By Saketh Lingisetty ’27
Artificial intelligence (AI) has recently taken the world by storm, especially in the field of healthcare. Predictive algorithms powered by AI have been able to detect early signs of disease, improve patient behavior, and quickly process clinical data with precision, all while reducing the workload and resources required for these tasks. Companies like Johnson&Johnson have utilized AI to kickstart a new era for medical advancement.
Predictive algorithms could make hosting clinical trials easier and more precise. A clinical trial is conducted with selected patients to evaluate the safety and efficiency of a medical treatment or procedure. A major challenge in this process is efficiently recruiting and enrolling patients who meet the trial’s criteria. Researchers at Johnson&Johnson have recently utilized machine learning (ML) algorithms with de-identified patient data sets to quickly locate clinical research sites where patients would benefit from the treatments being studied.
Historically, clinical trials have involved patients going to major academic medical centers for testing. However, not all patients have access to these centers, leading to a biased representation of patients. Bias in medical research often delays diagnosis, reduces the quality of healthcare, and even denies treatment for minorities altogether. To integrate diversity into clinical trials, Johnson&Johnson is leveraging the power of machine learning algorithms to effectively bring more trials to qualifying patients and locate healthcare centers where diverse patients are more likely to be treated.
Predictive algorithms also help improve the efficiency and accessibility of healthcare diagnoses. Some patients, especially those with uncommon diseases, lack proper diagnosis, often leading to unsuccessful medical treatment. According to the American Cancer Society, the diagnosis of colorectal cancer in its first stage yields a five-year survival rate of around 90 percent, but this rate decreases to 14 percent in later stages. In extreme cases, diagnostic odysseys have been documented to take five to six years on average before patients receive an accurate diagnosis of their condition.
Medical specialists like Andy Beam affirm that AI can be used to help diagnose patients. By being trained on large sets of medical imaging and patient data, predictive algorithms can accurately make diagnoses, detecting adverse patterns a human might miss. By 2030, AI-powered diagnostics are predicted to improve the accuracy of detecting health outcomes by 40 percent and reduce treatment costs by up to 30 percent. A cardiology study by the Mayo Clinic reported that their machine learning algorithm successfully identified patients at risk of left ventricular dysfunction, even though the individuals didn’t report any notable symptoms.
While predictive algorithms can help optimize healthcare, a growing reliance on AI implies a rise in privacy issues regarding the sensitive nature of healthcare data. Predictive models, which store, transfer, and analyze patient data on a larger scale, pose a greater risk of data breaches and cybersecurity attacks. As of this year, healthcare breaches incur an average cost of $9.77 million, twice that of a normal data breach. Implementing AI-powered algorithms may require hospitals and medical institutions to create a secure and resilient framework that protects patient data and preserves the trust of individuals.
A major issue with AI-powered models is racial bias. For example, an algorithm reported black patients to be healthier than white patients, effectively denying proper healthcare to 30 percent of Black patients. Medical experts have discovered similar algorithms that only give the same resources, diagnosis, and treatment to racial minorities if they are considerably more ill than their white counterparts. According to Harvard Medical School, people strive to increase the overall accuracy of such algorithms but end up making them racist. To overcome algorithmic bias, datasets used in predictive algorithms can adopt fairness metrics that evaluate training sets like neural networks for better data diversity.
Given its benefits and drawbacks, there is no straightforward way to approach AI with the future of healthcare in mind. When tested in medical diagnoses, a hybrid human-AI model from MIT produced the best results. Through similar cases of hybrid intelligence, healthcare companies continue to tread the line between human reliance and dependence on AI.
Currently, many hospitals and clinics lack adequate guidance to safely and successfully implement machine learning models into their current protocols. As we navigate a rapidly growing era in AI, it is crucial to recognize the benefits of predictive algorithms while consciously working to resolve existing flaws in their application. That way, we can truly move forward in the right direction to ensure a better future for healthcare.
Read more articles like this in our Fall 2024 Issue!
