The Impact of AI on Oncology Care Efficiency and Mortality Rates

The Impact of AI on Oncology Care Efficiency and Mortality Rates
Study: Uses and limitations of artificial intelligence for oncology

In a recent publication featured in the esteemed journal Cancer, scholarly articles explore the merits and constraints of AI precision medicine methodologies in the realm of oncology research and therapy.

Background:

The inquiry delves into the diagnostic and prognostic efficacy of artificial intelligence (AI) algorithms, particularly emphasizing the role of AI-driven chatbots (generative AI) in fostering favorable outcomes in cancer treatment over the past decades.

Additionally, it addresses the current obstacles hindering widespread AI integration and proposes regulatory frameworks to enhance the efficacy of these algorithms in the foreseeable future.

Precision Medicine and its Clinical Application in Anti-Cancer Therapies: Referred to colloquially as ‘personalized medicine,’ precision medicine represents a therapeutic paradigm that takes into account a patient’s individual genetic profile, environmental exposures, and lifestyle factors.

In contrast to conventional medical paradigms, which adhere to a ‘one size fits all’ approach, precision medicine offers manifold advantages, particularly in fields like oncology, where patient-specific characteristics (such as tumor profiles) significantly influence treatment outcomes, surpassing those achieved through generic chemotherapy.

Advancements in oncological practices have garnered significant scientific interest, with studies indicating a remarkable 33% reduction in cancer mortality rates over the past three decades.

However, escalating environmental pollutants and suboptimal lifestyle choices have concurrently impeded progress in the field, owing to the escalating variability of carcinogenic agents.

Precision medical interventions, particularly those harnessing artificial intelligence (AI) algorithms, hold promise in overcoming this limitation inherent in conventional, standardized medical approaches. These AI-driven methodologies empower researchers and clinicians to discern previously undetected patterns in patients’ radiological scans through the application of machine learning (ML) and deep learning (DL) technologies.

Regrettably, despite the development and rigorous testing of numerous AI algorithms for the management of cancer care, the integration of these technologies into mainstream medicine remains uncommon.

Prominent barriers to the adoption of AI models in research include their substantial initial implementation expenses, the lack of human interpretability regarding algorithmic outcomes, and the limited human oversight and validation of algorithms post-implementation.

Moreover, research endeavors across various facets and stages of cancer care lack uniformity, with a significantly larger body of literature focused on cancer diagnosis (exceeding 80%) compared to treatment and post-chemotherapy care.

Nevertheless, notwithstanding these challenges, the implementation of AI in oncology has significantly advanced the field, facilitating innovative diagnostic, prognostic, and chat-based information access for clinicians and their patients alike.

The current review delves into these advancements, delineating the merits and demerits of existing AI implementations while also addressing conventional and prospective hurdles in the widespread adoption of AI.

Furthermore, it proposes policy reforms that have the potential to alleviate the global burden of cancer, one of the most lethal and incapacitating chronic illnesses worldwide.

About the Review:

The present review endeavors to contextualize three common applications of precision medicine (particularly AI implementations) in cancer care:

  1. Cancer classification and diagnosis,
  2. Cancer prognostication, and
  3. The utility of AI chatbots and other sophisticated language model (LLM) technologies in streamlining clinical workflows.

Drawing from an extensive analysis of over 40 primary research studies, the review aims to elucidate policy and implementation enhancements that could further enhance reductions in cancer mortality rates in the years ahead.

Diagnosis:

Cancer diagnosis poses a formidable challenge, particularly in early-stage and recurrent cancers, as patients at these stages often appear clinically asymptomatic to human observers.

AI algorithms, especially those employing machine learning (ML) techniques, trained on vast datasets of cancer diagnostic images (including radiology scans, pathology images, and even patient-provided smartphone photographs), demonstrate efficacy in identifying, categorizing, and diagnosing such cancers, particularly when subtle image data features elude human perception.

Even in cases where human oversight is indispensable, AI technologies, including computer-aided detection (CAD) algorithms, variants of deep learning (DL) frameworks, can pinpoint regions of interest (suspicious pixels in cancer diagnostic images), thereby assisting clinicians in their diagnostic assessments.

Surprisingly, in certain instances, AI algorithms have exhibited superior diagnostic precision and efficiency compared to their human counterparts.

“Commonly used AI algorithms for image classification are convolutional neural networks (CNN), deep learning architectures that extract identifying features for each group and use the resulting schema for a new classification task. The algorithm assigns a probability for each output class, and the image is classified into the group assigned the highest probability. The accuracy of the AI tool is measured by comparing the algorithm classifications with clinician classifications, referred to as “ground truth”.”

The principal advantage of integrating AI into diagnostics lies in melanoma and breast cancer screening, where early detection stands as the pivotal factor influencing favorable mortality and morbidity outcomes. Regrettably, AI is plagued by profound biases associated with its training, significantly impeding its application in the field.

The underrepresentation of training data, coupled with disparities in data presentation and variability in data acquisition and processing, renders most AI models non-generalizable. This, in turn, obstructs their integration into global oncology protocols.

“Modifications along the algorithm development pipeline can help mitigate these concerns. Training data can be expanded to include representative images from all demographics (e.g. skin color, ages, and body types). Training sets with image data should include samples taken from different angles, lighting, and equipment; and AI technologies should accommodate changes in image acquisition technology by retraining the model with new images.”

Prognostication:

Anticipating patient outcomes stands as one of the paramount initial clinical intervention measures undertaken by medical practitioners. This enables the customization of clinical interventions to enhance or avert the most adverse clinical consequences.

Regrettably, prognostication conducted by humans has historically been prone to significant inaccuracies. Reports suggest that 63% of prognoses tend to overestimate outcomes, while 17% underestimate patient survival.

“The consequences of inaccurate predictions in oncology include increased emotional burden on patients and their caregivers, inappropriate allocation of resources, decreased trust in the patient–physician relationship, and delay in crucial therapeutic or end-of-life interventions. AI-based risk prediction models that generate individualized estimates on prognosis have augmented clinician assessments of risk and aided personalized care decisions in oncology.”

Electronic health records (EHR)-based machine learning (ML) models exhibit tremendous potential in this realm. They have demonstrated the ability to forecast cancer outcomes months or even years in advance, equipping clinicians with the foresight necessary to adequately prepare for oncological eventualities.

Furthermore, these models possess the capability to assess the most efficient and cost-effective clinical intervention pathway, thereby conserving substantial clinical manpower and minimizing patients’ financial outlay, consequently alleviating the overall disease and socioeconomic burden.

Alas, the deterministic nature of most of these models renders them vulnerable to alterations in model outcomes upon the incorporation of novel, yet unaccounted for, data generation methodologies.

The phenomenon known as ‘performance drift,’ characterized by the gradual deterioration in model efficacy over time, may render subsequent model predictions inaccurate and unreliable unless regular updates to the modeling algorithm and rigorous human results validation are diligently conducted.

In this arena, the quality of training data, frequent human model validation, and cross-cancer data-sharing initiatives may serve as potential remedies to overcome these challenges in the future.

Chatbots and Conclusions: Cutting-edge conversational chatbots, epitomized by platforms such as ChatGPT, Google Gemini, Microsoft Copilot, and others, are redefining the landscape of information acquisition and processing for both professionals and laypersons across the World Wide Web.

These generative AI applications harness the prowess of large language models to generate novel content tailored to the user’s requirements.

Regrettably, exploration of chatbot applications in oncology indicates that the technology is still in its infancy, lacking adequate support and, more critically, lacking policy-sanctioned implementation in clinical settings.

“The adoption of chatbots for medicine relies on achieving both understandable language and conveying complex medical topics accurately, which current algorithms cannot do consistently because readability scores vary by the user’s verbiage of the prompt. Although medical knowledge expands each day, algorithms are not continuously updated to accommodate this change. As a result, the chatbots that are not trained on updated information can become unreliable and more inaccurate with time.”

Collectively, these individual, discipline-specific advantages and disadvantages portray a compelling narrative – while the significance and pertinence of AI integration in oncological research are undeniable, the computational and raw data requisites of these models have only recently begun to be fulfilled.

With the advancement of enhanced modeling frameworks, the availability of larger and more detailed datasets, and the heightened scientific validation of their precision and dependability, AI models emerge as potent instruments in the oncologist’s arsenal against this formidable ailment and hold the potential to alleviate the majority of the burden associated with cancer care from human medical practitioners in the foreseeable future.

For more information: Uses and limitations of artificial intelligence for oncology, Journal Cancer, https://doi.org/10.1002/cncr.35307