🚨 Pulse of Innovation 🚨

Breaking news in the healthcare AI

AlphaFold 3 is Open Source! But how will it help the future of research?

AlphaFold 3 is poised to significantly influence both the cost and timeline of drug discovery in several key ways:

Accelerated Early-Stage Research

AlphaFold 3 enables researchers to rapidly screen and model thousands of molecular interactions computationally, dramatically reducing the need for costly and time-consuming laboratory experiments in the early stages of drug discovery. This computational approach allows testing tens of thousands of options in silico, compared to the traditional method of trialing only hundreds of options physically in the lab.

Reduced Time to Market

• The efficiency gains from AlphaFold 3 could potentially shorten the overall drug development timeline by several years.

• Estimates suggest that, with this technology, the typical 10-year timeline for bringing a drug to market could be reduced to around 7 years.

Cost Savings

• While exact figures are not provided, AlphaFold 3's computational approach is described as "a small fraction" of the cost of traditional medicinal chemistry methods.

• Reducing physical lab tests and resources required for early-stage discovery translates to significant cost savings.

Improved Success Rates

• AlphaFold 3's ability to predict protein-ligand interactions and binding sites more accurately increases the likelihood of identifying viable drug candidates early in the process.

• This improved accuracy could lead to fewer failed candidates in later stages, potentially reducing overall costs and timelines.

Democratization of Drug Discovery

• The availability of AlphaFold 3 through the AlphaFold Server for non-commercial research use democratizes access to cutting-edge computational tools

• This accessibility could enable smaller research groups and startups to engage in drug discovery with more realistic chances of success, potentially increasing innovation in the field

Challenges and Limitations

• The technology primarily impacts early-stage drug discovery. Later stages, including clinical trials and regulatory processes, will still require significant time and resources

• Success rates in later stages of drug development remain to be seen and will still face traditional challenges.

• Deploying AlphaFold 3 at scale requires significant investment in IT infrastructure, data pipelines, and domain expertise.

AlphaFold 3 has the potential to substantially reduce both the cost and timeline of early-stage drug discovery by enabling rapid, accurate computational modeling of molecular interactions. While it doesn't eliminate all challenges in drug development, it represents a significant step towards more efficient and cost-effective drug discovery processes.

🧑🏼‍🔬 Bench to Bedside👨🏽‍🔬

Developments in healthcare AI research and innovations

AI based Screening for Diabetic Retinopathy: Trends, Challenges and Opportunities.

• Diabetic Retinopathy is a common complication in patients with diabetes; however screening for this condition is often not conducted optimally.

• Programs such as Stanford’s Tele-ophthalmopathy Autonomous Testing and Universal Screening initiative underscore AI’s potential to enhance the detection of diabetic retinopathy. A research letter published in JAMA Ophthalmology highlighted the use of AI based tools for retinopathy screening in the United States.

• A retrospective cohort study utilised TriNetX federated database to analyse national trends in retinopathy screening in patients with diabetes. The study specifically tracked the use of Current Procedural Terminology(CPT) code 92229, which was introduced in 2021 as a reimbursement mechanism for AI based screening.

Key Findings :

• Among 4,959,890 patients with diabetes in the TriNetX database, 4.2% of the patients received one of the targeted codes. Of these,

  1. 80.3% underwent OCT imaging

  2. 35.0% underwent fundus photography

  3. Traditional remote imaging using CPT codes 92227 and 92228 was performed in 1.0% and 2.5% cases respectively.

• Since 2021 only 0.9% of diabetic patients received AI based screening for diabetic retinopathy. This rate increased by 1% between 2021 and 2023.

• In contrast, traditional diabetic retinopathy screening rates rose by 185.4% during the same period.

• AI- based imaging resulted in higher referral rates to OCT imaging compared to traditional remote imaging.

• Over 80% of the patient who underwent AI- based imaging were from the South with nearly half of these patients identifying as Black.

Conclusion:

Although FDA’s approval of AI- based systems holds promise for the early detection of diabetic retinopathy , widespread adoption remains limited dude to challenges such as availability, costs, integration into existing workflows and limited awareness of such technologies.

JAMA Ophthalmology 

Matching Clinical Trial Eligible Patients to Relevant Studies/Investigators: Kitsa.ai

• Clinical trials are often conducted in highly regulated setup with restrictive inclusion/exclusion criteria to ensure certain patient characteristics are met. This often leads to delays in patient recruitment and increase in the cost to conduct the trial. Clinical trials are usually conducted at reputed academic medical centers however in some instances many such sites go on to recruit zero patients for a clinical trial despite their perceived reputation.

• Kitsa is a Patient Discovery Platform powered by AI and Large Language Models designed to address critical challenges in patient recruitment for clinical trials. It enables Sites and Providers to perform deep searches across structured and unstructured health data that enables highest level of patients identification for recruitment.

• Clinical trial sponsors could benefit with Kitsa by performing pre-site selection run through that identifies high volume sites matching their trial’s requirements. This would ensure selection of high performing sites with diverse patients population.

• Kitsa could be beneficial to non-academic centers and physician practices who would have high volume but would often not be part of any clinical trials due to lack to research infrastructure. Kitsa through its strategic partnerships could support research activities at such centers by not only identifying such sites but also providing research related technology and human resources.

Navigating the EU AI Act in Healthcare

The EU AI Act, approved in 2024, introduces a comprehensive legal framework for regulating AI technologies, including their applications in healthcare. It classifies AI systems based on risk levels, with medical AI/ML-enabled devices categorized as “high-risk.” Obligations include risk management, data governance, transparency, and human oversight, applying to providers globally if their systems are used within the EU.

Implications for Medical AI Products

Compliance Requirements:

• High-risk AI systems in medical devices must meet stringent technical documentation, risk management, and performance testing standards.

• Alignment with existing EU Medical Device Regulation (MDR), In Vitro Diagnostic Regulation (IVDR), and GDPR is required.

Quality Management Systems (QMS):

• Integration of AI-specific QMS with existing medical device QMS (e.g., ISO/IEC 13485).

• Mandatory documentation of AI design, testing, and performance metrics.

Market Impact:

• Challenges for SMEs due to high compliance costs and technical demands.

• Extra-territorial application akin to GDPR, impacting international providers.

These regulations bring a lot of challenges.

Regulatory Complexity: Dual compliance with MDR/IVDR and AI Act may cause confusion and burden manufacturers, especially SMEs.Delays are expected due to the limited capacity of notified bodies to certify high-risk systems.

Innovation Risks: High documentation and risk management requirements might deter startups and new entrants. Rapidly evolving AI technologies like large multimodal models face static and rigid regulatory frameworks.

Operational Hurdles: Misalignment between stakeholders, including policymakers, developers, and certifying bodies, could delay effective implementation.