Computed tomography derived morphomics in the prediction of long-term outcomes in acute pancreatitis

Abstract

BACKGROUND

Acute pancreatitis is a common abdominal pathology with significant morbidity and mortality in select patients. Currently, prediction of outcome and prognosis in patients with acute pancreatitis is based on a variety of indices, scores, and criteria calculated from biochemical, clinical, and imaging parameters, but all currently available predictors of outcome have limitations. Analytic morphomics, a quantitative technique that assesses body composition from cross-sectional imaging, may offer a novel approach to outcome prediction.

AIM

To evaluate whether analytic morphomics-derived parameters can predict recurrence and mortality after an index episode of acute pancreatitis.

METHODS

Following ethical approval, a retrospective single-centre study was conducted at Cork University Hospital. Adult patients presenting with acute pancreatitis between January 2012 and December 2013 were included and followed for 10 years. Patients with pancreatic malignancy, equivocal diagnoses, or age < 18 years were excluded. Cases of acute pancreatitis were classified as acute interstitial oedematous pancreatitis or necrotising pancreatitis. Demographic, biochemical, and clinical data were collected, including the modified Glasgow Imrie severity score. Computed tomography examinations were analysed using CoreSlicer for semi-automated segmentation of skeletal muscle and adipose tissue at the L3 vertebral level. Statistical analysis was performed using Microsoft Excel and Jamovi, including the Mann-Whitney U test, Pearson correlation, and logistic regression.

RESULTS

Seventy-two patients were included with a median modified Glasgow Imrie severity score of 2. Wall muscle (WM) density correlated inversely with disease severity [WM density (Hounsfield units): r = -0.480, P < 0.001; psoas muscle density (Hounsfield units): r = -0.465, P < 0.001]. Higher WM density and greater subcutaneous fat (SCF) area were associated with reduced 30-day mortality (mean difference 14.4 HU, P = 0.003; 65.3 cm², P = 0.033, respectively) and improved 10-year survival. On multivariate analysis, lower WM density and lower SCF area independently predicted short- and long-term mortality, while no morphomic variables were independently associated with recurrence.

CONCLUSION

Skeletal muscle quality and SCF are protective factors in acute pancreatitis. Integration of morphomics with existing severity scores may enhance prognostication. Larger studies are needed to validate these findings.

Key Words: Acute pancreatitis; Computed tomography; Analytic morphomics; Body composition; Skeletal muscle density; Subcutaneous fat; Disease severity; Mortality; Long-term outcomes

Core Tip: Computed tomography-derived analytic morphomics provide objective measures of body composition that can be extracted from routine imaging in acute pancreatitis. In this retrospective cohort with 10-year follow-up, higher skeletal muscle density and greater subcutaneous fat area were independently associated with reduced short- and long-term mortality, whereas other morphomic parameters were not independently associated with recurrence. These findings suggest that integrating computed tomography-based morphomic metrics with established clinical scoring systems may improve prognostic assessment in acute pancreatitis.

INTRODUCTION

Acute pancreatitis is one of the most common abdominal pathologies encountered in medicine, with potentially devastating sequelae in some patients[1]. Multiple causes have been identified, with gallstones and alcohol consumption being by far the most common worldwide. Recurrence of pancreatitis following an acute episode has been reported in up to 32% of patients, while mortality occurs in approximately 5% of cases[2]. Multiple scoring systems based on demographics, biochemical markers, and imaging findings have been developed to predict disease severity and prognosis[1]. However, despite the disease being widely studied, literature on imaging-related predictors of long-term outcomes in acute pancreatitis is scarce. Although evidence suggests that abdominal visceral fat (VF) and subcutaneous fat (SCF) may be correlated with pancreatitis severity and short-term mortality, the association between body composition metrics and long-term sequelae has not been extensively studied[3,4].

Analytic morphomics (AM) is a comprehensive whole-body imaging analysis method that enables quantitative assessment of body composition. Various disease processes, including obesity, sarcopenia, and cachexia, alter body composition, and AM can be used to accurately quantify these alterations. A key advantage of AM is that all analyses can be performed on clinically indicated imaging studies without the need to subject the patient to any further tests or examinations[5].

Several studies have demonstrated that reduced muscle bulk and density measured with AM are predictive of poorer outcomes after surgery, trauma, and malignancy[5]. Additionally, AM has been shown to predict survival and complications after surgery[6,7]. Indeed, Sabel et al[8] found that AM provided objective data that stratified complications and outcome better than age, body mass index, or comorbidities in patients undergoing surgery for colorectal cancer. Literature on AM has suggested that sarcopenia and increased SCF are associated with severe pancreatitis and increased rates of readmission within 1 year. Furthermore, various studies have shown an association between abdominal VF and pancreatitis severity[3,4]. However, the association between body composition and long-term sequelae of pancreatitis is less well studied.

Given the potentially devastating consequences of acute pancreatitis, including chronic pancreatitis, pancreatic necrosis, pseudocyst formation, recurrence, and death, we proposed exploring computed tomography (CT)-derived body composition morphomics for the prediction of recurrence, short-term mortality (within 30 days), and long-term all-cause mortality (within 10 years). In addition, we explored the correlation between body composition morphomics and pancreatitis severity as calculated using the modified Glasgow Imrie severity (mGIS) score[9].

MATERIALS AND METHODS

Study population

Following institutional ethics board approval, a retrospective analysis of adult patients presenting with acute pancreatitis was performed. The local picture archiving and communication system of a tertiary centre was searched electronically using a keyword search function for patients with a diagnosis of “acute pancreatitis” on CT reports from January 2012 to December 2013, inclusive. Patients with a pancreatic malignancy, an equivocal diagnosis, or age < 18 years were excluded. For all patients included in this study, all imaging studies during the subsequent 10-year period were reviewed for evidence of pancreatitis recurrence and all-cause mortality. Patient demographics, including age at diagnosis and sex, were recorded. Aetiology and biochemical markers at presentation were also recorded. Mortality occurring within 30 days of diagnosis was recorded separately.

Pancreatitis severity

The included patients were divided into two groups based on the CT report: Those with acute interstitial oedematous pancreatitis (AIOP) and those with necrotising pancreatitis (NP). Biochemical markers were collected for calculation of the mGIS score, thereby classifying disease severity. Data collected included age, white cell count, calcium, blood urea nitrogen (BUN), lactate dehydrogenase (LDH), albumin, and glucose. Data on partial pressure of oxygen were not available; thus, an mGIS system with a minimum score of 0 and a maximum score of 7 was used, with higher scores indicating more severe disease[9].

Image analysis

Morphomic analysis was performed on previously acquired CT imaging of the abdomen. All imaging data were accessible through the institutional picture archiving and communication system. Intravenous contrast was administered in all cases, and the portal venous phase was used for analysis. Images were analysed using CoreSlicer (Mullie L, Afilalo J; Montreal, QC, Canada), a web-based tool enabling semi-automated segmentation of muscle [-29 to +150 Hounsfield units (HU)] and fat (-190 to -30 HU)[10]. On a single axial slice at the level of the midportion of the 3^rd lumbar vertebra, psoas muscle (PM) cross-sectional area (CSA) and density in HU were calculated. In addition, wall muscle (WM) (which includes both the paraspinal muscles and abdominal WMs) CSA and density, SCF CSA and density, and VF CSA and density were calculated at the same level. Minimal manual correction was required in a few cases. Average values were derived for the left and right PM CSA and attenuation and used for analysis (Figure 1).

Figure 1 Morphomic analysis using CoreSlicer. Semi-automated morphomic analysis conducted at the L3 level using CoreSlicer. Blue: Subcutaneous fat; Green: Right psoas muscle; Teal: Left psoas muscle; Yellow: Visceral fat; Red: Paraspinal and abdominal wall muscles.

Statistical analysis

Data compilation and statistical analysis were performed using Microsoft Excel 2011 (Microsoft Corporation, Redmond, WA, United States) and Jamovi version 2.3.21.0 (The Jamovi project; retrieved from https://www.jamovi.org). The Mann-Whitney test was used to compare non-parametric data between the AIOP and NP groups. Bivariate analysis using the Pearson correlation coefficient was performed to determine the correlation between clinical and morphomic variables. A P value of < 0.05 was considered statistically significant. The Pearson correlation coefficient was also used to determine the correlation between morphomic variables and the mGIS score. Univariate and multivariate logistic regression analyses were used to determine the relationship between clinical/morphomic variables and outcomes (recurrence, 30-day mortality, and 10-year all-cause mortality). As a secondary analysis, non-parametric comparisons of morphomic variables across pancreatitis aetiologies were performed using the Kruskal-Wallis test, with post hoc pairwise comparisons where appropriate.

RESULTS

Patient demographics

In total, 72 patients (26 female, 46 male) with a mean age of 56.0 years (standard deviation = 18.8) were included in the study. Most patients were classified as having AIOP based on the CT report (n = 55, 76.0%). Alcohol (n = 18, 25.0%), gallstones (n = 28, 38.9%), and idiopathic (n = 23, 31.9%) were the most common aetiologies in our population and accounted for 95.8% of patients. The median mGIS score was 2 (interquartile range = 2) (Table 1).

Table 1 Baseline patient demographics and pancreatitis aetiology.

Baseline characteristics	All patients (n = 72)	Men (n = 46)	Women (n = 26)
Age (years)	56.0 ± 18.8	55.6 ± 18.8	56.7 ± 19.2
Modified Glasgow Imrie severity score	2 [2]	2 [2]	1.5 [2]
Aetiology
Alcohol	18 (25.0)	13 (28.3)	5 (19.2)
Gallstones	28 (38.9)	14 (30.4)	14 (53.9)
Idiopathic	23 (31.9)	16 (34.8)	7 (26.9)
Other	3 (4.2)	3 (6.5)	0 (0.0)

Data are presented as mean ± SD, median [interquartile range], or n (%).

Comparison of the three most common aetiologies (alcohol, gallstones, and idiopathic) showed that the risk of recurrence was significantly higher in the alcohol group (χ² = 5.03, P = 0.025); however, there were no differences in the risk of death within 30 days among the three groups (χ² = 0.0724, P = 0.964). Similarly, no differences were noted in the rates of all-cause mortality over 10 years among the three groups (χ² = 2.17, P = 0.337). Additionally, Kruskal-Wallis testing revealed that pancreatitis severity determined by the mGIS score did not differ by pancreatitis aetiology (χ² = 3.56, P = 0.169).

Morphomics

The mean WM CSA was 126 ± 30.7 cm² with a mean density/attenuation of 27.8 ± 13.3 HU, while the mean PM area was 9.35 ± 3.87 cm² with a mean attenuation of 44.5 ± 12.0 HU. A positive correlation was noted between WM CSA and PM CSA (r = 0.752, P < 0.001) and between WM and PM attenuation (r = 0.842, P < 0.001). The mean SCF CSA was 233 ± 137 cm² with a mean attenuation of -95.2 ± 11.5 HU. The mean VF CSA was 205 ± 106 cm² with a mean attenuation of -77.2 ± 14.0 HU. A positive correlation was noted between SCF CSA and VF CSA (r = 0.257, P = 0.029) and between SCF and VF attenuation (r = 0.662, P < 0.001) (Table 2; Supplementary Table 1).

Table 2 Morphomic variables.

Baseline morphomics	All patients (n = 72)	Men (n = 46)	Women (n = 26)	P value
Wall muscle area (cm2)	126 ± 30.7	137 ± 30.4	105 ± 17.7	< 0.001
Wall muscle density (HU)	27.8 ± 13.3	30.1 ± 12.4	23.8 ± 14.2	0.039
Psoas muscle area (cm2)	9.35 ± 3.87	11.1 ± 3.35	6.23 ± 2.52	< 0.001
Psoas muscle density (HU)	44.5 ± 12.0	44.8 ± 11.7	44.0 ± 12.8	0.784
Subcutaneous fat area (cm2)	233 ± 137	191 ± 97	309 ± 165	< 0.001
Subcutaneous fat density (HU)	-95.2 ± 11.5	-93.1 ± 13.2	-98.8 ± 6.15	0.055
Visceral fat area (cm2)	205 ± 106	230 ± 115	162 ± 69.1	0.008
Visceral fat density (HU)	-77.2 ± 14.0	-78.5 ± 15.3	-74.8 ± 11.2	0.053

Data are presented as mean ± SD. The Mann-Whitney U test was used for data that were not normally distributed, whereas Student’s t-test was used for data that followed a normal distribution.

HU: Hounsfield units.

Morphomic correlation with biochemical parameters and the mGIS score: Negative linear correlations were noted between WM HU (r = -0.480, P < 0.001) and the mGIS score as well as between PM HU (r = -0.465, P < 0.001) and the mGIS score. In addition, a weak positive correlation was noted between VF CSA and the mGIS score (r = 0.239, P = 0.043) (Table 3). However, morphomic variables did not vary with pancreatitis subtype in our cohort, suggesting that body composition is not predictive of AIOP vs NP (Supplementary Table 2).

Table 3 Correlation analysis between morphomic variables and modified Glasgow Imrie severity score.

Variable	mGIS score (0-7)
	Pearson’s r	P value
Wall muscle area (cm2)	-0.190	0.875
Wall muscle density (HU)	-0.480	< 0.001
Psoas muscle area (cm2)	-0.153	0.199
Psoas muscle density (HU)	-0.465	< 0.001
Subcutaneous fat area (cm2)	0.079	0.509
Subcutaneous fat density (HU)	0.050	0.675
Visceral fat area (cm2)	0.239	0.043
Visceral fat density (HU)	-0.127	0.290

Pearson’s correlation coefficient was used for univariate analysis.

HU: Hounsfield units; mGIS: Modified Glasgow Imrie severity.

Associations were observed between albumin and several morphomic variables, including WM HU (r = 0.265, P = 0.028) and PM HU (r = 0.337, P = 0.005). LDH was also associated with WM density; however, a negative correlation was observed with WM HU (r = -0.241, P = 0.050). Similarly, BUN was negatively correlated with muscle density: WM HU (r = -0.309, P = 0.009) and PM HU (r = -0.364, P < 0.001). The full correlation matrix between all body composition variables and biochemical markers is available in Supplementary Table 1.

Morphomic correlation with patient outcomes: Body composition correlated with both the risk of death and recurrence. Mann-Whitney U testing revealed that higher WM HU [mean difference (MD) = 14.401 HU, P = 0.003], PM HU (MD = 10.469 HU, P = 0.010), and SCF CSA (MD = 65.279 cm², P = 0.033) were associated with an increased chance of short-term (30-day) survival. Interestingly, higher WM HU (MD = 7.936 HU, P = 0.009) and higher PM HU (MD = 6.366, P = 0.014) were also predictive of an increased risk of recurrence. While higher SCF CSA predicted short-term survival, lower SCF CSA predicted recurrence (MD = -58.451, P = 0.037). Higher WM HU (MD = 13.82, P < 0.001), PM HU (MD = 9.52, P < 0.001), and PM CSA (MD = 2.14, P = 0.033) were predictive of long-term (10-year) survival (Table 4).

Table 4 Association between morphomic variables and survival or recurrence.

Variable	Survival (30-day)		Survival (10-year)		Recurrence
	MD	P value	MD	P value	MD	P value
WM area (cm2)	-1.398	0.905	9.73	0.145	0.793	0.931
WM density (HU)	14.401	0.003	13.82	< 0.001	7.936	0.009
PM area (cm2)	1.927	0.153	2.14	0.033	1.117	0.366
PM density (HU)	10.469	0.010	9.52	< 0.001	6.366	0.014
SCF area (cm2)	65.279	0.033	38.04	0.216	-58.451	0.037
SCF density (HU)	-8.240	0.059	-3.33	0.125	-0.100	0.951
VF area (cm2)	0.673	0.986	-2.86	0.833	-42.016	0.056
VF density (HU)	-0.895	0.762	-3.00	0.393	3.686	0.347

Values were generated using the Mann-Whitney U test.

HU: Hounsfield units; MD: Mean difference; PM: Psoas muscle; SCF: Subcutaneous fat; VF: Visceral fat; WM: Wall muscle.

In a multivariate logistic regression model including variables that were statistically significant in the univariate analysis and clinically significant variables (age, sex, mGIS score, and pancreatitis subtype), lower WM HU and lower SCF CSA remained statistically significant predictors of both short- and long-term mortality in patients with acute pancreatitis. No morphomic variables were predictive of recurrence in the multivariate model (Tables 5 and 6).

Table 5 Multivariate logistic regression in predicting death and recurrence.

Variable	Death (30-day)		Recurrence
	Estimate1	P value	Estimate1	P value
Wall muscle density (HU)	-0.44137	0.019	-0.08008	0.283
Psoas muscle density (HU)	0.00581	0.957	0.11924	0.134
Subcutaneous fat area (cm2)	-0.06031	0.018	-0.00642	0.119
Age (years)	-0.10983	0.096	-0.01033	0.635
Sex (male-female)	2.23529	0.310	1.35976	0.149
Pancreatitis subtype (NP-AIOP)	0.44864	0.814	1.03349	0.184
mGIS score (0-7)	0.53263	0.447	-0.39655	0.230

¹The estimate represents the log odds of death vs survival and recurrence vs no recurrence.

AIOP: Acute interstitial oedematous pancreatitis; HU: Hounsfield units; mGIS: Modified Glasgow Imrie severity; NP: Necrotising pancreatitis.

Table 6 Multivariate logistic regression in predicting all-cause mortality over 10 years.

Variable	Multivariate
	Estimate1	P value
Wall muscle density (HU)	-0.21702	0.004
Psoas muscle density (HU)	0.07638	0.222
Psoas muscle area (cm2)	-0.15725	0.266
Subcutaneous fat area (cm2)	-0.01374	0.038
Age (years)	-0.00200	0.947
Sex (male-female)	1.07564	0.328
Pancreatitis subtype (NP-AIOP)	-0.77738	0.455
mGIS score (0-7)	0.25253	0.458

¹The estimate represents the log odds of death vs survival in the multivariate analysis.

AIOP: Acute interstitial oedematous pancreatitis; HU: Hounsfield units; mGIS: Modified Glasgow Imrie severity; NP: Necrotising pancreatitis.

Morphomic analysis by pancreatitis aetiology: As a secondary analysis, CT-derived morphomic parameters were compared across pancreatitis aetiologies (alcohol, gallstones, idiopathic, and other). SCF CSA differed significantly between groups (Kruskal-Wallis χ² = 13.36, df = 3, P = 0.004), driven by higher SCF CSA in gallstone-related than alcohol-related pancreatitis (P = 0.002). No significant pairwise differences were observed for abdominal WM HU despite a significant overall test (P = 0.043), and PM HU did not differ by aetiology (P = 0.368) (Table 7).

Table 7 Morphomic analysis by pancreatitis aetiology.

Variable	Alcohol (n = 18)	Gallstones (n = 28)	Idiopathic (n = 23)	Other (n = 3)	P value
Subcutaneous fat area (cm2)	157 [120]	262 [122]	184 [145]	184 [205]	0.004
Wall muscle density (HU)	34.4 [13.0]	27.0 [9.4]	29.1 [20.1]	38.3 [7.3]	0.043
Psoas muscle density (HU)	47.0 [12.2]	44.9 [12.7]	44.9 [12.8]	52.3 [3.3]	0.368

Data are presented as median [interquartile range].

Group comparisons were performed using the Kruskal-Wallis test.

HU: Hounsfield units.

DISCUSSION

Acute pancreatitis is a common condition, with devastating complications and sometimes poor outcomes in a select cohort of patients. The use of CT-derived AM in predicting these outcomes in patients with acute pancreatitis demands further investigation and research. Early studies, such as that by Edwards et al[11], suggest that morphometric measurements are a useful adjunct for identifying patients at increased risks of complications and subsequent readmission after pancreatic resection. In combination with physical examination and biochemical markers, contrast-enhanced CT plays a pivotal role in diagnosing acute pancreatitis by providing characteristic findings such as pancreatic enlargement and oedema, fat stranding, and heterogenous attenuation of the gland. Additionally, CT plays a vital role in differentiating AIOP and NP, which has significant prognostic implications for the patient[12].

An additional and sometimes overlooked advantage of CT is that it provides precise and highly reproducible data regarding SCF distribution and muscle density. In routine clinical practice, these additional data are not quantified or analysed during review of the CT imaging study by the reporting radiologist. The aim of this study was to quantify relevant morphomic data from CT scans performed on admission in patients with acute pancreatitis and to investigate whether these body composition metrics can predict prognosis in patients with acute pancreatitis.

The findings of the current study showed that higher muscle density correlated with an increased chance of survival in both the short and long term. Higher muscle density also correlated with a lower mGIS score and therefore with less severe acute pancreatitis on admission. This suggests that patients with fatty infiltration of muscle, reduced muscle quality, and thus lower muscle attenuation on CT are at greater risk of developing severe acute pancreatitis and experiencing greater disease severity, morbidity, and mortality. This therefore highlights sarcopenia as a key parameter in morphomic investigations and as a predictor of pancreatitis outcomes. Although more commonly associated with chronic pancreatitis, sarcopenia has been shown to correlate with greater severity of acute pancreatitis and a greater risk of mortality in previous morphomic studies[13].

Interestingly, greater muscle density/quality was also associated with a greater risk of recurrent acute pancreatitis; however, this may be attributed to survivorship bias. Farquhar et al[13] reported similar findings in their study, whereby patients without sarcopenia (i.e. greater muscle quality) were more likely to develop complications such as peripancreatic collections and pseudoaneurysms, likely because they survived for longer.

In addition to outcomes, our study also examined the relationship between biochemical markers and body composition metrics. Some of our findings aligned with our expectations: Muscle density showed a negative correlation with BUN, LDH, and age, while it positively correlated with the albumin level. These relationships have already been established by other studies. For instance, a study comparing LDH levels in young athletes vs untrained subjects demonstrated a correlation between muscle density and LDH[14]. Similarly, several studies have identified a simultaneous decline in albumin levels and muscle density with ageing, with research involving elderly patients demonstrating that higher albumin concentrations are associated with greater muscle strength[15]. However, while these studies provide important context, none have specifically examined these correlations in the setting of acute pancreatitis, highlighting the need for further research. When using scoring systems such as the standard GIS and mGIS scores, which incorporate traditional markers such as age, LDH, BUN, and glucose to predict the course of acute pancreatitis, the addition of morphomic measurement of body composition metrics from CT scans may enhance the accuracy of prognostic models and improve clinical decision-making. Although the sample size analysed herein was small and underpowered for the development of such a model, the study suggests that there is potential for this approach and highlights the need for further research in this area.

Although several significant findings were identified, this study has limitations. All data were obtained from a single tertiary centre through retrospective analysis, which may introduce selection bias. In addition, the relatively small cohort size limited statistical power for more complex outcome modelling and detailed subgroup analyses. The revised Atlanta severity classification could not be reliably assigned because the time-resolved clinical data required to define organ failure and local or systemic complications were not consistently available throughout the index admission[16]. Similarly, body mass index was not uniformly documented at presentation, precluding meaningful body mass index-based analyses. Although there is no universally accepted scoring system for pancreatitis severity, the GIS score remains a widely used and validated tool for risk stratification in clinical practice; however, the mGIS score was required in this study because of missing partial pressure of oxygen data in a subset of patients.

Despite the limitations, this study has several strengths. First, to the best of our knowledge, this study is the first to explore body composition morphomics and their relationship with long-term (10-year) pancreatitis outcomes. While morphomics has been extensively studied in the fields of surgery, oncology, and trauma, its role in pancreatitis outcomes is poorly recognised. Additionally, this study utilises real-world patient data and describes a method for body composition analysis that is both highly reproducible and readily achievable for centres worldwide. Future research in this area should focus on including larger data sets and developing predictive models incorporating body composition metrics, biochemical markers, and demographics.

CONCLUSION

Body composition metrics can be easily and accurately assessed using standard morphomic techniques applied to CT scans performed for the investigation of acute pancreatitis in routine clinical practice. These metrics, particularly muscle density, may be predictive of disease severity, recurrence, and short- and long-term survival in patients presenting with acute pancreatitis. Further research with larger cohorts is required to validate our findings.