Artificial intelligence in hemovigilance: A narrative review on advancing blood safety and monitoring systems

Literature search strategy

This review includes studies from multiple reputable sources such as PubMed, Scopus, IEEE Xplore, Web of Science, Google Scholar, and Embase. The following keywords and phrases were used in various combinations to conduct the search:

“Artificial intelligence in hemovigilance”

These search terms were selected to identify relevant studies on the integration of AI in improving blood transfusion safety, adverse event reporting, and operational efficiency in hemovigilance systems.

Timeframe and selection criteria

The chosen studies were the ones that are relevant to the review's focus on AI and its role in hemovigilance. At the very first stage, they were meaningful to the designated titles and abstracts; therefore, they were subjected to inclusion and exclusion criteria. Full texts of eligible articles were then inspected for further evaluation. The Pollock et al.'s systematic review was relevant to the research, so papers published from 2010 to 2024 were used to come up with the state-of-the-art studies concerning AI applications. The selection covered peer-reviewed articles, and research, and reviews, as well as case studies, the latter of them focused on AI applications in the field of hemovigilance and adverse event detection and blood transfusion safety. However, sources that are not scientific peer-reviewed, such as gray literature or conference abstracts, as well as studies not AI-focused or have nothing to do with hemovigilance or nonhealthcare applications were not included.

Data extraction

Relevant data from the selected studies were extracted, including:

Study characteristics: Authors, publication year, study design (e.g. case study, review, experimental study), and geographical region.

OPEN IN VIEWER

Figure 1.

PRISMA flowchart for AI in hemovigilance review. AI: Artificial intelligence.

Critical analysis and interpretation

Transforming hemovigilance with AI innovations

The integration of AI into hemovigilance systems has significantly improved blood transfusion safety and operational efficiency. AI technologies enhance various aspects of hemovigilance, including real-time data acquisition, adverse event detection, predictive analytics for transfusion needs, and blood supply chain management optimization, despite ongoing challenges related to data standardization and system interoperability.

Seamless data collection and integration in hemovigilance

AI has become a revolutionary instrument in hemovigilance, especially in overcoming the problems of collecting and integrating data. The key aspect of this progress is the employment of NLP techniques, which make it possible to handle unstructured data from electronic health records (EHRs) and other sources. Unstructured clinical narratives in EHRs provide extensive and patient-specific data that are necessary for hemovigilance. NLP-enabled AI systems can pull this data out and examine it rapidly, thus the discovery of transfusion-related incidents and other main hemovigilance indicators becomes possible. ⁶ AI plays a pivotal role in hemovigilance, not only facilitating data collection but also linking and harmonizing this data across different sources, through the introduction of NLP, which is necessary for the analysis of unstructured data (Figure 2). For example, NLP techniques have been successfully utilized to detect adverse drug reactions from pediatric EHRs through the pairing of drugs–symptoms with regulatory repository resources like the food and drug administration (FDA)'s Adverse Event Reporting System. ⁷ Pointing out the methods that result in a breakthrough of AI in terms of data extraction and integration through the use of techniques. The evolution of high-tech, clinical language models has also completely changed hemovigilance data processing. Models like GatorTron, which have been trained by de-identified generated text sourced from over 82 billion words, have achieved near-perfect performance for a wide array of clinical NLP tasks including concept extraction and medical question answering to a high degree. These strengths enable AI models to extract detailed insights, identify relevant information, and generate meaningful data from large datasets, thereby enhancing the effectiveness and efficiency of hemovigilance systems. ⁸

OPEN IN VIEWER

Figure 2.

Role of AI in hemovigilance. ⁴⁵ This figure illustrates the integration of AI tools in the monitoring and reporting of adverse events in blood transfusion practices, highlighting the potential for predictive analytics and automated data processing in improving patient safety. AI: Artificial intelligence.

Enhanced detection and reporting of blood products induced adverse events through AI

AI and ML have served as tremendous means for the identification and notification of adverse events related to blood transfusions as well as for the general hemovigilance. ML models are utilized to predict outcomes, assess risks, and detect complications such as hemorrhage in trauma patients that require transfusions. These models have exhibited notable qualities over traditional standards by enhancing predictions of mortality and the development of patient-specific scoring systems. AI-driven ML technologies are moving progressively to become a tool in trauma management that aids doctors to come up with a patient-specific diagnosis. ⁹ Existing national hemovigilance systems, such as the National Healthcare Safety Network (NHSN) Hemovigilance Module in the United States, play an important role in monitoring transfusion-related adverse events. However, although these systems are not AI-based, their comprehensive databases provide the data that can be used for ML analysis, which will increase the accuracy and efficiency of adverse event detection. Also, resources like the notify library import hemovigilance data to be used for both educational and reference purposes, therefore very supportive of safety in transfusion practices. ¹⁰ Despite the advancements, the following disadvantages exist: the majority of the ML research on hemovigilance and trauma care is retrospective in nature, with little validation in real-world clinical settings, either prospective or ongoing. Even though prediction models are available for transfusions requirements and coagulopathy, they are not widely used in clinical practice which is an indication of the theory–practice gap.^9,10

Leveraging predictive analytics for risk management in hemovigilance

AI has the capability to be a game changer in hemovigilance programs through predictive analytics and personalized risk assessments, which in turn, improve safety of the patient and transfusion practices. Through processing of vast arrays of data consisting of patients’ background information, transfusion histories, and adverse event reports, AI-based predictive algorithms can map out intricate patterns and risk factors connected to transfusion reactions. These revelations serve as stepping stones toward the early identification of high-risk individuals, thereby equipping clinicians with the power to put in place-specific preventive measures. AI models, for example, can predict the likelihood of the most common reactions such as the febrile nonhemolytic transfusion reactions or allergic responses, therefore enhancing the application of proactive care strategies.^11–13 AI-based decision support systems have the potential to integrate patient-specific data with clinical guidelines and risk stratification tools, which, in turn, allow for the individualization of transfusion plans. The systems suggest data-driven best practices for the choice of blood components, the assessment of the need for medication before the transfusion, and the development of monitoring protocols suitable for personalization. The risk of serious complications, such as transfusion-associated circulatory overload and transfusion-related acute lung injury, which are potentially fatal, is diminished as this level of personalization is implemented. ¹⁴ Besides these individual strategies result in increased safety of the transfusion process, they also provide the method to the efficient use of resources and the reduction of healthcare costs. ¹²

Optimizing blood supply chain efficiency with AI integration

AI has become a groundbreaking technology in blood supply chain management, with better demand forecasting, reduction in waste, and safety and traceability becoming easier as results. AI-based decision support systems that use ML and time series forecasting models can be the best ones in all the blood supply chain processes. This might mean things like how much you will need, how you deal with a donor, or the fixed blood donor schedules that together bring to the table heightened blood collection, reduced inventory wastage, and eliminated shortages. ¹⁵ The sophisticated inventory models and AI algorithms contribute to better blood inventory management, however, management practices are more important compared to the most cutting-edge tech, according to the latest research. For example, incorporating ploys of commercial supply chains like stock sharing or lateral transshipment due to expiring blood units between hospitals can incredibly boost the entire supply chain as a whole. ¹⁶ The mashup of AI together with upcoming technologies like blockchain only further ignites the possibility of blood supply chain management. Blockchain provides means of increasing transparency, traceability, and security, thereby, mitigating the potential risks like errors or frauds that may occur during the donation-to-transfusion process. ¹⁴ Globally, during the COVID-19 pandemic, AI-based multivariate time-series deep learning models have proven to be effective enough to predict the demand for donations and maintain a satisfactory inventory level, hence allowing the continuous delivery of patients care.^17,18

AI technologies transforming hemovigilance practices

AI increases efficiency in hemovigilance by using ML algorithms such as SVMs, random forests, and neural networks which are extensively used in the health sector for pattern and anomaly detection in huge datasets from transfusion records or medical IoT devices.^19,20 Random forest is an algorithm, which is noted for its resilience and great performance in different areas of application and is also an algorithm which is used in the healthcare domain. It is very effective when dealing with datasets that have a high number of features and is less likely to overfit compared to other algorithms. ²¹ SVMs are especially useful for classifying objects in the medical record and for anomaly detection and can deliver solutions with less training data. ¹⁸ Furthermore, neural networks with their deep learning capabilities are well-suited for the discovery of concrete patterns in the very large dataset, and as a result of such patterns they are the key technology for hemovigilance. ²² For the processing of the large numbers of the data that is generated by the medical IoT devices and the transfusion records, big data analytics frameworks such as Apache Spark are the only viable solutions, and they enable the processing of the datasets to be distributed. The addition of techniques that involve dimensionality reduction and feature extraction will give you more tools for proper handling of high-dimensional medical data. ²³ Moreover, IoT devices with edge-computing capability are able to conduct real-time monitoring during transfusions and thus anomalous events will be immediately detected and assist in the increase of patient's safety. The employment of these AI-driven techniques in combination ensures that the hemovigilance systems can be highly effective in monitoring and improving transfusions. ²⁴

Real-world applications and case studies of AI in hemovigilance

The incorporation of AI in hemovigilance has emerged as a feasible strategy to enhance healthcare outcomes and the credibility of adverse event reporting in the systems. An analysis detected 125 adverse event signals among 26 System Organ Classes, comprising 108 previously unknown signals, thereby confirming the potential of AI to detect security issues that do not appear to be clear but are hidden in real-world data. ²⁵ Moreover, one of the most ingenious uses of AI is demonstrated through DeepSAVE, which is a deep learning framework designed to precisely identify user query logs that contain adverse events. It presented a solution to the problem of comprehension in the diverse search settings and the variety of user behavior levels. In contrast to the available methods, DeepSAVE led in achieving the best results while it was used in three vast authentic datasets, thus supporting the fact that AI has the power to utilize the online search data for quick learning of potential adverse events. ²⁶ The AI safety monitoring assistance in educational situations has been put under investigation. A research project that involved health science students who participated in AI-enhanced simulations discovered that the intervention group reported more medication-related adverse events than the control group. The results of this study suggest that AI-based simulations can enhance the participants’ ability to think systemically and accurately report, thus resulting in the practice of the safety monitoring and reporting processes. ²⁷ In spite of the promises brought by the development, some issues still remain to be addressed regarding the use of AI for adverse event detection. ²⁸

Emerging trends and future prospects of AI in hemovigilance

The future of AI in hemovigilance and healthcare relies on the creation of standards that incorporate the integration of AI with other systems and exercising of privacy-protective techniques such as federated learning. ³⁸ Federated learning is a new way that enables data privacy while AI models are improved. This technique enables the training of the neural networks in remote, distributed edge devices collaboratively by avoiding the case to pass the raw data. Consequently, patient detail confidentiality is maintained, and privacy regulations such as the General Data Protection Regulation and the Health Insurance Portability and Accountability Act are upheld. ⁴⁰ It aids in the establishment of a healthcare open ecosystem by enabling knowledge sharing to be done from widespread and varied datasets but with privacy being maintained. ⁴¹ However, federated learning has both advantages and disadvantages, with the drawbacks being the distortion of the model due to the channel fading and poor aggregation of locally trained models on the unbalanced data. ⁴⁰ Further development in the future might be achieved by combining digital twin technology with federated learning, which could shape smart city healthcare network applications. ⁴² Besides federated learning, global AI standards in hemovigilance and healthcare should be implemented. Consistency, interoperability, and fairness will be the main guarantees of these standards, while at the same time they will guarantee the security of trust and the acceptance of them clinically. The incorporation of AI in different datasets and automating of workflows would even make AI more valuable in hemovigilance programs. Hence, AI's growing role evolves, the problems of generalizing algorithms, validating the clinical procedures as well as integrating advanced technologies such as large language models will be of utmost importance in the healthcare delivery transformation. ⁴³

Limitations

A key challenge is the variability in data quality. Hemovigilance data, often collected through multiple sources, can be inconsistent, incomplete, or poorly standardized, which hinders AI systems from making accurate predictions. Inaccurate or missing information about patient history, transfusion details, or reaction severity may affect AI's ability to detect trends or assess risks properly. Another limitation is the AI's reliance on existing data patterns. If a reaction is rare or atypical, AI might struggle to identify it, as it depends on historical data to recognize patterns. This can delay the detection of novel adverse events that were not previously encountered or documented in databases.

A comprehensive review of the state-of-the-art automatic trigger tool-based methods that were used for adverse event detection in EHRs showed a sharp divergence in the prevalence of adverse events and positive predictive values across the studies. ⁴⁴ The discrepancies were due to the differently described methodologies, which caused the interpretation difficulties. This discovery pointed to the fact that there is a need to have unified procedures and reporting guidelines in place to ensure uniformity and trust in AI systems.

Additionally, AI systems may not fully account for the complexity of individual patients. Variations in underlying health conditions, comorbidities, or genetic factors may influence how a patient reacts to a transfusion. AI models often lack the depth of human clinical expertise needed to assess these nuanced variables. Finally, AI in hemovigilance requires continuous updates and training to adapt to emerging knowledge. Without regular input from medical professionals and the latest research, AI systems risk becoming outdated or unable to respond to new challenges in transfusion safety.

Purpose of the review

This review aims to explore the potential and challenges of integrating AI into hemovigilance systems to enhance blood safety and monitoring. It aims to highlight how AI can improve the detection, reporting, and analysis of adverse reactions during and after blood transfusions, thereby enhancing patient outcomes. This review highlights AI's capacity to process large datasets, identify patterns, and predict risks more accurately and efficiently than traditional methods. However, the methodology does not explicitly detail how these capabilities were assessed. Additionally, it addresses the limitations and hurdles in implementing AI, such as data quality issues, the need for standardization, and its ability to detect rare or novel reactions. By critically analyzing the current state of AI in hemovigilance, this review provides insights into its potential for revolutionizing blood transfusion safety and discusses areas for future research and development. The goal is to inform healthcare professionals, researchers, and policymakers about the promising role AI could play in improving transfusion practices and patient care.