Predictors of repeated acute hospital attendance for asthma in children: A systematic review and meta‐analysis

Abstract Background Asthma attacks are common and have significant physical, psychological, and financial consequences. Improving the assessment of a child's risk of subsequent asthma attacks could support front‐line clinicians’ decisions on augmenting chronic treatment or specialist referral. We aimed to identify predictors for emergency department (ED) or hospital readmission for asthma from the published literature. Methods We searched MEDLINE, EMBASE, AMED, PsycINFO, and CINAHL with no language, location, or time restrictions. We retrieved observational studies and randomized controlled trials (RCT) assessing factors (personal and family history, and biomarkers) associated with the risk of ED re‐attendance or hospital readmission for acute childhood asthma. Results Three RCTs and 33 observational studies were included, 31 from Anglophone countries and none from Asia or Africa. There was an unclear or high risk of bias in 14 of the studies, including 2 of the RCTs. Previous history of emergency or hospital admissions for asthma, younger age, African‐American ethnicity, and low socioeconomic status increased risk of subsequent ED and hospital readmissions for acute asthma. Female sex and concomitant allergic diseases also predicted hospital readmission. Conclusion Despite the global importance of this issue, there are relatively few high quality studies or studies from outside North America. Factors other than symptoms are associated with the risk of emergency re‐attendance for acute asthma among children. Further research is required to better quantify the risk of future attacks and to assess the role of commonly used biomarkers.

asthma attacks, as this risk is not directly correlated with daily control.
Children with largely well-controlled symptoms are still at risk of developing severe asthma attacks. 5 Asthma attacks are common 6 and are associated with high healthcare costs 7 as well as missed school and workdays. They cause anxiety 8 and carry a risk of death and long-term effects such as loss of lung function. 9 These acute events are especially relevant for children among whom there is the greatest potential for loss of lung function.
Attacks often follow a viral respiratory tract infection 10,11 but secondary care attendance is generally preventable. Attacks may occur despite the use of inhaled corticosteroids, 5,12 though it is neither affordable nor safe to provide all children with more aggressive treatment, particularly in low-middle income countries. It is therefore important to be able to identify patients at greatest risk of further attacks and hospital admission to better prioritize limited resources and provide additional education and support or adjustments to treatment where most needed. Currently, it is not possible to predict which children among those treated for an acute asthma attack, are at a greater risk of suffering repeated attacks. Physicians treating asthmatic patients are identifying rather poorly who is at risk of asthma attacks. 13 The development of a tool to enable clinicians to identify such children could be useful to optimize treatment strategies and address modifiable risk factors. This is especially relevant when treating patients with discordant manifestations of asthma such as few daily symptoms but evidence from biomarkers of active eosinophilic inflammation in the airways and therefore a high risk of exacerbations, and vice versa, 14 and when healthcare resources are stretched. Individualizing therapy has the potential to reduce the patient's risk of adverse outcomes from their disease and from medications.
Several factors have been associated with a higher risk of attack among asthmatic children, 15 both aspects of clinical history such as past attacks and objective measures such as low FEV1. However, findings from large database or cohort studies do not necessarily reflect the population seen in the emergency department (ED) or hospital ward. They therefore may not be informative for the common clinical scenario of reviewing a child in the ED or ward and deciding who needs changes in treatment or specialist referral. We therefore set out to identify predictors (personal and family history, and biomarkers) for subsequent asthma attacks in children attending hospital with an acute episode from the published literature. The aim of the study was to collate information that could support targeted secondary prevention interventions.

| Data sources and search strategy
We conducted systematic searches of bibliographic databases as described in the Supporting Information (E- Table S1). All databases were searched from their inception to the present with no language of publication restriction. Searches were carried out by a Cochrane Information Specialist up to 9th January 2017. Duplicate references were removed using reference management software (EndNote X7).
The reference list of each selected publication was hand-searched for relevant studies.

| Study selection
Studies with the following criteria were included: (1) cohort and case-cohort observational design analyzing factors related to asthma clinical history, previous treatment, lung function, biomarkers, or readily measured environmental exposures; or controlled trials that involve a lifestyle or social (not educational or pharmaceutical) interventions; (2) asthmatic children aged between 5 and 15 years old (among the age range of the study) recruited from the ED or ward, treated for an acute asthma exacerbation, included as participants; (3) emergency re-attendance or hospital readmission due to an asthma attack listed as outcomes.
The list of abstract and titles was reviewed to exclude publications that were clearly not contributory on this basis and duplicate titles. Full text articles of selected papers were obtained via University library and inter-library loan and reviewed for eligibility, excluding those not fulfilling inclusion criteria.

| Data extraction
Data were extracted by three independent authors using a standard data extraction form. Possible disagreements were resolved by discussion. RevMan 5 and Endnote X7 software were used to assist in the collection and management of data from abstracts and papers.

| Quality and risk of bias assessment
Studies' accuracy and risk of bias was assessed using the criteria of the Cochrane Handbook for Systematic Reviews of Interventions by three independent authors. Observational studies' quality was also assessed using the Newcastle-Ottawa Quality Assessment Scale, 16 with a score of seven or more defined as of high methodological quality.

| Analysis
Data from comparable studies were combined in quantitative analyses.
We pooled data using a random effect model in RevMan5, creating pooled estimates of effects for Hazard and Odds Ratios separately.
Reports that presented the results using mean values, with no other data, were not included in the meta-analyses. We used the generic inverse variance as the analysis method for some of the pooled estimate of effects, as some studies only reported Odds or Hazard Ratios.
We represented an estimate of the degree of variation between study outcomes using the I 2 statistic. 17 Overall pooled estimates of effect are presented on the basis of being informative to some degree if I 2 was high, if either there were few studies (a situation where I 2 can be imprecise or biased), or if the studies found a consistent direction of effect which would imply that factor was worth consideration in future prospective studies.

| RESULTS
A total of 3259 records were identified and screened for eligibility with one additional paper obtained through reference list screening ( Figure 1). Forty-three papers fulfilled our inclusion criteria after full-text screening, accounting for 36 studies. Participants were recruited among inpatients admitted for acute asthma in most of the studies (27) ( Table 1).

| Outcome
The study outcome was ED re-attendance in 4 reports, ED or hospital readmission in 6, hospital readmission alone in 21 studies and a further 5 studies analyzed data for both outcomes separately.

| Predictors
The risk factors or predictors studied varied amongst studies, including: socio-demographic characteristics such as age, gender, sex, and socioeconomic status (SES, including household/neighborhood income, private vs public insurance and working rank); and asthma characteristics (severity, treatment, previous admissions).

| Risk of bias
The number of reports with low, unclear, or high risk of bias was 22, 7, and 7, respectively. Those with an unclear risk of bias lacked information on relevant aspects of the methods, mainly sample selection, or did not state clearly the number or factors studied in order to assess reporting bias. The details of the risk of bias assessments are shown in E-Table S2.

| Factors related to the person Age
The effect of age on the future risk of ED or hospital readmission was examined in 16 studies (E- Table S3). There was a marked variation in the age group classification and statistical methods used for the comparisons, precluding a meta-analysis of this factor. Studies were consistent in reporting that younger children had a higher risk of ED or hospital readmission.

Sex
Six reports analyzed sex as a risk factor for ED and 17 for hospital readmission, though some of them stratified its effect by age. There was a decreased odds of hospital readmission among boys compared to girls (OR 0.91, 95%CI: 0.86-0.97; N = 67706; I 2 = 52%) in the pooled analysis of data from 17 studies ( Figure 2C), but no difference in the pooled analysis of other two studies reporting hazard ratios ( Figure 2D). There was no difference in ED re-attendance by sex in the studies reporting either odds or hazard ratios (Figures 2A   and 2B).

Ethnicity
The effect of ethnicity on ED or hospital readmission was included in 18 reports, using different classifications. The most frequent was the comparison between black or African-American and white or other origins, and the risk of readmission for asthma. There was an increased rate of ED re-attendance among African-American compared to white    Figure 3A).
The results of seven studies analyzing the association between black ethnicity and the odds of hospital readmission for acute asthma are shown in Figure 3B. The pooled result is not shown given the marked heterogeneity in study results (I 2 = 95%). Four other studies reporting Hazard Ratios for hospital readmission showed no association between black ethnicity and hospital readmission rate ( Figure 3C), but again were apparently heterogeneous in their findings (I 2 = 81%). One paper studying hospital readmission was excluded from the meta-analysis as results were stratified by age and sex. 19

Socioeconomic status (SES)
Twenty-one reports examined SES as a predictor for ED or hospital readmission for asthma. The specific predictor used differed, with FIGURE 2 Forest plots for the association of sex with emergency department re-attendance and hospital readmission for acute asthma in children using a random effects model. (Figure 2A-2D showing separate estimations for odds and hazard ratios) public insurance (vs private or other) or low household income being the most frequent markers adopted for low SES.  Figure S1).

| Factors related to asthma characteristics Previous ED or hospital admission
Eight reports included data on the risk of ED or hospital readmission according to a history of previous hospital or ED admissions, either in the previous 12-24 months (the most common) or ever in life. The three reports studying ED re-attendance odds ratios are shown in Figure 5A. The pooled OR for two of these studies was 2.94 (95%CI:  (Figures 5C and 5D). One of the studies (Taylor 1999) was not included in the pooled analysis of the odds ratio for hospital readmissions, for the same reason as exposed above. Although there was apparent heterogeneity between studies in these analyses, the direction of effect was consistent and individual studies found similar effect sizes so indicative pooled effect sizes are shown.

Asthma severity and controller treatment
Ten studies assessed the association between asthma severity, control or controller treatment received with the risk of ED or hospital readmission (E- Table S4). The type of predictor and definition of severity varied greatly between studies, precluding a meta-analysis.
The findings related to severity as defined by treatment were inconsistent.

| Factors related to potential follow-up
Asthma follow-up Ten papers examined aspects of follow-up after the index ED or hospital admission (E -Table S5). Three studies explored the effect of the Children's Asthma Care (CAC) measures set implementation, which comprises providing reliever medication and systemic corticosteroids for children admitted to hospital for asthma, and discharging them with a home management plan. They showed no effect on ED re-attendance Five studies analyzed the effect of different follow-up visit characteristics and ED or hospital readmission for asthma with disparate outcomes. Three reports studied the effect of receiving an asthma action plan at discharge on the risk of ED or hospital readmission. Two of them were combined (E- Figure S2) showing no association.

| Other factors
Exposure to tobacco smoke (ETS) Five studies included data on ETS, producing a pooled odds ratio of 1.60 (95%CI: 0.94-2.72; N = 1041; I 2 : 49%) for hospital readmission for acute asthma for those exposed to tobacco smoke (E- Figure S3). One

| Strengths
This systematic review answers a relevant question with public health implications for future asthma management. It has been developed in accordance with best practice and using an extensive search with no time or language publication restrictions to ensure the inclusion of potentially suitable studies. Risk of bias was assessed and presented for each included study separately. We were also able to undertake a quantitative analysis of the most relevant predictors, increasing the relevance of our findings.

| Limitations
The data collected had a moderate to low strength of evidence, due to the quality of the included studies and the inconsistent findings in some of the factors reported. 23 America were asthma has emerged as an important public health issue. 25 Decisions on who should be the focus of treatment are crucial in such settings where resources are likely to be very limited. Another important factor is the inclusion of children younger than 2 years old in several studies, an age at which it is difficult to ascertain an asthma diagnosis. 26 The outcomes used in the studies are also a possible source of bias, as there is no consensus on when a child should be admitted to the ED or hospital for acute asthma. However, this is the current definition used by the ATS for an episode of severe asthma. 27

| Findings in relation to other studies
The use of different study designs and effect measures meant that for most factors under study there was not a simple way to summarize the effect of a given risk factor. However, the effect sizes reported were similar in the most relevant predictors identified, such as low socioeconomic status or history of ED or hospital admission for acute asthma during the previous year. In particular, we did not find a substantial difference for any given risk factor between effect sizes in studies considering OR or HR: That is to say there was no clear difference in effect size when considering whether an exacerbation would happen in the follow-up period, and frequency of exacerbations.
This is likely to be because most children did not have multiple exacerbations during follow-up. There are also statistical challenges of understanding factors that influence the time to event for a potentially recurrent event.
Children younger than 5 years old were at a higher risk of ED or hospital readmission for acute asthma when compared to different age groups in more than half of the studies that explored age as a predictor of future risk. Preschool children suffer a larger number of acute asthma attacks driven mostly by respiratory viruses. 28,29 It is also difficult to diagnose asthma in this age group, 26 potentially leading to inadequate management. Lintzenich et al 30 showed that children 1-6 years old hospitalized for asthma were less likely to receive ICS baseline treatment and asthma education than older children.
Lower SES was associated with a higher risk and a higher rate of ED or hospital readmission. 31 Similar associations had been previously described for other diseases. 32 It may reflect poorer long-term management due to inadequate access to primary and specialist care, and that caregivers may be less able to adequately manage a longterm condition thus relying more on ED attendance. Flores et al 33 showed that among ethnic minority children with asthma in urban settings, poorer children were less likely to have an asthma specialist than wealthier children.
Children of African-American origin living in Anglophone countries were at higher risk of re-attendance for asthma. Non-white ethnicity has also been described as a predictor of hospitalization or ED visits in adults with severe or difficult-to-treat asthma. 34 Beck et al 35 reported that up to 80% of the readmission disparity between African-American and white asthmatic children could be explained by other associated factors, such as access to care or disease management.
However, it is uncertain how applicable these findings are outside an urban United States setting.
Co-existing allergic diseases (allergic rhinitis/rhinoconjunctivitis and eczema) were also associated with a greater risk of hospital readmission for asthma. Previous work has shown that treatment for allergic rhinitis (nasal corticosteroids and antihistamines) is associated ARDURA-GARCIA ET AL.
with lower rates of unscheduled care use. 36 This may indicate that untreated comorbid allergic diseases are associated with higher risk of asthma hospital readmissions.
Asthmatic children with a history of a previous ED or hospital admission for acute asthma had 2-5.8 times more risk of ED re-attendance and 2.5-3 times more risk of a hospital readmission.
This was therefore the clearest factor related to asthma that was associated with future risk. Similarly, other studies have identified several variables related to previous healthcare utilization for acute asthma that are associated with future risk of severe asthma attacks. 37