Methodology for analysis and reporting patterns of failure in the Era of IMRT: head and neck cancer applications

Traditionally, failure reporting for HNCs has simply classified disease as “local”, “regional”, or “locoregional”, thus relating location of failure to a crude anatomic reference. However, such a reporting language gives no information regarding radiation fields/volumes, or delivered dose. In the pre-IMRT era, when large fields of homogenous dose were used, the definition of “in-field” failure (i.e. within the field borders) or “marginal” failure (i.e. adjacent to the block edges) were intuitive descriptors relating treatment parameters to sites of failure. However, in the current era of conformal therapy [23, 24], dose gradients, and multiple TVs make relating spatially accurate information about dose and recurrence far more complicated for IMRT plans. In the same way that a standardized method for analysis and reporting for TVs has been undertaken successfully [23‐25], a similar effort is desirable for pattern of failure reporting. In our opinion, reporting failure using only anatomic/field referents is insufficient for complex multi-volume/multi-dose plans, and obscures clinically useful information which might lead to improvements in future studies.

Likewise, the requirement of rigorous assurance for correctly localizing disease after-therapy is increased in terms of required spatial accuracy. The steep dose gradients of modern IMRT plans and proximate transition from high-risk CTVs to intermediate- or lower-risk CTVs implies inaccurate registration will erroneously assign the location of failure to incorrect dose/prescription volume. We, in fact, show RIR for IMRT plans resulted in incorrect localization relative to prior TVs and dose in 16 % of failures in the current pilot dataset. Consequently, this study presents a methodology and workflow that involves the application of quality assured DIR software as a tool to standardize co-registration and to correctly attribute sites of loco-regional failure.

Almost all previous studies have used RIR to describe the patterns of loco-regional failure after IMRT [2, 10, 12, 13]. Chao et al. [10] reported 17/126 loco-regional failures treated by definitive or postoperative IMRT; 53 % of failures were inside CTV1, 12 % marginal to CTV1, 6 % marginal to CTV2, and 28 % were out of field. Eisbruch et al. [2] reported on 21 recurrences in 133 patients with non-nasopharyngeal HNCs treated with parotid-sparing IMRT; 17/21 were in-field. Daly et al. [12] reported on 69 HNCs treated with parotid-sparing IMRT; 8 patients developed a loco-regional failure, 7 relapsed within the high-dose CTV, with one junctional failure observed. Sanguineti et al. [13] described the patterns of failure for 50 patients with IMRT for oropharyngeal SCC; 5 recurrences were related to high dose regions while 4 were at the low dose regions. All these reports relied on RIR, known to be less spatially accurate than DIR [18]; it is conceivable these results might be altered if DIR methods were used. Due et al. [7] reported that DIR showed slightly better reproducibility in identification of the site of recurrence origin compared to RIR. Our previous work [18], as well as the current study, confirm the qualitative superiority, in HNC applications, of DIR for CT-CT registration.

In our classification scheme, we designed a combined geometric/dosimetric typology definition to avoid the drawbacks of using each method separately. Centroid only method suppose a single point of origin and ignore the dose given to the whole area of recurrence while the dosimetric only analysis is agnostic to the geometric recurrence origin. Due et al. [7] previously reported that focal methods, such as the centroid method we used, are more accurate to localize the origins of loco-regional recurrences than volume overlap methods, which may incorrectly assign recurrences to more peripheral TVs. Raktoe et al. [8] further confirmed the superiority of focal methods like centroid expansion to the volumetric method in identifying the origin of loco-regional recurrences. The combined method we used identify the estimated site of recurrence origin relative to the respective TV in the planning CT and then compare the dose to the mapped recurrence volume with the dose prescribed to the TV of origin. Using this method, our results showed that DIR significantly assigned failures to more central TVs and doses compared to RIR concordant with the results of Due et al. [7].

Our proposed nomenclature allows granular reporting of different types of failure. In our classification, type A “central high dose” failures, are considered to be biological failures, as they likely represent resistance to maximal therapy, and thus could not conceivably be prevented by technical/operator dependant processes including IMRT QA or delineation alteration. Type A failures motivate future investigation of alternative treatment stratgies (e.g. integration of novel targeted drug therapies or dose escalation to identifiable biologically aggressive subvolumes). Likewise, type E “extraneous dose” failures cannot be modified by IMRT QA processes. They represent a possible diagnostic or decision error rather than a target delineation error (i.e. “One will never hit what one does not aim at.”). However, type B, C, and D failures are of a special concern since they entail potential technical or radiotherapy process failures. Type C “central elective (intermediate or low) dose” failures may be prevented by prescribing higher doses (i.e. shifting to higher CTV levels). Importantly, type B “peripheral high dose” or D “peripheral elective dose” failures necessitate a rigorous QA process including triple DIR registration of pre-therapy diagnostic imaging (diagnostic CT, MRI, and/or PET-CT) to pCT and the earliest rCT, to assess the potential causes: potential target delineation or dose delivery error (modifiable) versus overgrown recurrence that represents actual type A or C failure which is converted to type B or D, respectively, due to rapidly progressive disease or neglected late diagnosed recurrence (not modifiable). This involves multi-physician review of planning and recurrence contours, and review of IGRT data (i.e. set-up error, adaptive replanning datasets), as well as examination of the follow-up interval between surveillance images. By cataloging type B/D errors, we can then address the relevant issues dynamically for future patients. For instance, the only type B patient (i.e. using DIR methodology), was noted on secondary review of diagnostic imaging to have subsequent intracranial extension, route of failure, despite optimum delineation and dose coverage.

The secondary qualitative review by expert radiation oncologists (CDF, DIR) of all the clinical and imaging data of the four additional recurrences that were classified as peripheral high dose (type B) using RIR while were type A using DIR, concurred with DIR classification that those recurrences are actually central rather than peripheral in origin.

In this limited pilot dataset, our results showed the majority (84.6 %) of DIR mapped failures were of type A indicating, that biological, non-technically/non-operator dependant explanations for failure predominated. However, using RIR type A failures would have been erroneously reported as comprising only 69 %. These results assert the need for a robust, quality assured image registration technique, as error in the registration process would invalidate subsequent results and thus might deceptively indicate a greater rate of technical/operator-attributable therapy failures than DIR demonstrates. The current study, while underpowered to make clinical extrapolations due to limited number of patients, nonetheless serves as a benchmark to describe our standardized analytic and reporting method. Already, RTOG 1216, for example, contains provisions regarding collection of imaging data post-failure [26], which will allow careful analysis, and process quality improvement for future trials and large scale datasets.