A Novel Imaging Score for Prognostication in Cirrhosis
The South West Liver unit (Plymouth, United Kingdom), assesses and performs the post-operative care for patients living in the South West of England who are being considered for, or who have received, a liver transplant. The unit maintains very close links with Kings College Hospital (KCH), London where the surgery is performed. The unit also manages patients with general hepatological problems. An abdominal CT is routinely performed as part of the LT assessment. Patients with cirrhosis who were not part of the transplant assessment or hepatoma programme were included in the study if a CT scan had been performed for clinical reasons. Abdominal CT scanning was not performed routinely in patients with early cirrhosis (Child-Pugh A), although these patients did have 6-monthly liver ultrasound assessment for hepatoma surveillance. As the data were retrieved from existing retrospective sources, ethics committee approval was not required.
A retrospective–prospective cohort was identified from the departmental liver transplant database (maintained by JDM). The time period patients were identified for the study was May 2003 to May 2012. All patients who had computed tomography (CT) performed as part of the pre-transplant assessment were identified. To prevent cohort bias, patients with cirrhosis in whom a CT had been performed were identified from the in-patient database (maintained by TJSC) and from out-patients. The diagnosis of cirrhosis was made using well-defined parameters including clinical signs, biochemical criteria, radiological criteria and histological criteria as previously described. On the database, the following patient demographics were recorded: gender, age (years), liver disease aetiology, acute or chronic liver disease, the presence of hepatocellular carcinoma (HCC), serum sodium (mmol/L), serum creatinine (μmol/L), serum bilirubin (μmol/L) and international normalised ration (INR). With this information, it was possible to calculate the MELD, UKELD, MELD-Na and iMELD. The Childs-Pugh Score was not calculated due to incomplete data at the time of CT scanning. The overwhelming majority of the patients were Caucasian, so race was not a variant entered into the analysis. Patients with acute liver failure, an existing liver graft, and hepatocellular carcinoma were excluded from the final analysis. The validation set was collected from the cirrhosis and liver transplant database at the Royal Liverpool University Hospital (RLUH) from June 2008 to June 2013 and patients were also identified from the radiology database (CF). The same baseline parameters were collected as the training set. In the training and the validation sets, the reader of the LAAR score was aware of the patient outcome, but was not aware of the results of the other prognostic score results.
Computed tomography (CT) cross-sectional liver imaging (General Electric –GE Medical, Milwaukee, IL, USA) was reviewed and measurements were made when the maximum liver area was identified from a single 'slice.' For standard liver imaging, slices were taken at every 0.625 mm. On this image, a 'region of interest' (ROI) was conceptualised on the liver. A 'best-fit' ellipsoid was made around the liver to approximate liver area. The area of this ellipsoid was calculated using the formula πab (where π = 3.142, a = radius of ellipse height and b=radius of ellipse width). An approximated abdominal area was calculated using the formula πcd [where π = 3.142, c=radius of abdominal area height (diameter measured from sternum to base of spinal dorsal vertebrae) and d=radius of width (diameter measured from intercostal ribs in cross-sectional horizontal midline of the abdomen)]. All measurements were performed by the same individual (TJC). How the liver and abdominal areas were calculated is shown in Figure 1. If the liver extended round the midline to the left, often two separate liver areas with best-fit ellipses were calculated and added together to represent the liver area for that slice. The liver abdominal area ratio (LAAR) was calculated by dividing the liver area by the abdominal area and then multiplying the result by one hundred:
(Enlarge Image)
Figure 1.
Derivation of the LAAR score. A region of interest (ROI) was identified on a computed tomography slice where the greatest liver area was observed (a). This was placed within a best-fit ellipse. A best-fit ellipse was then measured assessing the abdominal circumference between the ribs in the midline and the anterior posterior diameter was measured from the sternum to the dorsal spinous process. The measurements were then made as shown in (b). The LAAR score is calculated by determining the liver area divided by the abdominal area Ă—100.
A control group was collected to determine a normal range for the LAAR score. Controls were identified from the hospital computer radiology system over a 10-day period in patients in whom a CT abdomen had been performed. Patients were included if the CT report did not identify any features of chronic liver disease or any liver abnormalities on imaging and in whom all patients had completely normal liver biochemistry on blood testing. Patients with cirrhosis were censored at the time of death or at the time of LT, when patients were no longer exposed to the risk of dying due to the complications of cirrhosis.
All values were expressed as median values with interquartile ranges. The primary end point was to determine if the LAAR score could predict death or LT. Continuous variables were assessed using the Mann–Whitney test for nonparametric populations. Categorical variables were compared using the χ test or Fishers exact test where appropriate. The optimal cut-off was calculated using the method described by Youden, after derivation of area under receiver operated characteristic curves (AUROC) had been constructed. The accuracy of the LAAR score was analysed by calculating its sensitivity, specificity, negative predictive value, positive predictive value, and negative and positive likelihood ratios. Kaplan–Meier survival curves were plotted and the significance of the cut-off value expressed using the log-rank test. The prognostic accuracy of the different models of ESLD was calculated by AUROC curves. Variables derived from the univariate analysis where P < 0.1 were entered into a multivariate analysis to determine factors independently associated with the primary end point. Single variables incorporated into composite scores for prognostication e.g. bilirubin, creatinine and INR, were excluded from the multivariate analysis to prevent redundancy. Calculations were performed with the SPSS package version 15 (SPSS Inc. 1989–2006, Chicago, IL, USA).
Methods
The South West Liver unit (Plymouth, United Kingdom), assesses and performs the post-operative care for patients living in the South West of England who are being considered for, or who have received, a liver transplant. The unit maintains very close links with Kings College Hospital (KCH), London where the surgery is performed. The unit also manages patients with general hepatological problems. An abdominal CT is routinely performed as part of the LT assessment. Patients with cirrhosis who were not part of the transplant assessment or hepatoma programme were included in the study if a CT scan had been performed for clinical reasons. Abdominal CT scanning was not performed routinely in patients with early cirrhosis (Child-Pugh A), although these patients did have 6-monthly liver ultrasound assessment for hepatoma surveillance. As the data were retrieved from existing retrospective sources, ethics committee approval was not required.
Patients
A retrospective–prospective cohort was identified from the departmental liver transplant database (maintained by JDM). The time period patients were identified for the study was May 2003 to May 2012. All patients who had computed tomography (CT) performed as part of the pre-transplant assessment were identified. To prevent cohort bias, patients with cirrhosis in whom a CT had been performed were identified from the in-patient database (maintained by TJSC) and from out-patients. The diagnosis of cirrhosis was made using well-defined parameters including clinical signs, biochemical criteria, radiological criteria and histological criteria as previously described. On the database, the following patient demographics were recorded: gender, age (years), liver disease aetiology, acute or chronic liver disease, the presence of hepatocellular carcinoma (HCC), serum sodium (mmol/L), serum creatinine (μmol/L), serum bilirubin (μmol/L) and international normalised ration (INR). With this information, it was possible to calculate the MELD, UKELD, MELD-Na and iMELD. The Childs-Pugh Score was not calculated due to incomplete data at the time of CT scanning. The overwhelming majority of the patients were Caucasian, so race was not a variant entered into the analysis. Patients with acute liver failure, an existing liver graft, and hepatocellular carcinoma were excluded from the final analysis. The validation set was collected from the cirrhosis and liver transplant database at the Royal Liverpool University Hospital (RLUH) from June 2008 to June 2013 and patients were also identified from the radiology database (CF). The same baseline parameters were collected as the training set. In the training and the validation sets, the reader of the LAAR score was aware of the patient outcome, but was not aware of the results of the other prognostic score results.
The Liver Abdominal Area Ratio (LAAR)
Computed tomography (CT) cross-sectional liver imaging (General Electric –GE Medical, Milwaukee, IL, USA) was reviewed and measurements were made when the maximum liver area was identified from a single 'slice.' For standard liver imaging, slices were taken at every 0.625 mm. On this image, a 'region of interest' (ROI) was conceptualised on the liver. A 'best-fit' ellipsoid was made around the liver to approximate liver area. The area of this ellipsoid was calculated using the formula πab (where π = 3.142, a = radius of ellipse height and b=radius of ellipse width). An approximated abdominal area was calculated using the formula πcd [where π = 3.142, c=radius of abdominal area height (diameter measured from sternum to base of spinal dorsal vertebrae) and d=radius of width (diameter measured from intercostal ribs in cross-sectional horizontal midline of the abdomen)]. All measurements were performed by the same individual (TJC). How the liver and abdominal areas were calculated is shown in Figure 1. If the liver extended round the midline to the left, often two separate liver areas with best-fit ellipses were calculated and added together to represent the liver area for that slice. The liver abdominal area ratio (LAAR) was calculated by dividing the liver area by the abdominal area and then multiplying the result by one hundred:
(Enlarge Image)
Figure 1.
Derivation of the LAAR score. A region of interest (ROI) was identified on a computed tomography slice where the greatest liver area was observed (a). This was placed within a best-fit ellipse. A best-fit ellipse was then measured assessing the abdominal circumference between the ribs in the midline and the anterior posterior diameter was measured from the sternum to the dorsal spinous process. The measurements were then made as shown in (b). The LAAR score is calculated by determining the liver area divided by the abdominal area Ă—100.
A control group was collected to determine a normal range for the LAAR score. Controls were identified from the hospital computer radiology system over a 10-day period in patients in whom a CT abdomen had been performed. Patients were included if the CT report did not identify any features of chronic liver disease or any liver abnormalities on imaging and in whom all patients had completely normal liver biochemistry on blood testing. Patients with cirrhosis were censored at the time of death or at the time of LT, when patients were no longer exposed to the risk of dying due to the complications of cirrhosis.
Statistical Analysis
All values were expressed as median values with interquartile ranges. The primary end point was to determine if the LAAR score could predict death or LT. Continuous variables were assessed using the Mann–Whitney test for nonparametric populations. Categorical variables were compared using the χ test or Fishers exact test where appropriate. The optimal cut-off was calculated using the method described by Youden, after derivation of area under receiver operated characteristic curves (AUROC) had been constructed. The accuracy of the LAAR score was analysed by calculating its sensitivity, specificity, negative predictive value, positive predictive value, and negative and positive likelihood ratios. Kaplan–Meier survival curves were plotted and the significance of the cut-off value expressed using the log-rank test. The prognostic accuracy of the different models of ESLD was calculated by AUROC curves. Variables derived from the univariate analysis where P < 0.1 were entered into a multivariate analysis to determine factors independently associated with the primary end point. Single variables incorporated into composite scores for prognostication e.g. bilirubin, creatinine and INR, were excluded from the multivariate analysis to prevent redundancy. Calculations were performed with the SPSS package version 15 (SPSS Inc. 1989–2006, Chicago, IL, USA).
SHARE
