Harnessing AI to expedite evidence generation & reporting


Authored by:   MW & Evidence Team at Genpro Research Inc.

Given the ever-increasing advocacy and preferences for evidence-based healthcare models, clinicians and the research community have been emphasizing and relying on high-quality evidence. On the other hand, the rapid proliferation of biomedical literature has resulted in the infodemic, posing great challenges for medical and scientific writing professionals. Also, critical conditions like COVID-19 have compelled pharmaceutical industries to act fast in terms of research and development as well as the production of repurposed drugs and vaccines. There seemed a huge and dire demand for expedited development of regulatory dossiers for the review and approval of pharmaceutical/vaccine products for emergency use in COVID-19. Furthermore, the increased popularity of real-world evidence (RWE) and digital transformation has inspired the knowledge synthesis community to innovate newer methodologies with patient-centric outcomes. Although it is essential to have a human component in diversified evidence generation, the importance of timeliness cannot be overlooked. The use of various automation algorithms in the medical writing and/or medical informatics process is no longer an enigma. But, it is a fact that limited digital awareness has been a strong barrier against the effective diffusion of automation tools in the Medical Writing (MW) industry. On the world evidence-based healthcare day 2021, Genpro’s experienced medical writing and evidence generation team brings to you a succinct summary of various application areas for MW/HEOR automation.

Strategic Applications of AI & Automation:

Automation in medical and scientific publication

The use of automation in publication writing may span minor activities like identification of synonyms from a thesaurus, creating an abbreviations list to intuitively prepare an initial draft for the writer’s perusal. This may involve textual analysis & generation, statistical analysis, and suggested interpretation. Also, it could have features like context generation (for identification of a problem statement from the reviewed literature), section content parser, and logical sequence tagging. This would help a medical writer to develop a preliminary draft of various sections of the proposed manuscript in no time. Finally, for effective knowledge dissemination, tools to generate quick and simple plain language summaries will become popular amongst MW and the patient communities.

Automation in HEOR/HTA

The most common automation needs from the HEOR perspective have been realized in terms of literature screening. According to published research, the majority (around 80%) of preference was seen in terms of the automated process of literature screening/selection of primary studies. Desired characteristics from an ideal automation tool may include but are not limited to auto-classification of study types, highlighting of key terms, AI-driven re-prioritization, and auto importing of updated references and their full-texts. Another important process requiring sheer attention and yet subject to introducing manual error is data extraction. An intuitive AI algorithm may auto-extract data with higher accuracy. Such tools bring to the table a value-added benefit for complex computational analysis like network meta-analysis with suggested statistical models.

Moreover, AI could help in differentiating between more useful and less useful articles based on their bibliometric analysis. This kind of feature is highly desired by health economists while developing particular disease models or determining values for any decision-analytic modeling inputs.

Lastly, AI can play an important role in developing rapid epidemiology reports to assess disease burden. Pharmacoepidemiologic applications of AI tools have dramatically changed the market scenario of predictive analytics firms deploying AI technologies for predicting treatment patterns, ideal drug doses, or even the incidence/prevalence data of any disease. Furthermore, complementary pattern augmentation framework like algorithms helps to curb the issue of false-positive prediction in low prevalence diseases.

Given such promising utilities, AI has gained immense interest in rare diseases and oncology research besides other therapeutic areas.

Advanced futuristic applications for AI in the RWE domain might cover areas like target trial emulation using RWD. Also, AI may predict long-term survival outcomes in oncology trials beyond the time horizon of an RCT using a transparent machine learning approach using Bayesian networks and patient-focused response model. Additionally, automated curation algorithms may synthesize data of a hypothetical historical comparator arm in cases of an unavailable head-to-head comparison of treatment options.

Regulatory affairs and clinical development

The use of AI in clinical study reports and other regulatory submissions may see a steep increase. Automated/Semi-automated generation of clinical study reports based on the ingestion of source documents is being attempted. This could be possible through intuitive text-mining algorithms applied to unstructured documents, which not only prepare a partial or full CSR but also generate semi-automated safety narratives as well from supplemental data such as Case Report Forms (CRF), CIOMS forms (Council for International Organizations of Medical Sciences) and other relevant datasets from various systems. Another utility of AI algorithms can be considered in terms of SOP auto-update while ingesting regulatory changes. Considering the medical device industry, natural language processing (NLP)-aided CER is the breakthrough for medical device-related regulatory writing.

Thus, it is clear that AI has a lot more to contribute beyond conventional expectations of a medical/HEOR writer or researcher. Cross-functional efforts at the intersection point of life‑sciences and computer science would be essential to unleash and understand novel utility scope of automation tools for faster and accurate generation of credible research evidence. Considering its huge potential, automation efforts are to be planned systematically while setting practical goals without over engineering processes. The right set of technology stakes deployed through a simplistic platform is the key for success of automation in evidence generation and reporting.

References/ Further insights: Available on request


  • ai in clinical trials
You may also like: