with and those without hypoglycemia during hospitalization. RESULTS: Of the 3,031 patients, 379 experienced at least one episode of hypoglycemia during hospitalization (HYPO group). Hypoglycemia occurred more frequently particularly in cases of premixed insulin therapy. Compared with the control group, the HYPO group was older (61.0±16.8 years vs. 59.1±16.5 years, P=0.035), with more females (60.4% vs. 49.6%, P<0.001), lower body mass index (BMI) (23.5±4.2 kg/m² vs. 25.1±4.4 kg/m², P<0.001), and higher prevalence of type 1 diabetes mellitus (6.1% vs. 2.6%, P<0.001), They had longer hospital stay (11.1±13.5 days vs. 7.6±4.6 days, P<0.001). After discharge the HYPO group had lower glycosylated hemoglobin reduction rate (-2.0%±0.2% vs. -2.5%±0.1%, P=0.003) and tended to have more frequent cases of cardiovascular disease. CONCLUSION: Hypoglycemia occurred more frequently in older female patients with lower BMI and was associated with longer hospital stay and poorer glycemic control after discharge. Therefore, clinicians must carefully ensure that patients do not experience hypoglycemia during hospitalization.

Expansion of EHR-Based Common Data Model (CDM).

We expanded and constructed a Common Data Model (CDM) based on hospital EHR to enable analysis and comparison of Adverse Drug Reactions(ADRs) integrated with external organizations with different data structures. This is significant in that it is possible to conduct joint research, analysis, and comparisons among institutions with the same type of CDM constructed, and provide the basis for conducting the same research simultaneously on various data sources.

Clinical experiences and case review of angiotensin II receptor blocker-related angioedema in Korea.

Although angiotensin-converting enzyme inhibitor-related angioedema is well known, angiotensin II receptor blocker (ARB)-related angioedema has not been extensively studied because of its lower incidence. Therefore, ARB-related angioedema is likely to be overlooked in the clinical setting. We analysed the medical records of adults who had been prescribed ARB and diagnosed with angioedema between 2009 and 2015. All adults over the age of 18 years who were initially administered ARB between 1 January 2009 and 31 December 2015 were selected as participants in this study. To confirm whether the angioedema was actually due to the administration of ARB, we conducted a chart review. A total of 35 584 patients were prescribed ARB for the first time when visiting the Seoul St. Mary's Hospital during the study period. Twenty-four patients diagnosed with angioedema for other reasons prior to their first prescription of ARB were excluded from this study. ARB-related angioedema was suspected in six of 35 560 patients (0.02%) who were initially prescribed ARB during the study period. The manifestation of ARB-related angioedema ranged from several days (1/6 case) to several years (3/6 cases). Some patients continued taking ARB with intermittent antihistamine or steroid therapy. In such cases, angioedema symptoms improved but did not completely resolve. Its diagnosis can be delayed and the symptoms may be recurrent as symptoms improve with antihistamine use. In some cases, the same person had different reactions depending on the type of ARB. Definitively diagnosing ARB-related angioedema is difficult, and physicians often overlook angioedema without suspecting that it is an adverse effect of ARB. Close attention of physicians and improved patient education can reduce the incidence of ARB-related angioedema.

Long-term effects of various types of 3-hydroxy-3-methylglutaryl coenzyme A reductase inhibitors on changes in glomerular filtration rate in Korea.

Few long-term follow-up studies have compared the changes in renal function according to the type of statin used in Korea. We compared the long-term effects of statin intensity and type on the changes in the glomerular filtration rate (GFR). We extracted data of patients who took statin for the first time. We analyzed whether or not different statins affect the changes in GFR at 3 months after baseline and 4 years after. We included 3678 patients and analyzed the changes in GFR. The GFR decreased by 3.2% ± 0.4% on average 4 years after the first statin prescription, indicating statistically significant deterioration (from 83.5 ± 0.4 mL/min/1.73 m2 to 79.9 0.4 mL/min/1.73 m2, P < 0.001). When comparing the GFR among different statins, significant differences were observed between atorvastatin and fluvastatin (-5.3% ± 0.7% vs. 1.2% ± 2.2%, P < 0.05) and between atorvastatin and simvastatin (-5.3% ± 0.7% vs. -0.7% ± 0.8%, P < 0.05). In pitavastatin (odds ratio [OR]= 0.64, 95% confidence interval [CI]= 0.46-0.87, P < 0.005) and simvastatin (OR = 0.69, 95% CI = 0.53-0.91, P < 0.008), the GFR rate that decreased by < 60 mL/min/1.73 m2 was significantly lower than that of atorvastatin. Regarding long-term statin intake, GFR changed with the type of statin. This work is the first in Korea to compare each statin in terms of changes in the GFR after the statin prescription.

Developing a multi-center clinical data mart of ACEI and ARB for real-world evidence (RWE).

BACKGROUND: Randomized controlled trials can be expensive and time-consuming, leading to medical researchers utilizing real-world evidence (RWE) based on already-collected data. We aimed to conduct various RWE studies on angiotensin-converting enzyme inhibitors (ACEI) and angiotensin II receptor blocker (ARB), commonly used as first-line therapy for blood pressure, and to develop a multi-center clinical data mart (CDM) of ACEI/ARB for various clinical purposes. METHODS: Data from electronic medical records of St. Mary's Hospital and the Seoul National University Hospital were collected. We obtained blood and urine test results of patients within the 30 days prior to their first prescription of ACEI or ARB, as well as the first date of diagnosis and presence of various chronic and cardiovascular diseases using the International Classification of Diseases-10 classification. One researcher managed data quality and collation for each hospital in order to facilitate patient anonymity. When results were unclear, the responsible investigator for each hospital attempted to resolve ambiguities by direct chart review. RESULTS: A total of 102,333 patients who were prescribed ACEI or ARB for the first time were included (21,481 ACEI, 80,551 ARB, and 301 both). Our ACEI/ARB-CDM included short-term studies (within 12 months) to observe changes in various blood or urinary laboratory test values after the initial prescription of ACEI or ARB and long-term studies to confirm the incidence of various diseases. CONCLUSION: We established a CDM of RWE for ACEI/ARB prescription, which included various clinical studies. As we accumulate experience in this process, we expect that the use of RWE research will grow and develop.

Change in ALT levels after administration of HMG-CoA reductase inhibitors to subjects with pretreatment levels three times the upper normal limit in clinical practice.

AIM: Prescription of statins to patients with chronic liver disease whose alanine transaminase (ALT) is over three times the upper normal limit (UNL) is not recommended. In this study, we attempted to evaluate patients with baseline ALT levels > 3 ×  UNL who were prescribed statins without ethical problems, using electronic medical records. METHODS: We enrolled subjects with ALT levels > 3 ×  UNL. The patients were divided into three groups consisting of those who had been taking agents affecting liver function (HEPA) and continued to do so after the statin prescription (HepCon), those who had not previously taken HEPA and began doing so after statin prescription (HepNew), and those who had never taken HEPA (HepNo). All ALT levels were determined within 3 months of statins administration, and changes were monitored. RESULTS: From January 2009 to December 2012, 61 patients with baseline ALT levels > 3 ×  UNL were prescribed statins for the first time. During the 3-month ALT monitoring, levels of the HepCon, HepNew, and HepNo groups decreased by 45 ± 8%, 64 ± 10%, and 42 ± 8%, respectively; however, intergroup differences were not significant (P = .386). All the subjects who were administered statins showed improvement or maintained their ALT levels, except for two subjects, which showed deterioration. However, the ALT levels of the two subjects subsequently remained stable. CONCLUSION: It is not clear whether it is safe to prescribe statins to patients with ALT > 3 times the UNL. Our study showed that prescription of statins in combination with HEPA did not cause deleterious effects, suggesting that ALT levels > 3 times the UNL do not have harmful effects.

The differences in the incidence of diabetes mellitus and prediabetes according to the type of HMG-CoA reductase inhibitors prescribed in Korean patients.

BACKGROUND: Very few studies conducted in Korea have investigated the relationship between statins and the incidence of diabetes. Therefore, we analyzed the progression from normal blood glucose to prediabetes and then to diabetes mellitus (DM) according to the type, intensity, and dose of statin prescribed. METHODS: Data of patients who were first prescribed statins between 2009 and 2011 were extracted from electronic medical records. Patients with normal blood glucose or prediabetes were observed for 4 years after initiation of statin therapy. RESULTS: A total of 2890 patients were included in our study and analyzed on the basis of the first statin they were prescribed. The incidence rate of DM in patients with prediabetes was 1.72 times that of patients with normal glucose levels (odds ratio = 1.72, 95% confidence interval = 1.41-2.10, P < .001). Regarding progression from normal blood glucose to prediabetes, the incidence rate of prediabetes was significantly lower in patients prescribed pitavastatin (odds ratio = 0.62, 95% confidence interval = 0.40-0.96, P = .031) compared to that in patients prescribed atorvastatin. Regarding the progression from normal blood glucose or prediabetes to DM, there were no significant differences among all statins. CONCLUSIONS: Lower DM incidence in patients prescribed pitavastatin appears to be primarily because of the lower rate of progression from normal blood glucose to prediabetes. These findings indicate that avoiding statins because of DM risk is unjustified and that clinicians should prescribe statins from the appropriate potency group.

Physician-Directed Diabetes Education without a Medication Change and Associated Patient Outcomes.

BACKGROUND: When patients with diabetes mellitus (DM) are first referred to a hospital from primary health care clinics, physicians have to decide whether to administer an oral hypoglycemic agent (OHA) immediately or postpone a medication change in favor of diabetes education regarding diet or exercise. The aim of this study was to determine the effect of diabetes education alone (without alterations in diabetes medication) on blood glucose levels. METHODS: The study was conducted between January 2009 and December 2013 and included patients with DM. The glycosylated hemoglobin (HbA1c) levels were evaluated at the first visit and after 3 months. During the first medical examination, a designated doctor also conducted a diabetes education session that mainly covered dietary management. RESULTS: Patients were divided into those who received no diabetic medications (n=66) and those who received an OHA (n=124). Education resulted in a marked decrease in HbA1c levels in the OHA group among patients who had DM for <1 year (from 7.0%±1.3% to 6.6%±0.9%, P=0.0092) and for 1 to 5 years (from 7.5%±1.8% to 6.9%±1.1%, P=0.0091). Those with DM >10 years showed a slightly lower HbA1c target achievement rate of <6.5% (odds ratio, 0.089; P=0.0024). CONCLUSION: For patients who had DM for more than 5 years, higher doses or changes in medication were more effective than intensive active education. Therefore, individualized and customized education are needed for these patients. For patients with a shorter duration of DM, it may be more effective to provide initial intensive education for diabetes before prescribing medicines, such as OHAs.

Development of Clinical Data Mart of HMG-CoA Reductase Inhibitor for Varied Clinical Research.

BACKGROUND: The increasing use of electronic medical record (EMR) systems for documenting clinical medical data has led to EMR data being increasingly accessed for clinical trials. In this study, a database of patients who were prescribed statins for the first time was developed using EMR data. A clinical data mart (CDM) was developed for cohort study researchers. METHODS: Seoul St. Mary's Hospital implemented a clinical data warehouse (CDW) of data for ~2.8 million patients, 47 million prescription events, and laboratory results for 150 million cases. We developed a research database from a subset of the data on the basis of a study protocol. Data for patients who were prescribed a statin for the first time (between the period from January 1, 2009 to December 31, 2015), including personal data, laboratory data, diagnoses, and medications, were extracted. RESULTS: We extracted initial clinical data of statin from a CDW that was established to support clinical studies; the data was refined through a data quality management process. Data for 21,368 patients who were prescribed statins for the first time were extracted. We extracted data every 3 months for a period of 1 year. A total of 17 different statins were extracted. It was found that statins were first prescribed by the endocrinology department in most cases (69%, 14,865/21,368). CONCLUSION: Study researchers can use our CDM for statins. Our EMR data for statins is useful for investigating the effectiveness of treatments and exploring new information on statins. Using EMR is advantageous for compiling an adequate study cohort in a short period.

Use of Moderate-Intensity Statins for Low-Density Lipoprotein Cholesterol Level above 190 mg/dL at Baseline in Koreans.

The ACC/AHA 2013 guideline recommends high-intensity statin therapy for a decrease in low-density lipoprotein cholesterol (LDL-C) level by >50% among patients with baseline values of ≥190 mg/dL (approximately 4.872 mmol/L); however, this value should be modified before applying it to Korean populations. We investigated the statin-specific LDL-C-lowering effects in Korean patients with baseline LDL-C value ≥4.872 mmol/L. Data of patients prescribed a statin for the first time from January 2009 to December 2013 were assessed. In patients with baseline LDL-C value ≥4.872 mmol/L, laboratory data for a maximum of 6 months from the date of first statin prescription were collected. Among 33,721 patients who were prescribed a statin for the first time, 655 patients had a baseline LDL-C value ≥4.872 mmol/L (1.9%). Of these, 179 patients were analysed. Patients receiving moderate-intensity statins were divided into two groups based on LDL-C reduction rate (p = 0.0002), defined as moderate-high-intensity (atorvastatin 20 mg, rosuvastatin 10 mg, simvastatin 20 mg) and moderate-low-intensity (atorvastatin 10 mg, pitavastatin 2 mg, pravastatin 40 mg) statin groups. LDL-C reduction rates did not significantly differ between the moderate-high- and high-intensity statin groups (p = 0.4895). We found that some moderate-intensity statins demonstrated a LDL-C-lowering effect of more than 50% in Korean patients with a baseline LDL-C value ≥4.872 mmol/L. Our results reflect the need of a large-scale, randomized, controlled trial on partial reclassification of statins for patients with baseline LDL-C value ≥4.872 mmol/L before adopting ACC/AHC guidelines in Korea.

Social Networking Services-Based Communicative Care for Patients with Diabetes Mellitus in Korea.

OBJECTIVES: Social networking services (SNS)-based online communities are good examples of improving quality of care by incorporating information technology into medicine. Therefore, we created an SNS-based community care webpage for communication among patients with diabetes mellitus (DM). We aimed to identify what diabetic patients wanted to know and were interested in by analyzing their posts and classified content in which users were interested. METHODS: As opposed to the existing physician-focused health information websites, we built a patient-focused experience exchange website, "I love insulin (http://www.iloveinsulin.co.kr)." The DM communication webpage was divided into "My Web-Chart," "My community-free board," and "Life & Health." The contents analysis targeted users' postings, and replies were classified by theme from May 2012 to June 2013. The data included number of questions asked, answers, and question-to-answer (QA) ratio in each category. RESULTS: A total of 264 patients registered on the "I Love Insulin" website. The most frequent topic of posts classified as questions were about diabetes itself (23%), diet (22%), and glucose levels (19%). Conversely, most answers and information provided by users were about daily life with no relationship to diabetes mellitus (54%). While there were many questions about diet, there were very few answers (2%). Whereas there was much provision of knowledge about general DM, sharing diet information was rare. The ratios of answers to questions on diet (ratio=0.059, 1/17), glucose level (ratio=0.067, 1/15), insulin regulation (ratio=0.222, 2/9) and webpage (ratio=0.167, 1/6) were significantly low compared to DM itself (all p < 0.001). DISCUSSION: Patients in Korea with DM tend to have insufficient knowledge about diet and insulin regulation; continuously providing diet and insulin regulation information are desirable. It is hoped that the patients would be motivated to participate actively by "knowledge sharing." Through this process, patients learn about their diseases not from the physicians but from among themselves.