TO THE EDITOR:
The quest for a correlation between tumor radiation-absorbed dose and response in radioimmunotherapy has been a difficult and, so far, mostly marginally productive effort. By the usual P < 0.05 requirement, Sgouros et al. (1) recently did not find statistically significant correlations for dose mean, maximum, minimum, and uniformity for tumors in 15 non-Hodgkin’s lymphoma patients participating in a phase II study of therapy with a combination of unlabeled tositumomab plus 131I-labeled tositumomab. For all patients, previous chemotherapy had failed. At the University of Michigan, we have studied patients undergoing the same treatment procedure. However, in some of our research, including research with similar measurements of mean radiation-absorbed dose, the patients were all previously untreated (2–5).
In their discussion, Sgouros et al. (1) correctly commented that in one of our publications (4) we presented results for a restricted dataset, that is, not for all patients potentially available to us for evaluation and not for all time points at which they were scrutinized after therapy. The reasons for this practice were 3-fold: First, we eliminated axillary tumors from the study because we have found that they have a considerably lower radiation dose estimate than do abdominal and pelvic tumors (2,5). Second, we chose to include tumors from only those patients who went on to achieve a partial response, rather than patients who achieved a complete response, because we anticipated that the former would have a more widely distributed set of volume-reduction values at any time after therapy and might be a more homogeneous group. Third, we chose to look at our results only at 12 wk after therapy to reduce the work of the initial evaluation.
Sgouros et al. (1) also correctly commented that we determined several different dose–response relationships. In fact, we produced 4 probit-fit relationships (4). We used a time series of diagnostic conjugate views for 1, and we used those same data supplemented by a single intratherapy SPECT image for 3 others. Those 3 were for the dataset independent of the initial tumor mass, a data subset of tumors with an initial mass greater than 10 g, and a data subset of tumors with an initial mass less than or equal to 10 g.
Sgouros et al. (1) also wrote: “In no case was a statistically significant relationship observed” with our data. In fact, the P value was significant for 1 of our 4 probit-fit relationships. That is, for the SPECT-supplemented data subset consisting of tumors with an initial mass less than or equal to 10 g, a statistically significant P value of 0.029 was determined (4). This significance occurred for the best-fit sigmoidally shaped relationship between tumor volume reduction and radiation dose compared with no dose–response relationship, that is, compared with a constant volume reduction. (Note that we previously stated that the significance test was in comparison with a constant volume reduction of 50% (4). That statement was in error.) The data subset comprised 15 tumors in 6 patients. The curve was a slightly truncated version of the classic sigmoidal shape (4).
In addition to this result, since the time of our publication (4) one of the patients has been reclassified from PR to CR and so should be removed from the study as it was defined. Because the patient’s 2 tumors were larger than 10 g, the removal has no effect on the data subset discussed above. However, the P value for the entire SPECT-supplemented dataset now has become statistically significant in the same sense as above (P = 0.0496). With the removal of the 2 tumors, this dataset now comprises 41 tumors in 9 patients. The curve is a considerably truncated version of the classic sigmoidal shape.
These observed correlations between radiation dose and response involve only a limited number of tumors and are only modestly robust. We expect that the modest robustness exists because the tumor mean radiation-absorbed dose, although quite important to response, is not likely to be the only determinant. For example, the uniformity of the radiation dose distribution may also be an important contributor. In addition, there are certainly measurement errors for both radiation dose and volume reduction.
Sgouros et al. (1) gave a plot of volume reduction versus mean radiation dose for their data in Figure 4A of their article (1). This plot separately represented volume reduction at 3 different times after therapy, including 12 wk (designated 75 d because of a difference in time-zero definition). They reported that for these data no statistically significant correlation was observed. They did not give a P value for each relationship, but the range given for all their relationships was 0.25 to greater than 0.5. However, they did not examine their data for only nonaxillary tumors in only partial-response patients, and they did not fit a sigmoidally shaped probit function to the data. It appears that at 12 wk the probit fit would be statistically insignificant or exhibit a very weak dependence on mean dose. We assume it would also be so if the data were restricted to those for nonaxillary tumors in partial-response patients. Given that assumption, we are of the opinion that the crucial difference between this particular result of theirs and our result is that they examined patients who had disease relapse after chemotherapy, whereas we studied previously untreated patients.
Sgouros et al. (1) made a statement that can be used as a possible physiologic reason for the difference between their results and ours. That is, they said: “the effects of prior treatment … would be expected to differentially impact tumor radiosensitivity and, thereby, confound an absorbed dose–response relationship… ” for patients who had disease relapse after chemotherapy. They also cited a reference for their statement (6). In that reference, Williams presages the result that is the subject of this letter by saying “Wisely, Koral et al. chose only untreated non-Hodgkin’s lymphoma patients so as to minimize any analysis difficulty due to prior therapies.” This statement was specifically made in regard to one of our publications (3), but it can equally well be applied to all (2–5), including that on which Sgouros et al. commented (4).
If the crucial difference between the 2 results cited is indeed pretreatment versus no pretreatment, it lends scientific weight to the supposition that dose–response relationships are easier to find for previously untreated patients than for patients with disease relapse. If this supposition is true, it implies that, while tumor dosimetry is being improved, researchers looking for dose–response relationships in radiopharmaceutical therapy should initially concentrate on previously untreated patients in the unusual situations in which both types of patient are available.
On the other hand, even in their patients with disease relapse, Sgouros et al. reported that “a trend toward increased response with increasing [dose] uniformity was observed (r = 0.37; P = 0.06)…” (1). So, when uniformity of radiation dose can be assessed and a large number of tumors are available, it may be possible to find a statistically significant dose–response relationship, even for patients who have relapse of disease after previous chemotherapy.
REFERENCES
REPLY:
Koral et al. have highlighted an important aspect related to tumor-absorbed dose versus response. We completely agree with them and thank them for emphasizing this point. It is encouraging that radioimmunotherapy has evolved to a stage at which it is being used as a first-line therapy. As Koral et al. suggest, the availability of such studies will remove an important confounding factor in establishing tumor (and normal organ) absorbed dose–response relationships.