Predicting the long-term prognosis of children with symptoms suggestive of asthma at preschool age

Article Outline

• Abstract
• Methods
  • Study population
  • Predictor variables
  • Outcome measures
  • Data analyses
• Results
  • Study population
  • Outcome of asthma
  • Univariate analysis
  • Multivariate analysis
• Discussion
• Acknowledgment
• Methods
• Results
  • Comparison with previous literature
  • External validity of the prediction rule
• Fig E1. 
• Fig E2. 
• Table E1. 
• Table E2. 
• Table E3. 
• References
• References
• Copyright

Background

Clinicians have difficulty in diagnosing asthma in preschool children with suggestive symptoms.

Objective

We sought to develop a clinical asthma prediction score for preschool children who have asthma-like symptoms for the first time.

Methods

The Prevalence and Incidence of Asthma and Mite Allergy birth cohort followed 3,963 children for 8 years. Between 0 and 4 years of age, 2,171 (55%) children reported “wheezing,” “coughing at night without a cold,” or both. In these children possible predictor variables for asthma were assessed at the age respiratory symptoms were first reported. Asthma was defined as wheezing, inhaled steroid prescription, or a doctor's diagnosis of asthma at both age 7 and 8 years of age.

Results

Eleven percent of children with symptoms at 0 to 4 years of age had asthma at 7 to 8 years of age. Eight clinical parameters independently predicted asthma at 7 to 8 years of age: male sex, postterm delivery, parental education and inhaled medication, wheezing frequency, wheeze/dyspnea apart from colds, respiratory infections, and eczema. In 72% of the cases, the model accurately discriminated between asthmatic and nonasthmatic children. A clinical risk score was developed (range, 0-55 points). Symptomatic children with a score of less than 10 points had a 3% risk, whereas children with a score of 30 points or greater had a 42% risk of asthma.

Conclusion

A risk score based on 8 readily available clinical parameters at the time preschool children first reported asthma-like symptoms predicted the risk of asthma at 7 to 8 years of age.

Key words: Asthma, children, wheeze, cough, prognosis, prediction, longitudinal, birth cohort

Abbreviation used: PIAMA, Prevalence and Incidence of Asthma and Mite Allergy

Most preschool children with symptoms suggestive of asthma do not have asthma and are unlikely to respond to asthma treatment.¹, ² If simple clinical parameters could identify children with early symptoms who have a high risk of asthma, this would allow for better targeting of secondary prevention measures and treatment for those children who are most likely to benefit. Also, doctors could be more restrictive when prescribing treatment to those who probably have transient conditions other than asthma.

A number of factors that might help to identify early-onset asthma have been reported, including a family history of atopy, eczema, and wheezing or wheezing frequency.³, ⁴, ⁵, ⁶, ⁷, ⁸ Only a few studies predicted asthma in children at the age when symptoms occurred.⁴, ⁵ Recently, Frank et al⁵ followed 201 wheezing children and reported “wheeze after exercise” and “family history of atopy” as the only significant predictors of later wheeze. However, because these data were not prospectively collected from birth, these authors were unable to exclusively select children with a first episode of symptoms. This would be important because prediction of asthma becomes clinically relevant as soon as suggestive symptoms first appear.

The aim of the present study was to develop a prediction rule for the development of asthma in children who have their first symptoms between the ages of 0 and 4 years, that could be useful in primary care settings, and that is based on data collected prospectively over a long period of time. For this purpose, we used data from the Prevention and Incidence of Asthma and Mite Allergy (PIAMA) birth cohort, in which 3,963 children were followed from birth up to the age of 8 years and in which adherence was very good. In children who had reported symptoms suggestive of asthma between the ages of 0 and 4 years, the combination of factors that best predicted asthma at the age of 7 to 8 years was assessed by using variables that can be easily obtained in practice. The resulting model was translated into a simple score that is feasible in a primary or secondary care setting to quantify the likelihood of asthma.

Back to Article Outline

Methods 

Study population 

Recruitment took place in 1996-1997. A screening questionnaire was distributed to 10,232 pregnant women who attended one of 52 prenatal clinics in The Netherlands. Based on this screening, 7,862 women (2,779 allergic and 5,083 nonallergic women) were invited to participate in the study; 4,146 agreed and provided written informed consent. Questionnaires for parental completion, partly based on the International Study of Asthma and Allergies in Childhood core questionnaires,⁹ were sent to the parents during pregnancy, when the children were aged 3 and 12 months, and yearly thereafter up to the age of 8 years. The eligibility criterion for the present study was at least 1 positive response to the following questions in the annual questionnaires at age 1 to 4 years: “Has your child had wheezing or whistling in the chest in the last 12 months?,” “Has your child had cough during the night, when he/she did not have a cold or a chest infection, in the last 12 months?,” or both. Children who did not have these early symptoms were not eligible and were excluded from our analyses. Details of the study design have been published previously.¹⁰ The PIAMA birth cohort study included an intervention part investigating the effect of mite-impermeable mattress covers. Of the population eligible for the current study, 242 (11%) children had used these covers. Because this had no effect on any of the investigated associations, these children were included in the final analyses. A screening questionnaire was used to select only atopic mothers for the intervention part. Importantly, the final study population (including the intervention and the natural history parts) was not enriched for atopic mothers. See the Methods section and Fig E1, Fig E2 in this article's Online Repository at www.jacionline.org for further details on study design. The study protocol was approved by the medical ethics committees of the participating university hospitals.

Predictor variables 

Based on previous literature and availability within our dataset, we selected 26 candidate predictor variables. All variables had to be easy to assess in general practice and not involve invasive tests or time-consuming measurements. Parental questionnaire data were used to asses all variables at the age of first report of symptoms. To mimic clinical practice, when composing the predictor variables, we only used information gathered up to the year of first presentation. Candidate predictor variables were a family history of atopic diseases (6 variables tested), perinatal factors (8 variables tested), environmental factors (4 variables tested), and the child's pattern of symptoms (8 variables tested, Table I). Age at first symptoms was defined as a categorical variable. Parental report of the number of serious infections per year was recorded in 3 classes (0, 1-2, and ≥3) and included in the model as a linear variable, with the assumption that the effect of frequent infections was 1.5 times as strong as the effect of infrequent infections. All remaining predictors were included as dichotomous variables. For both the categorical variables “wheezing” and “pregnancy duration,” 2 dichotomous dummy variables were included in the model. The symptom “dyspnea” was not asked for at 1 or 2 years because it was considered unreliable at this age. Therefore the variables “wheezing and/or dyspnea apart from colds” and “exercise-induced wheezing and/or dyspnea” were scored negative by default in children aged 1 or 2 years.

Table I. General characteristics of the study population of children who reported symptoms between the ages of 0 and 4 years and univariate relationship with asthma at 7 to 8 years of age

	All children∗ (n = 2,171), no. (%)	Asthma at 7-8 y (n = 240), no. (%)	OR (95% CI) for asthma	P value†
Age at onset of symptoms‡
0-1 y	1,157 (53)	143 (60)	1.7 (1.0-2.9)	.042
1-2 y	423 (19)	47 (20)	1.5 (0.9-2.7)	NS
2-3 y	366 (17)	33 (14)	1.2 (0.7-2.2)	NS
3-4 y (reference)	225 (10)	17 (7)	Reference	–
Reported symptom(s)
Cough at night§	1,314 (62)	137 (57)	0.8 (0.6-1.1)	NS
Wheezing
No wheezing (reference)	941 (43)	62 (26)	Reference	–
1-3 times/y	860 (40)	98 (41)	1.8 (1.3-2.5)	<.001
>3 times/y	370 (17)	80 (33)	3.9 (2.7-5.6)	<.001
Family history
Parental asthma	389 (18)	67 (28)	1.9 (1.4-2.6)	<.001
Parental inhaled medication	372 (17)	77 (32)	2.6 (1.9-3.5)	<.001
Parental hay fever	885 (41)	129 (54)	1.8 (1.4-2.4)	<.001
Sibling asthma	129 (6)	22 (9)	1.7 (1.1-2.8)	.025
Sibling hay fever	57 (3)	10 (4)	1.7 (0.9-3.5)	NS
Sibling eczema	402 (19)	51 (21)	1.2 (0.9-1.7)	NS
Perinatal factors
Male sex	1,196 (55)	162 (68)	1.8 (1.4-2.4)	<.001
Cesarean section	191 (9)	31 (13)	1.6 (1.1-2.5)	.018
Low birth weight (<2500 g)	94 (4)	15 (6)	1.6 (0.9-2.8)	NS
Breast-feeding ever	1,793 (83)	190 (79)	0.8 (0.6-1.1)	NS
Older sibling(s)	1,109 (51)	134 (56)	1.2 (0.9-1.6)	NS
Medium/low parental education parent(s)‖	1,659 (76)	199 (83)	1.6 (1.1-2.2)	.013
Smoking during pregnancy¶	325 (15)	41 (17)	0.8 (0.6-1.2)	NS
Delivery
Term (reference)	1,948 (90)	204 (85)	Reference	–
Preterm (<37 wk)	118 (5)	15 (6)	1.2 (0.7-2.2)	NS
Postterm (>42 wk)	105 (5)	21 (9)	2.1 (1.3-3.5)	.003
Environmental factors at age of report of symptoms
Day care	819 (38)	71 (30)	0.7 (0.5-0.9)	.006
Smoking in parental house#	632 (29)	67 (28)	0.9 (0.7-1.3)	NS
Pet ownership	1,221 (56)	128 (53)	0.9 (0.7-1.1)	NS
Avoidance of pets (due to allergy)	606 (28)	98 (41)	1.9 (1.5-2.5)	<.001
Child's other symptoms at age of report of symptoms
Wheezing/dyspnea apart from colds§∗∗	79 (4)	20 (8)	2.9 (1.7-4.9)	<.001
Wheezing/dyspnea, exercise induced∗∗	68 (3)	14 (6)	2.2 (1.2-3.9)	.013
Nasal symptoms§	908 (42)	126 (53)	1.6 (1.2-2.1)	<.001
Respiratory tract infections††
No infections (reference)	655 (30)	43 (18)	Reference	–
1-2 times/y	993 (46)	111 (46)	1.8 (1.2-2.6)	<.001
≥3 times/y	523 (24)	86 (36)	2.8 (1.9-4.1)	<.001
Eczema
Doctor's diagnosis ever	809 (37)	127 (53)	2.1 (1.6-2.7)	<.001
Eczematous rash present‡‡	465 (21)	89 (37)	2.4 (1.8-3.2)	<.001
Ever admitted for respiratory problem	65 (3)	17 (7)	3.0 (1.7-5.3)	<.001

All numbers and odds ratios refer to the imputed dataset (n = 2,171).

OR, Odds ratio.

Outcome measures 

The following 3 items of the questionnaires were used for the case definition of asthma: (1) at least 1 episode of wheezing; (2) inhaled steroids prescribed by a medical doctor; and (3) a doctor's diagnosis of asthma (a parental report of a doctor's diagnosis of asthma at any time and a parental report of asthma during the past 12 months). In the analyses children were only considered positive for asthma if they had 1 or more positive items at age 7 years and 1 or more positive items at age 8 years. Thereby we aimed to select only children with clinically relevant chronic asthma symptoms.

Data analyses 

Univariate associations between candidate predictor variables and the outcome of asthma were investigated with SAS software version 9.1 (SAS Institute, Inc, Cary, NC). In the eligible study population the overall proportion of missing data on candidate predictor variables was very low (<1%), and follow-up to the age of 8 years was 95%. In a standard complete case analysis, subjects with 1 or more missing values (on any variable) are excluded from analysis. Missing data were imputed by using the Multivariate Imputation by Chained Equations package in the statistical program R (version 2.5.1) to avoid any bias that might result from such a complete case analysis.¹¹, ¹² Multivariable analysis was performed with backward stepwise logistic regression to select combinations of predictor variables. Likelihood ratio statistics were used as a criterion for selection.

Candidate predictors were removed from the model if the corresponding P value was greater than .01. Interaction terms between different candidate predictors were tested if effect modification was considered likely based on previous literature. Interaction with age and “atopic mother” was tested for all predictor variables.

Calibration of the model was assessed both graphically and with the Hosmer-Lemeshow χ² statistic. The final model's ability to discriminate between asthmatic and nonasthmatic children was assessed by using the C-index (equivalent of area under the curve in the receiver operating characteristic curve). We considered a C-index of greater than 0.7 to be acceptable for a prognostic model.¹³ Internal validation of the final model was performed by using the bootstrap resampling technique.¹⁴, ¹⁵ For 200 bootstrap samples, a new model was selected according to the above procedure and tested on the original sample. This enabled us to estimate the bias (ie, optimism) caused by overfitting. Regression coefficients and the C-statistic of the final model were corrected for the calculated optimism. These regression coefficients were multiplied by a factor of 10 to create a simple risk score for use in clinical practice. Multivariate regression analyses and model validation were performed with R version 2.5.1 (Free Software Foundation, Inc, Boston, Mass).¹⁶

Back to Article Outline

Results 

Study population 

Of the 4,146 included mothers, 183 (4.5%) dropped out before returning the first postnatal questionnaire for various reasons (eg, stillbirth, language barrier, not interested, and moved). Of the 3,963 remaining children, 2,171 (55%) reported an episode of wheezing, coughing at night, or both between the ages of 0 and 4 years. Because only those children with early symptoms were eligible, the population for the present analysis consisted of 2,171 children. In more than half of the children, first symptoms were reported before the age of 1 year (Table I). The symptom of coughing at night (62%) was slightly more prevalent than wheezing (57%). At the age of 8 years, complete data on the outcome measure were available from 1,921 children (88% of the eligible study population). Complete data on all predictor variables were available from 1,854 (85%) children. The percentage of missing data per individual predictor variable did not exceed 2%. Children with at least 1 missing value on predictor variable, outcome variables, or both (n = 486) were less likely than children with complete data (n = 1,685) to have parents who both had a high level of education (19% vs 25%), to have been breast-fed (79% vs 84%), and to attend day care (33% vs 40%) and were more likely to have been exposed to cigarette smoke during pregnancy (19% vs 13%) and have had a low birth weight (8% vs 4%). This shows that complete case analyses would refer to a study population that is not representative of the original study population. To match the distribution of characteristics of the original dataset, these variables were taken into account in constructing the imputed dataset. Thus our results pertain to the imputed dataset, including all 2,171 children.

Outcome of asthma 

Two hundred forty (11%) children had symptoms, medication, or both at the ages of both 7 and 8 years and were thus defined as cases (Table II). The prevalence of asthma symptoms and medication use at the age of 7 years was similar to the prevalence at the age of 8 years. Only 10% of the children with asthma (n = 25) exclusively reported wheezing; the majority (90%) also used medication, had a doctor's diagnosis of asthma, or both. Asthma prevalence in the imputed dataset was 1.5% higher than in the complete dataset. Considering all children initially included in the birth cohort (n = 3,963), 85% of the children with asthma at 7 to 8 years of age reported cough, wheeze, or both at some point in the first 4 years of life. Apparently, children without these early symptoms accounted for only 15% of asthma diagnoses at 7 to 8 years of age.

Table II. Prevalence of features of asthma at age 7 and 8 years

All numbers refer to the imputed dataset (n = 2,171).

Univariate analysis 

The risk of asthma decreased with increasing age of report of first symptoms. A family history of asthma, hay fever, or both was clearly associated with an increased risk of asthma, especially parental use of inhaled medication. Male sex, cesarean section, low level of parental education, relatively long pregnancy duration, and avoidance of pets because of allergies in the family were all significantly associated with a higher prevalence of asthma in this population of children with early symptoms. Children attending day care at the age that symptoms were first reported had a lower prevalence of asthma at 7 to 8 years of age. Apart from the symptom “cough at night” (which was an inclusion criterion), all symptoms investigated were significantly positively associated with asthma. Results of all univariate analyses are shown in Table I.

Multivariate analysis 

In the multivariate analysis male sex, postterm delivery, parental education, parental inhaled medication, wheezing frequency, wheezing/dyspnea apart from colds, respiratory tract infections, and eczema remained as independent predictors of asthma (Table III). Age did not significantly contribute nor did interaction terms of predictors with age. Odds ratios for individual predictors are reported before and after internal validation. The discriminative power of the model, represented by the C-index, was 0.72 after validation. A clinical prediction score was developed by assigning points for each predictor variable based on its regression coefficient (Table III). A score for each child was calculated by using the equation shown in the legend of Table III. The score ranged from 0 to 55, with a median of 15.5. Fig 1 depicts the predicted risk of asthma at 7 to 8 years for every prediction score value. The risk of asthma increases with an increasing prediction score. In Fig 2 the predicted risk is compared with the observed risk in our population per score category with 5-point intervals. Because only a minority (2%) of children had a score of greater than 35 points, they were combined into a single category. Fig 2 shows that the model is well calibrated, which is confirmed by the Hosmer-Lemeshow test (Table III). Of the 561 children with a predictor score of less than 10, only 18 (3%) had asthma at the age of 7 to 8 years. In contrast, 50% (17/34) of the children with a score of at least 35 at first presentation of symptoms had developed asthma by the age of 7 to 8 years. Dichotomizing the prediction score at a certain cutoff value leads to a test with a dichotomous (positive/negative) result, which could be used to assist in clinical decisions in general practice.

Table III. Multivariate prediction model for asthma

Predictor variable		Before validation,∗ OR (95% CI)	After validation,∗ OR (95% CI)	Points in prediction score‖
1	Male sex	1.7 (1.3-2.3)	1.6 (1.2-2.1)	4.6
2	Postterm delivery	2.3 (1.3-4.0)	2.1 (1.2-3.6)	7.3
3	Medium/low parental education	1.6 (1.1-2.3)	1.5 (1.1-2.2)	4.2
4	Inhalation medication, parent(s)	2.4 (1.8-3.3)	2.2 (1.6-3.0)	7.7
5	Wheezing frequency†
	1-3 times/y	1.6 (1.1-2.3)	1.5 (1.1-2.1)	4.2
	≥4 times/y	2.8 (1.9-4.2)	2.5 (1.7-3.6)	9.1
6	Wheezing/dyspnea apart from colds‡	2.3 (1.3-4.1)	2.0 (1.1-3.7)	7.1
7	Serious infections†
	1-2 times/y	1.7 (1.3-2.2)	1.6 (1.2-2.1)	4.6
	≥3 times/y	2.2 (1.4-3.3)	2.0 (1.3-3.0)	6.9
8	Doctor's diagnosis of eczema and eczematous rash present	2.6 (1.9-3.5)	2.3 (1.7-3.1)	8.2 +
C-index§		0.743	0.717	55.1
Hosmer-Lemeshow (P value)		.822	.538

OR, Odds ratio.

• 
  • View Large Image
  • Download to PowerPoint

• Fig 1. 

  Predicted risk of asthma development at 7 to 8 years of age by prediction score.

• 
  • View Large Image
  • Download to PowerPoint

• Fig 2. 

  Predicted and observed risk of asthma development at 7 to 8 years of age per prediction score category. The number of children observed per category is shown in brackets. Expected risk is estimated by using the average score per 5-point category. ∗Estimate for the merged category (≥35 points) is weighed by the number of children actually observed at each 5-point subcategory between 35 and 55 points.

Sensitivity and specificity corresponding to ascending cutoff values are displayed in a receiver operating characteristic curve (Fig 3). Table IV further reports the predictive values and percentages of children with positive scores at the different cutoff values. For comparison with current practice, the predictive value of a doctor's diagnosis of asthma in the year when symptoms were first reported is given in Fig 3 and Table IV. Thereby it should be noted that for a “doctor's diagnosis of asthma,” all information that is available in clinical practice can be considered (eg, allergy tests, lung function tests, medication trial, and referral to secondary/tertiary medical centers). Our prediction rule merely uses simple items from the medical history and, with an appropriate cutoff, performs better.

• 
  • View Large Image
  • Download to PowerPoint

• Fig 3. 

  Receiver operating characteristic curve of categorized prediction score on the outcome of asthma at 7 to 8 years of age. Cutoff values of prediction scores are reported in the curve (dots). The area under the receiver operating characteristic curve (C-index) for the categorized score is 0.736 (before validation). Also, the sensitivity and specificity for a doctor's diagnosis of asthma at the age when symptoms were first reported (between 0 and 4 years) are displayed (square).

Table IV. Test characteristics at various cutoff points of prediction score for asthma at 7 to 8 years of age

The proportion of children with asthma at 7 to 8 years of age accounted for by children at highest risk according to the prediction rule is plotted in Fig 4. It can be seen from this figure that half of the population with the highest prediction score account for 80% of all asthma cases at 7 to 8 years of age. By selecting a cutoff of 20 points, 28% of the population will have a positive score (high risk), and this group accounted for 60% of all asthma cases.

• 
  • View Large Image
  • Download to PowerPoint

• Fig 4. 

  Proportion of children with asthma at 7 to 8 years of age explained by the proportion of children at highest risk according to the prediction score. Cutoff values of prediction scores are reported in the curve (dots). Also, data for a doctor's diagnosis of asthma at the age when symptoms were first reported (between 0-4 years) are displayed (square).

Back to Article Outline

Discussion 

In this study we developed a prediction rule for the risk of asthma at 8 years of age to be used when preschool children present with symptoms suggestive of asthma for the first time. From a large prospective database, we identified 8 easily obtainable clinical parameters that best characterized the risk of asthma at 7 to 8 years of age.

Algorithms to predict the development of asthma in children have been reported previously. However, differences in design and analysis should be considered when comparing the results. Although some authors developed a model for use in the general population,³, ¹⁷ others restricted their analysis to a selection of children with early symptoms, such as wheeze or cough.⁴, ⁵, ⁶, ⁷ We restricted our analysis to children who reported “wheezing,” “coughing at night without a cold or chest infection,” or both. We based our selection on these symptoms because they are prevalent in young children and suggestive of childhood asthma.¹⁸, ¹⁹

Another important consideration is that most studies developed a prediction rule using all available information at a fixed age.³, ⁶, ⁷, ¹⁷ Consequently, these prediction rules are valid only for this fixed age rather than at the age of symptom onset. In a recent study by Frank et al,⁵ the investigators followed children of different ages with a parental report of wheeze. In 201 children aged 0 to 4 years, they found only “wheeze after exercise” and “history of eczema or hay fever” to be predictive for wheezing after at least 6 years of follow-up. Interestingly, wheezing frequency or severity was not associated with symptom persistence. In their study data were not collected longitudinally from birth. Therefore the analysis could not be focused on the age of first symptoms, when prediction of asthma becomes relevant. Also, the fact that 51% of their original population was lost to follow-up might have been a source of bias.⁵

How do our findings compare with those from previous studies? A risk index developed in the Tucson Children's Respiratory Study also included wheezing frequency, eczema, and parental asthma as major criteria.³ This study reported wheezing apart from colds as a minor risk factor. The German Multicentre Allergy Study birth cohort reported male sex and parental atopy as independent predictors for symptom persistence in children with wheeze before the age of 3 years,⁷ and an algorithm to predict asthma from the Isle of Wight birth cohort included parental history of asthma and frequency of respiratory infections.⁶ All these earlier studies also included invasive allergy tests in their algorithms. This was also done in a Dutch case-control study by Eysink et al.⁴

For reasons of practical applicability, we included only clinical information that is easily obtainable in a primary care setting. In contrast to earlier studies, we found postterm delivery as a predictor of asthma development. The fact that we found this association does not necessarily imply a causal relationship. It might well be explained by correlated risk factors, such as meconium aspiration, cesarean section, birth order, and/or high birth weight.²⁰, ²¹, ²², ²³ High socioeconomic status was consistently linked with a lower prevalence of childhood asthma,²⁴, ²⁵ but this is the first study to report it as an independent predictor in a prognostic model.

The predictive power of our rule is not easily comparable with that of other studies because of differences in study design and objectives. The test characteristics of the stringent risk index by Castro-Rodriguez et al³ to predict asthma at 8 years are similar to our prediction rule at a cutoff of 30 points (see Table E1 in this article's Online Repository at www.jacionline.org). Considering that our rule predicted asthma in children at a much younger age (mean, 1.8 vs 3 years), did not require any laboratory tests, and was corrected for overoptimism by means of bootstrapping, we propose that our prediction rule can have an added value over the currently available risk scores.

Second, we compared the index by Castro-Rodriguez et al³ with our prediction score by applying both algorithms to our own dataset using only variables from clinical history (eosinophils not included) at the age symptoms were first reported (see the Results section and Table E2 in this article's Online Repository at www.jacionline.org for further details.) We found that our prediction rule had better predictive power in symptomatic preschool children at the age of first symptoms.

A strong point of our study is its size. The PIAMA birth cohort is the largest birth cohort used to predict asthma in children. Further strengths are good adherence and little missing data. In combination with the multivariate imputation, this limits the risk of selection bias.¹², ²⁶, ²⁷ Unlike most previous studies, we performed internal validation by using bootstrapping, which leads to more realistic estimates of regression coefficients and discriminative power of the model.¹⁵ By closely following clinical practice in the selection of our eligible population and predictor variables, we have developed a clinical tool that can be used at the moment it is of most clinical relevance. Finally, because the original study population of the PIAMA birth cohort is a reflection of the general population, our results might be valid for The Netherlands and, perhaps, other Western countries.

There are some limitations that should be considered when interpreting our results. First, information on dyspnea in children younger than 3 years is lacking in our database and difficult to retrieve reliably from parental reports. We cannot exclude that such information would contribute to the prediction. Consequently, we might have underestimated the predictive value of this variable.

Second, in the absence of a universal gold standard, the definition for asthma remains arbitrary. We decided to consider asthma as reported asthma symptoms, medication, or both at both ages 7 and 8 years because we aimed to predict chronic disease with clinical relevance. In a sensitivity analysis in which asthma was defined as a doctor's diagnosis plus symptoms at 7 to 8 years of age, our prediction rule did not lose any of its predictive power (C-index = 0.752).

We developed our prediction rule in a population selected on the basis of symptoms reported by questionnaire. Will these children indeed see a general practitioner? We propose that most parents will visit a general practitioner when their children have wheezing, nightly cough, or both, especially when their children are young (1-2 years of age). This is supported by our data: 1,577 (73%) children received a doctor's diagnosis of a respiratory illness in the year of first symptoms. Clearly, these children consulted a medical doctor with respiratory symptoms, and the prediction rule could have been used. A validation of our prediction rule in this subgroup showed that it had a similar predictive value (C-index = 0.726) and that all included questions remained significant predictors of later asthma at 7 to 8 years of age (P = .05, see the Results section and Table E3 in this article's Online Repository at www.jacionline.org for details). This strongly supports that our prediction rule is relevant and valid for children presenting in general practice. Before our prediction rule can be implemented, an external validation on a separate dataset should be performed.

Wheezing and coughing are very common in children.¹⁹ More than half of the children in this birth cohort reported one of these symptoms before the age of 5 years, the majority already in the first year of life. Treatment with inhaled medication could be effective for some but is costly and can be troublesome to administer to young children, and there is little evidence base for the use of inhaled corticosteroids in virally induced intermittent wheeze.² It remains a challenge for the clinician in general practice to differentiate children with transient symptoms from children who will have chronic asthma, especially if only a few risk factors are present. This is illustrated by the fact that in our birth cohort only 10% of children who received asthma medication in the first year of life were still using this medication 3 years later.²⁸ A trial with asthma medication could be helpful to confirm the diagnosis,²⁹ but spontaneous improvement while taking steroids can easily be misinterpreted as a favorable treatment effect.

Despite these difficulties, clinicians will need to make treatment decisions. Our prediction score might help to improve the accuracy of the prognosis and reduce undertreatment or overtreatment in individual children simply by combining readily available clinical information. The positive predictive value of our prediction score was twice as high as that of reported doctors' diagnoses at the same age, with similar negative predictive values. Apparently, the combination of 8 simple clinical questions had a better predictive value than the complete workup that is current practice for preschool children.

We therefore propose that our prediction score could serve as a practical and useful tool for clinicians who deal with preschool children presenting with symptoms suggestive of asthma, especially in a primary care setting. At a cutoff of 15 points, half of the children will have a negative test result. Because 96% of these children will not have asthma, this might well justify a conservative approach with reassurance of the parents. Increasing cutoff points could be used to imply different clinical actions, such as a specific IgE test, start of a trial with inhaled medication, or even referral to secondary care. Optimal cutoff points might vary between different settings, and an extensive analysis falls outside the scope of this article. Nonetheless, a reliable risk assessment in individual children is an essential step toward individualized asthma therapy.

In conclusion, we have developed an asthma prediction score that uses 8 easily obtainable clinical parameters and can be used in preschool children who present with asthma symptoms to predict asthma at the age of 7 to 8 years. Children with a low score (<10) at onset of symptoms had a risk of only a few percent of having asthma at 8 years, whereas children scoring 30 or higher had a risk of greater than 40%. The score could potentially be helpful for prognosis and treatment decisions in these young children.

Back to Article Outline

We thank Dr Hans van der Wouden for carefully reading the manuscript and for his valuable comments.

Back to Article Outline

Methods 

All children participated in the Prevention and Incidence of Asthma and Mite Allergy (PIAMA) birth cohort study. This multicenter study was conducted in 3 different regions of The Netherlands: north (Groningen and surroundings), central (Bilthoven, Wageningen and surroundings), and southwest (Rotterdam and surroundings). Recruitment took place between March 1996 and May 1997 by means of a validated screening questionnaire^E1 distributed by midwives to 10,232 pregnant woman visiting one of 52 prenatal clinics. According to the results of this screening, the women were divided into an allergic and a nonallergic group. Women with any of the following self-reported symptoms were defined as allergic: asthma, hay fever, house dust allergy, house dust mite allergy, or pet allergy. Children of allergic women were defined as high risk. Based on the screening, 7,862 women (2,779 allergic and 5,083 nonallergic women) were invited to participate in the study; approximately 50% (n = 4,146) agreed and provided informed consent (1,327 allergic and 2,819 nonallergic women). The PIAMA birth cohort includes an intervention part, which studied the effect of impermeable mattress covers, and a natural history part. fig E1 shows the flow diagram for recruitment of the intervention and natural history arms. Only high-risk children could participate in the intervention study. As previously published, the intervention had no effect on the incidence of allergy or respiratory symptoms.^E2 Furthermore, the intervention was no significant predictor of asthma in our study and had no significant interaction with any of the candidate predictor variables tested. Therefore we considered the complete PIAMA study population (both the intervention and the natural history part) in the present study. After birth, this study population consisted of 3,963 children. Importantly, the proportion of children with an allergic mother (31%) in the PIAMA birth cohort is very similar to the proportion in the general Dutch population. For the current study, only children with a positive response to the questions in the annual questionnaires at age 1 to 4 years (ie, “Has your child had wheezing or whistling in the chest in the last 12 months?,” “Has your child had cough during the night, when he/she did not have a cold or a chest infection, in the last 12 months?,” or both) were eligible. Fig E2 presents how the final study population was arrived at in the form of a flowchart.

Back to Article Outline

Results 

Comparison with previous literature 

Because of differences in the study design and objective, the predictive power of our rule is not easily comparable with that of other studies. Castro-Rodriguez et al^E3 developed a risk index for the prediction of asthma in preschool children. An important difference between their study and ours is that they predicted asthma in children at the age of 3 years regardless of the age of onset of symptoms. Second, their algorithm included blood eosinophils. We chose not to include any tests that were invasive, time-consuming, or both because our aim was to develop a prediction rule with which a doctor can instantly calculate the risk of later asthma.

Despite these differences, it is worthwhile to compare the yield of their algorithm with our risk score. Table E1 shows the test characteristics of the loose and stringent risk index, as reported by Castro-Rodriguez et al,^E3 to predict asthma at age 8 years. The same information is given for our prediction rule at a cutoff of 21 and 30 points (which leads to comparable groups of children with a positive test result). It can be seen that sensitivity and specificity, as well as predictive values, are similar at both cutoff values. In summary, we found similar test characteristics of the 2 algorithms despite the following differences: (1) the majority of children in our study were 1 year old at the time of prediction (compared with 3 years old in the article by Castro-Rodriguez et al^E3); (2) our prediction rule did not include any invasive tests (eg, measurement of blood eosinophils); (3) unlike the study by Castro-Rodriguez et al, our study adjusted for overoptimism by using an internal bootstrap validation; and (4) our study used a more strict definition of asthma. Therefore we propose that our prediction rule can have an added value over this currently available risk score.

Second, we aimed to assess whether the new clinical predictor variables that were found are of added value over those previously reported. We did this by comparing the variables included in the article by Castro-Rodriguez et al^E3 with those in our prediction rule using our own dataset. Apart from eosinophil counts, we annually collected all the information included in the risk index by Castro-Rodriguez et al: infrequent wheezing, frequent wheezing, doctor's diagnosis of parental asthma, doctor's diagnosis of eczema, allergic rhinitis, and wheezing in the absence of a respiratory infection. Using information up to the year of first symptoms and following their algorithm (without considering the factor eosinophilia), we calculated the risk of later asthma for every child within our own dataset. This resulted in 3 categories: (1) low risk of asthma (“negative loose index”); (2) intermediate risk of asthma (“positive loose index but negative strict index”); and (3) high risk of asthma (“positive strict index”). We then compared the risk score according to Castro-Rodriguez et al with our own internally validated PIAMA birth cohort risk score. For this purpose, we also categorized our continuous PIAMA birth cohort risk score in 3 identical categories (at similar cutoff points as the Castro-Rodriguez et al risk score). The C-index of the Castro-Rodriguez et al risk score was 0.628 versus the C-index of 0.696 for the PIAMA birth cohort risk score. This implies that although using easily obtained variables from clinical history, the PIAMA birth cohort score could better discriminate between asthmatic and nonasthmatic children. Importantly, by restricting the PIAMA birth cohort score to only 3 categories, predictive information was lost. In clinical practice a continuous score could provide a more accurate individual risk assessment. Indeed, with the PIAMA birth cohort score as a continuous variable, the C-index increases to 0.740.

Detailed results of this analysis, including sensitivity, specificity, and predictive values of the 2 risk scores, are shown in table E2. This table illustrates that all indices are higher in our prediction score than in the risk index by Castro-Rodriguez et al. Apparently, the use of the additional variables (sex, pregnancy duration, infection frequency, parental education, and parental inhaled medication rather than parental asthma) is of added value in the prediction of later asthma.

External validity of the prediction rule 

We developed our prediction rule in a population selected on the basis of symptoms reported by questionnaire. Will these children indeed see a general practitioner? We assume that most parents will present to a general practitioner when their child has respiratory symptoms, such as wheezing or coughing at night without a cold, especially if the child is very young. (The majority of children were less than 1 year old when they had first symptoms.) This is an important assumption because a general practitioner will only be able to use a prediction rule if children seek medical attention. To check the assumption and thereby the external validity of our results, we performed a validation of our prediction rule in a subgroup of children who reported a doctor's diagnosis of a respiratory illness in the same year as they had symptoms. Of these children, we can be certain that they presented to a medical doctor with respiratory symptoms. The following doctor-diagnosed respiratory illnesses were included: bronchitis, pneumonia, pertussis, flu, throat infection, croup, pseudocroup, otitis media, and asthma. Of the 2,171 children, 1,577 (73%) reported at least 1 of these diagnoses in the year of first symptoms. The percentage of children with asthma at 7 to 8 years of age in this subgroup (11.7%) did not differ significantly from the percentage in the total population (11.0%).

Table E3 shows the results of the developed model in the total population and in the subgroup with a doctor's diagnosis of a respiratory illness. It can be seen from the table that the regression coefficients of all variables remained more or less stable. The level of significance of some factors decreased. This can be partly explained by the weaker associations with asthma at 7 to 8 years of age but also partly by the decrease in sample size (n = 1,577 vs n = 2,171). Most factors remained strong and very significant independent predictors of asthma at 7 to 8 years of age. The overall predictive power of the model decreases slightly from 0.743 to 0.726.

In conclusion, we find that all variables in our original prediction rule were also significant predictors in this subpopulation of children with a doctor's diagnosis of a respiratory illness. This strongly supports our assumption that the prediction rule that was developed in this study has good validity in children who present to a medical doctor in the year of first symptoms.

Back to Article Outline

Fig E1. 

• 
  • View Large Image
  • Download to PowerPoint

• The PIAMA birth cohort recruitment scheme.

Back to Article Outline

Fig E2. 

• 
  • View Large Image
  • Download to PowerPoint

• Flowchart of the study population. ∗Two thousand seven hundred seventy-nine allergic and 5,083 nonallergic mothers were invited, as determined before initiation of the study based on a power calculation. Of the 4,146 women included in the study, the proportion of allergic women (31%) was very similar to that in the general Dutch population. †Symptoms were defined as a positive response to the following questions: “Has your child had wheezing or whistling in the chest in the last 12 months?,” “Has your child had cough during the night, when he/she did not have a cold or a chest infection, in the last 12 months?,” or both. Reasons for loss to follow-up included lack of motivation, illness of child, repeated nonresponse, moved, and personal reasons.

Back to Article Outline

Table E1. 

Comparison of predictive yield of PIAMA birth cohort risk with predictive yield as reported by Castro-Rodriguez et al^E3

Back to Article Outline

Table E2. 

Comparison of risk index by Castro-Rodriguez et al^E3 with the PIAMA birth cohort risk score categorized to 3 equal categories within the PIAMA birth cohort dataset (using only variables from clinical history)

PPV, Positive predictive value; NPV, negative predictive value.

Back to Article Outline

Table E3. 

Validation of multivariate model in subgroup of children with a doctor's diagnosis of respiratory illness in the year of first symptoms (n = 1,577)

OR, Odds ratio.

Back to Article Outline

References 

1. Martinez FD. What have we learned from the Tucson Children's Respiratory Study?. Paediatr Respir Rev. 2002;3:193–197
   • View In Article
   • Abstract
   • Full Text
   • Full-Text PDF (93 KB)
2. McKean M, Ducharme F. Inhaled steroids for episodic viral wheeze of childhood. Cochrane Database Syst Rev. 2000;(2):CD001107
   • View In Article
3. Castro-Rodriguez JA, Holberg CJ, Wright AL, Martinez FD. A clinical index to define risk of asthma in young children with recurrent wheezing. Am J Respir Crit Care Med. 2000;162:1403–1406
   • View In Article
   • MEDLINE
4. Eysink PE, ter Riet G, Aalberse RC, van Aalderen WM, Roos CM, van der Zee JS, et al. Accuracy of specific IgE in the prediction of asthma: development of a scoring formula for general practice. Br J Gen Pract. 2005;55:125–131
   • View In Article
   • MEDLINE
5. Frank PI, Morris JA, Hazell ML, Linehan MF, Frank TL. Long term prognosis in preschool children with wheeze: longitudinal postal questionnaire study 1993-2004. BMJ. 2008;336:1423–1426
   • View In Article
   • CrossRef
6. Kurukulaaratchy RJ, Matthews S, Holgate ST, Arshad SH. Predicting persistent disease among children who wheeze during early life. Eur Respir J. 2003;22:767–771
   • View In Article
   • MEDLINE
   • CrossRef
7. Matricardi PM, Illi S, Gruber C, Keil T, Nickel R, Wahn U, et al. Wheezing in childhood: incidence, longitudinal patterns and factors predicting persistence. Eur Respir J. 2008;32:585–592
   • View In Article
   • CrossRef
8. Wever-Hess J, Kouwenberg JM, Duiverman EJ, Hermans J, Wever AM. Prognostic characteristics of asthma diagnosis in early childhood in clinical practice. Acta Paediatr. 1999;88:827–834
   • View In Article
   • MEDLINE
   • CrossRef
9. Asher MI, Keil U, Anderson HR, Beasley R, Crane J, Martinez F, et al. International Study of Asthma and Allergies in Childhood (ISAAC): rationale and methods. Eur Respir J. 1995;8:483–491
   • View In Article
   • MEDLINE
   • CrossRef
10. Brunekreef B, Smit J, de Jongste J, Neijens H, Gerritsen J, Postma D, et al. The prevention and incidence of asthma and mite allergy (PIAMA) birth cohort study: design and first results. Pediatr Allergy Immunol. 2002;13(suppl 15):55–60
    • View In Article
    • CrossRef
11. Van Buuren S, Oudshoorn CGM. MICE: Multivariate Imputation by Chained Equations. R package version 1.16. Available at: http://web.inter.nl.net/users/S.van.Buuren/mi/hmtl/mice.htm. Accessed March 10, 2008.
    • View In Article
12. van der Heijden GJ, Donders AR, Stijnen T, Moons KG. Imputation of missing values is superior to complete case analysis and the missing-indicator method in multivariable diagnostic research: a clinical example. J Clin Epidemiol. 2006;59:1102–1109
    • View In Article
    • Abstract
    • Full Text
    • Full-Text PDF (126 KB)
    • CrossRef
13. Murphy-Filkins R, Teres D, Lemeshow S, Hosmer DW. Effect of changing patient mix on the performance of an intensive care unit severity-of-illness model: how to distinguish a general from a specialty intensive care unit. Crit Care Med. 1996;24:1968–1973
    • View In Article
    • MEDLINE
    • CrossRef
14. Harrell FE, Lee KL, Mark DB. Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med. 1996;15:361–387
    • View In Article
    • MEDLINE
    • CrossRef
15. Steyerberg EW, Bleeker SE, Moll HA, Grobbee DE, Moons KG. Internal and external validation of predictive models: a simulation study of bias and precision in small samples. J Clin Epidemiol. 2003;56:441–447
    • View In Article
    • Abstract
    • Full Text
    • Full-Text PDF (124 KB)
    • CrossRef
16. R Development Core Team (2007). R: a language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. Available at: http://www.R-project.org. Accessed March 23, 2007.
    • View In Article
17. Devulapalli CS, Carlsen KC, Haland G, Munthe-Kaas MC, Pettersen M, Mowinckel P, et al. Severity of obstructive airways disease by age 2 years predicts asthma at 10 years of age. Thorax. 2008;63:8–13
    • View In Article
    • CrossRef
18. Levy ML, Fletcher M, Price DB, Hausen T, Halbert RJ, Yawn BP. International Primary Care Respiratory Group (IPCRG) Guidelines: diagnosis of respiratory diseases in primary care. Prim Care Respir J. 2006;15:20–34
    • View In Article
    • MEDLINE
    • CrossRef
19. Kuehni CE, Davis A, Brooke AM, Silverman M. Are all wheezing disorders in very young (preschool) children increasing in prevalence?. Lancet. 2001;357:1821–1825
    • View In Article
    • Abstract
    • Full Text
    • Full-Text PDF (85 KB)
    • CrossRef
20. Macfarlane PI, Heaf DP. Pulmonary function in children after neonatal meconium aspiration syndrome. Arch Dis Child. 1988;63:368–372
    • View In Article
    • CrossRef
21. Tollanes MC, Moster D, Daltveit AK, Irgens LM. Cesarean section and risk of severe childhood asthma: a population-based cohort study. J Pediatr. 2008;153:112–116
    • View In Article
    • Abstract
    • Full Text
    • Full-Text PDF (156 KB)
    • CrossRef
22. Bernsen RM, de Jongste JC, van der Wouden JC. Birth order and sibship size as independent risk factors for asthma, allergy, and eczema. Pediatr Allergy Immunol. 2003;14:464–469
    • View In Article
    • MEDLINE
    • CrossRef
23. Remes ST, Patel SP, Hartikainen AL, Jarvelin MR, Pekkanen J. High birth weight, asthma and atopy at the age of 16 yr. Pediatr Allergy Immunol. 2008;19:541–543
    • View In Article
    • CrossRef
24. Cesaroni G, Farchi S, Davoli M, Forastiere F, Perucci CA. Individual and area-based indicators of socioeconomic status and childhood asthma. Eur Respir J. 2003;22:619–624
    • View In Article
    • MEDLINE
    • CrossRef
25. Duran-Tauleria E, Rona RJ. Geographical and socioeconomic variation in the prevalence of asthma symptoms in English and Scottish children. Thorax. 1999;54:476–481
    • View In Article
    • MEDLINE
    • CrossRef
26. Greenland S, Finkle WD. A critical look at methods for handling missing covariates in epidemiologic regression analyses. Am J Epidemiol. 1995;142:1255–1264
    • View In Article
    • MEDLINE
27. Moons KG, Donders RA, Stijnen T, Harrell FE. Using the outcome for imputation of missing predictor values was preferred. J Clin Epidemiol. 2006;59:1092–1101
    • View In Article
    • Abstract
    • Full Text
    • Full-Text PDF (339 KB)
    • CrossRef
28. Zuidgeest MG, Smit HA, Bracke M, Wijga AH, Brunekreef B, Hoekstra MO, et al. Persistence of asthma medication use in preschool children. Respir Med. 2008;102:1446–1451
    • View In Article
    • Abstract
    • Full Text
    • Full-Text PDF (226 KB)
    • CrossRef
29. British Thoracic Society Scottish Intercollegiate Guidelines Network. British Guideline on the Management of Asthma. Thorax. 2008;63(suppl 4):IV1–121
    • View In Article
    • CrossRef

Back to Article Outline

References 

1. Lakwijk N, Van Strien RT, Doekes G, Brunekreef B, Gerritsen J. Validation of a screening questionnaire for atopy with serum IgE tests in a population of pregnant Dutch women. Clin Exp Allergy. 1998;28:454–458
2. Brunekreef B, van Strien R, Pronk A, Oldenwening M, de Jongste JC, Wijga A, et al. La mano de DIOS…was the PIAMA intervention study intervened upon?. Allergy. 2005;60:1083–1086
3. Castro-Rodriguez JA, Holberg CJ, Wright AL, Martinez FD. A clinical index to define risk of asthma in young children with recurrent wheezing. Am J Respir Crit Care Med. 2000;162:1403–1406