A population-based study of the effectiveness of stereotactic ablative radiotherapy versus conventional fractionated radiotherapy for clinical stage I non-small cell lung cancer patients

PubMed for published reports using the keywords ([stereotactic radiotherapy] OR [stereotactic body radiotherapy] OR [stereotactic ablative radiotherapy] OR [SBRT] OR [SABR]) AND ([non-small cell lung cancer] OR [NSCLC]) AND ([survival] OR [OS]) was searched on Sep 2nd 2017, for evidence regarding the efficacy of SABR vs. CFRT. In addition to the aforementioned SPACE trial, we identified another small (n = 50) randomized study showing better treatment efficacy for SABR compared to CFRT in peripheral NSCLC.11 However, patients of various stages (stages I–IV) were included in the study, and the results of stage I patients were not reported. We also found a meta-analysis (published in 2010) that reported better overall survival (OS) for SABR compared to CFRT, but all of the included studies were nonrandomized.12 In addition, none of the included studies directly compared SABR and CFRT.12 We also found four subsequent single institutional nonrandomized studies from Europe or North America and two subsequent population-based studies from North America.13,14,15,16,17,18 However, to the best of our knowledge, no population-based study from Asia has compared SABR vs. CFRT for treating early-stage NSCLC.

Given the relatively limited evidence on this topic, we investigated the effectiveness of SABR vs. CFRT for non-operated early-stage NSCLC in a population-based sample from Taiwan.

The Health and Welfare Data Science Center (HWDC) database is a set of databases providing complete information regarding the Taiwan cancer registry, death registry, and reimbursement data for the whole Taiwanese population provided by the Bureau of National Health Insurance (NHI).19 The high quality of this cancer registry has been reported.20 NHI is a single-payer, compulsory social insurance program that provides insurance coverage to the majority of citizens in Taiwan.21 All of the above data were included in the HWDC with deidentified personal identifiers.

A flowchart showing the identification of study cases appears in Figure 1 as suggested by the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guideline.22 Briefly, we identified stage I histology-documented NSCLC patients diagnosed from 2007 to 2013 who received either CFRT or SABR without surgery. We used the date of diagnosis as the index date. We determined the explanatory variable of interest (CFRT vs. SABR) based on the record in the cancer registry using the dose/fractionation recommended by the National Comprehensive Cancer Network (NCCN) NSCLC guideline (CFRT: 60–70 Gy in 1.8–2 Gy/fraction; SABR: 25–34 Gy/1 fraction, or 45–60 Gy/3 fractions, or 48–50 Gy/4 fractions, or 50–55 Gy/5 fractions, or 60–70 Gy/8–10 fractions).1 We also collected other covariate and outcome data from the HWDC. We decided on covariates (age, sex, residency, comorbidity, histology, T stage, period, use of positron emission tomography [PET], use of systemic therapy, and previous cancer) based on our clinical and HWDC-related research experiences as well as previous reports.23,24,25 The covariates were defined as follows. Patient residency was classified as northern Taiwan or elsewhere. We included this covariable because geographic practice variation had been report in the literature26 and we felt it might influence treatment choice in our clinical and research experiences.24 Comorbidity was defined as with or without a modified Carlson comorbidity score ≥1, as used in our previous NHI cancer study.24 Histology was classified as adenocarcinoma or non-adenocarcinoma. T stage was classified as T1 vs. T2. Period was classified as 2007–2009 or 2010–2013 because staging was changed since 2010. Use of PET, systemic therapy, and previous cancer was classified as yes or no. We used the national death registry to determine survival status and used OS as our endpoint, as initially completed in the SPACE trial.10 This study was approved by the Research ethics committee at our institute (CMUH104-REC-002).

Figure 1

We used the Kaplan-Meier method and log-rank test to compare crude OS between patients treated with SABR vs. CFRT. We further used inverse probability weighting (IPW) based on the propensity score (PS) as the primary means of analysis to address the nonrandomization of treatment.27 We modeled the use of SABR vs. CFRT as the dependent variable and the above covariates as independent variables and used logistic regression to model the probability of receiving SABR. Then we used the logit of the probability as the PS, as described previously.27 Tabulation and standardized differences were used to assess the balance of covariates between treatment groups.27,28 We used a weighted Cox model to compare OS between treatment groups for the entire follow-up period (censored on December 31, 2015).27,29 We used bootstrap analysis to obtain confidence intervals and p-values, as described previously.30 For OS results with statistical significance, we further calculated the E-factor to evaluate the robustness of our finding regarding potential unmeasured confounder[s] as suggested in the recent literature.31

In the first supplementary analysis (SA-1), we constructed a subgroup based on PS matching and used a robust variance estimator to compare OS and lung cancer-specific survival of patients treated with SABR vs. CFRT. We also used cause of death to obtain lung cancer-specific survival (LCSS). In the second supplementary analysis (SA-2), we constructed another subgroup by PS matching limited to cases from 2011 to 2013 to use the additional covariate (performance status, classified as Eastern Cooperative Oncology Group [ECOG] 0–2 vs. 3–4) in PS modeling to compare the survival of patients treated with SABR vs. CFRT. We limited to this period [2011–2013] because performance information was available in Taiwan cancer registry since 2011. SAS 9.4 (SAS Institute, Cary, NC) was used for all analyses.

As shown in Figure 1, we found 238 clinical stage I NSCLC patients who received either SABR or CFRT from 2007 to 2013 were included in our primary analysis. The characteristics of these patients are described in Table 1. Although an imbalance in covariate distribution was observed before PS weighting such as higher percentage of patients with comorbidity received SABR [32%] than those without comorbidity [17%], a good balance of covariates and small standardized differences (≤ 0.25) were observed for all covariates after we adjusted for PS weighting.28,32

Table 1

Patient characteristics for the whole study population

After a median follow-up of 28 months (range 2–105), 171 patients were found to have died (40 SABR and 131 CFRT). We found that SABR led to higher crude OS compared to CFRT, as shown in Figure 2. The 5-year OS rates for SABR and CFRT were 31% and 20%, respectively (log-rank test, p = 0.0008). After IPW, OS was not significantly different between those treated with SABR vs. CFRT (SABR vs. CFRT: IPW adjusted hazard ratio [HR] 0.586, 95% confidence interval 0.264–1.101, p = 0.102).

Figure 2

In SA-1, a good balance of covariates was observed with small standardized differences (≤ 0.25) for the PS-matched subgroup (n = 120; see Table 2). Compared to CFRT, the OS (HR 0.672, p = 0.039) and LCSS (HR 0.529, p = 0.007) of patients receiving SABR were superior. The observed HR 0.672 for OS could be explained away by an unmeasured confounder that was associated with both selections of SABR/CFRT and live/death by a risk ratio of 1.96 fold each, but weaker confounding could not do so. The OS curve is shown in Figure 3. In SA-2, well-balanced covariates were observed with small standardized differences (≤ 0.25) when cases were limited to 2011 to 2013 with an available performance status (n = 52; see Table 3), although there were some imbalances before matching such as those with poor performance status [ECOG 3~4] were more likely to receive SABR [60%] than those with acceptable performance status [33%]. We found SABR was associated with further improvement in hazard for death (HR 0.381, p = 0.016) compared to CFRT, as seen in Figure 4. The observed HR 0.381 for OS could be explained away by an unmeasured confounder that was associated with both selections of SABR/CFRT and live/death by a risk ratio of 3.29 fold each, but weaker confounding could not do so.

Table 2

Patient characteristics in the first supplementary analysis

Table 3

Patient characteristics in the second supplementary analysis

Figure 3

Figure 4