Background: Few studies have explored the relation between air pollution and general practitioner (GP) consultations in Asia. Clinic attendance data from a network of GPs were studied, and the relationship between daily GP consultations for upper respiratory tract infections (URTI) and non-URTI respiratory diseases and daily air pollutant concentrations measured in their respective districts was examined.
Methods: A time series study was performed in 2000–2002 using data on daily patient consultations in 13 GP clinics distributed over eight districts. A Poisson regression model was constructed using the generalised additive model approach for each GP clinic, and associations with daily numbers of first visits for URTI were sought for daily concentrations of the following air pollutants: SO2, NO2, O3, PM10, and PM2.5. A summary relative risk of first visits to the GP for URTI per unit increase in concentration for each air pollutant was derived using a random effect model. First visits for non-URTI respiratory diseases were analysed in three GP clinics.
Results: Significant associations were observed between first visits for URTI and an increase in the concentrations of NO2, O3, PM10, and PM2.5. The excess risk was highest for NO2 (3.0%), followed by O3 (2.5%), PM2.5 (2.1%), and PM10 (2.0%). Similar associations with these air pollutants were found for non-URTI respiratory diseases.
Conclusions: These results provide further evidence that air pollution contributes to GP visits for URTI and non-URTI respiratory diseases in the community.
- NO2, nitrogen dioxide
- O3, ozone
- PM10, PM2.5, particulates with an aerodynamic diameter less than 10 μm or 2.5 μm
- RR, relative risk
- SO2, sulphur dioxide
- TSP, total suspended particles
- URTI, upper respiratory tract infections
- air pollution
- GP consultation
- respiratory diseases
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
- NO2, nitrogen dioxide
- O3, ozone
- PM10, PM2.5, particulates with an aerodynamic diameter less than 10 μm or 2.5 μm
- RR, relative risk
- SO2, sulphur dioxide
- TSP, total suspended particles
- URTI, upper respiratory tract infections
Many epidemiological studies on air pollution focus on hospital admissions and mortalities as health outcomes.1,2,3,4,5,6,7,8,9,10,11,12 These outcomes represent, respectively, serious morbidities and the ultimate health consequences of air pollution. Illnesses seen in primary health care settings, by contrast, form the much wider base of the pyramid of air pollution related illnesses. Although most of these illnesses are minor in nature with minimal long term effects on health, they represent a substantial proportion of the overall morbidity in the community. In a local cross sectional morbidity study, 41% and 48% of outpatient consultations in the private and public sector, respectively, were for respiratory illnesses.13,14 The total number of GP consultations for respiratory illnesses has been estimated to be about 23 million in Hong Kong in 2001.15 The direct medical cost exceeds HK$3.4 billion (US$442 million). With the associated productivity loss, these illnesses constitute a substantial economic burden on the community. Despite the significant health and economic impact of air pollution, few epidemiological studies on air pollution and health have been conducted in the primary healthcare sector.16–22 An important reason for the paucity of such studies is the absence of routinely collected data on primary health care in both private and public sectors in most countries.
To study the effect of air pollution on morbidities in the primary care setting, we prospectively collected data on daily visits for upper respiratory tract infections (URTI) and other diseases from 13 GP clinics over a 3 year period from 2000 to 2002 inclusive, and looked for associations with daily variations in the concentrations of the following air pollutants: NO2, SO2, O3, PM10 and PM2.5, using a time series approach. Our preliminary findings, based on 1 year data (2000) from seven GP clinics, have been reported earlier.23
Data from GP clinics
From January 2000 to December 2002 we recruited 13 GPs in eight districts of Hong Kong to collect data on the daily number of first visits for respiratory and other diseases at their clinics. The length of their participation in the study varied from 12 to 36 months, depending on the time of their recruitment and withdrawal. The following diseases/symptoms were recorded:
Respiratory diseases/symptoms: upper and lower respiratory tract infections, influenza, asthma, chronic obstructive pulmonary diseases, allergic rhinitis, cough and other respiratory diseases.
Cardiovascular diseases/symptoms: cardiac arrhythmias, angina, hypertension and other circulatory diseases.
Diseases of the other systems.
Diseases or symptoms were coded according to the International Classification of Primary Care 2nd edition (ICPC-2).24 This is an alphanumeric 3 digit system which identifies the “episode” of doctor consultation by system and diagnosis or reason for encounter. The system is partly compatible with the 10th revision of the International Classification of Diseases (ICD-10). The following information was recorded: age and sex of the patients; their place of residence and work by district; whether it was a first visit or a follow up visit for the same disease episode; the diagnoses (categorised according to the ICPC-2); their employment status; and the duration of sick leave granted by the doctor.
Data on air pollutants
Daily mean concentrations of four “criteria pollutants”—nitrogen dioxide (NO2), sulphur dioxide (SO2), ozone (O3), and particulates with an aerodynamic diameter less than 10 μm (PM10)—were measured in air monitoring stations located in eight districts throughout Hong Kong by the Environmental Protection Department (EPD). In addition, particulates with an aerodynamic diameter less than 2.5 μm (PM2.5) were monitored in one station. SO2 was measured by UV fluorescence (Model 43A SO2 Analyzer, Thermo Environmental Instruments Inc, USA); NO2 was measured by chemiluminescence (Model 42 Chemiluminescence NO-NO2-NOX Analyzer, Thermo Environmental Instruments Inc, USA); O3 was measured by UV absorption (ML8840 NOX Analyzer, Monitoring Lab Inc, USA); PM10 was measured by the Tapered Element Oscillating Microbalance (TEOM) (Rupprecht & Patashnick Co Inc, TEOM Series 1400a-AB PM10 Monitor, USA), and PM2.5 was measured by TEOM (R & P Partisol Plus, Model 2025, USA). Each of the 13 GP clinics was matched with a corresponding set of air pollutant data in his/her district.
We used a statistical model developed from the APHEA II protocol that has been used for time series analyses of mortality and hospital admissions in Europe.25 Missing air pollutant data in one station were predicted by a regression of data from that station on corresponding data from the nearest station. A regression was fitted for the daily concentrations of the pollutant with missing values (Y) on the corresponding values of the pollutant on the same day from the nearest station (X). The predicted values (Y1) obtained from the regression equation were then used to replace Y.
A generalised additive model26 using a Poisson distribution with log-link function was constructed as a core model. This regressed the daily numbers of GP visits for respiratory diseases in each clinic on the time variable (day), day of the week variable, daily mean temperature and humidity, and a holiday indicator. Only first visits of the disease episodes by patients who lived, worked, or attended school in the same district as the GP clinic were included in the model. Smoothing of the time variable was used to control for long term seasonal patterns of GP visits and smoothing splines were used as smoothers. The quasi-likelihood method was used to correct for overdispersion.27 Each core model was chosen based on the degree of freedom that gave the minimum AIC value.27 Autocorrelation was adjusted by adding autoregressive terms (GP consultations in the previous day up to 3 days) to the model. After the confounding effects of seasonality, days of the week, and climatic variables had been controlled, daily concentrations of PM10, NO2, SO2, and O3 (obtained from the monitoring station located in each district) were added to each core model to determine the relative risk (RR) of GP visits for a 10 μg/m3 increase in each of these air pollutants. Data on daily concentrations of PM2.5 were used in three models which took data from GP 9 (in the same district as the PM2.5 monitoring station) and GPs 1 and 4 (in a contiguous district). Concentrations of all the air pollutants for the same day (lag 0) up to 3 lag days (lag 3) and cumulative lags by 2 (lag 0 and 1), 3 (lag 0, 1 and 2), and 4 days (lag 0 to 3), were tested in each model. The lag day with the air pollutant concentration that yielded the smallest p value (that is, the largest χ2) was chosen. The standard errors of the estimates were computed using the supplementary program used in the re-analysis of the National Morbidity, Mortality, and Air Pollution Study (NMMAPS).28,29 The RRs obtained from individual GPs were combined using a random effect model.30
All calculations were performed with the software S-plus 4.0 using a more stringent convergence criteria in the gam(.) function.31
Ethical approval was not required as only counts of daily patient consultations were obtained from the GPs, stratified by various parameters without the individual patient’s identity. Data on air pollution were provided by the Environmental Protection Department.
Table 1 shows the summary statistics of consultations by 13 GPs in eight districts. The duration of their participation ranged from 12 months (2 GPs) to 36 months (4 GPs), with a mean of 28.8 months and a median of 33 months. A total of 314 514 consultations were recorded from the 13 GPs; 296 372 (94.2%) consultations were by patients living, working, or attending schools in the same district. About three quarters of these were first visits of disease episodes, two thirds of which were for respiratory diseases (all types). Upper respiratory tract infection (URTI) accounted for 82.7% of all respiratory diseases. Other respiratory diseases and cardiovascular diseases were few in number.
Table 2 shows the means, ranges, and percentiles of daily air pollutant concentrations in eight air monitoring stations during the study period. NO2 and PM10 are the dominant air pollutants in five districts (Eastern, Kwun Tong, Tsuen Wan, Kwai Chung and Yuen Long). The corresponding concentrations of O3 in these districts (with the exception of Eastern and Yuen Long) are low. This pattern is reversed in three districts (Shatin, Tai Po and Central and Western), with relatively low NO2 and PM10 levels but high levels of O3. PM2.5 was highly correlated with PM10 (correlation coefficient r = 0.94). NO2 was moderately correlated with PM10 (median r = 0.68; range 0.38–0.75) and SO2 (median r = 0.47; range 0.36–0.64), but poorly correlated with O3 (median r = 0.12; range −0.15–0.31). O3 was moderately correlated with PM10 (median r = 0.40; range 0.14–0.48) but poorly correlated with SO2 (median r = −0.13; range −0.04 to −0.42). SO2 was poorly correlated with PM10 (median r = 0.28; range 0.19–0.51).
The degrees of freedom used to control the seasonality in our statistical models differed for individual GPs. The mean, standard deviation, and median degrees of freedom for the 13 models are 80.8, 51.2, and 70, respectively.
The individual relative risks (RRs) of patients’ first visits for URTI by clinic, and the summary RRs for all clinics (per 10 μg/m3 increase in the concentrations of each air pollutant) are shown in table 3. Statistically significant summary RRs (at 95% confidence level) of first visits to GP clinics for URTI were found for O3, NO2, PM10 and PM2.5, but not for SO2. The magnitudes of the excess risk of GP visits for URTI, given by 100 × (RR – 1)%, were 3.0% for NO2, 2.5% for O3, 2.0% for PM10, and 2.1% for PM2.5. The individual and combined RR of first visits for non-URTI respiratory diseases were calculated for three GPs (GPs 9, 12 and 13) with sufficiently large numbers of visits (>2 per day) for modelling (table 4). The findings were similar to those for URTI. The daily numbers of first visits for individual respiratory diseases and for cardiovascular diseases as a group were too few for model construction.
As the RRs obtained from GP 2 (who contributed 1 year’s data with a much lower proportion of patients coming from his own district) were considerably higher than the corresponding figures from other GP clinics for four pollutants, a sensitivity analysis was performed by excluding GP 2 in calculating the summary RRs. The summary RRs and 95% confidence intervals were little changed, which did not affect the overall interpretation of our results.
A location map of the GP clinics and the air monitoring stations is shown in fig 1.
This is one of few time series studies on air pollution and community morbidity (as reflected in visits to GP) reported in the literature. For populations where routinely collected data are not available, GP consultation data have to be prospectively collected over a sufficiently long period for time series analysis, which accounts for the paucity of similar studies.
The summary RR of first visits to the GP for URTI are statistically significant and above unity for a 10 μg/m3 increase in four out of five air pollutants—namely, NO2, O3, PM10, and PM2.5. The association between air pollution and GP visits due to non-URTI respiratory diseases was analysed in three GP datasets and the results were similar to those for URTI. The RRs for PM2.5 closely mirror those for PM10, which reflects the strong correlation between the two pollutants. The magnitude of the RR for PM2.5 was uniformly higher than that for PM10. While the RRs for particulates of different sizes were found to be similar in some studies,32 our results are consistent with a number of studies which found that PM2.5 was more strongly associated than PM10 with acute health outcomes.33–35 It should be cautioned, however, that only one of three GP datasets (GP 9 in Tsuen Wan) used for calculating the RR for PM2.5 was from the district where PM2.5 concentrations were monitored. The other two GP clinics (GPs 1 and 4) were situated in the neighbouring district of Kwai Chung. Our decision to use PM2.5 data from Tsuen Wan for these two GPs was based on a higher propensity for PM2.5 to be more evenly dispersed than PM10.
Several GP clinics were spread over geographically diverse districts (corresponding to the locations of air monitoring stations) in Hong Kong. The clinic data were audited by systematic checks and the data quality was satisfactory, with few errors or missing data. Data on air pollution were subject to quality control measures as stipulated by the Environmental Protection Department. Daily data obtained from individual GPs could not be combined to produce a large single dataset for statistical modelling, as there were differences in the time period of their enrolment into the study. Instead, individual datasets were used to create models that yielded district based RR. These were then combined into a single summary RR for each air pollutant by methods used in meta-analysis. This approach circumvented the inherent instability of datasets from the individual clinics. Another advantage of the district based approach was that only first GP visits for respiratory diseases by patients living, working, or attending schools in the district were regressed with the air pollutant concentrations in the same district (with the exception of PM2.5). Hence, there was a better match between exposure and health outcome than if the pollutant concentrations were averaged over all stations in a territory wide dataset. This district based approach probably resulted in less exposure misclassifications than in the London study16–18 which used the mean values from different numbers of stations for different pollutants in various locations. There have been concerns that exposure estimates based on central monitor values do not accurately reflect individual personal exposure.36 However, Schwartz et al34 pointed out that, in a time series approach, the number of health outcomes per day was calculated over the population. The relevant exposure should be the mean of personal exposures on that day, which was probably more tightly correlated with monitoring station data than individual exposures. The generalised additive model used is the most widely used statistical model in recent studies, and the more stringent criteria have been applied in the modelling to obtain a more accurate estimate of the RR and its standard error.28,29 From the clinical data we could not detect any major influenza epidemics during the study period. Hence, the latter were not included in our models.
Our participant GPs were a selected volunteer group with the inherent problem of selection bias. To address this issue, we have checked the patient profile, the distribution of the diagnostic categories, the consultation hours, and the average workload of the doctors. In general, the above parameters for most of our GP clinics are fairly typical of those seen in a private GP practice in Hong Kong.13,14 There is no reason to believe that the observed associations between air pollutant concentrations and visits to GP for URTI and non-URTI respiratory diseases could only have occurred in our GPs but not in others. Moreover, the association could not be explained by some unidentified confounders. Hence, we conclude that the positive association between air pollution and respiratory diseases was likely to be true, as has been reported elsewhere.16–22 These findings are supported by similar associations between hospital admissions for respiratory diseases and mortalities that have been reported in other local studies6,7 and numerous epidemiological studies elsewhere.1–5,8,9,10,11,12 Nevertheless, as in all ecological studies, we cannot conclude that a cause-effect relation exists between air pollution and GP visits.
The London study16–18 on air pollution and GP consultation is one of few studies to link air pollutants with specific respiratory morbidity seeking primary medical care in the community. That study differed from ours in that it used retrospective datasets from a larger number of GPs (45–47) for a 3 year period. Asthma and lower respiratory tract infections were found to be significantly associated with an increase in air pollution (NO2 and CO in children, and PM10 in adults).16 Allergic rhinitis was associated with SO2 and O3, and other upper respiratory diseases were associated with PM10.17,18
In another study in a hospital outpatient department in Beijing, positive associations were found between an increase in total suspended particulates (TSP) and SO2 and an increase in outpatient visits in certain specialties (“non-surgical”, paediatrics and internal medicine). However, data on specific diseases or diagnostic codes were not available.19,20
In our dataset, except for URTI (and non-URTI respiratory diseases for three GPs), daily consultations for individual respiratory diseases and cardiovascular diseases are too few for statistical modelling. URTI (the largest component of all respiratory diseases seen in general practice) are significantly associated with all air pollutants except SO2. Gaseous air pollutants are irritants to the respiratory tract and can cause inflammation and respiratory symptoms or diseases. PM2.5 and PM10 can penetrate deeply into the lower respiratory tract and would be expected to be associated with lower respiratory tract diseases rather than URTI. However, many respiratory symptoms such as cough, phlegm, chest tightness, and others could have resulted from irritation of both upper and lower airways. Some of these visits might have been labelled by our GPs as URTI. Moreover, both pollutants are statistically highly correlated with TSP, which is predominantly made up of larger particles that may cause irritation and inflammation of the upper respiratory tract. Hence, the associations of PM2.5 and PM10 with URTI are biologically plausible. We could not explain the lack of significant association between URTI and non-URTI respiratory diseases and SO2, other than the small effect size of the latter. The concentration of SO2 in Hong Kong has decreased substantially since the mandatory use of low sulphur fuel in 1990.37 However, the magnitude and significance of the RR derived from the individual GPs were not related to the mean SO2 concentrations in their corresponding districts.
The excess risk in the London study cannot be directly compared with our findings because of the difference in the disease coding and the units of air pollutant concentration used (10–90th centiles in the London study and 10 μg/m3 in this study). Owing to small numbers, our data could not be used for subgroup analysis by specific diseases or groups such as asthma, lower respiratory diseases, and allergic rhinitis except for non-URTI in three GP clinics. Likewise, we could not analyse our data separately for different age groups. In general, the excess risks in the London study for SO2 in allergic rhinitis were high (15.7% and 9.3%, respectively, for the age groups 0–14 and 15–64 years). SO2 was also significantly associated with asthma, lower respiratory diseases, upper respiratory diseases, and allergic rhinitis. By contrast, the RR for SO2 was not significant in this study. The magnitudes of the excess risk of other pollutants (PM10 and NO2) were broadly similar in both studies, despite the differences in the health outcomes. An anomalous finding in the London study is the significant negative association between O3 and lower respiratory diseases and upper respiratory diseases (for all three age groups) but the significantly positive association with allergic rhinitis. The authors gave no plausible explanation for this observation, but noted a negative association between O3 and most other pollutants in winter. In our study O3 was positively associated with URTI and non-URTI respiratory diseases. The correlation between O3 and most other air pollutants is weakly positive (except SO2, where r was negative). Pollutants such as black smoke and carbon monoxide have not been measured or analysed in our study. Although there are differences in the association between individual pollutants and the diseases/disease groups between the London study and ours, we share the overall conclusion that air pollutants are positively associated with GP consultations for respiratory diseases.
This study provides additional evidence for the effect of air pollutants encountered at current levels on the health of the public. The results are broadly in agreement with other studies, with minor differences in specific relationships between individual air pollutants and diseases. In some time series studies on mortality rates there have been suggestions that air pollution might only exert a “harvesting effect” on susceptible groups such as the elderly and chronically sick whose lifespan might have been short anyway. In hospital admission studies the risk estimates were especially obvious for susceptible groups such as the elderly, the young, or the chronically sick. However, most time series studies have, in fact, shown a real increase in the disease burden and a substantial shortening of life.38,39 Our study shows that air pollution, besides affecting the “at risk” populations, also affects the relatively healthy population—those who usually consult their GPs for URTI and other respiratory diseases. Instead of attributing the association between air pollution and respiratory diseases to “harvesting” of susceptible individuals, the hypothesis that air pollution could lead to otherwise healthy people falling ill and visiting GPs seems a more plausible explanation. Despite the minor nature of these ailments, the overall morbidity burden on the community and on the healthcare system is heavy. The epidemiological evidence of the impact on health of the air pollutants PM10, PM2.5, NO2 and O3 (and SO2, as shown in other studies) at current ambient levels on the apparent absence of a “threshold effect”, in particular for particulates and O3,40 provides a scientific basis for setting more stringent air quality standards than those currently used in Hong Kong. Our findings suggest that a reduction in the concentrations of air pollutants could be associated with a corresponding decline in the numbers of GP visits for URTI and non-URTI respiratory diseases. This hypothesis has important implications in the formulation of environmental health policies. Like most cities, the major local sources of air pollutants in Hong Kong are motor vehicles and power plants. More vigorous control of these sources is likely to confer substantial health benefits.
The authors thank all the general practitioners for providing their patients’ data.
Published Online First 14 March 2006
This project was partly funded by the Environmental Protection Department of Hong Kong.
Competing interests: none declared.