In the search for practical prognostic decision support, numerous childhood asthma prediction tools (including a recent Pediatric Asthma Risk Score [PARS]) with modest prognostic accuracy have been developed, however, the prognostic utility of these tools using existing electronic health records (EHR) in clinical settings is unknown. To test the hypothesis that childhood asthma can be predicted using EHR, we sought to externally validate and update the PARS as a passive digital marker (PDM) for asthma risk.

Using a retrospective, population-based observational study design, children born between 2010 and 2017 who were consecutively enrolled at any of the pediatric healthcare institutions that contribute EHR data to the Indiana Network of Patient Care (INPC) databases were included in our analyses. Logistic and Cox proportional hazards models were used to validate and update the EHR-based PARS as a PDM for the prediction of physician documented diagnosis of asthma between ages 4-11 years.

Among 69,109 eligible children, of whom 5290 (7.65%) had a confirmed asthma diagnosis after age 4-years, our PDM had a higher prognostic accuracy (Area Under the Curve (AUC): 0.79; 95% CI: 0.78, 0.80; sensitivity-0.71 and specificity-0.74) than the EHR-based PARS (AUC: 0.76; 95% CI: 0.75, 0.76; sensitivity-0.74 and specificity-0.68) for early case detection. Both the PDM and EHR-based PARS had satisfactory calibration. For children classified as high-risk at age 3-years, the incidence of asthma was higher using the PDM than the EHR-based PARS (37% vs. 26%, p < 0.0001).

It is feasible to use EHR data for childhood asthma risk prediction by updating existing tools (e.g., PARS) with relevant clinical context to assure high prognostic accuracy and clinical utility during early childhood, a period of diagnostic uncertainty.

This study was supported by National Institutes of Health grants, K01HL166436 (AHO), R03HS029088 (AHO) and P01HL158507 (BG).