Pancreatic size and fat content in diabetes: A systematic review and meta-analysis of imaging studies

Objectives Imaging studies are expected to produce reliable information regarding the size and fat content of the pancreas. However, the available studies have produced inconclusive results. The aim of this study was to perform a systematic review and meta-analysis of imaging studies assessing pancreas size and fat content in patients with type 1 diabetes (T1DM) and type 2 diabetes (T2DM). Methods Medline and Embase databases were performed. Studies evaluating pancreatic size (diameter, area or volume) and/or fat content by ultrasound, computed tomography, or magnetic resonance imaging in patients with T1DM and/or T2DM as compared to healthy controls were selected. Seventeen studies including 3,403 subjects (284 T1DM patients, 1,139 T2DM patients, and 1,980 control subjects) were selected for meta-analyses. Pancreas diameter, area, volume, density, and fat percentage were evaluated. Results Pancreatic volume was reduced in T1DM and T2DM vs. controls (T1DM vs. controls: -38.72 cm3, 95%CI: -52.25 to -25.19, I2 = 70.2%, p for heterogeneity = 0.018; and T2DM vs. controls: -12.18 cm3, 95%CI: -19.1 to -5.25, I2 = 79.3%, p for heterogeneity = 0.001). Fat content was higher in T2DM vs. controls (+2.73%, 95%CI 0.55 to 4.91, I2 = 82.0%, p for heterogeneity<0.001). Conclusions Individuals with T1DM and T2DM have reduced pancreas size in comparison with control subjects. Patients with T2DM have increased pancreatic fat content.


Introduction
The pancreas plays a key role in diabetes mellitus, a progressive disease characterized by chronic hyperglycemia [1] in the context of insulin resistance [2] and/or beta cell dysfunction and death [3]. Beta cell loss secondary to apoptosis leads to a reduction in beta cell mass [4,5]. Although islets of Langerhans represent only 1% of the total pancreas, autopsy studies have demonstrated reduced pancreas size in both type 1 [6] and type 2 diabetic subjects [7].
Insulin deficiency and the lack of a trophic effect of insulin on acinar cells [6]as well as the chronic inflammation associated with insulitis [8,9]may explain the reduction in pancreas size in type 1 diabetes (T1DM), whereas atherosclerosismight play a role in type 2 diabetes (T2DM) [10,11]. In both type 1 and type 2 diabetes, pancreatic size reduction may also be associated to exocrine pancreatopathy [12,13]. However, the reduction in pancreatic size may also be the cause, and not a consequence of diabetes [14,15].
Imaging studies are expected to produce reliable information regarding pancreas size. However, while some studies using ultrasound (US), computed tomography (CT), and magnetic resonance imaging (MRI) to assess pancreas size in diabetes have shown reduced pancreatic size in individuals with diabetes as compared to controls [16][17][18], no differences were observed in others [19,20]. Such inconclusiveness may be related to the small sample size of most studies evaluating this issue.
Interestingly, CT and MRI are widely used to measure liver steatosis [21,22], which is closely related to obesity and diabetes [22]. More recently, imaging protocols have produced accurate non-invasive measurements of pancreatic fat content in humans [23,24]. Excess ectopic fat storage has been linked to insulin resistance [22], and pancreatic fat content has been negatively associated with insulin secretion [25].
The aim of the present study was to systematically review the literature and synthesize data regarding pancreatic size and fat content in diabetes using meta-analysis.

Data sources and searches
To identify observational studies evaluating pancreatic size or fat content by imaging in diabetes, the literature (Medline and Embase) was searched for studies using the three major imaging methods (US, CT, and MRI) for pancreas evaluation from inception until May 2017. No language or date restrictions were applied. Medical subject heading (MeSH) terms and key words included in the search were as follows: pancreas, diabetes, imaging, radiology, ultrasound, tomography, and magnetic resonance. Detailed Medline and Embase search strategies are shown as Fig 1. Also, the references of selected articles were manually searched. Titles and abstracts of all retrieved articles were independently reviewed by two physicians, T.S.G (radiologist) and T.H.R. Disagreements were resolved by consensus. The full text of selected articles was examined. The study is registered at PROSPERO under the number 2016: CRD42016039853.

Study selection
We included prospective and retrospective observational studies evaluating pancreatic size and/or fat content by US, CT, or MRI in T1DM and/or T2DM patients as compared to nondiabetic subjects. Exclusion criteria were as follows: case report design, inclusion of pediatric patients (<12 years), absence of control group, and no clear description of imaging and postprocessing technique. If duplicate studies were detected, the most complete report with the longest follow-up was included.

Data extraction
One reviewer (T.S.G) conducted data extraction and a second investigator checked data extraction for accuracy (T.H.R). Data were extracted on year of publication, number of T1DM, T2DM, and non-diabetic subjects, and imaging method used for pancreatic assessment. In addition, the following data were extracted on pancreatic parameters (mean and standard deviation) in the three groups of interest (T1DM, T2DM, and non-diabetic subjects): pancreatic diameter in cm, area in cm 2 , volume in cm 3 , density in Hounsfield units (HU), and percentage of pancreatic fat.

Quality assessment
Quality assessment of studies included in meta-analyses was performed using the Newcastle-Ottawa scale [26], a scoring system that takes into account selection of groups, comparability between groups, and ascertainment of the exposure (in the case of studies included in this meta-analysis, it refers to the imaging method and to the technique by which parameters were measured).

Data synthesis and analysis
Absolute changes in size (diameter, area, or volume), density, and fat percentage in patients with diabetes and control groups are presented as means ± standard deviation (SD). Cochran's Q test was used to evaluate heterogeneity between studies. A p value <0.1 was considered statistically significant. The I 2 test was also conducted to evaluate the magnitude of the heterogeneity between studies. Heterogeneity was defined as I 2 >50%. A random effects model was used for all analyses.
The contribution of individual studies to the overall heterogeneity was explored using meta-regression, subgroup analyses, and sensitivity analyses by removing each study at a time and re-running the meta-analyses. In some cases, these procedures were not feasible due to an insufficient number of studies/patients.
Funnel plot asymmetry was evaluated by Begg and Egger tests. The impact of small-study bias was considered as significant if p value <0.1 [27]. Analyses were conducted using Stata software version 11.0 (Stata Inc, College Station, Texas).

Results
A total of 5,634 potentially relevant studies were initially identified, 1,532 in Medline and 4,102 in Embase. Hand search of reference lists resulted in the inclusion of an additional eight articles (5,642 articles). After removal of 1,047 duplicates, 4,595 citations were screened based on titles and abstracts. Twenty-nine were selected for full-text review, and finally 23 articles and one poster fulfilled the inclusion criteria.
Of the 24 studies selected, seven were not included in the meta-analysis: in six, data were not extractable [12,24,[28][29][30][31], and in one study measuring pancreatic fat divided by splenic fat the parameters of interest could not be combined [23]. The remaining 17 studies were included in meta-analyses: two evaluating diameter [19,32], two evaluating area[16, 33]eight evaluating volume [17,18,20,[34][35][36][37][38], two evaluating density [17,39], and six evaluating fat content [17,18,[40][41][42][43]. Studies assessing multiple parameters were included in more than one meta-analysis. The flowchart of study selection is depicted in Fig 2. Characteristics of the studies included in the systematic review and meta-analyses are presented in Table 1. The studies were published between 1985 and 2015 and included 3,403 participants: 284 T1DM patients (minimum-maximum: 12 to 60 patients), 1,139 T2DM patients (11 to 198 patients), and 1,980 control subjects (9 to 660 patients). Overall mean age was 59.4 years (minimum-maximum: 13 to 100 years) (33.9 in T1DM, 58.1 in T2DM, and 63.0 years in controls), and overall mean BMI was 26.84 kg/m 2 (22.72 kg/m 2 in T1DM, 27.05 kg/m 2 in T2DM, and 27.21 kg/m 2 in controls). Mean duration of disease was 8.9 years in T1DM and 6.5 years in T2DM. CT scanners used for evaluation of pancreatic volume ranged from one to 64-detector-rows. MRI evaluating pancreatic volume and/or pancreatic fat percentage was performed in 1.5 T or 3.0 T scanners. In both CT and MRI scans, the contour of the pancreas in each slice was annotated, automatically generating the area of each slice. Pancreatic volume was then calculated by multiplying the area of pancreatic tissue on each section by the interval between slices. MRI methods for estimation of pancreatic fat content included spectroscopy or chemical shift sequences for separation of water and fat. CT quantification of pancreatic fat was performed by means of a histogram.
Heterogeneity was explored by sensitivity analysis and each study was excluded at a time. Heterogeneity was reduced to 47.8% (p for heterogeneity = 0.147) after omission of the study by Goda et al [20], while no change was observed when other studies were excluded. This may have resulted from patient selection bias, as the mean age of T1DM patients in this study was 48.7 years, while the mean duration of diabetes was 9.4 years-suggesting that T2DM patients may have been misdiagnosed withT1DM. Interestingly, this was the only study using CT for volume assessment, and thus sensitivity and subgroup analysis of pancreatic volume based on image technique were coincident; in this study, pancreatic volume in T1DM patients was 26.3 cm 3 smaller than that of controls (95%CI, -37.16 to -15.44). Moreover, the present meta-analysis of MRI studies shows a mean reduction of -44.08 cm 3 (95%CI -57.16 to -30.99) in pancreatic Table 1. Summary of studies evaluating pancreas size and fat content by imaging methods in diabetes.

No. of subjects Results
Authors  Despite the small number of studies, we performed meta-regression with age, BMI, and duration of diabetes as covariates. Although not statistically significant, a reduction in heterogeneity from 70.2% to 52.76% (p = 0.355) was observed in the model considering duration of diabetes. Interestingly, the T1DM patients with longer diabetes duration had the lowest pancreatic volume (Fig 4).
Interestingly, Yokota et al demonstrated a decrease in pancreatic density with increasingly impaired glucose homeostasis [29]. Healthy individuals had higher pancreatic densities, which decreased progressively from impaired glucose tolerance to diabetes [29]. However, Begovatz et al did not find differences in pancreatic fat content between subjects with normal glucose, impaired fasting glucose, or T2DM patients when pancreatic fat was evaluated by MRI [31].

Quality of studies and small-study bias
The studies included in meta-analyses were assessed for quality using the Newcastle-Ottawa scale (Table 2). Overall, studies had low/moderate quality; most had a score of 6 or 7 points from a maximum of 9.
The funnel plot asymmetry test revealed no major small-study bias regarding volume or fat content in T2DM patients (p = 0.458 and 0.484 respectively). However, a possible small-study bias was detected for volume in T1DM patients (p = 0.041).

Discussion
In this systematic review with meta-analysis of imaging studies, a reduction in pancreatic size was observed in both T1DM and T2DM patients. In addition, an increase in pancreatic fat content was seen in T2DM subjects.
Pancreatic size was evaluated in terms of diameter, area, and volume. In T1DM patients, the results show decreased pancreatic size in comparison to non-diabetic controls for all three parameters. In turn, volume, but not diameter, was reduced in T2DM patients; area was not meta-analyzed because only one study focusing on this aspect included T2DM subjects. Interestingly, a comparison between T1DM and T2DM revealed smaller pancreatic diameter in T1DM individuals. A single study assessing pancreatic area showed smaller dimensions in T1DM individuals vs. T2DM individuals, and the only study assessing volume observed no differences between T1DM and T2DM patients.
Volume, which provides three-dimensional data, is the best parameter to assess organ size. The present findings show smaller pancreatic volume in both T1DM and T2DM patients in relation to controls, but data are insufficient to establish a firm conclusion regarding the comparison between these two types of diabetes. However, our meta-analyses focusing on volume suggest that T1DM subjects may in fact have smaller pancreatic volume in relation to T2DM individuals: a difference of -38.72 cm 3 (95%CI -52.25 to -25.19) was observed for T1DM vs. controls, and a difference of -12.18 cm 3 (95%CI -19.1 to -5.25) was observed for T2DM vs. controls. Although a formal statistical test was not performed, it is fair to assume that pancreatic volume was smaller in T1DM than T2DM patients, since the 95%CIs did not overlap.
An intriguing finding of this systematic review is the low heterogeneity of MRI studies, as opposed to the high heterogeneity of CT studies. This might be due to the fact that the MRI studies cover a shorter time interval and more recent years (2007-2015) as opposed to the CT studies (1992-2014). Moreover, the magnitude of volume reduction detected by each imaging method was remarkably different (T1DM: -44.08 cm 3 for MRI vs. -26.3 cm 3 for CT; T2DM:-21.65 cm 3 for MRI vs. -7.5cm 3 for CT), with MRI showing consistently higher differences in volumes between patients with diabetes and controlsthan the results obtained by CT. These differences were unexpected, since the tool used to measure pancreatic volume is similar in The number of stars indicates the quality of each item evaluated: minimum 0, maximum 1 star for selection and outcome; and minimum 0, maximum 2 stars for comparability. The maximum possible overall score is 9. both imaging methods and no plausible technical reason can justify lower volumes measured by MRI. Furthermore, a recent study evaluating T1DM patients with MRI or CT did not report differences in pancreas size measured by the two methods [44].
There is a large inter-individual variation in pancreas morphology and volume related to body size and age in healthy populations [45,46]. This may be a relevant source of confusion in studies with T1DM and T2DM individuals-T1DM patients are usually younger, and, as shown in the present study, possibly have a smaller pancreas; conversely, T2DM patients might be older than controls, and pancreas size may decrease with age [35]. However, most studies in the present review included BMI-and age-matched controls, and neither subgroup analysis nor meta-regression considering these possible confounders showed an impact on heterogeneity. Reduced pancreatic volume and weight are present from early phases of T1DM, as demonstrated by a study comparing the pancreasofT1DM donors and controls [14], even after correction for confounders [15]. Recently, Virostko et al [44] have suggested progressively smaller pancreatic volume with increased duration of T1DM (decline rate of 0.013 cm 3 /kg per year). This is supported by the findings of our meta-regression showing that TD1M patients with longer disease duration had lower pancreatic volumes. Thus, monitoring variations in pancreatic volume might be useful to predict diabetes in high-risk individuals [14]. Recently, Yun et a lhave reported that pancreatic volume reduction rate calculated by serial CT volumetry is a significant predictor of new-onset diabetes in patients undergoing pancreaticoduodenectomy [47].
Pancreatic fat content is evaluated by means of density or fat percentage, with percentage being more precise. Some studies suggest an association between increased pancreatic fat and diabetes. Kim et al [23] have shown that two CT indexes-the difference between pancreatic and splenic density and the pancreas to spleen density ratio-are higher in patients with impaired glucose tolerance. In line with this, a study designed to compare pancreatic fat content and beta cell function found increased lipid deposition in the pancreas of diabetic patients as compared to healthy subjects [24]. Furthermore, in T2DM patients, obesity was associated with lower pancreatic density evaluated by CT, indicating higher pancreatic fat content [30]. In reverse order, return of normal insulin secretion and reduction of pancreatic fat content has been demonstrated after eight weeks of a low energy intake diet [40]. Our data indicate that pancreatic fat content is increased in T2DM patients, which may reflect a paracrine effect of insulin. Insulin resistance causes increased insulin secretion by beta cells, and the higher local insulin concentration may induce fat deposition. A similar phenomenon occurs in the liver when pancreatic islets are transplanted into the portal vein [48][49][50]. Pancreatic islets delivered to the hepatic sinusoids engraft and produce insulin, and focal steatosis is observed in 20% to 60% of islet recipients [48][49][50]. More interestingly, transplanted islets surrounded by fat have reduced function, probably as a result of lipotoxicity [50,51]. Conversely, a low-fat diet and leptin overexpression have been shown to reduce fat content around islets, improving islet function in an animal model [51]. Taken together, these findings suggest that pancreatic fat accumulation might be a result of the higher local insulin levels in an insulin-resistant environment, and that pancreatic lipid deposition may further impair islet function.
Our results have some practical implications. First, the finding of a small or fatty pancreas using imaging techniques should prompt a recommendation for proper biochemical investigation of diabetes. Second, as there is some evidence in the literature linking pancreas atrophy in T1DM and T2DM patients with pancreatic exocrine deficiency [12,13], differential diagnosis of chronic diarrhea in diabetic patients should consider exocrine pancreatopathy, a hypothesis that could be corroborated by diagnostic imaging.Important to say, ultrasonographic evaluation of the pancreas is not as accurate as evaluation by CT or MRI.
The present review has limitations that must be addressed. First, there are few studies assessing each parameter, precluding adequate exploration of heterogeneity and increasing the risk for small-study bias. Secondly, CT and MRI have different contrast mechanisms to quantify pancreatic fat content, and both lack detail in measurement methodologies. However, MRI is considered to be more reliable. Pancreatic fat measurement by CT is based on a histogram analysis in which anarbitrary threshold of -190 to -30 HU is applied to identify fat containing voxels [17], whileMRI uses spectroscopy or chemical shift techniques for this purpose [42]. These differencesmay explain discrepancies that can possibly be found between CT and MRI regarding pancreatic fat calculation.Thirdly, pediatric patients were excluded, although T1DM has an early age of incidence. As pancreatic volume is constitutionally associated with body size, and children have a relatively small body size, we think it would be better not to compare children with the adult population. Additionally, the overall quality of studies ranged from low to moderate. Furthermore, the studies included in the analysis date from 1985 to 2015, a time interval in which dramatic changes happened in imaging techniques, which might possibly influence the results of volume/area determinations.However, we believe that the findings of increased fat content and decreased pancreas size consistently point in the same direction and should not be dismissed.
In summary, the present data indicate that reduced pancreas size and increased fat content are features of diabetes. Further longitudinal studies are required to elucidate the cause and effect relationship between pancreatic size and diabetes, as well as the possible causes of pancreas shrinkage and fat deposition. A better understanding of the mechanisms of altered pancreas morphology and fat deposition in diabetes may lead to new insights in preventing, predicting, and treating patients with diabetes.