Facebook
  1. Home
  2. Insights
  3. Blog
  4. Can AI improve R&D productivity enough to restore ROI to sustainable levels?

Can AI improve R&D productivity enough to restore ROI to sustainable levels?

As costs continue to rise with no end in sight, what can be done to return ROI to sustainable levels?

  • $1.1 billon

    In 2010

  • $2.1 billion

    In 2018

The mean cost of bringing a new pharma product to market exploded to $2.1 billion in 2018 from $1.1 billion in 2010 – with clinical trials making up a large and growing share (1).

Transforming how clinical trials are conceived, designed and conducted will rely on harnessing the power of digital technologies. The current wave of emerging innovations offers opportunities to improve R&D productivity through automating processes, making efficient use of massive data sets, and supporting early decision-making with predictive analytics and statistical models, among others. 

According to an ICON survey, artificial intelligence (AI) was considered the digital technology with the most potential to improve R&D productivity. Moreover, nearly 80 percent of respondents said their firm plans to use or is using AI or Big Data approaches to improve R&D performance. 

Additionally, as the use of AI rapidly grows, regulatory support for such technologies will increase. For example, last year, the FDA quickly approved an AI-powered device for detecting diabetic retinopathy in primary care offices (2).

ICON survey respondents were optimistic that developments such as these will significantly increase R&D returns. Two-thirds reported they have the potential to increase productivity by 26 percent or more, with 22 percent expecting 51 to 99 percent and 5.5 percent expecting 100 percent or more. Less than one percent expects no improvement. 

Given that an increasing number of sponsors and developers are looking to incorporate AI into their R&D processes, here we discuss different approaches to AI and how to best deploy them.

Expert systems

Expert systems use rules-based algorithms to mimic specific human expertise. One example includes decision-support trees for routine diagnostic tasks, such as differentiating between bacterial and viral respiratory infections for prescribing antibiotics, which are built into every electronic health record (EHR) drug-ordering module.  

Robotic process automation

Robotic process automation (RPA) are specialised computer programs that automate and standardise processes based on pre-defined rules. Of itself, RPA has no ‘intelligence’ – however, increasingly it is typically integrated with other AI technologies to create faster automation, and it’s organisational impact is proving to be significant. When applied to clinical trials this includes:

  • Capturing routine clinical data, such as patient vital signs
  • Collecting operational data, such as drug administration dose and time
  • Testing data to flag safety issues, such as an out-of-range lab result
  • Assessing potential data entry errors, such as duplicated or missing data points
  • Detecting potential protocol deviations, such as the emergence of a non-random variation trend
  • Forwarding clean data to the trial master file and alerting trial monitors to anomalies

Benefits of robotic process automation include eliminating the need for manually transferring data from clinical sites to trial master files, reducing errors and delays, and reducing data loss by detecting anomalies more quickly and reliably than manual review.  

Not only does robotic process automation yield immediate efficiency benefits, but also, it lays the groundwork for incorporating massive data sets from EHRs, mobile devices, automated image scanning, and individual patient genomic and molecular data. 

Linking trial stages

Leveraging robotic process automation includes linking processes across study stages. This involves considering the final outputs — which include data supporting regulatory approval and commercial payment — in the design of every step of a study and automatically adjusting those steps when a change occurs. Automatically linking study requirements from end-to-end can significantly reduce delays and the manual effort required to fully implement a protocol amendment. 

Also, adopting a linked process automation approach facilitates portfolio management decisions by allowing sponsors to model how specific changes in study protocols might affect development timelines.  

Machine learning 

Machine learning is defined as algorithms and statistical models that computers use to perform tasks. This allows for more flexibility than rules-based expert systems because it allows the computer to improve its performance, or “learning,” based on training, instead of relying on programmers to provide a fully worked out set of rules.

Moreover, the greater processing power of modern computers is now enabling deep machine learning, in which the device, itself, extracts features from a raw data set and has multiple layers of optimisation processing modelled on how neurons process information. This allows the machine to discover patterns in the data that do not depend on the insight or expertise of a human programmer, making deep machine learning more powerful for assessing images and complex data sets. For example, AI deep learning machine techniques have improved the formulae for predicting the power of intraocular lenses needed to reach uncorrected 20/20 vision after cataract surgery (3).

Deploying AI and its challenges

The power of AI for revolutionising pharma R&D is evident. For instance, Alphabet’s Deep Mind program can predict the risk of heart attack and stroke from retinal images. Additionally, image analysis is used to assess oncology pathology and heart rhythm with accuracy that rivals or exceeds experienced clinicians. AI has many applications for improving clinical research returns, including:

  • Patient identification
  • Site selection
  • Patient monitoring and support 
  • Cohort composition

Yet, AI must be handled with care to ensure it is producing valid, reliable results. Its unstructured nature can lead to results that are not useful or defy logic. In addition, the cost and complexity of developing AI solutions can prevent the adoption of its implementation. Harnessing digital technology to transform clinical trials will require sponsors to develop or acquire a range of capabilities. To successfully deploy AI techniques, collaboration with outside experts, including clinical study process experts, will be critical. Steps for moving forward include:

  1. Identifying and developing operational and IT expertise and capacity
  2. Developing statistical expertise
  3. Developing global reach
  4. Managing change 

AI has the potential to fundamentally improve pharma R&D. Overcoming the obstacles of integrating AI, while ensuring trial data and process integrity, requires an understanding of study processes. Therefore, adopting a strategic partnership to acquire insights from CROs and others with extensive experience will be critical to designing and testing AI processes to avoid potential pitfalls, and maximise benefits.

Digital Disruption in Biopharma

References:

(1) Terry C, Lesser N. Unlocking R&D productivity: Measuring the return from pharmaceutical innovation 2018. Deloitte Centre for Health Solutions, 2018. https://www2.deloitte.com/content/dam/Deloitte/global/Documents/Life-Sciences-Health-Care/deloitte-uk-measuring-return-on-pharma-innovation-report-2018.pdf

(2) Lee KJ. AI device for detecting diabetic retinopathy earns swift FDA approval. American Academy of Ophthalmology, April 12, 2018. https://www.aao.org/headline/first-ai-screen-diabetic-retinopathy-approved-by-f

(3) Hill, W. American Society of Cataract and Refractive Surgery, San Diego, 2019.

In this section

Connect with us