Preoperative thoracic muscle area on computed tomography predicts long-term survival following pneumonectomy for lung cancer

Abstract

Open in new tabDownload slide

OBJECTIVES

To assess the prognostic role of thoracic muscle as quantified on preoperative computed tomography (CT) for the estimation of overall survival (OS) following pneumonectomy.

METHODS

Muscle cross-sectional area (CSA) at the level of the fifth (T5) and eighth (T8) thoracic vertebra was measured on CT scans of consecutive patients with lung cancer prior to pneumonectomy. We stratified patients into high and low muscle groups using the gender-specific median of muscle CSA as separator and estimated associations of muscle CSA and OS using the Kaplan–Meier analysis. Multivariable logistic regression adjusted for body mass index, Charlson comorbidity index (includes age), forced expiratory volume in the first second as a % of predicted, sex, race, smoking status, tumour stage and prior lung cancer treatment was performed.

RESULTS

A total of 128 patients were included (61.0 ± 10.6 years of age, mean body mass index of 26.9 kg/m², 55.5% men). The T8 level showed fewer artefacts and strong correlation with the T5 level (Pearson’s rho = 0.904). T8 CSA was therefore used for subsequent analyses. Mean T8 CSA was 118.5 cm² (median 115.3 cm²) in men and 75.2 cm² (median 74.0 cm²) in women. During a median follow-up of 23.6 months (interquartile range 39.3), 65 patients (50.8%) died, of whom 41 were in the low muscle group. The Kaplan–Meier analysis showed significantly longer OS in the high muscle group (log-rank P = 0.02). Multivariable analysis showed an independent association of muscle CSA and OS (P = 0.02) with a hazard ratio of 0.80 (confidence interval 0.67–0.98) per 10-cm² increment.

CONCLUSIONS

Thoracic muscle is independently associated with long-term overall survival following pneumonectomy for lung cancer and may contribute to refined survival estimates in this population.

IRB Protocol

Protocol #2017P000650, approved 21 April 2017.

INTRODUCTION

Pneumonectomy for locally advanced lung cancer still carries significant morbidity and mortality [1]. Surgeons are called upon to consider operative and recurrence risks as well as the postoperative quality of life and long-term survival when determining the most appropriate management [2].

Pulmonary function tests remain at the centre of preoperative work-up, as outlined in guidelines of the American and European Specialty Society [3–5]. However, pulmonary function tests depend on patient cooperation and are known to vary importantly [6]. Moreover, there is ample evidence that pulmonary resection can be safely performed in some patients with very low forced expiratory volume in the first second and diffusing capacity of the lung for carbon monoxide [7, 8]. As a result, there remains a need for additional objective parameters to preoperatively assess a patient’s performance status [4, 9].

The skeletal muscle measured as cross-sectional area (CSA) on computed tomography (CT) images has been shown to predict postoperative outcomes after a number of surgical procedures, including lung resection [10–12]. The analysis of skeletal muscle on CT makes use of readily available data from imaging studies obtained for tumour staging and surgical planning. Previously, most investigations focused on the muscle at the level of the third lumbar vertebral body (L3). For instance, a recent study reported that low psoas muscle area predicts worse survival in patients undergoing pneumonectomy for non-small-cell lung cancer [13] but accounted for only some of the many known confounders [12, 14, 15]. Furthermore, the assessment of thoracic rather than abdominal muscle seems more suitable for patients with lung cancer because thoracic muscles contribute to respiration, which is an important aspect of recovery and function following lung resection [12, 16–19].

We hypothesized that patients with increased thoracic muscle would have better overall survival (OS) following pneumonectomy for lung cancer. To test this hypothesis, we investigated the relationship of the thoracic muscle with the survival of lung cancer patients who underwent pneumonectomy.

PATIENTS AND METHODS

All patients who underwent pneumonectomy for lung cancer at the Massachusetts General Hospital from 1 January 2005 to 30 June 2017 were identified in the institutional sample of the Society of Thoracic Surgeons General Thoracic Surgery Database. Study inclusion required that patients were ≥18 years of age, had a pathologically confirmed lung cancer diagnosis and underwent CT of the thorax at most 90 days prior to surgery. After meeting these inclusion criteria, patients were retained for analysis if the CT was available in the institutional database and clinical data were complete. Patients with incomplete imaging of relevant chest wall muscles or soft tissue oedema were excluded from this study if this precluded muscle analysis. We also elected to exclude patients requiring extrapleural pneumonectomy due to known greater functional impairment after resection of the diaphragm, parietal pleura or pericardium compared to other pneumonectomy procedures [15]. The reasons for exclusion are detailed in Fig. 1.

The primary outcome measure was OS, which was defined from the time of pneumonectomy to the time of last follow-up or death. A secondary outcome was the availability and recency of CT examinations of the thorax compared to CT examinations of the abdomen.

Deaths were confirmed by reviewing the hospital cancer registry, the medical record and online obituary search tools [20]. Patients without evidence of death were censored from the analysis at either the time of last contact or the date of 12 December 2017—whichever occurred first.

Availability of CT examinations of the thorax and abdomen in the institutional database within 90 days prior to pneumonectomy was assessed. The most recent preoperative CT of the thorax of each patient was loaded onto a research workstation. Thoracic muscles were analysed using OsiriX Lite software (version 7.0.2, Pixmeo SARL, Bernex, Switzerland). A primary analyst measured the CSA of skeletal muscles on an axial image at the level of the fifth (T5) and eighth (T8) thoracic vertebral bodies using semi-automated threshold segmentation (Hounsfield unit thresholds −29 and +150), as described previously [12]. The measured muscles included the thoracic erector spinae, the external and internal intercostals, the serratus anterior, the subscapularis, the infraspinatus, the teres major, the trapezius, the latissimus dorsi and the pectoralis major and minor (Fig. 2).

An intra- and inter-reader agreement was assessed 4 weeks after the primary analysis by having the primary analyst and a second analyst repeat image-level selection and segmentation independently, each for a different random sample (15%) of all cases. All analysts were blinded to clinical outcomes.

Statistical analyses were performed with STATA software (version 13.0, StataCorp, College Station, TX). A type-I error rate of 5% was used for all confidence intervals (CIs) and hypothesis tests. Availability and recency of chest and abdominal CT scans were compared using the χ² and Mann–Whitney U-tests, respectively. Correlation between muscle measurements at the T5 and T8 levels was assessed using the Pearson correlation. Patients were stratified into the high and low muscle groups according to their T8 muscle CSA. Patients with CSA below the gender-specific median were considered the low muscle group, while patients with CSA above or equal to the gender-specific median constituted the high muscle group [12, 18]. Differences in patient characteristics between these groups were assessed with 2-tailed χ² tests for categorical variables, Student’s t-tests for numerical variables with normal distribution and Mann–Whitney U-tests for numerical variables not normally distributed. An inter- and intra-analyst agreement was evaluated using intraclass correlation coefficients.

The OS of the low versus high muscle groups was explored with the Kaplan–Meier method, and the log-rank test was used to assess for overall differences in survival curves between the groups. A multivariate Cox proportional hazards regression model was used to estimate whether muscle CSA (expressed per 10 cm²) as a continuous explanatory variable was independently associated with OS. Based on the literature [14, 15], this model was adjusted for the following a priori selected covariables: body mass index (BMI), which includes patient height; Charlson comorbidity index (CCI) score, which accounts for patient age; preoperative forced expiratory volume in the first second as a % of predicted, sex, race, smoking status; pathological tumour stage group; and prior treatment for lung cancer. CCI is a validated measurement of comorbidity severity in patients undergoing surgical care for lung cancer [21, 22]. CCI scores were determined retrospectively, and patients were considered to have a comorbid condition if it was noted in the medical record. The tumour stage was based on an assessment of the pneumonectomy specimen according to the 7th edition of the tumor, node, and metastasis (TNM) classification for lung cancer and reattributed if it was documented using a prior or subsequent edition of the TNM classification [13, 23]. The category ‘occult cancer’ was assigned if tumour stage could not be determined on pathological examination [23]. Prior therapy was defined as prior lung resection for lung cancer or preoperative chemoradiotherapy (CRT), both of which are associated with an increased risk of adverse postoperative events [15, 24].

The Partners Human Research Committee approved this study (protocol #2017P000650) and waived the need for informed consent. This study was conducted in accordance with the Health Insurance Portability and Accountability Act. No outside or industry funding was provided.

RESULTS

Of the 179 eligible patients who underwent pneumonectomy for lung cancer, 128 (73.1%) met the inclusion criteria. All included patients underwent standard thoracotomy for pneumonectomy. CT scans of the thorax were significantly more often available (n = 135, 75.4%) compared to scans of the abdomen (n = 93, 52.0%; P < 0.001). Furthermore, CT scans of the thorax were obtained more recently before the date of surgery [median 28 days, interquartile range (IQR) 33] compared to CT examinations of the abdomen (median 36 days, IQR 24; P = 0.01).

Muscle measurement was more frequently possible at level T8 compared to level T5 due to incomplete visualization. Muscles of all but 3 patients were completely imaged at level T8 whereas the field of view at level T5 was too small in 12 patients. As muscle measurements at these levels showed excellent correlation (Pearson’s r = 0.905), CSA measurements at T8 were used for subsequent analyses. By definition, the low muscle group and the high muscle group each comprised 64 patients, with genders distributed nearly equally between the 2 groups. Table 1 details the patient characteristics, which is stratified by the low and high muscle groups. Characteristics of the 51 excluded patients were not significantly different from the final study cohort.

Patients did not receive specific pre- or postoperative exercise therapy. Ten patients (7.8%) were discharged to rehabilitation or extended care facilities, while 114 patients (89.1%) were discharged home. Three patients (2.3%) died in the hospital, and no discharge details were available for 1 patient.

Among the 54 patients who had undergone prior therapy, 33 patients had received neoadjuvant CRT. Muscle measurements were performed on CT examinations obtained between the last week of CRT and surgery, at a median of 28 days prior to surgery (IQR 33). Among the 25 patients with either stage 1 disease or occult cancer, 17 had undergone neoadjuvant CRT with either a complete (pTx, n = 10) or near-complete (pT1-2, n = 7) response. Patients with stage 1 disease or occult cancer were nearly evenly distributed between the low and high muscle groups (Table 1), and their survival time was not significantly different than that of patients with stage II–IV disease (log-rank P = 0.57). Four patients underwent completion pneumonectomy for stage 1 disease, and 4 patients underwent pneumonectomy for central tumours preventing parenchyma-sparing resection.

The mean T8 muscle CSA was 118.5 ± 22.9 cm² in men and 75.3 ± 11.4 cm² in women. Values of gender-specific CSA showed unimodal and symmetric distribution with median values of 115.3 cm² for men and 74.0 cm² for women (Fig. 3). Excellent intra- and inter-class agreement was achieved with intraclass correlation coefficients of 0.993 (CI 0.983–0.997) and 0.991 (CI 0.978–0.997), respectively.

Overall, 65 patients (50.8%) died during the study period, of whom 41 were in the low muscle group. Two patients died within 30 days after surgery, and 6 patients died within 90 days after surgery. These early deaths were distributed equally between the low and high muscle groups and were included in the OS analysis. The median follow-up time was 23.6 (IQR 39.3) months for all patients and 25.2 (IQR 55.7) months for surviving patients. Univariable analysis considering deaths from any cause at any time during the study period showed that patients in the high muscle group had better survival compared with patients in the low muscle group (log-rank P = 0.02) (Fig. 4).

After multivariable adjustment (Table 2), lower T8 muscle CSA as the continuous explanatory variable was found to be significantly associated with worse survival [hazard ratio (HR) 0.80, CI 0.67–0.97, P = 0.02]. In addition, occult cancer (HR 4.50, CI 1.16–17.47, P = 0.03) and stage IIIA disease (HR 3.22, CI 1.05–9.81, P = 0.04) were significant predictors of survival.

DISCUSSION

The thoracic skeletal muscle area on routine preoperative chest CT is independently associated with long-term OS in patients undergoing pneumonectomy for lung cancer. This finding holds true in a model accounting for key variables known to be associated with outcomes following resection for lung cancer, including race, height (as part of BMI) and age (as part of the CCI) [13–15, 21, 22, 24]. We also found that CT scans of the thorax were significantly more often available and more recently obtained in this study population compared to CT scans of the abdomen, as postulated previously [16–19].

Extensive literature exists regarding the role of abdominal muscles for risk prediction based on a lumbar muscle assessment at the L3 level [25]. This study adds to the findings of a recent report linking psoas muscle area to survival following pneumonectomy [13] because our model accounts for all major confounders known to affect outcomes following lung resection. Furthermore, this study focuses on muscles that contribute to respiratory function and are included on routine preoperative chest CT examinations.

This study is part of an emerging set of literature outlining the role of the thoracic muscle for risk stratification of patients with lung cancer, including the association of the thoracic muscle with postoperative complications following lobectomy [12] and OS, regardless of the treatment modality [18]. In the absence of reference values defining the range of normal for the thoracic muscle [17, 19], a gender-specific median was used to separate patients into the low and high muscle groups [12, 18]. Efforts to define the normal thoracic muscle in men and women of different ages are ongoing [16].

Our findings suggest that the quantitative assessment of the thoracic muscle on CT may aid risk assessment in patients with lung cancer eligible for pneumonectomy. While we do not measure sarcopenia as previously defined by others [25], CT-derived muscle measurements likely reflect multiple factors that influence postoperative outcomes. Muscles could serve as a surrogate prognosticator and improve the assessment of patients’ overall physical fitness [4] while overcoming drawbacks of other risk assessment tools. For example, BMI is a commonly recorded known risk factor but fails to discriminate between fat mass and lean body mass [25, 26]. Also, tests such as stair-climbing or the 6-min walk test require additional patient effort and cooperation [27, 28] while muscle quantification relies on already available objective data, given that CT scans of the thorax are commonly obtained in patients with lung cancer for staging and surgical planning. Muscle quantification on CT does not require patient cooperation and requires <10 min per case. In addition, fully automated segmentation algorithms may be ready for use at the clinical workstation before long [29]. This in turn will facilitate the collection of data on a large scale and the establishment of reference values.

We do not propose that thoracic muscle measurements replace established risk assessment tools. Our preliminary findings suggest, however, that they may contribute to refined survival estimates and personalized follow-up strategies, given their independent association with long-term OS. Early identification of patients with low muscle at risk for adverse outcomes may improve the allocation of resources and services to mitigate risk and improve patient care during follow-up. Moreover, if inferior survival is anticipated, parenchymal sparing resections or even non-operative approaches may be considered. If pneumonectomy is required to achieve the desired oncological outcome, muscle quantification could be used to identify patients who would benefit from pulmonary exercise interventions, nutritional optimization and close follow-up [30].

Limitations

Several limitations of our study warrant consideration. First, our cohort is from a single academic tertiary care centre in a patient sample with limited racial diversity. Thus, larger studies with data from other centres are required to assess the generalizability of our findings and to further define the role of the thoracic muscle in preoperative risk prediction. Also, our sample may have been too small to detect associations between the thoracic muscle and early mortality, given only a very small number of events. What is more, the retrospective design resulted in a 27% exclusion rate due to unavailable imaging or clinical data elements. Furthermore, the retrospective design did not allow us to determine whether the muscle is associated with disease-specific survival, which remains to be evaluated in future prospective studies. Importantly, only patients deemed fit enough for resection were included in this analysis, and the thoracic muscle in this group is unlikely to be representative of patients assigned to other treatment modalities. Finally, muscle is just one body composition metric that can be derived from CT examinations. Future studies could explore the role of fat and bone density in this context, as well as identify extremes of the thoracic muscle that preclude meaningful survival after pneumonectomy. Prospective data collection would also allow the inclusion of important factors such as nutritional parameters and exercise testing as well as the diffusion capacity for carbon dioxide.

CONCLUSION

In conclusion, thoracic muscle on routine preoperative chest CT scans is independently associated with long-term OS following pneumonectomy for lung cancer. Thoracic muscle quantification may contribute to refined survival estimates and personalized follow-up strategies for pneumonectomy patients.

Conflict of interest: none declared.

REFERENCES

Author notes