Article Text
Abstract
Objectives Lung cancer caused by radon in miners is a well-known risk. However, the risk estimates vary between studies and between mines. We have studied the dose response–relationship in a Swedish iron ore mine where two other studies have previously reached different risk estimates. As this mine has relatively low radon levels, the results are highly relevant for risk estimation in non-uranium underground mines.
Methods A new cohort of 5486 male workers employed from 1923 to 1996 was established. Cumulative radon exposures were assessed based on a large number of measurements, including reconstructions of historical conditions. 122 lung cancer cases occurred during the follow-up period of 1958–2000.
Results The average cumulative exposure in underground workers was 32 kBq year/m3 (65 working level months (WLM)), experienced over 14.6 years. The excess RR (ERR) per kBq year/m3 was 0.046 (95% CI 0.015 to 0.077; 0.022 ERR/WLM). Confounding by quartz may affect these results but appears to account only for 10–20% of the risk. The results for squamous cell and small cell lung cancer were 0.049 and 0.072, respectively. However, no increased risk was observed for adenocarcinoma (0.000 ERR per kBq year/m3, 95% CI −0.017 to 0.017).
Conclusion Our overall risk estimate is about half of that found in the first Malmberget study but twice that found in the same cohort in the previously published pooled analysis. Radon did not increase the risk for adenocarcinoma in the lung.
- Lung cancer
- mining
- radon
- underground
- non-ionizing radiation
- epidemiology
- cancer
Statistics from Altmetric.com
What this paper adds
The risk estimates for lung cancer in underground miners due to radon vary between studies and mines; in the case of the Malmberget mine, there are considerable differences between analyses of the same cohort.
Using a new cohort and a new exposure assessment, we still see a high excess RR in the Malmberget mine, with values between those of the first study in the mine and a pooled analysis of the same mine.
Quartz exposure appears to contribute to a limited extent to lung cancer risk.
No increased risk for adenocarcinoma was seen when different cancer subtypes were analysed.
The risk estimates indicate a substantial risk at radon levels below the occupational exposure limits.
Introduction
The association between lung cancer and radon exposure is well established in underground miners.1 However, the dose–response relationship varies between studies and between mines.2 The dose rate is high in uranium mines, while it is lower in, for example, iron mines. The magnitude of the lung cancer risk in underground mines is relevant for the protection of miners.
An often cited pooled analysis of 11 cohorts of miners included workers from eight uranium mines. The iron mine in Malmberget, Sweden had the lowest radon concentration; historical radon progeny levels were estimated to be 2–7 WLM (working level months) per year.3 Studies of miners from this mine are therefore very important for estimating risk in non-uranium miners.
There is a discrepancy in the risk estimates obtained from miners in this mine between the first study published in 1984 by Radford and Renard3 and the report published in 1995 by Lubin et al2 using basically the same exposure assessment and cohort (the excess RR per working level month (ERR/WLM) was estimated to be 0.036 and 0.0095, respectively). The exposure assessment in the first study3 has been questioned in a report to the American Mining Congress4, in which the authors claimed that the historical exposure must have been higher than estimated, leading to an overestimation of the RR coefficient.
To further examine these contradictory results, we defined a new cohort of miners, independent of earlier studies, and used all available radon measurements for a new exposure assessment.
Methods
Exposure assessment
The iron ore mine was established in 1890. The mine consists of several iron ore bodies (see online figure A1) with somewhat different geology. The bedrock in the western part of the field contains about 4–6 ppm uranium, while the south-eastern area contains 2–3 ppm uranium.
In the early 20th century this mine was mainly open cast, but by 1920 approximately 80% of mining took place underground. Today, most mining is carried out about 1000 m below ground level.
Originally the mine only had natural ventilation, but forced ventilation was gradually introduced, especially during the 1950s and 1960s. Fresh air was preheated before being passed through rock shafts backfilled with waste rock (in this region snow lies on the ground from about September to May, and temperatures between −20°C and −30°C are common in winter). The disadvantage was that the fresh air became contaminated with radon emitted from the waste rock, walls and groundwater. Therefore, the early ventilation systems did not reduce radon levels. This contamination was detected in 1969 and a new ventilation system was installed in 1972–1973, using closed fresh-air channels.
Radon measurements
The Swedish Radiation Protection Authority began measurements of radon in 1969 in collaboration with the mining company. In 1972, the company started taking their own measurements. Personal dose measurement started in 1991. Details of the methods are given elsewhere.5 6
Our exposure assessments were based on data from 5481 radon measurements carried out during 1969–1998. Figure 1 shows the mean radon gas activity concentration for the mine. The estimated average radon activity concentrations for the periods 1969–1972 and 1974–1998 for the different ore bodies are shown in online figure A2.
The measurement strategy was a combination of worst case sampling, evaluation of effects following measures to reduce exposure, and monitoring of the general background exposure in the mine. This strategy probably resulted in more measurements from areas of the mine with high radon levels, as Swedish radiation legislation prescribed frequent measurements in environments where high radon levels were suspected. Therefore, the measurement strategy used in Malmberget probably resulted in an overestimation of the radon exposure. No attempt to adjust the exposure assessments was made due to lack of information concerning the magnitude of this bias.
To investigate if the radon concentrations during 1969–1972 could be regarded as representative of previous years, 120 measurements were carried out in five abandoned unventilated parts of the mine. The radon concentrations in these areas were expected to be the same or higher than when these areas were mined. Measurements were carried out with a continuous recording instrument over 1 or 2 consecutive days and with a nuclear track detector over 3 weeks. Measurements were performed when there was little inflow of ground water (March–April) and when there was a large inflow of ground water (May–June). The results from one area were disregarded due to few measurements. Data from the four remaining areas showed no systematic deviation from the data originating from 1969–1972 (see online figure A3). The mean difference among the 116 measurements was 14% higher than the corresponding data for 1969–1972.
Modelling of radon activity concentration
A theoretical calculation of radon activity concentration was performed with the assumptions that radon is emitted from rock surfaces and there was no ventilation in the mine.7 The calculation did not take into account radon inflow through water. The results showed a maximum concentration of 4000 Bq/m3 in the air in a drift in red vulcanite where the uranium concentration in the rock was 5 ppm (60 Bq/kg) (data not shown).
Exposure matrix
All the historical and reconstructed radon measurements were used to construct an exposure matrix. The exposure matrix was specified over time for different ore bodies. The radon levels in the matrix were set as constant from 1974 until 1998. The rationale for this is that there was no clear trend over time in the measurements and the variation from year to year was small compared to the period before 1974 and between different ore bodies.
For 1973, the radon level was set as equal to the mean of all data from that year. As the measurements in the old areas of the mine were comparable with data from 1969 to 1972, we assumed that these levels were representative of 1925–1969. Further support for this assumption is given by the theoretical calculations. The level in 1910 was set to 380 Bq/m3 followed by a linear increase from 1910 until 1925. By 1925, almost all mining activity had moved underground. The level of 380 Bq/m3, based on measurements under fairly similar conditions in another mine containing less uranium, is probably not a very accurate estimate. However, the level chosen only affects the exposure assessment marginally because only 59 person-years with radon exposure occurred before 1925. Radon exposure was set to zero before 1910. The estimated radon concentration for five ore bodies over time is shown in online figure A4.
Quartz exposure
Respirable quartz has been measured by the company since 1965 and an exposure matrix for quartz was established as described elsewhere.8
Study cohort
The cohort consisted of surface and underground workers who had worked at the mine between 1923 and 1996. Information concerning employees was collected from manual work records for the years 1923–1981 and from a computerised register available since 1982. All but 53 individuals were identified using the unique personal identification number assigned to each Swedish resident. Combining the two sources and restricting the cohort to male workers who had worked for at least 1 year at the mine, resulted in 5588 men being included in the cohort.
The cohort in this study was constructed from company records and not based on the previous Malmberget cohort which consisted of 1294 workers.3 We are not aware of the identity of those workers but probably most of them were also included in the present study.
Detailed job histories were registered for 5588 men. Each worker usually had different types of jobs giving a total of 23 192 work periods. Each period included date for the beginning and end of the period, type of occupation and ore body (for about 50% of the time periods). Almost all periods in the cohort could be assigned information about surface or underground mining (99.97%), while surface mining was assumed for the remaining periods.
By combining information about each work period with the exposure matrix, the individual cumulative radon dose was calculated for each calendar year of underground work. The radon dose was set to zero for periods of surface work. If a period was a combination of surface and underground work, the radon level was estimated to be 50% of the level for underground work.
Dates of death for the period 1952–2000 were obtained by record linkage, using the worker's unique personal identification number, to the national cause of death register. Records were also linked to the national population register which contains information about current residence to determine if workers were currently living in Sweden. The 101 men not found in either the cause of death register or the population register were excluded. These people had either died before 1952 or emigrated from Sweden. One man registered as dead before his first occurrence in the company records was also, excluded leaving 5486 men in the cohort.
Lung cancer morbidity and information about histological type were obtained for the years 1958–2000 by record linkage to the national cancer register. Therefore, 37 individuals not alive at the start of 1958 were excluded from the study, giving a final number of 5449.
Information on smoking habits was available for workers employed during 1972–1992 and participating in health examinations as part of an environmental health project. Data on smoking for 2310 workers were available.
Statistical methods
For all risk calculations, cumulative exposure was lagged by 5 years. The cumulative radon dose was divided into five categories, with approximately the same number of cases in each (0, 0–40, 40–65, 65–95 and ≥95 kBq year/m3). Data were cross-classified by radon dose, calendar year and attained age, and for each strata the number of lung cancer cases and person-years were computed. For each category of exposure, the mean exposure was calculated by summing the cumulative exposure for each year divided by the total number of person-years. To adjust for cumulative quartz, we used four categories (0, 0–2, 2–5 and >5 mg years/m3) and data were cross-classified in the same manner as above.
Relative risks (RR), in categories, were calculated by (i) the ratio between the observed number of lung cancer cases in the cohort and the expected number of cases based on age and calendar year specific population incidence rates (indirect standardisation, SIR) and (ii) using the non-exposed strata of the cohort as an internal reference group. The SIR computations were adjusted for calendar year between 1958 and 2000 and attained age in 5-year groups. Incidence rates from the northern region of Sweden (the four northernmost counties) were used for the calculations of expected number of cases. For estimation of RR (ii), a log-linear Poisson model, adjusted for age and calendar period, was used.
To estimate excess RR per unit radon increase, a linear excess RR model, adjusted for age and calendar period, was used.9 The latter model was also used to estimate a more extended model used in BEIR VI1 taking time since exposure and exposure duration into account. For details see online appendix.
S-plus10 and Epicure9 were used for the statistical analyses.
Conversion between units
Historically, radon progeny has often been focused and mostly expressed as working level (WL), and cumulative dose has been expressed as working level months (WLM).11 We have expressed radon exposure as the activity concentration of radon gas in air (Bq/m3) and cumulative exposure as kBq year/m3.
Based on parallel measurements of radon gas and radon progeny in the Malmberget mine, an equilibrium factor (sometimes referred to as an F-factor11) of 0.7 for conversion between radon gas and progeny was calculated. When we investigated the equilibrium factor in two different iron ore mines at different time periods, our estimates ranged from 0.52 to 0.81 but with no apparent time trend. Therefore, the uncertainty introduced by the choice of 0.7 as factor has only a minor impact on the final risk estimates. This factor was used in all conversions of radon progeny data to radon concentration.
At radioactive equilibrium (when the equilibrium factor is 1), WL correspond to a radon progeny concentration of 3700 Bq/m3. With an equilibrium factor of 0.7, the level of 1 WL corresponds to a radon gas concentration of 5286 Bq/m3. A cumulative exposure of 1 WLM corresponds to an exposure level of 1 WL over 166.67 h.11 Assuming a yearly work time of 1800 h, 1 WLM in this study corresponds to 0.489 kBq year/m3 (given an equilibrium factor of 0.7).
Sensitivity analysis
A sensitivity analysis was performed by modifying the exposure matrix. The radon level during 1925–1950 was increased by 50% and by 100%. The values for 1910 and 1969 were unchanged and for 1911–1924 and 1951–1968 linear interpolation was used. The reason for this analysis was to investigate how the risk estimates would be affected by significantly higher exposure levels before forced ventilation became common.
Results
Most workers were employed before 1970 (75%) (table 1).
Among the 5449 men in the study, 3597 had been exposed to radon. The average cumulative radon exposure of underground miners at last year of observation was 32 kBq year/m3 and the maximum was 325 kBq year/m3. Mean age at beginning underground work was 27 years (median 25) and the average time from first to last underground work period was 14.6 years.
In total, 122 lung cancer cases found in the cancer register for the follow-up years of 1958–2000. Eight of those cases were known non-smokers. Age at diagnosis varied between 29 and 85 years. The lung cancer risk in the group with no work underground was similar to that in the general population (SIR=1.00). There was a dose–response relationship between lung cancer risk and radon exposure (table 2). The SIR value for the highest dose category (272 WLM on average) was 4.27. Similar RR estimates were obtained using an internal reference group of non-exposed workers (table 2, figure 2). The excess RR was 0.046 per kBq year/m3 (95% CI 0.015 to 0.077) (figure 2).
When we applied the BEIR VI model, ERR/kBq year/m3 was estimated at 0.037 (95% CI 0.021 to 0.062) without adjustment for calendar period. Adjusted for calendar period, the estimate was 0.050 (95% CI 0.029 to 0.089).
In the cancer registry, 51 of the 122 cases were classified as squamous cell carcinoma, 55 were small cell cancer and 12 adenocarcinoma. Four cases were of other or unknown histological type. Squamous cell and small cell cancer showed a dose–response relationship (SIR), although the risk tended to level off at higher radon doses (figure 3). There was no indication of an excess risk for adenocarcinoma. The dose–response patterns for squamous cell and small cell cancer were quite similar using internal comparison in the log-linear model (not shown). The estimates of excess RR per kBq year/m3 were 0.049 (95% CI −0.003 to 0.102), 0.072 (95% CI −0.003 to 0.147) and 0.000 (95% CI −0.017 to 0.017) for squamous cell cancer, small cell cancer and adenocarcinoma, respectively.
Lung cancer was also studied in workers first exposed in 1973 or later, corresponding to the period after installation of the improved ventilation system which led to decreased radon levels. No worker was exposed to more than 10.6 kBq year/m3 during this period. No lung cancer case was observed in this subgroup (1.2 cases expected, SIR 0, 95% CI 0 to 3.04).
To examine if there was an association between smoking habits and radon exposure, we compared the radon exposure of smokers with non-smokers. Data regarding smoking habits were available for 945 smokers and 1365 non-smokers for 1972–1992. Overall, 75% of the smokers and 71% of the non-smokers had worked underground. Thus, 42% of the radon exposed and 37.5% of the unexposed men were smokers. Among underground workers, the mean radon dose was 28.3 and 27.7 kBq year/m3 for smokers and non-smokers, respectively. The average year of birth was 1936.8 and 1938.5 and the age at last exposure was 45.0 and 44.6 for smokers and non-smokers, respectively.
A sensitivity analysis showed an ERR/kBq year/m3 of 0.036 (95% CI 0.011 to 0.060) when the exposure levels for 1925–1950 were assumed to be 50% higher, and when raised to 100% over the assumed level the estimate was 0.031 (95% CI 0.010 to 0.052).
When the radon excess RR was adjusted for cumulative quartz together with attained age and calendar period, the estimate decreased from 0.046 to 0.031 ERR/kBq year/m3 (95% CI −0.009 to 0.070).
Discussion
Analysis of the 11 published cohorts gave an excess relative lung cancer risk ranging from 0.006 to 0.6 ERR/WLM. Our estimate of 0.022 ERR/WLM lies within this huge interval. It also lies between the two estimates for this particular mine as reported by Radford and Renard3 and Lubin et al.2 An excess lung cancer risk of radon exposure was found for small cell and squamous cell neoplasms but not for adenocarcinoma.
The present study is based on a very well defined cohort and with complete follow-up thanks to the unique personal identification numbers and the Swedish population and cancer registries. The Swedish cancer register has also made it possible to investigate the different histological subgroups of lung cancer.
The strength of our study compared to the previous study of a Malmberget cohort, is that we included unexposed workers. We could thus compare exposed workers with both unexposed colleagues and the general population in the region. Similar results were obtained indicating that the cohort members, at least those who were unexposed, had risk factors for lung cancer (eg, smoking habits) that were similar to those among the general population in the region. We did not study the relationship between lung cancer and radon exposure in non-smokers, since there were only eight workers in this group and dose–response calculations would therefore be very imprecise. A possible co-variation between smoking habits and radon exposure was investigated: data for the miners with known smoking habits showed that smokers and non-smokers had similar radon exposure, indicating that smoking is not a strong confounder in the study.
Exposure assessment and risk estimate
No radon measurements were obtained before 1969. We have investigated if it is plausible to apply measurements for 1969–1972 to previous years. Measurements in old areas of the mine as well as theoretical calculations indicate that radon levels prior to 1969 are likely to be of the same magnitude as levels for 1969–1972.
We used all available data from two different sources (the company and the Swedish Radiation Protection Authority) in exposure assessment. Measurements derived from these two sources gave very similar results.6 Moreover, for half of the employed miners individual information about which ore body the miner had worked in were available. However, there might be a bias in the exposure assessment and this has probably caused an overestimation of the radon activity concentration (see the Methods section).
We assessed the radon activity concentration at 2900 Bq/m3 (6 WLM/year) for a miner working in an unspecified ore body any time between 1925 and 1972. For a similar worker, Radford and Renard3 assessed the exposure at approximately 5–7 WLM/year, except for one ore body where they estimated the level to be 2 WLM/year (data extracted from figure 1).3 These two exposure assessments appear to be fairly similar.
This similarity in exposure assessment is reflected by a similar risk estimate. Our estimate of excess RR is 0.046 m3/kBq year, which corresponds to an ERR/WLM of 0.022 (0.046×0.489). The ERR/WLM result of Radford and Renard3 was 0.036, and that of Lubin et al2 0.0095. Lubin et al2 used the lowest exposed group as a reference, which had more than a doubled risk of lung cancer compared to the general population. The risk estimate was based on eight cases, making the risk estimate uncertain.
So, which of these risk estimates is the most accurate? The strength of our study is that it is based on a larger cohort, relies on a thorough exposure assessment and utilises both internal and external (SIR) comparisons. Lubin et al2 pooled the previous study by Radford and Renard,3 and this led to the choice of the lowest exposed group as reference group. That group, however, had an increased risk of lung cancer (standardised mortality rate (SMR)=2.4) which of course influenced the final risk estimate (see Lubin et al2 and Radford and Renard3 for details). Although the pooled analysis of Lubin et al2 made an admirable contribution to the radon risk assessment, the choice of control group probably biased their risk estimate for the Malmberget cohort towards the null. The main weakness of the present study (ie, the lack of exposure data before 1969) is common for all the studies in Malmberget.
The validity of the exposure assessment in the study by Radford and Renard3 has been criticised.4 The authors concluded that the exposure assessment prior to 1969 was underestimated. The report was never scientifically published but is sometimes cited in reviews and consensus reports. Our results show that radon exposure before 1969 was not much higher than between 1969 and 1972. Furthermore, a sensitivity analysis with 100% increased levels during 1925–1950 and a gradual decrease during 1950–1969 towards the average levels of 1969–1972, showed moderate effects on the risk estimates. Our risk estimate of 0.031 m3/kBq year corresponding to an ERR/WLM of 0.015 did not reach the level of 0.0095 ERR/WLM.2
Our risk estimates are high compared to other cohorts of miners. In the pooled analysis of BEIR VI, the overall risk estimate (11 cohorts), after adjustments for time since exposure, attained age and duration of exposure, was 0.0055 ERR/WLM. When we used the BEIR VI model, we got 0.034 ERR/kBq year/m3, corresponding to 0.017 ERR/WLM, that is about three times higher.
Our conclusion is that the higher lung cancer risk estimate in our study and in previous studies of Malmberget miners compared to other studies is not affected by our exposure assessment. A possible explanation is that low-level radon exposure with long duration gives a larger relative increase in risk per unit exposure than mines with higher radon concentration.12 Exposure to diesel exhaust and dust may have an effect.
Confounding by quartz
Positive confounding by quartz exposure is in accordance with earlier studies showing similar effects after adjustment for other exposures in the mine environment, including silica dust and arsenic.13 14 In the present study, the workers were exposed before the mid-1970s to levels of respirable quartz above the present Swedish exposure limit of 0.1 mg/m3. The effect of quartz exposure on the risk estimate for radon has been studied in the mines of Malmberget and the nearby mine in Kiruna.8 Bergdahl and coworkers concluded that exposure to quartz had an impact on the risk estimates of lung cancer. We adjusted for quartz in our calculations. Since there is a strong association between cumulated quartz and radon exposure in the Malmberget cohort, the adjustments may be incorrect as they may overadjust or underadjust the risk estimate depending on the degree of misclassification of the inter-related exposures. The adjustment decreased the risk estimate for radon gas by about 30%. This effect is somewhat larger than would be expected from quartz exposure. For further discussion on the relationship between lung cancer and quartz, please refer to Bergdahl et al.8 Given the unexpectedly high lung cancer risk ascribed to quartz by this model, it can be argued that the quartz adjusted model, due to intercorrelation of quartz and radon exposure, ascribes too much lung cancer risk to the quartz exposure and, as a consequence, too little lung cancer risk to radon exposure. Therefore, we conclude that the best estimate of the radon modelled risk should be decreased with less than 30% compared to the model not adjusted for quartz. Thus a decrease in the ERR of 10–20% for quartz confounding appears reasonable.
Histological subtypes
The proportion of small cell lung cancer among the miners in this study was high compared to the general population in Sweden and other European countries.15 Other studies of radon exposed miners have also shown an increased proportion of small cell lung cancer.16 17 When histological subtypes were studied at low radon exposure (ie, in homes), a significantly increased risk appeared only for small cell lung cancer.18 The estimated linear risks in the present study were similar for squamous and small cell cancer, which is in accordance with a cohort study of uranium miners in the Czech Republic.19 Tomasek et al studied miners with a relatively short exposure time (≤8 years) and found differences in risk between small cell lung cancer and squamous cell lung cancer if time since last exposure was included; small cell cancer typically occurred earlier than squamous cell lung cancer. The miners in our cohort have on average longer periods of exposure and lower dose rates. The magnitude of the excess RR estimates for these cell types was 0.049–0.072 per kBq year/m3 (0.024–0.035 ERR/WLM) in our study and 0.024–0.029 per WLM in the Czech study, that is the estimates are similar.
We see no increased risk regarding adenocarcinoma related to radon exposure. We conclude that radon does not appear to cause an increased risk of adenocarcinoma in the lung among Malmberget miners (based on 12 cases). However, more studies must be performed to verify our conclusion regarding adenocarcinoma.
Public health aspects
The aetiological fraction for lung cancer due to underground mining was 57% among the miners in the present cohort. For miners in the highest exposure category (>95 kBq year/m3), the aetiological fraction was 77%, that is more than three out of four lung cancer cases could have been prevented if there had been no work underground. In the lowest exposure category (0–40 kBq year/m3), the aetiological fraction was 9%, indicating that a relatively low exposure has an impact on miners' health. Lung cancer risk in a population depends greatly on tobacco smoking prevalence. Lung cancer was the underlying cause of death in approximately 4% of Swedish men who died at 45 years of age or older in 2002. Miners within the highest exposure category had a RR for lung cancer of 4, meaning that 11% of all deaths among them can be ascribed to lung cancer, assuming that all lung cancer cases are fatal. The Swedish occupational exposure limit (OEL) for radon gas in underground workplaces allows a yearly dose of 2.5 MBq h/year m3, which corresponds to 1389 Bq/m3 assuming the miners work 1800 h per year. From our results, we conclude that radon exposure at half of the Swedish OEL over 30 years gives a RR of 1.7. This RR corresponds to an aetiological fraction of about 3% for death from lung cancer if this disease is the cause of death in 4% of non-exposed individuals. If the risk for lung cancer is higher among the non-exposed population (ie, there is a higher proportion of smokers), the attributable risk is higher. Thus, exposure to radon in underground mining has a considerable impact on the health of miners.
Acknowledgments
We thank Anna-Maja Åberg, Karin Andersson, Eva Juslin, Göran Larsson and Katarina Örnkloo for help with the coding and registration of cohort data and radon measurements. We also thank Ulf Hedlund for discussion and valuable comments.
References
Supplementary materials
Web Only Data oem.2009.047449
Files in this Data Supplement:
Footnotes
Linked articles 047456.
Funding This study was financially supported by the Swedish Council for Working Life and Social Research, the Swedish Radiation Protection Authority and the Cancer Research Foundation in Northern Sweden.
The study has been carried out with the permission of the mining company, LKAB.
Competing interests One of the authors was previously employed and one is still employed by the company.
Ethics approval This study has been approved by the Ethics Committee at Umeå University (ref. 03-040).
Provenance and peer review Not commissioned; externally peer reviewed.