Reirradiation is currently regarded as the main local treatment for locally advanced rNPC. Our study showed that the prognosis of patients in rT3-4 stages was greatly improved in the IMRT era owing to the improvement of radiotherapy techniques and the modified total dose and fraction size compared with previous studies [24, 25]. For instance, 54.7% (29/53) of the patients were prescribed 60 Gy in 27 fractions to rGTVnx, which did not significantly reduce the local control rate of the patients, but decreased the incidence of fatal complications caused by a higher total dose of reirradiation so that they may have longer survival [6]. Nonetheless, how to avoid severe radiation-related toxicities needs to be addressed. Thus, it is prudent to explore a novel salvage local treatment for locally advanced rNPC. Recently, the development of endoscopic surgery has made it possible to resect locally advanced disease without causing severe complications. Wong et al. reported 15 patients with rT3-4 rNPC receiving salvage ENPG and their 2-year OS rate was 66.7% [17]. Li et al. retrospectively analyzed 120 patients with locally advanced rNPC undergoing salvage ENPG, and their 3-year OS rate was 55.2% [16]. These encouraging results indicate that patients with locally advanced rNPC undergoing ENPG may achieve favorable survival, but it is unclear whether salvage ENPG is not inferior to IMRT. The recently published data of Li et al. [26] tried to answer this question by comparing salvage ENPG with IMRT and 3D-CRT in patients with locally advanced rNPC. However, the results should be interpreted cautiously due to potential confounding factors. Firstly, it is a single-center study and the numbers (192 vs. 51 vs. 11) of three groups of patients were in great disparity. Meanwhile, they did not perform any matching to ensure comparability between the three groups of patients. Additionally, the median follow-up was 19 months (1–118 months) in their study, which was insufficient to observe survival endpoints. Therefore, we conducted, to the best of our knowledge, the first multicenter, case-matched study to compare ENPG with IMRT in efficacy and safety in treating locally advanced rNPC. We found that the patients in the ENPG group had similar OS, CSS, LRFFS and PFS but a lower incidence of severe adverse events compared with those in the IMRT group.

To better screen candidates for ENPG, we divided the patients into three subgroups in terms of their invaded structures and difficulty of resection: unresectable tumors, potentially resectable tumors and easily resectable tumors. Unresected tumors, invading critical structures such as the functional brain area and posterior cervical vertebra, are usually too extensive to resect and repair, so resection can be life-threatening. Easily resectable tumors are either superficial or located in the midline, e.g., ethmoid sinus and axillary sinus, so resection can be safe and efficacious. Perhaps it is controversial with regard to potentially resectable tumors. These tumors are usually adjacent to or even involve important blood vessels and nerves or natural barriers, such as the ICA, CS, optic nerve and meninges, which require exquisite surgical skills and accessory instruments (stent or embolization) to prevent undesirable damage. Therefore, operating on this subgroup of patients will be greatly dependent on surgeons’ experience and skills and a qualified multidisciplinary team (skull base surgery, vascular intervention, neurosurgery, radiology, etc.) is needed. On the other hand, although entire resection is technically feasible, surgeons also take patients’ quality of life into consideration so that they will still retain these structures. After gross tumors are removed, the introduction of adjuvant treatment, e.g., chemotherapy or immunotherapy, may help to inhibit the proliferation of tumor cells or even eliminate residual tumors.

The patients in the ENPG group had even more favorable outcomes than those reported in previous studies, with a 3-year OS rate of 68.4% in our study; several factors may account for this finding. First, the advancement of medical instruments provided superior conditions for effective and safe endoscopic resection. For example, navigation can help surgeons recognize anatomical structures and avoid injury to critical vessels and nerves; low-temperature plasma technology can facilitate hemostasis and keep the operative field clear. ICA invasion is an adverse prognosticator and the 3-year OS rates of patients with or without ICA invasion were 15.7% and 65.1%, respectively [16]. The embolization of the ICA or the use of film-coated stents in our study made it possible to resect the tumor involving the ICA and reduce the rate of residual tumor and fatal ICA hemorrhage, which may be another contributor to favorable outcomes; this is also consolidated by the same study that the 3-year OS rate of patients with ICA invasion undergoing embolization and resection of the ICA has been improved from 15.7% to 100%. The third contributor to favorable prognosis was that reconstruction of the skull base was performed on a higher proportion (35/53) of patients, as this procedure can protect the wound, accelerate the healing of the operative area and reduce the risk of postoperative bleeding. The decrease in fatal perioperative complications may transform into a survival benefit.

The survival rate of the ENPG group was similar to that of the IMRT group; except for the improvement in surgical techniques and skill, we should not neglect the influence of selection bias and potential noncomparability of some characteristics. First, patients with unresectable tumors usually have a wide extent of tumor invasion and were excluded from the ENPG group, which may result in prolonged survival compared with the total population in this group. Second, although the rT category was balanced and the statistical results of resectability characteristics indicated that the distribution of involved structures was not significant different between the two groups, in the IMRT group, there was a higher number of petrous apex, ICA and cavernous sinus involvement, which may lead to survival outcomes favoring ENPG. Third, although we matched the baseline characteristics of the patients between the ENPG and IMRT groups with the PSM method, a disparity in N stage still existed between the two groups; e.g., the proportion of patients in N0 stage was higher in the ENPG group than in the IMRT group (70.8% vs. 62.2%), as lower N stage was a favorable prognosticator for rNPC. We also noticed that a higher proportion of patients in the ENPG group received subsequent treatment (56.6% vs. 41.5%) when they experienced tumor progression after salvage treatment, which we speculated may play a vital role in prolonging survival.

Although advances in medical instruments and endoscopic surgery techniques have greatly enhanced the tumor resection rate of rNPC, in this study, the positive rate of surgical margins and/or postoperative imaging of patients was 71.7%, which was similar to that reported by Chan et al. (71.4% in clivus bone invasion and 80.0% in sphenoid sinus lateral wall invasion) [27]. Some studies also report different results in that the negative surgical margin was shown to range from 50% to 70.8% [16, 17]. The high positive rate of surgical margins in our study may be due to the following reasons. First, it was difficult to distinguish tumor invading bones from radiation-induced osteitis, and the resected bones could not be examined by intraoperative frozen sections. Second, to avoid causing severe cerebrospinal fluid leakage, we performed palliative resection of lesions adjacent to the dura mater behind the clivus. Third, we performed palliative resection of CS lesions to avoid damage to important structures such as optic nerves. Fourth, we palliatively resected lesions involving cranial nerves and brain tissues to ensure the quality of life of patients after the operation. Finally, even if the surgical margin is negative, it is difficult to achieve a safe surgical margin of 0.5–1.0 cm outside the tumor [28].

Despite the high positive rate of surgical margins in our study, the patients in the ENPG group could still achieve long-term survival by using adjuvant treatment or re-salvage ENPG or IMRT. We noticed that 5 patients received ENPG twice or more times after tumor progression, which indicated that compared with salvage IMRT, ENPG could be repeatedly performed on patients as long as they were still suitable for surgery. Meanwhile, another advantage of ENPG over IMRT was that it could resect radioresistant tumors and acquire histological specimens to guide subsequent treatment. These aforementioned traits of ENPG avoided the accumulation of radiation-induced toxicities.

No grade 3 or worse nasal adhesion occurred in the ENPG group; it may be that the endoscopic transnasal pterygoid process approach was mostly used and that the middle and inferior turbinates had been removed due to the extensive lesions. In addition, the incidence of severe treatment-related complications in patients in the ENPG group was lower than in the IMRT group (37.7% vs. 67.9%, P = 0.02), indicating that ENPG may be a safe method for the treatment of locally advanced rNPC.

Our study has some limitations. Initially, due to the intrinsic nature of retrospective studies, the clinical data of the patients during or after treatment were difficult to record completely. In addition, doctors’ experience and skills could also affect the therapeutic quality of each procedure. Finally, although we adjusted for some prognostic factors between the ENPG and IMRT groups by the PSM method, potential noncomparability of some characteristics and selection bias between groups still existed, e.g., the exclusion of patients with unresectable tumors in the ENPG arm, resectability and the adjuvant therapy; this is expected to be solved by designing prospective, random controlled clinical studies in the future.