We analyzed data from the Spanish Society of Thoracic Surgery GEVATS database, which recruited patients undergoing anatomic pulmonary resections between December 20, 2016 and March 20, 2018 who were followed up until July 31, 2022.10 This database prospectively included patients from 33 thoracic surgery units. The study was approved by the ethics committees of all participating centers. In this analysis, we included the 30 centers in which oncological follow-up for survival had been completed.

Of the 3533 patients included in the registry, patients with lung cancer (3085) were selected, and of these, individuals with a follow-up at least 90 days post-procedure (2726) were included in the study cohort. All left trisegmentectomies (anatomic resection S1–S3) and all left upper lobectomies in this group were selected, resulting in a total of 623 patients, comprising 540 lobectomies and 83 trisegmentectomies.

The following variables were compared for the 2 types of resection: age, smoking habit, tumor size, histological type, radiological density of the tumor, surgical access, forced expiratory volume in the first second (FEV1), diffusing capacity of the lungs for carbon monoxide (DLCO), hypertension, chronic heart failure, ischemic heart disease, arrhythmia, stroke, peripheral vascular disease, diabetes, cN status by positron emission tomography/computed tomography (PET/CT), pTNM, and postoperative N. Continuous variables were compared using a t-test, and categorical variables were compared using Fisher's exact test. A difference in any variable between both groups was considered significant if the p value was less than 0.05 (Fig. 1A).

Fig. 1.

Flowchart of the main methodology used. (A) Represents the statistical analysis performed comparing lobectomies and trisegmentectomies. This statistical analysis consisted of a t-test for continuous variables and Fisher's test for categorical variables. Statistical significance was set at a p value<0.05, obtaining a list of “variables that differed between surgeries”. (B) Represents the 3 types of survival analyses performed: B1: overall survival in which the event was death; B2: cancer-specific survival, in which the event was cancer-specific mortality; and B3: disease-free survival. A Cox regression was performed for the 3 analyses, and a variable was considered to have an effect if the p value of the Cox regression likelihood ratio test was <0.05. For the 3 analyses, propensity score matching was performed in which the “variables that differed between surgeries” plus the variables related to risk were considered. (C) Once the groups were obtained after propensity score matching, the effect of the type of surgery on the risk of overall survival, cancer-specific survival and disease-free survival was analyzed. Finally, for B3, the type of recurrence was analyzed using Fisher's test.

(0.58MB).

A survival analysis of each variable was performed using the Cox regression model. Specifically, 3 separate survival analyses were performed: (I) overall survival, in which the event being studied was all-cause mortality; (II) lung cancer-specific survival, in which the event being studied was cancer-specific mortality; and (III) disease-free survival, in which the event being studied was recurrence. A variable was considered to affect each of the analyses if the likelihood ratio test of the model had a p value less than 0.05 (Fig. 1B).

Variables that were significant in the statistical analysis or survival analysis were used to perform propensity score matching using the 2:1 nearest neighbor technique (Matchit library). Once comparable groups had been created, group homogeneity was verified using the t-test (continuous variables) and the Fisher's exact test (discrete variables) for the aforementioned variables (Fig. 1B).

Once the comparable groups had been created, overall survival, lung cancer-specific survival and disease-free survival were estimated using Kaplan–Meier curves, comparing both groups with the log-rank test. A Cox regression was also performed to compare the risk between both surgical approaches in all patients, using inverse probability of treatment weighting (IPTW) methodology.11 This method uses all patients, conferring to each one a weight according to the value obtained in the propensity score. These weights were calculated for each patient as 1/propensity score for patients with trisegmentectomy and 1/(1−propensity score) for the remaining patients, where propensity score is the score obtained.

Finally, recurrence and the recurrence pattern were compared using Fisher's test.

The methodology described above and in Fig. 1 was applied to a subgroup of patients with tumor size between 20 and 30mm.

The statistical analysis was performed using R software and the Matchit library was used to calculate the propensity score.

In patients with a diagnosis of lung cancer and survival of more than 90 days, 540 left upper lobectomies and 83 trisegmentectomies were retrieved. Statistical analysis was performed to compare the variables of interest between the lobectomy and trisegmentectomy subgroups. The results of this analysis are summarized in Table 1. The variables associated with significant differences (p value<0.05) were: tumor size, tumor density, surgical access, cN, postoperative N and pTNM. Among these variables, tumor size is notable, with an average of 30mm in lobectomies and 18.7mm in trisegmentectomies. The differences in pTNM (p value=5.32e−08) between the 2 types of surgeries are also worth pointing out.

Table 2 shows the results for the 3 survival analyses (overall, cancer-specific, and disease-free survival). This table shows the p value of the Cox regression likelihood ratio test. See Supplementary Table 1 for the p value corresponding to each Cox regression coefficient.

In the overall survival analysis, the following variables were significant (p value<0.05): tumor size, histological subtype, FEV1, DLCO, stroke, diabetes, cN, postoperative N, and pTNM.

In the analysis of cancer-specific survival, the following variables were significant: tumor size, histological subtype, surgical access, FEV1, DLCO, diabetes, cN, postoperative N and pTNM.

In the disease-free survival analysis, the following variables were significant: tumor size, histological subtype, surgical access, FEV1, DLCO, arrhythmia, diabetes, cN, postoperative N, and pTNM. In this analysis, 26 patients were excluded due to a lack of information on recurrence, leaving a total of 597 patients, with 517 lobectomies and 80 trisegmentectomies.

Propensity score matching was performed with the variables that were significant in these analyses (Fig. 1B). In total, 3 types of matching were conducted, 1 for each type of analysis. To be able to use all variables, patients with incomplete information were excluded. In this way, a group of 195 patients with 130 left upper lobectomies and 65 trisegmentectomies was obtained for the first 2 analyses and 1 group of 189 patients was obtained for the third analysis, consisting of 126 lobectomies and 63 trisegmentectomies. Table 3 shows the variables used for each of the 3 analyses, the number of patients with complete and incomplete information, and finally the number of patients in each group after the propensity score procedure.

No differences were found between the 2 surgical groups for the 3 analyses in any of the technical aspects (fissureless lobectomy or the need for conversion to thoracotomy), or in postoperative complications or hospital stay. Table 4 provides information on complications after propensity scoring for overall survival. Supplementary Tables 2 and 3 show these results for the other 2 analyses.

A Cox regression was performed to determine if trisegmentectomy increases the risk of survival, cancer-specific survival, and disease-free survival compared to lobectomy. This Cox regression was conducted with comparable groups obtained from propensity score matching for the 3 analyses. The Kaplan–Meier curve was also applied for the 3 analyses and compared with the log-rank test (Fig. 2). Supplementary Tables 5–7 show the 95% confidence interval for the risk in each case.

Fig. 2.

Kaplan–Meier curves after propensity score matching for overall survival, cancer-specific survival, and disease-free survival analyses. The 95% confidence interval and the p value corresponding to the log-rank test are shown for the 3 analyses.

(0.36MB).

In the case of overall survival, the p value obtained is not significant (p=0.76) showing that the type of surgery does not affect the risk of survival. Furthermore, the Kaplan–Meier curves corresponding to the types of surgery (Fig. 2) show no significant differences (log-rank test, p=0.76).

With regard to cancer-specific survival, the results are very similar, with a p value of 0.62 for the Cox regression and a p value of 0.62 in the log-rank test comparing the 2 Kaplan–Meier curves (Fig. 2).

In the case of disease-free survival, the p value of the Cox regression is 0.15 and the p value for comparison of the 2 Kaplan–Meier curves (Fig. 2) is 0.15. These results show that the type of surgery does not affect the risk of recurrence.

Comparable groups determined by propensity score matching were used in the 3 analyses: 291 patients were excluded out of 486 cases with complete data in the first 2 analyses; and 281 were excluded out of 470 cases with complete data in the third analysis. To avoid patient loss, a Cox regression was also performed with all patients using the IPTW method. The Cox regression p value for overall survival was 0.23; for cancer-specific survival, it was 0.73; and for disease-free survival, it was 0.28. These results are consistent with previous reports suggesting that trisegmentectomy does not affect the risk of overall survival or disease-free survival compared to lobectomy.

The analysis of disease-free survival using the set of patients with complete information (597 patients, 517 lobectomies and 80 trisegmentectomies) showed a recurrence rate of 37.52% for lobectomies and 28.75% for trisegmentectomies. This difference was not significant according to Fisher's exact test (p value=0.1359).

In comparable groups (Table 5) comprising 189 patients (126 lobectomies and 63 trisegmentectomies), recurrence was 32.53% after lobectomy, compared to 20.63% after trisegmentectomy. This difference was not significant according to Fisher's exact test (p value=0.12). It should be noted that, in the lobectomy group, 14.3% of patients had distant recurrence compared to 3.2% in the trisegmentectomy group (p=0.02). If we combine these recurrences with mixed recurrence rates (locoregional+distant), a total of 19% of the patients in the lobectomy group had recurrence compared to 11% in the trisegmentectomy group (p=0.21). Table 5 shows the recurrence rates in both groups after the creation of comparable groups.

In this second analysis, patients with tumors measuring between 20mm and 30mm were studied, resulting in a total of 97 patients, of whom 82 underwent lobectomies and 15 trisegmentectomies.

The variables used for the propensity score for the 3 types of analyses are shown in Supplementary Table 4. After propensity score matching, a subset of 45 patients (30 left upper lobectomies and 15 trisegmentectomies) was obtained for the 3 analyses. For the disease-free survival analysis, 3 patients with incomplete data were excluded from the calculation of the propensity score.

For the overall survival analysis, there is no difference in risk between the 2 types of surgery (p value=0.834 for the Cox regression and p value=0.83 for the log-rank test). For the cancer-specific survival analysis, the Cox regression p value was 0.59, and also 0.59 for the log-rank test. Finally, for the disease-free survival analysis, the p value of the Cox regression was 0.646, and for the log-rank test, it was 0.65 (Fig. 3).

Fig. 3.

Kaplan–Meier curves after propensity score matching for overall survival, cancer-specific survival, and disease-free survival analyses in the subgroup of patients with tumors measuring 20–30mm. The 95% confidence interval and the p value corresponding to the log-rank test are shown for the 3 analyses.

(0.36MB).

The results obtained using the IPTW method are also consistent, with p values for the 3 analyses of 0.582, 0.743 and 0.611, respectively.

Table 6 summarizes the type of recurrence for each surgical approach. Differences are observed, but they are not statistically significant.