14 Comments

  1. Rod, thanks for this. I had to look up abbreviations to refresh my statistics. Tell me if I’m wrong and I’ll fix the comment.

    ERR is excess relative risk. So if normal risk of something is 1%, then with ERR = 2.38 the observed risk is 2.38%.

    p is probability the observation is just noise. So p = 0.24 means that if the experiment were repeated 100 times and there really is no correlation, a similar correlation result would be observed 24 times.

    Gy is gray (not gigawatt-year) which is 1 watt-second of ionizing energy absorbed per kilogram of flesh — a large amount. [Below I assume 1 Sv = 1 Gy for simplicity.]

    The mean dose of 132 mGy is high compared to low level ionizing radiation concerns. The article is concerned with doses as high as 1.4 Gy. Looking at the graph on page 30 the confidence error bars are consistent with ERR = 1.0 up to doses of 500 mGy.

    It’s not clear whether the doses received are chronic or acute. I would think cleanup workers would be exposed to most of the radiation at the beginning of the cleanup. The concerns people have with low-level chronic ionizing radiation are not addressed in this study.

    At Fukushima, 9900 members of the public were exposed to radiation < 10 mGy and none more than 23 mGy. These exposures are way smaller than those in the study.

    The US and ICRP limits public exposure to 1 mGy/year. Jaworowski proposed raising the limit to 10 mGy/year — the level of background radiation as life evolved on earth. Wade Allison proposes 100 mGy/month.

    But the takeaway is that only 19 people, 0.017%, developed leukemia attributable to radiation from the Chernobyl cleanup.

    1. @Robert

      That’s just about perfect.

      The only correction I will offer is that there is an indication in the study that each person received their total doses during a relatively brief period of performing clean up work – the median time in the zone for all workers was 35 days with just 38% of the people in the zone for more than 60 days (two months). The maximum duration of the work inside the zone was 711 days.

      All of that can be found on page 12 of 31.

    2. Note that these results exclude more cases than they attribute to radiation. The authors exclude 20 cases for which the ERR/Gy was negative!

  2. Let me add another couple of translations:
    The 20 cases were excluded because they were “incompatible with our current understanding of radiation-related leukemia risk.”
    = We took out the cases that conflicted with our expectations.

    “However, it must be recognized that our final results derived from a post-hoc subgroup analysis”
    = We cherry-picked.

    I am still mystified, though, as to how these 20 cases with an “unexpectedly” LOW dose were assessed as being a negative ERR. And even more mystified as how taking them out would increase overall ERR, too, since last time I looked zero is less than twenty.

    1. @Joffan

      Can you point to the section that led you to say that the 20 excluded cases had an unexpectedly “LOW dose”?

      I do not think their doses were low; they showed a negative ERR/Gy which indicates to me that they must have had higher than average doses and/or lower than expected excess risk. (Or am I getting confused here?)

      1. page 18: “However, a preliminary examination of differences in various characteristics of participating cases, ascertained using the two methods described above, indicated that cases with direct in-person interviews <2 years from start of chemotherapy treatment had mean bone marrow radiation dose estimates significantly lower than other cases interviewed in-person (16.8 vs.121.4 mGy, 7-fold difference in means, p Wilcoxon=0.036), and these doses were uniformly lower across all types of work performed in the 30-km zone, while the mean doses for controls from both groups were almost identical."

        I think it is intrinsically confusing. I would be surprised (and delighted) to find someone who was not confused.

        1. They’re bad investigators for not looking more closely at this. These directly-interviewed subjects starting chemo within 2 years of being interviewed got lower rad doses than other subjects. Why?

          For one thing, their proxy interviews do seem to exaggerate doses compared to their direct interviews (see my reply to Rod here).

          For another, I’m not sure I believe them when they say there’s no relationship between ERR/Gy estimates and “calendar period of diagnosis”. The calendar periods they consider are ’86-94, ’95-00, and ’01-06… relatively long compared to 2 years. Presumably these recent-chemo folks are the last to be diagnosed. That’s consistent with a low dose giving protection.

          But you’re right, I don’t think any conclusions can be drawn from the work as presented. Aside from issues already mentioned, their confidence intervals look a bit crazy. They used EPICURE, and BEIR warns about how easy it is to misuse.

          -Carl

      2. “…cases with direct in-person interviews < 2 years from start of chemotherapy
        treatment had mean bone marrow radiation dose estimates significantly lower than other cases interviewed in-person (16.8 vs.121.4 mGy …)… while the mean doses for controls from both groups were almost identical."

        A negative ERR/Gy means less risk of leukemia than the general population. To get the relative risk per Gy, add 1 to the ERR/Gy. So -0.47/Gy + 1 = 0.53/Gy, which means: every Gy of radiation cuts your risk of leukemia approximately in half compared to the population at large. Of course that's the relationship only in the low-dose cohort. For the high-dose cohort, it's 1 + 2.38 = 3.38 times more risk per Gy. Which is to say, they observed a nonlinear dose-response.

        However, their data looks pretty crap. All doses were inferred by interview (basically, "how long did you spend in each area?" times the known contamination in each area). And despite them saying "There was no significant difference in ERR/Gy estimates by proxy or direct interviews", the differences do seem significant. From the "S2" supplemental table

        Proxy 69 5.10 (<-0.81, 29.29)
        Direct 68 -0.10 (<-0.38, 1.74)

        The third column is ERR/Gy. Proxy interviews give a much bigger bang than direct interviews. When they say the difference wasn't significant, they mean the p-value wasn't < 0.05. But it's not clear this "p-interaction" calculation is meaningful. And the p-interaction is 0.42 for their cherry-picked group of 117 cases, but is already down to 0.1 for the full 137 cases (see the supplemental table PDF).

        Now, a lot of proxy interviews may have been necessary because the subject had died, and perhaps they were more likely to die if they got a higher dose. So some difference in dose may be expected, who knows. The potential for coworkers and family members to exaggerate after the death of a loved one is also something to consider.

  3. What a find! You can’t make this stuff up: “The ERR/Gy estimates for cases with direct interviews < 2 years from start of chemotherapy (ERR/Gy=-0.47) and the remaining cases (ERR/Gy=2.38) differed substantially (p=0.021), with the former estimate incompatible with our current understanding of radiation-related leukemia risk. ERR/Gy estimates in the former group were negative overall and by time since first exposure, for cases diagnosed in 1986-2000 and 2001-2006, and for CLL and non-CLL cases (data not shown). The discrepancy could have arisen by chance or from an unknown ascertainment anomaly. Other possible reasons were that the 20 cases were undergoing therapy at the time of interview or were in poorer health compared to other cases, which could have influenced the accuracy of recall. In our primary analyses, we omitted these 20 cases, so results were not unduly influenced."

  4. A question. If the 20 excluded cases were for people interviewed less than 2 years from starting chemo, then by definition these people must have had leukemia. So how could the ERR be negative? A negative number would indicate some hormetic effect, would it not? But they DID get cancer. I am obviously confused here.

    1. A negative ERR does mean exposed subjects had a lower incidence of disease than the general population. The authors basically divided their subjects into groups by estimated dose. The ERR for a given dose D can be solved from

      (leukemia rate at dose D / normal leukemia rate) = 1 + (ERR/Gy * D)

      The ERR/Gy figures in the paper are basically averages of the these ERR/Gy figures across all groups.

  5. Thank you for writing such a detailed synopsis of this study. If it weren’t for this article, I wouldn’t have even noticed this study. I’m always tuned in to these sorts of studies because I’m interested in hearing the evidence against nuclear. Being a nuclear engineer, my livelihood depends on a strong nuclear industry and it’s upsetting to hear so much negative propaganda about nuclear energy.

    After reading the paper, it seems clear the researchers were strongly influenced by a predisposed belief in the LNT model. They even say that the 20 cases thrown out are inconsistent with their current understanding of “radiation-related leukemia risk”. I read this as, “it is inconsistent with some monotonic relationship between leukemia risk and radiation dose.” They are admitting that they have to throw out cases because it doesn’t fit their thesis.

    They go on to present results for the entire data set. The ERR seems significantly lower (which is underplayed by the authors) and the p-value is higher. The higher p-value indicates that the 0.74 ERR/Gy is not even statistically significant and that the high ERR calculations could be entirely due to chance. (Note: I base that statement on my rudimentary understanding of hypothesis testing and I don’t really know how they are actually calculating the p-value).

    This paper is why I generally dislike statistics. Authors go in with a strongly held belief, blatantly skew statistics to fit their belief, then hide their malpractice by using complicated arguments. Also note that the entire data set and equations used are not readily available. God forbid somebody else come in and do a full peer-check of their results.

  6. What model does NASA employ to gauge rad exposure for Shuttle/ISS crews? What’s the solar storm cosmic ray flux tipping point to abandon ship? Can’t Google it up. I assume NASA would use a model with the most fidelity and accuracy — and can such be applied to terrestrial situations?

    James Greenidge
    Queens NY

  7. My father worked at Oak Ridge for two years (1944-1946) and got a diagnosis of CLL at age seventy. He died in 1993 at 78. We applied for EEOICPA compensation and were successful. I think it was because the cause of death stated on his death certificate was “lymphoma”.

Comments are closed.

Recent Comments from our Readers

  1. Avatar
  2. Avatar
  3. Avatar
  4. Avatar
  5. Avatar

Similar Posts