The Pediatric Asthma Risk Score (PARS): more does not mean better

Asthma is a chronic inflammatory lung disease that affects 6.2 million children worldwide¹. The vast heterogeneity in endotypes makes it difficult to predict which children will develop asthma. The Pediatric Asthma Risk Score (PARS) is a quantitative risk score for asthma developed using clinical and demographic data that can be used by healthcare providers, parents and researchers². PARS was developed using data from the Cincinnati Childhood Allergy and Air Pollution Study (CCAAPS) birth cohort³ to identify risk factors predicting asthma development at age 7. Using just six factors, PARS outperforms and/or is less invasive than 30 published models for asthma development in children included in a 2015 systematic review^{2, 4}. The factors include eczema, early wheezing, wheezing apart from colds, and polysensitization to aero or food allergens before age 3 years, parental asthma and African-American race. PARS has increased sensitivity compared to the Asthma Predictive Index (API) and is significantly more robust in identifying children with mild-to-moderate asthma risk².

As asthma is a complex disease with a myriad of environmental risk factors not originally included in the PARS model, we sought to determine if adding these and other additional demographic and clinical factors would further increase the performance of PARS in CCAAPS. Children participating in CCAAPS completed exams at ages 1–4 and 7 where questionnaires collected allergy and asthma symptoms, demographics and exposure information and children underwent skin prick testing (SPT) to aeroallergens and foods³. All predictors for PARS were collected from the visits at ages 1–3. Asthma was defined at age 7 years in CCAAPS based on reported symptoms and objective measures of lung function².

In the present analysis, new factors considered in addition to the 6 original PARS factors were: environmental exposures including traffic pollution (defined as elemental carbon attributable to traffic (ECAT))⁵, secondhand smoke (defined as a positive response to a smoker living in the child’s home, smoking in the car when the child is present or >0 hours per day in the same area as someone smoking⁶), cat and dog ownership, demographic factors including family income, insurance type and parental education level, breastfeeding and daycare attendance.

These factors were chosen because they have previously been shown to be associated with asthma in the CCAAPS cohort^{7, 8} or in the literature^{9, 10}. We also included early frequent wheezing (defined as 10 or more episodes of wheezing in the past 12 months (top 15th percentile) at age 1, 2, or 3 years) and allergic rhinitis (AR, defined as clinician report of “probable” or “definitive” AR at age 1, 2, or 3 years based on SPT results and symptoms) which were considered but not retained in the original PARS model².

The association of each potential predictor with asthma was assessed using univariate logistic regression, followed by log-likelihood ratio tests. All significant predictors were included in a multivariable logistic model, and backward selection was used to remove factors that were least significant. For each predictor, the odds ratio (OR) was calculated, and a weight was assigned by rounding the OR to the nearest whole number. These weights were then used to calculate the updated PARS score for each subject in the CCAAPS cohort. To evaluate the predictability of the updated PARS model on asthma, a logistic regression model was generated using PARS score as the primary predictor of asthma risk. This logistic regression model was used to calculate the area under the curve (AUC) for continuous PARS measure, and the sensitivity, specificity, positive (PPV) and negative predictive values (NPV) were estimated by using a threshold of 6 and 5 for the original model and updated model, respectively, to maximize both sensitivity and specificity. Model discriminatory power was assessed by constructing receiver operating characteristic (ROC) curves and comparing the AUCs using DeLong’s test. To minimize the potential effect of sample size in model comparison, we fitted the original model restricted to the n=550 subjects included in the updated model. Statistical significance was defined as P < 0.05. Statistical analyses were performed in R version 4.1.0 (R core team, 2021).

Asthmatics were more likely to have been formula fed (p=0.001), have a family annual income less than $30,000 (p<0.001), attend daycare (p=0.027), have no commercial insurance (p<0.001), have at least one parent with a college degree (p<0.001), been exposed to secondhand smoke (p<0.001), have an average ECAT ≥0.41 ug/m³ (p=0.042) and be less likely to own a dog (p=0.030) in their first 3 years of life (Table 1).

Table 1.

Association of original and additional PARS risk factors with asthma, updated multivariable PARS model and comparison to original PARS model.

	Unadjusted Associations			Factors Retained in Multivariable Model
	Asthmatics	Non-asthmatics
	n = 89	n = 461	P-value	OR	95% CI	P-value	Weight
Original PARS Risk Factors	N (%)	N (%)
Parental asthma	50 (56.2)	171 (37.1)	<0.001	1.0 9	1.03−1.15	0.00 5	1
Eczema before age 3	40 (44.9)	112 (24.3)	<0.001	1.0 8	1.02−1.15	0.01 4	1
Wheezing apart from colds age before age 3	39 (43.8)	58 (12.6)	<0.001	1.1 7	1.07−1.29	0.00 1	1
Early wheezing before age 3	61 (68.5)	138 (29.9)	<0.001	1.1 4	1.05−1.22	0.00 1	1
≥ 2 positive SPT response to aero/food allergens age 1–3	56 (62.9)	176 (38.2)	<0.001	1.1 3	1.07−1.20	<0.0 01	1
African-American race	34 (38.2)	88 (19.1)	<0.001
Additional Risk Factors Considered
Clinical Factors
Early frequent wheezing age before age 3	32 (36.0)	49 (10.6)	<0.001
Allergic rhinitis before age 3	49 (55.1)	158 (34.3)	<0.001
Demographic and socioeconomic risk factors
Exclusively formula fed	37 (41.6)	112 (24.3)	0.001	1.0 7	1.00−1.14	0.04 6	1
Age 1 household income <$30k	37 (41.6)	99 (21.5)	<0.001	1.0 8	1.01−1.16	0.02 2	1
Daycare attendance before age 3	54 (60.7)	221 (47.9)	0.027	1.0 6	1.01−1.13	0.03 1	1
No commercial insurance age 1	36 (40.4)	91 (19.7)	<0.001
≥1 parent with college degree age 1	279 (60.5)	36 (40.4)	<0.001
Environmental risk factors
Secondhand smoke exposure before age 3	47 (52.8)	146 (31.7)	<0.001
ECAT mean age 1–3 ≥ 0.41 ug/m3	37 (41.6)	140 (30.4)	0.042
Cat ownership before age 3	20 (22.5)	139 (30.2)	0.13
Dog ownership before age 3	29 (32.6)	207 (44.9)	0.03
Comparison of Performance of the original and updated PARS models.
	Thresh old	Sensitivity	Specificity	PPV	NPV	AUC (95%CI)
Original PARS Model	6	0.69	0.77	0.37	0.93	0.80 (0.75–0.85)
Updated PARS Model	5	0.64	0.81	0.39	0.92	0.80 (0.75–0.85)

AUC: area under the curve, CI: confidence interval, NPV: negative predictive value. OR: odds ratio PARS: Pediatric Asthma Risk Score. PPV: positive predictive value.

Backward selection of all factors yielded a new model which included 5 of the 6 original PARS factors (parental asthma, eczema, wheezing apart from colds, early wheezing, polysensitization to two or more allergens) and 3 new factors (formula feeding, family annual income at age 1 <$30,000, and daycare attendance before age 3) as significant risk factors (Table 1). To examine the predictive accuracy of the original and updated PARS model on asthma, we compared the sensitivity, specificity, PPV and NPV (Table 1). The updated model had a slight increase in sensitivity (0.64 vs. 0.69) and a slight decrease in specificity (0.81 vs. 0.77, Table 1). The NPV and PPV between the two models were similar and the accuracy was unchanged (both AUCs 0.80 [0.75–0.85]; p=0.89, Table 1).

Although these additional factors are associated with asthma in CCAAPS, including them in the PARS model does not increase model performance. The updated model contained 8 factors, including 5 of the 6 original factors. The 5 original factors were all more strongly associated with asthma than the 3 new factors, further supporting the robustness of the original PARS model. The original PARS model contained African-American race, while household income age 1 <$30,000 was retained in the updated model. It is important to recognize that race is a proxy for sociodemographic factors¹¹, so it is not unexpected as race and income are highly correlated in CCAAPS (spearman correlation coefficient 0.50, p<0.001). Collectively, these results suggest that the 6 factors identified in the original PARS model are the minimum set of questions required to predict asthma in CCAAPS.

In conclusion, the performance of the PARS model was not enhanced by the addition of environmental exposures, demographics, or clinical diagnoses. Future studies should continue to evaluate the performance of the PARS model in diverse populations.

Abbreviations:

AR

allergic rhinitis

AUC

area under the curve

CCAAPS

Cincinnati Childhood Allergy and Air Pollution Study

ECAT

elemental carbon attributable to traffic

NPV

negative predictive value

OR

odds ratio

PARS

Pediatric Asthma Risk Score

PPV

positive predictive value

ROC

receiver operating characteristic

SPT

skin prick test