This study has shown that CT derived muscle mass at T4 is comparable to those measures obtained from the L3 level in patients with rectal cancer. Body composition measurements at T12 revealed a larger distribution of differences and a proportional bias was noted (Fig. 4, Table 3).
In method comparison studies, the appropriate analysis should aim to uncover systematic differences. There are two potential sources of systematic disagreement between methods of measurement: fixed and proportional bias. Fixed bias means that one method gives values that are higher (or lower) than those from the other by a constant amount; proportional bias means that one method gives values that are higher (or lower) than those from the other by an amount that is proportional to the level of the measured variable [13,14,15,16].
Previous studies on T4 and T12 CSA [5, 7, 20] have reported findings contradictory to our own results. High values of Pearson’s correlation coefficients (r) had been used as metrics to assess agreements with L3 measurements in other publications [21, 22]. However, it has been well documented in the statistical literature that Pearson’s correlation coefficient merely indicates the scatter of values around the line of best fit, (Fig. 3) regardless of whether the slope of that line differs from unity (proportional bias) or whether its intercept differs from zero (fixed bias) [13,14,15,16]. It does no more than indicate the strength of the linear association between the x and y variables in the examined population. The information provided by r is, therefore, of no value in detecting systematic biases between methods. Ordinary least product (OLP) regression and or Bland–Altman plot [21, 22] are the appropriate analysis to use in these situations.
Utilising CT derived muscle mass at the L3 is currently the gold standard for body composition analysis, which serves a marker for total body skeletal muscle quantity [9, 10, 23, 24]. Although artificial intelligence (AI) derived 3D total body composition measurements have been developed and have been described within the literature, there are very few studies using validated patient data, and most clinical data linked research is generally only based on single level semiautomated assessments . There is also growing evidence that other levels including T4 and T12 are equally representative of patient body composition, as shown in conditions such as lung cancers [5, 25], and patients undergoing interventional cardiothoracic and vascular surgery [19, 26].
In our study, we found that both total muscle cross sectional area (CSA) and skeletal muscle index (SMI) were significantly higher in males than in females; however female patients had significantly higher BMI compared to our male patients. We found that muscle measurement at all vertebral levels was relatively easy to complete by manual segmentation. When interobserver agreement was compared, the much larger variability seen for L3-T12 CSA compared to L3-T4 CSA implied that T4 CSA measurements had higher test–retest reliability. When we looked at time and consistency to complete body composition analysis between T4, T12 and L3 levels, we found that overall, the T12 vertebra reading required slightly less time to annotate compared to L3. However, the readings at T4 vertebral levels had a more consistent segmentation result between graders.
We also performed a comparison of the degree of muscle quantity between our thoracic and lumbar levels within our patient cohort. We found that the cross-sectional muscle areas were greatest at the vertebral T4 level, followed by L3 then finally T12. Comparing methods of measurement utilising percentile ranked differences and ordinary least product regression, we found a stronger agreement between both CSA and SMI measured at L3 with that at T4 as compared with T12. However, despite these agreements, we noticed that when trying to annotate the muscle groups within the T4 level, some of the patients’ skeletal muscles were “cut off” from the CT scan thereby affecting our ability to complete the total calculation of body composition in those images.
With regards to clinical outcomes, our study did identify evidence of sarcopenia within our cancer cohort; however, we did not find an association between sarcopenia and post-operative complications, recurrence rates or hospital LOS in patients undergoing curative resection. We did however identify that sarcopenia was related to a reduced survival.
The strength of this study is the robust analysis of measurement errors (fixed and proportional bias) as distinct from using correlation coefficients to assess agreements. We recognised that Pearson’s (r) value does not provide clinicians with any insight into systematic errors that may be inherent in the measurement obtained with a specific assessment tool. We therefore utilised ordinary least product (OLP) regression for our analysis.
There were several limitations to our work. Our study relied on retrospective data from a single-centre and only 5 scans at each vertebral level were read to determine the interobserver reliability. Whilst our study was able to determine a positive relationship between both T4 and T12 vertebral levels with L3, half of the images (40 out of 80) acquired at T4 were noted to have “cut offs”, where the outer circumference of the muscle mass was missing; a problem also seen in other studies assessing skeletal muscle at the T4 level [5, 27]. To account for this “cut off”, the “arm” muscles were not counted in the final cross-sectional area. This therefore led to a poorer reproducibility as it was not clear as to where the boundaries of the muscles around the scapula and arm might be. Previous studies encountered a similar problem and overcame this by including only the pectorals, intercostals and muscles of the back in their CSA measurements [27, 28].
The findings of a strong association between T4 and L3 measurements suggested that the thoracic muscles, like those at the lumbar level, would be reasonably representative of total body skeletal muscle quantity. However, consideration must be given to the functions of the muscles at their respective levels. For example, at the T12 and L3 levels, muscles include the rectus abdominis, external and internal oblique and erector spinae (the core muscles). These muscles are thought to initiate full-body functional movement and are essential for stabilising the body in dynamic movements . Although some of these muscles (erector spinae) also extend upwards, the major muscles annotated at the T4 level, (pectoralis muscles, and supra and infraspinatus muscles) primarily function in a different way, through mobilisation of the arm and shoulder gridles, resulting in potentially different CSA and SMI results [5, 29].