Article Text
PDF
Abstract
Objectives To examine the association among acute bronchiolitis-related hospitalisation in children, meteorological variation and outdoor air pollution.
Methods We obtained the daily counts of acute bronchiolitis-related admission of children≤2 years old from all public hospitals, meteorological data and outdoor air pollutants’ concentrations between 1 January 2008 and 31 December 2017 in Hong Kong. We used quasi-Poisson generalised additive models together with distributed lag non-linear models to estimate the associations of interest adjusted for confounders.
Results A total of 29 688 admissions were included in the analysis. Increased adjusted relative risk (ARR) of acute bronchiolitis-related hospitalisation was associated with high temperature (ambient temperature and apparent temperature) and was marginally associated with high vapour pressure, a proxy for absolute humidity. High concentration of NO2 was associated with elevated risk of acute bronchiolitis admission; the risk of bronchiolitis hospitalisation increased statistically significantly with cumulative NO2 exposure over the range 66.2–119.6 µg/m3. For PM10, the significant effect observed at high concentrations appears to be immediate but not long lasting. For SO2, ARR increased as the concentration approached the 75th percentile and then decreased though the association was insignificant.
Conclusions Acute bronchiolitis-related hospitalisation among children was associated with temperature and exposure to NO2 and PM10 at different lag times, suggesting a need to adopt sustainable clean air policies, especially to target pollutants produced by motor vehicles, to protect young children’s health.
- asthma epidemiology
- infection control
- respiratory infection
- viral infection
Statistics from Altmetric.com
Key messages
What is the key question?
Are meteorological variation and outdoor air pollution associated with acute bronchiolitis-related hospitalisation in children?
What is the bottom line?
Acute bronchiolitis-related hospitalisation among children was associated with temperature and exposure to NO2 and PM10 at different lag times.
Why read on?
This is one of the very few studies that explored the association of interest in a subtropical setting.
Introduction
Acute bronchiolitis is typically caused by viral infection. Respiratory syncytial virus (RSV) is the most common pathogen in childhood respiratory tract infection, especially among those ≤2 years old.1–3 Clinical signs and symptoms of acute bronchiolitis include cough, tachypnoea, hyperinflation of lungs, chest retraction, inspiratory crackles and expiratory wheeze.2 3
Acute bronchiolitis causes significant morbidity and mortality among infants and young children worldwide.1 It is the most common cause of lower respiratory tract infection (LTRI)-related hospitalisation.1–3 In the USA, acute bronchiolitis is the leading cause of hospitalisation among infants ≤1 year old.4 Nearly all children would be infected at least once by the time they reached the age of 2 years, and an estimated 2%–3% of all children would be hospitalised with acute bronchiolitis during their first year of life.2 3 5 6 In temperate regions, the annual epidemics usually last from November all the way till April with a peak observed in January or February.5 6 Similar patterns could be seen in European countries where the average RSV season commences in the beginning of December, peaks in early February and ends in early April.7
In Hong Kong, acute bronchiolitis is one of the major causes of hospital admissions among infants under the age of one. Up to 5% of total paediatric discharge from all hospitals under the Hong Kong Hospital Authority are attributed to this condition.8 In a single hospital study, the estimated incidence of bronchiolitis hospitalisation was 21 per 1000 among children ≤2 years old with 86% of them having been infected with RSV.9 Different from the USA and Europe, Hong Kong is located within the subtropical region where RSV season remains ill-defined but usually overlaps with the rainy season (ie, from March to September).3 8
Global surveillance suggested that the seasonal periodicity of RSV infection is related to climatic factors.10 Meteorological conditions including ambient temperature, rainfall and humidity were reported to be associated with RSV activity,6 10–12 while altitude, wind speed, dew point and ultraviolet B radiance were reported to be associated with bronchiolitis.11 12 On the other hand, air pollutants such as primary traffic pollutants, ozone (O3), fine suspended particulates (PM2.5) and environmental nitrogen dioxide (NO2), as well as living close to major roadways have been reported to be associated with exacerbation of respiratory infections in young children.13 14
Hong Kong is a densely populated cosmopolitan city in southern China with hot and humid summers, and mild and dry winters. According to the 2016 Population By-census, Hong Kong had a population of 7.34 million.15 Motor vehicles are the main source of air pollutants in Hong Kong. According to the air quality guideline of the WHO, 8 hours daily mean of O3 should not exceed 100 μg/m3, while annual mean level of respirable suspended particulates (PM10) and NO2 should not exceed 20 μg/m3 and 40 μg/m3, respectively.16 Unfortunately, in Hong Kong, the highest 8 hours mean concentration recorded for O3 was 305 μg/m3 and the highest annual mean concentrations recorded for PM10 and NO2 were 46 μg/m3 and 97 μg/m3, respectively, in 2017, all of which exceeded the WHO’s guideline limits.17
Previous studies suggested that acute bronchiolitis-related hospitalisation in children is associated with meteorological variation but its association with outdoor air pollution has not been well explored.6 9–14 18 19Furthermore, only limited studies have been conducted in subtropical regions, most of which yielded inconsistent results.8 9 18 19 Therefore, in this study, we aim to evaluate the possible short-term association among acute bronchiolitis-related hospitalisation in children, meteorological variation and outdoor air pollution in Hong Kong.
Methods
Study design
This is a retrospective time-series study to assess the short-term impacts of meteorological variation and outdoor air pollution on acute bronchiolitis-related hospitalisation among young children. We obtained the daily counts of acute bronchiolitis-related admission of children≤2 years old from a total of 12 public hospitals with acute paediatric inpatient services, daily meteorological data and daily outdoor air pollutants’ concentrations between 1 January 2008 and 31 December 2017.
Hospital admission data
In Hong Kong, paediatric care services are provided by both the public and private sectors. Approximately, 71% of all hospitalisations in 2016 occurred in public hospitals.20 The retrospective data of admissions of children ≤2 years old to all public hospitals for acute bronchiolitis (International Classification of Diseases 9th version (ICD-9): 466.1 in principal and secondary diagnosis) from 1 January 2008 to 31 December 2017 were extracted for statistical modelling. Data on influenza-associated hospital admission (ICD-9: 487.0, 487.1, 487.8) were also collected.
Meteorological data
Daily meteorological records measured at a single central monitoring station run by the Hong Kong Observatory were obtained for the study period, including mean ambient temperature (oC), mean relative humidity (RH%) and total rainfall (mm), from the open-access data available on their website (https://www.hko.gov.hk/cis/climat_e.htm). Actual vapour pressure (hPa), a proxy for absolute humidity well adopted in many environmental studies21 22 was then derived using Teten’s formula23:
where e, RH and TEMP denote actual vapour pressure, relative humidity and ambient temperature, respectively. Apparent temperature (oC) was derived using the following formula24:
where TEMP and e are as defined above and WIND denotes wind speed (m/s).
Ambient air pollutants data
Daily average levels of air pollutants measured at 13 general air monitoring stations (Central/Western, Eastern, Kwai Chung, Kwun Tong, Sham Shui Po, Shatin, Tai Po, Tap Mun, Tseng Kwan O, Tsuen Wan, Tuen Mun, Tung Chung and Yuen Long), including nitrogen dioxide (NO2), sulphur dioxide (SO2), ozone (O3) and particulate matters (PM10), were retrieved from the official website of the Environmental Protection Department of Hong Kong (https://cd.epic.epd.gov.hk/EPICDI/air/station/?lang=en). Data from Tap Mun station were excluded because it is situated in a remote area with very low population density, thus, might not represent the exposure of the general population. Data from the remaining 12 general air monitoring stations were averaged for downstream analysis.
Statistical analyses
Descriptive statistics were used to describe the demographic characteristics of the patients. A quasi-Poisson generalised additive model was used in conjunction with the distributed lag non-linear models (DLNMs) to assess the potentially lagged non-linear short-term associations between acute bronchiolitis-related hospitalisation of young children, meteorological variation and outdoor air pollution. In the DLNM, cross-basis is a bidimensional functional space created by combining two preselected sets of basis functions, which specify the relationships in the dimensions of an independent variable and lags, respectively. Algebraically, the cross-basis function of an independent variable X at time t with a maximum lag of L can be expressed in the following form:
where 
is an 
vector representing the jth basis variable of X across the (L+1) lags, 
is an 
vector obtained by transforming the lag vector 
using the kth basis function in the lag dimension, 
and 
are the dimensions of the basis functions of 
and
l
respectively, and 
is the unknown regression coefficient.25
The form of the primary model is as follows:
where E(Yt) is the daily expected count of acute bronchiolitis admissions on day t; TEMP, RH, RAINFALL denote ambient temperature, relative humidity, and total rainfall, respectively; 
and 
denote day of study and day of week on day t respectively; α is the overall intercept; factor(.) is a set of indicator functions for any categorical independent variable, and s(.) denotes a smoothing spline function with maximum basis dimension k. In this study, natural cubic spline functions (ns) with 2 and 3 knots placed at equally spaced intervals were chosen as the basis function of the meteorological parameters (except for rainfall) and pollutants, respectively, which is equivalent to 3 and 4 degrees of freedom (df) respectively; whereas an ns function with 2 knots placed at logarithmically equal intervals, which is equivalent to 4 df, was selected as the basis function of lag. DOS was modelled using an ns function with 7 df per year. The incubation period for acute bronchiolitis is usually 5–7 days and disease period is usually 1–2 weeks5; hence, the maximum lags for independent variables were assumed to be 21 days. Total rainfall was categorised into three groups, namely 0 mm, 0 to 38.8 mm (the 95th percentile) and >38.8 mm, so as to better examine the effect of extreme rainfall and reduce the influence of outliers.26 SO2 and PM10 were log transformed to reduce skewness and the influence of outliers. An indicator variable HOLIDAY which denotes whether a specific day was a holiday, DOS, DOW and the square root of the number of influenza-associated hospital admission on the same day (FLU) were included in the model for confounder control. To further assess the adjusted effect of absolute humidity and apparent temperature on the daily counts of acute bronchiolitis admissions, two secondary models with vapour pressure in place of ambient temperature and relative humidity, and apparent temperature in place of ambient temperature were fitted due to high correlations between these meteorological variables (online supplemental appendix section A).
Supplemental material
All results were reported in terms of overall cumulative adjusted relative risks (ARR), which accumulated the ARRs over the lag period of 21 days, along with the corresponding 95% CI, and ARR plots by lag at selected percentiles, at which the adjusted risks at a specific lag and percentile were compared with that at the reference value on the concurrent day. The reference values for rainfall and all pollutants were 0 mm and their corresponding 5th percentile, respectively, whereas the medians were chosen as the reference values for ambient temperature, apparent temperature, relative humidity and vapour pressure.
Model diagnostics and sensitivity analysis
Adjusted R2 and residual plots were used to assess the goodness of fit of the final models. Partial autocorrelation function (PACF) plots were used to assess partial autocorrelation of residuals, with absolute values of pacfs <0.1 at the first 30 lag days regarded as adequate fit.27 Sensitivity analyses were conducted to assess the robustness of the final results to the initial model choices. To be specific, changes in the overall cumulative ARRs induced by the alternation of the dfs of the basis functions for meteorological parameters (2 and 4 except for rainfall), pollutants (3 and 5), and lag dimensions (3 and 5), as well as that of ns function for DOS (6 and 8 per year) were examined. The cut-off values for categorising total rainfall were also altered (online supplemental appendix section B).
All statistical analyses were performed using R V.3.6.0. (R Development Core Team, 2018: https://www.rproject.org/) and the raw outputs were listed in online supplemental appendix section E.
Results
A total of 31 523 admissions from 2008 to 2017 were retrieved from the database, out of which 1835 were elective admission, and thus, were excluded in the analysis. Among the remaining 29 688 admissions, 66% were infants ≤12 months, 4% were preterm babies (<37 weeks), 68% were male and 99% were Asian. The median (IQR) length of stay was 3 days (2–4 days). 21.5% of cases were admitted more than once with acute bronchiolitis. During the study period, a total number of 10 deaths were recorded, corresponding to a mortality rate of 0.04% (table 1). The major causes of these deaths were respiratory failure and multiple organ failure. Each hospital applies different practices based on different priorities. Some hospitals might not examine nasopharyngeal aspirates (NPA) for pathogens since the results would not affect their medical management. From the available NPA results, RSV was the most common pathogen causing acute bronchiolitis in infants.
Characteristics of the 29 688 patient episodes
During the study period, the median (IQR) number of daily acute bronchiolitis admission was 8 (5–10) (table 2). The median (IQR) mean daily ambient temperature and mean relative humidity were 24.8°C (19.2°C–28.2°C) and 79% (74%–86%), respectively. According to the results, acute bronchiolitis was shown to exhibit seasonal pattern, with a major peak in spring (February–April) which could attain >400 cases per month, and a minor peak in late summer (September–October) having around 250 cases per month (figure 1).
Seasonal variation in the weekly numbers of acute bronchiolitis-related admissions, meteorological parameters (monthly average ambient temperature, apparent temperature, relative humidity, vapour pressure and total rainfall), and air pollution level (monthly average nitrogen dioxide (NO2), ozone (O3), sulphur dioxide (SO2), and PM10) in Hong Kong during 2008 to 2017.
Summary statistics of daily acute bronchiolitis hospital admissions, meteorological conditions and air pollutant concentrations in Hong Kong during 2008 to 2017
Figure 2A–D show the overall cumulative ARRs and percentile-specific (5th, 25th, 75th and 95th percentiles) ARRs of acute bronchiolitis-related hospitalisation among children≤2 years old. The patterns of the overall cumulative ARR for ambient temperature and apparent temperature were similar, which showed an overall increasing trend (figure 2A,B). For ambient temperature, when the median (24.8°C) was taken as reference, the overall cumulative ARRs at the 5th, 25th, 75th and 95th percentiles were 0.990 (95% CI 0.721 to 1.361), 1.048 (95% CI 0.846 to 1.298), 1.034 (95% CI 0.904 to 1.182), and 1.108 (95% CI 0.871 to 1.409), respectively; whereas for apparent temperature, when the median (23.6°C) was taken as reference, resulted ARRs at the four percentiles were 1.005 (95% CI 0.761 to 1.326), 1.044 (95% CI 0.873 to 1.247), 1.112 (95% CI 0.962 to 1.286), and 1.323 (95% CI 1.011 to 1.731), respectively. It was found that, at the 75th and 95th percentiles of both ambient temperature and apparent temperature, the ARRs were statistically significant since lag 16–17 days.
Overall cumulative adjusted relative risk along lags and adjusted relative risk by lag at selected percentiles against different meteorological variables. The reference values for (A) ambient temperature, (B) apparent temperature, (C) relative humidity and (D) vapour pressure for comparison were their corresponding median.
For relative humidity (figure 2C), the overall cumulative ARR was comparatively higher when relative humidity was low compared with when it was high. When the median relative humidity (79.0%) was taken as reference, the overall cumulative ARRs at the 5th, 25th, 75th and 95th percentiles were 1.012 (95% CI 0.828 to 1.236), 0.987 (95% CI 0.901 to 1.080), 1.034 (95% CI 0.936 to 1.143) and 1.087 (95% CI 0.836 to 1.414), respectively. A delayed effect of high relative humidity (95th percentile) on the admission was evident: the estimated ARR was <1 from the concurrent day up to lag 4 days and was >1 afterwards.
The adjusted association between vapour pressure and the risk of acute bronchiolitis-related hospitalisation looked like a combination of the association of ambient temperature and that of relative humidity (figure 2D). Although the association appeared to be flattened U-shape, the risks were relatively higher at the higher end. Using median (24.7 hPa) as the reference value, the overall cumulative ARRs at the 5th, 25th, 75th and 95th percentiles were 1.056 (95% CI 0.851 to 1.311), 1.019 (95% CI 0.884 to 1.173), 1.070 (95% CI 0.903 to 1.267) and 1.125 (95% CI 0.864 to 1.466), respectively. While at the 5th percentile, the risk was relatively lower from the concurrent day to lag 3 days and was relatively higher up to lag 17 days. At moderately high (75th percentile) and very high (95th percentile) vapour pressures, the association was more complicated. The estimated ARR was >1 from lag 1 to 5 or 6 days as well as lag 15 days onwards, and was statistically significant since lag 18 days.
Total rainfall did not show a statistically significant relationship with the risk of hospitalisation adjusted for other variables in the model (figure 3). Nevertheless, there was apparently a harvesting effect at extreme rainfall (>38.8 mm), at which the risk was relatively higher during the first few lag days. We have tested the robustness of different cut-off values for categorising total rainfall and our results were not sensitive to the variation (online supplemental appendix section B).
Adjusted cumulative relative risk along lags and adjusted relative risk by lag at selected percentiles against total rainfall. The reference value for a comparison was set to zero.
The overall cumulative effects and percentile-specific (25th, 50th, 75th and 95th percentiles) effects of the air pollutants on acute bronchiolitis-related hospitalisation among children≤2 years old are shown in figure 4A–D. High concentrations of NO2 were associated with elevated risks of acute bronchiolitis admission. In particular, the overall cumulative risk of bronchiolitis hospitalisation increased significantly when NO2 was between 66.2 and 119.6 µg/m3. The overall cumulative ARRs at the 25th, 50th, 75th and 95th percentiles were 1.013 (95% CI 0.849 to 1.208), 1.038 (95% CI 0.831 to 1.297), 1.237 (95% CI 0.961 to 1.592) and 1.643 (95% CI 1.155 to 2.338), respectively. At high concentration of NO2, the estimated risk was statistically significantly higher after a lag of 2 weeks.
Overall cumulative adjusted relative risk along lags and adjusted relative risk by lag at selected percentiles against different pollutants, including (A) NO2, (B) PM10, (C) O3, and (D) SO2. The reference values for comparison were their corresponding 5th percentile.
At very high concentrations of PM10, the estimated overall cumulative adjusted risks tended to be higher than that at their corresponding 5th percentile despite statistical insignificance. For PM10, the overall cumulative ARRs at the 25th, 50th, 75th and 95th percentiles were 1.084 (95% CI 0.916 to 1.282), 0.962 (95% CI 0.730 to 1.267), 0.793 (95% CI 0.572 to 1.099) and 0.756 (95% CI 0.522 to 1.094), respectively, and its effect appears to be immediate but not long lasting: the risk on the concurrent day was significantly higher at all percentile levels than if the exposure level had been the fifth percentile. The ARRs on the concurrent day at the four percentiles were 1.046 (95% CI 1.009 to 1.083), 1.056 (95% CI 1.005 to 1.110), 1.063 (95% CI 1.001 to 1.128) and 1.072 (95% CI 1.000 to 1.150), respectively.
For O3, the overall cumulative ARRs at the 25th, 50th, 75th and 95th percentiles were 0.838 (95% CI 0.703 to 1.001), 0.934 (95% CI 0.751 to 1.162), 0.965 (95% CI 0.758 to 1.230) and 0.962 (95% CI 0.711 to 1.301), respectively, except for the 25th percentile the risk tended to be higher during lag 1–7 days and was lower afterwards.
For SO2, the estimated ARR was above 1 when log(SO2) was between 1.65 and 3.35, which was equivalent to the interval 5.21–28.5 µg/m3 in the original scale. The overall cumulative ARRs at the 25th, 50th, 75th and 95th percentiles were 1.083 (95% CI 0.916 to 1.279), 1.179 (95% CI 0.952 to 1.460), 1.234 (95% CI 0.955 to 1.596) and 1.065 (95% CI 0.740 to 1.534), respectively. The patterns of ARR by lag at the four selected percentiles were similar. The estimated ARR was greater than 1 across lag 1–15 days.
Concerning model diagnostics, the adjusted R2 of the primary model was 47.4%, and that of the two secondary models with apparent temperature and vapour pressure were 48.0% and 47.2%, respectively. The absolute values of PACFs of the residuals for the first 30 lags were all <0.1. In the sensitivity analyses, reducing the annual number of df for calendar time from 7 to 6 had the largest effect on the results (online supplemental appendix section C and D).
Discussion
To our knowledge, few studies have investigated the short-term effects of meteorological variables and outdoor air pollution on acute bronchiolitis admissions in subtropical settings. In this study, a total of 29 688 emergency paediatric acute bronchiolitis admissions from all Hong Kong public hospitals over a 10-year period were analysed and studied whether they were associated with meteorological variables and outdoor air pollutants. Our results showed that increased acute bronchiolitis-related hospitalisation was associated with high temperature (ie, ambient temperature and apparent temperature), and was marginally associated with absolute humidity. Although the result is generally in line with the study from Mäkinen et al 28, which showed an increasing trend in the risk of LTRIs when temperature and absolute humidity increased, we acknowledge that an effect of high temperature on bronchiolitis has not been well established in the literature. An investigation conducted in Malta suggested that majority of bronchiolitis-related hospital admissions occurred in winter in subtropical region,29 while another study conducted in China indicated that ambient temperature was negatively correlated with bronchiolitis-related hospitalisations among children.30 Yet, by now, there has not been much consensus within the field on ‘risky’ temperature range for acute bronchiolitis. This was similar with another study on pneumonia in children aged <5 years old where the temperature effect was inconsistent.31 Apart from that, a large proportion of admissions for LRTI (many of which were attributed to RSV) was shown to be independently associated with absolute humidity in children in northern Spain.32 Nevertheless, only limited literatures have documented the association between absolute humidity and acute bronchiolitis-related or RSV-related admissions, especially when compared with other respiratory infections such as influenza.21 22
Among the air pollutants, high level of NO2 was associated with increased risk of acute bronchiolitis admission and the overall effect became statistically significant over the range 66.2–119.6 µg/m3 when compared with value at the 5th percentile. NO2 is a common outdoor air pollutant primarily produced by motor vehicles and power plants as a byproduct of high-temperature combustion. At ambient pressure and temperature, NO2 exists in its gaseous state and is relatively insoluble in water. These properties allow deeper penetration of NO2 into the respiratory tree (ie, the lower respiratory tracts) where it would react with water from mucosal lining of the bronchioles and form corrosive chemicals including nitric and nitrous acids.33 The nitrous vapour could cause microlesions in the respiratory tract and increase susceptibility to infections, including those caused by RSV.33 According to a recent large birth cohort study conducted in Indonesia among children aged 0–3 years, an IQR increase in NO2 exposure would increase the risk for acute respiratory tract infection by an estimated 18%.34 With reference to the WHO guideline for air quality, the annual mean level of NO2 should not exceed 40 μg/m3.16 During our study period, the median (IQR) daily mean NO2 level was 49.4 µg/m3 (39.6 to 62.5 µg/m3), and in 1770 days (48.5%) it exceeded the WHO’s annual safe level.17 Considering the aforementioned, it is thus very important for Hong Kong to adopt sustainable clean air policies as soon as possible to protect young children’s health. Regarding the sustained effect of NO2 observed in our study, we tend to interpret the findings from toxicological perspectives. Seaton and Dennekamp35 proposed the epidemiological association between illness and NO2 might be confounded by particles number. Given the toxic effects of ultrafine particles has been evident at low concentrations in animals, we speculate the delayed effect of NO2 might actually be the cumulative effect of successive exposure to ultrafine particles.35 Follow-up studies should be conducted to test and verify this speculation.
A number of studies have been showing that exposure to PM would increase the risk of respiratory infections and other illnesses.13 36 37 In an investigation conducted in the USA, Karr et al 13 found both chronic and subchronic exposures to PM2.5 were associated with elevated risk of bronchiolitis hospitalisation in infants (ie, around a 9% increase in risk for every 10 μg/m3 increase in PM2.5). PM10 was found to be related to bronchiolitis consultation and hospitalisation in both Malaysia and France.36 37 We found a significant positive effect of PM10 on acute bronchiolitis admissions only within a short time window, suggesting that most of the effects of PM10 were acute. This might explain why we did not find cumulative effects of PM10 over 21 days. One possible reason might be the relatively low PM10 concentration in Hong Kong (ie, 43.2 μg/m3) as compared with the WHO guideline level of 50 μg/m3 for 24 hours mean. At low concentration, the adverse effect of PM10 on pulmonary health might not be clinically severe enough to cause hospitalisation, which might also explain why we could not find a cumulative effect of PM10 on bronchiolitis hospitalisations. Given that PM2.5 has the ability to go even deeper into the lungs than PM10 does, and furthermore to enter the blood system, the damaging effect of PM2.5 on respiratory health might be stronger than that of PM10 at low concentrations. It is thus speculated that replacing PM10 with PM2.5 might be a better way to explore the overall cumulative risk of exposure to particular matters on acute bronchiolitis hospitalisation. Further research is needed to verify this hypothesis.
Similarly, no positive association was observed between O3 and acute bronchiolitis admission. In a recent study, a 20-ppb increase in O3 was associated with a significant increase in emergency department visits for respiratory conditions (by 1.7%–5.1%).38 Children exposed to high ozone levels had significantly decreased lung function, which may have reduced their defence against viral infections (eg, RSV) and increased their susceptibility to bronchiolitis.39 Nevertheless, it was argued that O3 could reduce the risk of respiratory infection that might be attributed to the virucidal activity of O3 and its relationship with host defence.40
The present study has several potential limitations. First, acute bronchiolitis is a clinical diagnosis, and therefore subject to variation among public hospitals due to differences in the availability of virus testing, differences in the diagnostic criteria used by doctors to diagnose bronchiolitis, and differences in coding practices. It is thus possible that acute bronchiolitis might be coded in the medical record system as other respiratory conditions. Meanwhile, criteria adopted in determining the need for hospitalisation changes over time. That being so, we might have underestimated the frequency of acute bronchiolitis hospitalisation in our study. Nevertheless, since over 90% of admissions in our dataset were principally diagnosed with acute bronchiolitis, our results shall remain robust (table 1). Second, this study used surveillance data in public hospitals which does not include data from private hospitals. However, the public healthcare services accounted for over 90% inpatient services in Hong Kong. Third, exposure measurement error is a common concern in environmental epidemiology, especially when studying the short-term impacts of environmental variables, given that meteorological condition and air pollutant levels change across time and space. For example, a child might be exposed to more air pollutants if she/he has smoking parents and lives in a household with 24 hours air conditioning that located close to busy roads. The error in exposure measurement might lead to inaccurate estimation of relative risks which, unfortunately, can hardly be avoided. Fourth, the increased incidence rate observed in this study might be affected by residual confounding. Unmeasured factors, such as parental smoking status, household crowding, indoor metrological condition (eg, temperature, relative humidity and air movement) and indoor air pollution (eg, NO2, PM10, PM2.5, O3, formaldehyde, volatile organic compounds, radon and airborne bacteria level) could be potential confounders and/or effect modifiers that might distort the results of the study. Fifth, given that different pathogens prefer different environment for survival and transmission,21 the findings of our study may not be consistent with results from pathogen-specific (eg, RSV, influenza, etc) analyses of acute bronchiolitis. We acknowledge the lack of information about pathogen causing the admissions is one of the main limitations in this study. Also, as RSV infection is a major determinant for acute bronchiolitis hospitalisation in children, we took the incubation period into account in addition to the symptomatic period and assumed a maximum lag of 21 days for the delayed effects in our study. Our results may thus be less compatible with other similar investigations that generally studied the short-term effects of pollutants using a shorter maximum lag (eg, 3–7 days). Last but not least, interaction effects between the meteorological factors and pollutants were not considered in this study. Our preliminary analysis has shown that although the interaction effects were not strong in general, when meteorological factors were at medium level and pollutants were at medium or high level, their joint effects on the outcome were usually statistically significant (detailed results not given), hence worth further investigation in future studies.
In conclusion, our findings showed that high temperature (ambient temperature and apparent temperature) and exposure to NO2 and PM10 were associated with acute bronchiolitis-related hospitalisation among children at different lag times. The significant relationship with pollutants suggests the need to adopt sustainable clean air policies in Hong Kong, especially to target pollutants produced by motor vehicles, to protect young children’s health. Since acute bronchiolitis continues to be a public health burden on the already-stressed healthcare system, policy-makers are urged to develop more cost-effective approaches to manage acute bronchiolitis in children. We believe that our findings could rationalise an evidence-based allocation of resources and formulation of environmental policies to minimise the burden of disease associated with acute bronchiolitis in young children.
Acknowledgments
The authors thank Lee Tsz Cheung from Hong Kong Observatory providing his expertise advice on the interpretation of meteorological effects.
References
Supplementary materials
Supplementary Data
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Footnotes
Contributors SYL and KCC conceived the study. SYL, KCC and SYFL performed the analysis. KLK and SYL contributed to the acquired data of the study. SYL, PKSC, KCC and KM contributed to the results interpretation. SYL, KCC, SYFL, PKSC and KM drafted the paper. All authors have read and approved the final paper.
Funding This work was supported by the National Natural Science Foundation of China (71974165, 81473035) and partially supported by Health and Medical Research Fund (INF-CUHK-1, 19181132).
Competing interests None declared.
Patient consent for publication Not required.
Provenance and peer review Not commissioned; externally peer reviewed.
Data availability statement Data may be obtained from a third party and are not publicly available. The sharing of data is restricted by the Hong Kong Hospital Authority.