Journal of Medical Internet Research最新文献

Background: Interoperability has been a challenge for half a century. Led by an informatics view of the world, the quest for interoperability has evolved from typing and categorizing data to building increasingly complex models. In parallel with the development of these models, the field of terminologies and ontologies emerged to refine granularity and introduce notions of hierarchy. Clinical data models and terminology systems vary in purpose, and their fixed categories shape and constrain representation, which inevitably leads to information loss.

Objective: Despite these efforts, semantic interoperability remains imperfect. Achieving it is essential for effective data reuse but requires more than rich terminologies and standardized models. This methodological study explores the extent to which the SNOMED CT (Systematized Nomenclature of Medicine - Clinical Terms) compositional grammar can be leveraged and extended to approximate a formal descriptive grammar, allowing clinical reality to be expressed in coherent, meaningful sentences rather than preconstrained categories.

Methods: Building on a decade of semantic representation efforts at the Geneva University Hospitals, we developed a framework to identify recurring semantic gaps in clinical data. We addressed these gaps by systematically modifying the SNOMED CT Machine Read` Concept Model and extending its Augmented Backus-Naur Form syntax to support necessary grammatical structures and external vocabularies.

Results: This approach enabled the semantic representation of over 119,000 distinct data elements covering 13 billion instances. By extending the grammar, we successfully addressed critical limitations such as negation, scalar values, uncertainty, temporality, and the integration of external terminologies like Pango. The extensions proved essential for capturing complex clinical nuances that standard precoordinated concepts could not represent.

Conclusions: Rather than creating a new standard from scratch, extending the grammatical capabilities of SNOMED CT offers a viable pathway toward high-fidelity semantic representation. This work serves as a proof-of-concept that separating the rules of composition from vocabulary allows for a more flexible and robust description of clinical reality, provided that challenges regarding governance and machine readability are addressed.

Background: Cancer remains one of the foremost global causes of mortality, with nearly 10 million deaths recorded by 2020. As incidence rates rise, there is a growing interest in leveraging machine learning (ML) to enhance prediction, diagnosis, and treatment strategies. Despite these advancements, insufficient attention has been directed toward the integration of sociodemographic variables, which are crucial determinants of health equity, into ML models in oncology.

Objective: This review aims to investigate how ML techniques have been used to identify patterns of predictive association between sociodemographic factors and cancer-related outcomes. Specifically, it seeks to map current research endeavors by detailing the types of algorithms used, the sociodemographic variables examined, and the validation methodologies used.

Methods: We conducted a systematic literature review in accordance with the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines. Searches were executed across 6 databases, focusing on the primary studies using ML to investigate the association between sociodemographic characteristics and cancer-related outcomes. The search strategy was informed by the PICO (population, intervention, comparison, and outcome) framework, and a set of predefined inclusion criteria was used to screen the studies. The methodological quality of each included paper was assessed.

Results: Out of the 328 records examined, 19 satisfied the inclusion criteria. The majority of studies used supervised ML techniques, with random forest and extreme gradient boosting being the most commonly used. Frequently analyzed variables include age, male or female or intersex, education level, income, and geographic location. Cross-validation is the predominant method for evaluating model performance. Nevertheless, the integration of clinical and sociodemographic data is limited, and efforts toward external validation are infrequent.

Conclusions: ML holds significant potential for discerning patterns associated with the social determinants of cancer. Nevertheless, research in this domain remains fragmented and inconsistent. Future investigations should prioritize the integration of contextual factors, enhance model transparency, and bolster external validation. These measures are crucial for the development of more equitable, generalizable, and actionable ML applications in cancer care.

Background: Living evidence (LE) synthesis refers to the method of continuously updating systematic evidence reviews to incorporate new evidence. It has emerged to address the limitations of the traditional systematic review process, particularly the absence of or delays in publication updates. The emergence of COVID-19 accelerated the progress in the field of LE synthesis, and currently, the applications of artificial intelligence (AI) in LE synthesis are expanding rapidly. However, in which phases of LE synthesis should AI be used remains an unanswered question.

Objective: This study aims to (1) document the phases of LE synthesis where AI is used and (2) investigate whether AI improves the efficiency, accuracy, or utility of LE synthesis.

Methods: We searched Web of Science, PubMed, the Cochrane Library, Epistemonikos, the Campbell Library, IEEE Xplore, medRxiv, COVID-19 Evidence Network to support Decision-making, and McMaster Health Forum. We used Covidence to facilitate the monthly screening and extraction processes to maintain the LE synthesis process. Studies that used or developed AI or semiautomated tools in the phases of LE synthesis were included.

Results: A total of 24 studies were included, including 17 on LE syntheses, with 4 involving tool development, and 7 on living meta-analyses, with 3 involving tool development. First, a total of 34 AI or semiautomated tools were involved, comprising 12 AI tools and 22 semiautomated tools. The most frequently used AI or semiautomated tools were machine learning classifiers (n=5) and the Living Interactive Evidence synthesis platform (n=3). Second, 20 AI or semiautomated tools were used for the data extraction or collection and risk of bias assessment phase, and only 1 AI tool was used for the publication update phase. Third, 3 studies demonstrated the improvement in efficiency achieved based on time, workload, and conflict rate metrics. Nine studies applied AI or semiautomated tools in LE synthesis, obtaining a mean recall rate of 96.24%, and 6 studies achieved a mean F1-score of 92.17%. Additionally, 8 studies reported precision values ranging from 0.2% to 100%.

Conclusions: AI and semiautomated tools primarily facilitate data extraction or collection and risk of bias assessment. The use of AI or semiautomated tools in LE synthesis improves efficiency, leading to high accuracy, recall, and F1-scores, while precision varies across tools.