Methods

The National Cancer Database (NCDB) is a joint project of the Commission on Cancer (CoC) of the American College of Surgeons and the American Cancer Society, and consists of deidentified information regarding tumor characteristics, patient demographics, and patient survival for approximately 70% of the US population.^15–24 All pertinent cases are reported regularly from CoC-accredited centers and compiled into a unified data set, which is then validated. The NCDB contains information not captured in the SEER database, including details regarding use of systemic therapy. The data used in the study were derived from a deidentified NCDB file (2004–2014). The American College of Surgeons nor the CoC have not verified and neither are responsible for the analytic or statistical methodology used or the conclusions drawn from these data by the investigators. As all patient information in the NCDB database is deidentified, this study was exempt from Institutional Review Board evaluation.

The inclusion criterion for this study was newly diagnosed, primary NSCLC treated with upfront surgery followed by postoperative chemotherapy and RT. Patients who received neoadjuvant chemotherapy and/or RT were excluded. Surgery was defined as an oncologic-quality procedure, and hence local tumor destruction/excision techniques (including laser, cryosurgery, and electrocautery) constituted exclusion. RT was defined as external-beam RT to a dose of 45 to 74 Gy.^5,6 Per multiple other NCDB publications, concurrent therapy referred to starting chemotherapy and RT within 14 days of each other, with the remainder designated as sequential.^25,26 Patients with an interval of >6 months from surgery to adjuvant therapy, or between adjuvant RT and chemotherapy, were excluded, per other work, in order to differentiate between potential salvage and definitive therapy.⁵ No patients were excluded who died within a certain point after surgery, similar to prior investigations⁵ and sensitivity analyses demonstrating no impact on results when this cohort was eliminated.⁶ Lastly, although PORT is recommended for pN2 disease, we allowed pN0 and pN1 disease for numerous reasons, including the lack of prospective assessment to date on whether these populations are at higher risk of locoregional recurrence; retrospective studies have suggested no differences.²⁷ We also allowed inclusion of patients with pN1, M− disease owing to the lack of direct randomized comparisons between postoperative CRT and chemotherapy alone, together with the fact that PORT is still often delivered in these patients.⁶ We also allowed patients with pN0, M− disease (which constituted a large minority of this cohort) for multiple reasons. First, postoperative CRT cannot be disregarded altogether for pT3–4, pN0 disease; these cases may still be at enough risk of locoregional failure (eg, based on extent of lymphovascular invasion, lymph node ratio, and other factors) to warrant adjuvant therapy. Additionally, many institutions make individualized decisions on adjuvant local therapy based on precise surgical findings, in light of data showing a connection between local failure and distance from tumor to ink.^28,29

In accordance with the variables in NCDB files, information collected on each patient broadly included demographic, clinical, and treatment data. All statistical tests were 2-sided, with a threshold of P<.05 for statistical significance, and were performed using STATA, version 14 (StataCorp LP). Multivariable logistic regression modeling was used to determine characteristics that were predictive for receipt of cCRT. The Kaplan-Meier method was used for survival analysis, and comparisons between groups were performed with the log-rank test. Overall survival (OS) was defined as the interval between the date of diagnosis and the date of death, or censored at last contact. Univariate analysis was performed to determine factors associated with OS, and subsequently Cox multivariate analysis was performed including variables that were either significant or showed a strong trend to statistical significance on univariate analysis. The proportional hazards assumption was checked graphically using log–log plots.

To account for indication bias, propensity score matching (PSM) was used to compare patients between groups. PSM is a method that creates quasi-case/control pairs using a retrospective cohort in an effort to account for the recorded and unrecorded confounding variables.^30–32 Propensity scores were calculated using a multivariable logistic regression model, with the dependent variable being receipt of particular treatment paradigm and the independent variables being those that were statistically significant for correlation with OS on multivariate analysis. Patients were matched 1:1 without replacement

Table 1.

Cohort Characteristics and Factors Associated With Receiving Concurrent CRT

View Table

View Table

Results

A complete flow diagram of patient selection is provided in supplemental eFigure 1 (available with this article at JNCCN.org). A total of 4,921 patients met study analysis criteria (Table 1). Of these, 3,475 (71%) were M− and 1,446 (29%) were M+; 2,271 (46%) received sCRT and 2,650 (54%) underwent cCRT. Multivariable logistic regression analysis revealed that independent predictors of cCRT delivery included lower income, treatment at a nonacademic facility, diagnosis at earlier periods (2004–2008), histology of NSCLC not otherwise specified, and M+ disease (P<.05 for all). The association with pathologic nodal classification is more difficult to interpret; with pN0 as the reference, patients with pN2 disease were less likely to receive cCRT (P<.001), but the very few patients who were evidently pN3 more often underwent cCRT (P=.019).

Median follow-up was 30 months (interquartile range, 17–52 months). In all patients, median OS for patients who underwent sCRT was 50 versus 37 months for cCRT (P<.001) (supplemental eFigure 2A. When separating these patients by margin status, sCRT displayed a higher OS in the M− (55 vs 40 months; P<.001) and M+ groups (36 vs 31 months; P=.011) (Figure 2A,B). Following PSM (supplemental eTables 1–3), these findings persisted among all patients (50 vs 39 months; P=.001; supplemental eFigure 2B) and in the M− group (55 vs 40 months; P<.001; Figure 2C), but showed equivalence in the M+ group (36 vs 33 months; P=.745; Figure 2D).

Given that not all patients with M+ disease had available data on extent of resection, when substratifying them according to resection status and evaluating sCRT versus cCRT, both groups experienced similar OS in R1 (38 vs 37 months; P=.368) and R2 resections (25 vs 21 months; P=.553) (Figure 3A, B). These populations, namely the patients with R2 resections, were too small to reliably undergo PSM.

When analyzing the sCRT subset only, patients who received chemotherapy followed by RT experienced improved OS compared with those receiving RT followed by chemotherapy (53 vs 39 months; P=.002; Figure 3C). Following PSM (supplemental eTable 4), statistical equipoise remained between groups (46 vs 40 months; P=.229; Figure 3D).

In the overall cohort, there were several predictors of OS on univariate analysis (Table 2). After multivariate adjustment for potential confounding factors (Table 2), factors independently associated with poorer OS included advancing age, male sex, high comorbidity index, Medicaid insurance, treatment at a non-academic facility and at earlier periods (2004–2008), adenosquamous histology, differentiation status, lower lobe location, T and N classification, surgical margin status, and receipt of cCRT (P<.05 for all).

Discussion

Our study of a contemporary national database most notably demonstrates that sCRT is associated with improved outcomes for patients with M− disease, and statistical equipoise for those with M+ disease. Additionally, sCRT performed in a chemotherapy-first manner produced statistically equivalent survival as administering RT-first. Collectively, these hypothesis-generating data do not imply causation (as discussed further) and should ideally be tested prospectively, but do illustrate comparative outcomes between sCRT and cCRT (and sequencing of sCRT) for the first

View Full Size

Kaplan-Meier overall survival curves comparing sequential chemoradiotherapy (gray) versus concurrent chemoradiotherapy (black) in patients with (A) margin-negative disease and (B) margin-positive disease, and in propensity score–matched patients with (C) margin-negative disease and (D) margin-positive disease.

Citation: Journal of the National Comprehensive Cancer Network J Natl Compr Canc Netw 16, 5; 10.6004/jnccn.2018.7007

View Full Size

Kaplan-Meier overall survival curves comparing sequential chemoradiotherapy versus concurrent chemoradiotherapy in patients with (A) R1 resection and (B) R2 resection. Comparison of the sequential chemoradiotherapy cohort with receipt of (C) chemotherapy-first versus radiotherapy-first, and (D) after propensity score matching.

Citation: Journal of the National Comprehensive Cancer Network J Natl Compr Canc Netw 16, 5; 10.6004/jnccn.2018.7007

In the absence of prospective data, it is readily acknowledged that retrospective selection biases may limit interpretation of this work. Notably, the cCRT group consisted of higher-risk disease (illustrating adherence to national guidelines), thus likely necessitating more aggressive therapy. These patients may have also likely received suboptimal chemotherapy doses if concurrent RT was delivered. PSM is a powerful tool to control for many of these biases, but only the variables available in the NCDB can be controlled. Although equipoise between both groups was demonstrated for M+ disease, it is entirely possible that cCRT offers a real benefit to higher-risk patients (as observed by examining both R1 and R2 subgroups). However, that benefit is not enough to overcome multiple poor-prognostic factors when comparing to the sCRT group. A converse argument, however, is that delivering aggressive cCRT after lung resection may incur substantially more treatment-related morbidities, some of which may predispose toward earlier mortality, and that the increased toxicity from cCRT could diminish oncologic benefit (potentially even in the higher-risk patient population). Because the NCDB does not record other end points such as locoregional control, cancer-specific survival, or toxicities, it is unclear whether OS differences are cancer-related or morbidity-related. There is less reason to believe both groups experienced similar postoperative or 30-day mortality, given the similar initial slopes of the Kaplan-Meier curves in both groups, without a drastic decline in either group. Nevertheless, single- or multi-institutional data may be able to provide further insight, albeit at the cost of much smaller sample sizes. For instance, Kim et al³⁴ demonstrated an increase in locoregional recurrence-free survival, but no differences in OS, with cCRT in M− cases (n=68).

Likewise, the finding of statistically similar OS for chemotherapy-first and RT-first sCRT can be biased in either direction, especially in the absence of locoregional control as an end point in the NCDB. A major shortcoming of the NCDB is the absence of information on number of received chemotherapy cycles; although all patients received full-dose RT, those who received initial RT may not have tolerated optimal postoperative chemotherapy, thus compromising outcomes. This group may also have had larger-volume residual disease and/or rapid postoperative recurrence, neither of which is captured by the NCDB. However, the counterargument is the principal rationale for why chemotherapy is most commonly delivered first: namely, that virtually all postoperative patients who warrant combined-modality therapy are at high risk of distant failure, for which early chemotherapy administration is most necessary.

Based on these unavoidable biases in this and other similar NCDB studies,^5,6 we posit a conservative interpretation of these data: there may not be a benefit to routinely delivering cCRT in patients with M− and M+ disease, but well-selected patients may very well benefit from this approach. It is more intuitive that the relative benefit to concurrent therapy is greater with the larger amount of postoperative disease remaining. Although the NCDB does not code for size/volume of gross residual disease, it is still reasonable to consider cCRT or RT-first sCRT for larger-volume gross residual disease, or potentially even in patients with rapid growth following surgery. Although, most commonly chemotherapy was administered before PORT in patients undergoing sCRT, select patients may very well have an initially higher risk of local/regional failure instead of distant failure.¹⁴ We recommend that decision-making should continue to be made judiciously, while assessing the balance between oncologic benefit and potential adverse events. Factors such as age, performance status, and tolerance of concurrent therapy (along with tolerance of surgery and/or any postoperative events) are essential to consider when determining optimal postoperative management, and despite these data, we highly advocate for individualized management in this circumstance.

Nevertheless, it is also interesting to evaluate factors and trends associated with cCRT versus sCRT delivery. Based on results from multivariable

Table 2.

Univariate and Multivariate Cox Proportional Hazards Model for Overall Survival

View Table

View Table

The independent association between treatment at an academic facility and OS on multivariate analysis has far-reaching implications on patient counseling and management by oncologists and referring providers. There are several potential reasons for this, including greater multimodality coordination, streamlined diagnostic processes, technical expertise, ancillary support staff for close toxicity monitoring, and potentially the availability of salvage therapies (or clinical trials). Nevertheless, this finding may impact any case of postoperative CRT and could warrant revisions in patterns of patient education.

Although the NCDB provides a unique platform with which to study this important clinical question, the present investigation has some additional limitations to those previously discussed herein. First, the NCDB does not keep track of several other factors, including chemotherapy cycles/agents, performance/functional status, lymph node ratio, lymphovascular invasion, or nature of preoperative workup (including brain MRI and PET staging). It also offers very limited or no information on RT field design, treatment volumes, or techniques. Information in technique is especially important because advanced techniques such as intensity-modulated RT can potentially reduce toxicity risk in the postoperative setting. Second, no information is provided regarding surgeon experience, reexcision rates, and toxicity-related deaths. Third, the NCDB does not allow for an assessment of subsequent lines of treatment (eg, reirradiation, further systemic and/or targeted therapy), which could influence OS. Nevertheless, the known shortcomings of a national, large-volume database do not undermine the necessity for further prospective investigation; the ongoing LungART trial will likely help address several unresolved PORT-related controversies (ClinicalTrials.gov identifier: NCT00410683.

Conclusions

This is the largest study to date evaluating sCRT versus cCRT in the postoperative setting. There seems to be no noticeable benefit to routinely delivering cCRT in M− or M+ disease, and sequencing sCRT with either chemotherapy-first or RT-first seemed to