Clinical heterogeneity and tumor control probability

Abstract

Heterogeneity in tumor sensitivity to therapy within patient populations is a major factor affecting tumor control probability (TCP) models, typically flattening dose-response curves. Knowledge of the major clinical factors affecting TCP would allow incorporation of these factors into TCP models, or stratification of patients and tumors, yielding steeper and more specific and predictive TCP curves. Methods. Literature review of the results of therapy for oropharyngeal cancer. Results. In recent years an improvement in control rates of cancers such as oropharyngeal cancer has been reported as potentially gained due to the dosimetric superiority of intensity modulated radiotherapy (IMRT). However, the emergence of Human Papilloma Virus (HPV)-related oropharyngeal cancer in recent years is an example of a major clinical factor which likely supersedes the dosimetric gains of IMRT. Conclusions. Clinical factors such as the example discussed in this paper: HPV-related oropharyngeal cancer, must be taken into account in TCP models in order for such models to be relevant.

Current TCP models typically utilize dose distributions within the tumor and population-based estimates of tumor control [1]. Individual variability in tumor radiosensitivity is taken for granted: for example, while the surviving fraction at 2 Gy (SF2) was reported to be quite similar in two studies, mean 0.39 [2] and 0.33 [3], the range of SF2 in both publications was very wide; 0.2–0.7 and 0.1–0.9, respectively. Inter-tumor heterogeneity shallows the average TCP curve, and the effect of in vitro radiosensitivity differences was found to be larger than heterogeneities in clonogen numbers [4]. The evidence of the crucial effect of tumor sensitivity differences on pre-clinical models of tumor control probability (TCP) models translates clearly into clinical data, where inter-patient heterogeneity flattens TCP curves based on models derived from in vitro experiments [5]. Furthermore, even if in vitro heterogeneity in tumor sensitivity to therapy was taken into account, the clinical dose-response curve was still shallower than predicted by TCP models, and it could only be explained by a higher variation in radiosensitivity than apparent in pre-clinical studies [5]. Webb and Nahum stated that there was an enormous variation in TCP parameters depending on whether or not patient heterogeneity was incorporated [6]. Thus, inter-patient heterogeneity in the radiosensitivity of tumors seems to be a dominant contributor to the shallowness of the observed dose-response curves.

Inter-patient heterogeneity in radiosensitivity has been taken into account in models of TCP. For example, the Nahun-Tait-Webb TCP model incorporates inter-patient variation by assuming that a is distributed normally, and the final TCP is the average of TCPs of patients with different radiosensitivities. Nahum and Sanchez-Nieto demonstrated that TCP curves for patient sub-populations divided according to tumor radiosensitivities would re-generate steep slopes that are similar to those obtained using pre-clinical experiments and data [1].

Methods

Prominent examples for the importance of dosimetric factors assumed to affect TCP, and clinical factors associated with tumor response heterogeneity, include the use of IMRT, and Human Papilloma Virus (HPV)-related oropharyngeal (OP) cancer, respectively. Both have emerged simultaneously in the developed world during the past decade. Relevant literature is reviewed below, aiming to compare their relative importance in achieving improved tumor control for patients with OP cancer.

Results

Intensity modulated radiotherapy (IMRT) has been used extensively in academic centers since the late 1990's. Its proponents cited better dosimetric tumor coverage compared with conventional RT in cases of locoregionally advanced OPC thanks to reduced constraints regarding spinal cord doses and the lack of dose deficiencies related to off-cord photons and posterior neck electron match lines. These dosimetric advantages, which have been translated into improved calculated TCPs [7], have been cited as explanations for a high rate of local/regional tumor control, typically > 85%, in published clinical IMRT series [8–13]. These rates are higher than rates typical of previous series using conventional RT. Furthermore, adequate dosimetric coverage of the target was found to be associated with improved tumor control. RTOG 00-22 protocol was a phase II multi-institutional study of IMRT in early oropharyngeal cancer. Seven of 69 (10%) patients in this study had local/regional (LR) failures, of whom two had major dosimetric variations in PTV66 coverage (both failed locally), and two patients had inevaluable treatment plans. The major variations in the two who failed locally were 66 Gy (PTV prescribed dose) covering 88% or 80% of PTV66 (rather than 95% per protocol). Overall, two of the four patients with major under-dose variations and evaluable plans failed locally, compared with three LR failures in 49 patients with evaluable plans and no major variations (p = 0.04). The conclusions were that major dosimetric deviations in IMRT plans were significantly associated with LR failures [14]. Had such variations been avoided, it is possible that LR control rates would be higher. These results seem to validate assumptions underlying TCP models. However, other factors may explain the high control rates observed in series of HN cancer treated with IMRT.

An interesting retrospective analysis from the University of Wisconsin [15] found an improvement of outcome of patients treated with IMRT compared with previous-era patients, and IMRT was a significant prognostic factor. However, multiple factors, including patient selection and the addition of concurrent chemotherapy, could explain these improvement. The most interesting finding was that similar patients treated with conventional RT in recent years enjoyed similar improvement in outcome.

Very recently, a powerful clinical factor has emerged which may explain the improvement of treatment results in the IMRT era. While in previous years almost all cases of OPC were related to heavy smoking and alcohol abuse, in recent years about half or more of OPC cases in western countries do not have any present or past smoking history [16–18]. Their cancer is attributed to HPV, and their prognosis is much better than smoking-related ones [19]. This improvement in prognosis is striking and is likely more relevant than the minor improvement in tumor doses achieved by IMRT. If no stratification of OPC to HPV-related or smoking-related is performed, it is likely that any TCP modeling or calculation in this population will err substantially. No similar emergence of HPV-related OPC tumors has been observed in developing countries, where almost all cancers are still smoking-related [20]. Thus, models of TCP developed using data from developed countries may not be relevant to patients in developing countries. An example can be made of the long-term results of a randomized study of RT alone vs. RT concurrent with cetuximab, an antibody to the Epidermal Growth Factor receptor (EGFR). Statistically significant improvement in survival was found in the RT+cetuximab arm, making it a standard of care for HN cancer [21]. The authors have provided a detailed analysis of the sub-groups of patients which benefited from cetuximab: patients with oropharyngeal cancer treated in the US (this was an international study), with relatively early tumors, high performance status, males, younger than 65 years old, and whose tumors contained less, rather than more, EGFR positive cells (Figure 3 in [21]). All these factors are characteristics of HPV-related OP cancer. It is possible that cetuximab was indeed more effective in HPV-related OP cancers than in other HN cancers. However, taking into account the excellent prognosis of these patients, another possibility is lack of balance in the numbers of HPV-related OP cancers in the two arms: a relatively small difference in the allocation of these patients between the two arms may explain all the advantage in survival observed in the RT + cetuximab arm. We cannot assess this possibility as neither tumor HPV status nor patient smoking history were provided by the authors. Thus, potentially small imbalances in important clinical factors (which were not known at the time of the study) may have affected critically the results of an important randomized study, notwithstanding the post-hoc nature of the sub-group analysis.

Discussion

In the case of OP cancer in recent years, the data presented above suggests that clinical factors such as HPV-related OP tumors, especially in non-smokers, is a powerful prognostic factor which likely supersedes dose-related considerations. Are we likely to develop in vitro tests assessing the sensitivity of HPV-related tumors, which will enable calculating HPV-specific predictive TCP curves? Very little relevant in vitro data has been published thus far [22]. It is possible that immunological, host-related, factors are associated with the higher sensitivity of HPV-related tumors to therapy, rather than intrinsic sensitivity [23]. If this is the case, then in vitro testing will be futile and population-based data will be the only available sources for TCP parameters for these patients.

In conclusion, tumor and patient heterogeneities are major factors affecting TCP. Some of these factors, such as the emergence of HPV-related OP cancer in recent years, have such an important effect on TCP that disregarding it would make relevant models obsolete. The identification of such clinical factors, and their incorporation into TCP modeling, is crucial.

Acknowledgements

This paper was supported by NCI grant PO1 CA59827 and by the Newman Family Foundation.