Imagine you’re not feeling well and you trudge into a doctor’s office. As usual, your blood pressure, temperature, weight, and blood tests are taken, and the results are entered into your record.
But then, rather than ask you a bunch of questions you’ve already answered before, the doctor hits a button and calls up your entire healthcare history—all your past test results and your risks for hereditary and lifestyle-influenced diseases—along with findings from the latest, relevant scientific studies.
After asking some computer-assisted questions about your symptoms, the doctor has the system churn through all this information to make recommendations for treatments, medicines, and lifestyle changes to make you better. The data-informed evaluation is tailored specifically to you and costs less than a visit to the doctor’s office today.
This level of personalized medicine is the promise behind the growing influence of machine learning and artificial intelligence (AI) in healthcare. Unfortunately, however, we’re still a long way from doctors and other healthcare professionals being able to make that visit a reality.
It’s not for lack of trying. In 2017, venture capitalists in the United States poured US$800 million into AI healthcare startups. These efforts tend to be focused on specific tasks, such as improving diagnoses. For example, San Francisco-based Freenome is applying machine learning to cancer detection. The four-year-old company has already raised $78 million from investors. Also in the Bay Area, Arterys, which brings AI to radiology, has raised $44 million.
The technology that is being developed by these kinds of companies will change healthcare forever, and it will save lives by enabling more accurate diagnoses, more precise treatments, and improved collaboration among patients and healthcare providers. However, our progress is being impeded by one big, persistent problem: our enormous, tangled mess of health data.
Free the Data
Healthcare is overflowing with data. Primary care doctors, specialists, hospitals, pharmacies, and research institutes, among others, all collect it. Test results, images, clinical study results, genome data, and more are stored in a whole range of formats. Some providers still keep handwritten notes, which may or may not be legible. And none of it is easy to aggregate.
This is a global problem. Most developed economies have siloed health data. Israel is an exception: it has a highly consolidated healthcare system because four nonprofit health insurance companies provide care and insurance for the entire country.
Most countries will never be able to replicate Israel’s model because their populations are larger and there are many more stakeholders in their healthcare systems. The systems we’re now using are so embedded across the healthcare industry—including hospitals, insurance companies, and practitioners—that it’s not realistic to replace them completely. That means working with what we have. We need platforms that can be adapted to many different types of data in many different formats.
The most realistic approach is to develop platforms that can be overlaid on top of the current systems to gather and organize data in ways that make it more useful. With a flexible way to pick and sort massive amounts of information, AI and machine learning can enable care recommendations, help researchers with drug discovery, and facilitate predictive analysis. Google and Apple are both working on the healthcare data problem by creating such platforms.
All this data innovation might seem abstract and removed from the aforementioned doctor’s visit, but we must not forget about the patient’s role. Patients are some of the best data collectors we have. Digital innovations to date have given them the ability to research their health concerns. People living with chronic diseases routinely collect data on their health and share it with the multiple healthcare professionals who treat them.
Patients expect to have access to their own health data and to be involved in making decisions about their healthcare. We can turn these expectations to our collective benefit by designing systems that, by improving data access, make all of their interactions with healthcare easy for them, and allow for fully integrated case management. Which is to say that everyone involved in care brings value; machine learning and AI in healthcare present a great opportunity to connect all the participants.
With a flexible way to pick and sort massive amounts of information, AI and machine learning can enable care recommendations, help researchers with drug discovery, and facilitate predictive analysis.
For this to happen, however, we need clearly defined patient privacy rules. One reason is that people have become much more sensitive to who sees their personal information and what it’s used for. Another is that patients need to trust the healthcare system; if they don’t, they are more likely to withhold information about their health, to the detriment of their care.
The General Data Protection Regulation (GDPR) just instituted in the European Union has rules for governing data access, for incorporating privacy and data protection into system design, and for ensuring the right to be forgotten. All of these rules are significant to healthcare. In some ways, the GDPR will likely make using data easier. For example, it allows researchers to request a patient’s permission to use their data for broad medical research, not just for specific research projects.
But burgeoning concerns about privacy and increasing regulations could also create more hurdles for machine learning and AI. For example, any patient data used for research must have each individual’s opt-in consent, must be anonymized, can be kept only for a limited time, and is subject to right of erasure. For researchers working after the implementation of the GDPR, creating a study with data from thousands of patients means checking a lot of boxes to comply with the regulation.
We must also be realistic about the privacy of patient data because it is not possible to completely anonymize information and maintain its usefulness. There will need to be a compromise, a “good enough” standard that benefits the patient, healthcare professionals, researchers, and institutions so that barriers to accessing data do not hinder the development of medical innovations.
The objective is to create a useful balance of data and privacy that will become the foundation of future AI-powered healthcare. Governments, the healthcare and tech industries, and patients should all work together to create this foundation now, while these innovations are emerging.
Only then will we be able to create the life-saving, life-changing medical care that will improve every trip to the doctor’s office. D!