AI at the Helm: Transforming the Medical Industry

AI at the Helm: Transforming the Medical Industry

James Doulgeris,Chairman Population Health Advisory Board RSDSA

In 2025 or 2026, if the rate of progress in AI development keeps apace, today’s capabilities will become rapidly outdated by a seminal transformation of AI from a language-based model to a science-based model. This will be brought forth by the integration of images, sounds, and multi-media and the ability of these systems to integrate, collate, stratify, and comprehend them on a superhuman level. A good analogy is that interactive models like Gemini and Chatgpt will evolve from a humanities professor to a superhuman genius theoretical physics professor with access to all human knowledge and the ability to use it intuitively in mere seconds.

1. Can you elaborate on how AI is currently being utilized in healthcare, particularly in areas like diagnosis, treatment optimization, and patient care management?

Firstly, regardless of size, providers will be purchasing third-party AI powered clinical decision-making systems. Trying to build a system from scratch is simply like trying to invent a new computer operating system when superior ones are available at a fraction of the cost. Each off the shelf system has its strengths and weaknesses, so the selection process is the most important, as is the decision that will be made. A selection committee comprised of physicians representing each affected discipline, IT leadership, administration and, most importantly, an impartial third-party expert is essential to make the best choice for your facility or practice. The following provides a guide to what to look for in assessing the best system for your needs.

The top ten AI powered healthcare decision making systems are GE Healthcare, Siemens Healthineers, Canon Medical Systems, Philips, Aidoc, Google, Merative, Microsoft, NVIDIA, and in no particular order.

Taking a broad brush, Artificial Intelligence (AI) is revolutionizing healthcare in three primary areas, each of which are invaluable contributors to value based care:

a. Diagnosis: AI aids in accurate and swift diagnoses by analyzing medical imaging, such as X-rays, CTs and MRIs, to detect abnormalities at a level of accuracy and detail not always a strong point with human radiologists. Conversely, while it also assists in analyzing patient data to identify patterns and predict diseases, AI does not have the intuitive ability humans have in identifying those patterns and associating them with various disease states. The two working together, however, represent a potent working relationship that elevates the quality and efficiency of medical imaging analysis to a level not achievable by either by themselves.

b. Treatment Optimization: AI analyzes vast amounts of patient data, including medical history, diagnostics and genetic information looking for patterns and matches to known disease states. Also known as precision medicine, it customizes the most effective therapies to individual patients, enhancing efficacy and safety by reducing adverse effects.

c. Patient Care Management: Using predictive analytics using patient historical data, patient vitals and other real-time data, imaging, and other diagnostic results, AI presents personalized recommendations. This guides healthcare providers in delivering proactive and personalized care to improve patient outcomes and satisfaction with optimized efficacy, efficiency, and safety.

2. What are some of the most significant advancements AI has brought to medical imaging analysis, and how has this impacted diagnostic accuracy and speed?

I see three primary areas where AI has enhanced medical imaging, bringing it to an entirely new level of effectiveness and efficiency:

a. Enhanced Accuracy: AI algorithms, while not yet perfect, analyze complex medical images with a level of precision, detecting subtle abnormalities that might be missed by even the most skilled professionals. This aids radiologists in making more accurate diagnoses while reducing the risk of errors, stress and fatigue caused by volume and monotony.

b. Improved Speed: AI systems process medical images in seconds, allowing for high throughput of routine exams and more time to examine complex cases. These capabilities provide treating physicians with more accurate and detailed reports from which to develop treatment plans with greater confidence and improved outcomes.

c. Early Disease Detection: AI imaging analytics have proven their ability to detect early disease states through their analysis of medical images with far greater accuracy than human review. Early detection and treatment are proven to be the most effective means to prevent disease states from reaching more mature stages where they cause irreversible damage and are much harder to treat. One of the most promising areas is the early detection of rare diseases when they are the most treatable.

3. AI-driven personalized medicine is gaining traction. Could you discuss how AI algorithms are helping tailor treatments to individual patients based on their genetic makeup, medical history, and other factors?

Tailoring treatments by assessing their genetic makeup, medical history, diagnostics, and physical examinations using AI powered algorithms is becoming a best practice in more complex cases where conventional diagnostic techniques prove to be inadequate. By analyzing diverse data sources, AI algorithms create personalized therapy recommendations adapted to a patient’s unique needs. This approach maximizes treatment effectiveness while minimizing adverse effects resulting in optimized outcomes and safety using three primary methodologies:

a. Genetic Analysis: By analyzing genomic data to identify genetic variations that influence treatment responses using AI powered algorithms, targeted therapies can be identified to optimize treatment and pharmaceutical protocols enhancing outcomes and safety.

b. Medical History Integration: When genetic analysis is integrated with a patient's medical history, past treatment responses, comorbidities, and medication tolerances further enhance outcomes and patient safety.

c. Considering Additional Factors: In the most complex cases where a definitive or differential diagnosis is elusive, additional factors including lifestyle, environmental exposures, and socio-economic determinants can fine-tune personalized medicine to an exacting level.

4. Ethics and privacy are crucial in healthcare AI. How do you ensure that AI systems maintain patient confidentiality and adhere to ethical guidelines?

As with any internal data system, there are rules to protect patient privacy and ethical guidelines within established best practices. Since healthcare AI systems typically bridge to external sources to retrieve or interact with disparate data sources from imaging to data lakes and outside analytic systems, the following ten requirements must be embedded in the system to ensure compliance with HIPAA and ethical requisites:

a. An Ethical Review Board: A purpose directed internal ethics committee or board represented by clinical, IT and administrative sectors should collaborate with the vendor in the deployment process to ensure that AI systems comply with HIPAA, ethical guidelines and regulations particular to your facility or practice.

b. Data Encryption: Strong end to end encryption is an absolute necessity so that even if unauthorized access occurs, the data remains unreadable.

c. Access Control: Robust role-based access control (RBAC) with frequent password refreshment is your responsibility.

d. Anonymization and Pseudonymization: Patient deidentification before using systems for training or statistical analysis by non-clinical personnel protects patient privacy while allowing for these functions to be meaningfully conducted.

e. Transparent Logic Trees: Look for algorithms that are transparent, easily followed, and explainable to allow clinicians to understand how decisions are made to ensure they align with current clinical best practices and ethical principles.

f. Regular Audits: Regular audits ensure systems are operating as intended and are not inadvertently exposing patient data or making biased decisions. Audits are also essential to ensure clinical practices are up to date with best practices. Your vendor should provide these services; however, belt and suspender programs ensure nothing is missed.

g. Compliance with Regulations: Ensure that the system is and will remain up to date with relevant regulations such as GDPR, HIPAA, and other federal and state privacy laws.

h. Continuous Training: Regular training and testing of IT staff and clinicians on ethical principles and the importance of patient confidentiality ensure that these principles are upheld throughout the working life of the programs.

i. Informed Consent: Obtain informed consent from patients before using their data for training and analytic models or making clinical decisions based on AI recommendations when PII and PHI are disclosed. If your vendor does not provide a program for this, develop one internally.

j. Accountability and Responsibility: Clearly define roles and responsibilities for everyone involved in the development and deployment of AI systems, including defined levels of access and accountability for each role that is signed by each participant. This should include consequences for any breaches of confidentiality or HIPAA violations.

5. One of the challenges with AI in healthcare is the interpretability of algorithms. How do you address the "black box" issue to ensure transparency and trustworthiness in AI-driven decisions?

Explainable AI (XAI), or AI algorithms and models that are easily transparent and explainable allows clinicians to understand the rationale behind AI-driven recommendations. There are seven elements that can achieve this goal. Look for them in any purchased system:

a. Interpretable Models: While machine learning models such as decision trees, rule-based systems, or linear models may sacrifice some predictive performance, they offer greater interpretability. Complex black-box models like deep neural networks are inherently opaque due to their complexity unless the user interface includes a transparent decision-making tree that allows the clinician to follow the logic of the clinical recommendation.

b. Post-hoc Explanation Techniques: Features such as SHAP (SHapley Additive exPlanations), LIME (Local Interpretable Model-agnostic Explanations), or Integrated Gradients provide post-hoc explanations for AI model predictions by highlighting which features contributed most to a decision.

c. Model Documentation: Every system should include comprehensive developmental data preprocessing steps, model architecture, training procedures, and validation methods. This documentation helps clinicians and regulators understand how the AI system works.

d. User-Friendly Interfaces: Design user interfaces need to provide recommendations in a clear and intuitive manner with explanations of how recommendations were arrived upon. The best systems include visualization packages to aid in conveying complex information effectively.

e. Transparency Reports: Up to date reports detailing the performance, limitations, and potential biases of AI powered clinical decision systems.

f. Continuous Evaluation and Validation: Regular evaluation and validation of AI models to ensure they are up to date with real-world best practices are essential responsibilities for the review committee or board.

g. Regulatory Compliance: Ensure that AI-driven healthcare systems comply with up-to-date regulatory requirements for transparency and accountability, such as the FDA's guidelines for AI in medical devices by conducting an annual review.

6. Integrating AI into existing healthcare systems can be complex. What are some strategies for successful AI adoption in hospitals and clinics, considering factors like interoperability, data security, and staff training?

Purchasing a system that is compatible with your existing EMR or EHR system is an absolute must. Your vendor should be responsible for system integration in collaboration with your existing EMR provider. If they are not compatible and cross-licensed, they are automatically disqualified.

7. The role of AI in drug discovery and development is expanding. Can you discuss how AI algorithms are revolutionizing the pharmaceutical industry, from molecule design to clinical trials?

AI powered analytic systems are revolutionizing the pharmaceutical industry from molecular design to clinical trials, accelerating drug discovery, optimizing drug development processes, and personalizing treatments. Here are the major areas of impact:
Molecular Design and Drug Discovery:

a. Virtual Screening: By analyzing vast databases of molecular structures, these systems can predict which compounds are most likely to interact with a target protein or biological pathway, speeding the process of identifying potential drug candidates sometimes by years.

b. Generative Models: Generative adversarial networks (GANs) and variational autoencoders (VAEs) can create novel molecular structures facilitating the discovery of new chemical entities with targeted therapeutic potential.

c. Drug Repurposing: By analyzing large-scale biomedical data, existing drugs may be repurposed for new indications, vastly reducing the time and cost required to discover new drugs for the same purpose.

Optimization of Clinical Trials:

a. Patient Stratification: By analyzing diverse data sources, including genomic data, electronic medical records (EMRs), medical imaging, and other disparate health data to identify patient subgroups that are more likely to respond to a particular treatment, the clinical trial process can be significantly streamlined.

b. Predictive Analytics: The outcomes of clinical trials, such as patient response to treatment or the likelihood of adverse events, can be predicted with statistically significant accuracy. This helps pharmaceutical companies make data-driven decisions to optimize trial design.

c. Real-world Evidence (RWE): AI systems can collect and analyze real-world data from novel sources such as wearables and social media to generate real-world indications on treatment effectiveness, safety, and patient preferences, supplementing traditional clinical trial data.

Drug Development and Manufacturing:

a. Process Optimization: Optimizing drug development processes such as formulation design, dosage optimization, and manufacturing processes improves efficiency, reduces costs, and minimizes waste.

b. Quality Control: Real time monitoring and analysis of production processes can detect deviations from expected quality standards and enable proactive interventions to maintain product quality and safety.

Personalized Medicine:

a. Precision Prescribing: AI algorithms can analyze individual patient data, including genetic information, biomarkers, and clinical history, to predict optimal treatment regimens tailored to each patient's unique characteristics. This approach, known as precision medicine, aims to maximize treatment efficacy while minimizing adverse effects.

b. Companion Diagnostics: AI algorithms can identify biomarkers or molecular signatures associated with drug response or disease progression, facilitating the development of companion diagnostics to guide treatment decisions and monitor patient outcomes.

Overall, AI algorithms are reshaping the pharmaceutical industry by accelerating drug discovery, optimizing clinical trials, streamlining drug development processes, and enabling personalized medicine approaches. These advancements have the potential to significantly improve patient outcomes, reduce healthcare costs, and drive innovation in healthcare.

8. AI has the potential to reduce healthcare disparities. How can AI technologies be leveraged to improve healthcare access and outcomes for underserved communities?

AI can provide access to specialists in rural areas and third world countries where specialists are not available by providing these services to primary and secondary care providers who would not normally have access, particularly to provide more complex diagnoses including differential diagnoses. This expanded access can also more importantly identify potential nascent pandemics among the most vulnerable populations and bring first world medicine to areas where it has not been available before. In all, by improving the quality of medical services where none has been available before, it benefits the health status of regional populations overall and potentially world health.

9. Regulatory frameworks for AI in healthcare are evolving. What are the current challenges and opportunities in ensuring regulatory compliance while fostering innovation in AI applications?

AI has the inherent capability to ensure regulatory compliance on a local, state, and country basis freeing itself as a dynamic platform to grow and thrive in different regulatory environments without diluting its overall capabilities or diminishing its ability to grow and advance as a system.

10. Interdisciplinary collaboration is essential for AI success in healthcare. How do you encourage collaboration between data scientists, healthcare professionals, policymakers, and industry stakeholders to drive AI innovation?

Because AI platforms are dynamic machine learning environments, they not only lend themselves to collaboration, but they are also inherently built to upgrade within collaborative distributed networks.

11. Looking ahead, what are the key trends and future developments you anticipate in AI's role in transforming the medical industry, and how should stakeholders prepare for these changes?

Soon, there will be rapid adoption of clinical AI systems in imaging and other diagnostics where speed and superhuman accuracy and the ability to detect early disease states followed by deployment in rural and other areas where access to specialists is difficult or impossible. AI is already well established in assembling disparate information in EMR systems to provide an easily consumable clinical summary to increase physician efficiency, productivity, patient safety and reduce physician burnout by companies like Navina, Innovaccer and EPIC.

12. In your opinion, what do you see as the most significant opportunity or challenge that AI will face in the next five years in its ongoing transformation of the medical industry, and how do you think stakeholders should address it?

As I noted in the beginning, clinical AI systems will quickly transition from language-based systems to science-based systems. This will begin to displace and repurpose traditional healthcare jobs, particularly in the clinical sectors beginning with primary care, radiology, pathology, and a range of diagnostic professionals. That does not mean these positions will be eliminated; however, their roles will be transitioned from more routine tasks to practicing at the top of their licenses. Preparing for these transitions should begin now with advanced training to better understand managing differential diagnoses and recognizing disease patterns best suited to human intuition.

Allied health positions will be the last to transition because people will still have to collect samples, perform imaging studies and care for patients in traditional roles, however, eventually, cost pressures will eventually cede these roles to robotics and increasing automation where the precision and repeatability of such mechanical systems will prove to be far superior to human intervention. In these cases, and in physicians and other clinicians, human interaction is still and will remain an important part of care that cannot be replaced by machines. The decision process and planning in making those determinations is best started now so proper thought can be applied to these eventual changes making adaptations much easier when the time comes to change in real life.