Improving Patient Outcomes With AI: Case Studies And Success Stories

Introduction

Imagine being diagnosed with prostate cancer and your doctor tells you a treatment plan will be tailored precisely to your unique case, thanks to a cutting-edge AI system. This isn't a scene from a futuristic movie but a reality today. With prostate cancer treatments often being hit-or-miss, an AI tool now offers a solution that outperforms traditional methods by considering the complexity of individual cases. This is just one of the many ways AI is revolutionising healthcare, improving patient outcomes, and transforming lives. Let's explore the remarkable success stories of AI in healthcare.

Prostate cancer care revolution

Prostate cancer treatment faces challenges due to its variable prognosis. Traditional methods sometimes lead to over-treatment or under-treatment. Enter Artera, an AI system that predicts long-term clinical outcomes better than the U.S. National Comprehensive Cancer Network (NCCN) guidelines. Artera's AI continually learns from new data, improving its accuracy over time. It's scalable, accessible, and can be implemented in any clinical setting with basic digital scanning technology and internet access. This AI tool promises to transform prostate cancer care by providing personalised, accurate, and dynamic risk assessments, reducing the risks associated with traditional treatment methods​​.¹

AI in cardiology: Cognitive computing and virtual assistants

Cardiology is also benefiting from AI. Cognitive computing systems are under development, using machine learning (ML), pattern recognition, and natural language processing (NLP) to mimic human thought processes. Voice-powered virtual assistants can understand and process spoken data, aiding in tasks such as equipment control and navigating electronic medical records (EMR). These AI tools enhance efficiency and accuracy in the catheterisation laboratory, allowing for hands-free operation and better decision-making​​.²

Early detection of diabetic retinopathy

In ophthalmology, early diagnosis is critical to prevent irreversible damage. The IDx-DR system, an FDA-approved AI tool, autonomously diagnoses diabetic retinopathy by analysing eye images captured with a fundus camera. This system delivers results in under a minute, boasting 87% sensitivity and 90% specificity. By enabling timely intervention, IDx-DR helps prevent vision loss and improves patient outcomes​​.³

AI for skin cancer detection

The World Health Organisation reports that 325,000 cases of melanoma and 1.5 million cases of non-melanoma skin cancer are diagnosed worldwide each year. Digital health tools with AI capabilities are leading the charge in the battle against the widely-spread illness.

Skin cancer diagnoses have also become more precise with AI. A robust system developed at Stanford University was trained on over 1.28 million images and fine-tuned with nearly 130,000 scans of skin lesions. This AI tool can accurately classify a range of skin cancers, offering a powerful diagnostic aid that outperforms traditional methods. By leveraging vast datasets, this AI system helps dermatologists make more accurate diagnoses and improve treatment plans​​.⁴

Predicting death risk with AI

AI's predictive capabilities extend to life-and-death scenarios. Researchers at Stanford University developed an AI system to predict the death risk among inpatients. By analysing EMRs, the system flags high-risk patients, allowing timely palliative care interventions. This approach ensures that patients receive the necessary care when they need it most, improving end-of-life care and patient comfort​​.⁵

Sepsis watch: Early detection and response

Sepsis, a life-threatening condition, requires rapid intervention. Duke University developed the Sepsis Watch algorithm, which assesses patient risk and alerts the rapid response team. This AI tool guides care administration in the critical first hours, preventing complications. Similarly, HCA Healthcare's Sepsis Prediction and Optimisation of Therapy algorithm detects sepsis six hours earlier than clinicians, reducing mortality rates by nearly 30% across their hospitals​​.⁶

Using AI to help identify Long Stay Patients and improve outcomes

The Accelerated Capability Environment (ACE) supported the NHS AI Lab in developing a proof of concept (PoC) to identify patients at risk of unnecessary long hospital stays. Gloucestershire Hospitals NHS Foundation Trust, serving a population of 660,000, found that 4% of admissions resulted in stays of 21 days or longer, making up 34% of all bed stays. These extended stays often lead to negative outcomes such as an 11% mortality rate, a 23% chance of readmission, and significant muscle mass loss in patients over 80.

ACE worked with Polygeist to create a long-stay stratification tool using an AI model trained on 460,000 anonymised records. This model produced a risk score for long stays, accessible to all reception and clinical staff. Clinicians could then take preventive actions, such as avoiding catheterisation or referring patients to specialists, to reduce the risk of long stays.

In just 12 weeks, ACE delivered the PoC, which accurately detected 66% of long stayers within the highest risk categories. This tool not only improved patient outcomes but also provided significant economic benefits, saving £1.7m for Gloucestershire Hospitals alone with just a single-day reduction in average stay. Following successful initial testing, the PoC was integrated into the trust’s EMR system for further validation.

AI-enhanced decision support systems (DSS) in healthcare

AI-enhanced Decision Support Systems (DSS) are software designed to assist the medical team in decision-making. They handle organisational, diagnostic, and therapeutic problems by using patient data combined with models and algorithms. DSS provides advice in the form of alerts, colour codes, or visual messages, enhancing the quality of care without replacing human operators.

One example is an online DSS for Prostate-Specific Antigen (PSA) tests, which helps GPs in Denmark to decide if a patient should have a PSA test, interpret results, and assist in the diagnostic process​​.⁷ 

Another application is therapeutic guidance, where DSSs aid in developing treatment protocols and recommending therapeutic actions. For instance, the DICTA system supports Danish GPs in prescribing treatments for type 2 diabetes using algorithm-based decision support​​.⁷

Furthermore, DSSs enhance decision-making by providing evidence-based recommendations. The Data Capture Module (DCM) improves diabetes care by offering Danish GPs updated data on care quality, helping identify patients who are not optimally treated​​. These systems not only improve the accuracy and efficiency of clinical decisions but also ensure adherence to clinical guidelines, ultimately enhancing patient outcomes.⁷

Challenges: Impact of biases in AI algorithms

AI can inherit biases from both the data it’s trained on and the developers who create these systems. Sampling bias occurs when training data is not representative of the entire population, leading to skewed results favouring certain groups. For instance, if AI is trained primarily on data from a specific demographic, it may be less accurate for underrepresented groups, reinforcing existing healthcare disparities​​.^8,11

A 2022 study identified significant gender bias, showing men were three times more likely than women to be first and last authors of clinical AI papers.⁸ This reflects a male-dominated perspective in research, influencing AI outcomes. Moreover, AI algorithms often poorly account for biological sex differences, affecting treatment results. The full impact of this gender imbalance on patient outcomes remains unclear​​.

CC Perez, in "Invisible Women: Data Bias in a World Designed for Men," analysed how male-biassed medical research adversely affects women’s health.⁹ Hence, male-biassed data in DSS can lead to less accurate diagnoses and treatments for women, worsening existing inequalities​​.

There was a biased machine learning algorithm implemented in healthcare. It was created in 2007 to predict safe vaginal births after C-sections. This algorithm inaccurately suggested that Black and Hispanic women had lower chances of success compared to Caucasian women. This led to more unnecessary C-sections for these minority women before the algorithm was subsequently amended almost a decade later​​. Factors like race, insurance type, and marital status could influence birth outcomes, however, the study published in 2019 showed that the algorithm only considered race factors, while other, more important biological factors like age, BMI, and prior labour history were left out.¹⁰

Also, automation bias can occur when healthcare professionals rely too much on AI recommendations, ignoring their own judgment. This can result in patient care errors, as AI may not always account for individual patient complexities​​.¹¹

Solutions to Address AI Biases

To tackle these issues, several solutions are proposed:¹¹

• Ethical Governance Frameworks: Ensures AI systems are developed and deployed responsibly
• Model Explainability: Provides clear insights into AI decision-making, promoting transparency
• Diverse and Representative Datasets: Reduces biassed predictions and improves generalisability
• Regular Audits and Bias Detection: Identifies and addresses biases in AI systems
• Continuous Monitoring: Ensures ongoing improvement and reliability
• Comprehensive Training: Educates healthcare professionals and AI developers on limitations and potential biases​​

These measures help maintain high standards of care and equity in healthcare delivery.

Summary

AI's integration into healthcare is no longer a distant dream but a present reality transforming patient outcomes.

• AI for Prostate Cancer: AI predicts long-term outcomes better than traditional methods, continuously learning and improving its accuracy, making it scalable and accessible​​
• AI in Cardiology: Cognitive computing and virtual assistants enhance efficiency and accuracy in cardiology by mimicking human thought processes and aiding in procedural tasks improving safety and reducing patient anxiety​​​​
• AI for Diabetic Retinopathy: AI tool autonomously diagnoses diabetic retinopathy quickly and accurately, preventing vision loss​​
• AI for Skin Cancer: AI trained on vast datasets accurately classifies skin cancers, aiding dermatologists in diagnosis and treatment​​
• AI Predicting Death Risk: AI predicts inpatient death risk, enabling timely palliative care and improving end-of-life care​​
• AI for Sepsis Watch: AI detects sepsis early, guiding care and reducing mortality rates significantly​​.
• AI-Enhanced DSS: AI-driven decision support systems assist in medical decision-making, improving accuracy and adherence to clinical guidelines​​
• AI for Long Stay Patients: AI predicts patients at risk of long hospital stays, allowing preventive measures, improving patient outcomes, and saving resources​​
• Challengesin AI: AI systems can inherit biases from training data and developers, leading to healthcare disparities. Examples include gender bias in research and racial bias in predictive algorithms
• Solutions to address biases: Proposed solutions include ethical governance, model explainability, diverse datasets, regular audits, continuous monitoring, and comprehensive training

These points highlight AI's transformative impact on healthcare while addressing challenges and proposing solutions to ensure equitable and accurate patient care.