Data’s big three: how learning health systems, artificial intelligence and predictive analytics are transforming healthcare

Three areas transforming how we use data in healthcare, learning health systems (LHSs), artificial intelligence (AI) and predictive analytics, are enhancing the quality, efficiency and personalisation of care. By leveraging data from everyday clinical practice, LHSs create a feedback loop where new knowledge is constantly generated and applied, ensuring best practices are embedded in care processes, leading to better patient outcomes and more efficient healthcare delivery. A recent study reported digital data capture as the most common LHS characteristic, whereas other characteristics such as patient engagement, aligned governance and a culture of rapid learning and improvement were less likely to be reported as impactful characteristics.1 AI provides effective tools for diagnostics, treatment planning and patient care. AI algorithms can analyse medical images with high precision, predict disease progression and personalise treatment plans based on individual patient data.2 And predictive analytics uses data mining, machine learning and statistical modelling to forecast future health events and trends. In healthcare, predictive analytics can identify high-risk patients, optimise resource allocation and enhance disease prevention strategies.3 By anticipating healthcare needs, predictive analytics helps in proactive intervention, reducing hospital readmissions and improving overall patient care. Together, these technologies create a more data-driven, efficient and patient-centred healthcare system, ultimately leading to improved health outcomes and reduced costs. It is for this reason that this editor’s choice reflects on three articles from this issue related to these three data technologies transforming health.

The article ‘Barriers and facilitators to learning health systems in primary care: a framework analysis’ explores the challenges and enablers of implementing LHS in primary care settings.4 This study provides a comprehensive analysis of the factors influencing the adoption of LHS principles in general practice, particularly through the lens of implementation science. While primary care serves as the first point of contact for most patients, playing a crucial role in reducing overall health costs and alleviating pressure on other parts of the healthcare system, primary care faces significant challenges, including an ageing population, an increase in chronic diseases and workforce shortages. These issues have been exacerbated by the COVID-19 pandemic, which has necessitated system-wide reorganisation and added stress to general practitioners and their practices. The concept of an LHS, as defined by the National Academy of Medicine, involves a system that continuously improves by systematically learning from each care experience. Key characteristics of an LHS include real-time access to knowledge, patient–clinician partnerships, incentives for high-value care, a continuous learning culture, and supportive structure and governance. While LHSs have been successfully implemented in tertiary hospital settings, their application in primary care remains underexplored.

Fisher et al examine the barriers and facilitators to achieving an LHS in primary care and to provide practical recommendations for general practices. Through an analysis of an LHS in a university-based general practice in Sydney, Australia, the study used a framework analysis approach, coding data from semistructured interviews with clinic staff according to the theoretical domains framework and an LHS framework. The study identified several barriers to LHS implementation, including lack of knowledge and skills, unclear policy and roles, physical distance between teams and poor data quality. The facilitators of LHS implementation included leadership and support, person-centred care and shared and communicated organisational goals. The study makes several recommendations, including the need for established clear policies and procedures that support LHS principles, including guidelines for data collection, analysis and sharing, investment in interoperable health information technology systems that facilitate seamless data integration and real-time access to knowledge, engagement of all staff in the LHS vision and ongoing education and training to build the necessary skills and knowledge and creating a culture of continuous learning and improvement by encouraging feedback, collaboration and innovation.

The article ‘Building a house without foundations? A 24-country qualitative interview study on artificial intelligence in intensive care medicine’ explores the perspectives of intensive care professionals on the use and implementation of AI technologies in intensive care units (ICUs).5 This study provides valuable insights into the readiness and challenges faced by ICUs across different countries in adopting AI by providing an understanding of the views of ICU staff from both high-income countries (HICs) and lower-to-middle-income countries (LMICs) regarding AI in ICUs. Through semistructured qualitative interviews with 59 intensive care professionals from 24 countries, the study identified several key themes and insights. Participants generally had positive views about the potential of AI to enhance ICU care. They recognised AI’s ability to assist in disease identification, predict disease progression, phenotype diseases and guide clinical decision-making. However, they also expressed concerns about the practical implementation of AI in clinical settings.

McLennan et al identified several barriers to AI use in ICUs. Technical challenges included data quality and integration, such as the challenge of obtaining high-quality, standardised data, infrastructure, due to the lack of the necessary digital infrastructure to support AI technologies, lack of knowledge and skills required to use AI among ICU staff, ethical, legal and regulatory issues, including data privacy, security and the ethical use of AI, and trust and acceptance, which is crucial for the successful adoption of AI. The study found that the readiness to implement AI varied significantly between ICUs. Interestingly, the differences were not strictly between HICs and LMICs but rather between a small number of well-resourced ICUs in large tertiary hospitals and the majority of other ICUs. The well-resourced ICUs had the necessary digital infrastructure and skilled staff, while most other ICUs, regardless of their country’s income level, did not. The authors emphasise that resourcing into developing AI without first building the necessary digital infrastructure is unethical, with several of the above barriers needing to be addressed before AI can be effectively implemented in ICUs:

The final article ‘Promising algorithms to perilous applications: a systematic review of risk stratification tools for predicting healthcare utilisation’ is a systematic literature review that critically examines the effectiveness and real-world application of risk stratification tools in primary care.6 These tools are designed to predict healthcare utilisation and are integral to population health management strategies, aiming to identify high-risk individuals for targeted interventions. The article reviews 51 studies evaluating 28 risk prediction models, half of which were externally validated models. The authors identified that while many models demonstrated acceptable discriminatory power within the contexts they were developed, their performance in different global contexts was less clear. The majority of real-world evaluation studies reported no significant change or even increases in healthcare utilisation within targeted groups. Only one-third of the reports showed some benefit, indicating that accurate identification of high-risk individuals does not reliably translate to improved service delivery or reduced morbidity. The real-world application of these models often produced disappointing results. The study highlighted that the predictive performance of these models might diminish when applied to demographically and culturally distinct populations or when using electronic health data with different characteristics. This variability can lead to inconsistent intervention thresholds and varying clinical effectiveness, potentially exacerbating healthcare inequalities.

Oddy et al identified that high-quality, standardised data are crucial for the effective use of risk stratification tools, emphasising the need for a cautious approach to integrating risk stratification tools into primary care. While risk stratification tools have the potential to enhance primary care, their real-world application is fraught with challenges. The authors recommended deployment of individual-level risk prediction being subjected to the same controls as other technologies due to their impact on clinical care pathways. Further, the authors call for national bodies to be more involved in the procurement of commercial risk stratification services and regulatory bodies to have greater regulation and involvement in the implementation of risk stratification in health.

These three articles all highlight similar challenges that need to be addressed as the integration of LHS, AI, and predictive analytics increases. Investment in the technology, workforce and regulation of these areas will improve health outcomes, service delivery and workforce retention. This proactive approach reduces hospital readmissions and costs and improves health and social care, fostering a data-driven, efficient and person-centred system. The sector is only starting to appreciate the safety challenges that novel types of technology bring, especially at the systems and societal level. Regulation and ethical review need to be in place before these technologies are integrated into care planning.7 Investment in the infrastructure to support these technologies is vital to ensure they are implemented and adopted in a safe manner. And workforce capabilities need to be addressed so the staff using these technologies have the skills and knowledge to do so responsibly.8 The sector needs a shift from a technology-centric approach towards a human-centred approach, with more robust evidence for how we can safely and responsibly adopt and evolve this technology across the health and social care sector.