Machine Learning and Artificial Intelligence in Evidence Generation and Evidence Synthesis

INTRODUCTION Suboptimal decision-making in healthcare contributes to decreased patient outcomes (health status and/or quality of life), increased healthcare costs, inequitable access to care, patient dissatisfaction and legal and ethical issues.[1] To mitigate these negative consequences, healthcare organisations and professionals should prioritise evidence-based practices, promote interdisciplinary collaboration, invest in on-going education and training and implement quality improvement initiatives to optimise decision-making. Evidence-based medicine (EBM) is a knowledge driven framework, while machine learning (ML) and artificial intelligence (AI) provide data-driven knowledge. Bridging the gap between these approaches requires a balanced approach that addresses ethical, legal, biased and transparent aspects during the process of evidence generation, evidence gap analysis and evidence synthesis. Integrating reliable processes in EBM decision-making is essential for clinical practice.[2,3] Transforming evidence into algorithmic clinical guidelines helps standardise practice and improve treatment outcomes. Ethical challenges e.g., informed consent, transparency, fairness of algorithms used in AI and privacy and security of data. This can be addressed by leveraging scientific advancements from ML and AI tools.[4] In evidence-based practice slow evidence generation process[5] makes its adoption challenging.[6] The time lag between generating evidence and effectively implementing it in clinical settings can be as long as 17 years.[7] Studies have shown that citizens only receive half of the recommended healthcare solutions available under EBM.[8] The healthcare industry is witnessing a revolutionary change from emerging technologies, this progress requires high-throughput sophisticated algorithms enabling practitioners to analyse evidence more quickly and efficiently.[9] With patients increasingly relying on social media and smart devices for health information and self-management, it is crucial to consider the role of AI and ML in evidence-based practice in today's tech-driven world,[10] where social media can influence the healthcare seeking behaviour among patient. In the rapidly evolving landscape of research and evidence-based practice, AI has emerged as a game-changer. Its transformative potential is evident in various stages of the research process, from evidence generation, to evidence gap analysis and evidence synthesis. As the scientific community strives to push the boundaries of knowledge, AI offers unprecedented opportunities to augment research capabilities, accelerate progress and bridge critical gaps.[11] The emergence of e-Health records has generated vast data in healthcare sectors. More than 10 million consultations have happened on e-Sanjeevani alone in India, which is based on e-Health records used for telemedicine. Use of AI in this vast data opens the opportunities to improve clinical decision-making and bridge gaps in knowledge. ML has capacity to handle this vast data and addressing the challenges in decision making. However, the adoption of ML in healthcare is still limited, and its full benefits remain unrealised. Integration between ML and EBM faces epistemological, methodological and ethical challenges. EBM's dominance in clinical sciences poses significant hurdles to the adoption of data science.[11] Individual patient database has been considered as the gold standard form of meta-analysis can make use of AI and ML tool easily. MACHINE LEARNING AND ARTIFICIAL INTELLIGENCE IN EVIDENCE-GENERATION The process of evidence generation is resource-intensive, time-consuming and ethically challenged in terms of informed consent, transparency, privacy and security of data. However, ML and AI-powered tools and algorithms can significantly enhance and streamline this process. ML algorithms assist in identifying research questions, robust study designing and optimising data collection methodologies liberates researchers to focus on higher-level tasks, increasing efficiency and productivity. Previous studies have shown the importance of ML in preclinical research[12] and assistance in identification, generation and mechanistic elucidation of any drug target molecule[13] and autonomous ML-powered laboratories.[14] A Start-up company Trials. AI explored the role of ML in the identification of potential pitfalls and barriers from a study protocol.[15] For information received from wearable and smart devices ML tools assist in monitoring, collection and surveillance of data[12] to selection, identification and recruitment of study participants.[16] However, there are philosophical and operational barriers which limit the full potential of ML in clinical research and domain expert required for further exploration.[12] MACHINE LEARNING AND ARTIFICIAL INTELLIGENCE IN EVIDENCE-SYNTHESIS Evidence synthesis serves as the foundation for evidence-based decision-making, offering a comprehensive understanding of existing knowledge. The integration of AI techniques has significantly enhanced the speed, accuracy and rigor of evidence synthesis processes. AI encompasses a broad range of tools and techniques focused on planning, predicting, understanding and classification of given task.[17] Traditional systematic reviews (without automation tools), although regarded as the gold standard in evidence synthesis, often require months or even years to complete, leading to delayed dissemination of knowledge.[18] However, the integration of AI has revolutionised evidence synthesis in various disease domains, such as diabetic retinopathy, glaucoma and chest imaging during the COVID-19 pandemic. By employing automation and AI, the ability to synthesise research literature has been greatly improved, reducing the workload, increasing speed and enhancing quality. These advancements have demonstrated a significant reduction in evidence synthesis time, from 12 months to a mere 2 weeks.[16] These techniques have been used in conducting 'Rapid Review' during the COVID-19 pandemic. Various automation tools are available to support different tasks in traditional systematic reviews. These tools either assist humans in completing tasks or autonomously perform them. Notable tools include COVID-19-SEE, Search Refiner, Scientific Evidence Explorer, Polyglot Search Translator, De-duplicator, SARA, SRA Helper, RobotSearch, Citation Spider, RevMan, Replicant, Robot Reviewer and EvidenceGRADEr.[17] DistillerSR an online reviewing tool has been used to write rapid reviews.[19] The integration of AI techniques has revolutionised evidence synthesis in healthcare research, enabling rapid assessments of the literature, reducing workload and enhancing speed and quality. These advancements hold great promise in addressing the challenges faced by the healthcare sector and facilitating evidence-based decision-making for improved patient outcomes. REPLICABILITY, BIAS, SAFETY AND ETHICAL CHALLENGES IN ARTIFICIAL INTELLIGENCE RESEARCH Control over confounding variables, limited generalisability may lead to inappropriate conclusion. Lack of reproducibility, complexity and lack of transparency, lack of standardisation and absence of defined policies are potential challenges in AI research. All these challenges complicate the synthesis of ML-based evidence and negatively affect the reliability of outcomes.[4] Studies estimated that only 50% of research outcomes are reproducible.[20] Poor reproducibility due to flaw in design, errors in statistical tests, sample sizes, switching and selective reporting in outcome,[21] and not reporting the negative results.[17] Replication issue can arise due to ML algorithm if taken from same setting, where COVID-19 mortality model was not feasible to replicate by different studies conducted separately. Bias may arise: When an ML algorithm generated from different ethnic group data is applied to different setting where burden data is not available for disease prediction, this will generate bias for the individual being treated based on the ML algorithm. Similarly, automation bias was highest where clinician accept machine advice, instead of maintaining vigilance or validating that advice.[22] Ethical challenges may also arise: There are chances of false negative prediction due to ML algorithms which may cause serious harm to individual. Similarly, privacy and data ownership right may arise as a result of technological advancement. This limits the AI powered tools to evaluate the studies.[23] Emergence of SPIRIT-AI[24] and CONSORT-AI[25] can improve the reproducibility issue in different clinical settings. There are limited studies that report patient safety outcome in AI-enabled healthcare research.[26] Safety issue may arise: In a recent review study conducted by Laranjo et al., on AI safety, found that only one study has reported the patient safety as an outcome major.[26] CONCLUSION AI is revolutionising evidence-based research by empowering scientists and researchers with advanced tools to enhance evidence-generation, identify evidence gaps and synthesise knowledge. By automating multiple tasks, analysing vast datasets and extracting insights from a multitude of sources, AI expedites research processes, accelerates discoveries and facilitates evidence-based decision-making. As AI continues to evolve, it is imperative for the scientific community to embrace this technology responsibly, address ethical considerations and collaborate across disciplines to fully harness its potential. Through the judicious integration of AI, we can unlock new frontiers in knowledge and drive transformative advancements in research for the betterment of society. Challenges cum limitation of automation tools (machine learning and artificial intelligence) Relevant data collection from concerned authorities, potential distorted algorithms due to missing as well as un-cleaned datasets, ethical concerns as well as social concerns and clinical implementation due to the absence of uneven data set increases the quality and credibility of research and causes difficulty to implement the AI and ML in evidence generation.