Enhancing readmission prediction models by integrating insights from home healthcare notes: Retrospective cohort study.

BACKGROUND: Hospital readmission is an important indicator of inpatient care quality and a significant driver of increasing medical costs. Therefore, it is important to explore the effects of postdischarge information, particularly from home healthcare notes, on enhancing readmission prediction models. Despite the use of Natural Language Processing (NLP) and machine learning in prediction model development, current studies often overlook insights from home healthcare notes. OBJECTIVE: This study aimed to develop prediction models for 30-day readmissions using home healthcare notes and structured data. In addition, it explored the development of 14- and 180-day prediction models using variables in the 30-day model. DESIGN: A retrospective observational cohort study. SETTING(S): This study was conducted at Ajou University School of Medicine in South Korea. PARTICIPANTS: Data from electronic health records, encompassing demographic characteristics of 1819 participants, along with information on conditions, drug, and home healthcare, were utilized. METHODS: Two distinct models were developed for each prediction window (30-, 14-, 180-day): the traditional model, which utilized structured variables alone, and the common data model (CDM)-NLP model, which incorporated structured and topic variables extracted from home healthcare notes. BERTopic facilitated topic generation and risk probability, representing the likelihood of documents being assigned to specific topics. Feature selection involved experimenting with various algorithms. The best-performing algorithm, determined using the area under the receiver operating characteristic curve (AUROC), was used for model development. Model performance was assessed using various learning metrics including AUROC. RESULTS: Among 1819 patients, 251 (13.80 %) experienced 30-day readmission. The least absolute shrinkage and selection operator was used for feature extraction and model development. The 15 structured features were used in the traditional model. Moreover, five additional topic variables from the home healthcare notes were applied in the CDM-NLP model. The AUROC of the traditional model was 0.739 (95 % CI: 0.672-0.807). The AUROC of the CDM-NLP model was high at 0.824 (95 % CI: 0.768-0.880), which indicated an outstanding performance. The topics in the CDM-NLP model included emotional distress, daily living functions, nutrition, postoperative status, and cardiorespiratory issues. In extended prediction model development for 14- and 180-day readmissions, the CDM-NLP consistently outperformed the traditional model. CONCLUSIONS: This study developed effective prediction models using both structured and unstructured data, thereby emphasizing the significance of postdischarge information from home healthcare notes in readmission prediction.

Neuroimaging and natural language processing-based classification of suicidal thoughts in major depressive disorder.

Suicide is a growing public health problem around the world. The most important risk factor for suicide is underlying psychiatric illness, especially depression. Detailed classification of suicide in patients with depression can greatly enhance personalized suicide control efforts. This study used unstructured psychiatric charts and brain magnetic resonance imaging (MRI) records from a psychiatric outpatient clinic to develop a machine learning-based suicidal thought classification model. The study included 152 patients with new depressive episodes for development and 58 patients from a geographically different hospital for validation. We developed an eXtreme Gradient Boosting (XGBoost)-based classification models according to the combined types of data: independent components-map weightings from brain T1-weighted MRI and topic probabilities from clinical notes. Specifically, we used 5 psychiatric symptom topics and 5 brain networks for models. Anxiety and somatic symptoms topics were significantly more common in the suicidal group, and there were group differences in the default mode and cortical midline networks. The clinical symptoms plus structural brain patterns model had the highest area under the receiver operating characteristic curve (0.794) versus the clinical notes only and brain MRI only models (0.748 and 0.738, respectively). The results were consistent across performance metrics and external validation. Our findings suggest that focusing on personalized neuroimaging and natural language processing variables improves evaluation of suicidal thoughts.

Time Series AI Model for Acute Kidney Injury Detection Based on a Multicenter Distributed Research Network: Development and Verification Study.

Background: Acute kidney injury (AKI) is a marker of clinical deterioration and renal toxicity. While there are many studies offering prediction models for the early detection of AKI, those predicting AKI occurrence using distributed research network (DRN)-based time series data are rare. Objective: In this study, we aimed to detect the early occurrence of AKI by applying an interpretable long short-term memory (LSTM)-based model to hospital electronic health record (EHR)-based time series data in patients who took nephrotoxic drugs using a DRN. Methods: We conducted a multi-institutional retrospective cohort study of data from 6 hospitals using a DRN. For each institution, a patient-based data set was constructed using 5 drugs for AKI, and an interpretable multivariable LSTM (IMV-LSTM) model was used for training. This study used propensity score matching to mitigate differences in demographics and clinical characteristics. Additionally, the temporal attention values of the AKI prediction model's contribution variables were demonstrated for each institution and drug, with differences in highly important feature distributions between the case and control data confirmed using 1-way ANOVA. Results: This study analyzed 8643 and 31,012 patients with and without AKI, respectively, across 6 hospitals. When analyzing the distribution of AKI onset, vancomycin showed an earlier onset (median 12, IQR 5-25 days), and acyclovir was the slowest compared to the other drugs (median 23, IQR 10-41 days). Our temporal deep learning model for AKI prediction performed well for most drugs. Acyclovir had the highest average area under the receiver operating characteristic curve score per drug (0.94), followed by acetaminophen (0.93), vancomycin (0.92), naproxen (0.90), and celecoxib (0.89). Based on the temporal attention values of the variables in the AKI prediction model, verified lymphocytes and calcvancomycin ium had the highest attention, whereas lymphocytes, albumin, and hemoglobin tended to decrease over time, and urine pH and prothrombin time tended to increase. Conclusions: Early surveillance of AKI outbreaks can be achieved by applying an IMV-LSTM based on time series data through an EHR-based DRN. This approach can help identify risk factors and enable early detection of adverse drug reactions when prescribing drugs that cause renal toxicity before AKI occurs.

Comparison of Finasteride and Dutasteride on Risk of Prostate Cancer in Patients with Benign Prostatic Hyperplasia: A Pooled Analysis of 15 Real-world Databases.

PURPOSE: Finasteride and dutasteride are used to treat benign prostatic hyperplasia (BPH) and reduce the risk of developing prostate cancer. Finasteride blocks only the type 2 form of 5-alpha-reductase, whereas dutasteride blocks both type 1 and 2 forms of the enzyme. Previous studies suggest the possibility that dutasteride may be superior to finasteride in preventing prostate cancer. We directly compared the effects of finasteride and dutasteride on the risk of prostate cancer in patients with BPH using a pooled analysis of 15 real-world databases. MATERIALS AND METHODS: We conducted a multicenter, cohort study of new-users of finasteride and dutasteride. We include patients who were prescribed 5 mg finasteride or dutasteride for the first time to treat BPH and had at least 180 days of prescription. We excluded patients with a history of prostate cancer or a prostate-specific antigen level ≥ 4 ng/mL before the study drug prescription. Cox regression analysis was performed to examine the hazard ratio (HR) for prostate cancer after propensity score (PS) matching. RESULTS: A total of 8,284 patients of new-users of finasteride and 8,670 patients of new-users of dutasteride were included across the 15 databases. In the overall population, compared to dutasteride, finasteride was associated with a lower risk of prostate cancer in both on-treatment and intent-to-treat time-at-risk periods. After 1:1 PS matching, 4,897 patients using finasteride and 4,897 patients using dutasteride were enrolled in the present study. No significant differences were observed for risk of prostate cancer between finasteride and dutasteride both on-treatment (HR=0.66, 95% confidence interval [CI]: 0.44-1.00; p=0.051) and intent-to-treat time-at-risk periods (HR=0.87, 95% CI: 0.67-1.14; p=0.310). CONCLUSIONS: Using real-world databases, the present study demonstrated that dutasteride was not associated with a lower risk of prostate cancer than finasteride in patients with BPH.

Comparative effectiveness and safety outcomes between denosumab and bisphosphonate in South Korea.

Both bisphosphonates and denosumab are the mainstays of treatment for osteoporosis to prevent fractures. However, there are still few trials directly comparing the prevention of fractures and the safety of 2 drugs in the treatment of osteoporosis. We aimed to compare the efficacy and safety between denosumab and bisphosphonates using a nationwide claims database. The database was covered with 10 million, 20% of the whole Korean population sampled by age and sex stratification of the Health Insurance Review and Assessment Service in South Korea. Among 228 367 subjects who were over 50 yr of age and taking denosumab or bisphosphonate from January 2018 to April 2022, the analysis was performed on 91 460 subjects after 1:1 propensity score matching. The primary outcome was treatment effectiveness; total fracture, major osteoporotic fracture, femur fracture, pelvic fracture, vertebral fracture, adverse drug reactions; acute kidney injury, chronic kidney disease, and atypical femoral fracture. Total fracture and osteoporotic major fracture, as the main outcomes of efficacy, were comparable in the denosumab and bisphosphonate group (HR 1.06, 95% CI, 0.98-1.15, P = .14; HR 1.13, 95% CI, 0.97-1.32, P = .12, respectively). Safety for acute kidney injury, chronic kidney disease, and atypical femoral fracture also did not show any differences between the 2 groups. In subgroup analysis according to ages, the denosumab group under 70 yr of age had a significantly lower risk for occurrences of acute kidney injury compared to the bisphosphonate group under 70 yr of age (HR 0.53, 95% CI, 0.29-0.93, P = .03). In real-world data reflecting clinical practice, denosumab and bisphosphonate showed comparable effectiveness for total fractures and major osteoporosis fractures, as well as safety regarding acute kidney injury, chronic kidney disease, and atypical femoral fracture.

This study compared the effectiveness and safety of denosumab and bisphosphonates, 2 primary treatments for osteoporosis, using a large South Korean nationwide claims database. Analysis of data from 91 460 individuals over 50 yr old showed no significant difference in preventing fractures or in safety outcomes such as kidney injury and atypical femoral fractures between the 2 drugs. However, among patients under 70, denosumab was associated with a lower risk of acute kidney injury. Overall, both medications demonstrated similar effectiveness and safety in the real-world treatment of osteoporosis.

Association between pre-diagnostic serum uric acid levels in patients with newly diagnosed epilepsy and conversion rate to drug-resistant epilepsy within 5 years: A common data model analysis.

PURPOSE: Drug-resistant epilepsy (DRE) poses a significant challenge in epilepsy management, and reliable biomarkers for identifying patients at risk of DRE are lacking. This study aimed to investigate the association between serum uric acid (UA) levels and the conversion rate to DRE. METHODS: A retrospective cohort study was conducted using a common data model database. The study included patients newly diagnosed with epilepsy, with prediagnostic serum UA levels within a six-month window. Patients were categorized into hyperUA (≥7.0 mg/dL), normoUA (<7.0 and >2.0 mg/dL), and hypoUA (≤2.0 mg/dL) groups based on their prediagnostic UA levels. The outcome was the conversion rate to DRE within five years of epilepsy diagnosis. RESULTS: The study included 5,672 patients with epilepsy and overall conversion rate to DRE was 19.4%. The hyperUA group had a lower DRE conversion rate compared to the normoUA group (HR: 0.81 [95% CI: 0.69-0.96]), while the hypoUA group had a higher conversion rate (HR: 1.88 [95% CI: 1.38-2.55]). CONCLUSIONS: Serum UA levels have the potential to serve as a biomarker for identifying patients at risk of DRE, indicating a potential avenue for novel therapeutic strategies aimed at preventing DRE conversion.

Inter hospital external validation of interpretable machine learning based triage score for the emergency department using common data model.

Emergency departments (ED) are complex, triage is a main task in the ED to prioritize patient with limited medical resources who need them most. Machine learning (ML) based ED triage tool, Score for Emergency Risk Prediction (SERP), was previously developed using an interpretable ML framework with single center. We aimed to develop SERP with 3 Korean multicenter cohorts based on common data model (CDM) without data sharing and compare performance with inter-hospital validation design. This retrospective cohort study included all adult emergency visit patients of 3 hospitals in Korea from 2016 to 2017. We adopted CDM for the standardized multicenter research. The outcome of interest was 2-day mortality after the patients' ED visit. We developed each hospital SERP using interpretable ML framework and validated inter-hospital wisely. We accessed the performance of each hospital's score based on some metrics considering data imbalance strategy. The study population for each hospital included 87,670, 83,363 and 54,423 ED visits from 2016 to 2017. The 2-day mortality rate were 0.51%, 0.56% and 0.65%. Validation results showed accurate for inter hospital validation which has at least AUROC of 0.899 (0.858-0.940). We developed multicenter based Interpretable ML model using CDM for 2-day mortality prediction and executed Inter-hospital external validation which showed enough high accuracy.

Explainable multimodal prediction of treatment-resistance in patients with depression leveraging brain morphometry and natural language processing.

Although 20 % of patients with depression receiving treatment do not achieve remission, predicting treatment-resistant depression (TRD) remains challenging. In this study, we aimed to develop an explainable multimodal prediction model for TRD using structured electronic medical record data, brain morphometry, and natural language processing. In total, 247 patients with a new depressive episode were included. TRD-predictive models were developed based on the combination of following parameters: selected tabular dataset features, independent components-map weightings from brain T1-weighted magnetic resonance imaging (MRI), and topic probabilities from clinical notes. All models applied the extreme gradient boosting (XGBoost) algorithm via five-fold cross-validation. The model using all data sources showed the highest area under the receiver operating characteristic of 0.794, followed by models that used combined brain MRI and structured data, brain MRI and clinical notes, clinical notes and structured data, brain MRI only, structured data only, and clinical notes only (0.770, 0.762, 0.728, 0.703, 0.684, and 0.569, respectively). Classifications of TRD were driven by several predictors, such as previous exposure to antidepressants and antihypertensive medications, sensorimotor network, default mode network, and somatic symptoms. Our findings suggest that a combination of clinical data with neuroimaging and natural language processing variables improves the prediction of TRD.

Incidence of mental health diagnoses during the COVID-19 pandemic: a multinational network study.

AIMS: Population-wide restrictions during the COVID-19 pandemic may create barriers to mental health diagnosis. This study aims to examine changes in the number of incident cases and the incidence rates of mental health diagnoses during the COVID-19 pandemic. METHODS: By using electronic health records from France, Germany, Italy, South Korea and the UK and claims data from the US, this study conducted interrupted time-series analyses to compare the monthly incident cases and the incidence of depressive disorders, anxiety disorders, alcohol misuse or dependence, substance misuse or dependence, bipolar disorders, personality disorders and psychoses diagnoses before (January 2017 to February 2020) and after (April 2020 to the latest available date of each database [up to November 2021]) the introduction of COVID-related restrictions. RESULTS: A total of 629,712,954 individuals were enrolled across nine databases. Following the introduction of restrictions, an immediate decline was observed in the number of incident cases of all mental health diagnoses in the US (rate ratios (RRs) ranged from 0.005 to 0.677) and in the incidence of all conditions in France, Germany, Italy and the US (RRs ranged from 0.002 to 0.422). In the UK, significant reductions were only observed in common mental illnesses. The number of incident cases and the incidence began to return to or exceed pre-pandemic levels in most countries from mid-2020 through 2021. CONCLUSIONS: Healthcare providers should be prepared to deliver service adaptations to mitigate burdens directly or indirectly caused by delays in the diagnosis and treatment of mental health conditions.

Comparative estimation of the effects of antihypertensive medications on schizophrenia occurrence: a multinational observational cohort study.

BACKGROUND: The association between antihypertensive medication and schizophrenia has received increasing attention; however, evidence of the impact of antihypertensive medication on subsequent schizophrenia based on large-scale observational studies is limited. We aimed to compare the schizophrenia risk in large claims-based US and Korea cohort of patients with hypertension using angiotensin-converting enzyme (ACE) inhibitors versus those using angiotensin receptor blockers (ARBs) or thiazide diuretics. METHODS: Adults aged 18 years who were newly diagnosed with hypertension and received ACE inhibitors, ARBs, or thiazide diuretics as first-line antihypertensive medications were included. The study population was sub-grouped based on age (> 45 years). The comparison groups were matched using a large-scale propensity score (PS)-matching algorithm. The primary endpoint was incidence of schizophrenia. RESULTS: 5,907,522; 2,923,423; and 1,971,549 patients used ACE inhibitors, ARBs, and thiazide diuretics, respectively. After PS matching, the risk of schizophrenia was not significantly different among the groups (ACE inhibitor vs. ARB: summary hazard ratio [HR] 1.15 [95% confidence interval, CI, 0.99-1.33]; ACE inhibitor vs. thiazide diuretics: summary HR 0.91 [95% CI, 0.78-1.07]). In the older subgroup, there was no significant difference between ACE inhibitors and thiazide diuretics (summary HR, 0.91 [95% CI, 0.71-1.16]). The risk for schizophrenia was significantly higher in the ACE inhibitor group than in the ARB group (summary HR, 1.23 [95% CI, 1.05-1.43]). CONCLUSIONS: The risk of schizophrenia was not significantly different between the ACE inhibitor vs. ARB and ACE inhibitor vs. thiazide diuretic groups. Further investigations are needed to determine the risk of schizophrenia associated with antihypertensive drugs, especially in people aged > 45 years.

Towards global model generalizability: independent cross-site feature evaluation for patient-level risk prediction models using the OHDSI network.

BACKGROUND: Predictive models show promise in healthcare, but their successful deployment is challenging due to limited generalizability. Current external validation often focuses on model performance with restricted feature use from the original training data, lacking insights into their suitability at external sites. Our study introduces an innovative methodology for evaluating features during both the development phase and the validation, focusing on creating and validating predictive models for post-surgery patient outcomes with improved generalizability. METHODS: Electronic health records (EHRs) from 4 countries (United States, United Kingdom, Finland, and Korea) were mapped to the OMOP Common Data Model (CDM), 2008-2019. Machine learning (ML) models were developed to predict post-surgery prolonged opioid use (POU) risks using data collected 6 months before surgery. Both local and cross-site feature selection methods were applied in the development and external validation datasets. Models were developed using Observational Health Data Sciences and Informatics (OHDSI) tools and validated on separate patient cohorts. RESULTS: Model development included 41 929 patients, 14.6% with POU. The external validation included 31 932 (UK), 23 100 (US), 7295 (Korea), and 3934 (Finland) patients with POU of 44.2%, 22.0%, 15.8%, and 21.8%, respectively. The top-performing model, Lasso logistic regression, achieved an area under the receiver operating characteristic curve (AUROC) of 0.75 during local validation and 0.69 (SD = 0.02) (averaged) in external validation. Models trained with cross-site feature selection significantly outperformed those using only features from the development site through external validation (P < .05). CONCLUSIONS: Using EHRs across four countries mapped to the OMOP CDM, we developed generalizable predictive models for POU. Our approach demonstrates the significant impact of cross-site feature selection in improving model performance, underscoring the importance of incorporating diverse feature sets from various clinical settings to enhance the generalizability and utility of predictive healthcare models.

Real-world treatment trajectories of adults with newly diagnosed asthma or COPD.

BACKGROUND: There is a lack of knowledge on how patients with asthma or chronic obstructive pulmonary disease (COPD) are globally treated in the real world, especially with regard to the initial pharmacological treatment of newly diagnosed patients and the different treatment trajectories. This knowledge is important to monitor and improve clinical practice. METHODS: This retrospective cohort study aims to characterise treatments using data from four claims (drug dispensing) and four electronic health record (EHR; drug prescriptions) databases across six countries and three continents, encompassing 1.3 million patients with asthma or COPD. We analysed treatment trajectories at drug class level from first diagnosis and visualised these in sunburst plots. RESULTS: In four countries (USA, UK, Spain and the Netherlands), most adults with asthma initiate treatment with short-acting ß2 agonists monotherapy (20.8%-47.4% of first-line treatments). For COPD, the most frequent first-line treatment varies by country. The largest percentages of untreated patients (for asthma and COPD) were found in claims databases (14.5%-33.2% for asthma and 27.0%-52.2% for COPD) from the USA as compared with EHR databases (6.9%-15.2% for asthma and 4.4%-17.5% for COPD) from European countries. The treatment trajectories showed step-up as well as step-down in treatments. CONCLUSION: Real-world data from claims and EHRs indicate that first-line treatments of asthma and COPD vary widely across countries. We found evidence of a stepwise approach in the pharmacological treatment of asthma and COPD, suggesting that treatments may be tailored to patients' needs.

RHEA: Real-World Observational Health Data Exploration Application.

We developed a standardized framework named RHEA to represent longitudinal status of patient with cancer. RHEA generates a dashboard to visualize patients' data in the Observational Medical Outcomes Partnership-Common Data Model format. The generated dashboard consists of three main parts for providing the macroscopic characteristics of the patient: 1) cohort-level visualization, 2) individual-level visualization and 3) cohort generation.

A De-Identification Model for Korean Clinical Notes: Using Deep Learning Models.

To extract information from free-text in clinical records due to the patient's protected health information PHI in the records pre-processing of de-identification is required. Therefore we aimed to identify PHI list and fine-tune the deep learning BERT model for developing de-identification model. The result of fine-tuning the model is strict F1 score of 0.924. Due to the convinced score the model can be used for the development of a de-identification model.

Prediction Models for Readmission Using Home Healthcare Notes and OMOP-CDM.

This study developed readmission prediction models using Home Healthcare (HHC) documents via natural language processing (NLP). An electronic health record of Ajou University Hospital was used to develop prediction models (A reference model using only structured data, and an NLP-enriched model with structured and unstructured data). Among 573 patients, 63 were readmitted to the hospital. Five topics were extracted from HHC documents and improved the model performance (AUROC 0.740).

Multi-Institutional Collaborative Research Using Ophthalmic Medical Image Data Standardized by Radiology Common Data Model (R-CDM).

Observational Medical Outcome Partners - Common Data Model (OMOP-CDM) is an international standard model for standardizing electronic medical record data. However, unstructured data such as medical image data which is beyond the scope of standardization by the current OMOP-CDM is difficult to be used in multi-institutional collaborative research. Therefore, we developed the Radiology-CDM (R-CDM) which standardizes medical imaging data. As a proof of concept, 737,500 Optical Coherence Tomography (OCT) data from two tertiary hospitals in South Korea is standardized in the form of R-CDM. The relationship between chronic disease and retinal thickness was analyzed by using the R-CDM. Central macular thickness and retinal nerve fiber layer (RNFL) thickness were significantly thinner in the patients with hypertension compared to the control cohort. It is meaningful in that multi-institutional collaborative research using medical image data and clinical data simultaneously can be conducted very efficiently.

Real-World Effectiveness of Statin Therapy in Adult Asthma.

BACKGROUND: Blood lipids affect airway inflammation in asthma. Although several studies have suggested anti-inflammatory effects of statins on asthmatic airways, further studies are needed to clarify the long-term effectiveness of statins on asthma control and whether they are an effective treatment option. OBJECTIVE: To evaluate the long-term effectiveness of statins in the chronic management of adult asthma in real-world practice. METHODS: Electronic medical record data spanning 28 years, collected from the Ajou University Medical Center in Korea, were used to conduct a retrospective study. Clinical outcomes were compared between patients with asthma who had maintained statin use (the statin group) and those not taking statins, whose blood lipid tests were always normal (the non-statin group). We performed propensity score matching and calculated hazard ratios with 95% CIs using the Cox proportional hazards model. Severe asthma exacerbation was the primary outcome; asthma exacerbation, asthma-related hospitalization, and new-onset type 2 diabetes mellitus and hypertension were secondary outcomes. RESULTS: After 1:1 propensity score matching, the statin and non-statin groups each included 545 adult patients with asthma. The risk of severe asthma exacerbations and asthma exacerbations was significantly lower in the statin group than in the non-statin group (hazard ratios [95% CI] = 0.57 [0.35-0.90] and 0.71 [0.52-0.96], respectively). There were no significant differences in the risk of asthma-related hospitalization or new-onset type 2 diabetes mellitus or hypertension between groups (0.76 [0.53-1.09], 2.33 [0.94-6.59], and 1.71 [0.95-3.17], respectively). CONCLUSION: Statin use is associated with a lower risk of asthma exacerbation, with better clinical outcomes in adult asthma.