Radiologically Determined Sarcopenia Predicts Morbidity and Mortality Following Abdominal Surgery: A Systematic Review and Meta-Analysis

The Meta-analysis of Observational Studies in Epidemiology (MOOSE) and Preferred Reporting Items for Systematic Review and Meta-analyses (PRISMA) were consulted throughout this review. A qualified medical librarian conducted the literature search. The following databases were searched for relevant studies: CENTRAL (via Cochrane Library September 2015), EMBASE (via OVID 1974 to September 2015) and MEDLINE (via PubMed 1946 to September 2015). The search strategy used text words and relevant indexing to capture the concept of radiologically assessed sarcopenia used to predict post-operative complications in abdominal surgery. The full search strategy can be viewed in the supplementary material. The following trial registers were searched: ClinicalTrials.gov (www.​clinicaltrials.​gov September 2015), WHO International Clinical Trials Registry Platform (www.​who.​int/​ictrp September 2015) and UK Clinical Research Network Study Portfolio (http://​public.​ukcrn.​org.​uk/​search September/2015). A total of 6526 records were retrieved after removal of duplicate manuscripts. Searches did not exclude studies based on publication status or language. Reference lists of key articles and the grey literature were hand-searched.

Inclusion criteria were established prior to the literature search. Studies reporting the prevalence of sarcopenia and outcomes in adult patients (>18 years) following abdominal surgery were sought. At least one of the following outcomes was required: post-operative mortality (30 days following surgery), post-operative complications, Clavien–Dindo complications, critical care dependency, length of stay, disease-free survival (recurrence of the primary tumour or metastases in cancer patients), overall survival and graft loss (transplantation).

Abdominal surgery was defined as surgery involving the abdominal cavity, including patients undergoing gastrointestinal, hepatobiliary, pancreatic, endocrine, urological, gynaecological and transplantation surgery for both elective and emergency indications. Assessment of lean muscle mass was limited to studies reporting radiological assessment methods, including computed tomography, magnetic resonance and dual-energy X-ray absorptiometry.

Patients undergoing abdominal interventions other than surgery, including percutaneous radiological procedures were excluded. Studies reporting lean muscle mass as a continuous measure or failing to define a sarcopenic population were also excluded. Subcutaneous surgery not breaching the peritoneum, including abdominoplasty, and patients undergoing oesophagectomy via a thoracic approach were also excluded.

Following removal of duplicates, two investigators screened abstracts independently, and those meeting the inclusion criteria were selected for full-text review.

Data were extracted independently by two investigators and discrepancies resolved following further review of the full article. Extracted data included age, sex distribution, ethnic characteristics, malignant status, site of primary pathology, grade and stage of tumour, exposure to chemotherapy or radiotherapy, imaging modality and image analysis technique, sex-specific muscle measures, body mass index (BMI), length of stay, complications (any complication and Clavien–Dindo grades), mortality, disease-free survival and overall survival. Authors were contacted in order to obtain raw data where summary data or odds or hazard ratios were reported, or if any further data clarification was required.

Investigators independently reviewed each full-text article, assessing quality using the Newcastle–Ottawa assessment scale for each of the outcome measures.

The Cochrane collaboration risk of bias assessment tool (chapter 8.5a) was used with additional domains relevant to this review. Domains included image capture, training of assessor, inter-observer reliability, selection bias, allocation concealment, blinding of assessors, incomplete outcome data and selective reporting. Each domain was allocated either a low, unclear or high-risk score and summary data presented using the traffic light system. Funnel plots for each meta-analysis were visually inspected and interpreted in the context of the individual comparisons (Supplementary Fig. 1).

Heterogeneity was estimated using Cochran’s Q statistic, and the percentage of variation in meta-analysed outcomes that could be attributed to sources other than sampling error (I ²) also was calculated. An I ² > 50% was considered to represent a chance of substantial heterogeneity, and >75% considerable heterogeneity.

Freeman–Tukey arcsine transformation was applied for analyses where abstracted proportions had values of zero or one [17]. Heterogeneity amongst study estimates was quantified using the I ² and associated test for heterogeneity. Where significant heterogeneity (>75%) was apparent, the DerSimonian and Laird random effects [18] method was used to pool estimates, with inverse-variance weights. Otherwise, the Mantel–Haenszel fixed effect (FE) method was applied [19].

A total of 8272 records were identified from database searching, of which 24 were included in the review [8‐10, 20‐38]. Nine studies involved patients undergoing hepatobiliary surgery, 4 pancreatic surgery, 4 colorectal surgery, 3 urological surgery, 2 oesophago-gastric surgery and 2 transplant surgery (liver). Five authors were contacted via email on at least two separate occasions to request clarification or to provide further data, three authors responded with raw data which was included in the meta-analysis. The PRISMA flow diagram can be seen in Fig. 1, which includes the reasons for removal of studies.

All of the studies were cohort studies, and quality was high (median 8, range 5–9, see Table 1). A number of parameters in the risk of bias assessment were not reported and therefore scored as unclear. Where bias was assessed, it was generally determined to be low risk, with selection bias being the most frequently reported high-risk domain (Table 2). Following inspection of funnel plots for all analyses, asymmetry was apparent for total complications. An absence of studies in the bottom left side of the plot suggests the possibility of reporting bias, where small studies demonstrating no risk reduction in non-sarcopenic patients are not published. This asymmetry may also be explained by true heterogeneity as those studies towards the right-hand side of the plots represent hepatobiliary cohorts where the greatest relative risk increase was detected.

All of the included studies used computed tomography to quantify muscle mass, with no alternative quantification methods used in studies excluded following full-text review. In addition, all included studies used the third lumbar vertebra as the landmark for muscle measurement. Alternative landmarks used in other studies include L4 [39], the umbilicus [40] and iliac crests [41]. A majority of studies used image quantification software to calculate surface area, manually drawing around the border of the muscle, or alternatively measuring the antero-posterior and transverse diameter [10]. Quantification by either a trained assessor or radiologist was reported in eight studies [10, 23, 33, 34, 37, 38, 42, 43]. Most used either total lumbar muscle area (TLA) or total psoas area (TPA) with one study using total psoas volume [34] and one using both TPA and TPV [8]. All area measurements were corrected for patient height (Table 1). Sarcopenia was defined as lean muscle mass below a specific threshold based on either previously published parameters or internally derived based upon sensitivity analyses. Thirteen (54%) of the included studies used previously published thresholds to define sarcopenia [9, 10, 12, 13, 21, 28, 32, 37, 38, 44‐46], whilst the remaining 11 (44%) defined sarcopenia internally [8, 22, 26, 29‐31, 34, 36, 43, 47, 48]. Sex-specific thresholds were frequently calculated in-house for each study based on the local population, with values for TPA ranging from 391 to 414 mm²/m² in women, and 468 to 562 mm²/m² in men. For TLA, values ranged from 370 to 414 mm²/m² in women, and 437 to 550 mm²/m² in men. The median values were 475 mm²/m² for men and 386 mm²/m² for women. The largest thresholds were reported in Western study populations [13], with the difference between Western and Asian populations prompting Higashi et al. [21] to use different thresholds within their study cohort.

A total of 5267 patients were included across the 24 studies. Thirty-five percentage (1820) patients were sarcopenic, with rates varying from 15% (colorectal [10]) to 66% (liver transplantation [28]). Median age was 65 years, and 60% of patients were male. Ninety-three percentage (4876) patients were operated on for malignancy, with the remaining patients undergoing liver transplantation for both malignant and benign indications [26, 28]. Of the patients who were staged using the TNM classification, 12% (359) were stage 1, 40% (1137) stage 2, 38% (1078) stage 3 and 10% (298) stage 4. Thirty-three percentage (613) had well-differentiated tumours, 56% (1037) moderately differentiated tumours and 11% (217) poorly differentiated tumours.

Total complications included all Clavien–Dindo graded complications as well as all other study specific complications. There was a 15% increased risk of any complication, with 54% of sarcopenic patients suffering complications compared with 37% of non-sarcopenic patients (RR 1.15 95% CI 1.04–1.28 p = 0.009), though there was significant heterogeneity amongst studies (I ² 67%, see Fig. 2). The greatest effect size was seen in colorectal and hepatobiliary surgery (RR 1.75 95% CI 1.06–2.87 p = 0.03, OR 1.72 95% CI 1.15–2.58 p = 0.008, respectively). Sarcopenia increased the risk of major complications (Clavien–Dindo > grade 3) by 61%, with 25% of sarcopenic patients suffering major complications compared with 16% of non-sarcopenic patients (95% CI 1.24–4.15 p = <0.00001 I ² 43%, see Fig. 3). Following subgroup analysis, although there was a 7% increased risk of major complications in patients undergoing pancreatic surgery, which failed to reach significance (95% CI 0.83–1.39 p = 0.61, see Fig. 3). The baseline characteristics, including tumour stage and grade, were comparable across the three studies involving pancreatic surgery [8, 30, 49]. Of note, the study period overlaps for the cohorts included in the studies by Peng et al. [30] and Amini et al. [8], attributed to the same institution. The more recent report by Amini et al. [8] used total psoas volume to quantify muscle mass, compared with psoas cross-sectional area which was used by the previous study by Peng et al. [8, 30]. Whilst both methods determined 25% of patients to be sarcopenic, the volumetric technique predicted both overall and major complications. Exclusion of the earlier paper by Peng et al. from the subgroup analysis reduced heterogeneity to 0% and increased the relative risk from 9 to 33%, though this failed to reach statistical significance (p = 0.09).

The presence of sarcopenia significantly increased the risk of post-operative mortality with 2.7% of sarcopenic patients dying within 30 days compared with 0.8% in the non-sarcopenic group (RR 2.06 95% CI 1.02–4.17 p = 0.04, see Fig. 4). Sarcopenia was also associated with an increased risk of 90-day mortality (10% sarcopenic vs. 2.5% non-sarcopenic, RR 3.66 95% CI 2.10–6.38 p = <0.0001, Supplementary Fig. 2) with no heterogeneity in both analyses (I ² 0%, see Fig. 4).

Sarcopenia predicted 1-, 3- and 5-year survival, with the risk decreasing from 61% for 1-year, 45% for 3-year and 25% for 5-year (29% sarcopenic vs. 18% non-sarcopenic RR 1.61 95% CI 1.36–1.91 p = <0.0001 (1-year), 66% sarcopenic vs. 45% non-sarcopenic RR 1.45 95% CI 1.33–1.58 p = <0.0001 (3-year), 50% sarcopenic vs. 56% non-sarcopenic RR 1.25 95% CI 1.11–1.42 p = 0.0003 (5-year), see Fig. 5). Studies reporting 1-, 3- and 5-year survival included liver (1019), pancreas (1022) and transplant (169) patients. In addition, median overall survival was shorter in patients with sarcopenia in 8 from a total of 10 studies [11, 20, 31, 34, 37, 43, 47, 50], ranging from 17.7 to 69.7 months in sarcopenic patients, to 18–146 months in non-sarcopenic patients (Supplementary Fig. 3). In the 2 studies where median survival in sarcopenic patients exceeded non-sarcopenic patients, the difference was non-significant [8, 45].

All five studies reporting disease-free survival included patients with malignant disease. One-year disease-free survival was significantly worse in patients with sarcopenia, with 55% of sarcopenic patients suffering recurrence compared with 45% of non-sarcopenic patients (RR 1.30 95% CI 1.12–1.52 p = 0.0008, see Fig. 6). Whilst the risk was increased for 3-year disease-free survival (81% sarcopenic vs. 80% non-sarcopenic, RR 1.04 95% CI 0.96–1.13 p = 0.04, see Fig. 6), the relative risk increase was significantly reduced at 4%, and for 5-year disease-free survival, there was no difference between sarcopenic and non-sarcopenic groups (RR 1.06 95% CI 1.00–1.13 p = 0.06, see Fig. 6).

Ten studies including 2766 patients reported data pertaining to length of hospital stay. Length of stay was increased in patients with sarcopenia in 8 studies, with 6 identifying a significant difference (see supplementary Fig. 4). The remaining two studies showed no significant difference in length of stay between sarcopenic and non-sarcopenic patients [30, 35]. There were also comparable overall complications [30] and major complications [35] between sarcopenic and non-sarcopenic patients in these studies including oesophago-gastric and pancreatic cancer patients. In addition, three of the studies collected data on intensive care unit (ICU) stay, reporting significantly increased ICU stay in patients with sarcopenia [24, 29, 51, 52].