Power of computed-tomography-defined sarcopenia for prediction of morbidity after pancreaticoduodenectomy

Pancreaticoduodenectomy (PD) is a well-established procedure with low mortality rate in centers of high patient volume [1]. However, the morbidity rate still reaches up to 50% [2], most prominent due to postoperative pancreatic fistulas (POPF). In preparation for PD, patients receive computed tomography (CT) and derived parameters are widely available. Therefore, various efforts were undertaken to predict and stratify the risk of major complications and in particular of POPF based on preoperative CT data with partly controversial results:

The majority of recent studies evaluated the influence of visceral adipose tissue (VAT), on complications and POPF [3‐6]. However, the measures were often not in accordance to the standard procedure, e.g. in regards to the anatomical site of measurement [4, 7], or dependent on expensive segmentation tools [4, 5, 8]. Both facts are limiting a more standardized and clinical use of this parameter. Results were heterogeneous, but most publications showed a positive influence of visceral adipose tissue on complications [3‐5], only few showed contradicting results [7].

Another focus of previous studies was to characterized certain structures of the pancreas on CT images, such as the diameter of the pancreatic duct (DPD), the diameter of the pancreatic parenchyma (DPP), the mean density of the gland or the estimated postoperative remnant volume [4‐6, 8‐17]. Most of these parameters were too complex for clinical routine relying on special image reformation. Therefore only the diameters of the pancreatic parenchyma and duct seem to be reliable enough for further evaluation and showed promising results in earlier reports .

The evaluation of sarcopenic obesity using CT data was introduced in 2013 [18], but only recently this concept was also applied to predict major complications and POPF after PD [8]. Nishida et al. could show that preoperative sarcopenia is a strong and independent risk factor to predict POPF. In this publication 49.6% of the patients were classified as sarcopenic and POPF was significantly higher in these patients, however the results of a Japanese cohort can hardly be applied to a western cohort due to differences in body composition [19].

Although the results found by Nishida et al. are promising they cannot be directly transferred to a western cohort and additionally they lacked a key parameter of sarcopenic obese, the mean attenuation (MA) of skeletal muscle. MA is reported in Hounsfield units and can indicate elevated levels of intramuscular fat accumulation, which is ignored when solely quantifying the muscle area. MA seems to be a powerful prognostic tool, however its capability to predict major complication after PD was never evaluated.

The goal of our study was therefore first to provide an overview over the reported parameters for prediction of major complications and POPF, second to validate selected methods on our own patient cohort using a freely available, custom-made software tool and third to further evaluate the role of sarcopenic obesity by quantifying MA for the first time in this specific context.

The majority of the CT (118 out of 139; 84.9%) was in portal venous contrast. Most relevant imaging data is also shown in Table 2. DPD and DPP could be measured in 125 (89.9%) and 136 (97.8%) patients, respectively. Median DPD was 2.9 mm (Range 0.7–10.7) and more narrow in patients with complications equal to or greater stadium IIIb (p < 0.04) and severe POPF (< 0.01). Median DPP value was 17.7 (6.9–37.9) mm and there was no significant difference regarding major complications or POPF.

A_VAT could be quantified in all 139 patients with a median value of 127.5 (14.5–473.0) cm². A_SAT could only be quantified in 124 (89.2%) patients and was not further evaluated. A_VAT was significantly larger in patients with severe POPF (p < 0.01), in regards to major complications only a trend was seen (p < 0.06).

Areas of psoas, paraspinal, ventral abdominal and total muscle (A_MPSO, A_MSPI, A_MVEN and A_MTOT) showed no significant difference between major complications and POPF. Median muscle attenuation was both lower in groups with major complications (p < 0.01) and POPF (p < 0.01).

Median SMI was 44.7 (19.6–71.0) cm²/m² in patients with or without major complications. There were no significant differences between the different groups mentioned above. Following the published thresholds for SMI and MA for prediction of survival [18] in our cohort there were 35 (11 females and 24 males) and 36 (14 females and 12 males) patients classified as sarcopenic, respectively, accounting for a total of 60 (21 females and 25 males) sarcopenic individuals. There were 28 sarcopenic out of 37 patients (75.7%) that developed major complications and 22 sarcopenic patients out of 26 patients (84.6%) that suffered from severe POPF.

Binary logistic regression was performed for prediction of major complications or severe POPF was tested. Included parameters were sarcopenia, A_VAT and DPD. For major complication no significant results were achieved. For severe POPF binary logistic regression revealed significant results for all three parameters sarcopenia (p < 0.03, OR 4.30, CI 1.154–16.005), A_VAT (p < 0.05, OR 1.006, CI 1.000–1.012) and DPD (p < 0.04, OR 0.725, CI 0.537–0.978).

In the analysis of the Receiver operator characteristics (ROC) curve, the area under the curve (AUC) were 0.638 and 0.723 for DPD and MA, respectively, in regards to major complications (see Fig. 3). For severe POPF corresponding AUC were 0.716 and 0.723, respectively.

The morbidity rate after PD is still a relevant problem and many earlier reports tried to predict major complications perioperatively. In general there is still no standard approach to predict major postoperative complications and severe POPF. Several studies aimed to predict preoperatively POPF by using various CT parameters, such as VAT, pancreatic features and surrogate parameters reflecting sarcopenic obesity:

In our data we were first able to reproduce the known effect of visceral obesity on the development of complications and POPF as reported earlier by others [3‐5, 7, 29, 30]. This is in line with the increasing number of publications that estimate VAT as an overall negative factor to human health by favoring cardiovascular complications known as the metabolic syndrome [31].

Next we demonstrated that the diameter of the pancreatic duct is negatively associated with complications and POPF after PD. This was reported earlier by various studies: Frozanpor used 2 mm [13], three independent publications by Rosso et al., Liu et al. and Su et al. used 3 mm [9, 15, 16], Fang et al. used 3.1 mm [17] and Callery et al. [11] as well as Sandini et al. [5] both used 5 mm as opimum cutoff values to predict POPF. To the contrary Schröder et al. [4] and Nishida et al. [8] did not see significant impact of DPD on POPF. In our data 69 out of 139 patients showed a duct diameter under 3 mm. 23 and 17 of them (33.3 and 24.6%) showed major complications or severe POPF, respectively. There were 14 (20%) and 9 (12.8%) cases of major complications or severe POPF in the 70 patients with a duct size over 3 mm. All in all this methods has some potential, but exactness of measurement is limited and specific cutoffs might therefore be hard to validate.

In contrast to existing studies, our analysis could not confirm that a pancreas thicker than 12 mm significantly increases the incidence of POPF, as shown after DP using stapler closure in 122 distal pancreatectomies [14]: In our cohort only 14 patients (8 women) had a DPP of under 12 mm. This comparison is limited by different surgical resection sites, though.

Several reports addressed the diameter or volume of the pancreatic parenchyma: Frozanpor et al. used data from 182 patients to show that a large pancreas volume is significantly higher in patients suffering from POPF [10]. In a similar approach Roberts et al. used data from 217 patients to show that the pancreas width is significantly higher in patients suffering from POPF [7]. This observation was validated in 266 patients by Nishida et al. [8]. In the data reported in this article, DPP was not relevant risk factor for complications. We did not assess pancreatic volume because this requires 3D software and therefore this method is too complex for a standard approach.

Nishida et al. also stated preoperative sarcopenia to favor POPF [8]: Authors included skeletal muscle areas and SMI into their analysis. Both were reduced in patients with POPF. In the same publication MA was not assessed and therefore characterization of sarcopenic obesity stayed incomplete. In our data MA was the strongest overall predictor of POPF. Therefore we can only speculate this discrepancy between our approach and the work Nishida et al. to be based on technical limitations since the most relevant publication on that issue came out in 2013 [18]: Martin et al. demonstrated that muscle depletion and low muscle attenuation can be used as surrogate parameters for cancer cachexia and thereby assessed for survival prognosis [18]. Additionally, Martin et al. provided threshold values for skeletal muscle index and mean attenuation to diagnose sarcopenia. Looking at our cohort we could only partially confirm these results: MA was associated the strongest with major complications and POPF. Following the thresholds for SMI and MA 4 (2 females and 2 males) and 56 (27 females, 29 males) patients were classified as sarcopen, respectively. Okumura et al. demonstrated psoas muscle mass index and intramuscular adipose tissue content to be associated with mortality after pancreatic resection in 230 patients [32]. Van Dijk et al. could link a low MA to reduction of survival in 199 patients [29]. Namm et al. showed similar results applying a semi-automated technique in 116 patients [33]. Sui et al. linked sarcopenia as defined by total muscle area to the 5 year survival [34]. Most recently Sugimoto et al. showed data from 323 patients and favored a sex-standardized skeletal muscle index as best factor of prediction of both overall and recurrence-free survival [35]. In another quite recent multi-center observation on 120 ptients, Pecorelli et al. stated sarcopenic obesity to be associated with failing rescue from major complications [36].

Sandini et al. in contrary contradicted these findings by normalizing the muscle area to height when they only found an index of both parameters to be predictive for complications [30]. Also contradicting most findings mentioned above, Clark et al. argued that more precise and robust measures of visceral fat, subcutaneous fat, and muscle mass fail to predict cancer progression. They see an explanation in the complexity of cancer biology [37].

In unresectable pancreatic cancers, Ishii et al. stated concluded that the psoas muscle index was the best predictor for survival [38]. Similar results were achieved by Rollins et al. [39]. Further work is needed to develop reliable and most likely gender-, age and ethnicity-dependent cutoff values.

Another question is how to quantify CT parameters in a most efficient way. For analysis of A_VAT, skeletal muscle areas and muscle attenuation, the axial slice at the level of the intervertebral space L3-L4 was chosen. This selection refers to a variety of studies on large cohorts of normal weight to obese subjects, stating best agreement between VAT areas on single imaging slices and total VAT volume for levels cranial of L4-L5 intervertebral space [21, 40], more specific at level L3-L4. For simultaneous evaluation of skeletal muscle tissue and A_VAT, also the level of L3 was shown to be most precise [26]. The software application used for segmentation in this project is easy to use and can be made available upon request. Overall segmentation time is under 5 min. Therefore the presented approach can smoothly be integrated in the clinical setting in order to evaluate sarcopenia on a large scale abdominal CT images.

The presented study has some limitations. First this was a retrospective descriptive study and results should be interpreted with care. Second patient number seems small: we used all consecutive CT datasets of a specific period of time available to perform this study. This resulted in 139 patients (55 women), which seems few to make general assumptions, but is in line with similar publications. In our study the main reason for this limitation was the lack of external CT scans. Due to legal issues, external CTs are not stored in the archive of our institution. Therefore we mainly relied of in-house imaging data.

Next, with a minority of 55 females of the 139 patients, there most likely is a potential gender bias. This can be explained by the also male-dominated prevalence of the of pancreatic adenocarcinoma, that is the most common disease that leads to PD.

Several comorbidities, especially a slightly increased BMI and arterial hypertension, indicate associations with postoperative complications and POPF which may introduce a selection bias. Besides patient-related limitations, there are technical drawbacks of our study: The segmentation of visceral and muscle tissue was accomplished using published density values. This ignores the actual variability in tissue density, e.g. due to ascites. Although this could have been addressed by individual histogram-dependent selection of the respective tissue, we restricted our analysis to the published density range to grant comparability with existing data. Another technical limitation is due to the often incomplete coverage of subcutaneous adipose tissue in the field of view of the CT scanner, especially in men with a large abdominal circumference. Therefore we had to exclude this parameter from our evaluation in order to avoid a selection bias. In future studies this might be overcome by using complementary methods for SAT quantification e.g. ultrasound [41, 42]. Furthermore, most datasets lacked native CT scans. These are mandatory for valid measurement of pancreatic density [3, 12]. With a majority of contrast-enhanced protocols in our CT data we could not quantify pancreatic density in an appropriate way, which is a general limitation in similar retrospective approaches. Last, in 14 cases DPD could not be quantified, most likely because of a very narrow duct, which biases the analysis towards extended DPD widths. The magnitude of this effect could be examined by setting the value of a non-viewable DPD to a standard width, e.g. the respective slice thickness.

Looking at other imaging techniques, magnetic resonance imaging and spectroscopy can replace histological assessment e.g. by Mutli-Echo-Dixon techniques [43]. Earlier it was showed that a fatty infiltration of is associated with POPF [9]. Taken together MR offers even more potential in predictinfg POPF than CT. The major drawback of MR is the generalization of the data, because it is often complex, technically heterogeneous, and above all rare in a common patient collective. Thus computed tomography stays the gold standard for image-based assessment before pancreatic surgery.

All in all and in comparison to existing studies, the presented approach is comprehensive and integrates the most reliable markers as the diameters of pancreatic parenchyma and ductus, areas of adipose and muscle tissue. The quantification of muscle attenuation is new in this field of interest.

Besides the known factors visceral obesity and narrowness of the pancreatic duct, the mean muscle attenuation can easily be examined on routine preoperative CT scans and shows to predict POPF. Even though our study has some limitations, the data presented emphasizes that these parameters should be included in future prospective analyses of morbidity after pancreatic resection.