Comparison of the Effectiveness of Radiofrequency Ablation with Stereotactic Body Radiation Therapy in Inoperable Stage I Non-Small Cell Lung Cancer: A Systemic Review and Pooled Analysis

Abstract

Introduction

Stereotactic body radiation therapy (SBRT) and Radiofrequency ablation (RFA) are two important treatment options for medically inoperable patients with stage I non-small cell lung cancer (NSCLC). Here we performed a systemic review and pooled analysis to compare clinical outcomes of these two techniques.

Methods

A comprehensive literature search for published trials from 2001 to 2012 was undertaken. Pooled analyses were performed to obtain overall survival (OS), local tumor control rates (LCR), and the adverse events. Regression analysis was conducted considering each study's proportions of stage IA and age.

Results

Thirty-one studies on SBRT (2767 patients) and 13 studies on RFA (328 patients) were eligible. LCR (95% confidence interval) at 1, 2, 3 and 5 years for RFA was 77% (70 – 85%), 48% (37 – 58%), 55% (47 – 62%), and 42% (30 – 54%) respectively, which was significantly lower than that for SBRT 97 % (96 – 98%), 92% (91 – 94%), 88% (86 – 90%), and 86% (85 – 88%) (P<.001). These differences remained significant after correcting for stage IA and age (P<.001 at 1 year, 2 years and 3 years; P=.04 at 5 years). The effect of RFA was not different from that of SBRT on OS (P>.05). The most frequent complication of RFA was pneumothorax, occurring in 31% of patients, while that for SBRT (≥grade 3) was radiation pneumonitis, occurring in 2% of patients.

Conclusions

Comparing to RFA, SBRT seems to have higher LCR, but similar OS. More studies with larger sample sizes are warranted to validate such findings.

Introduction

Nearly 25% of patients with early stage or stage I non-small-cell lung cancer (NSCLC) are considered medically inoperable,[28,45] with five-year survivals only about 15% prior to the era of stereotactic body radiotherapy (SBRT), which are significantly poorer than those who receive surgery.[45,61] Local failure was the primary cause of failure[25,48]. SBRT, which is also referred to as stereotactic ablative radiotherapy (SABR), has emerged as an important treatment option for inoperable stage I NSCLC over the last 10 years. This radiation technology allows accurate delivery of a very high dose to the tumor, while sparing normal adjacent tissue. In prospective and large retrospective multicenter studies, when an adequate dose (BED≥100 Gy) is utilized, high local control rate (LCR) of over 90% has been reported, as well as promising overall survival (OS) advantages and good tolerance compared to conventional radiotherapy.[4,46,56] The recently published results from the Radiation Therapy Oncology Group (RTOG) 0236 trial demonstrated 3-year actuarial LCR of 98%, while altogether avoiding severe toxicity in more than 75% of treated patients.[56] With this and other evidence, SBRT is the current standard of care for medically inoperable stage I NSCLC.

Radiofrequency ablation (RFA) is another promising non-surgical therapeutic alternative for small, peripheral tumors. It can be performed percutaneously, under conscious sedation and image guidance, and as an outpatient or with a short hospital stay. The use of RFA to treat lung malignancies in human was first reported in 2000.[13] The largest prospective multicenter trial of RFA for lung tumors was the RAPTURE study published in 2008, which enrolled 106 inoperable patients with 183 lung tumors including 33 primary NSCLC patients.[30] A 75% 2-year OS has been shown in patients with stage I NSCLC (n=13), and a confirmed complete response lasting at least 1 year was 88%. RFA treatment was well tolerated. Major complications were symptomatic pneumothorax needing drainage (n=27) and pleural effusion needing drainage (n=4). Minor complications were pneumothorax (n=28) or pleural effusion (n=11) not needing treatment and self-limiting intrapulmonary haemorrhage (n=3). However, the majority of RFA studies in early stage NSCLC are small case series and the reported LCR ranged greatly from 58% to 95%. [18,19,37,40]

At present, no direct comparison of clinical outcomes between RFA and SBRT has been published. Furthermore, it is often challenging for tumor board to decide between RFA and SBRT, as many patients are candidates for both treatments and it is unknown which treatment would provide superior outcome. In addition, outcome data from randomized studies are lacking, and no such prospective trial is ongoing. Therefore, we performed this systemic review and pooled analysis to provide a comparison between RFA and SBRT, for treatment of medically inoperable stage I NSCLC, with a focus on LCR, OS, and treatment-related toxicities.

Methods

A comprehensive literature search was performed using MEDLINE, Embase and Cochrane Library from 2001 to 2012. Specific search terms used included radiofrequency ablation, thermal ablation, stereotactic radiotherapy, stereotactic radiation therapy, stereotactic ablative radiotherapy, non–small cell lung carcinoma, and NSCLC. Reference lists of obtained articles were searched as well.

Inclusion and exclusion criteria

The eligibility criteria included as follows: (1) stage I NSCLC diagnosis, (2) medically inoperability, and (3) reporting on the outcome of patients after RFA or SBRT. Outcome data included local tumor control rates (LCR) at 1, 2, 3 and 5 years, and overall survival (OS) rates at 1, 2, 3 and 5 years. Local control was assessed by follow-up radiological image and defined as no tumor recurrence/progression in the primary site. Procedure-related morbidity and mortality were also analyzed. Studies about RFA followed by immediate resection or radiotherapy, or SBRT with BED <100 Gy, fraction dose <8 Gy, or using more than 5 fractions, or with less than 5 patients in RFA group and 30 patients in SBRT group were excluded. We limited the SBRT to 5 fractions or less to make the patient population more comparable to that of RFA, i.e. peripheral location. Case reports, review articles, unpublished data, and publications in languages other than English were also excluded from further analysis. Studies and/or subgroups were included if reporting sufficient data on LCR, OS and treatment related morbidity. In case of potential duplication or overlapping studies only the largest and most complete data set was included, based on published information.

Data extraction

One investigator (N.B.) identified the articles for this topic. Two investigators (N.B., X.Z.) extracted the data on study characteristics, study population, technical details and outcome measurements using a standardized Excel file and reached consensus on all items. No attempt was made to get missing data from the authors. BED10 was calculated based on the following linear-quadratic equation: BED10 = nd*[1 + d/(a/b)], where a/b=10, n and d represent the number of fractions and the dose per fraction, respectively. The 1-, 2-, 3-, and 5-year LCR and OS were extracted from each study if available, as well as the occurrence of adverse events. In studies that survival rates were not explicitly stated but Kaplan-Meier survival curves were provided, survival data were extracted from survival curves.

In stage I NSCLC the occurrence of severe adverse events (CTCAE grade 3– 5) was infrequent for all treatment modalities. The majority of SBRT studies reported zero adverse events; therefore, we added up the adverse events occurring for each treatment modality instead of pooling the estimates using a regression. Proportions and 95% confidence intervals were calculated using a binomial distribution.

Data analysis

The primary endpoint of this study was LCR at 3 years. Secondary endpoints were OS and treatment related thoracic toxicity.

A pooled analysis was used to determine weighted summary statistics for each of the treatments. The results were presented as pooled proportions of patients surviving (or local-progression free surviving) 1-, 2-, 3-, and 5- years from diagnosis, with 95% confidence intervals (95%CI). Since standard errors for the surviving proportions of patients were not consistently reported, we inferred them from the reported median follow-up times using the parametric bootstrap[14]. Application of the parametric bootstrap to survival data requires models for both the survival and censoring processes. To obtain a model for the population survival process, for each study, a model was fit to the reported survival times (at 1-, 2-, 3-, and 5- years as reported). For studies with only one reported follow-up time, a one-parameter exponential model was used. For studies with two or more reported follow-up times, a two-parameter Weibull model was used. The right tail of the population survival curve was fit to the reported surviving patient fractions using least squares. The censoring process was modeled as an exponential distribution, with median equal to the reported median follow-up time. Simulation was then used to estimate the standard errors for estimating the surviving patient fractions, based on data sets of the reported sample size, from the estimated survival and censoring processes. Since differences between the characteristics of the study populations may influence the effectiveness of different treatment modalities, regression analysis was performed considering each study's proportions of Stage IA and patients’ age. All the analyses were done using R language (www.r-project.org).

Results

Literature search

The literature search performed in July 2012 yielded 1,270 articles (Figure 1). Among them, 478 articles were found to be pertinent. 306 studies were excluded as they failed to meet the eligibility criteria based on the screening of the titles and abstracts. Of the remaining 172 references the full text were retrieved and the data were analyzed. One additional reference on RFA was retrieved through manual searches of the reference lists. Ultimately, 44 articles were found to report data of interest and fulfilled the inclusion criteria of the present study: 13 studies including 328 patients on RFA [1,5,18,19,21,27,29,30,37,40,49,50,62] and 31 studies including 2767 patients on SBRT.[3,4,7–9,11,15–17,22–24,26,32,34–36,38,39,42,44,46,52–58,60,64] There were no RCTs or non-randomized studies directly comparing the differences between RFA and SBRT. All eligible studies were single arm observational studies or subgroup of comparative studies. Table 1 and Table 2 summarize the main characteristics of these studies.

Figure 1.

Flowchart of literature search and review.

Table 1.

Characteristics for RFA Studies (2001–2012) Included in Current Meta-analysis.

First author	Year	Study. design	Sample size	Male (%)	Median/ mean age (range)	IA/I (%)	Path. (%)	Median F/U (mo)
Belfiore G16	2004	R	33	79	66 (44–75)	39	100	12
Pennathur A11	2007	R	19	42	78 (68–88)	58	100	28 (9–52)
Simon CJ17	2007	R	75	57	69 (17–94)	75	NA	21(3–74)
Lencioni R10	2008	P	33	76	67 (29–82)	100	100	15 (1–30)
Okuma T12	2010	R	7	78	70 (31–94)	NA	NA	12 (3–60)
Zemlyak A18	2010	R	12	56	74 (62–83)	NA	100	33
Ambrogi MC19	2011	P	59	79	74 (40–88)	75	100	46 (12–82)
Hess A13	2011	R	15	60	64 (42–82)	93	0	17.6 (2–31)
Hiraki T14	2011	R	50	58	75 (52–88)	76	32	37 (2–88)
Lee H20	2011	R	16	75	73±8	NA	100	56 (6–64)
Sofocleous CT21	2011	R	12	67	65 (44–81)	75	100	23
Kim SR22	2012	R	8	88	72 (61–78)	25	100	9 years
Lanuti M23	2012	R	45	40	70 (51–89)	82	100	32 (2–75)

Note: (%):F/U (mo): follow-up period in months;; Pts, patients with Stage I NSCLC; Path (%): percentage of disease with pathological confirmation; P: prospective; R: retrospective; N /A: not reported or not specified.

Table 2.

Characteristics of SBRT Studies (2001–2012) Included in Current Meta-analysis.

First author	Year	Study. design	Sampl e size	Male (%)	Median/ mean age	IA/I (%)	Path. (%)	Dose range	BED10	F/U (mo)
Nagata Y24	2005	P	45	74	77 (51–87)	71	100.0	12Gy × 4	105.6	30
Nyman J25	2006	R	45	57	74 (58–84)	40	80.0	15Gy × 3	112.5	43 (24–74)
Timmerman R26	2006	R	70	74	70 (51–86)	50	/	20Gy × 3	180	17.5 (0.6–44)
Zimmermann F27	2006	R	68	71	76	100	100.0	12.5Gy× 3; 7Gy× 5 (at 60%-isodose)	84.4;59.5	17(3–44)
Koto M28	2007	P	31	81	77 (60–83)	61	100.0	15Gy × 3; 7.5Gy × 8	112.5;105	32 (4–87)
Onishi H29	2007	R	257	NA	74 (39–92)	64	NA	4.4~35/1-14F×	117 (100–180)	38 (2–128)
Fritz P30	2008	R	40	80	74 (59–82)	55	100	30Gy×1	120Gy (isocenter)	20 (6–61.5)
Lagerwaard FJ31	2008	R	206	57	73	59	31.0	20Gy × 3; 12Gy × 5; 7.5Gy × 8	132	12
Onimaru R32	2008	P	41	69	76 (52–85)	61.0	NA	10Gy × 4; 12Gy × 4	80;105.6	27
Salazar OM33	2008	R	60	NA	NA	75	NA	13Gy × 4	119.6	NA
Baumann P7	2009	P	57	44	75 (59–87)	70.2	NA	15Gy × 3	112.5	35 (4–47)
Fakiris AJ34	2009	P	70	NA	NA	48.6	NA	20Gy × 3 (IA); 22Gy × 3 (IB)	180; 211.2	50.2
Kopek N35	2009	R	88	50	72 (47–88)	58.6	100.0	15/22.5Gy × 3	112.5; 219.5	44 (1.6–96.5)
Stephans KL36	2009	R	56	52	72 (49–89)	75.0	NA	10Gy × 5	100	19.8
Takeda A37	2009	R	63	63	78 (56–91)	60.3	82.5	10Gy × 5	100	31 (10–72)
Baba F38	2010	R	124	68	77 (29–89)	70.2	91.9	11Gy × 4 (1.6%); 12Gy × 4; 13Gy × 4	92.4; 105.6; 119.6	26 (7–66)
Bradley JD39	2010	P	91	47	71 (31–93)	72.5	86.3	18Gy × 3	151.2	18 (6–42)
Burdick MJ40	2010	R	72	52	74 (44–89)	68.1	68.1	20Gy × 3; 10Gy × 5; 5Gy × 10	180; 100; 75	36
Dunlap NE41	2010	R	40	NA	73 (54–87)	67.5	100.0	42-60/3-5FX	150	12.5 (2–35)
Hamamoto Y42	2010	R	52	70	78 (58–90)	76.8	67.3	12Gy × 4	105.6	14 (3–34)
Ricardi U43	2010	P	62	84	73 (53–83)	69.4	100.0	15Gy × 3	112.5	28 (9–60.7)
Timmerman R8	2010	P	59	38	72 (48–89)	80.0	100.0	18Gy × 3	151.2	34.4 (4.8–49.9)
Van Der Voort Van Zyp NC44	2010	P	39	NA	77 (55–87)	58.9	NA	20Gy × 3	180	17
Bral S45	2011	P	40	83	73 (54–86)	65	100	20Gy × 3; 15Gy × 4	180; 150	16 (5–33)
Lanni TB46	2011	R	45	40	76 (63–90)	71	100	12Gy × 4; 12Gy × 5	105.6; 140	36
Matsuo Y47	2011	R	101	73	77 (62–87)	58.9	100.0	12Gy × 4	105.6	31.4 (4.2–118)
Nath SK48	2011	R	58	63	79 (60–88)	58.9	100.0	10Gy × 5; 12Gy × 4; 13Gy × 4	100; 105.6; 119.6	17 (4–42)
Turzer M49	2011	R	36	28	74 (54–85)	53	73.7	15Gy × 3	112.5	13.8 (0–21)
Widder50	2011	P	202	73	76 (46–93)	NA	29	20Gy × 3; 12Gy × 5; 7.5Gy× 8	180; 140; 105	13
Senthi S6	2012	R	676	61	73 (47–92)	56	35	20Gy × 3 or 18Gy × 3; 12Gy × 5 or 11Gy × 5; 7.5Gy × 8	151·2; 115·5; 105	32.9(14.9–50.9)
Taremi M51	2012	P	108	49	73 (48–90)	79.6	69.4	20Gy × 3; 18Gy × 3; 12Gy × 4; 7.5Gy× 8; 5Gy × 10 (9.6%)	180; 151.2; 105. 6; 105; 75	19.1 (1–55.7)

Note: BED10: biological equivalent dose with α/β=10; F/U (mo): follow-up period in months; Path (%): percentage of disease with pathological confirmation; P: prospective study; R: retrospective study; NA: not reported or not specified

Local control for medically inoperable Stage I NSCLC

Forty studies reported local control rates (LCR, Fig. 1 and Table. 3). In the RFA group, 5 studies (142 patients) reported 1-year LCR, 4 studies (113 patients) reported 2-year LCR, 6 studies (248 patients) reported 3-year LCR, and 4 studies (106 patients) reported 5-year LCR. In the SBRT group, 20 studies (2,107 patients) reported 1-year LCR, 22 studies (2,137 patients) reported 2- year LCR, 21 studies (2,151 patients) reported 3-year LCR, and 6 studies (1,192 patients) reported 5-year LCR. The results of the fixed effect pooled analysis on LCR are presented in Table 3. The 1-, 2-y, 3-, 5-year LCR estimates and their 95%CI were 77% (70–85%), 48% (37–58%), 55% (47– 62%), 42% (30–54%) for RFA, and 97% (96–98%), 92% (91–94%), 88% (86– 90%), 86% (85–88%) for SBRT, respectively. The uncorrected pooled 1-, 2-, 3-, 5-year LCR for RFA were all significantly lower than that for SBRT (P-values all <.001).

Table 3.

Results of Meta-analysis for Local Control Rate (LCR)

	SBRT			RFA
	No. of study	LCR	95% CI	No. of study	LCR	95% CI	P value	P value*
1 year	20	0.97	0.96–0.98	5	0.77	0.70–0.85	< .001	< .001
2 year	22	0.92	0.91–0.94	4	0.48	0.37–0.58	< .001	< .001
3 year	21	0.88	0.86–0.90	6	0.55	0.47–0.62	< .001	< .001
5 year	6	0.86	0.85–0.88	4	0.42	0.30–0.54	< .001	.04

^*Corrected by age and percentage of stage IA; LCR: local control rate; SBRT: stereotactic body radiation therapy; RFA: radiofrequency ablation; CI: confidence interval.

Overall survival for medically inoperable Stage I NSCLC

All studies reported overall survival (OS, Fig. 2 and supplemental Table. S1). In the RFA group, 12 studies (313 patients) reported 1-year OS, 12 studies (295 patients) reported 2-year OS, 9 studies (240 patients) reported 3-year OS, and 8 studies (216 patients) reported 5-year OS. In the SBRT group, 27 studies (2,467 patients) reported 1-year OS, 26 studies (2,377 patients) reported 2-year OS, 21 studies (2,003 patients) reported 3-year OS, and 10 studies (1,503 patients) reported 5-year OS. The results of the fixed effect pooled analysis on OS are presented in Table S1. The 1-, 2-y, 3-, 5-year OS estimates and their 95% CI were 85% (80–89%), 67% (61–74%), 53% (45– 61%), 32% (22–43%) for RFA, and 85% (84–87%), 68% (66–70%), 56% (53– 59%), 40% (36–45%) for SBRT. Neither treatment had significantly different uncorrected 1-, 2-, 3- and 5-year OS (P-values all >.05).

Figure 2.

Overview of local control rate (LCR) and 95% confidence intervals for all studies and pooled estimates. A. LCR at 1 year. B. LCR at 2 years. C LCR at 3 years. D. LCR at 5 years.

Comparison between RFA and SBRT: Regression analysis

Regression analysis was conducted to adjust for the different influence of clinical factors on effectiveness of different treatment modalities. After adjusting for age and percentage of stage IA, the corrected pooled 1-, 2-, 3-, and 5-year LCR for SBRT were significantly higher than that for RFA (P-values <.05, respectively; Table. 3). RFA and SBRT therapy did not have statistically significantly different 1-, 2-, 3-, and 5-year OS (Table. S1).

Occurrence of adverse events

The total occurrence of each adverse event per treatment modality, as well as the number of patients at risk, is listed in supplemental Table S2. Overall, both RFA and SBRT studies reported limited severe adverse events. The most frequent complication of RFA was pneumothorax (Grade ≥1), which reported in 31% (95%CI: 19–43%) patients. Severe pneumothorax that required intervention (Grade ≥3) occurred in 13% (95%CI: 0–27%) of patients. The most frequent grade 3 or greater toxicity for SBRT was radiation pneumonitis (RP), occurring in 2% of patients (95%CI: 1–4%). The second frequent toxicity was rib fracture, occurring in 2% of patients (95%CI: 1–3%). The incidence of severe (grade ≥3) acute esophageal toxicity, however, was uncommon. Several studies also reported cases of grade 3/4 adverse events, but because they did not specify the types of adverse events, these were not incorporated into the count.

Discussion

To our knowledge, this is the first systemic review and pooled analysis to compare the efficacy and morbidity of RFA and SBRT for medically inoperable early stage NSCLC. A total of 44 studies, 13 for RFA and 31 for SBRT, were used in the comparison. The 1-, 2-, 3-, and 5-year LCRs, corrected for differences in each study’s proportions of stage IA and age, were significantly higher after SBRT than that after RFA treatment. The OS rates for RFA and SBRT were comparable. Both SBRT and RFA reported limited risk of severe toxicity.

A number of literature reviews that separately addressed the effectiveness of SBRT and RFA were published recently.[10,33,41,51,63] However no actual pooled analysis has been published until now. Recently, Bial et al performed an English literature review to compare RFA with SBRT in patients with early stage medically inoperable NSCLC.[6] A total of 16 studies (9 studies for RFA and 7 studies for SBRT) were used to answer this question. They found that the OS at 1 year (68.2–95% vs. 81–85.7%) and 3 years (36–87.5% vs. 42.7– 56%) was similar between patients treated with RFA and SABR, while 5-year OS was higher in SBRT (47%) than RFA (20.1–27%). Local progression rates were lower in patients treated with SBRT (3.5–14.5% vs. 23.7–43%). These results were different from the pooled estimates presented in the current study, showing better local control without significant difference in 5-year survival.. The difference might be explained by two main factors. First, Bial et al. did not use a statistical approach to analyze all the data. Second, our analysis included more studies, and especially recently published studies. We utilized 8 studies for RFA and 10 studies for SBRT which reported 5-year OS results, while only 2 RFA studies and 1 SBRT study were included in Bial’s review. A cost-effectiveness analysis published by Sher et al. compared conventional RT, SBRT, and RFA for medically inoperable Stage I NSCLC.[47] They developed a Markov model to describe health states of 65-year-oldmen with medically inoperable NSCLC after treatment with 3D-CRT, SBRT, and RFA. They found that SBRT is the most cost-effective treatment for medically inoperable Stage I NSCLC. The incremental cost-effectiveness ratio for SBRT over 3D-CRT was $6,000/quality-adjusted life-year, and the incremental cost-effectiveness ratio for SBRT over RFA was $14,100/quality-adjusted life-year. However, these results were based on a wide range of assumptions, including the efficacy of each treatment modality. For example, the local recurrence rates they used for RFA treatment was only derived from one retrospective study published by Simon et al.[49] The 3-year LCR for RFA in that study was only 43%, which was significantly lower than pooled result (55%, 95%CI: 47–62%) from this study. The possible explanation of this difference is that the study by Simon et al. included not only primary early stage NSCLC but also metastatic lung cancer (more than 40%). Since the local recurrence risk of RFA is one key variable that could change the outcome of cost-effectiveness analysis, our pooled estimate results on SBRT and RFA might be helpful to improve the predictive accuracy of their model.

A strong correlation between the size of the targeted tumors and the RFA treatment results has been reported. Higher rates of relapse were noted in tumors larger than 2–4 cm. Simon et al. reported on 75 patients with stage I NSCLC that were treated with RFA and followed for 5 years; the 5-year progression-free survival (PFS) was significantly higher in tumors less than 3 cm (47%) compared to larger tumors (25%).[49] Okuma et al. identified that the significant risk factor for local progression after RFA was tumor size ≥ 2 cm (HR 1.5, 95% CI: 1.2–2.0),[37] while Huang et al. indicated that significant difference in the risk of local progression was found in tumors more than 4 cm.[20] A recent study including 55 ablations in 45 patients with stage I NSCLC reported more local failures in tumors more than 3 cm (80%), compared with lesions < 3cm (29%).[27] All these evidence suggested that the difference in the percentage of small tumor (<3cm, stage IA patients) between the study populations might significantly influence the outcome. Therefore, in the present study, we performed a regression analysis and found that after correcting for differences in tumor size (the percentage of stage IA patients), the local tumor control rates at 1-, 2-, 3- and 5-year were still significantly better for SBRT compared with RFA (P<.001 at 1 year, 2 years and 3 years; P=.04 at 5 years).

Although the local tumor control rate was significantly higher in patients treated with SBRT, the pooled estimates of OS between SBRT and RFA were similar. These results were consistent with a recent prospective study,[43] which included 116 patients treated with sublobar resection (n = 42), RFA (n = 25), or radiotherapy (n = 49). After adjusting for age and tumor size, a significant difference in the PR rate was observed, but no significant differences in OS. The comparison of long-term survival outcomes assessing alternate therapies for high-risk patients is challenging because the comorbidity and mortality rates are high. Evidence showed that even without any cancer, the median 5-year OS for patients with severe COPD was only about 40%.[59] As many as one-third of medically inoperable patients with NSCLC would die from comorbid conditions rather than cancer.[51] It has been suggested that for medically inoperable patients with NSCLC, local tumor control and relapse-free survival might be more valid endpoint than overall survival. [41]

Though our study demonstrated that the local tumor control was significantly better for SBRT than for RFA treatment, there are certainly a number of limitations. First, this is a systemic review and pooled analysis based on observational studies. When combining observational studies, heterogeneity of populations, design, and outcome are expected, and these differences may influence the pooled estimates. However, despite these challenges, when no RCTs or retrospective studies directly comparing RFA and SBRT are available, a systemic review and pooled analysis of observational studies might be a valid method for assessing efficacy and effectiveness, and could provide useful evidence to inform the decision whether more evidence is needed. Moreover, a regression analysis could correct for some potential differences between study populations or designs. Second, most RFA studies were small case series with relatively short follow up, which led to wide variance of LCR and OS estimated by formula. Four RFA studies were newly published.[2,12,31,43], while we were preparing this manuscript. A local progression rate of more than 20%, 1y-LCR 68.9%, and 2y-LCR 59.8% were reported, which were similar to our pooled results. Thirdly, definitions of local progression are not always consistent across the published reports. The wide range (3% to 50%) of local tumor progression after earlier RFA studies may be partially explained by these variations. Last, we did not look at trials presented only in their abstract form or at unpublished studies, nor at studies without separate data for early stage NSCLC patients. This emphasizes the importance that authors should fully document the characteristics of the study populations in their papers.

In conclusion, this systemic review and pooled analysis demonstrated that SBRT provided superior 1-, 2-, 3- and 5-year local tumor control over RFA, even when corrected for patients age and tumor size ≤ 3 cm. Therefore, at present, SBRT is still the most effective local treatment for inoperable Stage I NSCLC patients, and RFA should be offered only if patients are not candidates for SBRT. However, caution should be taken due to the relatively limited number of RFA trials and short follow-up time. More studies with larger sample sizes for RFA treatment are needed.

Figure 3.

Overview of overall survival (OS) and 95% confidence intervals for all studies and pooled estimates. A. OS at 1 year. B. OS at 2 years. C OS at 3 years. D. OS at 5 years.

Summary.

A systemic review and pooled analysis was performed to compare SBRT and RFA in inoperable stage I NSCLC. A comprehensive literature search for published trials from last 15 years was undertaken. LCR at 1, 2, 3 and 5 years for RFA was significantly lower than that for SBRT (P<.001). These differences remained significant after correcting for stage IA and age. However, caution is warranted due to the limited number of RFA studies.