Approval for this retrospective data analysis of prospectively collected data was obtained from the Institutional Review Board. Patients with known or suspected clinical stage I NSCLC who underwent surgical resection at Fox Chase Cancer Center (FCCC) during two periods of time, from mid-2003 until November, 2005 and from November 2005-through 2008 were analyzed. Patients undergoing lobectomy, bilobectomy or segmentectomy were included. During the study periods, patients underwent standard preoperative staging with chest computed tomography (CT) and positron emission tomography with CT (PET/CT). At the time of the study, brain MR was performed in patients with neurologic symptoms or for patients with T2 tumors. Cervical mediastinoscopy was performed selectively for patients with abnormal lymph nodes on any imaging study, or for those with T2 or more central tumors.
All of the operations in this series were performed by a single surgeon (WJS). Standard anesthetic management with single lung ventilation including restriction of intraoperative fluids was used in all patients. Open thoracotomy (THOR) was most commonly performed through a posterolateral incision with entry into the chest through the fifth intercostal space. Resection of a small portion of the 6th rib was routine. Muscle sparing incisions were used occasionally and rib resection was not performed in those instances. VATS lobectomy (VATS) was performed using three incisions, a 5 cm or smaller access thoracotomy and two additional 1 cm incisions. Rib spreading and rib resection were not performed. Visualization was achieved using the video thoracoscope.
Lymph node dissection was routinely performed in all cases (THOR or VATS). Postoperative pain control was achieved using patient-controlled analgesia delivered through thoracic epidural catheters (PCEA) or through the use of patient-controlled intravenous narcotics (PCA). Most recently, pain control for VATS lobectomies consists of PCA narcotics supplemented with a continuous local anesthetic infusion (0.2% ropivacaine) though subpleural catheters placed intraoperatively. Patients converted from VATS to thoracotomy were included in the VATS group. All complications were graded according to the National Cancer Institute Common Terminology Criteria for Adverse Events version 3.0 [12
Patients undergoing thoracotomy (open cases) were the control population and patients undergoing VATS procedures comprised the treatment group. The baseline characteristics of the two groups are shown in table (demographics). The endpoint of the study was postoperative length of stay (LOS), duration of chest tube placement, and overall complication rates. Because this was a nonrandomized comparison of two treatments, the two groups were adjusted for characteristics that were known to influence the main study outcomes. We chose age, sex, the preoperative FEV1 percent predicted and two additional measures, the Charlson Comorbidity Index (CCI) and the American Society of Anesthesiology (ASA) score. (Table ) The CCI consists of the sum of weighted scores for 19 medical conditions that have been shown to affect mortality. The CCI has been validated in a surgical population of lung cancer patients [13
]. The ASA score is a rating scale used by anesthesiologists to estimate the overall condition of a patient in the preoperative period Table [14
Charlson Comorbidity Index (CCI)
Patient Characteristics, Unadjusted data, % or mean (SD)
American Society of Anesthesiologist (ASA) Score
In order to adjust for baseline differences between those patients who did and did not have VATS, we used propensity score based adjustment through propensity score based weighting [15
]. Similar propensity score based weighting has been used in a variety of settings to investigate treatment effects using observational data [16
]. The propensity scores, which are the probabilities of receiving VATS given potential confounders of treatment assignment, were estimated by a multiple logistic regression. The adequacy of the propensity score model was verified by examining adjusted differences in potential confounder variables between the treatment groups. The lack of significant differences in propensity-score adjusted averages of the confounder variables suggested that they could not be confounders after adjustment.
We used Fisher's exact tests and T-tests to assess unadjusted differences. For inferences concerning length of stay and time until chest tube removal, we used Cox proportional hazards regressions weighted by the inverse of the probability of receiving the treatment actually received. The weight would be the inverse of the propensity score for those in the VATS group and the inverse of one minus the propensity score for those in the thoracotomy group. The bootstrap [18
] with 2500 resamples was used to calculate the standard errors. Cox models were used for the time to event variables since some patients were lost to follow-up (for example, discharged with a chest tube) and hence had censored data. We used simple linear regressions similarly weighted by the inverse of the probability of receiving the treatment actually received for adjusted inferences concerning the number of lymph nodes removed and the number of lymph node stations sampled. We used propensity score based weighted logistic models for adjusted inferences concerning complication rate differences.