Health professor Neil Wenger was deep into a years-long study on seriously ill primary care patients when he uncovered a different but persistent issue: Many patients who were targeted for follow-up interventions had actually died, and their hospitals did not know about it.
It wasn’t what Wenger had set out to research. But it was unacceptable, he said, for health systems to lose track of seriously ill primary care patients who’d been treated on-site for years. To be clear, these were not one-time out-of-state patients who’d come in for a single visit.
“When I found a bunch of them were dead and we didn’t actually know it, that’s when I said we need to describe it,” he said. “This is a cohort where we need to know about everyone who is dead.”
In an attempt to establish how widespread the problem was, he examined roughly 12,000 seriously ill primary care patients from UCLA Health, one of three California health systems in his earlier study. Wenger and his co-investigators found that several hundred — a fifth of deceased patients in his sample — were marked alive in electronic records, but dead in state public health files.
Most of those patients had also been targeted for different kinds of contact after they’d died: more than 2,000 combined phone calls, patient portal messages, notices about outstanding appointments and mailings about preventive care like flu shots and cancer screenings, among other messages.
Now, Wenger and his team are leading a charge to eradicate these errors, urging health systems to update their verification process and pressuring legislators to make death data more accessible. He says hospitals haven’t had much incentive to do so until now, as the issue only arises among a subset of patients — though it’s likely more common than in his study, which only covered seriously ill primary care patients, suggests.
The advent of insurance contracts rewarding hospitals for healthy patients — called value-based care in industry jargon — could push them to collect more accurate data, if only to prove that patients died elsewhere. And incomplete patient data could seriously impair health systems’ charge toward AI and predictive algorithms.
There’s no nationwide benchmark for how often these mix-ups happen, and Wenger appears to be among the first to quantify them. But health leaders STAT spoke with said it likely happens at every health system, though the prevalence varies according to internal backstops and state data regulations.
“This is a totally solvable problem, which is actually the reason we wrote the [JAMA Internal Medicine] research letter,” Wenger said. “It’s a health system’s responsibility to be accountable for their patient and that means being able to follow when someone has passed away.”
The problem might be solvable, but it’s not easy. While some clinical notes do reference a patient’s death, the file typically isn’t labeled “deceased” until another department verifies it, health leaders said. It’s especially easy to lose track of patients who’ve only visited a health system once — though those fell outside the scope of Wenger’s research — or the ones who died at home or in an outside facility or out of state.
Cross-referencing death databases is also a challenge. The Social Security Administration’s master file only accounts for people with social security numbers, and states typically maintain their own death records. Hospitals rely on those state records, and fees and accessibility vary by jurisdiction. While California does maintain a death file, health IT leaders there told STAT that what health systems are given access to is not detailed enough to fully verify the death of some patients with common names.
“In California, this file is available to health care facilities only in a form that contains inadequate specificity to inform clinically about death,” Wenger’s research letter reads. “In contrast, a more complete file is available ‘for purposes of law enforcement or preventing fraud.’”
For patients who aren’t in those databases but still have electronic health records, hospitals must find other ways to ascertain their death — sometimes by using technology that mines the internet for obituary data or death records, flagging a patient as likely deceased, and then manually verifying their death by obtaining a death certificate. Companies like LexisNexis and Veritas Data Research sell those services.
But many hospitals simply do not have a robust verification process, said Eric Cheng, chief medical informatics officer at UCLA, where Wenger’s research was conducted.
“If a patient were to call the clinic or a doctor and say a family member died, we don’t necessarily do the best job in documenting that the same way. Physicians don’t know whether that’s stored, the front desk clinic may not be comfortable if they’ve never heard of the patient — they would all document in the note, but not in the official place where it should be,” said Cheng, who oversees UCLA’s medical records department.
“We know we should do it, but we need a little push to do it — and I think this [research] will be helpful for us to demonstrate and prioritize work on this for the rest of the organization,” he said.
Fixing the problem might require inventing an entirely new workflow for clinical and administrative staff, he said. “We have to make it easier for them to fill out a form saying [a patient] is likely deceased, and then someone who does this more often independently confirms and does the final changing the status,” he said. “The front desk, or provider or clinician wouldn’t bear the responsibility of getting this important decision right.”
Health system leaders told STAT that the issue has grave implications for artificial intelligence and machine learning systems that draw on patient records.
“We’ve tried to develop models to predict who’s likely to be deceased and those models rely on accurate data — we don’t have good data to train our own prediction models [and] they’re going to be off,” Cheng said.
“If the system doesn’t know the patient has passed away, they’re still included in various population health cohorts,” said Paul Fu, chief medical information officer at City of Hope cancer center and co-chair of informatics trade group the American Medical Informatics Association’s health systems committee.
Lawmakers have tried to address the issue, but with little success so far: A bill introduced early in Covid, which died in the House Energy and Commerce committee, aimed to enforce better data sharing — including death data — between the federal government, states and health systems, said Michael Hogarth, a UCSD researcher who worked with Wenger on the study. Industry groups have also lobbied for a national patient identifier to better match records, but the effort has faced resistance in Congress from Republicans worried about surveillance.
The march toward value-based care, and the promise of insurance payouts for healthier patients, could nudge hospitals to shore up their record keeping, Hogarth said.
But until it’s fixed, “you have to read the obituary of your patient in the newspaper to know that they’ve died,” said Deepti Pandita, chief medical information officer at UC Irvine Health and a member of AMIA’s board. “It happens more often than you think.”