In this study, we examined the impact of dosimetric metrics on local recurrence and analyzed recurrent characteristics of NPC patients treated with IMRT. We found that eight dosimetric metrics and HI# were significantly associated with local recurrence of NPC patients while only D95 and D5 were independent prognostic factors. More importantly, a novel model constructed with these two factors could effectively predict the risk of local recurrence. Moreover, we found that in field recurrence was still the main failure pattern of NPC with IMRT, and nasopharynx cavity, clivus, and pterygopalatine fossa were the frequently recurrent sites.
IMRT was a major break-through in the treatment of NPC which dramatically enhanced the local control rate of NPC, with a 5-year LCR of 95% for T1-2 disease and 80%-88% for T3-T4 disease [20–22]. The improved LCR was associated with highly target dosimetry coverage and conformity in an IMRT planning [23, 24]. However, sometimes it is difficult to balance the conflict between the potential serious late injuries and the risk of local recurrence due to inadequate target coverage in an IMRT planning, especially in NPC patients with advanced stage [9]. Additionally, it should be also noted that quality of IMRT planning also differs due to physician’s capabilities and personal experience. The dose coverage and uniformity of the target might be inferior in certain patients, thus leading to tumor relapse. Therefore, identifying reliable dosimetric metrics associated with their treatment outcomes are significant.
In the current study, univariate and multivariate Cox regression analysis were carried out to identify important dosimetric index to predict local recurrence. First, we found that T classification was not correlated with LRFS, indicating that T classification alone had less power in dividing patients into different risk groups in IMRT era, which was similar to the result of other studies [14, 15]. Although patients with advanced T classification usually had a larger tumor volume, the sophisticated IMRT technique greatly improved the dose distribution and reduced the proportion of insufficient dose-related recurrence compared to two-dimensional or three-dimensional conformal radiotherapy. However, the univariate analysis found that eight dosimetric metrics were associated with LRFS, among which D5, D2, D1, and Dmax reflected the near-maximum dose of the target volume, while D95, D98, D99, and Dmin reflected the near-minimum dose of the target volume to some degree, suggesting that local failure might probably associate with dose homogeneity of the target volume. Similarly, other studies also indicated that both D95 and Dmin were significantly associated with local recurrence [10, 11]. Subsequently, the multivariate analysis demonstrated significant prognostic value of D5 and D95 in the LRFS of NPC patients. A cumulative risk score consisted of this two dosimetric metrics was calculated, which indicated that this two-dosimetric parameter signature independently predicted LRFS in NPC patients. And the AUC value of the ROC curve was more than 0.7 when assessing the accuracy of the signature over 3-year LRFS, suggesting that the established risk model was reliable.
In this risk model, D95 and D5 were ultimately incorporated to predict local recurrence. Although ICRU 83 reports have recommended D50 as dose-volume parameter for evaluating IMRT planning [16], it was poorly adopted in academic institutions according to a survey [25]. Furthermore, consistent with other studies, there was no correlation between D50 and local recurrence [26]. However, D95 was a commonly used dose-volume constraint in clinical practice and some clinical trials use this metric to determine prescription dose [27]. In fact, the significance of D95 in an IMRT planning was similar to D98 to some extent. As D95 increased, the near-minimum dose of the target increased, thus increasing the whole absorbed dose of the target volume, which was helpful to tumor local control. One study suggested that a Dmin to the GTVnx ≥ 54.0 Gy conferred better local control in NPC patients with T3 and T4 [11], and another study also indicated that patients who received at least 66.5 Gy to primary GTV were less likely to have local failure [14]. By contrast, the significance of D5 was similar to D2. As D5 increased, the near-maximum dose of the target correspondingly raised. Hence, the dose uniformity of target was decreased, which might be harmful to tumor local control. In essence, the two dose-volume metrics of this risk model determined the shape and trend of a dose line in DVH (vertical drop or not) to some extent, reflecting a homogeneous absorbed-dose distribution in the target [16]. From this point of view, the risk model developed by this study was sensible.
Due to this risk model included the metrics of D5 and D95, which were also the key parameters to be used for calculating the dose HI#. HI# is also a commonly used dosimetric parameter for treatment plan reporting recommended by TG10118. Hence, the univariate analysis was performed to examine the relationship between HI# and local recurrence. We found that patients with higher HI# had significantly shorter LRFS than that with lower HI#. Dose HI reflected the uniformity of the absorbed dose distribution of the target volume. As the HI# increased, the “hot spot” of the target volume increased, and the “cold spot” of the target volume decreased. This meant that this IMRT planning itself was difficult, and dosimetrists might sacrifice dose coverage and uniformity of the target to reduce doses of organ at risk (OAR), thus increasing the risk of tumor relapse. In addition, we did not found that HI* was statistically associated with LRFS. The possible reason might be that the formula used to calculate HI* included three parameters according to ICRU 83: D2, D98, and D50. However, these parameters were not independent prognostic factors in our multivariate analysis, which was also consistent with the study of Wang et al [26]. Therefore, our study added evidence that HI# might be a more promising parameter for IMRT evaluation compared with HI*.
Although HI# was demonstrated as an indicator for predicting local recurrence, the predictive power of HI# was lower than that of the risk model according to the ROC value, especially the signature of 3-year LRFS. Therefore, compared with the HI#, the risk model that we established was more preferable to predict local recurrence of NPC patients.
In addition, the results of one study exploring the influence of target dosimetry on tumor recurrence in NPC differs from our results [26]. They concluded that D90 was the independent dosimetric parameter for predicting tumor recurrence and patients with D90 < 101% had higher incidence of local-regional recurrence than those with D90 > 101%. The possible reason underling the inconsistent conclusion might be that their studies focused on both local and regional lymph nodes recurrence, while ours only focused on local recurrence. Theoretically, the impact of dosimetric metrics on primary tumor is greater than on regional lymph nodes, because the latter is more influenced by anatomical change and positioning errors during radiotherapy [28–30], which might ultimately affect the analysis of dosimetric metrics on treatment outcome. From this point of view, it might be more reasonable to only focus on local recurrence when analyzing the effect of dosimetric metrics on treatment outcome, and include more factors when analyzing the factors influencing the local-regional recurrence. Nevertheless, more studies are needed to validate these conclusions.
Previous studies have showed that the local relapse of NPC mainly occurred in high dose area. In the study of Yang et al, they analyzed 212 NPC patients undergoing IMRT and found that 18 patients developed local recurrence, 15 (83.3%) of which were confirmed with in field failure [31]. Wang et al also reported that in field failure was the main pattern associated with local-regional recurrence of NPC [15].The present study further confirmed this conclusion: in field failure was found in 68.2% recurrent patients, while marginal and outside field failure were not common. Together with those results, it was suggested that the definition and delineation of CTV currently used was large enough, with low incidence of outside field failure. Hence, in the future, further reducing CTV coverage to reduce late complications of patients is an important direction to explore in IMRT era in the future [31, 32].
This study has several limitations. Although PSM method was adopted, the selection bias was inevitable. Besides, due to the lack of more patients with local recurrence in our center, we did not have enough patients to construct another independent cohort to validate the risk model. Finally, we just focused on the dosimetric metrics of PGTVnx, other targets and OAR sparing might have some effects on outcome of patients. Given these limitations, more studies or multicenter researches are warranted.