What the Claim Text Reveals

The structured claims data classifies every admission into one of a few dozen ailment categories. That classification is clean, complete, and the spine of every other cost-signals page in this report. It is also lossy. A single admission gets one primary category — cardiac, or renal, or injury — and the metabolic disease sitting underneath it disappears. The free text is where that underlying disease is still visible. This page reads the text for what the categories drop.

Two in three claims never name a condition#

Of the 74,339 claims in the latest twelve months, only 33.9% carry a diagnosis specific enough to name a disease. The other 66.1% record the admission at the level of an ICD chapter heading, or leave the diagnosis line blank.

This is the first and most important fact about the dataset, and the reason the rest of the page is framed as signal rather than measurement. A keyword scan over claim text can only ever see the third of the book that was written up specifically. Whatever pattern it finds is a pattern in that third — useful as a directional read, useless as a denominator.

Chronic-disease claims cost more, and comorbid ones cost more again#

Among the claims that do name a condition, the ones whose text mentions a chronic disease cost materially more than the ones that don't.

The chronic-disease claim runs 1.47× the cost of the other specific-condition claims. That gap widens sharply when the text names more than one chronic condition at once.

A claim whose text names two or more chronic conditions costs 2.05× a single-condition claim. The arithmetic is not subtle: comorbidity, where it is documented, roughly doubles the bill.

Cardiac and coronary language carries the heaviest cost#

Scanning the diagnosis text for chronic-disease keywords, ranked by the incurred expense on the claims that mention them, the order is led by the heart.

The shares are small because they are shares of mentions, capped by the third of claims that name anything at all. The cost behind them is not small. Cardiac and coronary language appears on the records carrying ₹7.1 Cr of incurred expense, at an average of ₹1.11 lakh a claim — the most expensive chronic-disease signal in the text by a wide margin. Renal failure mentions are frequent but comparatively cheap per claim, which fits a population where a meaningful share of renal admissions are recurring dialysis sessions rather than acute crises.

The two metabolic conditions at the root of most of this — diabetes and hypertension — sit low on the list by their own incurred totals. That is exactly the limitation the free text exposes. Diabetes is rarely the line a hospital admits a patient under. It is the condition named underneath the admission, and the structured category never sees it.

Diabetes and hypertension are the comorbidity the categories drop#

The clearest thing the free text adds to the structured data is co-occurrence — which chronic conditions get written down together on the same claim.

Every one of the five most-common pairs contains diabetes or hypertension. The metabolic pair leads, and from it the links run out to the heart, the kidney, and the brain — the three organs the biomarker pages of this report flag as the downstream destinations of poor glucose and lipid control. The text is drawing the same map the blood panels draw, from the other end of the disease.

Where metabolic disease is named alongside one of those downstream organ failures, the cost confirms the severity — the same cardiac and renal lines that sit at the top of the Top Categories by Incurred page. Of the claims whose text names diabetes or hypertension, 14.5% also name a separate organ disease — heart, kidney, liver, or brain — and those claims average ₹1.39 lakh, more than double the all-specific-claim average. These are the admissions where the metabolic process has already done structural damage and the hospitalisation is paying for it.

Chronic-disease language concentrates in the parent layer#

The chronic-disease keywords are not spread evenly across the people on the policy. They cluster in the oldest layer.

Chronic-disease language appears in 6.1% of parent claims against 1.7% of employee claims — a 3.6× concentration. This is the expected direction, given the parent layer is retirement-age and the employee layer is working-age. What makes it worth surfacing is the magnitude relative to the mention rate overall, and the fact that it lines up exactly with the dependent-layer thesis running through the rest of the report: the spouses and parents on the employer's policy are in worse metabolic shape than the employee, and they drive the heaviest end of the bill. The comorbidity claims bear this out — the parent layer accounts for the clear majority of the claims whose text names two or more chronic conditions at once.

The employee number is the one to sit with. It is low — 1.7% — but it is text on working-age people, the population the structured data and the biomarker panels both show carrying early glucose and lipid dysfunction that has not yet reached a hospital. The parent layer is what that text looks like twenty years on, once the metabolic disease has become a cardiac or renal admission. The two rows of the table are the same disease at two ends of a career.

Most free-text notes are about evidence, not illness#

The diagnosis line is one of three free-text fields scanned. The other two — the deficiency and deduction notes the insurer writes when querying or trimming a claim — are about the handling of the claim, not the patient. Mined the same way, they reveal the workflow rather than the workforce.

The single most common note, on 13.2% of all claims, is a request for an investigation or diagnostic report — the insurer asking for the lab work or imaging that substantiates the admission. The second, on 5.9%, flags a congenital condition, a pre-existing disease, or a waiting-period question. These are friction signals, not health signals, and they belong on this page only as the counterweight to the diagnosis mining above: most of the writing on a claim record is the administrative back-and-forth of adjudication, which is exactly why the third of records that do describe the patient are worth reading closely.