The question comes up in every sales meeting in Hyderabad: why are hospital bills going up so much? In Mumbai and Bangalore, it rarely comes up. In Hyderabad, it’s the first thing on the table.

We cover about a million people across India at Loop Health, and in our claims data, Telangana has been an outlier for years. Consultation fees and inpatient costs climbing faster than our national book.

I’d always brushed it off as vibes. Every time I fly into Hyderabad, there’s another enormous hospital going up. Gleaming towers along the Outer Ring Road, hoardings promising “world-class oncology” and “super-specialty care.” The city looks like it’s building its way to cheaper healthcare, but the data says the opposite. Checking meant querying India’s official statistics infrastructure, which I’d never have done on my own. Then the National Statistics Office launched an MCP server for the eSankhyiki API, and the friction dropped to near zero.

A Health CPI of 223.9 means prices have more than doubled since the 2012 base year. The national average grew 72% over the same period. Telangana grew 107%, nearly 1.5 times the national rate — the highest of any major state. For every ₹100 a Telangana household spent on healthcare in 2012, they now spend ₹224.

It wasn’t always this way. Through 2018, Telangana and All-India were basically neck-and-neck: the gap was 0.6 points. Then Telangana pulled away, and kept pulling.

The divergence started before COVID, but COVID made it dramatically worse. This isn’t general inflation either: Health CPI was actually below Telangana’s General CPI through 2017, dragged down while food prices kept the overall index high. In 2018, health overtook general. General CPI barely moved that year (138.6 to 138.5, essentially flat). Health CPI jumped 10% in a single year.

Something repriced, and it was specific to healthcare.

The Markup Trap

Here’s the mechanism I think explains it: the markup trap. When government reimbursement rates freeze but hospitals still need to hit their margins, they don’t absorb the loss. They pass it to other patients. The trap is that everyone assumes private healthcare is expensive because hospitals are greedy, but the dynamic is more specific than that. Stagnant public payment creates aggressive private pricing.

You’d think more patients flowing into Hyderabad would push prices down — economies of scale. But in a concentrated market where five chains set the floor, more volume just means more paying patients to load markups onto. Scale economies work when there’s price competition. Oligopolies absorb volume without passing savings through.

If this is right, you’d expect the burden to fall hardest on areas with the weakest public infrastructure, where patients have no alternative to private care. That’s exactly what the data shows.

Rural Telangana’s Health CPI grew 119% over the decade. Urban grew 89%. A CAG audit found that only 16 out of every 100 required Community Health Centers in rural Telangana actually exist. Districts like Jangaon and Rajanna Sircilla had zero functioning CHCs. When there’s no secondary care locally, patients bypass the system entirely and travel to Hyderabad’s tertiary centres. The rural CPI captures not just the price of care but the effective cost of reaching it.

In rural Telangana, healthcare costs 2.3 times what it did in 2012.

For context, 41.6% of Telangana’s regular wage workers lack any social security benefit. Not rich enough for corporate rack rates, not poor enough for government schemes. The CPI captures exactly their experience. The squeezed middle, absorbing these numbers directly from household budgets.

The Natural Experiment

Telangana was carved from Andhra Pradesh on June 2, 2014. The CPI series for both states starts in January 2014. Near-identical starting points. Same hospitals, same doctors, same infrastructure. As close to a controlled comparison as policy analysis gets.

They tracked identically through 2016. Then Telangana jumped 10% in 2018 while AP grew 2.7%. One decade later, a 35-point gap: a hospital visit that costs ₹10,000 in AP effectively costs ₹11,800 in Telangana for the same basket of care.

The policy choices diverged sharply after bifurcation. AP expanded Aarogyasri aggressively: coverage pushed to roughly 90% of the population with a ₹25 lakh annual cap and 3,257 covered procedures. It invested in village clinics, capped COVID treatment rates, and enforced the Clinical Establishments Act with compliance raids.

Telangana kept Aarogyasri targeted at BPL families with a ₹2–5 lakh cap covering 1,835 procedures, invested in tertiary infrastructure (big hospitals in Hyderabad), and took a laissez-faire approach to private pricing. Reimbursement rates to hospitals were largely frozen from 2013 to 2024. Enforcement of the Clinical Establishments Act has been described as “lethargic” in multiple reports.

The market structures also differ. Hyderabad is a mature corporate oligopoly — five or six PE-backed chains (Apollo, KIMS, Yashoda, AIG) competing on technology rather than price. AP’s hospital market is more fragmented across Visakhapatnam, Vijayawada, and Tirupati, with regional chains and doctor-owned nursing homes holding significant share. Less consolidation, less pricing power.

Twin states, different policy bets, and a 35-point gap in what people actually pay.

(To be fair: AP’s model has its own problems. Out-of-pocket expenditure remains at 52% despite “universal” coverage, and fiscal stress from premiums consumes most of the health budget. Both states rank as front-runners in the NITI Aayog Health Index. The divergence is in what healthcare costs, not in every outcome. The contrast, though, is hard to explain without the policy differences.)

Where the Money Goes

The CPI doesn’t break health into sub-items at the state level — the single biggest data gap, frustratingly. Nationally, though, hospital and nursing home charges grew 89% over the decade, the fastest component, while diagnostics grew just 46%, held down by technology and competition. This matches what we see in Loop’s claims data for Telangana: the growth is concentrated in consultation fees and inpatient costs. Not medicines, not diagnostics, but hospitals.

If retail health costs are spiking, what’s happening at the input level? WPI (Wholesale Price Index, the factory-gate prices manufacturers charge) for pharmaceuticals nationally is at 144.0, up about 44% since 2012. The raw inputs to healthcare have gotten moderately more expensive. But what consumers actually pay has grown at more than double that rate nationally, and nearly triple in Telangana.

Factory gate to consumer: pharmaceutical inputs grew 44%, but what patients pay grew nearly triple that. The markup isn’t at the factory level. It’s at hospitals and in the service delivery chain.

There’s an irony here. Telangana’s pharma industry is one of India’s largest. Output grew 158% in a decade, employment nearly doubled. Hyderabad is a global pharmaceutical hub. Local production hasn’t translated into lower local prices. The factories serve national and export markets. The CPI reflects what hospitals charge, not what factories produce.

The Strongest Hypothesis

I don’t have a causal model — I’m being intentionally unnuanced for the sake of making a point. The data does, however, point toward a specific mechanism.

Telangana’s Aarogyasri reimbursement rates (many unchanged since 2013) created a financial squeeze on private hospitals. Outstanding dues reached an estimated ₹1,400 crore by late 2024, triggering repeated service suspensions. Hospitals responded the way any business would under margin pressure: they rationed access for government-scheme patients and recouped margins by raising rack rates for everyone else. ARPOB (Average Revenue Per Occupied Bed) for listed hospital companies has been growing at 8–9% annually. That number translates directly into higher patient bills.

This is the cross-subsidisation channel. When the government underpays, hospitals don’t absorb the loss — they shift it. Stagnant government reimbursement doesn’t reduce costs. It redistributes them to the people least equipped to absorb it: the family earning ₹8–10 lakh a year in Hyderabad, facing a ₹3 lakh inpatient bill at a corporate hospital, with no government scheme to fall back on. That’s the squeezed middle.

AP faced similar payment delays but compensated with broader coverage, rate revisions, and aggressive price regulation. Its Health CPI grew 66.7%, close to the national average. Telangana’s grew 96.4%. Same starting point. Different policy architecture. Different cost trajectories.

What I Don’t Know

The 2018 inflection is the sharpest feature in the data. Health CPI jumped 10% in a single year while general CPI was flat. AP didn’t experience the same spike.

I can identify when it happened but not what triggered it. Was it a coordinated wave of hospital rack rate revisions? A lagged response to margin compression from frozen reimbursements? Something else entirely? I’m not sure.

The Aarogyasri mechanism, while consistent with the data, hasn’t been quantified. If reimbursement rates had been revised annually and hospitals paid on time, how much of the 35-point gap between Telangana and AP would close? That’s the question that would move this from provocation to finding.

There’s also a cost-side story I haven’t ruled out. The WPI comparison shows the markup isn’t at the pharma factory gate, but hospitals aren’t just drug dispensaries — they’re labor-intensive service businesses. Specialist physician salaries, nursing staff competing with Hyderabad’s IT sector for talent, and a richer acuity mix as tertiary centres attract more complex cases could all be contributing to cost growth independently of the cross-subsidisation mechanism. Decomposing the hospital margin into input costs versus pricing power would require data I don’t have.

A Note on the Tools

This analysis exists because the National Statistics Office built an MCP server for its own statistics API. MCP (Model Context Protocol) lets an LLM call external tools mid-conversation. The eSankhyiki API enforces a 4-step query sequence: datasets, then indicators, then metadata codes (you can’t guess that Telangana is state_code “36” or that the Health subgroup is indicator_code “13”), then the actual data. That’s exactly the kind of structured, multi-step retrieval that LLMs with tool access handle well. I made about a dozen different queries for this analysis. That’s a dozen CSVs I didn’t download, a dozen pivot tables I didn’t build.

India has extraordinary public datasets (eSankhyiki, the Periodic Labour Force Survey, the Annual Survey of Industries, NFHS, HMIS) locked behind clunky portals. The NSO’s beta release covers seven datasets so far. If the rest get the same treatment, the next version of this analysis writes itself.

The markup trap is real. When government payment systems freeze, private markets don’t get cheaper. They get more expensive for everyone who can’t access the public scheme. That’s not a market failure. That’s a design choice.

The people bearing it are the rural family travelling 80 kilometres to Hyderabad because their district has zero functioning CHCs. The middle-class household with a ₹3 lakh hospital bill and no safety net. The 41.6% of wage workers without social security.

The data is public. The tools now exist to query it in real time. The question is what we do when we can finally see the problem this clearly.

All CPI figures from the eSankhyiki API (CPI dataset, indicator_code “13”, base_year_code “2”). WPI data from the Office of the Economic Adviser. Analysis enabled by the NSO’s eSankhyiki MCP server.