Hospital outliers
IIB Board AI Pack, page 16. First phase. Built here on synthetic data.
Today. Each insurer sees only its own claims from a hospital, so a hospital billing far above similar hospitals across the market is hard to spot.
What would change. Compares each hospital with similar hospitals on cost, length of stay and treatment, and flags those far outside the norm for review. It is the Health vertical's weekly alert.
Six hospitals bill far above similar hospitals
Week of 28 September to 4 October 2026. Each hospital is compared with similar hospitals over the 13 weeks to that Sunday, 6 July to 4 October 2026, so one busy week cannot tip it in or out. No hospital is new on the list this week. The flagged hospitals admitted 17 patients with a claim this week.
Flagged hospitals, 6 July to 4 October 2026
| Hospital | Where | Claims | Bill against similar | Stay against similar | Treatment mix | Insurers billed | Flagged since |
|---|---|---|---|---|---|---|---|
| Vaidyavan Super Speciality HospitalMultispeciality hospitals in tier 2 citiesBilled 2.3 times similar hospitals for the same procedures and kept patients 1.5 times as long. | Haryana | 66 | 2.26× | 1.49× | In line | 21 | week to 28 June 2026 |
| Shubhprana Hospital and Research CentreMultispeciality hospitals in tier 2 citiesBilled 2.2 times similar hospitals for the same procedures and kept patients 1.5 times as long. | Rajasthan | 38 | 2.24× | 1.54× | In line | 15 | week to 28 June 2026 |
| Ayushkiran HealthcareMultispeciality hospitals in tier 2 citiesBilled 2.2 times similar hospitals for the same procedures and kept patients 1.6 times as long. | Tamil Nadu | 77 | 2.18× | 1.64× | In line | 21 | week to 28 June 2026 |
| Pranavriksha Hospital and Research CentreMultispeciality hospitals in tier 2 citiesBilled 2.1 times similar hospitals for the same procedures and kept patients 1.4 times as long. | Chandigarh | 54 | 2.12× | 1.41× | In line | 18 | week to 28 June 2026 |
| Swasthkiran Super Speciality HospitalMultispeciality hospitals in tier 1 citiesBilled 2.4 times similar hospitals for the same procedures and kept patients 1.5 times as long. | Delhi | 51 | 2.36× | 1.52× | More high-cost work | 18 | week to 5 July 2026 |
| Sukhvardhan Super Speciality HospitalMultispeciality hospitals in tier 1 citiesBilled 2.2 times similar hospitals for the same procedures and kept patients 1.7 times as long. | Maharashtra | 80 | 2.16× | 1.68× | In line | 21 | week to 5 July 2026 |
Claims are patients sent home in the 13 weeks, so their bill is known. "Bill against similar" is what the hospital billed, divided by what similar hospitals bill for the same procedures. 2.00× means twice as much. "Stay against similar" does the same for days in hospital. Similar hospitals are the same type of hospital in the same city tier. Click a row to see it procedure by procedure. Places are shown by state for this role.
Where the flagged hospitals are
States in grey have no hospital with enough claims to compare.
Hospitals by state
Click a state on the map to see the hospitals that stand furthest from similar hospitals there. Click a hospital to open it below.
One hospital, procedure by procedure
Select a hospital
Week by week
Review note for an insurer's claims team
No note for this hospital.
Procedure by procedure
Insurers billed
Each insurer sees only its own rows. IIB sees them all.
Close to the line
Watch list
| Hospital | Where | Claims | Bill against similar | Stay against similar |
|---|---|---|---|---|
| New Life Medical CentreMultispeciality hospitals in tier 2 cities | Uttar Pradesh | 22 | 1.17× | 1.20× |
These score from 3 up to 4 on bills or stays: unusual, but not far enough out to flag. They are not counted as flags.
Unusual treatment mix
| Hospital | Where | Admissions | High-cost | High-cost share | Similar hospitals |
|---|---|---|---|---|---|
| Sanjeevkiran Nursing HomeNursing homes in tier 2 cities | Maharashtra | 10 | 4 | 40% | 1.0% |
| Arogya Niketan Nursing HomeNursing homes in tier 2 cities | West Bengal | 53 | 5 | 9% | 1.0% |
| Arogyam Multispeciality HospitalMultispeciality hospitals in tier 1 cities | Uttar Pradesh | 47 | 12 | 26% | 8.0% |
High-cost work is procedures with a package cost of ₹2 lakh or more: bypass surgery, angioplasty, knee and hip replacement and cancer surgery. A mix far from similar hospitals is shown, never flagged on its own, because the bill comparison already allows for it. It can point to a different problem, such as billing procedures a hospital is not equipped for (use 23). One of these is a hospital the test data built for use 23.
How we would measure it
Flags in each weekly run, last 52 weeks
How this test works. The problems were built into the test data. From June 2026, six hospitals were set to bill about 2.1 times their usual amount and keep patients about 1.6 times as long (story H1). The page reads that list only to score itself here. The method never sees it.
The measure counts a flag as right when it is one of the six. That shows the method works on data where the answer is known. It cannot show that it finds real problems. The real test is a pilot, where reviewers at IIB and the insurers judge each flag and record whether it needed a closer look.
The flags lag the change by a few weeks because each run looks at 13 weeks of claims. A shorter window would react sooner, but fewer hospitals would have enough claims to compare: 159 with 8 weeks, against 259 with 13.
How it works: the method, the thresholds and the peer groups
- Similar hospitals. The same type of hospital (multispeciality, single speciality, nursing home or day care) in the same city tier. Tier 1 is the largest metros.
- Expected bill. Each patient sent home is priced at what similar hospitals billed, as a median, for the same procedure in the same quarter. Adding these up gives what similar hospitals would have billed for the same patients. This is called indirect standardisation. It means a hospital that does more heart surgery is not flagged just for doing dearer work. Where fewer than 5 such bills exist, a wider pool is used. 99.3% of patients were priced against similar hospitals in the same quarter. The other 0.7% used a wider pool, such as the same hospitals over two years.
- Expected stay. The same, with the average days in hospital for the same procedure. The stay is scored only when a hospital's expected days add up to 20 or more, since day care has almost no stay to compare.
- Score. How far a hospital's ratio sits from the middle of similar hospitals, in units of their usual spread. It uses medians, so a few odd hospitals cannot hide each other. Groups with fewer than 10 hospitals to compare use the spread of all groups. A score of 4 is far outside the norm.
- Flag. A hospital with at least 20 patients sent home in the 13 weeks is flagged when its bill or its stay scores 4 or more and is at least 1.25 times what similar hospitals show. The rule runs every week, on the 13 weeks to that Sunday.
- Treatment mix. The share of admissions for high-cost procedures, against the share in similar hospitals. Shown beside the flags, never a flag on its own.
- Data. 15,699 health claims admitted in the window, 15,332 of them with a final bill. All synthetic. Patients still in hospital have no bill yet and are left out.
| Similar hospitals | Hospitals compared | Middle bill ratio | Usual spread | Middle from | Spread from |
|---|---|---|---|---|---|
| Day care centres in tier 2 cities | 1 | 1.02 | 0.055 | all groups | all groups |
| Multispeciality hospitals in tier 1 cities | 129 | 1.02 | 0.055 | the group | the group |
| Multispeciality hospitals in tier 2 cities | 80 | 1.00 | 0.049 | the group | the group |
| Multispeciality hospitals in tier 3 cities | 21 | 1.02 | 0.081 | the group | the group |
| Nursing homes in tier 1 cities | 5 | 1.02 | 0.055 | the group | all groups |
| Nursing homes in tier 2 cities | 8 | 1.03 | 0.055 | the group | all groups |
| Nursing homes in tier 3 cities | 11 | 1.00 | 0.071 | the group | the group |
| Single speciality hospitals in tier 1 cities | 3 | 1.02 | 0.055 | all groups | all groups |
| Single speciality hospitals in tier 2 cities | 1 | 1.02 | 0.055 | all groups | all groups |
The usual spread is on the log scale. 0.05 means similar hospitals typically sit about 5% either side of the middle. A group needs 5 hospitals for its own middle and 10 for its own spread.
Personal data is masked before any person or the AI sees it. A fixed rule flags, not the AI. The AI only drafts a note about what was flagged. A person decides. This use needs no outside data, so nothing on it is mock.