A Blog by Jonathan Low

 

Dec 12, 2021

The Temptation To Mine Informational Gold From Patient Medical Records

The insights may prove valuable for improving patient care and outcomes. 

But in the tradeoff between data monetization and respecting personal privacy, monetization always - always - wins. JL

Ron Winslow reports in the Wall Street Journal:

Details on the medical care of hundreds of millions of patients are piling up in electronic health records in clinics and hospitals around the world, comprising a growing treasure trove of real-world data on the daily practice of medicine—patient diagnoses, treatments and outcomes. Aided by advances in artificial intelligence, search capabilities and other analytics, researchers are now probing the huge databases for rapid insights into the performance of the healthcare system.

You have just been diagnosed with diabetes. The doctor opens your electronic medical record, which includes your latest test results, DNA sequence, history of arthritis, smoking status, prior Covid-19 infection and your age, gender, body-mass index and race. She updates as necessary and clicks a green button on her computer screen.

Moments later, a report comes back summarizing how thousands of patients just like you were treated in the past. In this hypothetical future appointment, your doctor draws on this volume of cases to come up with the best treatment.

Details on the medical care of hundreds of millions of patients are piling up in electronic health records in clinics and hospitals around the world, comprising a growing treasure trove of real-world data on the daily practice of medicine—patient diagnoses, treatments and outcomes. Hardly anyone has sought to tap the knowledge sequestered in those digital vaults that might benefit future patient care. Until now.

Aided by advances in artificial intelligence, search capabilities and other analytics, researchers are now probing the huge databases for rapid insights into the performance of the healthcare system.

For example, closely held Epic Systems Corp., maker of one of the most widely used electronic health record systems, searched a segment of its database in the spring of 2020 to find that routine breast, colon and cervical cancer screenings in the U.S. had each dropped by more than 85% during the first weeks of the Covid-19 pandemic. The report helped spur efforts to persuade people to make up for missed screenings.

Changing care for individual patients

But researchers have a much more ambitious vision for this data: to help guide how doctors treat individual patients in real time.

“The evidence from real-world data is a different and exciting new path,” says Jackie Gerhart, a physician who works with the informatics team at Epic. “You can get a lot of outcomes information from medical records that can help change care for individual patients.”

For almost any patient a doctor sees today—whether for, say, asthma, high cholesterol or sepsis—hundreds or perhaps tens of thousands of similar patients “have already had the care and have had outcomes, good or bad,” says Saurabh Gombar, co-founder and chief medical officer of Silicon Valley startup Atropos Health.

To be sure, patient records are observational, and thus subject to confounders and other shortcomings that can undercut their reliability in pointing to treatment options.

But the gold standard has its own issues. Randomized clinical trials, which control for differences in patient health status and other variables, are the preferred evidence to inform patient care. Yet such trials generally exclude an especially common group of patients—those with multiple ailments. Moreover, the elderly, children, women, minority groups and people who live far from medical research centers have long been underrepresented in such studies.


As a result, the highest-quality evidence that medicine produces doesn’t apply to most patients doctors see in daily practice. “There are so many clinical situations where the evidence that is needed does not exist,” says Nigam Shah, professor of medicine and biomedical data science at Stanford University Medical School.

Researchers have believed for at least a half-century that data in patient medical records could help fill the gaps. But until the last decade, most of it was stored in manila folders that lined shelves in doctors’ offices and hospital record rooms. Without the ability to digitally organize and analyze these records, they have had little value for patient care.


Thanks to a massive government investment and regulatory push early in the Obama administration, well over 80% of U.S. hospitals and physician practices now maintain computerized patient records.

Clifford Hudis, chief executive of the American Society of Clinical Oncology, is convinced that one big payoff is the emerging ability to tap this resource for help with clinical care.

Under a project called CancerLinQ, ASCO has built a database of 6 million patients provided by some 108 cancer practices. It has already led to progress in health equity by quantifying and encouraging oncologists to narrow gaps, for instance, in treatments offered to Black and white patients, Dr. Hudis says. Another plan for CancerLinQ is to gather data about outcomes that will help doctors better inform their own patients about the impact of treatments on quality of life. Eventually, the hope is that a doctor treating, say, a patient with breast cancer, will be able to use her clinical profile to query the database for what treatment regimens have yielded the best outcomes in similar patients.

Epic’s database, called Cosmos, currently boasts 122 million patient records, and the company’s researchers mine it to produce weekly reports on population-level trends in the healthcare system, such as the one on the pandemic’s impact on cancer screenings. Epic also expects that doctors eventually will be able to ask Cosmos for guidance in treating specific patients in real time, helping to “change care on the individual level,” Dr. Gerhart says.

Atropos Health launched late last year to commercialize technology being developed at Stanford to convert data in electronic health record databases to information for patient care. It markets a tool called a Prognostogram that uses a combination of the technology and human experts and that it says can now answer most physician inquiries about patient treatments within 24 hours.

Atropos’ vision is for a “green button” search function linked to a patient’s electronic health record that would fully automate real-time retrieval of a descriptive summary of how hundreds or thousands of similar patients were treated for the same ailment—something Stanford’s Dr. Shah suggests could be a routine part of a doctor visit within a decade.

‘A question of quality and accuracy’

All of this comes with daunting challenges. Electronic health records are a hodgepodge of data because few standards exist for entering patient variables. In developing the CancerLinQ database, ASCO found more than 60 different versions of how white blood cell counts—a fundamental biomarker for cancer patients—were recorded.

Without standardized reporting, “you have a question of quality and accuracy that follows you forever,” Dr. Hudis says.

had ambitions to develop a tool for cancer doctors that would mine patient health records and thousands of pages of research from the peer-reviewed medical literature for treatment advice. But it hit walls, including accuracy and the complexity of combining data from electronic health records, billing claims and published research to provide a cohesive product. Doctors who used the service rarely changed treatment plans. IBM says it discontinued Watson for Oncology at the end of 2020.

Controversies over privacy and ownership of patient data could also hinder use of the information. Especially important is the need to minimize biases inherent in observational data and to select the most meaningful variables for a patient’s condition. Doctors need to know they can trust the information and that it provides useful advice in the clinic or at a patient’s bedside.

“Observational data have errors,” says Stanford’s Dr. Shah, who is also co-founder and a technical adviser for Atropos Health. “But that’s not an excuse to say we have all this data and we’re not going to use it ever.”


0 comments:

Post a Comment