Research and analysis

INWORKS2: an update to the INWORKS study

Updated 18 December 2023

Background

The International Nuclear Workers Study (INWORKS) is an epidemiological study of workers in the nuclear sector launched in 2011 and coordinated by the International Agency for Research on Cancer (IARC) which combines data from nuclear workers in the UK, US and France for the purpose of undertaking a pooled analysis. It seeks to gain greater knowledge relating to the risks of cancer and non-cancerous diseases linked to chronic exposure to low doses of ionising radiation at low dose rates.

The UK component of the first pooled data set was drawn from the NRRW-3 cohort. The final paper from the INWORKS team based on this data set was published in 2021.

Since the first INWORKS study data set was compiled in 2011, each of the contributing studies has been updated with at least 10 years of additional follow-up. In 2022, new INWORKS2 data set was compiled including the extra follow-up information from each component study.

Study population

The number of deaths in the INWORKS2 data set has increased by 56% to 103,000 while the number of person-years of observation has increased by 3 million to 10.7 million compared to the original data set.

The average period of follow-up for the workers has increased from 27 years to 34 years while the average cumulative radiation dose per worker has decreased from 25mSv to 18mSv, reflecting the much lower annual doses received by workers in more recent years.

INWORKS2 data set characteristics	France	UK	US	INWORKS2
Calendar years of follow-up	1968 to 2014	1946 to 2011	1944 to 2016	1944 to 2016
Workers	60,697	147,872	101,363	309,932
Person-years (millions)	2.08	4.67	3.98	10.72
Males	1.80	4.27	3.17	9.24
Females	0.28	0.40	0.81	1.48
Deaths (all causes)	12,270	39,933	51,350	103,553
All cancer	4,885	12,556	13,568	31,009
Average duration of follow-up (years)	34.2	31.6	39.3	34.6
Average individual cumulative dose (mGy)	12.9	20.19	16.8	17.7

Methods

The same standard statistical modelling techniques are used in INWORKS2 as were previously used. The dose-response relationship between external radiation exposure and mortality rate was studied for a range of causes of death including the solid cancer, haematological cancers, and non-cancer outcomes.

All cancer and solid cancer results based on the INWORKS2 data set

The first analysis using the INWORKS2 data set examined risk from the groupings of all cancer and solid cancers and was published in the British Medical Journal in August 2023.

Cancer mortality after low dose exposure to ionising radiation in workers in France, the United Kingdom, and the United States (INWORKS): cohort study

The overall results were similar to the previous analysis. There was reasonable evidence for a linear association between solid cancer risk and cumulative dose over the whole dose range. The estimates of the slope of the linear association were slightly higher than for the previous analysis with narrower confidence intervals reflecting the greater information available from this data set. For all cancers as a group, the slope estimate was 0.53 per Gy (90% confidence interval 0.30 to 0.77), while for all solid cancers as a group, the slope estimate was 0.52 per Gy (90% confidence interval 0.27 to 0.77) for solid cancer mortality. The slope estimate for the group of solid cancers other than lung cancer was the same as found in the original analysis but with a narrower confidence interval. No one-country data was found to influence these results to a large degree.

For the grouping of all solid cancers, deviations from a linear association were examined and in contrast to the previous analysis, there was modest evidence for a downward curvature in risk per unit dose with increasing dose (above 200mSv) either by including a quadratic or exponential term.

A recent NRRW publication, based on cancer incidence data (Hunter and others, 2022), found a similar downward curvature in the all solid cancer dose response. However, this was demonstrated to be heavily influenced by the result for lung cancer. When lung cancer was excluded, the best fitting dose response was found to be linear. The INWORKS2 authors found the same thing when lung cancer was excluded. 

In the NRRW cancer incidence study, further work was undertaken, looking at the dose response for all solid cancers excluding lung and separately for lung cancers when dose response relationships were fitted separately for workers who were only ever exposed to external radiation and those who were additionally monitored for potential for internal radiation contamination. In the NRRW analyses, there was good evidence for a linear dose response relationship with cumulative dose for both lung cancer and solid cancers, excluding lung for those workers who were only monitored for external exposure but no evidence for any dose response among those workers additionally monitored for internal exposure. In the INWORKS2 analysis, a similar pattern was seen for all solid cancers excluding lung, but the authors did not perform this analysis separately for lung cancer.

INWORKS2 also examined how the dose response trend changed when the analyses were restricted to the lower cumulative dose range. The linear relationship for all solid cancers was refitted over a progressively lower dose range. As the dose range decreased, the slope of the linear relationship increased considerably. Even as low as 50mGy, the slope was still statistically significant and over double that of the main result but with wider confidence intervals indicating the increasing uncertainty in these results.

Several issues may explain these results. Firstly, dose measurements for early workers are thought to be far less reliable than in later years, so this difference could be partly due to inaccuracies in the dose measurements. Also, the healthy workers survivor effect whereby only health workers are in employment long enough to accumulate large lifetime doses might also impact on these results as early workers could accumulate greater doses in a shorter period than later workers could. It is also possible that early workers were potentially less healthy overall due to military conscription during World War II, while later workers are likely to be healthier due to general improvements in the health of the populations and increased medical surveillance of the workforce.

The authors have tried to examine the impact on the risk estimates of early workers by performing separate analyses for workers who were first employed prior to 1958 or 1965 and those first employed later.

For the later workers they found risks comparable to those found when examining all the workers but over the reduced dose range (<100mSv) but for the early workers much lower risk per unit dose was found. Such a difference was also found between earlier and later employed workers in a separate analysis of the US component of INWORKS (Kelly-Reif and others, 2022).

It is premature to suggest that this study provides good evidence that the assumption that a linear dose response is no longer the most appropriate general description of the relationship between cancer risk and cumulative dose. However, the INWORKS2 study does provide some interesting new evidence about variation in overall cancer mortality risks that needs further investigation to understand if or how it really differs from the evidence from the atomic bomb survivors that national and international radiation protection organisations use as the basis for their recommendations for worker and public protection.

All these issues will be investigated further in the fourth analysis of the NRRW (NRRW4). The NRRW4 is expected to contain around two-thirds of the number of cancer deaths used in this INWORKS2 analysis but will have the advantage of containing cancer registration information. The total number of registrations in NRRW4 could well exceed the number of cancer deaths in INWORKS2. This gives it the potential to be more informative. Furthermore, for most of these 30,000 workers in NRRW who have been monitored for internal exposure, NRRW4 will have dose estimates based on urinalysis data and not just a monitoring indicator allowing the impact of these doses to be evaluated. It will also have an additional 10 or more years of follow-up and increased numbers of workers.