The Pediatric Asthma Risk Score (PARS) Predicts Allergic and Non-Allergic Asthma

Asthma is a heterogenous disease including allergic, non-allergic, and other endotypes that complicate the development of tools to identify at-risk children. The Asthma Predictive Index (API) was developed in 2000¹ and has remained a useful clinical and research tool to identify children at-risk for developing asthma. The API, however, is more specific than sensitive and is dichotomous limiting its utility in gauging children at moderate risk. Subsequent iterations, such as the Modified API (mAPI)² have made only incremental improvements in accuracy.

The Pediatric Asthma Risk Score (PARS) is a quantitative scoring system developed using clinical and demographic data that can be collected during a typical outpatient visit with any healthcare provider³. The PARS risk factors are parental asthma, eczema ages 1-3, early wheezing age 1-3, wheezing apart from colds, polysensitization (aero or food allergens), African ancestry.

PARS has a higher sensitivity and similar specificity compared to the API and mAPI³. We sought to determine the PARS accuracy in predicting allergic and non-allergic asthma subtypes in two cohorts of children and compare these results to the API and mAPI.

The primary population was the Cincinnati Childhood Allergy and Air Pollution Study (CCAAPS), an at-risk cohort of 762 infants born to atopic parents. Children were identified by birth records from Greater Cincinnati, Ohio from 2001-2003⁴. Children had examinations at ages 1,2,3,4 and 7 years of age with allergy symptoms, wheezing frequency, wheezing apart from colds, and skin symptoms recorded. At age 7, 589 children were assessed and n=95 had asthma based on reported symptoms and measures of lung function⁵. Allergic asthma (n=71) and non-allergic asthma (n=24) were defined as asthma with or without a positive skin-prick test (SPT) at any time point, respectively.

The U.K. population-based, Isle of Wight 1989-90 birth cohort (IOW, n=1,456) of children of European descent served as the replication population⁶. Asthma and allergy data were collected at ages birth, 1, 2, 4, 10 years. Allergy was defined as a positive SPT at any time point. Asthma (n=156, 90 allergic and 66 non-allergic) was defined at age 10 years based on physician-diagnosis plus at least one episode of wheeze in the prior 12 months. For CCAAPS and IOW, PARS was calculated according to the scoring sheet (Supplemental Figure 1).

Receiver operator characteristics (ROC) for the API (both “stringent” (early frequent wheeze and 1 major or 2 minor criteria) and “loose” (early wheezing and 1 major or 2 minor criteria)) and mAPI for both CCAAPS and IOW cohorts were calculated based on published criteria. ROC for PARS for all possible scores was determined for both cohorts. The area under the curve (AUC) was compared using DeLong’s test (p<0.05 considered significant). Sensitivity, specificity, predictive values, and likelihood ratios were determined for all possible scores with the optimal sensitivity and specificity reported. Analyses were performed in R software⁷.

The AUC analysis (Figure 1, Supplemental Table 1) demonstrates that the PARS (red line, Figure 1) predicted allergic asthma significantly better than the API (loose criteria, green line, Figure 1) in CCAAPS (AUC=0.82, 95% C.I. [0.77-0.87] vs 0.71 [0.65-0.77], p=0.004) and the IOW cohort (0.87 [0.84-0.91] vs 0.74 [0.69-0.79], p=7E-5). PARS (red line, Figure 1) also predicted non-allergic asthma better than the API (green line) in both cohorts, though it was only significant in IOW (CCAAPS: 0.71 vs 0.63 p=0.32; IOW: 0.74 vs 0.64, p=0.04). Results for the mAPI (blue line, Figure 1) were similar to the API in both populations for allergic and non-allergic asthma.

Figure 1:

ROC for asthma prediction models applied to allergic and non-allergic asthma in the CCAAPS and IOW cohorts

For allergic asthma, the PARS (at the cut-point of optimal sensitivity and specificity) had an increased sensitivity when compared to API (loose) in the CCAAPS (0.72 vs 0.61) and the IOW (0.83 vs 0.62) cohorts (Supplemental Table I). Similarly, in non-allergic asthma, the PARS had an increased sensitivity in both CCAAPS (0.58 vs 0.46) and IOW (0.62 vs 0.42). The specificity of the PARS was lower than the other indices in both allergic (CCAAPS: 0.77 vs 0.81; IOW: 0.78 vs 0.86) and non-allergic asthma (CCAAPS: 0.77 vs 0.81; IOW: 0.84 vs 0.86). The NPV for all indices in both allergic and non-allergic asthma were similar and greater than 0.90. The PPV for all indices were low in both allergic and non-allergic asthma for the two cohorts.

The PARS outperformed both the API and mAPI in prediction of allergic and non-allergic asthma development in two very different cohorts of children with a significant increase in the AUC and an improved sensitivity with only a marginal decrease in specificity. When compared to the API there was little impact on NPV or PPV, and compared to the mAPI, a small decrease in PPV for allergic asthma. In non-allergic asthma, the AUC was found to be significant in the IOW but not in CCAAPS (possibly related to statistical power). The sensitivity for non-allergic asthma was improved with only a marginal decrease in specificity. The PPV was similar in all indices for non-allergic asthma.

It is not unexpected that PARS, mAPI, and API performed better in allergic asthma than non-allergic asthma given that the factors in these indices include sensitization, eczema, eosinophilia, and allergic rhinitis. However, the PARS, a continuous scoring system, was more accurate at predicting both allergic and non-allergic asthma than the mAPI and API (both dichotomous), reinforcing that a quantitative measure is more informative. Comparing the statistics of the indices beyond the AUC of the ROC is somewhat difficult. The PARS has variable sensitivity, specificity, predictive values for each possible score; whereas the API and mAPI, being dichotomous, have fixed values. The increased accuracy of these predictive tests for allergic asthma is clinically meaningful as this endotype is most prevalent and the rate is rising⁸.

In conclusion, the PARS had a significant performance improvement in the ability to accurately forecast the development of allergic and non-allergic asthma over the API and mAPI. Future studies need to be directed at 1) replicating and building upon these findings in larger and diverse cohorts with ethnicity not represented here, 2) application of the PARS to identify at-risk children for asthma intervention trials.