Abstract
Dear Editor:
Our article
1
and associated letter to the editor, which appeared in the previous issue of this journal,
2
discussed some alternative methods for analyzing data from patients receiving peptide radionuclide receptor therapy (PRRT) treatments who manifested differing levels of renal toxicity in response to estimated kidney radiation absorbed dose. Our work discussed mathematical and conceptual issues present in the analysis of data from two patient populations reported by Barone et al.
3
and Bodei et al.
4
in Pamphlet No. 20
5
of the Medical Internal Radiation Dose (MIRD) Committee. Two letters of response were also published, one from the MIRD Committee
6
and another from the European Association of Nuclear Medicine (EANM) Dosimetry Committee.
7
We thank the authors for providing these comments and would like to address a few specific points raised in these letters: We completely agree with the EANM authors that accurate image quantification is an essential element to quantitative dosimetry, but we disagree with their assessment that “no protocol has emerged to date as an absolute reference for clinical practice.” We and many others have published extensively on this subject and specific recommendations can be found in MIRD Pamphlet No. 16.
8
The EANM authors discussed the data of Barone et al.
3
and stated that “the biological equivalent dose (BED) provided for a much better correlation (r = 0.93; p < 0.0001), despite the fact that the parameters applied (α/β and DNA repair half-time) were taken from the literature.” Although it is true that the BED approach improved the correlation, compared with previous methodologies, we simply emphasized that using the assigned values of 2.6 and 2.8 hours for the α/β ratio and tissue repair time, respectively, is essentially no different than using 1/T
eff (refer to our Eq. 4), because the effective half-time for PRRT is on the order of 30 hours or more. This essential point was not addressed by either MIRD or EANM. The EANM authors neglected to note that good correlations were not obtained, however, in another study by Bodei et al.,
4
and that different BED thresholds for renal toxicity were found in patients with and without risk factors. As indicated in our article,
1
when the Barone and Bodei data were properly combined and analyzed, the time–dose fractionation (TDF) model provided better correlations for all patients with and without risk factors. The BED model did not provide good correlations in patients with or without risk factors, whereas the TDF model greatly improved the correlation (r = 0.90 versus r = 0.69 for BED) in patients without risk factors, which is a better population to examine when attempting to establish new methodology. Thus, radiation dose by itself, or even dose coupled to the linear quadratic (LQ) radiobiological model, was unable to predict renal toxicity accurately in patients with certain preexisting conditions (toxicity occurred almost exclusively in such patients), and because the TDF model did better in this patient population, this is certainly an early indication that this model is worth additional consideration. We were disappointed that several points that we considered of greatest interest were not addressed by the MIRD authors, but were dismissed as “secondary”: (1) Different dosimetric quantities were inappropriately combined in a single analysis in Pamphlet 20, without adjustment for calculational consistency. MIRD authors commented that MIRD 20 explicitly states all model limits and assumptions, for example, a “PRRT data uncertainty from 10% to 25%,” knowing that the data collection and analysis methods of Barone et al.
3
and Bodei et al.
4
differed. However, this is not true as MIRD 20 does not recognize any such methodological differences and only presents a generic uncertainty for such data in general. (2) The fitting of the NTCP curve using these incorrectly combined dose quantities was performed with arbitrarily binned and limited data and these BED values, unlike presented in MIRD 20, are not consistent with the appropriately generated external-beam BED data for predicting kidney toxicity.
These points present credible challenges to the conclusions alleged and the soundness of the analysis in Pamphlet 20 and of the associated “web-based software tool” and are therefore highly relevant. Such a tool's theoretical basis should first undergo a thorough independent evaluation before being tested for use in any clinical situation. All such tools require formal evaluation and exemption by the Food and Drug Administration before being used in the general scientific community. The MIRD authors reiterate that MIRD 20 provides the “best method currently available,” that is, the best predictive model, to minimize kidney toxicity in PRRT; however, insufficient justification for this model is offered and its effective use for predicting nephrotoxicity on a pretherapy basis is not described or even known at this time. According to Bodei et al.,
4
“BED is a relatively young concept applied to nuclear medicine and has still to be fully validated with a wider series of data.” The MIRD assertion is thus overstated, because we have also shown that absorbed dose coupled with radiobiological modeling may be of limited value in predicting renal toxicity in patients with risk factors. There must first be a critical evaluation of some important issues in more PRRT datasets, including: (1) absorbed dose methodologies (e.g., the method of background subtraction can greatly affect kidney activity quantitation
9
); (2) the suitability of using one dosimetry study to predict total renal absorbed dose of a multidose treatment (one study that examined patients who received only two 90Y-DOTATOC therapies indicated significant intrapatient differences in kidney absorbed doses
10
); (3) the need for the MIRD 19 multiregion kidney model; (4) need for clinical screening for patient preexisting conditions, such as hypertension and diabetes, and risk factor effect on BED threshold for renal toxicity; (5) LQ model and effect of selected parameter values for α/β ratio and tissue repair half-time on BED dose value; and (6) optimization and refinement of the proper radiobiological model. The MIRD authors state we have asserted “that we cannot move forward unless we absolutely know everything about the patient's ‘specifics’” and that such “an ivory tower” approach is likely to harm more patients than are helped. This implies that having all patient parameters is neither necessary nor required to practice “good and safe” clinical physics and dosimetry and further that it is unlikely that all patient-specific information can ever be known. As there are patients “on the table,” the good physician will provide treatment solutions based on the current state of knowledge. But, we have not suggested that we cannot move forward; on the contrary, our findings indicate that more study of these models is warranted and should be pursued. The MIRD authors state that the TDF model “was abandoned by the radiobiology community in the early 1980s” and describe its use in our article
1
as a “phenomenological reach into the past.” However, as we noted, Voit and Yi
11
in 1990 demonstrated that the LQ and TDF models, although different, are mathematically closely related and produce similar results. In addition, Howell et al.
12
compared the LQ and TDF models for use in radioimmunotherapy in 1994 and concluded that at this time, “it is not clear which model should be used in RIT planning.….It appears that the LQ approach is favored in conventional radiotherapy; however, firm clinical data are needed to establish the usefulness of any model in RIT.” O'Donoghue, a radiobiologist, noted in 2004
13
that the applicability of the external beam experience to radionuclide therapy is “difficult to answer with any degree of certainty” and that the LQ model “may underestimate effectiveness of low-dose rate irradiation,” as typically delivered in PRRT. Our article
1
demonstrated very clearly that firm clinical data to support BED in radionuclide therapy do not exist and, interestingly, showed that the TDF approach performed better than BED in all cases. Thus, either the TDF model is not worthy of “abandonment” in radionuclide therapy and should be considered as a viable alternative or the LQ model requires much more investigation and optimization. We do not strongly advocate for TDF or against BED; we simply note that the Barone et al.
3
and Bodei et al.
4
data show that the investigation of this issue is far from settled, as advocated in MIRD 20 and the recent MIRD letter. The MIRD authors state that we “missed an important opportunity to clarify scientific and technical issues for [the] readers. Akin to missing the forest for the trees, they focused on minor aspects of these two publications.…” To the contrary, we considered a vast array of literature, presented an in-depth analysis, and focused on major aspects of this dosimetry issue. The readership of Cancer Biotherapy and Radiopharmaceuticals may benefit significantly from considering a broader evaluation of dose/response models and their applicability to different PRRT datasets.
In summary, we were disappointed that the MIRD committee representatives did not consider our analysis and comments seriously. Indeed, we presented concrete evidence that challenges the premise that the recommendations of MIRD 19 and 20 are the best and only solutions to the issue at hand. The goal for the predictive model in PRRT, as yet unrealized, is to predict in an individual patient the response and potential for nephrotoxicity and to use such information in adjusting the individual's treatment plan. It is our hope that a constructive dialog can be pursued with the involved medical community so that this important issue is openly discussed and challenged, because such efforts can only lead to essential improvements in a methodology that has promising clinical impact.