🚨 Pulse of Innovation 🚨

Breaking news in the healthcare AI

Cyber Outage crippled computer systems across the world: A Wakeup Call and Opportunity

Computers are integral parts of our daily functioning especially at the hospitals. A cyber outage on July 19, 2024, impacted windows computers across the world preventing them from booting and showed what is widely known as the “Blue Screen of Death”. We are sure a lot you encountered this!! This impacted all the common EMR systems such as EPIC, Cerner, Radiology and many more. The healthcare community which is taking its initial steps in the world of artificial intelligence shall view this as a wakeup call and opportunity.

WHAT happened

Computer software have many pieces that require regular updates. Most of our computers get updated behind the scenes while we are not actively using them. These are critical updates from security standpoint and one such file was programmed inaccurately by Crowdstrike but likely still included in the update without proper testing. Once the update was complete, this file failed to boot the computers properly. A solution was found that required the computers to be started in safe mode and manually removing the corrupt file.

Impact on Hospitals (Account from an Intensivist)

No computers were working except 1 that also worked intermittently in the ICU. Handwritten progress notes and order sets were done. Some order sets were faxed to the lab and lab result were printed and faxed back or manually delivered. A pharmacist was physically in the ICU to help with all the medication requirements. All in all it severely impacted their efficiency and increased burden. Some of the elective surgeries were canceled.

Potential impact on AI tools in Hospital

The critical impact of such outage in terms of healthcare AI is losing access to required information. If the EMR systems are integrated with AI tools and portions of patient care depends on such AI tools, then it will obviously cause a critical failure. An earlier Microsoft outage in May 2024 caused issues with functioning of OpenAI, a core and widely used AI tool which is often used to build several healthcare AI applications.

Prevention and Precaution: Can AI be used for this

• All the software undergo testing before releasing for wider use and most of the testing is done by combination of other software programs and humans. An AI powered software testing widely implemented in addition to human testing could provide an additional layer of safety.

• An essential precaution should be availability of a backup computer system in case of such outage particularly in essential services such as hospitals and transportation. A backup system that can start functioning on basic level when the regular system fails to boot. The backup system shall be able to hold critical functions such as notes and orders.

• This outage was not a deliberate cyberattack however it could easily have been one and it has provided us with an opportunity to prevent such incidents in future with more of a hybrid intelligence platform that uses combination of human and artificial intelligence.

🩺 Start-Up Stethoscope 💵 

Trending startups and technologies impacting clinical practice

Evolutionary Scale: AI powered models to understand the proteins that make us all

• Scientists have identified 20 amino acids that assemble to form various proteins based on the genetic codes. The 20 amino acids and the process of genetic programming to make proteins of various sequences, shapes and functions are fundamentally same across the spectrum of biology including humans. The proteins have species level differences/similarities and individual level differences/similarities based on the same three parameters of sequences, shape and function.

• The proteins undergo evolution overtime based on changes in the programming aka genetic code aka the DNA. To better understand the differences in protein generation and the evolution it has undergone in last 500 million years the Evolutionary Scale AI company has generated the ESM3 generative AI model that takes in to account all the aspects of protein synthesis including the historical evolutions. It uses real data and simulated data to understand the process.

• It has over 1 trillion teraflops of computing power with 2.78 billion real world sampled protein data to make a billions of protein structures. These proteins can be simulated or real-world and can create opportunities to understand several processes diseases and treatment.

• The Evolutionary Scale has partnered with NVIDIA to provide the ESM3 model to scientist for their research activities and is available through the open access platform. The company recently received an investment of $142 million with venture partners led by the NVIDIA.

 Evolutionary Scale ESM3 article  

🧑🏼‍🔬 Bench to Bedside👨🏽‍🔬

Developments in healthcare AI research and innovations

Machine learning Enhancing Precision Medicine- Study on Predicting Blood Pressure Phenotypes

• Polygenic risk scores (PRS) are becoming increasingly relevant in studying and managing hypertension, a key risk factor for cardiovascular diseases. PRS quantifies an individual's genetic predisposition to hypertension by aggregating effects from numerous genetic variants identified through Genome-Wide Association Studies (GWAS).

• A recent study published in Nature Scientific Reports developed a non-linear ensemble model that accurately predicts blood pressure (BP) phenotypes using the Trans-Omics for Precision Medicine (TOPMed) dataset:

• Baseline model with demographic and clinical variables and Expanded Model with addition of PRS in the baseline model were evaluated.

• Including multiple PRSs further enhanced the predictive accuracy, particularly for Black, Asian, and Hispanic populations, demonstrating the superior performance of non-linear models in BP prediction especially with multiple variables.

•  Innovation: A key advantage of this study is the model's ability to adapt to new datasets by refitting one component while using the other component as previously trained. This flexibility enhances the model's applicability and accuracy.

•  Integration: The results highlight the potential of non-linear ML models to effectively combine traditional epidemiological risk factors with genetic scores for more accurate BP predictions.

• Circadian Rhythm: One limitation is the model's exclusion of BP's circadian rhythm, which can influence blood pressure readings.

• Sample Size and Diversity: The study had a limited sample size with unequal racial representation, which might affect the generalisability of the findings.

 Conclusion:

This study underscores the promising role of non-linear ML models in integrating genetic and clinical data for predicting blood pressure phenotypes, providing a pathway for more personalised and accurate hypertension management strategies.

🧑🏽‍⚕️ AI in Clinic 🏥

Developments in healthcare AI research and innovations

AI predictive modeling by the EMS helps hospital reduce length of stay by 1 day

Hospitals in post pandemic era have faced multiple challenges from staffing shortages, overcrowding in some units, underutilization other units and equipment redistribution needs. all these issues greatly impact the efficiency of the hospital and its staff.

• The AmeriPro Health, an EMS service provider that closely works with hospitals was also impacted by the issues of hospital level deficiencies in bed allocation causing long ques and delays.

• The AmeriPro Health generated a homegrown AI module that was layered on its existing system that will provide information of peak times, volumes, potential bottlenecks, staffing and equipment needs and many more things.

• The AmeriPro can transmit this information to the hospitals directly to its EHR system. This would allows the hospital get the ordering profile understand its own limitations and shortages and redirect as needed saving transportation costs for themselves and making hospitals better equipped ahead of patient arrival. They also used the same system in staffed the discharge lounge of the hospital that helps with efficient discharge planning allowing the hospital staff to focus on inpatient care.

• With the implementation of this AI module the hospital reduced average length of stay by 1 day with just few weeks of implementation and the company had plans to expand the service to more hospitals.

🤖 Patient First, AI Second🤖

Ethical and Regulatory Landscape of Healthcare AI

Recommendations for Common Ethical Standards for Trustworthy AI : A Summary from the Report Commissioned by the CDBIO

• On December, 2021 , the Council of Europe’s Steering Committee for Human Rights in the fields of Biomedicine and health (CDBIO) published a report addressing the impact of AI on the doctor-patient relationship.

• This report, prepared by Brent Mittelstadt, Senior Research fellow and Director of research at Oxford University offers a comprehensive examination of AI systems through the lens of human rights principles guidelines in the European Convention on Human Rights and Biomedicine of 1997, also known as the “Oviedo Convention”

• Key ethical themes explored in the report focus on higher positive standard of care with respect to the doctor-patient relationship to ensure an undisrupted integration into existing health systems.

  1. Generalisation of performance of AI from trials to real world practice is unproven due to the nascency of AI in healthcare systems.

  2. The shift of care and responsibilities of patients may only be augmented and in many ways can be disruptive to patient care.

  3. The human rights impact on AI include inequality in access, lack of transparency, risk of social bias, dilution of patients wishes and risk of automation bias, de-spilling and displaced liability.

  4. Inconsistency in deployment in terms of speed, scale and prioritisation.

  5. Impact on doctor-patient relationship with the use of AI remains uncertain . May lead to fewer in person interactions and delayed access in emergency situations.

  6. Intelligibility requirements for informed consent : AI models to be trained in a humanly understandable way.

  7. Public register of medical AI systems for transparency : To provide an “algorithmic literacy” among the general public to exercise human and legal rights.

  8. Collection of sensitive data for bias and fairness auditing

  Conclusion:

• The CDBIO report serves as a pivotal document in understanding the ethical implications of AI in the doctor-patient relationship. By addressing critical issues such as inequality, transparency, bias, patient voice, professional skills, and privacy, the report provides a comprehensive framework for embedding human rights in AI-driven healthcare.

• As AI continues to evolve and integrate into medical practice, these considerations will be essential in ensuring that technological advancements align with ethical standards and human rights principles.

• This report is a a must read for clinical practitioners as well as

  engineers working in healthcare AI.