A Novel Imaging Score for Prognostication in Cirrhosis
A Novel Imaging Score for Prognostication in Cirrhosis
Three hundred and sixteen (316) subjects were identified from the training data set. The median age of the study population was 58 years (IQR 48– 62). The underlying liver disease was available in 298 subjects. After exclusion criteria were applied and subjects identified in whom a CT had not been performed, 158 subjects with survival data were available for analysis, 100 (63%) were male. The underlying liver diseases were as follows: alcoholic liver disease, n= 80 (50%); chronic hepatitis C, n = 20 (12.7%); non-alcoholic steatohepatitis (NASH), n = 11 (7.0%); primary biliary cirrhosis, n = 6 (3.8%); primary sclerosing cholangitis, n = 9 (5.7%); others n = 32 (20.3%) where indications for review included conditions as diverse as auto-immune hepatitis, biliary atresia, chronic hepatitis B, cryptogenic cirrhosis, haemochromatosis and Wilson's disease. Of 158 patients at the end of the study period, 49 (31%) were transplanted and 8 (5%) remained on the LT wait list. The baseline characteristics of the study cohort are shown in Table 1. The median follow-up of the study cohort was 211 days (range 9–3525 days), the mean follow-up was 658 days s.d. ± 513 days.
The control group was derived from 91 consecutive patients. From this set, 30 patients were identified with no liver abnormalities on computed tomography and with normal liver function tests. The reason for the CT abdomen was as follows: abdominal pain n = 16, post-surgical n = 2, renal cell carcinoma n = 2, pelvic mass n = 2, to exclude GI malignancy n = 3, Crohns' disease n = 1, oesophageal cancer n = 1, bowel obstruction n = 2, lymphoma n = 1. The median age of the control group was 57 (48–62) and 15 (50%) were male.
The LAAR score was able to detect death/LT our cohort (P = 0.003). The median LAAR score for survivors was 40.2 (IQR 33.5– 47.2) and the median score for dead/LT was 32.2 (28.4–42.0). In the controls, the median LAAR score was 39.5 (IQR 35.8–43.8). Comparing the LAAR scores, there was no difference between the control group vs. survivors (P = 0.85), but there was a difference between the control group vs. dead/LT (P = 0.009) and the survivors vs. dead/LT (P = 0.004) (Figure 2). The difference between the survivors vs. dead/LT patients is shown in Table 2.
(Enlarge Image)
Figure 2.
Box plot graph of LAAR scores between survivors, dead/LT and controls. Middle line is median value with upper and lower quartiles represented by the upper and lower limit of the box, and lines give the full range of the scores for the cohort.
The respective LAAR scores for each single aetiology were as follows: ALD 35.6 (IQR 30.9–42.3), HCV 39.4 (IQR 25.6–41.7), NASH 35.8 (29.6–44.1), PBC 43.8 (25.6–45.4) and PSC 48.0 (43.1–52.8). The optimal cut-off for death/LT was 32.14. A score of <32.1 was associated with a greater risk of death/LT (P = 0.001, OR 1.51, 95% confidence interval 1.19–1.93). At the optimal cut-off for the overall study, the LAAR score had a specificity of 77%, sensitivity 49%, negative predictive value 59%, positive predictive value (69%) and positive likelihood ratio 2.1 (1.3–3.4), P = 0.001 odds risk 1.52 (95% CI 1.2–2.0). At 720 days, the LAAR score had specificity 72%, sensitivity 62%, negative predictive value 83%, positive predictive value 46%, and positive likelihood ratio 2.3 (1.5–3.4), P < 0.0001, odds risk 1.92 (95% CI 1.2–2.9).
A comparison was made between those patients who died and those patients who received a liver transplant. No statistical difference between these groups was found [P = 0.1 (OR 2.09, 95% CI 0.94–4.62)]. Yet, on comparing survivors against those patients who died without LT, a statistically significant difference between groups was demonstrated [P = 0.02 (OR 2.49, 95% CI 1.2–5.2)]. Excluding PBC and PSC patients from the analysis did not improve the diagnostic accuracy of the LAAR score (data not shown).
The prognostic accuracy of the LAAR score at 90, 360 and 720 days from the date of the CT to the death/LT using the optimal cut-off and the log-rank test was P = 0.28, P = 0.06 (OR 1.35, 95% CI, 0.94–1.94) and P < 0.0001 (OR 1.89, 95% CI 1.25–2.85) respectively. The Kaplan–Meier survival curves with the optimal cut-off are shown in Figures 3 and 4 One hundred patients had ascites (63%), 50/81 survivors (62%) had ascites, whereas 50/77 (65%) of nonsurvivors had ascites. The presence of ascites had no impact on survival, P = 0.27 (OR 1.46, 95% confidence interval 0.75–2.85). But the LAAR score was different between patients without ascites, 42.6 (IQR 37.9–48.2) in comparison with patients with ascites, 32.1 (28.7–40.2). There was a strong correlation between the LAAR score and survival (Spearman's test gave a correlation coefficient of 0.38, P < 0.0001), but the presence or absence of ascites did not predict survival (Kendall tau-b test, correlation coefficient 0.07, P = 0.61). Patient height was only available in 63 of the patients in the training set, but no correlation between the height and the LAAR score was seen (Spearman's test 0.10, P = 0.43).
(Enlarge Image)
Figure 3.
Kaplan–Meier survival curve demonstrating patient survival in patients with a LAAR score above and below the optimal cut-off at 720 days from date of CT scan.
(Enlarge Image)
Figure 4.
Kaplan–Meier survival curve demonstrating patient survival in patients with a LAAR score above and below the optimal cut-off at 360 days from date of CT scan.
The prognostic accuracy of LAAR was then compared with exiting models of end-stage liver disease including MELD, MELD-Na, iMELD and UKELD. The AUROCs for the prediction of death/LT at 90, 360, 720 days and overall are shown in Table 3.
All variables from the univariate analysis that had a P < 0.1 were entered into a multivariate analysis to determine what factors were independently associated with death or LT. At 720 days, the variables independently associated with death were: LAAR P = 0.008; MELD, P = 0.004 and MELD-Na, P = 0.03). Overall, LAAR, and MELD were the only variables independently associated with the primary outcome measure where P = 0.04, and 0.003 respectively; the data are shown in Table 4.
To eliminate the error that some patients transplanted may not have died, a further analysis was carried out looking at patients who died without LT. Overall, the LAAR score was still predictive of death P = 0.004, AUROC 0.71), as were MELD P = 0.004 (AUROC 0.72), MELD-Na P = 0.006 (AUROC 0.71) and UKELD P = 0.02 (0.68). However, on a multivariate analysis, none of the models was shown to be independently associated with death.
Sixty patients were identified from the validation set. The median age of the study population was 50 years (IQR 42–60). After exclusion criteria were applied, 52 subjects with survival data were available for analysis, 36 (69%) were male. The median follow-up was 304 days (range 4–1696 days). The mean follow-up was 446 days s.d. ± 437 days The underlying liver diseases were as follows: alcoholic liver disease, n = 28 (53.8%); chronic hepatitis C, n = 8 (15.4%); non-alcoholic steatohepatitis (NASH), n = 5 (9.6%); primary biliary cirrhosis, n = 4 (7.6%); others, n = 7 (13.5%). Of 52 patients, at the end of the study period, 15 (28.8%) were transplanted and 2 (3.8%) remained on the LT wait list. Four patients died (7.6%) without a liver transplant. The baseline characteristics of the study cohort are shown in Table 1.
The LAAR score was able to detect progression to death or liver transplantation in our cohort (P < 0.0001). The median LAAR score for survivors was 39.0 (33.4–44.7), and the median score for dead/LT was 29.8 (27.2–33.2).
The ability of the LAAR score in the validation set to determine likelihood of death/LT for the overall study period was statistically significant [P < 0.0001, AUROC 0.89 (0.78–0.97)]. This was compared with the other tests and shown to be superior iMELD P = 0.09, AUROC 0.61; UKELD P = 0.02, AUROC 0.66; MELD P = 0.13, AUROC 0.6. At 90, 360 and 720 days from the date of CT the AUROC curves for death/OLT using LAAR were 0.73 (0.56–0.89, P = 0.11); 0.79 (0.65–0.93, P = 0.005); and 0.79 (0.65–0.93, P = 0.005) respectively.
Results
Three hundred and sixteen (316) subjects were identified from the training data set. The median age of the study population was 58 years (IQR 48– 62). The underlying liver disease was available in 298 subjects. After exclusion criteria were applied and subjects identified in whom a CT had not been performed, 158 subjects with survival data were available for analysis, 100 (63%) were male. The underlying liver diseases were as follows: alcoholic liver disease, n= 80 (50%); chronic hepatitis C, n = 20 (12.7%); non-alcoholic steatohepatitis (NASH), n = 11 (7.0%); primary biliary cirrhosis, n = 6 (3.8%); primary sclerosing cholangitis, n = 9 (5.7%); others n = 32 (20.3%) where indications for review included conditions as diverse as auto-immune hepatitis, biliary atresia, chronic hepatitis B, cryptogenic cirrhosis, haemochromatosis and Wilson's disease. Of 158 patients at the end of the study period, 49 (31%) were transplanted and 8 (5%) remained on the LT wait list. The baseline characteristics of the study cohort are shown in Table 1. The median follow-up of the study cohort was 211 days (range 9–3525 days), the mean follow-up was 658 days s.d. ± 513 days.
The control group was derived from 91 consecutive patients. From this set, 30 patients were identified with no liver abnormalities on computed tomography and with normal liver function tests. The reason for the CT abdomen was as follows: abdominal pain n = 16, post-surgical n = 2, renal cell carcinoma n = 2, pelvic mass n = 2, to exclude GI malignancy n = 3, Crohns' disease n = 1, oesophageal cancer n = 1, bowel obstruction n = 2, lymphoma n = 1. The median age of the control group was 57 (48–62) and 15 (50%) were male.
Prediction of Death/Liver Transplant From Time of CT Using LAAR Score
The LAAR score was able to detect death/LT our cohort (P = 0.003). The median LAAR score for survivors was 40.2 (IQR 33.5– 47.2) and the median score for dead/LT was 32.2 (28.4–42.0). In the controls, the median LAAR score was 39.5 (IQR 35.8–43.8). Comparing the LAAR scores, there was no difference between the control group vs. survivors (P = 0.85), but there was a difference between the control group vs. dead/LT (P = 0.009) and the survivors vs. dead/LT (P = 0.004) (Figure 2). The difference between the survivors vs. dead/LT patients is shown in Table 2.
(Enlarge Image)
Figure 2.
Box plot graph of LAAR scores between survivors, dead/LT and controls. Middle line is median value with upper and lower quartiles represented by the upper and lower limit of the box, and lines give the full range of the scores for the cohort.
The respective LAAR scores for each single aetiology were as follows: ALD 35.6 (IQR 30.9–42.3), HCV 39.4 (IQR 25.6–41.7), NASH 35.8 (29.6–44.1), PBC 43.8 (25.6–45.4) and PSC 48.0 (43.1–52.8). The optimal cut-off for death/LT was 32.14. A score of <32.1 was associated with a greater risk of death/LT (P = 0.001, OR 1.51, 95% confidence interval 1.19–1.93). At the optimal cut-off for the overall study, the LAAR score had a specificity of 77%, sensitivity 49%, negative predictive value 59%, positive predictive value (69%) and positive likelihood ratio 2.1 (1.3–3.4), P = 0.001 odds risk 1.52 (95% CI 1.2–2.0). At 720 days, the LAAR score had specificity 72%, sensitivity 62%, negative predictive value 83%, positive predictive value 46%, and positive likelihood ratio 2.3 (1.5–3.4), P < 0.0001, odds risk 1.92 (95% CI 1.2–2.9).
A comparison was made between those patients who died and those patients who received a liver transplant. No statistical difference between these groups was found [P = 0.1 (OR 2.09, 95% CI 0.94–4.62)]. Yet, on comparing survivors against those patients who died without LT, a statistically significant difference between groups was demonstrated [P = 0.02 (OR 2.49, 95% CI 1.2–5.2)]. Excluding PBC and PSC patients from the analysis did not improve the diagnostic accuracy of the LAAR score (data not shown).
The prognostic accuracy of the LAAR score at 90, 360 and 720 days from the date of the CT to the death/LT using the optimal cut-off and the log-rank test was P = 0.28, P = 0.06 (OR 1.35, 95% CI, 0.94–1.94) and P < 0.0001 (OR 1.89, 95% CI 1.25–2.85) respectively. The Kaplan–Meier survival curves with the optimal cut-off are shown in Figures 3 and 4 One hundred patients had ascites (63%), 50/81 survivors (62%) had ascites, whereas 50/77 (65%) of nonsurvivors had ascites. The presence of ascites had no impact on survival, P = 0.27 (OR 1.46, 95% confidence interval 0.75–2.85). But the LAAR score was different between patients without ascites, 42.6 (IQR 37.9–48.2) in comparison with patients with ascites, 32.1 (28.7–40.2). There was a strong correlation between the LAAR score and survival (Spearman's test gave a correlation coefficient of 0.38, P < 0.0001), but the presence or absence of ascites did not predict survival (Kendall tau-b test, correlation coefficient 0.07, P = 0.61). Patient height was only available in 63 of the patients in the training set, but no correlation between the height and the LAAR score was seen (Spearman's test 0.10, P = 0.43).
(Enlarge Image)
Figure 3.
Kaplan–Meier survival curve demonstrating patient survival in patients with a LAAR score above and below the optimal cut-off at 720 days from date of CT scan.
(Enlarge Image)
Figure 4.
Kaplan–Meier survival curve demonstrating patient survival in patients with a LAAR score above and below the optimal cut-off at 360 days from date of CT scan.
Performance of LAAR vs. Existing Models of End-stage Liver Disease
The prognostic accuracy of LAAR was then compared with exiting models of end-stage liver disease including MELD, MELD-Na, iMELD and UKELD. The AUROCs for the prediction of death/LT at 90, 360, 720 days and overall are shown in Table 3.
All variables from the univariate analysis that had a P < 0.1 were entered into a multivariate analysis to determine what factors were independently associated with death or LT. At 720 days, the variables independently associated with death were: LAAR P = 0.008; MELD, P = 0.004 and MELD-Na, P = 0.03). Overall, LAAR, and MELD were the only variables independently associated with the primary outcome measure where P = 0.04, and 0.003 respectively; the data are shown in Table 4.
Prediction of Death From Time of CT Using LAAR Score
To eliminate the error that some patients transplanted may not have died, a further analysis was carried out looking at patients who died without LT. Overall, the LAAR score was still predictive of death P = 0.004, AUROC 0.71), as were MELD P = 0.004 (AUROC 0.72), MELD-Na P = 0.006 (AUROC 0.71) and UKELD P = 0.02 (0.68). However, on a multivariate analysis, none of the models was shown to be independently associated with death.
Validation Set
Sixty patients were identified from the validation set. The median age of the study population was 50 years (IQR 42–60). After exclusion criteria were applied, 52 subjects with survival data were available for analysis, 36 (69%) were male. The median follow-up was 304 days (range 4–1696 days). The mean follow-up was 446 days s.d. ± 437 days The underlying liver diseases were as follows: alcoholic liver disease, n = 28 (53.8%); chronic hepatitis C, n = 8 (15.4%); non-alcoholic steatohepatitis (NASH), n = 5 (9.6%); primary biliary cirrhosis, n = 4 (7.6%); others, n = 7 (13.5%). Of 52 patients, at the end of the study period, 15 (28.8%) were transplanted and 2 (3.8%) remained on the LT wait list. Four patients died (7.6%) without a liver transplant. The baseline characteristics of the study cohort are shown in Table 1.
The LAAR score was able to detect progression to death or liver transplantation in our cohort (P < 0.0001). The median LAAR score for survivors was 39.0 (33.4–44.7), and the median score for dead/LT was 29.8 (27.2–33.2).
The ability of the LAAR score in the validation set to determine likelihood of death/LT for the overall study period was statistically significant [P < 0.0001, AUROC 0.89 (0.78–0.97)]. This was compared with the other tests and shown to be superior iMELD P = 0.09, AUROC 0.61; UKELD P = 0.02, AUROC 0.66; MELD P = 0.13, AUROC 0.6. At 90, 360 and 720 days from the date of CT the AUROC curves for death/OLT using LAAR were 0.73 (0.56–0.89, P = 0.11); 0.79 (0.65–0.93, P = 0.005); and 0.79 (0.65–0.93, P = 0.005) respectively.
Source...