The possibilities of artificial intelligence in the realm of healthcare

assets/files/images/04_07_24/learning-ai-resize.jpg

By Brian Phillips, freelance writer.

Introduction

As data in healthcare continues to grow in complexity, the integration of artificial intelligence (AI) is becoming more prevalent. Across payers, care providers, and life sciences firms, various forms of AI are already in use. These encompass diagnosis and treatment suggestions, fostering patient involvement and adherence, and streamlining administrative processes. While AI demonstrates the potential to outperform humans in numerous healthcare tasks, widespread automation of professional roles faces hurdles in implementation. Additionally, ethical considerations surrounding AI's role in healthcare are under scrutiny.

Research findings are already indicating that AI exhibits comparable or even superior performance to humans in critical healthcare tasks, such as disease diagnosis. Presently, algorithms surpass radiologists in detecting malignant tumors and assist researchers in structuring cohorts for expensive clinical trials. Nonetheless, due to multiple factors, we anticipate a considerable timeframe before AI supplants humans in extensive medical processes. This article delves into the potential of AI to automate care elements and addresses impediments hindering its swift adoption in healthcare.

Types of AI of relevance to healthcare

Artificial intelligence comprises a diverse array of technologies rather than a singular entity. While these technologies hold immediate relevance in healthcare, their applications span a wide range of processes and tasks. Below, we outline and describe several key AI technologies pivotal to the healthcare sector.

Machine Learning - Neural Networks and Deep Learning

Machine learning, a statistical method for modeling data and learning through training with data, stands as one of the most prevalent forms of AI. According to a 2018 Deloitte survey of 1,100 US managers, 63% of companies already engaged in AI initiatives were utilizing machine learning. This versatile technique serves as the foundation for various AI approaches, with numerous iterations existing.

In healthcare, traditional machine learning finds common application in precision medicine, predicting effective treatment protocols based on patient attributes and treatment contexts. Typically reliant on supervised learning, these applications necessitate training datasets with known outcome variables, such as disease onset.

Deep learning represents the pinnacle of machine learning complexity, employing neural network models with multiple layers to predict outcomes. Enabled by advancements in graphics processing units and cloud architectures, deep learning finds widespread application in healthcare, particularly in radiology for identifying potentially cancerous lesions. Its integration with radiomics enhances the detection of clinically significant features in imaging data, surpassing the capabilities of previous automated tools like computer-aided detection (CAD).

Natural Language Processing

Since the 1950s, AI researchers have strived to decipher human language, forming the field of Natural Language Processing (NLP). NLP encompasses various applications like speech recognition, text analysis, translation, and language-related objectives. Two primary approaches define NLP: statistical and semantic. Statistical NLP, rooted in machine learning, notably deep learning neural networks, has significantly enhanced recognition accuracy, necessitating ample language data for effective learning.

In healthcare, NLP finds predominant use in generating, comprehending, and categorizing clinical records and research publications. NLP systems adeptly analyze unstructured patient notes, generate reports (such as radiology findings), transcribe patient interactions, and facilitate conversational AI.

Rule-Based Expert Systems

During the 1980s, expert systems built on sets of "if-then" rules stood as the prevailing AI technology, widely employed in commercial and healthcare domains. These systems, pivotal for clinical decision support over recent decades, continue to see extensive use. Many electronic health record (EHR) providers include rule sets within their systems today.

Expert systems rely on human experts and knowledge engineers to formulate rules within specific knowledge domains. While effective initially and comprehensible, they falter as the number of rules increases, often leading to conflicts and breakdowns. Adapting rules to evolving knowledge domains proves challenging and time-consuming. Consequently, they are gradually supplanted in healthcare by data-driven approaches and machine learning algorithms.

Physical Robots

The ubiquity of physical robots is evident, with over 200,000 industrial robots installed globally each year. These machines execute predefined tasks such as lifting, welding, and assembling in industrial settings, while also fulfilling roles like supply delivery in healthcare facilities. Recent advancements have made robots more collaborative with humans, facilitating easier training through hands-on interaction. Moreover, their intelligence is evolving with the integration of various AI capabilities into their operating systems. It is foreseeable that the intelligence enhancements observed in other AI domains will gradually permeate into physical robots.

Surgical robots, first sanctioned in the USA in 2000, endow surgeons with enhanced capabilities, enabling precise and minimally invasive procedures, including gynecologic, prostate, and head and neck surgeries. Despite this, critical decisions remain under the purview of human surgeons.

Robotic Process Automation

Robotic Process Automation (RPA) executes structured digital tasks within administrative spheres, emulating human actions following predefined scripts or rules. Compared to other AI forms, RPA is cost-effective, programmable, and transparent in operation. Contrary to its name, RPA comprises computer programs deployed on servers, leveraging workflow, business rules, and system integration to mimic semi-intelligent user behavior. In healthcare, RPA streamlines repetitive tasks like prior authorization, record updates, and billing. Integration with technologies such as image recognition enables data extraction from faxed images for input into transactional systems.

While traditionally viewed as distinct technologies, there is a growing trend towards their amalgamation and integration. Physical robots are evolving with AI-based enhancements, while image recognition is being fused with RPA. This convergence suggests that future solutions may lean towards composite frameworks, blending the strengths of various technologies for enhanced performance and versatility.

Diagnosis and Treatment Applications

AI software development services empowers healthcare with advanced diagnostic capabilities. Machine learning algorithms analyze vast datasets, aiding in the early detection of diseases and providing healthcare professionals with unprecedented insights. This not only enhances diagnostic accuracy but also paves the way for proactive and personalized treatment plans.

The quest for AI-driven diagnosis and treatment dates back to the 1970s, exemplified by MYCIN's development at Stanford for diagnosing blood-borne bacterial infections. Despite early promise, such rule-based systems failed to gain traction in clinical practice, primarily due to their limited superiority over human diagnosticians and poor integration with medical workflows and record systems.

In recent years, IBM's Watson garnered media attention for its precision medicine initiatives, particularly in cancer diagnosis and treatment, employing a blend of machine learning and NLP. However, initial enthusiasm waned as challenges emerged in teaching Watson to address specific cancer types and integrating it into care processes. Watson, a suite of cognitive services offered through APIs, faced competition from free open-source programs like Google's TensorFlow.

Patient Engagement and Adherence Applications

Long regarded as the "last mile" challenge in healthcare, patient engagement and adherence represent the crucial link between ineffective and favorable health outcomes. Active patient involvement in their well-being and care correlates with improved outcomes across utilization, financial metrics, and member experience, prompting a burgeoning reliance on big data and AI solutions.

While providers and hospitals expertly craft care plans to enhance patient health, the effectiveness of these interventions often hinges on patient behavior modification, such as weight management, appointment scheduling, medication adherence, and treatment plan compliance. Noncompliance, characterized by deviating from prescribed treatments, poses a significant hurdle.

A survey of over 300 clinical leaders and healthcare executives revealed that over 70% reported less than 50% of patients as highly engaged, with 42% indicating that less than a quarter of patients demonstrated high engagement levels.

Given the correlation between deeper patient involvement and improved health outcomes, AI-driven capabilities hold promise in personalizing and contextualizing care interventions. Machine learning and business rules engines are increasingly leveraged to orchestrate tailored interventions throughout the care continuum. This includes deploying messaging alerts and targeted content strategically to prompt actions at pivotal moments.

Furthermore, healthcare is witnessing a growing emphasis on designing effective "choice architectures" to proactively influence patient behavior based on real-world evidence. Leveraging data from provider EHR systems, biosensors, wearables, smartphones, and conversational interfaces, AI-powered software customizes recommendations by comparing patient data against successful treatment pathways for similar cohorts. These recommendations can be disseminated to various stakeholders, including providers, patients, nurses, call-center agents, and care delivery coordinators.

Administrative Applications

Healthcare administration encompasses numerous AI-driven applications, albeit with less potential for revolutionary impact compared to patient care, yet offering significant efficiency gains. This is crucial given the substantial time spent on regulatory and administrative tasks; for instance, the average US nurse devotes 25% of work hours to such activities.

Robotic Process Automation (RPA) emerges as a key technology in this realm, offering versatility in various healthcare applications, including claims processing, clinical documentation, revenue cycle management, and medical records upkeep.

Some healthcare entities have explored the use of chatbots for patient engagement, mental health support, and telehealth services. These NLP-based systems facilitate straightforward interactions like prescription refills or appointment scheduling. However, patient surveys indicate concerns regarding confidentiality, addressing complex health issues, and usability.

Machine learning also holds relevance in claims and payment administration, enabling probabilistic matching of data across disparate databases. Insurers must meticulously verify the accuracy of millions of claims, and machine learning aids in identifying, analyzing, and rectifying coding errors and erroneous claims. This streamlined process yields substantial time, cost, and effort savings for insurers, governments, and healthcare providers, while mitigating financial losses associated with erroneous claims slipping through the system.

Implications for the Healthcare Workforce

Considerable attention has been drawn to the apprehension that AI will usher in job automation, leading to significant workforce displacement. A collaboration between Deloitte and the Oxford Martin Institute projected that AI could automate up to 35% of UK jobs within the next 10 to 20 years. However, various external factors, including technology costs, labor market dynamics, broader benefits of automation, and regulatory frameworks, could mitigate actual job losses to 5% or less.

Thus far, AI has not eliminated any jobs in healthcare. The limited integration of AI into the industry and challenges in incorporating it into clinical workflows and EHR systems have tempered its impact on employment. The healthcare roles most susceptible to automation are likely those involving digital information management, such as radiology and pathology, rather than those with direct patient interaction.

Even within fields like radiology and pathology, AI's penetration is expected to progress gradually. While technologies like deep learning advance image diagnosis capabilities, radiologists' roles extend beyond image interpretation to include collaboration with peers, disease treatment, and image-guided interventions.

Furthermore, clinical adoption of AI-based image analysis is impeded by disparate technology focuses, the need for labeled data for deep learning algorithms, and regulatory and insurance challenges.

Similar complexities are observed in pathology and other digitally-driven medical domains, suggesting minimal healthcare employment disruption due to AI in the next two decades. Additionally, the emergence of new roles to support and develop AI technologies may counterbalance potential job losses. However, sustained or increasing human employment implies limited cost reductions in medical diagnosis and treatment facilitated by AI technologies within this timeframe.

Ethical Implications

The integration of AI into healthcare raises myriad ethical concerns. Historically, healthcare decisions have been predominantly human-driven, and the introduction of intelligent machines to either make or assist in these decisions brings forth issues surrounding accountability, transparency, consent, and privacy.

Transparency emerges as a paramount challenge, particularly with current AI technologies. Many AI algorithms, notably deep learning algorithms utilized for image analysis, are exceedingly complex and inscrutable. Patients understandably desire explanations for diagnoses, yet deep learning algorithms, and even physicians versed in their operation, often struggle to provide comprehensible explanations.

Inevitably, AI systems will err in patient diagnosis and treatment, posing challenges in establishing accountability for these errors. Moreover, instances may arise where patients prefer to receive medical information from empathetic clinicians rather than from AI systems. Additionally, machine learning systems in healthcare risk perpetuating algorithmic biases, potentially associating disease likelihood with gender or race where such factors are not causal.

Navigating the ethical, medical, occupational, and technological implications of AI in healthcare demands proactive measures. Healthcare institutions, along with governmental and regulatory bodies, must establish robust frameworks to monitor critical issues, respond responsibly, and implement governance mechanisms to mitigate adverse consequences. Given AI's profound impact on human societies, sustained attention and prudent policy-making are imperative for the foreseeable future.

The Future of AI in Healthcare

We envision AI playing a pivotal role in shaping the future landscape of healthcare services. Particularly through machine learning, it serves as the cornerstone for advancing precision medicine, a much-needed breakthrough in healthcare. Although initial endeavors in diagnosis and treatment recommendations have encountered hurdles, we anticipate AI mastering these domains. With the rapid strides in AI for imaging analysis, it's foreseeable that machine evaluation will become commonplace for most radiology and pathology images. Moreover, the utilization of speech and text recognition for patient interactions and clinical documentation is poised to surge.

The primary hurdle facing AI in healthcare lies not in its technical capabilities but in ensuring its integration into daily clinical practices. Achieving widespread adoption necessitates regulatory approval, seamless integration with EHR systems, standardization, clinician education, reimbursement by payer organizations, and ongoing updates. While these challenges will eventually be surmounted, their resolution may lag behind the technological advancements. Consequently, we anticipate limited AI utilization in clinical settings within five years, with broader adoption within a decade.

Moreover, it's increasingly evident that AI systems will complement rather than replace human clinicians on a large scale, enhancing patient care efforts. As AI assumes routine tasks, human clinicians can focus on tasks that tap into their uniquely human skills, such as empathy, communication, and holistic patient care. Those resistant to embracing AI collaboration may find their roles evolving, while those adaptable to working alongside AI are likely to thrive in the evolving healthcare landscape.

Add a Comment

No messages on this article yet

Editorial: +44 (0)1892 536363
Publisher: +44 (0)208 440 0372
Subscribe FREE to the weekly E-newsletter