27,111
Requests in the California book
270,150
Processing notes read, line by line
1,979
Patients behind the Boston book
321
FALSE "NO RECORDS" ANSWERS PROVEN WRONG
Ask anyone what is wrong with health information and you get the same word back: fragmented. It is the most agreed-upon diagnosis in American healthcare, and nobody has put a number on it. This case study uses the complete operating records of two retrieval engagements to try.
Your medical history does not exist
Not in one place. Break your wrist tonight and by morning five organizations hold a piece of it: the hospital's record, the hospital's bill, the emergency physician's bill, the radiologist's report and the radiologist's billing company. None of them talks to the others. None of them holds a list of what the others have. A complicated case runs to forty.
EXHIBIT 1 · Boston book, by request type
Same patient. Same night, frequently the same building.
Mean wait
Hit a problem
Clinical records
3,501 REQUESTS
33.2 days
32.4%
The hospital's bill
3,411 REQUESTS
44.9 days
38.0%
The ER physician's bill
410 REQUESTS
43.9 days
42.2%
The radiologist's bill
308 REQUESTS
54.9 days
54.9%
Fragmentation is not a metaphor. It behaves like a dose.
The cost nobody counts
Every organization standing between a question and an answer can say no, and no is the one answer nobody can check. A "no records" letter means someone typed a name into a search box and did not see anything. Testing it means proving the treatment happened some other way and making them look again, so it is done for almost nothing. We did it anyway, across 27,111 requests for one national firm over seven months.
EXHIBIT 2 · "No records" answers, California book
Two in five of the denials we tested were wrong.
How big is the hole
~19%
Low case· ~498 false denials
~40%
High case· ~1,025 false denials
Somewhere between one in five and two in five of the absences reported to us were probably not absences.
The same institution, several strings
Fragmentation also happens inside provider organizations. In Boston, one firm, one statute and one regional market still produced a 23-day mean wait at one provider group and a 54-day mean at another. Before any of it could be measured, the names had to be fixed: hospital names in an export are not stable institution identifiers.
EXHIBIT 3 · Same institution, different strings
The institution did not change. The string did.
"Mass General Hospital"
118 requests · 48.9 days
"Massachusetts General Hospital"
108 requests · 47.6 days
"Brigham and Women's hospital"
54 requests · 60.7 days
"Brigham and Womens Hospital"
59 requests · 52.5 days
Resolving it reduced 1,368 raw provider names to 1,334 organizations across 74 clusters.
Two ways a request dies
Just over a third of the Boston book, 2,809 requests, hit at least one problem. Where the fault landed matters more than how often. Provider failures are slow, but four in five eventually deliver. Client-side failures are faster and far more lethal. Wrong date of service, wrong provider, misspelled name: each a field that could have been checked before the request went out.
Provider failures nearly triple the wait
27.2
Avg days clean request
74.6
Avg daysprovider-caused failure
Requests that were simply never answered account for nearly a fifth of all the time lost in the book.
The middlemen
About a quarter of the time, a records request never reaches the hospital at all. It goes to a company handling release of information on the hospital's behalf, chosen by the provider, not the requester. The failures at this layer have nothing to do with whether records exist. They are about handling.
EXHIBIT 4 · Platform handling, California book
Lost the request on their own system
218
Requests routed through a release-of-information
platform, of 27,111
6,193
Through Ciox, ChartSwap and Datavant
combined
1,274
Bounced or misrouted by a platform and sent
again
1,057
Resubmission events across ~6,601 requests, absorbed without an extra fee
12,648
We count delay, because delay is easy to count. What fragmentation actually does is quieter and considerably worse. It destroys information.
Fragmentation Has a Number
A fee is easy to verify. Completeness is not.
The answer to concentration is not to presume bad intent. It is to demand better instruments.
Independent measurement adds
-
False "no records" responses, and whether the file arrived complete
-
Fees by outcome, plus bounce and resubmission rates
-
Provider identities resolved before scoring
-
Provider delay separated from requester error and platform failure
What the market measures today
-
Whether the request closed
-
Fees by transaction
Scorecards built on raw provider names
Turnaround as a single number
What a useful scorecard would measure
Most retrieval scorecards collapse performance into one number: turnaround. The data argues for separating at least five things. Three of them:
Cause
Cause, not just age
A sixty-day request caused by a provider backlog is not the same event as one waiting on a wrong date of service.
Completeness
Completeness, not just closure
A request closed as "no records" may still be wrong.
Exposure
Case-level exposure
A patient with several requests waits on the slowest fragment. Maximum delay often matters more than the mean.
Read the full case study
The complete 14-page analysis, every exhibit, the methodology, and the per-requester appendix.
About The Records Company
Better records. Better outcomes.
The Records Company is a national medical-record retrieval partner serving legal, insurance, healthcare, and government clients. It pairs disciplined intake with transparent status on every request and a fully auditable record of the work, so clients can see at any moment exactly where each request stands and what it has cost. This case study is based on a real, anonymized client engagement, published for educational purposes.