After obtaining institutional review board approval, the electronic medical record was reviewed to identify 349 consecutive patients with advanced melanoma treated with ICI (either anti-PD-1/PD-L1 monotherapy or combination ipilimumab and nivolumab) at Vanderbilt University Medical Center from October 2009 through October 2018. Of these, 290 patients had pretreatment scans available for analysis (defined as CT or positron emission tomography-computed tomography (PET-CT) obtained within 6 months of treatment start). After exclusion of three scans for excess artifact, a total of 287 patients were included. Chart review was performed to assess immune-related adverse events related to treatment as defined by Common Terminology Criteria for Adverse Events Version 4.03. Response, progression-free survival (PFS) and overall survival (OS) were investigator determined based on review of electronic medical records. Objective response was defined by RECIST V.1.1. PFS was defined as time of treatment start to progression by RECIST V.1.1, and OS was defined as time of treatment start to death or last follow-up.

Pretreatment scans (either non-contrasted CT images from PET-CT scans or CT images) were analyzed using Slice-o-matic (Tomovision V.5.0) and ABACS (automated body composition analyzer using CT) software according to previously established methods (figure 1).11 Briefly, patient scans were viewed using AGFA IMPAX software (V.6.6.1.3525) and the L3 level was identified by study personnel. Axial, cross-sectional images at the L3 level were then uploaded to Slice-o-matic and automatically segmented into muscle tissue, subcutaneous adipose tissue (SAT) and visceral adipose tissue (VAT) using ABACS automatic segmentation software. This software identifies muscle tissue as tissue with a radiodensity between −29 and +150 Houndsfield units (HU). Given the radiodensity of organs also falls within this range, the software incorporates knowledge of L3 muscle shape to avoid erroneously labeling organs as muscle tissue.12 Once muscle tissue was identified, SAT was defined as tissue lying outside the border of the defined muscle area with a radiodensity between −190 and −30 HU and VAT was tissue lying inside the border of the defined muscle area with a radiodensity between −150 and −50 HU. This software has been previously validated through comparison with manual segmentation and was found to have excellent concordance.11

Skeletal muscle index (SMI) was used to normalize muscle area for height and was calculated as follows: (skeletal muscle area (cm²))/(height (m²)). Sarcopenia was defined according to Martin et al.13 For patients with BMI <25, sarcopenia was defined as SMI <43 cm/m² for men and <41 cm/m² for women and for BMI ≥26, sarcopenia was defined as <53 cm/m² for men and <41 cm/m² for women. Sarcopenic obese was defined as patients who met above criteria for sarcopenia with BMI ≥30. Skeletal muscle density (SMD) is a measure of muscle attenuation and was determined by Slice-o-matic software by taking the average HU of muscle present at the L3 level. SMD is known to be inversely related to myosteatosis and has also been shown to be associated with survival among patients with cancer in several studies.13–16 Low SMD was defined according to previously established cut-offs by Martin and colleagues.13 For patients with BMI <20 to 24.9, low SMD was defined as <41 HU. For patients with BMI >25, low SMD was defined as <33 HU. Of note, there are multiple definitions of SMD in the literature. We chose the Martin et al definition because it was the first pivotal study to address SMD in oncology and it is the most commonly employed in the literature.17 Total adipose tissue index (TATI) was used to normalize adipose tissue for height and was calculated as follows: (subcutaneous adipose tissue area (cm²) +visceral adipose tissue area (cm²))/(height (m²)). We used tertiles to categorize TATI because there is no clinically established threshold for TATI and we felt that tertiles would be the most appropriate way to categorize a continuous variable without an established threshold. Skeletal muscle gauge (SMG) incorporates both muscle area and muscle density and has strongly correlated with patient outcomes including toxicity and functional status among patients with breast cancer treated with chemotherapy.18–20 SMG was calculated as follows: (SMI cm²/m²)×(skeletal muscle density in HU). Patients were divided into high and low SMG groups based on previously established cut-offs by Shachar et al (SMG cut-off 1475).19 VAT index and SAT index were calculated as follows: (visceral adipose tissue area (in cm²)/(height (m²)) for VAT index and (subcutaneous adipose tissue (in cm²)/(height (in m²)) for SAT index.

Continuous variables were compared between groups using t-test or non-parametric Mann-Whitney U test. Categorical variables were compared using χ² test. PFS and OS were assessed using the Kaplan-Meier method and compared between groups using the log-rank test. Multivariable Cox regression models were developed to estimate patient survival in association with TATI and SMG, as well as their interaction term, adjusted for covariates including age, sex, stage, and prior therapy. Regarding the SMG:TATI interaction term, we thought that patients with high skeletal muscle and high adipose tissue would behave differently than those with low skeletal muscle and high adipose tissue and thus we included an interaction term to allow the estimated effect of SMG to differ at different TATI levels. The interaction term was significant for response (p_interaction=0.0499) but not statistically significant for PFS and OS (p_interaction >0.1). Given the study sample size as well as the number of OS/PFS events, we were able to afford to include the interaction term along with other specified covariates in the multivariable analyses without causing overfitting concerns. Of note, the effect of SMG on PFS appeared to be lower for patients in the highest TATI tertile compared with patients in the lowest tertile, independent of age and gender (adjusted HR=0.55, p=0.08). Age appeared to have a non-linear relationship with survival outcomes and therefore was fitted using restricted cubic splines with three knots. A sex-stratified analysis was performed to assess for body composition differences that are sex-specific. All statistical analyses were performed using R V.3.6.0.

There were 287 patients with metastatic melanoma included (see table 1 and online supplementary table 1). Median follow-up time was 519 days. A total of 136 patients were alive at last follow-up. The median age was 63 and most (64.1%) patients were male. Most patients had stage IV M1c or M1d disease (51.9%) and had received prior treatment (53.3%). Pembrolizumab was the most common ICI (64.8%) followed by combination ipilimumab and nivolumab (21.6%). The median BMI was 28.9, slightly higher than the national average (26.6 for men and 26.5 for women).22 Over half (53.7%) of patients were sarcopenic at baseline.

We did not find any statistically significant associations between BMI and response, toxicities (any grade), PFS, or OS in univariable or multivariable analyses (table 2; online supplementary table 2). Next, we considered the possibility that sex differences were masking associations between BMI and outcomes by performing sex-stratified analysis (table 2). We found that this was not the case given that even in the sex-stratified analysis, we did not find any statistically significant associations between BMI and response, toxicities, PFS, or OS.

Next, we sought to determine whether different classes of obesity were associated with any of the above outcomes (online supplementary table 3 and online supplementary figure 1). Although there were no statistically significant differences in response, toxicity, or PFS when examining different classes of obesity, there was a statistically significant difference in overall survival among patients with class III obesity compared with patients with class I obesity (HR 2.4, 95% CI 1.1 to 4.9, p value = 0.03).

Lastly, to determine if associations differed between the monotherapy and combination therapy cohorts, we fitted the same multivariable Cox regression models for response, PFS and OS for the combination therapy cohort and the monotherapy cohort separately (online supplementary table 4). There were no major differences between the individual cohorts and the pooled analysis. Again, there were no statistically significant associations between BMI and Response, PFS or OS.

Given BMI is often used as a surrogate for body composition, we next examined the correlation between BMI and various body composition measures. BMI was most strongly correlated with TATI (correlation coefficient 0.88; online supplementary table 5). The weakest correlation was between BMI and SMG (correlation coefficient 0.15).

Given the strong correlation of TATI and BMI, we first assessed whether TATI correlated with clinical outcomes. In the univariable analyses (table 3 and online supplementary table 2), we did not find any statistically significant associations between TATI (assessed by tertiles) and response, toxicities, PFS or OS. Patients with sarcopenic obesity had inferior PFS (HR 1.47, 95% CI 1.02 to 2.12, p=0.037), although there were no differences in response or OS.

In multivariable analyses (table 4) adjusted for age, sex, SMG:TATI interaction, stage and prior therapy, there were no differences in response, PFS or OS among patients with high SMG versus low SMG. Although there was no difference in response among different TATI tertiles when examining the total cohort, when stratified by sex, females in the highest TATI tertile were significantly less likely to respond to ICI compared with females in the lowest TATI tertile (OR 0.18, 95% CI 0.04 to 0.76, p=0.02). This difference was not seen among males. Additionally, multivariable analyses showed that patients in the highest TATI tertile were more likely to experience progression (HR PFS 1.71, 95% CI 1.01 to 2.87, p=0.04) than those in the lowest tertile. In the sex-stratified analysis, this association was even stronger among women (HR PFS 2.06, 95% CI 1.06 to 3.98, p=0.032) but did not persist among men (HR PFS 1.40, 95% CI 0.59 to 3.31, p=0.45). There was, however, no statistically significant difference in OS between patients with high and low TATI in the total cohort or when stratified by sex.

We next sought to determine if associations differed between the monotherapy and the combination therapy cohorts. We fitted the same multivariable Cox regression models for response, PFS and OS for the combination therapy cohort and the monotherapy cohort separately (online supplementary table 6). The associations in the monotherapy cohort were consistent with the results of the pooled analysis. There were no statistically significant associations in the combination therapy cohort which may be due to limited power as a result of the small number of patients in this group (n=62).

To further investigate the relationship between SMG and TATI, we assessed different combinations of SMG and TATI. Although within the full group, outcomes were not statistically different. when comparing cohorts with the poorest outcomes (low SMG:high TATI) to those with the best outcomes (high SMG:mid TATI), there was a significant difference in both PFS and OS with patients in the low SMG:high TATI group having significantly worse outcomes (p=0.02 and 0.02 respectively, figure 2).

Given the inverse relationship between TATI and response, we were interested in whether one adipose tissue compartment was primarily driving this association. Thus, we assessed SAT and VAT in relation to response, PFS and OS (online supplementary table 7). We found that there were no statistically significant differences among patients with low versus high SAT or low versus high VAT with regard to these outcomes.