Background
The expansion of the coal seam gas (CSG) industry in Australia has raised concerns about potential human health impacts in part because of a current lack of human health impact assessment information, as well as accessible baseline studies in Australia [
1]. Furthermore, information on exposures to CSG-associated environmental hazards is minimal. There is a need for source-to-effect pathways to be fully mapped for relevant exposure media, including air and soil. In the broader unconventional natural gas development (UNGD) context, there is generally a lack of health research on the effects of UNGD [
2].
Some health-related studies have been conducted in other locations, predominantly in the United States. These include cross-sectional studies [
3,
4], ecological [
5], qualitative [
6,
7], retrospective cohort [
8], ‘difference-in-differences’ design [
9], as well as human health risk assessments [
10‐
12] and Health Impact Assessment (HIA) [
13]. Such studies have provided some evidence for adverse health outcomes potentially associated with UNGD, but have suggested the need for further research. These studies have predominantly examined cancer incidence [
5], birth outcomes [
8,
9,
14], cancer and non-cancer risks for air emissions [
10], and a range of other areas of concern identified in an HIA (including air pollution, water and soil contamination, and community wellness) [
13,
15].
Investigation of the association between exposures associated with CSG extraction and health outcomes are often limited by the relatively small population base exposed (as CSG wells are often established in less populated rural areas). This is combined with a focus on often rare health outcomes with long latencies (such as cancer incidence), or common outcomes and/or syndromes that could plausibly be attributed to other putative causes occurring contemporaneously to CSG development.
Adgate et al. [
2] noted that more epidemiological studies are needed to determine what disease patterns may exist and how UNGD may affect these patterns. This may be helpful for companies to improve practices to reduce exposures, or to assist in formulating stricter regulations where necessary. A recent literature review concluded that the majority of studies published on UNGD and environmental health concentrated on shale gas, focused only on a few key areas of environmental health, such air and water, and generally lacked methodological rigor [
16]. Many of the UNGD-related studies have focused on shale gas, which means that these outcomes do not necessarily translate to the CSG context due to the differences between both types of UNGD [
17].
To the authors’ knowledge, no epidemiological studies have been conducted on the impacts of CSG on human health, in Australia or elsewhere. This, as well as the level of public concern, prompted this exploratory study. The objective of this analysis was to explore trends in hospitalization rates for three designated geographical study areas (CSG, coal mining, rural) in Queensland over the period 1995–2011. The three study areas were used to determine whether there were increases in hospitalization rates for health conditions, as measured by ICD-10-AM codes, in the CSG area compared to the coal mining and rural study areas.
Discussion
The objective of this study was to assess potential health impacts of CSG development activities by examining increases in hospitalization rates for health conditions measured by ICD-10-AM chapter codes in three designated study areas (CSG, coal mining, rural) in Queensland over the period 1995–2011. CSG development activities only began to increase in 2001/2002 and steadily increased starting in 2005/2006 (see Fig.
1). To our knowledge, this is the first study of its kind in Australia.
In order to contextualize the results of the current analysis, available literature on UNGD impact evidence was reviewed to identify possible health conditions where increases in hospitalization rates due to CSG may be expected. Potential health outcomes included birth defects, cancer, cardiovascular outcomes, dermatological outcomes, injuries, neurological problems, psychosocial stress, respiratory disease, sexually transmitted infections, and vector-borne disease [
2,
8,
13,
30‐
32]. These outcomes were matched with the appropriate ICD chapters where such outcomes would appear if a person were to be hospitalized. The identified ICD chapters are shown in Table
3, along with the ICD chapters where increases in hospitalization rates over time were observed in the CSG area relative to the CHI or RLI areas. Due to the scarcity of previously published data (generally, but also specifically within Australia), it was considered important to examine changes over time in hospitalization rates for all ICD chapters, not just those chapters matched from previous literature.
Table 3
Potential health outcomes associated with UNGD and corresponding ICD chapters from the literature, as well as the observed outcomes (unadjusted and adjusted) from this study for the coal seam gas (CSG) area
STIs | Infectious disease | |
Vector-borne disease | Infectious disease | |
Cancer | Neoplasms | X |
Mental health | Mental disorders | |
Neurological/nervous system | Nervous system | Xc |
Noise-related outcomes | Ear | |
Cardiovascular outcomes | Circulatory | |
Respiratory outcomes | Respiratory | |
Dermatological outcomes | Skin | |
Nephrotoxicity | Genitourinary | |
Impaired fertility | Genitourinary | |
Urological outcomes | Genitourinary | |
Perinatal outcomes | Perinatal | |
Birth defects | Congenital | |
Injuries | Injuries | |
n.d. | All-cause | Xc
|
n.d. | Blood/immune | X |
n.d. | Endocrine | |
n.d. | Eye | Xc
|
n.d. | Digestive | |
n.d. | Musculoskeletal | |
Other symptoms that were discussed in the literature were not included in Table
3 as they were either specific codes within chapters, or such outcomes would most likely fall within a group for which a person would not be admitted to hospital. This is true for many of the symptoms that have been reported (e.g., eye irritation, headaches, nosebleeds). Table
3 shows that a number of outcomes observed in the unadjusted and adjusted models in this study have not been previously mentioned in the literature as potential health outcomes related to UNGD (i.e.,
‘Blood/immune’ and
‘Eye’ diseases).
Very few UNGD-related studies have examined hospitalization rates. One study examined all-age hospitalization rates for all-cause admissions across four counties with varying degrees of UNGD in the USA [
33]. Garfield County, the county with the highest level of UNGD, was found to have the lowest or second lowest rate of all-cause hospitalizations. This finding is inconsistent with the present study, where no significant differences were found between areas, after adjustment of key demographic and socioeconomic characteristics. Another study examined hospital admissions alongside well number and density data [
32]. While the study by Coons & Walker [
33] used hospital admissions data over a 6.25-year period and the study by Jemielita [
32] was over a 5-year period, our study was over a 17-year period.
The only overlap between health conditions identified in previous literature as being potentially associated with CSG, and for which increases in hospitalization rates were observed in the CSG area relative to another study area in the current analyses, was for neoplasms. Hospitalizations with a primary diagnosis code within this ICD chapter represent diagnosis of a neoplasm, where related codes (e.g., treatment) are within ICD chapters that were excluded from this study. However, this was not one of the strongest outcomes for all-age hospitalization rates because increases over time were noted in the CSG area compared only to the RLI area for adjusted models.
It is difficult to draw any conclusions with respect to possible changes in environmental exposures due to the fact that neoplasm trends typically reflect events 10–20 years prior to manifestation or are due to cumulative lifetime exposures [
33]. Any short-term trends may not be reflective of changes in the health hazard impact potential of CSG development. Gas well development activity only began a steady increase in 2005/2006. Considering a very conservative lag period of 4 years [
34], the
‘Neoplasms’ data presented here could only be reflective of changes after this period, with manifestation of disease after 2009/2010 (if the
‘Neoplasms’-related diagnoses are related to any exposures associated with CSG development). Additionally, such changes can be an artefact of changes in screening practices [
33].
While unrelated to UNGD, other health-related studies noted changes in hospitalization rates could be due to service changes in care for certain health outcomes (e.g., diabetes or pneumonia) [
35,
36], changes in medical technology or laws [
26], coding [
35,
37,
38], or a combination of these factors [
35]. Therefore, it could be hypothesized that the noted differences could be due to any of these changes, which have not been explored in this study. Such factors could also explain the increases in hospitalization rates that were noted in hospital admissions prior to the expansion of CSG development activities.
In these data, increases in hospitalization rates in the CSG area compared to the CHI
and RLI areas were observed for
‘Blood/immune’ and
‘Eye’ diseases for unadjusted models. Adjusted models showed increases in hospitalization rates in the CSG area compared only to the RLI area (
‘Neoplasms’ and
‘Blood/immune’ diseases). All-age RR estimates were greatest for
‘Blood/immune’ disease-related admissions in the CSG area. Admissions within the
‘Blood/immune’ chapter include sub-chapters such as
‘aplastic and other anemias’,
‘coagulation defects’,
‘hemolytic anemias’,
‘nutritional anemias’, and
‘purpura and other hemorrhagic conditions’. However, in absolute terms, admissions from this ICD chapter accounted for only 1.01, 0.52, and 0.79 % of each area’s total admissions for the CSG, CHI, and RLI areas, respectively (refer to Additional file
1).
The previously mentioned study by Coons & Walker [
33] used Diagnostic-Related Groupings (DRG), whereas ICD chapters were used for this study. Therefore, the results are not directly comparable across all categories. For example, there is no equivalent
‘Blood/immune’ DRG that was used in the Coons & Walker study, only a
‘Red cell/clotting’ DRG category, which showed that rates decreased steadily over time in Garfield County. Likewise, the study by Jemielita et al. [
32] did not find any significant associations for the
‘Hematology’ category.
These findings are dissimilar to those presented here, even after adjusting for covariates. Diseases from this ICD chapter (e.g., anemia and other blood disorders) have been discussed in the UNGD literature in relation to worker health and exposure to benzene, toluene, ethylbenzene, and xylene (BTEX) [
2]; however, such discussion is lacking in terms of community health. Generally, long-term exposure to benzene most often affects the blood, and such exposure can also affect the immune system [
39], for which such outcomes are found in the
‘Blood/immune’ chapter. The most common route of exposure to BTEX is through inhalation, typically through air contaminated by motor vehicle emissions and industrial use, as well as cigarette smoke [
40]. While BTEX compounds are naturally occurring and can be found in some water sources, the Queensland Government now has laws in place that ban the use of such compounds in hydraulic fracturing fluids [
41].
Sub-chapters within the
‘Eye’ ICD chapter include
‘disorders of the eyelid, lacrimal system and orbit’,
‘disorders of conjunctiva’,
‘disorders of lens’,
‘glaucoma’,
‘disorders of vitreous body and globe’, and
‘visual disturbances and blindness’, amongst others. In relation to the Coons & Walker [
33] study that used the DRG category for diseases of the eye, hospitalization rates were lowest in Garfield County [
33]. Additionally, there were no significant findings within the
‘Ophthalmology’ category used by Jemielita et al. [
32] Both of these findings are in contrast with the results from the unadjusted models for the
‘Eye’ disease-related hospitalization rates presented here; however, the results are similar (i.e., no significant findings) after adjusting for covariates.
Numerous studies have raised the issue of eye-related symptoms, such as burning, irritation or itching, associated with UNGD [
3,
42‐
45]; however, these studies have discussed outcomes in terms of self-reported symptoms rather than eye-related diseases for which a person would be admitted to hospital. In discussing UNGD operations, Brown et al. [
30] noted that short-term exposure to volatile organic compounds can irritate the eyes, and exposure to diesel emissions can also cause eye irritation. The data presented here would capture the most severe cases rather than residents reporting the symptoms that have typically been discussed in the literature.
In terms of chemicals affecting these systems, Colborn et al. [
46] assessed chemicals used in UNGD operations and found that more than 75 % of the chemicals assessed can affect sensory organs such as the eyes. Likewise, 40 % of chemicals can affect the immune system and 46 % can have possible health effects on the cardiovascular system and blood [
46]. However, it must be noted that this analysis focused on chemicals used in UNGD operations in the United States, which also includes shale gas. Hence, it was unclear whether chemicals used specifically in CSG operations were included in the analyses.
This study has several limitations. Firstly, the study is limited by the ecologic approach used and the associated possibility of ecologic fallacy. While individual, episode-based hospital admissions data were provided, the data were grouped according to the three geographic areas (classified by varying levels of environmental impact), which served as a proxy for unmeasured exposures. Grouping of data in this manner limited the analyses, and represented the smallest aggregations of geographic areas that were allowable by Queensland Health, due to privacy and confidentiality concerns. The unadjusted models showed there were increases in specific hospital admissions in the CSG area relative to the other two study areas; however, these increases were modest and RR estimates were generally small and confidence intervals generally narrow. After adjustment for all covariates, increases in admissions in the CSG area were significantly higher compared only to the RLI area for certain outcomes. Due to small sample sizes for some of the ICD chapters within given study areas, these results should be interpreted cautiously.
The hospital admissions database represents the highest level of morbidity data available, meaning that any data below this (i.e., General Practitioner (GP) or Emergency Department visits) were not captured. There is also lack of data on the percentage of people who do not seek health care in the three study areas, so true rates of health impact are likely to be underestimated. In addition, hospitalization data are episode-based and not person-based, hence, repeat admissions were included in this dataset. Therefore, a resident could have been admitted for the same primary diagnosis on more than one occasion within the same year.
Additionally, it is possible that residents moved from one area to another, moved out of the area entirely, or died, which could result in measurement errors. While we obtained hospital admissions data only on residents of the three study areas to exclude admissions of non-resident workers, it is possible that non-resident workers were included in population enumeration for these areas, depending on a worker’s interpretation of ‘usual residence’ [
25,
47]. Considering this, the rates could be underestimated for residents of the three study areas. While empirical data on the impact of fly-in, fly-out workers on health services is lacking [
48], a recent report found that non-resident workers did have a significant impact, with up to 30 % of health service presentations coming from non-residents [
49]. This may not be applicable to all mining communities; therefore it was suggested community-specific analyses be conducted by collecting and including the home address postcode for each patient, along with diagnoses [
48]. These are the methods that were used in our analyses, although data were collected by broader home address groupings for each patient due to limitations previously addressed.
Certain indicators have been linked to poorer health such as income, household size and overcrowding, and education attainment [
50,
51]. Indigenous Australians also have poorer health outcomes, with one of the highest levels of health inequality compared to Indigenous groups [
52] and a disproportionate level of chronic disease compared to non-Indigenous Australians [
53]. While our adjusted model controlled for factors that were available across the entire time period (e.g., proportion employed full-time was available across all Census years, but a uniform measure of education attainment was not), these were ecological adjustments for the demographic and socioeconomic factors given and were necessarily based on the geographic unit of analysis provided in the hospital admissions data.
The analyses presented in this paper do not allow for conclusions that CSG is a cause of any of the increased hospitalization rates reported. The present study was a descriptive-analytic study employing ecologic units of analysis using routinely available health indicator data. This study provides a preliminary assessment of hospitalization rates and serves to generate hypotheses for future research. As such, the results presented here suggest areas that should be explored further with more sophisticated study designs, and using higher resolution data (e.g., Emergency Department presentations, presentations to GPs) than what could be obtained for this study. Additionally, CSG development in Australia is a contentious issue [
1,
17,
54], and much of the data that are collected are predominantly used for legislative compliance purposes rather than monitoring and research purposes [
17]. Calls have been made for more environmental data, as well as health data, that are publicly available in a common repository [
17].
Further examination of these hospitalization data to determine trends over time in age-specific and gender-specific rates of the health conditions potentially related to CSG is recommended. It would be useful to include data on the working population of an area, where study time periods allow for inclusion of such data, to better understand populations with a high proportion of fly-in, fly-out workers. In addition, analyses on specific diseases identified in previous literature for which potential health outcomes may arise should be examined. Further research using robust methodology is required to investigate the potential causal association between CSG and the potential adverse health outcomes presented here.
Authors’ contributions
AKW and KW contributed to the design of the study, statistical analysis, and interpretation of the results. AKW had primary responsibility for the manuscript. CMC contributed to the statistical analysis, interpretation of results, and manuscript revisions. SV and PJ contributed to the design of the study, overall editing, and manuscript revision. AP contributed to statistical analysis, interpretation of results, overall editing, and manuscript revision. All authors have read and approved the final version of the manuscript.