Patient matching: cap overlays before chasing duplicate rates

0:05Host:This is MBA Training. On the table: 7 false merges per 1,000 is the overlay risk hiding inside your duplicate-rate target. Would you rather your hospital lose your chart, or hand it to a stranger?

0:19Expert:Oh, come on. That's a terrible choice to put first thing in the morning.

0:23Host:It's the choice every patient identity team is making right now. They just don't say it out loud.

0:29Expert:Fair. Losing your chart is a duplicate: you exist twice in the system. Handing it to a stranger is an overlay: two different people living inside one record.

0:41Host:And everybody's measured on the first one.

0:43Expert:Everybody's measured on the first one. That's the whole problem in a sentence.

0:48Host:Okay, but the duplicates aren't harmless. That study in February, BMJ Quality and Safety, the journal from the British Medical Journal group...

0:57Expert:Patients with duplicate records were about five times more likely to die after admission. Propensity-matched, so they compared similar patients, not sicker against healthier.

1:09Host:That one stopped me cold, honestly.

1:12Expert:Me too. You picture somebody's allergy sitting in the other chart. The one nobody opened.

1:18Host:So the answer is obvious. Merge harder.

1:21Expert:And that's exactly where it goes wrong.

1:24Host:Hang on, you just told me duplicates kill people.

1:28Expert:They do. And the fastest way to shrink a duplicate rate is to auto-merge more aggressively: let the software glue records together without a human checking. Every wrong glue job is an overlay.

1:41Host:Give me the number from the article. The one in the headline.

1:45Expert:Seven false merges per thousand. That's from OneFlorida, a big research network in Florida, part of PCORnet, the national clinical research network. Their validated matching precision, meaning how often a match is actually the same person, came out around ninety-nine point three percent.

2:04Host:Which sounds fantastic.

2:06Expert:It sounds like a report card you'd frame. Now picture a cleanup project that auto-merges twenty thousand pairs.

2:14Host:Oh no.

2:15Expert:A hundred and forty people now share a chart with somebody else. Notes, orders, lab results, all braided together.

2:23Host:Two Guys, One Chart. Worst sitcom ever pitched.

2:27Expert:Season one is just the blood bank calling.

2:31Host:Seriously though, why is an overlay worse? A duplicate hides information. An overlay at least keeps it all in one place.

2:39Expert:In the wrong place. A duplicate is a missing page. An overlay is someone else's pages stapled into your book, and nobody knows which ones. The article cites an ECRI case, ECRI being the patient safety research nonprofit, where one patient's documentation was used to clear a different patient for surgery. That patient died.

3:04Host:That's... yeah.

3:06Expert:And unmerging is brutal. Separating a merged chart means somebody going note by note asking whose kidney function this was.

3:15Host:Speaking of things nobody wants to untangle, it's Election Day.

3:20Expert:Is this going where I think it's going?

3:22Host:Tom's Guide, pointing to Axios: voters in ten states are deciding today whether to restrict AI data centers, the giant server farms running artificial intelligence. Electricity bills are the fight.

3:35Expert:Ten states voting on infrastructure most people never think about. Patient matching would love that kind of attention.

3:43Host:Nobody's putting the master patient index on a ballot.

3:46Expert:The EMPI, the master list that says which records belong to which human. No. Though Congress did come close this summer.

3:55Host:The MATCH IT Act.

3:57Expert:Senators Warner and Banks, bipartisan, asking the health department to set national standards for measuring match rates. The bill admits nobody even agrees on what "match rate" means.

4:09Host:Which is wild. We're grading people on a test nobody wrote.

4:14Expert:And the bill floats ninety-nine point nine as an aspiration, then has a line saying nothing in it requires anyone to hit it. Aspirational, legally optional, very Washington.

4:26Host:A goal with a built-in excuse note. Inspiring.

4:30Expert:Here's where I get opinionated, though. That ninety-nine point nine is the right number for something else. Cap overlays at one false merge per thousand automated merges. Set that ceiling first, tune the matching to respect it, and let the duplicate rate land wherever it lands.

4:48Host:See, this is where I push back. You're telling a hospital to accept more duplicates on purpose.

4:55Expert:To accept that some duplicates get a human instead of a robot.

4:59Host:Humans are slow. Health information management teams, the folks who adjudicate records, are drowning already. Now national exchange is shoving more records at them.

5:10Expert:TEFCA, the national framework for swapping records between networks. A new operating procedure took effect in August, so yes, more inbound records to file.

5:22Host:So the backlog grows, the duplicates sit there, and per that BMJ study, people are dying in the backlog while you protect your precision score.

5:31Expert:That's a real point, and I don't love it.

5:34Host:Thank you.

5:35Expert:But your version quietly manufactures a different body count, one that doesn't show up on the dashboard. "Twelve thousand duplicates resolved this quarter" gets applause. Nobody reports the wrong merges hiding inside it.

5:49Host:Maybe because there aren't many.

5:51Expert:Maybe. Most hospitals genuinely don't know. They can tell you their duplicate backlog to the decimal and they've never sampled a single past merge to check it.

6:02Host:Okay, that part is damning.

6:04Expert:It's like a restaurant bragging about how many plates went out without asking how many came back.

6:10Host:Fine. But you still haven't convinced me the duplicate patient isn't the more urgent emergency.

6:17Expert:I'm not sure I can. Both errors hurt people. I just think the one you can see shouldn't get to set the rules for the one you can't.

6:26Host:Let's talk money then, because the CFO will. What does an overlay cost compared to a duplicate?

6:32Expert:Figures attributed to AHIMA, the health records professional association, put an overlay at about five times the cost of fixing a duplicate. But those come via Staffingly, which sells registration services, so treat them as ballpark, not gospel.

6:48Host:Vendor math. My favorite genre.

6:51Expert:Still, even rough, it means a pair only deserves an automatic merge when you're quite sure. Below the very top band, a person looks. Below that, you leave the records separate but linked.

7:05Host:Linked, not merged. Explain that like I'm a board member.

7:09Expert:Merging pours two glasses of water into one. Linking ties a string between the glasses. Cut the string anytime. Try un-pouring the water.

7:20Host:Okay, that's good. Who actually does this well?

7:23Expert:Intermountain, years ago, set different matching thresholds for each source system. Their internal duplicate creation dropped noticeably within months. Old study, still the right idea: an emergency room registering an unconscious trauma patient is not the same as a clean feed from a clinic.

7:43Host:Or twins.

7:44Expert:Newborn twins are the final boss of patient matching. Same birthday, same address, same last name, names that are sometimes literally "Baby Girl A."

7:55Host:The algorithm sees that and goes, "Ah, one confused baby."

7:59Expert:Exactly.

8:02Host:There's an AI angle here that genuinely worries me.

8:05Expert:Me too. Picture a summarization agent reading an overlaid chart. It'll write a beautiful, confident paragraph about one patient who is actually two.

8:16Host:Fluent nonsense with excellent grammar.

8:19Expert:The article borrows from MIT Sloan Management Review, comparing AI governance to nuclear plants and air traffic control, places where safety is operating discipline, not just clever technology. Identity resolution is that discipline for clinical AI.

8:36Host:And the people who get missed by matching, is that random?

8:39Expert:No, and that's the part that bothers me. In England, missed matches skewed readmission rates, especially for some ethnic minority groups. So the errors aren't spread evenly. They land on specific people, and then on contracts that pay hospitals based on those numbers.

8:58Host:That lands hard. The patients the system already sees worst get seen even worse.

9:04Expert:Yeah. It's quiet, and it compounds.

9:08Host:Let me give you the last word on my side's fear. A hospital sets your overlay cap, slows merging, and the duplicate backlog balloons. Then what?

9:17Expert:Then you throw people and cleaner data at the middle band, the zone where the algorithm is most confidently wrong. Standardized addresses at the front desk alone made a big dent in the research. Front desk is cheaper than any software upgrade.

9:33Host:I'll admit the address thing is unglamorous enough to be true.

9:37Expert:The best data fixes always are. Nobody's giving a keynote about postal formatting.

9:43Host:I still think you're underweighting the duplicate deaths.

9:46Expert:And I think you're trusting a number that's only half the scoreboard.

9:50Host:So we agree to keep fighting.

9:52Expert:Over a single chart, ideally. Clearly labeled.

9:56Host:Two Guys, One Chart, season two.

9:59Expert:Renewed by nobody.

10:01Host:Here's what sticks with me. Seven in a thousand sounds like a rounding error, until you remember each one is a person walking into surgery carrying somebody else's history.

10:12Expert:And the dashboard calls it success.

10:14Host:Drawn from BMJ Group, Digital Health Insights and Nalashaahealth, links in the show notes. End of episode. The CDO calculators are running at mba-training.com.