By Kevin Thomas
Inside Precision Medicine’s inaugural issue focused on artificial intelligence (AI)—namely, how it can help solve the population health crisis and predict early warning signs of chronic disease. The healthcare industry is booming with AI-enabled solutions, from faster, more accurate analysis and diagnostic tools to helping clinicians and researchers make sense of huge amounts of patient data. AI is responsible for reducing MRI scanning times from an hour to minutes and cutting pre-discovery time for new drug development from years to months1. It is clear that AI is moving healthcare forward, making a positive difference in people’s lives, and dramatically cutting cost.
These advancements in technology, and in particular AI, come with enormous responsibility. We in the healthcare industry are obligated to remain vigilant about acting in the best interest of those who we ultimately serve—the patients—and to put safeguards in place to protect their privacy. Patients are increasingly concerned about the use of their data as they navigate the healthcare system and participate in clinical trials2. In broader society, AI is too often used invasively to infer individuals’ personal information for commercial gain. In healthcare, we must take the opposite approach and leverage this powerful technology to protect individuals’ privacy. Most medical records contain personal health information (PHI), and therefore carry risks to patient privacy if not protected. Legal guidelines for how to properly steward this information are laid out in the United States’ Health Insurance Portability and Accountability Act (HIPAA) and the European Union’s General Data Protection Regulation (GDPR). The doctors, hospitals, labs, clinical trial researchers, and health insurers who handle PHI are all subject to this legislation. Most safeguards required by HIPAA and GDPR are straightforward, but some are very time intensive and the regulatory landscape is evolving with technological advances. AI is being used to meet these challenges in some exciting new ways.
There is a lot of information for us to protect. A patient file may contain hundreds or even thousands of pieces of information of various types, including images, electrocardiograms (ECGs), videos, PDFs, and electronic medical records (EMRs) from multiple sources. While this vast amount of information leads to a better understanding of disease and how to treat or prevent it, the data from medical records can be used effectively in research and clinical trials even if not connected directly to a patient by their name, face, birthdate, or other identifier. A tremendous amount of work in clinical trials goes into removing the PHI from medical data as it is collected to protect privacy. AI stands to augment the critical human work that goes into this process, making it more efficient and reducing errors along the way.
In the case of imaging and photographs in which a patient’s face is visible, AI solutions help prevent patients from being identified by their appearance by automatically blurring or redacting identifiable facial features. This is particularly valuable when video footage is required for a clinical trial. This can be necessary when assessing neuromuscular and behavioral healthoutcomes, for example. Even brief videos will contain thousands of frames, which quickly become infeasible to redact if done entirely manually. Redacting patients’ faces has also become necessary in previously unexpected imaging modalities. Recent research has found that patients’ faces can be recognized from their brain MRIs when the image data is rendered in 3D. AI-enabled image redaction can obfuscate the full 3D face in brain MRIs in a reliable manner while leaving the portion of the image depicting the brain untouched to enable clinical evaluations. This would be infeasible at the scale of large clinical trials if done completely manually but can be quickly checked by human quality control professionals and refined if necessary.
Text in medical records is another type of data that must be screened for PHI when collected for clinical trials. These records may contain patients’ addresses, phone numbers, names, ages, and other sensitive information. The tremendous volume of text data that is collected poses a risk of overlooking instances of PHI during the redaction process when done completely manually. However, AI-enabled solutions that screen documents and highlight PHI that may have been missed stand to improve detection and redaction efforts. These solutions work with both typed and hand-written documents and can detect a diverse array of PHI.
While the PHI redaction process is extremely important for protecting patient privacy, it poses a risk to trial accuracy and safety. Once patients’ medical images and other source materials are captured at a medical center, they are often scrubbed of PHI on site before being uploaded to a central repository for review. As each file is uploaded, the patient’s name is replaced with an ID number, allowing their health to be tracked over time while protecting their privacy. Historically, if a given image was assigned an incorrect ID number, the error was difficult to detect and nearly impossible to fix due to the prior redaction of PHI. This jeopardized the accuracy and success of the clinical trial. AI now offers the ability to automatically compare images collected at two timepoints, rather than compare patient names or ID numbers, to verify whether images depict the same individual. Doing this during the site upload process means errors can be detected before it is too late to fix them. This ensures patients’ health trajectories are assessed accurately.
The people who participate in clinical trials generously give their time and energy with the hope of improving medicine for everyone. We owe them everything modern technology has to offer when protecting their privacy along the way. As the streams of data collected in trials become increasingly comprehensive, our safeguards must become increasingly sophisticated.
AI is already proving to be an incredibly powerful tool for protecting patient privacy, and over time, in partnership with human oversight, will make an even greater positive impact.
1. CB Insights, Healthcare AI Trends to Watch 2020
2. Inside Precision Medicine, AI Offers Window on Heart Health, February 2022
Kevin Thomas, PhD, Director of Artificial Intelligence, Clario. Before Clario, Kevin co-founded Saliency, a Silicon Valley-based startup developing AI-enabled medical imaging biomarkers that was acquired by Clario in 2020. Kevin has a PhD. in biomedical informatics from Stanford University and completed the first two years of medical school at Stanford School of Medicine. Before attending graduate school, he conducted research in Canada as a Fulbright Scholar.