Tumor temperature distributions predict hyperthermia effect

Abstract

Review of clinical hyperthermia (HT) trial results shows that there previously has not been a robust model relating efficacy of HT treatments to characteristics of the temperature distribution. Lack of a model has been an impediment in Phase II trials; these trials must include defining the prescription for HT treatment, optimizing the schedule of HT treatments, and defining quality assurance procedures. We propose a model that is based upon noting that the majority of a tumor volume is contained in the outermost “shell” of a solid tumor, across which shell the radial temperature distribution is assumed to be linear. Any linear distribution can be defined by coordinates of a point and a slope, and we choose the temperature at the radiographically defined edge of a tumor and the slope (dT/dr) across the outer shell as these determinants of the linear radial temperature distribution. A discriminant analysis of success or failure of treatment can then be based upon these two descriptors (T edge dT/dr). We have tested this model using data from patients with soft tissue sarcoma (Stage IIB or greater) that have entered an ongoing prospective trial of conventional preoperative radiotherapy (5000 cGy/25 Fx/5 wk) together with HT, the latter randomized to be given once or twice weekly during the 5 week course. Wide local excision of the primary tumor is done 1 month after completion of radiotherapy, and the extent of histologic change in the resected specimen is scored. Our model has an 86% predictive value for lack of complete or nearly complete necrosis in the resected specimen according to whether the time-averaged T edge and slope during each HT treatment satisfy the equation T edge + 1.2 (slope in °°C/cm) s 40.6°°C in all but one treatment at most. Conversely, in 85% of cases with complete or nearly complete tumor necrosis, temperature distributions satisfied Td, + 1.2 (slope in °C/cm) > 40.6°C during at least one HT treatment. Requiring ≥one third of treatments of a patient to satisfy the preceeding discriminant equation resulted in 80% of patients being correctly classified as a responder or nonresponder, with only one false positive prediction (patient incorrectly classified as a responder). The model can reveal systematic changes in the edge temperature distribution during the treatment course that are consistent with tumor perfusion changes inferred and measured by independent means. We propose that the model be tested further in prospective Phase II trials using other tumor histologies, heating techniques, and treatment endpoints.