How to Read a MedMalPredict Report

A MedMalPredict report is longer than a single number, on purpose. It walks a case from a plain historical baseline through the model, the human factors, and the probability of any payment at all, then surrounds that prediction with the benchmarks and drivers behind it. Read in order, each section answers a different question.

Here is what every section shows and how to use it. To follow along, open the sample report in another tab and match each section as you read.

Start with the case profile

The first page lays out the case profile the prediction is built on: the incident details (allegation, injury severity, year, setting), the defendant (state, license type, prior reports), and the plaintiff (age, sex, and the human-factors flag).

Read this first, and read it closely. Every number in the report follows from this profile. If the venue, the injury severity, or the allegation is wrong, the prediction is answering the wrong question. It also tells you whether human factors were applied, which changes how you read the payout.

The payment page has three parts

The prediction results page is the heart of the report, and it holds three distinct pieces. They are easy to blur together, so read them as three separate answers.

1. Predicted payout

This is the headline: a single recommended settlement amount, adjusted for human factors and expressed in current dollars. Next to it sit three quick reference points: where the case falls in its cohort (percentile), how far it sits above or below the median, and how many similar cases the figure is drawn from.

The percentile matters as much as the dollar figure. A payout in the 98th percentile is a high-value outlier; the same dollar amount at the 50th percentile would be an ordinary case. The number alone does not tell you which.

2. Payment probability

This is the chance the case results in any payment at all, shown against a baseline and a typical range. A case reading well above the baseline is in elevated-likelihood territory; one below it faces longer odds.

Payment probability answers a different question from payout: not how much, but whether. A high payout attached to a low probability is a different case from a modest payout that almost always pays, and this is the number that tells them apart.

3. Payout options: the four-stage build

This is the part most people miss, and it is the most useful. The report does not produce the payout in one step. It shows the four stages that build it, each one adjusting the last:

1. Empirical baseline: the median of similar historical cases, before any modeling. This is the plain "what have cases like this paid" number.
2. Hooper Model: the model's prediction from the case features alone, with a range that reflects real variability in outcomes.
3. Human factors: the feature-based prediction multiplied by the human-factors adjustment, when human factors were entered.
4. Expected value: the human-adjusted figure weighted by the payment probability, giving a risk-adjusted benchmark.

Reading the four together shows you where the number comes from and how much of it is baseline, model, human factors, or probability. If most of the lift comes from human factors, the case rests on non-economic arguments; if it comes from the baseline, the case is strong on the raw comparables alone.

Human factors analysis

When human factors were entered, the report devotes a full section to them. It shows a composite score and the multiplier it produces, built from four weighted categories: pain and suffering, loss of lifestyle, family impact, and jury appeal. Each category is scored from its own sub-factors.

This section explains the human-factors stage on the payment page. If the multiplier is doing a lot of work, this is where you see which categories drove it, which is the argument you would need to support at mediation. If human factors were not entered, this section does not apply.

Contextual intelligence

The prediction is one number; this section is the evidence around it. It gives you several ways to see how the case sits against history:

• Similar case benchmarks: where the prediction lands across the percentile range of matched historical cases.
• Resolution timeline: how long cases like this typically take to resolve, overall and for this license type.
• Payment distribution: how historical payments cluster across dollar ranges, with the predicted range highlighted, so you can see how common or rare this outcome is.
• Historical payment trend: whether awards in this license type and severity have been rising, falling, or flat over time.
• State vs national and license type vs national: how this jurisdiction and this license type compare to the national median, with a rank.

Use this section to pressure-test the prediction. A predicted amount in a sparse part of the payment distribution, or a jurisdiction that ranks far above the national median, tells you where the case's leverage and risk actually sit.

Outcome severities

This section shows two distributions side by side: how this allegation type has resolved across injury-severity levels historically, and the model's predicted severity distribution for this case. The claimed severity is marked on both.

Comparing the two is the point. If the model's predicted severity sits above or below the claimed severity, that gap is a flag worth understanding before you rely on the payout. The cumulative probability of catastrophic outcomes, shown here, is often the number that matters most for exposure.

Case summary: the drivers

The final page condenses everything into the factors that moved the prediction. Primary drivers are the factors pushing the number up or down, each one grounded in the count of historical records behind it. Moderating factors are the ones pulling the other way. A full analysis paragraph ties them together.

Read this section first if you are short on time and last if you want the detail, because it tells you not just what the prediction is, but why. When a driver is grounded in tens of thousands of records, it is a stable signal; when it rests on a few hundred, treat it with more caution.

Put it together

Read end to end, the report moves from a plain historical baseline to a case-specific prediction, then shows its work: the stages that built the payout, the human factors behind any adjustment, the benchmarks around it, and the drivers that moved it. A strong case is one where the baseline, the model, the probability, and the drivers point the same way. Where they disagree, that disagreement is the most useful thing in the report, because it shows you exactly where the case carries its risk.

Every report downloads as a PDF built for a case file. It is a record of what the data said on the day you ran it, ready for an intake memo, a mediation, or a conversation with a client.

Try It

Run a case you already know the outcome of, and read the report end to end. Comparing what the data said to what actually happened is the fastest way to learn to read the next one.

Try MedMalPredict

---

MedMalPredict AI is not legal advice. Predictions are based on historical data and represent probabilities, not guarantees.