Not from us. From a look-back audit of your own records. Records that were already coded and paid are too many for anyone to re-check, so nobody knows whether revenue leaked. The audit uses AI to check several hundred inpatient records, comparing the codes actually billed with what the record contains, such as recorded diagnoses, lab results and drugs given, under the published TDRG and ICD-10-TM rules. The AI flags each gap with the part of the record behind it, and your hospital's own coder confirms every one, counting both codes that were missed and codes the record does not support. Price the net difference in adjRW at your own payer rates. That measured delta, annualized, is R. R measured this way is a floor, because it counts only what the record's data can confirm. If the floor pays back, the investment does. This is the pilot we described in part one, and it is the only source of R we will accept in our own business case.
What we can say, with citations, is that when others have looked, they found discrepancies:
| Source | Sample | What it found |
| BMC Health Services Research, 2011 (reporting the NHSO's 2008 audit) |
The NHSO's 2008 Summary and Coding Audit: 57,828 records from 931 hospitals, chosen because their data looked abnormal |
Errors in 42 percent of abstractions from medical record to discharge summary; the most common error was the secondary diagnosis, at 28 percent |
| Healthcare Informatics Research, 2017 |
118,971 records from the NHSO's 2014 audit results, selected by abnormal-data criteria |
For the 20 most common principal diagnoses, discharge summaries captured only 7.3 to 37.9 percent of the cases the auditors found |
| Integrated Social Science Journal, Mahidol University, 2014 |
1,107 inpatient records at Mahidol University's Hospital for Tropical Diseases |
Discrepancies in 322 records; the corrections raised adjRW by 182.84 and, by the authors' own calculation, reimbursement by 2,420,928.35 baht (about 2.4 million) |
These numbers come from published research. They do not mean your hospital has the same problem; the only way to know is to check your own data.