International Journal of Medical Informatics最新文献

Background

Frailty is an age-related syndrome characterized by loss of strength and exhaustion and associated with multi-morbidity. Early detection and prediction of the appearance of frailty could help older people age better and prevent them from needing invasive and expensive treatments. Machine learning techniques show promising results in creating a medical support tool for such a task.

Methods

This study aims to create a dataset for machine learning-based frailty studies, using Fried's Frailty Phenotype definition. Starting from a longitudinal study on aging in the UK population, we defined a frailty label for each subject. We evaluated the definition by training seven different models for detecting frailty with data that were contemporary to the ones used for the definition. We then integrated more data from two years before to obtain prediction models with a 24-month horizon. Features selection was performed using the MultiSURF algorithm, which ranks all features in order of relevance to the detection or prediction task.

Results

We present a new frailty dataset of 5303 subjects and more than 6500 available features. It is publicly available, provided one has access to the original English Longitudinal Study of Ageing dataset. The dataset is balanced after grouping frailty with pre-frailty, and it is suitable for multiclass or binary classification and prediction problems. The seven tested architectures performed similarly, forming a solid baseline that can be improved with future work. Linear regression achieved the best F-score and AUROC in detection and prediction tasks.

Conclusions

Creating new frailty-annotated datasets of this size is necessary to develop and improve the frailty prediction techniques. We have shown that our dataset can be used to study and test machine learning models to detect and predict frailty. Future work should improve models' architecture and performance, consider explainability, and possibly enrich the dataset with older waves.

Background

Strabismus is a common eye condition affecting both children and adults. Effective patient education is crucial for informed decision-making, but traditional methods often lack accessibility and engagement. Chatbots powered by AI have emerged as a promising solution.

Aim

This study aims to evaluate and compare the performance of three chatbots (ChatGPT, Bard, and Copilot) and a reliable website (AAPOS) in answering real patient questions about strabismus.

Method

Three chatbots (ChatGPT, Bard, and Copilot) were compared to a reliable website (AAPOS) using real patient questions. Metrics included accuracy (SOLO taxonomy), understandability/actionability (PEMAT), and readability (Flesch-Kincaid). We also performed a sentiment analysis to capture the emotional tone and impact of the responses.

Results

The AAPOS achieved the highest mean SOLO score (4.14 ± 0.47), followed by Bard, Copilot, and ChatGPT. Bard scored highest on both PEMAT-U (74.8 ± 13.3) and PEMAT-A (66.2 ± 13.6) measures. Flesch-Kincaid Ease Scores revealed the AAPOS as the easiest to read (mean score: 55.8 ± 14.11), closely followed by Copilot. ChatGPT, and Bard had lower scores on readability. The sentiment analysis revealed exciting differences.

Conclusion

Chatbots, particularly Bard and Copilot, show promise in patient education for strabismus with strengths in understandability and actionability. However, the AAPOS website outperformed in accuracy and readability.

Introduction

Inherent variations between inter-center data can undermine the robustness of segmentation models when applied at a specific center (dataset shift). We investigated whether specialized center-specific models are more effective compared to generalist models based on multi-center data, and how center-specific data could enhance the performance of generalist models within a particular center using a fine-tuning transfer learning approach. For this purpose, we studied the dataset shift at center level and conducted a comparative analysis to assess the impact of data source on glioblastoma segmentation models.

Methods & Materials

The three key components of dataset shift were studied: prior probability shift—variations in tumor size or tissue distribution among centers; covariate shift—inter-center MRI alterations; and concept shift—different criteria for tumor segmentation. BraTS 2021 dataset was used, which includes 1251 cases from 23 centers. Thereafter, 155 deep-learning models were developed and compared, including 1) generalist models trained with multi-center data, 2) specialized models using only center-specific data, and 3) fine-tuned generalist models using center-specific data.

Results

The three key components of dataset shift were characterized. The amount of covariate shift was substantial, indicating large variations in MR imaging between different centers. Glioblastoma segmentation models tend to perform best when using data from the application center. Generalist models, trained with over 700 samples, achieved a median Dice score of 88.98%. Specialized models surpassed this with 200 cases, while fine-tuned models outperformed with 50 cases.

Conclusions

The influence of dataset shift on model performance is evident. Fine-tuned and specialized models, utilizing data from the evaluated center, outperform generalist models, which rely on data from other centers. These approaches could encourage medical centers to develop customized models for their local use, enhancing the accuracy and reliability of glioblastoma segmentation in a context where dataset shift is inevitable.

• •

  Background: Clinical Data Repositories (CDRs) lie at the core of both primary and secondary utilisation of vast amounts of health information. These data are intricate, heterogeneous and constantly evolving alongside advancements in biomedical sciences. The separation between clinical content and its persistence renders the archetype-based paradigm naturally well-suited to manage this complexity. This approach is adopted by a number of open source and commercial CDR solutions, mostly implementing the openEHR specifications, which also encompass the Archetype Query Language (AQL) to define portable queries independently of the persistence scheme.

• •

  Aim: To provide a wide knowledge base and a set of customisable tools as a support in the selection of openEHR CDRs according to a broad ensemble of features relevant to different use cases.

• •

  Approach: After conducting an extensive search of the existing openEHR CDRs, a survey consisting of fifty-four questions was administered to all the nineteen identified vendors/developers, covering an ample set of aspects such as licensing, implementation, interoperability and ecosystem. Subsequently, the answers from the eleven responders were processed and analysed, also applying statistical techniques.

• •

  Results: Two detailed tables depict the current landscape of openEHR CDRs, presenting a structured view of the most relevant survey answers. Unsupervised clustering led to the categorisation of CDRs into four groups, and a decision-making diagram has been designed to aid the CDR selection according to a restricted set of desired features.

• •

  Conclusions: Compared to a similar study conducted in 2013, the results indicate a worldwide rise in the number of openEHR CDRs, marked by a wider adoption of the dedicated query language, to leverage AQL universality across openEHR platforms. The evolution in the last ten years also included an increased attention to exchange data with non-openEHR solutions and to simplify the content creation from clinical models based on Archetypes and Templates. Materials and analytical tools hereby presented are publicly available for further reuse.

Objective

Norwegian health registries covering entire population are used for administration, research, and emergency preparedness. We harmonized these data onto the Observational Medical Outcomes Partnership common data model (OMOP CDM) and enrich real-world data in OMOP format with COVID-19 related data.

Methods

Data from six registries (2018–2021) covering birth registrations, selected primary and secondary care events, vaccinations, and communicable disease notifications were mapped onto the OMOP CDM v5.3. An Extract-Transform-Load (ETL) pipeline was developed on simulated data using data characterization documents and scanning tools. We ran dashboard quality checks, cohort generations, investigated differences between source and mapped data, and refined the ETL accordingly.

Results

We mapped 1.5 billion rows of data of 5,673,845 individuals. Among these, there were 804,277 pregnancies, 483,585 mothers together with 792,477 children, and 472,948 fathers. We identified 382,516 positive tests for COVID-19 in 380,794 patients. These figures are consistent with results from source data. In addition to 11 million source codes mapped automatically, we mapped 237 non-standard codes to standard concepts and introduced 38 custom concepts to accommodate pregnancy-related terminologies that were not supported by OMOP CDM vocabularies. A total of 3,700/3,705 (99.8%) checks passed. The 5 failed checks could be explained by the nature of the data and only represent a small number of records.

Discussion and conclusion

Norwegian registry data were successfully harmonized onto OMOP CDM with high level of concordance and provides valuable source for federated COVID-19 related research. Our mapping experience is highly valuable for data partners with Nordic health registries.

Purpose

ChatGPT has the potential to offer patient-friendly support. Thyroid carcinoma has become increasingly prevalent in recent years. This study aimed to assess ChatGPT’s accuracy and adequacy in answering questions about information, management, and emotional support related to thyroid cancer.

Methods

We conducted a three-step study. In the first step, ChatGPT responded to 30 questions about thyroid cancer. In the second step, we presented three different cases of thyroid cancer patients and asked ChatGPT about their diagnosis, treatment management, and follow-up. In the third step, we inquired about emotional support for patients and their families. Three expert endocrinologists graded these responses according to ATA guidelines.

Results

We showed that ChatGPT regurgitated extensive knowledge of thyroid cancer (76.66% correct), but only small proportions (6.66%) were labeled as mixed with correct and incorrect/outdated data. However, it was inadequate in the evaluation of clinical cases of thyroid cancer. It mentioned treatment and follow-up recommendations in a general framework, not patient-specific. Also, it provided practical and multifaceted emotional support advice to patients and caregivers regarding the next steps and adjusting to a new diagnosis.

Conclusion

Our study is the first to evaluate the competence and reliability of ChatGPT in thyroid cancer. Although ChatGPT is moderately competent in obtaining information about thyroid cancer, it has not yet been determined to be sufficiently competent and reliable in case management. It has been found effective in guiding patients and their relatives regarding emotional support.

Introduction

Delays in discharging patients from Acute Medical Units hamper patient flows throughout the hospital. The decision to discharge a patient is mainly based on the patients’ physiological condition, but may vary between physicians. An objective decision-support system based on patients’ physiological data may help minimizing unnecessary delays in discharge. The aim of this proof-of-concept study is to assess the feasibility of predicting whether patients in an Acute Medical Unit are physiologically fit-for-discharge using machine learning with commonly available hospital data. Furthermore, this study investigated how long before actual time of discharge from the Acute Medical Unit we could predict discharge fitness. Also, the predictive importance of features extracted from these data was assessed.

Methods

Electronic Medical Records of patients who participated in a Randomized Controlled Trial conducted in an Acute Medical Unit were used retrospectively (N = 199). Only commonly available hospital data were used. Logistic Regression and Random Forest models were applied to predict every hour whether patients were physiologically fit-for-discharge. Nested 5-fold cross-validation with 5 repeats was used to optimize the model hyperparameters and to estimate the predictive performances.

Results

Physiological discharge fitness was predictable with reasonable performance for Logistic Regression (mean AUROC: 0.67) and Random Forest (mean AUROC: 0.69). For an intuitively chosen classification threshold of 0.8, mean specificity was 93.3 % and sensitivity 14.1 %. Models could predict physiological discharge fitness more than 24 h earlier than actual time of discharge for most patients who were correctly predicted to be fit-for-discharge. Patient characteristics, vital signs and laboratory results were shown to be important predictors.

Conclusion

This proof-of-concept study showed that it is feasible to predict with machine learning whether patients in an Acute Medical Unit are physiologically fit-for-discharge using commonly available hospital data.

Background

Digital interventions are becoming increasingly popular in rehabilitation. Understanding of device features which impact clinician adoption and satisfaction is limited. Research in the field should be conducted across diverse settings to ensure digital interventions do not exacerbate healthcare inequities.

Objective

This study aimed to understand rehabilitation clinicians’ preferences regarding device attributes and included a cross-cultural comparison.

Materials and Methods

Choice experiment methodology (best-worst scaling) was used to survey rehabilitation clinicians across Australia and Brazil. Participants completed 10 best-worst questions, choosing the most and least important device attributes from subsets of 31 attributes in a partially balanced block design. Results were analysed using multinomial models by country and latent class. Attribute preference scores (PS) were scaled to 0–100 (least to most important).

Results

A total of 122 clinicians from Brazil and 104 clinicians from Australia completed the survey. Most respondents were physiotherapists (83%) working with neurological populations (51%) in the private/self-employed sector (51%) who had experience using rehabilitation devices (87%). Despite preference heterogeneity across country and work sector (public/not-for-profit versus private/self-employed/other), clinicians consistently prioritised patient outcomes (PS 100.0, 95%CI: 86.2–100.0), patient engagement (PS 93.9, 95%CI: 80.6–94.2), usability (PS 81.3, 95%CI: 68.8–82.5), research evidence (PS 80.4, 95%CI: 68.1–81.7) and risk (PS 75.7, 95%CI: 63.8–77.3). In Australia, clinicians favoured device attributes which facilitate increased therapy dosage (PS 79.2, 95%CI: 62.6–81.1) and encourage patient independent practice (PS 66.8, 95%CI: 52.0–69.2). In Brazil, clinicians preferred attributes enabling device use for providing clinical data (PS 67.6, 95%CI: 51.8–70.9) and conducting clinical assessments (PS 65.6, 95%CI: 50.2–68.8).

Conclusion

Clinicians prioritise patients’ needs and practical application over technical aspects of digital rehabilitation devices. Contextual factors shape clinician preferences rather than individual clinician characteristics. Future device design and research should consider preferences and influences, involving diverse stakeholders to account for context-driven variations across cultures and healthcare settings.

Objective

This scoping review aims to explore the current state of encounter notification systems (ENS) between emergency departments (EDs) and primary care providers (PCPs), focusing on their mechanisms, effectiveness, impacts, and challenges in healthcare settings.

Methods

A systematic search was conducted using PubMed/MEDLINE and Google Scholar to identify relevant literature on ENS between EDs and PCPs. Eligible studies were selected based on predefined criteria, and data were synthesized narratively.

Results

The initial search yielded 1,396 articles, with 29 included in the review. Studies highlighted the significance of encounter notifications in improving communication and care coordination between EDs and PCPs, leading to enhanced patient outcomes. However, challenges such as technological barriers, privacy concerns, and variations in healthcare settings were identified.

Conclusion

ENS play a crucial role in enhancing communication and care coordination between EDs and PCPs. Despite challenges, these systems offer substantial benefits and opportunities for improving patient care in the ED-primary care continuum. Future research should focus on addressing implementation barriers and evaluating long-term impacts to optimize the effectiveness of ENS in this context.

Background

Prediction of mortality is very important for care planning in hospitalized patients with dementia and artificial intelligence has the potential to serve as a solution; however, this issue remains unclear. Thus, this study was conducted to elucidate this matter.

Methods

We identified 10,573 hospitalized patients aged ≥ 45 years with dementia from three hospitals between 2010 and 2020 for this study. Utilizing 44 feature variables extracted from electronic medical records, an artificial intelligence (AI) model was constructed to predict death during hospitalization. The data was randomly separated into 70 % training set and 30 % testing set. We compared predictive accuracy among six algorithms including logistic regression, random forest, extreme gradient boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), multilayer perceptron (MLP), and support vector machine (SVM). Additionally, another set of data collected in 2021 was used as the validation set to assess the performance of six algorithms.

Results

The average age was 79.8 years, with females constituting 54.5 % of the sample. The in-hospital mortality rate was 6.7 %. LightGBM exhibited the highest area under the curve (0.991) for predicting mortality compared to other algorithms (XGBoost: 0.987, random forest: 0.985, logistic regression: 0.918, MLP: 0.898, SVM: 0.897). The accuracy, sensitivity, positive predictive value, and negative predictive value of LightGBM were 0.943, 0.944, 0.943, 0.542, and 0.996, respectively. Among the features in LightGBM, the three most important variables were the Glasgow Coma Scale, respiratory rate, and blood urea nitrogen. In the validation set, the area under the curve of LightGBM reached 0.753.

Conclusions

The AI prediction model demonstrates strong accuracy in predicting in-hospital mortality among patients with dementia, suggesting its potential implementation to enhance future care quality.