Overview

We receive frequent questions from our users about our metrics. Many similar questions start out the same but have different endings. Here it is, "Why doesn't this metric in Trella match:

the metric I am seeing in my EHR data?
what my doctor says about his/her patients?
the metric from some other data provider?

This question has numerous additional variations.

Summary:

In short, the answer is that the metrics don't match because they aren't calculated the same way. To make them match, we need to make sure we are using the same population of claims, looking for the same metric, and calculating it the same way.

Lots more detail:

There is a broad, conceptual answer to these questions, and a detailed one that explains, well, in more detail. But first, some background.

Where do Trella Health Metrics come from?

You are going to want to look at the article Data Sources if you want the detailed answer to this question.

In general, we write detailed queries that run against the data from every Medicare Fee for service claim each time new data is available. An example query would be "Give me counts of every distinct patient treated by every physician from January 1, 2020 through December 31, 2020." As a privacy restriction, we would not get back any counts of patients less than eleven. But what we would get is millions of numbers because there are lots of physicians. For every physician (who submitted Medicare claims under their NPI) we receive the number of distinct patients they treated in the specified year.

That patient count query is pretty simple. (Although the actual query is far more complex because it is written in a database language so that there is no ambiguity.) Other queries are super complex. To calculate a home health agency readmission rate, we need to identify all patient discharges that match with a home health admission that also matches with an inpatient admission, and those three claim events all need to happen within 30 days. And that only gives us the numerator for one agency. Yikes!

The point is that we use about 150 of these queries to generate about 1900 different metrics that we put into 200 different tables for millions of providers. That is a lot of individual counts and rates and scores.

Of course, we perform quality control checks to make sure that our final metrics are comprehensive and accurate.

Now let's turn our attention to why Trella Health metrics don't match some other metric.

Telling a Story

Trella Health metrics are not just a dump of every possible patient event. Trella Insights, for example, is designed to show facility and physician sources and post-acute destinations in the care continuum that have a close logical proximity. The reason for this is that Medicare claims don't include any patient referral information. Instead, we use analytics to create metrics that identify referral potential. Let's look at an example. The patient count we call "HHA Patients" is not a count of every patient discharged from a hospital and admitted to a home health agency in the same year. Although that broad, comprehensive number might be useful, we are looking for something more specific. "HHA patients" is a count of patients discharged from a hospital and admitted to home care within 30 days. That count gives us a number of patients where the source hospital could have or should have referred a patient to post-acute care.

In other words, we have identified a count of patients who you could compete for by presenting your competitive advantage. Another angle is that this count is the number of patients who are proven to be home health appropriate (since they were admitted to home care on the heels of a discharge) and shows the size of the patient population you can expect in the future. Go get 'em!

To tell this story, Trella health metrics have strict parameters, which allows us to segue to our next topic.

Lots of variables

When comparing metrics from different sources, even if the source data is the same, how you slice it can be very different. Let's look at some of the differences.

It is important to note that health networks of any size can submit claims from one facility under the NPI of another. We can only do our calculations on the claims as submitted. The NPI for the actual place of service is not required on a claim.

What you can do.

If you see a difference between a metric in your EHR and in a Trella product, there are some steps you can take to bring those metrics in alignment. In short, we need to figure out how each metric was calculated.

Each Trella metric has a definition that summarizes the details of the metrics. This is a very general definition, so you will probably need to:
Search the Trella Health Help Centers to see if we have a more detailed formula included there.
Send an email to Trella Health Support for a more specific formula.
Get in touch with your Customer Success Manager for help digging in on the details.

Parameters that impact why metrics are different
Reporting Period	For metrics to match, you need to make sure that the metrics are drawn from claims submitted during the same time period. Most Trella Health counts come claims submitted during a 1 year period. Performance rates are calculated from a 2 year period. Sometimes, in order to leave time for another claim event to occur, like a readmission, we need to offset the one year period to allow time for a readmission to take place after the discharge claim. Most trended counts have a single quarter as the reporting period. For metrics to match, we need to make sure we are counting from the same reporting period.
Reference Periods Examples of metrics that depend on a reference period: HHA/HOS/SNF patient counts for facilities Readmission rates	A reference period is that period of time that we use to limit a connection between two events. Imagine the following metric: Count all patients discharged from hospital H and admitted to agency P in 2021. H P - any time in 2021 As you can imagine, that will be a lot of patients, and would mostly likely be what you would see in your internal Medical records. But we are trying to draw a connection between each discharge and admission, so our reference period is 30 days: H P - must be within 30 days. This count is a sub-set of the first count. Some metrics have a "built in" reference period required by definition, for example, readmission rates, for which the reference period between the initial inpatient discharge and the subsequent inpatient admission must be 30 days.
Duplication	Most of our patient counts are for distinct patients during the reporting period. The easiest way to visualize this is to consider a patient who is admitted to home care three times in a one quarter. That patient would only be counted once for that period as a distinct patient.
Count type	In short, depending on what story we are trying to tell, our analytics can count: Distinct patients - Example: Patient counts are almost always unduplicated counts All admissions - Example: the Market Share report has admission counts All discharges - The Discharge Events by Setting table contains counts of discharges. All claims Each of these different counts tells a different story. Any organization will have more admissions/discharges than patients.
Counting Period	This is similar to a reference period. The difference is that a reference period is fixed; 30 days, 1 year, etc. A counting period is the number of days between two claim events. Example: Average Length of Stay. Starting with all discharges in a one year reporting period, we find the claim for the start of care going back for a specific period of time during which we count the number of days.
Inclusions/Exclusions/	There are dozens of other parameters that we use to either include or exclude data in our analytics. These tend to be highly specific in order to identify connections that aren't explicit from the claims or to remove data that would skew the final metrics. In most cases, these highly specific parameters are described explicitly in the Help Center content for the metric.