論文 - 木崎　速人

A scalable natural language processing framework for drug repurposing in chemotherapy-induced adverse events from clinical narrative records

Tsuchiya M., Inoue M., Kawazoe Y., Shimamoto K., Seki T., Imai S., Kizaki H., Shinohara E., Yada S., Wakamiya S., Aramaki E., Hori S.

European Journal of Cancer 232   116157 2026年01月

ISSN  09598049

　概要を見る

Background Preventing chemotherapy-related adverse events (AEs) remains an unmet clinical challenge. Preclinical studies have suggested protective effects of several existing agents, but translation into human evidence has been limited. We aimed to establish proof of concept (PoC) for drug repurposing by applying a natural language processing (NLP)-based framework to electronic health record (EHR) narratives, thereby bridging preclinical findings with clinical validation. Methods We retrospectively analyzed 56,326 patients with cancer treated at the University of Tokyo Hospital (2004–2023). A transformer-based NLP model extracted symptomatic AEs from clinical notes. Candidate preventive drugs identified from preclinical evidence were assessed using propensity score matching and Cox proportional hazards models. We evaluated angiotensin II receptor blockers (ARBs) for fluoropyrimidine-induced oral mucositis and ramelteon for platinum-induced peripheral neuropathy, with laxatives serving as a negative control. Results NLP demonstrated high accuracy (precision 0.81–0.83; recall 0.95–0.97). After matching, ARB co-administration was significantly associated with reduced mucositis incidence (hazard ratio [HR] 0.58, 95 % confidence interval [CI] 0.44–0.77; P < 0.001), representing a clinical PoC consistent with mechanistic preclinical data. Ramelteon showed an exploratory protective signal against neuropathy (HR 0.60, 95 % CI:0.38–0.93; P = 0.024). No preventive association was observed for laxatives. Conclusions This study introduces a scalable NLP-epidemiology framework for non-invasive, real-world validation of drug repurposing candidates. The ARB finding provides human-level PoC evidence supporting prospective clinical testing, while the ramelteon signal warrants further exploration. Our approach demonstrates how EHR narratives can operationalize translational research, prioritizing safe, accessible agents for improving the tolerability of cancer treatment.

• Access to Document (DOI)

A real-world pharmacovigilance study of adverse events associated with esketamine: disproportionality analysis and detection of potential drug-drug interaction signals

Pisanu C., Imai S., Tsuchiya M., Inoue M., Ikegami K., Zammarchi G., Kizaki H., Hori S.

European Journal of Clinical Pharmacology 82 （ 1 ） 13 2026年01月

ISSN  00316970

　概要を見る

Purpose: We conducted a comprehensive analysis of esketamine-related adverse events (AE) on the FDA Adverse Event Reporting System (FAERS) database, taking into account for the first time drug-drug interaction signals. Methods: We conducted a retrospective case/non-case study of esketamine-related AEs reported in the FAERS database up to the last quarter (Q4) of 2024. Potential signals were detected using the reporting odds ratio (ROR) and confidence intervals (CI), while drug-drug interactions were studied using different metrics such as lift, conviction and the combination risk ratio detection algorithm. An analysis of sex differences was also performed using the relative ROR and CI. Results: The analysis of 7,790 reports in which esketamine was a primary or secondary suspect identified potential safety signals for 173 AEs. Novel signals include homicidal ideation (ROR = 5.30, 95% CI: 2.38-11.82) and substance use disorder (ROR = 6.12, 95% CI: 2.54-14.73). Women showed a longer time to onset than men (p = 0.003). In addition, we detected sex differences in 23 AEs, seven of which were more likely to be reported in women, while 16 in men. Among these, four were significant exclusively in women (oxygen saturation decreased, abnormal behaviour, unresponsive to stimuli and aggression) and two in men (vision blurred and bradycardia). Potential signals of additive and multiplicative drug-drug interactions were detected for antidepressants (venlafaxine for“dizziness” and bupropion for “agitation”) and antipsychotics (risperidone for “vertigo”). Conclusions: Our results increase knowledge on potential risks related to esketamine AEs and potential drug-drug interaction signals in a real-world setting.

• Access to Document (DOI)

Validation of diagnostic coding for chronic kidney disease using a Japanese hospital-based database

Otani-Kono M., Imai S., Tsuchiya M., Mitsuboshi S., Kizaki H., Hori S.

Clinical and Experimental Nephrology 2026年

ISSN  13421751

　概要を見る

Background: The validity of the International Classification of Diseases 10th Revision (ICD-10) codes for chronic kidney disease (CKD) in Japan have not been evaluated. In this study we evaluated the validity of CKD-related ICD-10 codes in Japan. Methods: This retrospective cohort study used the JMDC hospital-based database, which comprises claims and laboratory data from over 1,000 medical institutions in Japan. Patients who underwent two serum creatinine measurements between April 2014 and August 2022 were identified; the second measurement was obtained 90–365 days after the first. The estimated glomerular filtration rate (eGFR) was calculated. CKD was defined by the Kidney Disease: Improving Global Outcomes (KDIGO) criteria: an eGFR < 60 mL/min/1.73 m² for both measurements. Markers of kidney damage such as albuminuria were unavailable. Patients assigned CKD-related ICD-10 codes (N183, N184, N185, N189, N19, E102, E112, E142, and I120) within 365 days of the initial serum creatinine measurement were identified. Subsequently, the positive predictive value (PPV), sensitivity, specificity, and negative predictive value (NPV) were calculated. Results: Among the 618,208 included patients, 59,139 were assigned CKD-related ICD-10 codes and 172,657 met the KDIGO criteria. The overall PPV, sensitivity, specificity, and NPV were 57.9%, 19.8%, 94.4%, and 75.2%, respectively. In contrast, the PPVs of N183, N184, N185, and N189 exceeded 80%. Conclusions: In Japan, the PPV of the ICD-10 codes for CKD was 57.9%, marginally lower than values reported in the United States (86.1%) and Canada (60.1%). Several ICD-10 codes may be useful for identifying patients with CKD, despite their limited sensitivity.

• Access to Document (DOI)

A questionnaire survey of healthcare access and dietary habits in a rural Japanese community: implications for potential community pharmacy roles.

Kizaki H, Tsukamoto M, Yamada M, Ito S, Sasaki M, Sakamoto K, Ikeda Y, Miyamoto K, Iino H, Hori S

Journal of pharmaceutical health care and sciences 12 （ 1 ） 7 2025年12月

ISSN  2055-0294

• Access to Document (DOI)

Analysis of factors affecting difficulty in handling oral medicine using electronic medication notebook-based personal health records

Shimizu Y., Tsuchiya M., Asano M., Imai S., Kizaki H., Ito Y., Tsuchiya M., Kuriyama R., Yoshida N., Shimada M., Sando T., Ishijima T., Hori S.

Scientific Reports 15 （ 1 ） 26867 2025年12月

　概要を見る

Tablets and capsules are widely used forms of oral medication, but some patients experience difficulty handling them, which can reduce medication adherence and affect health outcomes. This study aimed to identify factors contributing to perceived handling difficulty, using data from harmo^®, a nationwide electronic medication notebook system. A questionnaire was distributed to adult users who had been prescribed oral medications, and the responses were linked with personal health records to analyze medication characteristics and patient backgrounds. Among the 1,230 respondents, 24% reported difficulty with small tablets or capsules. A size threshold was identified: a combined long and short diameter of 13.3 mm or less was most associated with handling problems (ROC-AUC = 0.834). Binomial logistic regression analysis revealed that difficulty in applying force with the hands (OR = 2.64), prescription of small tablets or capsules (OR = 2.52), and medical histories of hypertension (OR = 1.69) and osteoporosis (OR = 4.99) were significantly associated with reported difficulty. These results suggest that both the physical characteristics of formulations and individual patient factors influence medication usability. Our results provide evidence to inform more patient-centered approaches to oral formulation design and prescribing practices, ultimately supporting better adherence and medication safety.

• Access to Document (DOI)

A patient-centered approach to developing and validating a natural language processing model for extracting patient-reported symptoms

Watabe S., Yanagisawa Y., Sayama K., Yokoyama S., Someya M., Taniguchi R., Yada S., Aramaki E., Kizaki H., Tsuchiya M., Imai S., Hori S.

Scientific Reports 15 （ 1 ） 27652 2025年12月

　概要を見る

Patient-reported symptoms provide valuable insights into patient experiences and can enhance healthcare quality; however, effectively capturing them remains challenging. Although natural language processing (NLP) models have been developed to extract adverse events and symptoms from medical records written by healthcare professionals, limited studies have focused on models designed for patient-generated narratives. This study developed an NLP model to extract patient-reported symptoms from pharmaceutical care records and validated its effectiveness in analyzing diverse patient-generated narratives. The target dataset comprised “Subjective” sections of pharmaceutical care records created by community pharmacists for patients prescribed anticancer drugs. Two annotation guidelines were applied to develop robust ground-truth data, which was used to develop and evaluate a new transformer-based named entity recognition model. Model performance was compared with that of an existing tool for Japanese clinical texts and tested on external patient-generated blog data to evaluate generalizability. The newly developed BERT-CRF model significantly outperformed the existing model, achieving an F1 score > 0.8 on pharmaceutical care records and extracting > 98% of physical symptom entries from patient blogs, with a 20% improvement over the existing tool. These findings highlight the importance of fine-tuning models using patient-specific narrative data to capture nuanced and colloquial symptom expressions.

• Access to Document (DOI)

Development of a novel person-centered question prompt list to talk with your pharmacists in Japanese community pharmacies: focus group and Delphi method

Hayakawa M., Kizaki H., Yanagisawa Y., Suzuki N., Kagawa Y., Sayama K., Imai S., Hori S.

Journal of Pharmaceutical Health Care and Sciences 11 （ 1 ） 87 2025年12月

ISSN  2055-0294

　概要を見る

Background: The concept of shifting from a patient-centered to person-centered approach in pharmacy practice has been proposed. Patient-centered care focuses on the individual patient, whereas person-centered care focuses on a holistic understanding of the individual, taking not only medications but also daily life into account. This shift requires effective communication between patients and pharmacists. Although question prompt lists (QPLs) exist for patients and doctors, the same is not available for pharmacists focused on a person-centered approach. This study aimed to develop a question prompt list to talk with your pharmacists (QPLP) focusing on a person-centered approach to facilitate communication with patients. Methods: This study aimed to develop the QPLP in three steps. In the first step, six medical pharmaceutical researchers with pharmacist qualifications and two patient researchers prepared an initial draft of the QPLP, referencing existing QPLs used by patients to prepare questions for doctors before consultation. A focus group interview was then conducted with eight patients, and a QPLP drafted. Finally, a modified Delphi method was used to evaluate and collect opinions, and the QPLP was finalized. Results: A QPLP comprising 16 questions was developed with patient participation in a three-step process. The content was categorized into five sections: “Medicines,” “How to Take Medicines,” “Daily Life,” “Treatment,” and “Consumer Health Information.” The questions covered concerns regarding medicines, difficulties in using medicines, issues in daily life during treatment, treatment-related problems, authenticity of health information, and community health and exercise information. Conclusions: In this study, we developed a novel QPLP to enhance communication between patients and pharmacists, focusing on a person-centered approach with active patient involvement. Future studies should investigate the contribution of the developed QPLP in improving patient–pharmacist communication. These questions may encourage both healthy people and patients to seek health advice from pharmacists.

• Access to Document (DOI)

Assessing the utility and challenges for implementation of a risk prediction system: a usability study with hospital pharmacists

Ikegami K., Tsuchiya M., Kizaki H., Imai S., Yasumuro O., Sato C., Fujita Y., Funakoshi R., Hori S.

Journal of Pharmaceutical Health Care and Sciences 11 （ 1 ） 96 2025年12月

ISSN  2055-0294

　概要を見る

Background: The clinical implementation of prediction models can face important barriers, particularly regarding user-friendliness and interpretability for healthcare professionals. We recently developed and externally validated a risk prediction model for denosumab-induced hypocalcemia. The present study aimed to evaluate the model’s utility and identify challenges for its clinical implementation through pilot testing conducted by hospital pharmacists. Methods: A paper-format prediction model was distributed to pharmacists at Kameda General Hospital, Miyagi Cancer Center, and Gunma Prefectural Cancer Center. Participants trialed the model outside their routine workflow by applying it to data from patients scheduled to receive their first dose of denosumab. A subsequent questionnaire survey, available in paper and electronic formats, was conducted to gather feedback on the model’s utility and limitations. Results: A total of 49 responses were obtained, predominantly from pharmacists in their 20 s and 30 s with diverse professional responsibilities. The model was positively evaluated in terms of its clear target population, predicted outcome (47/49, 95.9%), and simple, straightforward nature (47/49, 95.9%). However, some participants provided neutral feedback on its ease of use (10/49, 20.4%). While its potential as an auxiliary tool for risk prediction was acknowledged, there were also some neutral views on its practical utility. One concern was the inconvenience of implementing a paper-format tool in clinical environments that primarily operate on electronic health records. Conclusions: The paper-format prediction model was positively evaluated by frontline pharmacists, especially for its clear and straightforward nature. As anticipated, however, limitations such as manual data input and paper-based format were identified by some participants. Integration into electronic health records and broader clinical validation will be necessary to advance the clinical application of this prediction model and ensure real-world applicability.

• Access to Document (DOI)

Survey of perioperative treatment in muscle-invasive bladder cancer using Japanese hospital-based claims database.

Yamazawa A, Tsuchiya M, Imai S, Ikegami K, Kizaki H, Hori S

BMC urology 25 （ 1 ） 309 2025年11月

• Access to Document (DOI)

Natural Language Processing-Based Approach to Detect Common Adverse Events of Anticancer Agents from Unstructured Clinical Notes: A Time-to-Event Analysis.

Tsuchiya M, Shimamoto K, Kawazoe Y, Shinohara E, Yada S, Wakamiya S, Imai S, Kizaki H, Hori S, Aramaki E

Studies in health technology and informatics 329   703 - 707 2025年08月

ISSN  0926-9630

　概要を見る

This study assessed the effectiveness of natural language processing (NLP) in detecting adverse events (AEs) from anticancer agents by analyzing data from over 39,000 cancer patients. A specialized machine learning model identified known AEs from anticancer agents like capecitabine, oxaliplatin, and anthracyclines, revealing a significantly higher incidence in the treatment groups compared to non-users. While the NLP approach effectively detected most symptomatic AEs requiring manual review, it struggled with rarely documented conditions and commonly used clinical terms. Overall, the method shows promise for automated AE detection in medical records, particularly for symptoms without laboratory markers or diagnosis codes.

• Access to Document (DOI)

Development of an Automated Classification System for Medication-Related Incident Factors: A Practical Approach to Enhancing Patient Safety Management.

Takamatsu Y, Ebara S, Kizaki H, Watabe S, Imai S, Yada S, Aramaki E, Yasumuro O, Funakoshi R, Hori S

Studies in health technology and informatics 329   758 - 763 2025年08月

ISSN  0926-9630

　概要を見る

Analyzing medication-related incident reports is crucial for patient safety; however, systematically extracting the underlying factors contributing to incident occurrence remains challenging. We developed a multi-label classifier that automatically identified incident factors from 1,212 drug-related incident reports using the Bidirectional Encoder Representations from Transformers and its derivatives. Based on the P-mSHELL model, a comprehensive framework for incident factor analysis, we established seven distinct factor categories and evaluated various pre-trained models through five-fold cross-validation. Almost all models achieved macro F1 scores exceeding 0.6, with the lightweight A Lite BERT model showing comparable performance to BERT. This study demonstrates the practical feasibility of natural language processing techniques for systematic incident factor analysis, supporting enhanced patient safety management.

• Access to Document (DOI)

Elucidating Celecoxib's Preventive Effect in Capecitabine-Induced Hand-Foot Syndrome Using Medical Natural Language Processing.

Tsuchiya M, Kawazoe Y, Shimamoto K, Seki T, Imai S, Kizaki H, Shinohara E, Yada S, Wakamiya S, Aramaki E, Hori S

JCO clinical cancer informatics 9   e2500096 2025年08月

　概要を見る

PURPOSECapecitabine, an oral anticancer agent, frequently causes hand-foot syndrome (HFS), affecting patients' quality of life and treatment adherence. However, such symptomatic toxicities are often difficult to detect in structured electronic health record (EHR) data. This study primarily aimed to validate a natural language processing (NLP) approach to identifying capecitabine-induced HFS from unstructured clinical text and demonstrate its application in evaluating medication-associated adverse event trends in real-world settings.METHODSWe conducted a retrospective cohort study using EHRs from the University of Tokyo Hospital (2004-2021). HFS cases were identified using the MedNERN-CR-JA NLP model. After propensity score matching, we compared capecitabine users with and without celecoxib and assessed time to HFS onset using Cox proportional hazards models. NLP-based HFS detection was validated through manual annotation of aggregated clinical notes. Negative control and sensitivity analyses ensured robustness.RESULTSAmong 44,502 patients with cancer, 669 capecitabine users were analyzed. HFS incidence was significantly higher among capecitabine users (hazard ratio [HR], 1.93 [95% CI, 1.48 to 2.52]; P <.001) compared with nonusers. Celecoxib use showed a suggestive association with a reduced HFS risk (HR, 0.51 [95% CI, 0.24 to 1.07]; P =.073). The NLP model demonstrated high accuracy in identifying HFS, achieving a precision of 0.875, recall of 1.000, and F<inf>1</inf> score of 0.933, based on manual annotation of patient-level clinical notes. Outcome trends remained consistent when using manually annotated HFS case labels instead of NLP-detected events, supporting the method's robustness.CONCLUSIONThese findings demonstrate the effectiveness of NLP in detecting HFS from real-world clinical records. The application to celecoxib-HFS detection illustrates the potential utility of this approach for retrospective safety analysis. Further work is needed to evaluate generalizability across diverse clinical settings.

• Access to Document (DOI)

オンラインでの健康相談の役割と課題：消費者視点からの分析

大上 愛由，木﨑 速人，今井 俊吾，堀 里子

医療薬学 51 （ 6 ） 335 - 344 2025年06月

研究論文（学術雑誌）, 共著, 査読有り

オンラインでの健康相談の役割と課題　消費者視点からの分析

大上 愛由, 木崎 速人, 今井 俊吾, 堀 里子

医療薬学 （(一社)日本医療薬学会）  51 （ 6 ） 335 - 344 2025年06月

ISSN  1346-342X

医療機関に所属する臨床研究コーディネーターのレジリエンス要因

野呂 幾久子, 飯岡 緒美, 堀 里子, 中野 重行, 木﨑 速人, 石崎 雅人

臨床薬理 （(一社)日本臨床薬理学会）  56 （ 2 ） 97 - 108 2025年05月

研究論文（学術雑誌）, 共著, 査読有り,  ISSN  03881601

　概要を見る

It has been observed that CRCs, pivotal figures in clinical research, often grapple with a substantial mental burden. This study, which investigates the resilience factors of CRCs, aims to provide practical suggestions for their education and resilience development, empowering them with actionable knowledge. Ten CRCs, who had been working as CRCs for many years, were interviewed individually. They were asked about their challenging experiences while working as CRCs, how they successfully adapted to those challenges, and the factors involved in the process. We then inductively coded the factors involved in successfully adapting to challenges as the analysis theme. The qualitative analysis reveals that the factors of resilience of CRCs were classified into four groups(: 1)individual characteristics and ways of thinking,(2)meaning, goals, and motivation,(3)ways of dealing with work, and(4)support from those around them. Among the factors, the following were considered to be particularly important for CRCs: meaning, continuing attempt to find meaning in one's work, even in the face of adversity; metacognition, handling reflection on one's own and others' cognitive activities from a bird's-eye view; and balancing professional and patient perspectives that maintains the rigor of carrying out clinical research as a medical professional with empathy for patients' feelings.

• Access to Document (DOI)

Construction of Personalized Predictive Models for Missed Medication Doses Using Wearable Device Data: Prospective Observational Study

Iino H., Kizaki H., Imai S., Hori S.

Jmir Formative Research 9   e72113 2025年05月

研究論文（学術雑誌）, 共著, 査読有り

　概要を見る

Background: Declining medication adherence remains a critical health care issue, often assessed through unreliable self-reporting methods. Wearable devices (WDs) may offer an objective means to improve adherence monitoring by continuously recording physiological and activity data. Objective: This study aimed to develop and internally validate personalized predictive models, utilizing objective physiological and activity data from WDs, for identifying missed medication doses. Methods: A 30-day prospective observational study was conducted with 8 participants who wore Apple Watches and used a dedicated iOS app. The app collected demographics, medication details, psychological factors, mealtimes, and daily missed dose events. WDs recorded time-series data (ie, activity, heart rate, sleep) at 3-minute intervals. Data were aggregated into 1-hour segments, and lag (6 and 12 h) as well as rolling (24 h) features were generated. Light Gradient Boosting Machine models were constructed for each individual’s dosing regimen if the missed dose rate exceeded 20%. Two modeling approaches were compared: a group cross-validation (CV) model that grouped data by day to avoid data leakage from rolling features, and a nonrolling feature model that excluded rolling features and used leave-one-out CV. F<inf>1</inf>-score, accuracy, recall, and precision were assessed between the 2 models. Results: Of the 15 enrolled participants, 8 completed the study; 4 had a missed dose rate above 20%. In these 4 individuals, the group CV model achieved F<inf>1</inf>-scores of 0.435 to 0.902, with accuracy ranging from 0.711 to 0.911, recall from 0.278 to 0.822, and a precision of 1.000 for the most robust regimens. The nonrolling feature model yielded F<inf>1</inf>-scores of 0.667 to 0.910, with accuracy ranging from 0.800 to 0.906, recall from 0.500 to 0.835, and a precision of 1.000. Morning dosing regimens generally showed higher predictive performance than evening or afternoon. Time-series features, particularly those reflecting 6-, 12-, and 24-hour patterns, emerged as key predictors, indicating that physiological and lifestyle variations prior to dosing strongly influenced missed dose events. Conclusions: Personalized predictive models using WD-derived data demonstrated high precision for detecting missed medication doses, especially in morning and evening regimens. These findings underscore the feasibility of employing continuous, objective physiological and activity data from WDs to forecast nonadherence events. Although the sample size was limited, restricting the generalizability of the results, this study demonstrates the potential of WD-based personalized prediction of medication adherence. Future work should involve larger populations for external validation, strategies to improve recall, especially for clinically critical medications, and careful consideration of real-world implementation challenges.

• Access to Document (DOI)

Incidence Status and Factors Associated With Tyrosine Kinase Inhibitor-Induced Hypertension in Patients With Renal Cell Carcinoma

Nakanishi S., Ikegami K., Imai S., Kizaki H., Hori S.

Cancer Reports 8 （ 5 ） e70219 2025年05月

研究論文（学術雑誌）, 共著, 査読有り

　概要を見る

Background: Although hypertension is a common side effect of tyrosine kinase inhibitor (TKI) treatment for renal cell carcinoma (RCC), there is limited evidence regarding its occurrence and related risk factors. Preliminary studies suggest that proton pump inhibitors (PPIs) may mitigate the risk of TKI-induced hypertension; however, their clinical effectiveness remains unclear. Aims: In this study, we examined the prevalence of TKI-induced hypertension and the patterns of antihypertensive prescriptions among patients with RCC in Japan. Additionally, we investigated factors associated with TKI-induced hypertension to assess the potential impact of PPIs. Methods and Results: Data from patients diagnosed with RCC who were prescribed TKIs between April 2008 and July 2021 were retrospectively gathered from a Japanese administrative database. TKI-induced hypertension was detected following the diagnosis of hypertension and subsequently the prescription of an antihypertensive agent during TKI therapy. The prescription details for antihypertensive agents were organized in a tabular format. Cox proportional hazards regression analysis was conducted to examine factors contributing to TKI-induced hypertension. Among the 225 patients analyzed, 36.4% experienced hypertension, and calcium channel blockers were the most prescribed antihypertensive agents. Pre-existing hypertension was identified as a risk factor for TKI-induced hypertension, while the concurrent use of PPIs did not show a tendency to reduce the risk of TKI-induced hypertension. Conclusions: These findings indicate the importance of blood pressure management in patients with elevated baseline blood pressure.

• Access to Document (DOI)

Accuracy of Diagnostic Coding for Acute Kidney Injury in Japan-Analysis of a Japanese Hospital-Based Database.

Mitsuboshi S, Imai S, Tsuchiya M, Kizaki H, Hori S

Pharmacoepidemiology and drug safety 34 （ 4 ） e70146 2025年04月

研究論文（学術雑誌）, 共著, 査読有り,  ISSN  1053-8569

　概要を見る

Purpose: To evaluate the accuracy of diagnostic coding for acute kidney injury (AKI) in Japan. Methods: The data analyzed were obtained from the JMDC hospital-based administrative claims database from cases registered between April 2014 and August 2022. Only patients who underwent serum creatinine measurements two or more times with intervals of 7 days or less were eligible for inclusion. AKIs were identified by International Classification of Diseases 10th Revision (ICD-10) codes N14 and N17. These were assessed according to the Kidney Disease: Improving Global Outcomes (KDIGO) criteria. Results: A total of 467 019 patients (median age, 74 [range, 20–99] years; male, 50.9%) were eligible for inclusion. Among these patients, 1849 (0.4%) were assigned ICD-10 codes for AKI. Among these 1849 patients, the code was assigned within 7 days of the occurrence of AKI (as defined by the KDIGO criteria) in 212 patients, within 14 days in 294 patients, and within 30 days in 386 patients. The positive predictive values and 95% confidence intervals of the ICD-10 code for AKI at these timepoints were as follows: within 7 days, 11.5% (10.1%–13.0%); within 14 days, 15.9% (14.3%–17.6%); and within 30 days, 20.9% (19.1%–22.8%). Conclusions: The ICD-10 codes for AKI showed poor positive predictive values for AKI as defined by the KDIGO criteria, suggesting that it may be difficult to identify AKI using ICD-10 codes alone in the Japanese context.

• Access to Document (DOI)

Improving Systematic Review Updates With Natural Language Processing Through Abstract Component Classification and Selection: Algorithm Development and Validation.

Hasegawa T, Kizaki H, Ikegami K, Imai S, Yanagisawa Y, Yada S, Aramaki E, Hori S

JMIR medical informatics 13   e65371 2025年03月

研究論文（学術雑誌）, 共著, 査読有り

　概要を見る

Background: A challenge in updating systematic reviews is the workload in screening the articles. Many screening models using natural language processing technology have been implemented to scrutinize articles based on titles and abstracts. While these approaches show promise, traditional models typically treat abstracts as uniform text. We hypothesize that selective training on specific abstract components could enhance model performance for systematic review screening. Objective: We evaluated the efficacy of a novel screening model that selects specific components from abstracts to improve performance and developed an automatic systematic review update model using an abstract component classifier to categorize abstracts based on their components. Methods: A screening model was created based on the included and excluded articles in the existing systematic review and used as the scheme for the automatic update of the systematic review. A prior publication was selected for the systematic review, and articles included or excluded in the articles screening process were used as training data. The titles and abstracts were classified into 5 categories (Title, Introduction, Methods, Results, and Conclusion). Thirty-one component-composition datasets were created by combining 5 component datasets. We implemented 31 screening models using the component-composition datasets and compared their performances. Comparisons were conducted using 3 pretrained models: Bidirectional Encoder Representations from Transformer (BERT), BioLinkBERT, and BioM- Efficiently Learning an Encoder that Classifies Token Replacements Accurately (ELECTRA). Moreover, to automate the component selection of abstracts, we developed the Abstract Component Classifier Model and created component datasets using this classifier model classification. Using the component datasets classified using the Abstract Component Classifier Model, we created 10 component-composition datasets used by the top 10 screening models with the highest performance when implementing screening models using the component datasets that were classified manually. Ten screening models were implemented using these datasets, and their performances were compared with those of models developed using manually classified component-composition datasets. The primary evaluation metric was the F10-Score weighted by the recall. Results: A total of 256 included articles and 1261 excluded articles were extracted from the selected systematic review. In the screening models implemented using manually classified datasets, the performance of some surpassed that of models trained on all components (BERT: 9 models, BioLinkBERT: 6 models, and BioM-ELECTRA: 21 models). In models implemented using datasets classified by the Abstract Component Classifier Model, the performances of some models (BERT: 7 models and BioM-ELECTRA: 9 models) surpassed that of the models trained on all components. These models achieved an 88.6% reduction in manual screening workload while maintaining high recall (0.93). Conclusions: Component selection from the title and abstract can improve the performance of screening models and substantially reduce the manual screening workload in systematic review updates. Future research should focus on validating this approach across different systematic review domains.

• Access to Document (DOI)

Identifying Adverse Events in Outpatients With Prostate Cancer Using Pharmaceutical Care Records in Community Pharmacies: Application of Named Entity Recognition.

Yanagisawa Y, Watabe S, Yokoyama S, Sayama K, Kizaki H, Tsuchiya M, Imai S, Someya M, Taniguchi R, Yada S, Aramaki E, Hori S

JMIR cancer 11   e69663 2025年03月

研究論文（学術雑誌）, 共著, 査読有り

　概要を見る

Background: Androgen receptor axis-targeting reagents (ARATs) have become key drugs for patients with castration-resistant prostate cancer (CRPC). ARATs are taken long term in outpatient settings, and effective adverse event (AE) monitoring can help prolong treatment duration for patients with CRPC. Despite the importance of monitoring, few studies have identified which AEs can be captured and assessed in community pharmacies, where pharmacists in Japan dispense medications, provide counseling, and monitor potential AEs for outpatients prescribed ARATs. Therefore, we anticipated that a named entity recognition (NER) system might be used to extract AEs recorded in pharmaceutical care records generated by community pharmacists. Objective: This study aimed to evaluate whether an NER system can effectively and systematically identify AEs in outpatients undergoing ARAT therapy by reviewing pharmaceutical care records generated by community pharmacists, focusing on assessment notes, which often contain detailed records of AEs. Additionally, the study sought to determine whether outpatient pharmacotherapy monitoring can be enhanced by using NER to systematically collect AEs from pharmaceutical care records. Methods: We used an NER system based on the widely used Japanese medical term extraction system MedNER-CR-JA, which uses Bidirectional Encoder Representations from Transformers (BERT). To evaluate its performance for pharmaceutical care records by community pharmacists, the NER system was first applied to 1008 assessment notes in records related to anticancer drug prescriptions. Three pharmaceutically proficient researchers compared the results with the annotated notes assigned symptom tags according to annotation guidelines and evaluated the performance of the NER system on the assessment notes in the pharmaceutical care records. The system was then applied to 2193 assessment notes for patients prescribed ARATs. Results: The F1-score for exact matches of all symptom tags between the NER system and annotators was 0.72, confirming the NER system has sufficient performance for application to pharmaceutical care records. The NER system automatically assigned 1900 symptom tags for the 2193 assessment notes from patients prescribed ARATs; 623 tags (32.8%) were positive symptom tags (symptoms present), while 1067 tags (56.2%) were negative symptom tags (symptoms absent). Positive symptom tags included ARAT-related AEs such as “pain,” “skin disorders,” “fatigue,” and “gastrointestinal symptoms.” Many other symptoms were classified as serious AEs. Furthermore, differences in symptom tag profiles reflecting pharmacists’ AE monitoring were observed between androgen synthesis inhibition and androgen receptor signaling inhibition. Conclusions: The NER system successfully extracted AEs from pharmaceutical care records of patients prescribed ARATs, demonstrating its potential to systematically track the presence and absence of AEs in outpatients. Based on the analysis of a large volume of pharmaceutical medical records using the NER system, community pharmacists not only detect potential AEs but also actively monitor the absence of severe AEs, offering valuable insights for the continuous improvement of patient safety management.

• Access to Document (DOI)