Associations between daily hospital admissions for cardio-respiratory diseases and particulate matter less than or equal to 10 microns in aerodynamic diameter (PM₁₀) have been described in many settings worldwide including North America, Europe, Asia and Australia 1. In large cities, where the vast majority of research has been conducted, fossil fuel combustion in industry and transport are major sources of PM₁₀. However, depending on the setting, there are potential contributions from a range of other sources including crustal particles and biomass combustion such as forest fires and wood fuels 2. The relative effects of different sources of particulate pollution on adverse heath outcomes, and differences in these effects across population sub-groups, remain major gaps in the currently available evidence 1. In particular, the relative role of particulates derived from biomass as opposed to fossil fuel combustion remains unclear, although two empirical studies of PM₁₀ derived from vegetation fires 34, and one review of studies examining PM₁₀ from wood smoke 5 all observed that the magnitude of associations with respiratory outcomes is greater when PM was derived from biomass combustion. However, studies examining a single source of ambient PM₁₀ are infrequent because of the difficulty finding a site without a mixture of various pollutants, and the complexity of apportioning contributions from different sources 26. A few epidemiological studies have apportioned total PM according to a range of sources (such as biomass, crustal and motor vehicles) and have found a range of different clinical outcomes were associated with different exposure sources 78.

The city of Darwin enables the health effects of vegetation fire smoke to be assessed because the source of the PM pollution is due almost entirely to fire smoke. Particulate matter derived from biomass combustion has been identified as an increasing and unregulated source of outdoor air pollution. The use of wood burning for domestic heating is increasing in several countries 910, while the frequency and severity of uncontrolled vegetation fires is increasing the world over 11. Vegetation fires generate pollution episodes across wide geographic areas, and major population centers are frequently affected 12.

In Australia, the increasing use of deliberate fuel reduction burns as hazard reduction activities to avert major fire disasters is becoming more controversial in the light of the evidence of adverse health impacts of particulate air pollution 13.

The tropical city of Darwin (Latitude: -12.462, Longitude: 130.842) provides an opportunity to specifically examine the health associations of vegetation fire smoke. Here 50–70% of the surrounding savanna burns annually during the 8 month dry season between April and November 14. The months from December until March are referred to as the wet season when approximately 80% of Darwin's average annual total rain falls. Due to the rain, fires only occur during the dry season and the smoke from these fires are the source of 95% of measured PM₁₀ in the city 15. There are no other important sources of air pollution such as traffic or industry and so PM is negligible during the wet season 16. A comprehensive air quality study was conducted in the year 2000 15 and the average concentrations of other pollutants including ozone, sulfur dioxide and nitrogen dioxide are negligible.

During the dry season, prevailing south-easterly winds bring vegetation fire smoke over Darwin from a large region of savanna. The lower atmosphere of the airshed is characteristically stable during dry seasons and there is a persistent inversion at about 3000 meters 17. These conditions produce similar concentrations of ambient PM₁₀ across the city. This was validated in 2005 when PM₁₀ measurements at two monitors located 25 km to the west and south were shown to be of similar magnitude and highly correlated with the primary monitor 17. This evidence shows the monitor values for PM₁₀ are representative of the community's exposure.

The pattern of vegetation burning remained consistent throughout the study period. This was demonstrated by analysis of satellite data which have confirmed the ongoing regional and seasonal nature of annual landscape fires 18 and by an air quality monitoring campaign conducted 25 km north-west of Darwin in the mid 1990s 16.

Darwin has a population of approximately 110,000 people and also provides an opportunity to examine the relative impact on indigenous Australians, a high-risk population subgroup comprising 11% of the population of Darwin 19. Socio-economic disadvantage, chronic cardio-respiratory diseases and diabetes have all been shown to modify the effect of particulate air pollution on health outcomes 20. Indigenous Australians have a high prevalence of all these health risks and have been recognized as being likely to be at much greater risk from poor air quality than other Australians 21. This has been stated as a priority for Australian public health research 22.

A previous case-crossover study of the hospital admissions and observed PM₁₀ in Darwin showed a positive association with respiratory diseases, and disproportionately higher effect estimates in indigenous people 23. That study had limited statistical power as Darwin's population is relatively small and only three years of air quality data were available for analysis. Here we attempt to address these limitations by using PM₁₀ estimations over a 10-season period using a previously validated predictive model based on visibility records 17 and using the alternative method of time series modeling 24.

There were 2,410 days in the 10 dry seasons of our study period. There were 8,279 admissions during this period. The total numbers of hospital admissions (and proportion of patients under 15 years old) are given in Table 3, stratified by clinical grouping and indigenous status. Despite indigenous people representing 11% of the population of Darwin, they comprised 23% of these admissions.

Descriptive statistics for daily admissions in each disease category, estimated daily ambient PM₁₀ and meteorological parameters are summarized in Table 4.

Our modeling procedure used a sensitivity analysis similar to the method described by Dominici and colleagues 26 to select the optimal degrees of freedom for the smoothed function of time; to minimize bias in the estimates of the pollution coefficients. This sensitivity analysis was applied to the model for the estimated ambient PM₁₀ lag with the greatest absolute t-value. We adjusted the degree of smoothing on the time variable by applying different values of a multiplier (α) that ranged from 0.2 to 3 times the degrees of freedom which had been chosen a priori 2830. The influence that this had on the effect estimate was assessed using the change in the mean squared error. Theoretically there is lower bias in the estimate caused by smoothing at higher values of α, but there is larger statistical uncertainty. We conservatively selected the optimal smoothing function for minimizing bias in the point estimate.

The point estimates and 95% Confidence Intervals (CI) for the association between hospital admissions with estimated ambient PM₁₀ are reported here as the percentage change in the relative risk per 10 μg/m³ change in exposure.

Initial modeling without the interaction between indigenous status and PM₁₀ found a positive association for total respiratory admissions with same day estimated ambient PM₁₀ (4.81%; 95%CI: -1.04%, 11.01%). The subgroups of respiratory infections, asthma and COPD all had positive associations with same day estimated ambient PM₁₀. The small associations for all cardiovascular diseases and IHD were all negative or zero and not statistically significant. Due to small numbers in these groups the confidence intervals are wide.

We then compared the effects for indigenous and non-indigenous people. Figure 3 shows the point estimates and 95% confidence intervals for the association between hospital admissions with estimated ambient PM₁₀ when an interaction term with indigenous status is included. A statistically different association (p-value = 0.01) was observed for respiratory infections in indigenous people of 15.02% (95%CI: 3.73%, 27.54%) at a lag of 3 days while no association was evident for this condition in non-indigenous people at this lag (0.67%; 95%CI: -7.55%, 9.61%).

The point estimates for the effects in the other disease groups where not significantly different at the 95% confidence level; however the indigenous estimates were consistently higher than those for the non-indigenous population. The association of total respiratory admissions with same-day ambient PM₁₀ in indigenous residents was much higher (9.40%; 95%CI: 1.04%, 18.46%) compared with the estimate for non-indigenous residents (3.14%; 95%CI: -2.99%, 9.66%). For asthma admissions and estimated ambient PM₁₀ there was a non-significant estimated increase at a lag of 1 day: 16.27% (95%CI: -3.55%, 40.17%) for indigenous compared with 8.54% (95%CI: -5.60%, 24.80%) for non-indigenous people.

There were positive non-significant estimates for same day estimated ambient PM₁₀ with COPD admissions in both groups. This is in contrast to negative associations with COPD admissions and lagged estimated ambient PM₁₀ in both groups.

There were no clear associations with estimated ambient PM₁₀ and total cardiovascular admissions or IHD. For non-indigenous admissions the estimated association with total cardiovascular admissions for ambient PM₁₀ at lag 0 was -3.43% (95%CI: -9.00%, 2.49%) and the estimate for indigenous admissions was -3.78% (95%CI: -13.4%, 6.91%), although indigenous people did have positive (non-significant) estimates at lags 2 and 3.

We found generally positive associations between PM₁₀ with total respiratory admissions, asthma and respiratory infections especially among indigenous people for total respiratory admissions (at lag 0) and respiratory infections (at lag 3). Negative associations were apparent between lagged PM₁₀ and COPD. There were generally negative non-significant associations for cardiovascular outcomes in both population groups.

While we report our findings for several diagnostic sub-categories, the numbers in these groups, especially asthma and COPD, were much smaller and our confidence in effect estimates is greatest for the total respiratory and total cardiovascular classifications.

Our observed negative association with COPD admissions was unexpected as previous biomass studies have generally found strong positive associations with this outcome 33343536. This could be due to hospital admission practices as in many instances patients presenting with exacerbations of asthma and COPD will be discharged home from the emergency department and therefore not be included in admissions data. However insufficient emergency department data were available for examination. Alternately it may be that persons with pre-existing chronic respiratory conditions take extra precautions with their care during days when PM levels are noticeably extreme.

In addition these counterintuitive results may be due to chance reflecting the low precision of our estimates due to the relatively small numbers of daily admissions. The lack of precision in our estimates could also be a function of other factors, such as uncertainty in exposure estimates, variation in population response, and even the lack of any association. However our findings are consistent with other studies of ambient biomass smoke and contribute to the limited evidence concerning the health effects of vegetation fire PM₁₀.

A previous case-crossover study in Darwin had similar findings to this study with positive associations reported between observed ambient PM₁₀ and respiratory admissions with Odds Ratio (OR) 1.08 (95%CI: 0.98, 1.18) and a tendency towards negative associations with cardiovascular admissions (OR 0.91; 95%CI: 0.81,1.02) 23. Similarly, the estimates from that analysis for total respiratory admissions were also approximately double for indigenous rather than non-indigenous people.

A study in Christchurch, New Zealand, where ambient PM₁₀ predominantly arises from the combustion of wood for domestic heating, found a 3.37% (95%CI: 2.34%, 4.40%) increase in total respiratory admissions per interquartile rise in ambient PM₁₀ (IQR = 14.8 μg/m³) at a lag of 2 days 33. That study found an association with admissions for heart failure but not other cardiac diagnoses, while a later study in Christchurch found no association with cardiovascular admissions 37.

In a study of the South East Asian forest fires of 1997 Mott et al 34 found large fire-period related increases in respiratory hospitalizations for asthma and COPD, ranging from 40–80% in adults but no association with cardiovascular admissions although people with pre-existing cardio-respiratory diagnoses were at greatest risk.

A recent study from Brisbane, Australia, directly compared the association between bushfire and non-bushfire derived particulates on total respiratory hospital admissions excluding influenza 3. That study analyzed the PM₁₀ distribution as a three-level factor with levels defined as low (< 15 μg/m³), medium (15–20 μg/m³) and high (> 20 μg/m³). They found that for an increase in same-day PM₁₀ from low to high there was an increase in the relative risk for total respiratory hospital admissions of 19% (95%CI: 9%, 30%) whereas on non-bushfire days the associated increase was 13% (95%CI: 6%, 23%).

A similar study from Sydney, Australia, directly compared associations between cardio-respiratory hospitalizations and ambient PM₁₀ derived from vegetation fire smoke with associations between these outcomes and ambient PM₁₀ derived from other sources 4. They apportioned ambient PM₁₀ on vegetation fire days into particulate matter derived from burning biomass and particulates due to other sources. They found a 1.24% (95%CI: 0.22%, 2.27%) increase in relative risk for all respiratory admissions per 10 μg/m³ increase in vegetation fire derived ambient PM₁₀ at lag 0. Ambient PM₁₀ due to other sources at lag 0 was associated with an increase in all respiratory admissions of 1.04% (95%CI: 0.02%, 2.07%) per 10 μg/m³ increase. They also failed to find an association between cardiovascular outcomes and vegetation fire smoke in contrast to findings of a positive association between cardiovascular admissions and ambient PM₁₀ from all other (non bushfire) sources.

The magnitude of the point estimates for all respiratory admissions from our study, the studies discussed above and several other studies of outpatient attendances for respiratory conditions in association with vegetation fires 333438394041, are much greater than multi-city studies of associations between admissions for respiratory diseases (including asthma, COPD, and total respiratory admissions) with positive associations for a 10 μg/m³ change in ambient PM₁₀ of the order of just 1–1.5% 4243. Dominici et al 2006 found similar associations of around 1% increase in respiratory admissions per 10 μg/m³ change in PM_2.5 28. The greater magnitude of adverse respiratory effects reported in studies specifically examining biomass smoke might reflect a true difference in the adverse outcomes associated with this source of PM. However, studies of biomass smoke are usually conducted in cities and towns with small populations, or around short episodes of extreme exposures, and their results inevitably are less precise than those from multi-city studies making direct comparisons difficult to interpret. Similarly, the absent or negative associations between biomass smoke and cardiovascular disease outcomes in our study and in three previous studies of vegetation fire smoke 42334, might also reflect a different pattern of adverse health outcomes from biomass smoke. However these findings require replication as cardiovascular admissions have been clearly associated with ambient PM₁₀ in many large studies, usually conducted in urban settings where fossil fuel combustion is a major source of PM 1.

The primary strengths of this study are the spatially homogenous population exposure to particulates across Darwin 17, the specific source from vegetation fires 15, the hospital data collection which represents the admissions patterns for the entire population of the city and the inclusion of details of indigenous status in the health records. These factors all minimized the problems of exposure and outcome misclassification inherent in population-level studies. Additionally, due to Darwin's tropical climate, there was minimal variation of daily temperature and humidity minimizing confounding by meteorological changes.

An important limitation of this study is the lack of air quality data, necessitating our use of an estimate based upon daily visibility. This inevitably will have introduced exposure misclassification bias limiting our ability to detect associations that might be present. In addition, because of the small population of Darwin, there were low numbers of daily admissions for cardio-respiratory diseases in spite of our relatively long 10-season period of data for analysis. This limited the statistical power and reduced the precision of our point estimates. However, population-level studies of the health effects of ambient biomass smoke have inherent limitations. Vegetation fire events affecting large populations are rare, unpredictable and often of short duration. In addition settings where biomass is the predominant source of ambient particulate matter tend to have smaller populations as larger cities will have a more complex mix of pollutants often dominated by fossil fuel combustion by industry and transport. For this reason the results from studies specifically examining vegetation fire smoke pollution will almost inevitably have greater technical challenges than studies examining ambient PM regardless of source.

A key reason why PM₁₀ effect estimates may differ by region is the different sources and resulting chemical composition of particles, such as the biomass burning noted here. Most of the literature concerning chemical composition compares different size classes, such as PM₁₀, PM_2.5, PM_2.5–10 and PM₁ and while all size classes have been associated with adverse heath outcomes, smaller particles have generally been found to be relatively more toxic 1. Our study could not examine different size classes however, a detailed study of PM in Darwin during 2004–5 found that the total PM_2.5 comprised on average 56% of the total PM 44.

In addition to the different ratios of size fractions, bushfire derived PM is associated with a distinctive suite of toxic co-pollutants including metals, organic and inorganic compounds 9. In vitro studies have demonstrated that different chemical compositions induce different cellular responses 45. Moreover a few epidemiological studies have apportioned total PM according to a range of sources (such as biomass, crustal and motor vehicles) have found that the magnitude of a range of different clinical outcomes were associated with different exposure sources. A study of hospitalizations in Copenhagen, Denmark found respiratory outcomes to be predominantly associated with biomass particulates and crustal and secondary particulate sources with cardiovascular outcomes 7. However in that study asthma was more closely associated with markers of car exhaust. A study of mortality in Phoenix, USA found that secondary sulfate, traffic, and copper smelter-derived particles were most consistently associated with cardiovascular mortality while biomass derived particulates were not 8. These source apportionment studies are compatible with the few studies, including ours, that have specifically examined biomass smoke derived PM.

Our study also compared rates of admissions between indigenous and non-indigenous subpopulations finding the suggestion of disproportionate burdens of health effects due to the seasonal fire smoke pollution; especially a statistically different association between PM₁₀ and admissions for respiratory infections three days later. This is consistent with the only previous study examining this issue in Australia 23. Many factors could contribute to this including excess socio-economic disadvantage, chronic cardio-respiratory diseases and diabetes 21 which all modify the effects of ambient PM₁₀ on cardio-respiratory admissions 20. Other factors could include reduced access to health services and therefore early management of chronic conditions 46, and different patterns of smoking, physical activity or diet among this population sub-group 21. In addition, the two populations have differing age structures, with a greater proportion of people over sixty-five in the non-indigenous group, and a greater proportion of children less than 15 years in the indigenous group 19. The former factor could result in an underestimate of the difference between the two population groups, while the latter could have contributed to the differences in respiratory infections observed between the two groups. We have attempted to control for this age-structure effect by including a term for indigenous status which should capture this. Residential segregation is less likely to explain the difference in this setting as exposure is relatively uniform across the city 17.

Our results suggest associations between vegetation fire smoke and daily hospital admissions for respiratory diseases that were stronger in indigenous people. The analysis found approximately three-fold higher associations between same-day estimated ambient PM₁₀ and total respiratory admissions in indigenous people than non-indigenous people. This has implications for local public health policy and practice, such as the identification of sensitive sub-groups, the setting of air quality guidelines, targeting of public health messages in relation to air pollution and the regulation of deliberate burning practices 22.

This is an important research area to pursue. With global change bringing changes in vegetation burning regimes and increasing population exposures to pollution from vegetation fires, understanding and managing the health impacts of biomass combustion smoke will become an increasingly important public health activity.

CI: Confidence interval; COPD: Chronic Obstructive Pulmonary Disease; ICD10: International Classification of Diseases, 10th Revision; ICD9: International Classification of Diseases: 9th Revision. IHD: Ischemic Heart Disease; OR: Odds ratio; PM: Particulate Matter; The aerodynamic diameter of the particles is shown by the additional of the size range in microns as subscripts. For instance PM with diameter less than or equal to 10 microns is PM_10, PM with diameter less than or equal to 2.5 microns is PM_2.5 and so forth.

The authors declare that they have no competing interests.

ICH carried out the analysis and drafted the manuscript. FHJ conceived the study and helped to draft the manuscript. GGM provided theoretical and conceptual guidance and helped to draft the manuscript. All authors have read and approved this version of the manuscript.