Register for Updates | Search | Contacts | Site Map | Member Login


View Comment

Submitted by Manfred Tschurlovits, Austrian Radiation Protection Association
   Commenting on behalf of the organisation
Document Foundation docs Optimisation; Dose to Individual
Manfred Tschurlovits 13.6.05

Note: apparently this mode of communication can not forward some format and symbols. The full ( WORD) text can submitted on request when an appropriate address is made known MT


1.1 General comments

Conceptually, this report shows in part some progressive development not necessarily done so far by the ICRP. This is because it is explicitly recognized that dose estimates can not be carried reasonably out by deterministic methods only, to employ distributions instead of numbers, and statistical methods are proposed. This is an important step toward a reasonable consideration of the problem, because simple the application of simple deterministic models, as used previously, do no longer sufficiently mirror the problem.

When distributions are used, some procedures have also to be established to discriminate between different exposed groups.
In order to cover the problem more complete, the subject of the report should therefore extended to:

Assessing dose of the representative individual for the purpose of radiation protection of the public AND PROVING COMPLIANCE WITH CONSTRAINS OR LIMITS

because both subjects are clearly interrelated.

The application of statistical methods is not implemented at all in standards, as people were prepared in depth for decades to think in and to apply deterministic methods, and it might be difficult to convince possible users on the suggested stochastic methods.

However, it is not mentioned that a more or less identical formalism was, after some decades of development, eventually applied in Low Level counting to discriminate between background and sample contribution and to define a Lower Limit of Detection. A number of technical standards as ISO 11929 are widely distributed and eventually accepted by the community. A statement on this reference might be useful.

Regarding the nature of the annexes, it seems that they are contradicting. On the one hand it is shown in annex A that the resulting dose in different age groups is not very different, but the data sets pretend accuracy which does not exist at all. Annex B shows broad distributions.

(46) However, deterministic models might remain useful in some cases, but it has to be mentioned (I did not found any hint on this issue in the report) that in some cases uncertainty analysis might improve the total uncertainty substantially. Application of error propagation might be desirable

In general, the report should go into more details, as at present the statements are very general

1.2 Details

Proposal to add New 2.3: It seems important to explain a little more whether doses have a broad distribution and to define the terms ¡§uncertainty¡¨ and ¡§variability¡¨ e.g. as:

Result of individual radiation dose assessments show for different reasons large confidence intervals. The following reasons can be roughly distinguished:

a) Variability
a1) statistical randomness or variability of nature, e.g., variability due to differences between individuals in their susceptibility and response to low doses of radiation,
a2) insufficient understanding of mechanisms, i.e. lack of scientific knowledge

b) Uncertainty
b1) insufficient dosimetric data, lack of epidemiological and laboratory data about the stochastic effects of low doses of radiation, and
b2) imprecision in dose assessment procedures

Variability refers to real and identifiable heterogeneity or diversity between individuals within a population. For example, variability might refer to differences in background cancer rates in different countries, individual sensitivity, and different exposure conditions. Sources of variability can be classified into three categories: spatial variability, temporal variability, and inter-individual variability. As there is a background cancer rate, these variations establish indirectly a lower boundary below which a statistically proved answer on the question whether radiation doses will result in harmful effects becomes neither possible nor reasonable.

Uncertainty arises from some unavoidable limitations in the assessment. These limitations are, regarding stochastic risk of radiation, mainly associated with the retrospective assessment of investigated cases.

Although uncertainty and variability differ significantly in terms of their causes, they lead eventually to the same outcome, namely to a broadening of the distribution of results. The contribution to the randomness of the data have different potential for improvement, as shown in table 1. Even when systematic errors can be excluded, both variability and uncertainty are to be assumed to be distributed at random. This leads in effect to two distributions of data, and the interaction leads to a broader distribution than the originating events. This combined distribution is eventually the basis for application.

Table 1. Potential to reduce uncertainty and variability.
Reason for large confidence intervals Potential for improvement
Variability Individual response No
Understanding basic mechanisms to be expected
Dosimetry/ epidemiology Limited
Methodology Yes

After: M. Tschurlovits, R. Taghizadegan, R. Engelbrecht: Handling uncertainty and variability in risk communication, Proc. IRPA 11 Madrid (2004)

It is interesting to note that there is no broader term including both ¡§uncertainty¡¨ and ¡§variability

In addition, it should be mentioned that the dose distribution of predictive models is broader than of retrospective assessments.

As an example (reference provided if required) it can be shown that the range of dose prediction in the case of a reactor accident can be large (see figure below, paper submitted)

Something should be mentioned how to handle distribution vs constraints or limits as numbers, i.e. how to deal in comparison of a statistical quantity (with uncertainty) with a number without uncertainty
Possible relations of the dose distribution with the limiting number are, where G: median of distribution, ċ: standard deviation, L: limit
a) G + ƒã „T L
b) G „T L, G + ƒã „d L
c) G > L , G - ƒã „T L
When condition c) apply, compliance is not proved. This formalism can be modified for lognormal distributions also.

(69) It is not proved in general that even a homogenous group receives dose with a range less than a factor of ten.

(72) The annexes do not necessarily suggest that the goal of simplification of the recommendations is met with this report.

Table 3, colum ¡§ deterministic¡¨:
1) The term ¡§Single best estimate¡¨ has the same quality in precision of expression as ¡§representative¡¨.
2) 2nd line: ¡§ percentile suggest that a number of data are available. If this is true, a distribution exist. In this case, ¡§median¡¨ is more correct than an undefined term ¡§average¡¨

To convince reader on the use of distributions instead of single numbers, it seems reasonable (perhaps in an annex) to show examples of broad distributions.

1.3 Misprints

Please consider to correct on p.24 the reference Tscherlovits „_ Tschurlovits. In addition, reference to this paper should be made, if at all, somewhere in paragraphs (51) or (52)

A-1 (A2) 4th line from the bottom 5- year old child

1.4 Coherence with other drafts
Some issues addressed here are appearing in other ICRP documents in consultation as ¡§basic quantities¡¨ , but not mentioned. e.g collective dose etc