Response to “Critique and Discussion of ‘Multicenter Evaluation of Frequency and Impact of Activity Infiltration in PET Imaging, Including Microscale Modeling of Skin-Absorbed Dose’”

MONTE CARLO DOSIMETRY MODEL

Regarding the Monte Carlo dosimetry model, significant concern was expressed that our geometry excluded muscle from the distribution volume of an infiltrated radiopharmaceutical injection. We stand behind our distribution model that limits activity to the subcutaneous tissue and, to a lesser extent, the dermis. Muscle is encapsulated in the epimysium, which is a thick connective tissue layer that is composed of coarse collagen fibers in a proteoglycan matrix. The epimysium surrounds the entire muscle and largely isolates it from macroscopic rapid exchange of fluids from surrounding tissue, even under pressure. Unless the radiopharmaceutical is accidentally directly injected into the muscle, there is no direct pathway into muscle tissue. Further, our review of PET/CT infiltrations invariably shows the infiltrate limited to the skin layer, with no detectable component in the muscle above expected background. Figure 6A of the article demonstrates that, in an animal model, fluid introduced under pressure in the subcutaneous tissue is contained within the fat space and does not enter muscle. We strongly disagree with the criticism that “the muscle tissue adjacent to the injection site is valid as both a source and target volume” and is “inappropriately ignored…in the dosimetry model.” We stand by its inclusion as a target organ only. We agree that in the unlikely event of an intramuscular injection, muscle would need to be a source and target organ, but we would then exclude all skin structures as source volumes, as tissue exchange is improbable.

Even in the unlikely event of a direct intramuscular injection, the distribution volume in the muscle is large, which would dilute the infiltrate over a larger volume, thus reducing absorbed dose. Further, muscles are among the least proliferative and most radiation-resistant cells in the body. Only at doses in excess of 40 Gy (functional changes) (5) or 60–80 Gy (significant tissue injury) (6) do we see tissue effects in muscle tissue, and these absorbed doses are in excess of those achievable with diagnostic quantities of PET radiopharmaceuticals. The significant concern voiced for damage to muscle tissue as an unstudied risk is entirely unfounded and ignores decades of radiation biology experience from external-beam radiation therapy.

There was concern expressed that we did not compare our Monte Carlo results against existing published models. In fact, we did perform several dose estimates of the skin using several existing published models (not reported). The results of these dose calculations were entirely consistent with the literature, and only when simulating approximately 100% infiltration of administered activity did absorbed doses exceed values for which we would expect to see deterministic and observable skin reactions (2–10 Gy, mild temporary effects; >15 Gy, high probability of serious or permanent injury) (7). Yet we found no such reports in the literature for diagnostic PET radiopharmaceuticals. It was precisely the failure of conventional dosimetry methods to explain observed phenomena (or lack thereof) that prompted the development of the proposed model that accounts for skin tissue subanatomy. Modeling accounted for an approximately 40-min biologic half-life combined with the physical half-life of the radionuclide under study. The biologic half-life was derived from a typical 30-min combined biologic and physical half-life reported by Osborne (8).

We freely admit that this is only an early-phase model, but we think it holds promise to assess safety risk more accurately in the event of a significant infiltration event than do the current more simplistic methods, which appear to correctly calculate absorbed dose when using a somewhat arbitrarily assumed tissue mass but incorrectly predict risk. We are in the process of expanding the scope of simulation to include a wider range of radionuclides and geometries and expect these results to be published within a year.

The opinion piece further objects to our using a subcutaneous injection model to describe fluid dynamics when a radiopharmaceutical is infiltrated. They state, “Subcutaneous administrations are very different than intravenous and are not an appropriate basis for model definition.” We agree fully that subcutaneous administrations are very different from intravenous injections. However, we do firmly believe that subcutaneous administrations are precisely analogous to infiltrated intravenous administrations. Veins accessed for intravenous administrations reside exclusively in the subcutaneous tissue, and when injectate leaks into surrounding tissue under pressure through a blown vein or from around the puncture site of the vein itself, the leakage will invariably enter the subcutaneous fat layer given the anatomic confines. The fat layer is a remarkably accommodating and elastic structure to contain the excess fluid introduced under pressure. References supporting this were provided in the original article. We maintain confidence in our geometric model defining the behavior of infiltrated injectate and its time course and disagree strongly that our model is inappropriate; we consider our approach to be a substantially more appropriate physical model than currently used approaches.

FREQUENCY-OF-INFILTRATION STUDY

Regarding the frequency-of-infiltration study, it was initiated because of the discordant results between our institution and the reports in the literature describing much higher rates. It became apparent on reading the literature that the primary difference from our internal institutional analysis was that we measured activity at the injection site, where reports of significantly higher infiltration rates were based solely on “visualization” of activity.

We absolutely stand by our belief that visualization is an inappropriate criterion to characterize a meaningful infiltration event. The clinical utility of PET in oncologic applications is entirely dependent on the modality’s exquisite sensitivity. Virtually all PET scanners will clearly visualize a 2-cm-diameter tumor with ¹⁸F-FDG at an SUV of 4. With a typical injection activity and body weight, this tumor will have, very approximately, 37 kBq (1 µCi) of activity, or about 0.01% of the injected activity. This implies, particularly in the low-background injection site, that PET is capable of visualizing this amount of activity. Categorizing 0.01% of the injected activity as a reportable or significant infiltration event, by virtue of visibility, is categorically wrong and misleading, particularly since the activity could instead be trivially quantitated in less than a minute from the image data. It is based on these observations that we now believe we clearly understand the discrepancies between our institutional results and these other visualization-based literature reports, which we consider misleading for the above reasons.

Missing from the literature was a body of quantitative measurements of activity at the injection site. This was considered a significant information gap that this study intended to fill. The true incidence rate for significant infiltration events remains unanswered and will depend entirely on a formal definition, which is beyond the scope of the article and our expertise and responsibility. But it will hopefully be better informed because of the data reported.

Criticisms were made implying inherent bias in the data reported. We believe the study took reasonable efforts to avoid bias in data collection. Intentionally, a variety of institutions (10 total) were chosen, including an academic medical center, private radiology groups, private oncology groups, a community hospital, multispecialty groups, and a research facility. Consecutive patients who had the injection site in the field of view were studied. To avoid statistical overweighting, no single site was allowed to contribute more than 200 studies. Consistent analysis methods were used to quantitate activity at the injection site. Criticism was leveled that “training and experience levels of participating technologists” was not reported, and “an unknown number of images with injection sites outside of the field of view were excluded from the study.” Regarding the latter concern, we believe strongly this did not in any way statistically bias results. With regard to technologist training (many sites were small enough to not have on-site reading physicians), we are confident that this diverse array of 10 different institutions represented an array of different technologist skill levels and is almost certainly a more accurate sampling of the technologist population than the largely single-center studies on which the author and his company base their estimates.

The critique further states that “The results from this paper only reflect what happened in these few centers during undefined observation periods and cannot be applied to the practice of nuclear medicine generally.” As we believe ours is a largely unbiased sample, we believe strongly that it is entirely generalizable to the broader PET imaging community. Injection practices in the larger nuclear medicine community may or may not be similar, as we made no attempt to sample this broader space. However, regarding the Monte Carlo dose estimation methods, we do believe the approach is broadly applicable to the entire practice of nuclear medicine. This criticism about generalization from a well-sampled population is a particularly odd comment and concerning for several reasons. First, it flies in the face of the entire field of statistics, which is based on unbiased sampling of a much larger population where the sample is considered mathematically representative of that larger population—to within calculable confidence intervals. Second, this is a self-defeating argument coming from an individual and company who have continuously based comments to the Nuclear Regulatory Commission and other organizations on arithmetic extrapolation from much smaller, less controlled, and more statistically biased reports.

Finally, in our article we somewhat arbitrarily categorized infiltration of less than 1% of total activity at the injection site as being “not a clinically meaningful infiltration event” for the sake of simple statistical analysis. This in no way implies, nor means to imply, that an infiltration of more than 1% is a clinically meaningful infiltration event. The absorbed dose estimates from the Monte Carlo analysis suggest that even at a 100% injection infiltration, we would not expect a patient to experience deterministic skin injury, which is entirely consistent with the lack of reported events in the literature over the last several decades. The question of a threshold for compromised image quality or quantitation was not addressed by the article. As such, the frequencies of “clinically meaningful extravasations” calculated in the critique based on a 1% threshold are dramatically overstated.

John J. Sunderland*, Stephen A. Graves, Dusty M. York, Christine A. Mundt, Twyla B. Bartel

*University of Iowa, Iowa City, Iowa

E-mail: john-sunderland{at}uiowa.edu