Real World Fitness Score - How is it calculated?

We have previously given a brief overview of the Real World Fitness Score, what value it brings, and how to use it today, but we wanted to go into more detail on how it is calculated.

Briefly - the Real World Fitness Score is calculated to tell us "How well a dataset fits our specific question".

We do this by translating each of the questions asked of us into PICOT format - Population, Intervention, Control, Outcome, Time.

This allows us to standardize each question and understand exactly how our analysis will be crafted and coded in our proprietary query language, TQL.

Once translated in PICOT and coded in TQL, the analysis is run and returns results as you might typically expect - effect sizes, p-values, and e-values, for a number of matching scenarios (unmatched, basic matching, and propensity score matching - more on that in a separate article).

To compliment these traditionally methods, we also calculate the RWFS, that goes one step deeper than the PICOT to connect and calculate the fit of the dataset.

This includes evaluating the following metrics -

Volume

Intervention - How many patients have some occurrence of the intervention?

Data Coverage

History and Follow-up - What are the average days of history and follow-up for patients included in the cohort?

Outcomes

Diseases and Conditions of Interest - How many patients fit our outcome of interest (ICD9, ICD10, etc.)?

Breadth

Comorbidities - How sick is the population of interest? How many additional comorbidities does the cohort have?

The greatest value of the RWFS comes from asking the same question of different datasets. In that you get two analyses conducted and can compare which results to potentially trust more based on the RWFS.

We hope this gives you a slightly deeper look into the Real World Fitness Score (RWFS) and how it's defined. In the future, we will be publishing a manuscript diving into even greater detail on our methodology.