[Knowledge Graph-Based Prediction of Potentially Inappropriate Medication].

Objective: To improve the accuracy of potentially inappropriate medication (PIM) prediction, a PIM prediction model that combines knowledge graph and machine learning was proposed. Methods: Firstly, based on Beers criteria 2019 and using the knowledge graph as the basic structure, a PIM knowledge representation framework with logical expression capabilities was constructed, and a PIM inference process was implemented from patient information nodes to PIM nodes. Secondly, a machine learning prediction model for each PIM label was established based on the classifier chain algorithm, to learn the potential feature associations from the data. Finally, based on a threshold of sample size, a portion of reasoning results from the knowledge graph was used as output labels on the classifier chain to enhance the reliability of the prediction results of low-frequency PIMs. Results: 11 741 prescriptions from 9 medical institutions in Chengdu were used to evaluate the effectiveness of the model. Experimental results show that the accuracy of the model for PIM quantity prediction is 98.10%, the F1 is 93.66%, the Hamming loss for PIM multi-label prediction is 0.06%, and the macroF1 is 66.09%, which has higher prediction accuracy than the existing models. Conclusion: The method proposed has better prediction performance for potentially inappropriate medication and significantly improves the recognition of low-frequency PIM labels.

Predicting Hepatotoxicity Associated with Low-Dose Methotrexate Using Machine Learning.

An accurate prediction of the hepatotoxicity associated with low-dose methotrexate can provide evidence for a reasonable treatment choice. This study aimed to develop a machine learning-based prediction model to predict hepatotoxicity associated with low-dose methotrexate and explore the associated risk factors. Eligible patients with immune system disorders, who received low-dose methotrexate at West China Hospital between 1 January 2018, and 31 December 2019, were enrolled. A retrospective review of the included patients was conducted. Risk factors were selected from multiple patient characteristics, including demographics, admissions, and treatments. Eight algorithms, including eXtreme Gradient Boosting (XGBoost), AdaBoost, CatBoost, Gradient Boosting Decision Tree (GBDT), Light Gradient Boosting Machine (LightGBM), Tree-based Pipeline Optimization Tool (TPOT), Random Forest (RF), and Artificial Neural Network (ANN), were used to establish the prediction model. A total of 782 patients were included, and hepatotoxicity was detected in 35.68% (279/782) of the patients. The Random Forest model with the best predictive capacity was chosen to establish the prediction model (receiver operating characteristic curve 0.97, accuracy 64.33%, precision 50.00%, recall 32.14%, and F1 39.13%). Among the 15 risk factors, the highest score was a body mass index of 0.237, followed by age (0.198), the number of drugs (0.151), and the number of comorbidities (0.144). These factors demonstrated their importance in predicting hepatotoxicity associated with low-dose methotrexate. Using machine learning, this novel study established a predictive model for low-dose methotrexate-related hepatotoxicity. The model can improve medication safety in patients taking methotrexate in clinical practice.

Developing a Warning Model of Potentially Inappropriate Medications in Older Chinese Outpatients in Tertiary Hospitals: A Machine-Learning Study.

Due to multiple comorbid illnesses, polypharmacy, and age-related changes in pharmacokinetics and pharmacodynamics in older adults, the prevalence of potentially inappropriate medications (PIMs) is high, which affects the quality of life of older adults. Building an effective warning model is necessary for the early identification of PIMs to prevent harm caused by medication in geriatric patients. The purpose of this study was to develop a machine learning-based model for the warning of PIMs in older Chinese outpatients. This retrospective study was conducted among geriatric outpatients in nine tertiary hospitals in Chengdu from January 2018 to December 2018. The Beers criteria 2019 were used to assess PIMs in geriatric outpatients. Three problem transformation methods were used to tackle the multilabel classification problem in prescriptions. After the division of patient prescriptions into the training and test sets (8:2), we adopted six widely used classification algorithms to conduct the classification task and assessed the discriminative performance by the accuracy, precision, recall, F1 scores, subset accuracy (ss Acc), and Hamming loss (hm) of each model. The results showed that among 11,741 older patient prescriptions, 5816 PIMs were identified in 4038 (34.39%) patient prescriptions. A total of 41 types of PIMs were identified in these prescriptions. The three-problem transformation methods included label power set (LP), classifier chains (CC), and binary relevance (BR). Six classification algorithms were used to establish the warning models, including Random Forest (RF), Light Gradient Boosting Machine (LightGBM), eXtreme Gradient Boosting (XGBoost), CatBoost, Deep Forest (DF), and TabNet. The CC + CatBoost model had the highest accuracy value (97.83%), recall value (89.34%), F1 value (90.69%), and ss Acc value (97.79%) with a good precision value (92.18%) and the lowest hm value (0.0006). Therefore, the CC + CatBoost model was selected to predict the occurrence of PIM in geriatric Chinese patients. This study's novelty establishes a warning model for PIMs in geriatric patients by using machine learning. With the popularity of electronic patient record systems, sophisticated computer algorithms can be implemented at the bedside to improve medication use safety in geriatric patients in the future.

Predicting adverse drug events in older inpatients: a machine learning study.

BACKGROUND: Building an effective prediction model of adverse drug events (ADE) is necessary to prevent harm caused by medication in older inpatients. AIM: This study aimed to develop a machine learning-based prediction model for the prediction of ADE and explore the risk factors associated with ADEs in older inpatients. METHOD: Data were from an observational, retrospective study that included 1800 older Chinese inpatients. After dividing the patients into training and test sets (8:2), seven machine learning models were used. Demographic, admission, and treatment clinical variables were considered for model development. The discriminative performance of the model by the area under the receiver operating characteristic curve (ROC) was evaluated. We also calculated the model's accuracy, precision, recall, and F1 scores. RESULTS: Among 1800 patients, 296 ADEs were detected in 234 (13.00%) patients. The main cause of ADEs was antineoplastic agents (55.74%). Seven algorithms, including eXtreme Gradient Boosting (XGBoost), AdaBoost, CatBoost, Gradient Boosting Decision Tree (GBDT), Light Gradient Boosting Machine (LightGBM), Tree-based Pipeline Optimization Tool (TPOT) and Random Forest (RF), were used to establish the prediction model. The Adaboost model was chosen with the best predictive ability (accuracy 88.06%, precision 68.57%, recall 48.21%, F1 52.75%, and AUC 0.91). Ten significant factors associated with ADEs were identified, including the number of true triggers (+), length of stay, doses per patient, age, number of admissions in the previous year, surgery, drugs per patient, number of medical diagnoses, antibacterial use, and gender. CONCLUSION: Using machine learning, this novel study establishes an ADE prediction model in older patients. The sophisticated computer algorithm can be implemented at the bedside to improve patient safety in clinical practice.

A pharmacovigilance study of association between proton pump inhibitor and dementia event based on FDA adverse event reporting system data.

Proton pump inhibitor (PPI) was widely used around the world. Studies suggested conflicting results between PPI treatment and dementia event. This study examined the association between six PPI agents and dementia event by mining the US FDA Adverse Event Reporting System (FAERS) database from 2004 to 2020. We employed proportional reporting ratio (PRR) and information element (IC) methods to detect the signals of dementia relevant to PPI. We also analyzed characteristics of PPI and positive control reports, compared dementia event between long- and short-duration of PPI treatment. Finally, we identified 2396 dementia cases with PPI treatment. We did not detect significant signal between PPI and dementia event: PRR = 0.98, 95%CI 0.94 to 1.02, IC = -0.03, 95%CI - 0.17 to 0.10, even in gastroesophageal reflux disease cases: PRR = 0.65, 95%CI 0.59 to 0.72, IC = -0.62, 95%CI - 0.97 to - 0.27. No significant differences of dementia event were detected between long- and short- duration groups, the OR (95%CI) of the 3 years, 5 years and 10 years comparison were 0.70 (0.48 to 1.02), 0.72 (0.45 to 1.15) and 1.65 (0.75 to 3.63), respectively. Based on the current FAERS data mining, we discovered no association between PPI use and dementia event, even in long-term PPI therapy case.

Validating the Chinese geriatric trigger tool and analyzing adverse drug event associated risk factors in elderly Chinese patients: A retrospective review.

OBJECTIVE: The aim was to evaluate the performance of the initial Chinese geriatric trigger tool to detect adverse drug events (ADEs) in Chinese older patients, to attempt to shorten this list for improving the efficiency of the trigger tool, and to study the incidence and characteristics of ADEs in this population. METHODS: A sample of 25 cases was randomly selected per half a month from eligible patients who aged 60 years and older, hospitalized more than 24 hours, and discharged or died between January 1, 2015 and December 31, 2017 in West China hospital. A two-stage retrospective chart review of the included inpatients were conducted. ADEs were detected using a list of 42 triggers previously selected by an expert panel by means of a Delphi method. The number of triggers identified and ADEs detected were recorded and the positive predictive value (PPV) of each trigger was calculated to select the most efficient triggers. Several variables were recorded, including age, sex, number of diseases, length of hospital stay and so on, to analyze the risk factor of ADEs. RESULTS: Among 1800 patients, 1646 positive triggers and 296 ADEs were detected in 234 (13.00%) patients. Older patients who were younger, had more medications, longer stays or more admission, and did not experience surgical operation more likely experienced ADEs. Triggers with PPV less than 5% were eliminated, which resulted in the upgraded version of Chinese geriatric trigger tool of 20 triggers with a PPV of 28.50%. This upgraded tool accounted for 99.66% of all ADEs detected. CONCLUSIONS: The upgraded version of Chinese geriatric trigger tool was an efficient tool for identifying ADEs in Chinese older patients. Future, the trigger tool could be incorporated into routine screen systems to provide real-time identification of ADEs, thereby enabling timely clinical interventions.

Adverse events identified by the global trigger tool at a university hospital: A retrospective medical record review.

OBJECTIVES: The purpose of this study was to describe the level, preventability and categories of adverse events (AEs) in Chinese geriatric patients identified by medical record review using the Global Trigger Tool. The applicability of the GTT was also assessed to explore possible modifications for trigger tools. METHODS: The study was conducted at a 4300-bed tertiary teaching hospital. Twenty randomly-selected medical records for patients over 60 were reviewed every 2 weeks from January 1 2015 to December 31st, 2015. We studied 480 medical records in total. Two trained specialists reviewed the presence of AEs using 43 triggers, and a physician reviewed and validated the findings. The outcome measures included the number of AEs per 1000 patient days, AEs per 100 admissions, the percentage of entries with at least 1 AE and AE categorisation. Also, we carried out a descriptive analysis of the suspected factors of AEs, such as age, gender, length of stay, surgery. RESULTS: A total of 610 AEs were detected in the 480 medical records reviewed, corresponding to 127 injuries per 100 admissions. The number of AEs per 1000 patient days was 22.43. AEs occurred at least once in 329 (68.54%) patients. The rate of care harms ranked highest of all AEs, followed by the rate of medication harms and surgical harms. Patients with a more extended hospital stay or surgery was more likely to experience AEs. However, there was a negative correlation between age and the rate of AEs. CONCLUSION: The Global Trigger Tool was a useful method for detecting the characteristics of AEs in geriatric patients in a Chinese tertiary teaching hospital. To improve patients' safety, this tool should be incorporated into routine screening systems.

Development of a trigger tool for the detection of adverse drug events in Chinese geriatric inpatients using the Delphi method.

Background The global trigger tool is a method of retrospective medical record review that identifies possible harm in hospitalized patients using "triggers". Elderly patients with multiple co-morbid illnesses are especially vulnerable to adverse drug events (ADEs) that have high prevalence rates. Objective The purpose of this study was to develop an appropriate trigger tool to detect ADEs in Chinese geriatric inpatients by combining a literature review with the Delphi method. Setting Chinese geriatric inpatients. Methods Two steps were used to develop the trigger tool. First, we conducted a comprehensive literature review for existing ADE triggers (adult or elderly) to form the initial triggers for the Delphi process. Second, a group of clinical experts, including physicians, clinical pharmacists and nurses, was established to score candidate triggers for utility according to the usefulness and feasibility of implementing triggers in clinical practice. Main outcome measures The frequency of the full mark, arithmetic mean and coefficient of variation of each trigger. Results An initial set of 51 triggers was selected by literature review for evaluation. The group of experts was composed of 18 clinical experts: 13 physicians, 4 clinical pharmacists, and 1 nurse. Based on the two-phase Delphi process, 42 triggers in five categories (laboratory index, plasma concentration, antidotes, clinical symptoms and intervention) were retained. Conclusion The 42-trigger tool was developed to identify ADEs in Chinese geriatric inpatients. A pilot study that tests the list of triggers to identify ADEs in Chinese geriatric inpatients is the next step for establishing a specific trigger tool for Chinese geriatric inpatients.