Developing Prediction Scores & Risk Values: Collaboration is Key

Segment Description: Julie Ann Justo, PharmD, MS, BCPS-AQ ID, discusses the source of prediction scores and how antimicrobial stewards work in concert with other clinicians to develop prediction scores.

Interview Transcript: (modified slightly for readability)

Julie Ann Justo, PharmD, MS, BCPS-AQ ID: These risk scores come from a variety of areas. Risk models are usually multivariate logistic regression models, and these can be found in the medical literature going back decades. And so, it really depends—you can have some outbreak folks, [or] maybe clinicians putting it together; it can really vary.

What we’re seeing a lot of now is actually antimicrobial stewards are working in concert with frontline clinicians to figure out what it [is] that we actually want to predict for. For example, our institution at Palmetto Health Richland in Columbia, South Carolina, we’re really big on bacteremia because we feel like it’s a specific infectious syndrome that has real impacts if you’re not quite so good at guessing the empiric antimicrobial.

And so, a lot of what I think is some of the most interesting prediction scores that are coming out now are actually coming from this combination of frontline—either intensive care unit physicians or hospitalists—in concert with antimicrobial stewards. And the reason why I say that is, [is] because you can have existing risk scores that might have somewhat general risk factors. For example, a lot of risk scores will say [prior antibiotics] if you’re trying to predict for a resistant organism. And if I’m reading a paper like that, I’m kind of left scratching my head because that is exceedingly general. Is it the prior antibiotic that they got a week ago? Three years ago? Is it the 1-time dose? The 2-week course? Is it in the narrow-spectrum? Is it the broad-spectrum? All of these things are questions that you could be left asking with a characteristic and a risk score that is not well defined.

Some of the more innovative risk scores that are coming out now in the literature are doing a better job of testing some of these different characteristics. They might test prior antibiotics in terms of turning it into a continuous variable, measuring the amount of antibiotic exposure and creating a differential for how that changes the patient’s individual risk for resistant organism.