Thoracoscopic surgery versus open surgery for lung metastases of colorectal cancer: a multi-institutional retrospective analysis using propensity score adjustment†

Abstract

INTRODUCTION

Although there has been long-standing controversy over the use of thoracoscopic lung surgery for pulmonary metastasectomy [1–3], the ratio of the thoracoscopic approach to the open thoracotomy procedure has increased over time [4], and several papers have reported optimal outcomes from thoracoscopic surgery [5]. The main criticism of this method has been on the issue of tumour identification [6, 7]. The thoracoscopic approach may preclude precise tumour identification using finger palpation, although several reports have shown the superiority of recent multidetector computed tomography [7, 8]. Moreover, a thoracoscopic approach may preclude a sufficient surgical margin because of the limited insertion angle of surgical instruments. Conversely, a recent systematic review comparing thoracoscopic and open approaches reported that overall survival was equivalent between the methods [9].

In this study, we retrospectively compared the outcomes of pulmonary metastasectomy of colorectal cancer using the open thoracotomy approach or the thoracoscopic approach that were not examined in a previous study by the Metastatic Lung Tumour Study Group of Japan [10]. To adjust for possible bias inherent to the choice of approach (open thoracotomy versus thoracoscopy), propensity score adjustment was used [11].

METHODS

Patients

We retrospectively reviewed the database of the Metastatic Lung Tumour Study Group of Japan, which was established in 1984; the database consists of 25 institutions to date and it has collected data on patients undergoing pulmonary metastasectomy with curative intent. The surgical indications and the mode of operation were decided at each institution independently, and postoperative follow-up was performed at the discretion of each institution. Information on patients who were disease-free after lung resection was limited in this registry, and we used overall survival data for analysis, as performed in a previous study from our research group [10]. The date of pulmonary surgery was defined as the starting point of the study period, and the overall survival after pulmonary metastasectomy was evaluated as the end point of the study. The disease-free interval was defined as the time between the date of initial treatment for the primary tumour and the date of pulmonary metastasis diagnosis.

In this registry, 2234 patients who underwent pulmonary metastasectomy of colorectal cancer were identified. After excluding the patients with incomplete data (lack of age information, 223; lack of information of final observation, 119; lack of primary tumour operation date, 215; lack of information on mode of approach (open thoracotomy versus thoracoscopy), 178; lack of information of the number of pulmonary metastases, 88), data from 1411 patients were available for analysis. Because thoracoscopic surgery was rarely performed before 1998 (Table 1), data from patients after 1999 were used for subsequent propensity score adjustment analysis to compare the results of open thoracotomy and thoracoscopy. Thus, data from 1047 patients were analysed in this study.

The research review boards of the affiliated institutions approved this retrospective database study in accordance with the Declaration of Helsinki. The need for informed consent from patients was waived as long as patient data remained anonymous.

Statistical methods

The Wilcoxon rank sum test was used for continuous covariates, and Fisher’s exact test was used for categorical covariates when comparing the groups. Overall survival was calculated from the date of the first pulmonary metastasectomy to the date of the last follow-up or death. The probability of survival was estimated using the Kaplan–Meier method, and comparison between the groups was performed using the log-rank test.

To adjust for possible bias inherent to the choice of approach (open thoracotomy versus thoracoscopy), propensity scores were used. For data balancing, stratification was adopted [11]. The variables included in the propensity score model were as follows: age, sex, location of primary colorectal cancer, completeness of resection for primary tumour, presence/absence of chemotherapy for primary tumour, presence/absence of extra-pulmonary metastasis, disease-free interval, preoperative number of metastases, carcinoembryonic antigen level (low or high) and era of lung surgery. Because a significant difference in survival was noticed according to the era of lung surgery, the era was included in the propensity score model to adjust for patients’ background and management factors, which were not itemized in the model (Fig. 1). The population was then divided into quintiles according to the propensity score. Within each quintile, the mean propensity scores of the open group and the thoracoscopy group were compared. Hosmer–Lemeshow tests for goodness of fit and c-statistics were evaluated to confirm good calibration and good discrimination, respectively. After checking the balance between the groups, a multivariate Cox proportional hazards model including covariates of mode of approach (open or thoracoscopy) and other operative and postoperative covariates were evaluated. The latter included the pathological number of metastasis, procedure (non-anatomical or anatomical resection), presence/absence of hilar and/or mediastinal lymph node metastasis, completeness of resection for pulmonary metastasis, maximum tumour size, and presence/absence of chemotherapy for pulmonary metastasis. The propensity score was entered as a continuous variable to adjust for heterogeneity between the two groups. The definition of the radiological number of metastases was the number of metastases detected using preoperative computed tomography. The definition of the pathological number of metastases was the number of metastases that were resected and pathologically diagnosed as metastases from primary colorectal cancer. The definition for non-anatomical resection was partial resection, while the definition for anatomical resection was segmentectomy, lobectomy and pneumonectomy in this study. In the analysis, a combined procedure was categorized as non-anatomical, combined non-anatomical and anatomical, and anatomical. In the Cox model, the stepwise backward elimination method with a probability level of 0.15 was used to select the most powerful sets of outcome predictors. The validity of the proportional hazards assumption was tested by examining the −ln(−ln(survival)) plot and Schoenfeld residuals.

Figure 1:

Overall survival according to the era of lung operation. Significant differences in survival were noted (P = 0.0046).

Open in new tabDownload slide

To evaluate the difference in the power of tumour identification between open thoracotomy surgery and thoracoscopic surgery, the difference between the radiological tumour number and resected tumour number (delta_num) was also evaluated before and after propensity score adjustment.

All statistical analyses were performed using the statistical software package Stata/SE 13.1 (Stata-Corp, College Station, TX, USA). A P-value  <0.05 was considered statistically significant.

RESULTS

Patient characteristics

The characteristics of the 1047 patients enrolled in the subsequent analyses are shown in Table 2. Over time, patients’ survival rates improved (Fig. 1, P = 0.0046). There were 647 patients in the open group and 400 in the thoracoscopy group. Several differences in background characteristics were revealed (radiological number of metastases, carcinoembryonic antigen level, era of surgery, pathological number of metastases and chemotherapy for pulmonary metastasis). There were 671 cases with a single metastasis, 376 cases with multiple metastases, and 139 scheduled bilateral operations. In a crude analysis of overall survival before propensity score adjustment, the thoracoscopy group showed better survival than the open group (Fig. 2, log-rank test: P = 0.0121).

Figure 2:

Overall survival according to surgical approach before propensity score adjustment. Thoracoscopic surgery shows better survival than open thoracotomy surgery (P = 0.0121).

Open in new tabDownload slide

Table 2:

Open in new tab

Patients’ characteristics

Variables (number of objectives) .	Open n = 647 .	Thoracoscopy n = 400 .	P-value .
Preoperative variables
Age (1047)	64.4	64.7	0.344
	(95% CI: 63.6–65.1)	(95% CI: 63. 7–65.7)
Sex (1045)			0.217
Male	398	230
Female	248	169
Primary site (1047)			0.177
Colon	276	194
Rectum	343	192
Multiple or unknown	28	14
Completeness of resection for primary tumour (885)			0.167
Complete	496	301
Incomplete	48	40
Chemotherapy for primary tumour (483)			0.218
Yes	12	14
No	274	183
Extra-pulmonary metastasis (836)			0.092
Yes	195	132
No	333	176
Disease-free interval (months) (1028)	25.8	22.6	0.348
	(95% CI: 23.6–28.1)	(95% CI: 20.4–24.8)
Radiological number of metastasis (1047)			<0.001
1	382	285
2	129	73
3	70	24
≥4	66	18
Carcinoembryonic antigen level (958)			0.016
Normal	379	255
High	220	104
Era of operation (1047)			<0.0001
2014–2009	206	236
2008–2004	278	111
2003–1999	163	53
Operative and postoperative variables
Operative procedure (1043)			0.092
Non-anatomical	382	213
Combined non-anatomical and anatomical	20	19
Anatomical	241	168
Completeness of resection for pulmonary metastasis (1015)			1.000
Complete	599	380
Incomplete	22	14
Maximum tumour size (cm) (1018)	2.23	2.06	0.101
	(95% CI: 2.12–2.34)	(95% CI: 1.93–2.19)
Lymph node metastasis (604)			0.622
None	317	227
N1	17	11
N2	16	16
Pathological number of metastases (1047)			<0.001
1	383	288
2	101	48
3	69	39
≥4	94	25
Chemotherapy for pulmonary metastasis (1047)			<0.0001
Yes	161	151
No	486	249

Variables (number of objectives) .	Open n = 647 .	Thoracoscopy n = 400 .	P-value .
Preoperative variables
Age (1047)	64.4	64.7	0.344
	(95% CI: 63.6–65.1)	(95% CI: 63. 7–65.7)
Sex (1045)			0.217
Male	398	230
Female	248	169
Primary site (1047)			0.177
Colon	276	194
Rectum	343	192
Multiple or unknown	28	14
Completeness of resection for primary tumour (885)			0.167
Complete	496	301
Incomplete	48	40
Chemotherapy for primary tumour (483)			0.218
Yes	12	14
No	274	183
Extra-pulmonary metastasis (836)			0.092
Yes	195	132
No	333	176
Disease-free interval (months) (1028)	25.8	22.6	0.348
	(95% CI: 23.6–28.1)	(95% CI: 20.4–24.8)
Radiological number of metastasis (1047)			<0.001
1	382	285
2	129	73
3	70	24
≥4	66	18
Carcinoembryonic antigen level (958)			0.016
Normal	379	255
High	220	104
Era of operation (1047)			<0.0001
2014–2009	206	236
2008–2004	278	111
2003–1999	163	53
Operative and postoperative variables
Operative procedure (1043)			0.092
Non-anatomical	382	213
Combined non-anatomical and anatomical	20	19
Anatomical	241	168
Completeness of resection for pulmonary metastasis (1015)			1.000
Complete	599	380
Incomplete	22	14
Maximum tumour size (cm) (1018)	2.23	2.06	0.101
	(95% CI: 2.12–2.34)	(95% CI: 1.93–2.19)
Lymph node metastasis (604)			0.622
None	317	227
N1	17	11
N2	16	16
Pathological number of metastases (1047)			<0.001
1	383	288
2	101	48
3	69	39
≥4	94	25
Chemotherapy for pulmonary metastasis (1047)			<0.0001
Yes	161	151
No	486	249

Wilcoxon rank-sum (Mann–Whitney) test for continuous covariates.

Fisher's exact test for categorical covariates.

CI: confidence interval.

Table 2:

Open in new tab

Patients’ characteristics

Variables (number of objectives) .	Open n = 647 .	Thoracoscopy n = 400 .	P-value .
Preoperative variables
Age (1047)	64.4	64.7	0.344
	(95% CI: 63.6–65.1)	(95% CI: 63. 7–65.7)
Sex (1045)			0.217
Male	398	230
Female	248	169
Primary site (1047)			0.177
Colon	276	194
Rectum	343	192
Multiple or unknown	28	14
Completeness of resection for primary tumour (885)			0.167
Complete	496	301
Incomplete	48	40
Chemotherapy for primary tumour (483)			0.218
Yes	12	14
No	274	183
Extra-pulmonary metastasis (836)			0.092
Yes	195	132
No	333	176
Disease-free interval (months) (1028)	25.8	22.6	0.348
	(95% CI: 23.6–28.1)	(95% CI: 20.4–24.8)
Radiological number of metastasis (1047)			<0.001
1	382	285
2	129	73
3	70	24
≥4	66	18
Carcinoembryonic antigen level (958)			0.016
Normal	379	255
High	220	104
Era of operation (1047)			<0.0001
2014–2009	206	236
2008–2004	278	111
2003–1999	163	53
Operative and postoperative variables
Operative procedure (1043)			0.092
Non-anatomical	382	213
Combined non-anatomical and anatomical	20	19
Anatomical	241	168
Completeness of resection for pulmonary metastasis (1015)			1.000
Complete	599	380
Incomplete	22	14
Maximum tumour size (cm) (1018)	2.23	2.06	0.101
	(95% CI: 2.12–2.34)	(95% CI: 1.93–2.19)
Lymph node metastasis (604)			0.622
None	317	227
N1	17	11
N2	16	16
Pathological number of metastases (1047)			<0.001
1	383	288
2	101	48
3	69	39
≥4	94	25
Chemotherapy for pulmonary metastasis (1047)			<0.0001
Yes	161	151
No	486	249

Variables (number of objectives) .	Open n = 647 .	Thoracoscopy n = 400 .	P-value .
Preoperative variables
Age (1047)	64.4	64.7	0.344
	(95% CI: 63.6–65.1)	(95% CI: 63. 7–65.7)
Sex (1045)			0.217
Male	398	230
Female	248	169
Primary site (1047)			0.177
Colon	276	194
Rectum	343	192
Multiple or unknown	28	14
Completeness of resection for primary tumour (885)			0.167
Complete	496	301
Incomplete	48	40
Chemotherapy for primary tumour (483)			0.218
Yes	12	14
No	274	183
Extra-pulmonary metastasis (836)			0.092
Yes	195	132
No	333	176
Disease-free interval (months) (1028)	25.8	22.6	0.348
	(95% CI: 23.6–28.1)	(95% CI: 20.4–24.8)
Radiological number of metastasis (1047)			<0.001
1	382	285
2	129	73
3	70	24
≥4	66	18
Carcinoembryonic antigen level (958)			0.016
Normal	379	255
High	220	104
Era of operation (1047)			<0.0001
2014–2009	206	236
2008–2004	278	111
2003–1999	163	53
Operative and postoperative variables
Operative procedure (1043)			0.092
Non-anatomical	382	213
Combined non-anatomical and anatomical	20	19
Anatomical	241	168
Completeness of resection for pulmonary metastasis (1015)			1.000
Complete	599	380
Incomplete	22	14
Maximum tumour size (cm) (1018)	2.23	2.06	0.101
	(95% CI: 2.12–2.34)	(95% CI: 1.93–2.19)
Lymph node metastasis (604)			0.622
None	317	227
N1	17	11
N2	16	16
Pathological number of metastases (1047)			<0.001
1	383	288
2	101	48
3	69	39
≥4	94	25
Chemotherapy for pulmonary metastasis (1047)			<0.0001
Yes	161	151
No	486	249

Wilcoxon rank-sum (Mann–Whitney) test for continuous covariates.

Fisher's exact test for categorical covariates.

CI: confidence interval.

Propensity score allocation

The propensity score was calculated using preoperative variables. In this model, the P-value of the Hosmer–Lemeshow of goodness of fit was 0.1579, and the c-statistic was 0.7149. The optimal balancing properties between the groups were satisfied (Fig. 3).

Figure 3:

Propensity score distribution (A) and standardized % bias across covariates (B). Balance was satisfied in the model.

Open in new tabDownload slide

Analyses with propensity score adjustment

Kaplan–Meier survival estimation after adjustment showed better survival in the thoracoscopy group than in the open group (Fig. 4) (stratified log-rank test: P = 0.0353).

Figure 4:

Kaplan–Meier survival estimation after adjustment showed better survival in the thoracoscopy group than in the open group (stratified log-rank test: P = 0.0353).

Open in new tabDownload slide

Using the multivariate Cox proportional hazard model with stepwise backward elimination method with a probability level of 0.15, the most powerful predictive model for overall survival was found to be that which combined thoracoscopy (HR: 0.468, 95% CI: 0.262–0.838, P = 0.011) and anatomical resection (HR: 1.49, 95% CI: 1.134–1.953, P = 0.004). The proportional hazards assumption was validated by Schoenfeld residuals (global test: P = 0.9538) and a −ln(−ln(survival)) plot (Fig. 5). A one-sided Cox proportional hazard test detected a statistically significant 50% increase in the survival probability for the subjects in the case group (alpha = 0.05, power = 0.70), compared with those in the control group (58 events among 182 subjects).

Figure 5:

The proportional hazards assumption was validated using Schoenfeld residuals (A) (global test: P = 0.9538) and −ln(−ln(survival)) plot (B).

Open in new tabDownload slide

Before adjusting for the propensity score, the delta_num was significantly greater in the open group than in the thoracoscopy group (thoracoscopy: 0.06, open: 0.33, P = 0.001); however, after adjustment, there was no difference in the delta_num (thoracoscopy: 0.04, open: 0.19, P = 0.114) (Table 3).

DISCUSSION

Pulmonary metastasectomy of colorectal cancer has been accepted as an effective therapeutic strategy for resectable tumours, and several factors have been identified as prognostic indicators (e.g. tumour number, tumour size, preoperative carcinoembryonic antigen level, lymph node involvement and completeness of resection) [10, 12]. The main issues of the surgical approach for pulmonary metastasectomy have been the precision of tumour identification and subsequent survival outcomes [1–3, 5, 6]. Although there has been considerable doubt over the benefits of thoracoscopic lung surgery for pulmonary metastasectomy, the prognostic impact of this surgical approach has not yet been examined. In this study, thoracoscopic metastasectomy revealed an equivalent power of tumour identification to open surgery, and the survival outcome was better than that of open surgery.

Finger palpation is widely accepted as an indispensable tool for the detection of pulmonary nodules [7]. Thoracoscopic surgery that may preclude finger palpation has been regarded as an improper method of manipulating pulmonary metastasis when there may be multiple metastases, even if they appear as a solitary lesion using preoperative computed tomography [1–3, 13]. In spite of this background, the use of thoracoscopic surgery over open surgery has been increasing in clinical practice [4].

Repeated surgery for pulmonary metastases is commonly done because metachronous emergence of metastases is highly probable [14]. The following hypothesis might give a possible explanation of the equivalent power of open and thoracoscopic surgery. Truly curable patients with true synchronous or metachronous oligometastases might be rescued using repeated surgery, even if the metastases were missed in the initial operation, regardless of surgical approach. Patients with an initial solitary metastasis or oligometastases might be burdened with radiologically silent and non-palpable numerous metastases; in this condition, surgery may be ineffective, irrespective of the approach. The number of metastases is a strong prognostic indicator [10], and the larger the number of metastases, the lower the effectiveness of surgery, even if complete macroscopic resection can be achieved. This may be one reason why the surgical approach showed less influence on outcomes.

Because the preoperative radiological number of metastases was entered into the propensity score model, the pathological number of metastases that should be correlated to preoperative radiological number might be adjusted between the two groups at the time of post-adjustment analysis. This might be why the number of metastases, a strong prognostic indicator, was not an effective prognostic indicator after adjustment.

The increased ratio of thoracoscopic surgery over open surgery has reflected recent developments in computed tomography and pharmacotherapy [15, 16]. Additionally, radiologically depicted targets can be resected even though they are very small [17, 18]. A weak aspect of the thoracoscopic approach may be compensated by surrounding technologies [9].

The difference between the outcomes of primary lung cancer surgery after thoracoscopy and open thoracotomy remains controversial. Previous reports on the possible oncological benefits of thoracoscopic surgery over open surgery primarily came from retrospective studies [19–21]; however, recent propensity score analyses have reported that thoracoscopic surgery and open surgery showed equivalent survival outcomes [21–25]. In this study of lung metastasectomy, the thoracoscopic group demonstrated better outcomes than the open group in terms of survival, but this might have been caused by incomplete balancing between the two groups because of missing data in the database.

There were several limitations to this study. Because of the retrospective setting of this study’s database, opportunity for performing the thoracoscopic approach might be neglected (i.e. identification of small and deep nodules) [17, 18]. This multi-institutional study lacked uniform surgical indications and uniform selection criteria for the surgical approach. Moreover, surgical indications and selection criteria of the surgical approach in each institution might evolve over time. This database lacked information regarding tumour location (peripheral or central), method of tumour identification, type of lymphadenectomy (sampling or systematic), and the distance of the surgical margin. Moreover, the database is not complete and has missing data that may influence the analysis (Table 2). Information about chemotherapy for primary colorectal cancer that might influence survival was missing in approximately half of the cohort. Information about the presence or absence of lymph node metastasis was available in only 58% of the patients. The presence of lymph node metastasis might influence the type of approach (preference of open approach over thoracoscopic approach) and thus might influence survival differences between open and thoracoscopy approaches.

Treasure et al. [26] highlighted several controversial issues about the efficacy of lung metastasectomy. Similar to past reports based on retrospective analysis for very highly selected patients, the patients in this study were assigned to lung resection based on resectability in individual cases, on the institutional or each surgeon’s preference, and possibly on patient ‘survivability’, a concept introduced by Treasure et al. [26]. Although the purpose of this study was not to test the efficacy of lung metastasectomy, the early horizontal component of survival curves in Figs 1 and 2 might indicate inherent immortal time bias in our cohort [27].

CONCLUSION

Thoracoscopic metastasectomy revealed better overall survival than the open approach in this analysis. The thoracoscopic approach may be an acceptable option for the resection of pulmonary metastases in terms of tumour identification and survival outcome in this era, as shown by a recent systematic review [9].

Conflict of interest: none declared.

REFERENCES

Author notes