Underwriting Medical Malpractice with Historical Outcome Data

An underwriter prices claims that have not happened yet. The premium has to cover how often claims arrive, how often they pay, and how much they cost when they do. Two of those three questions have answers in the historical record. This post is about using those answers well.

It is written for carriers, captives, risk retention groups, and the underwriters who work for them. Our earlier posts spoke to the attorney valuing one case and the risk manager pricing a whole system. The underwriter sits between them, pricing many risks before any of them become claims.

The market is asking for sharper pricing

The line has been hard to price profitably. S&P Global Market Intelligence, as reported by Insurance Journal, estimates that medical professional liability ran a calendar-year combined ratio above 105% in 2025. That was the fifth time in eight years.

The same report shows how unevenly the pain is spread. In 2025, the direct incurred loss and defense cost ratio was 143.8% in Utah, 128.8% in New Mexico, and 125.7% in South Carolina. The U.S. aggregate was 75.9%. A book can look sound in total and still lose money in specific states.

When results swing that much by place, the question is not whether to segment. It is how finely.

What outcome data adds to an underwriting file

Most underwriting files already hold the basics: specialty, territory, limits, claims history, and practice setting. Outcome data does not replace those fields. It puts a measured value behind several of them.

Venue, measured rather than assumed

Territory rating captures some of the venue effect. Outcome data shows how large that effect really is. In the historical record, the median payout for comparable cases is about $73,000 in California and about $507,000 in Pennsylvania. That is roughly seven times apart for the same kind of case.

For a group practice or facility near a state line, or one with locations in several states, that gap belongs in the price of each location, not in a blended average.

Severity mix, not just claim count

Two specialties can produce the same number of claims and very different losses. The difference is in the severity of the injuries they tend to produce, and in what those injuries pay.

Anesthesia is a clear example. Anesthesia allegations have the highest median payout of any allegation category in the dataset, at $507,000. They are also among the least common. A rare, severe claim type is easy to underprice when experience is thin, because a small book may go years without seeing one.

Practitioner history as a priced factor

Practitioners with five or more prior adverse reports settle for about 43% more than first-time reporters. Many applications already ask about prior claims. The open question is whether the answer moves the price by a measured amount or only by a judgment call.

Survival severity, not just mortality

Cases that end in grave permanent injury carry a median payout of about $807,000. Cases that end in death carry about $311,000. For any specialty where a patient can survive a catastrophic injury, that ordering matters more than intuition suggests. A book priced as if death were the worst outcome is priced to the wrong ceiling.

Trend that has not finished developing

Payouts in the dataset trend upward, with a turn around 2018. The S&P report adds a caution: recent accident years look calmer than they are, because their claims have not matured. Loss picks built on recent experience alone are likely to start low.

Using case-level predictions for a book

A MedMalPredict prediction is built one case at a time. It returns the payment probability, a payout range in current dollars, and the distribution of outcome severity across comparable cases. Each one draws on 270,000+ resolved malpractice claims from 2004 to 2026, across all 50 states and U.S. territories, and returns in seconds.

An underwriter can still use that at the level of a book. The practical approach is to run the scenarios that drive the exposure:

List the scenarios that would hurt most. For a given account, these are the allegation and severity combinations its specialties are most exposed to, in the states where it practices.
Run each scenario as a case profile. Note the payment probability, the payout range, and how wide the severity distribution is.
Compare the results to the limits and the premium. If the high end of a likely scenario sits near the policy limit, the limit is doing real work. If several scenarios carry long tails in the same venue, that venue deserves a closer look.
Repeat when the account changes. A new location, a new specialty, or a new practitioner with prior history changes the picture.

This does not produce a rate. It produces evidence that a rate, a limit, or an attachment point can be checked against.

Where outcome data stops

Be clear about what a prediction does not measure.

It does not predict frequency. A payment probability is the chance that a claim with a given profile results in payment. It is not the chance that a practitioner will face a claim at all. Claim frequency still comes from your own experience and your actuarial work.

It does not predict defense costs, and it does not estimate incurred-but-not-reported losses.

It is a statistical estimate, not legal or actuarial advice. Underwriting judgment, loss control findings, and the details of each account still matter.

What outcome data offers is a shared factual floor. It shows what claims like these have actually paid, with the spread intact, so pricing decisions start from evidence rather than from the last large loss someone remembers.

Getting started

MedMalPredict is built for single predictions and for volume. Predictions cost $2,500 each, or $1,500 to $2,000 each in 5-packs and 12-packs. Organizations that need custom terms, scale, or the ability to rerun predictions can talk to our team about an Enterprise arrangement. Your first prediction is free for a limited time, so you can test the approach on a scenario you already know well before you rely on it.