| |
Ultrasonographic evaluation of liver surface and transient elastography in clinically doubtful cirrhosis - see attached pdf
|
| |
| |
Download the PDF here
Articles in Press
Journal of Hepatology March 2010
Annalisa Berzigotti123, Juan G. Abraldes13, Puneeta Tandon1, Eva Erice1, Rosa Gilabert23, Juan Carlos Garcia-Pagan13, Jaime Bosch13
Received 30 September 2009; received in revised form 29 December 2009; accepted 30 December 2009. published online 15 March 2010.
Uncorrected Proof
Background & Aims
Both transient elastography (TE) and left lobe liver surface (LLS) ultrasound may non-invasively detect cirrhosis (LC). We aimed to examine the diagnostic value of these methods in patients with a suspicion but not a definite diagnosis of cirrhosis.
Methods
We enrolled 90 patients with clinical suspicion of cirrhosis and a strong co-existing differential diagnosis requiring further invasive evaluation. They underwent hepatic venous pressure gradient (HVPG) measurement±transjugular liver biopsy, LLS and TE. Images of LLS were digitally post-processed to obtain a numerical value (quantitative LLS, qLLS). TE<12kPa was considered to exclude LC, 18kPa diagnosed LC, and 12-18kPa indeterminate. Technical failures were considered 'indeterminate'. Diagnosis of cirrhosis was confirmed by histology (84%) or by clinical data and HVPG10mm Hg. Diagnostic accuracy was evaluated by positive and negative likelihood ratios (+LR and -LR).
Results
Cirrhosis was diagnosed in 44 patients. There were 14 technical failures with TE and 1 with LLS (p=0.001). TE and LLS had similar diagnostic accuracy but gave complementary information: TE was mildly more accurate than LLS to rule out LC (-LR: 0.08 vs. 0.10), while it was less accurate to rule it in (+LR 5.05 vs. 11.15). Their combination offered the best diagnostic performance (+LR 9.15; -LR 0.06).
Conclusions
LLS is more technically applicable than TE. In patients with clinical suspicion of cirrhosis, LLS is the best non-invasive method to diagnose cirrhosis, while TE is preferable to rule it out. The combination of both holds the best diagnostic accuracy.
Introduction
Liver cirrhosis is defined by anatomical changes within the liver parenchyma, including fibrosis and the development of regenerating nodules [1], [2]. The appearance of cirrhosis is a hallmark in the natural history of chronic liver diseases, such as alcoholic HCV or HBV infection, because it identifies the point to initiate surveillance for hepatocellular carcinoma [3] and oesophageal varices [4]. Additionally, in alcoholic patients it is important to differentiate cirrhosis from alcoholic hepatitis, due to the different management and prognosis of these conditions [5], [6].
Although liver biopsy is considered the gold standard for the diagnosis of cirrhosis, it has important limitations, such as being invasive, having the potential for complications and false-negative results or underestimating disease severity due to sampling error [7], [8], [9].
In recent years, many studies have evaluated the possibility of diagnosing cirrhosis by non-invasive methods. Transient elastography (TE) estimates liver stiffness, which is thought to be mainly determined by fibrosis, and is proven to be accurate for the non-invasive diagnosis of severe fibrosis and cirrhosis [10]. However, technical limitations preclude the use of TE in obese patients and in patients with ascites, and liver stiffness is increased by hepatic necroinflammation, cholestasis and increased central venous pressure. Moreover, the cut-off of liver stiffness for the identification of cirrhosis appears to be different in different aetiologies of liver disease and has not yet been evaluated in all comers with clinical suspicion of cirrhosis.
Ultrasound is usually the first imaging method used in patients with suspected cirrhosis, as it is non-invasive and widely available. Among ultrasonographic signs of cirrhosis, liver surface nodularity, evaluated by high-resolution ultrasound at the left liver lobe (LLS), is the most accurate [11], [12], [13], [14]. As the main limitation of the technique is its operator dependency [11], it is possible that an objective method to measure LLS may increase its accuracy and applicability.
The aims of this study were (1) to investigate the performance of LLS in the diagnosis of cirrhosis in patients with clinical suspicion of cirrhosis but also a differential diagnosis for the presenting abnormalities, who were chosen over those with an established diagnosis of cirrhosis to mirror the clinical situation in which further invasive investigation would be pursued; (2) to assess whether a digital analysis of images of LLS permits a valid, objective and quantitative LLS measurement; (3) to test the accuracy of TE for the diagnosis of cirrhosis in patients with clinical suspicion of this condition; and (4) to test the hypothesis that the combination of TE and LLS is more accurate than one single method for the non-invasive diagnosis of cirrhosis.
Discussion
The main result of the present study is that a simple non-invasive technique based on the ultrasonographic examination of the LLS had very good diagnostic accuracy to detect cirrhosis, similar to that of TE measurements by Fibroscan, a technique that has received much attention in recent years. In order to minimise the observer-dependency of LLS assessment, we applied a novel method based on the computerised post-processing of the ultrasound images, which permits quantification of LLS length in a standardised segment (qLLS); this method had high intra- and interobserver reproducibility in the present study, and it let us compare the accuracy of TE and LLS, which proved to be analogous for the diagnosis of cirrhosis in an intention-to-treat analysis. However, the concordance between the two non-invasive techniques was only moderate. TE and LLS are applied to different lobes, and differences in the degree of fibrosis might explain part of the variability. Still, the lack of concordance most likely reflects the fact that TE and LLS evaluate different unique characteristics of cirrhosis: TE senses increased liver stiffness, mainly due to fibrosis, while LLS visualises the nodules. Because of this, it is not surprising that their combination is superior to either one alone [25]. In a sense, the combination of these two techniques integrates, in a quantitative and objective way, the information that skilled physicians have long recognised, that palpating a firm, rough liver edge strongly indicates cirrhosis, as do the visualisation of a nodular liver surface at surgery, peritoneoscopy, or CT/MR scans.
Of interest, the finding of nodular LLS was superior to high values of TE for ruling in cirrhosis, while low values of TE were slightly superior to smooth LLS to rule out this diagnosis. More important, the combination of the two methods markedly reduced the number of indeterminate cases, from 30% (TE) and 35% (LLS) to 18%.
This is relevant in the clinical diagnosis of individual patients in whom TE values are in the 'grey zone' or not feasible. In this regard, it is worth noting that in this series TE had a 15% rate of technical failure. This is higher than reported in previous studies [26] and may be partly due to the inclusion of patients with ascites in our population. LLS evaluation, on the contrary, could be obtained in all patients but one, including obese and ascitic patients in whom TE was not obtainable, suggesting that its applicability is very high and that LLS assessment may be helpful for the classification of patients in whom TE cannot be performed. Although it can be argued that the presence of ascites makes TE unnecessary because the diagnosis of cirrhosis may be obvious, this was not the case in the present series, which intentionally included patients with a suspicion of cirrhosis but in whom other diagnoses were also likely. Actually, ascites was not due to cirrhosis in one-third of our patients with ascites. It should be noted that the accuracy of TE improved only slightly, even including in the analysis only patients in whom TE was measurable.
In this study, the semi-quantitative assessment of LLS was superior to qLLS in diagnosing and excluding cirrhosis but gave a substantial number of indeterminate cases. Because of this limitation, it is important to note that qLLS, which provides a quantitative, operator-independent evaluation of LLS, ameliorated the performance of LLS in indeterminate cases.
As per STARD guidelines [27], we decided to apply previously published cut-offs for the diagnosis of cirrhosis by TE; we first used the cut-offs derived from a study conducted in patients with HCV-related chronic liver disease [17] because they offered the best methodological approach. Cut-offs for TE have been shown to vary according to the cause of the underlying liver disease [28], [29], so it has been suggested that it may be better to choose TE cut-offs according to the aetiology [30]. However, in actuality the cause of chronic liver disease is frequently multifactorial or unknown at the time of initial examinations, and specific cut-offs are not available for all aetiologies. Therefore, we also applied a single cut-off of 13kPa, which has been suggested as an 'optimal' cut-off in a recent meta-analysis [18]. TE performed poorly in our series using this cut-off.
On the other hand, it may be argued that, instead of assessing only LLS, as we chose, if we had performed a complete abdominal ultrasonographic study, this would have given more detailed information. Still, we wanted to identify and test a simple, user-friendly, rapid and operator-independent sign, which could be applicable at bedside similarly to TE by a non-specifically expert operator, and LLS and qLLS evaluation fulfilled all these requirements. In this regard it should be noted that even the subjective appreciation of smooth, irregular and nodular appearance of LLS exhibited excellent agreement between observers, and the portable ultrasonic equipment used is low-cost and readily available in many centres, as it is analogous to that commonly used at bedside for the US guidance of central venous access and other invasive procedures.
The importance of conducting clinical studies in patients who are typical of day-to-day clinical care has been recently emphasised [31]. To assess the actual usefulness of the two non-invasive methods in clinical practice, we included only patients with clinical suspicion of cirrhosis raised in highly specialised hospitals but with a strong differential diagnosis challenging the clinical assessment. As shown by our results, clinical uncertainty was justified, as half of the included patients did not have cirrhosis after full work-up, including TJLB and HVPG evaluation. Nonetheless, due to the peculiar clinical setting of our study, other studies are needed in order to confirm our results in other populations at risk of cirrhosis, such as asymptomatic patients with chronic liver diseases.
In conclusion, our data suggest that the diagnosis of cirrhosis can be achieved through non-invasive methods with high accuracy, even in difficult clinical circumstances. The combination of transient elastography and high-resolution ultrasonographic evaluation of the liver surface increases the ability of the two separate methods to diagnose and rule out cirrhosis.
Results
Diagnosis of cirrhosis by gold-standard techniques
Overall, cirrhosis was diagnosed in 44 patients (32 by means of histology) and ruled out in 46.
Liver biopsy was performed in 76/90 (84%) patients; in one case it was not diagnostic due to an insufficient specimen. In 32 of these patients the final diagnosis was cirrhosis, while the 43 non-cirrhotic patients were diagnosed with normal or near-normal liver (n=9), HCV-related chronic hepatitis with F2-F3 fibrosis (n=4), idiopathic portal hypertension (n=4), unspecific chronic hepatitis with no or minimal fibrosis (n=4), regenerative nodular hyperplasia (n=4), simple steatosis (n=3), alcoholic hepatitis with sinusoidal fibrosis (n=1), non alcoholic steatohepatitis (n=2), acute toxic hepatitis (n=2), acute alcoholic hepatitis (n=2), T cell lymphoma (n=2), cardiac congestive liver (n=3), alpha-1 antitrypsin chronic liver disease (n=1), infiltration from solid neoplasia (n=1) and plasma cell infiltration (n=1).
In patients without or with insufficient liver samples, cirrhosis was finally diagnosed on the basis of the presence of clinically significant portal hypertension (CSPH), physical/laboratory data and imaging techniques.
HVPG measurement was successful in 89/90 patients; in one patient the presence of veno-venous communications prevented HVPG measurement. This patient was diagnosed with idiopathic portal hypertension by TJLB and compatible clinical data.
LLS and qLLS
LLS could be studied in 89/90 patients; in one case obesity prevented LLS examination (Fig. 2). Agreement in the assessment between two observers, an expert ultrasonographist (AB) and a hepatologist with minimal teaching but without specific training in ultrasound (JB), was tested in 15 randomly chosen patients and was excellent (k=0.9).
On semi-quantitative evaluation, LLS was scored as smooth in 23 patients, irregular in 31 and nodular in 35. Control subjects showed smooth LLS in all cases.
On an intention-to-treat analysis, LLS diagnosed cirrhosis in 35 patients (39%) and ruled out cirrhosis in 23 (26%). In the remaining 32 patients (35%), LLS was indeterminate. Therefore, the finding of nodular LLS had a+LR of 11.15 (90% CI 4.38-28.36) for ruling in cirrhosis, while the finding of smooth LLS had a -LR of 0.10 (90% CI 0.03-0.32) for ruling out cirrhosis (Table 3).
On post-processing of images, qLLS was significantly different in the three categories (ANOVA p<0.0001): 2.00±0.01 in smooth liver surface, 2.04±0.04 in irregular liver surface (p=0.001 vs. smooth) and 2.09±0.04 in nodular surface (p<0.0001 vs. irregular and vs. smooth). Similarly, qLLS was 2.00±0.01cm (range 1.99-2.01cm; median 2.00cm) in healthy subjects and 2.09±0.03cm (range 2.02-2.22cm; median 2.09cm) in patients with proven cirrhosis (p<0.0001). In the study cohort (i.e., excluding controls) qLLS showed an AUROC of 0.88 (90% CI 0.81-0.96) (p<0.0001) for the diagnosis of cirrhosis. The best cut-off for identification of cirrhosis was 2.04, which had a sensitivity of 89% (90% CI 0.78-0.94), specificity of 82% (90% CI 0.71-0.90), +LR of 6.01 (90% CI 3.01-9.67) and -LR of 0.14 (90% CI 0.07-0.28).
The analysis of the subgroup of patients with irregular liver surface, who were indeterminate at semi-quantitative evaluation, showed that a qLLS<2.04 decreased the probability of having cirrhosis from 35% to 18%, and the observation of a value >2.04 increased this probability from 35% to 57%.
Intraobserver and interobserver reproducibility of qLLS were assessed in 10 patients randomly chosen from the studied population by two independent observers, both blinded to the final diagnosis results. Both performed three measurements on the images of the 10 subjects. Intraobserver reproducibility was near optimal: observer 1 showed an intraclass correlation coefficient (ICC) of 0.969 (95% CI 0.913-0.991) for absolute concordance. Observer 2 had an ICC of 0.954 (95% CI 0.875-0.987) for absolute concordance. The method was also highly reproducible between the two observers, as confirmed by an ICC of 0.915 (95% CI 0.682-0.979) for absolute concordance.
Transient elastography (Fig. 2)
TE could be correctly performed in 76/90 patients. In 13, the measurement was not feasible due to obesity (n=5), ascites (n=5), lack of cooperation by the patient (n=2) and small right hepatic lobe (n=1). In one case the measurement was performed but not valid due to a low success rate (44%). The rate of technical failures was significantly greater for TE as compared to LLS (15.5% vs. 1.1%, p=0.001). In patients in whom TE could be measured, it showed a near-optimal accuracy for the diagnosis of cirrhosis (AUROC 0.91; 90% CI 0.84-0.97; p<0.0001).
According to the previously mentioned cut-offs (<12kPa ruled out cirrhosis, 12-18kPa was indeterminate, >18kPa diagnosed cirrhosis), in an intention-to-treat analysis, TE diagnosed cirrhosis in 35 patients (39%) and ruled out cirrhosis in 28 (31%) (Table 3). In the remaining patients (30%) TE was indeterminate. According to these data, +LR was 5.05 (2.63-9.71) and -LR 0.08 (0.03-0.26).
Even including in the analysis only patients in whom TE was measurable (per-protocol analysis), the performance of TE improved only slightly. A value over 18kPa had a+LR of 5.97 (3.17-11.26), while a TE<12kPa had a -LR of 0.10 (0.03-0.30), and 17% of patients in this analysis were indeterminate.
By using the 13kPa cut-off, TE had a poor accuracy for diagnosing cirrhosis and a moderate accuracy for ruling out cirrhosis both in the intention-to-treat analysis [sensitivity 68% (90% CI 56-78%), specificity 59% (90% CI 47-70%),+LR 2.09 (90% CI 1.42-3.08) and -LR 0.15 (90% CI 0.07-0.35)] and in the per-protocol analysis [sensitivity 88% (90% CI 76-95%), specificity 64% (90% CI 52-75%), +LR 2.47 (90% CI 1.73-3.53) and -LR 0.18 (90% CI 0.08-0.41)].
Comparison of LLS and TE and their combination for the diagnosis of cirrhosis
TE correlated with qLLS (R=0.62, p<0.0001). The concordance between TE and semi-quantitative and quantitative evaluation of LLS for the diagnosis of cirrhosis was slight (k=0.27, 95% CI 0.05-0.46, p=0.015), while the concordance between the two methods for ruling out cirrhosis was moderate (k=0.40, 95% CI 0.18-0.59, p<0.0001). As shown in Table 3, LLS had the highest accuracy for ruling in cirrhosis (+LR: 11.15; 90% CI 4.38-28.36), while TE was the best non-invasive technique to rule out cirrhosis (-LR 0.08; 90% CI 0.03-0.26). The combination of the two methods markedly reduced the number of indeterminate patients (TE 30%, LLS 35%, combination 18%; Table 3) and maintained a good accuracy.
Results in patients diagnosed by liver biopsy
The performance of the two non-invasive methods in the 75 patients with histological diagnosis (cirrhosis n=32) was similar to that observed in the whole study population (Fig. 3).
TE was technically feasible in 63/75 patients (84%). In these patients TE had an AUROC of 0.887 (90% CI 0.807-0.967; p<0.0001) for the diagnosis of cirrhosis. In an intention-to-treat analysis, TE over 18kPa had a+LR of 4.93 (90% CI 2.56-9.49), while TE<12kPa had a -LR of 0.08 (90% CI 0.02-0.41); 33% of patients were indeterminate.
In a per-protocol analysis, TE performed similarly; TE>18kPa had a+LR of 5.22 (90% CI 2.77-9.82), while TE<12kPa had a -LR of 0.07 (90% CI 0.01-0.35) for the diagnosis of cirrhosis.
By using the 13kPa cut-off, TE had the following performance. Intention-to-treat analysis: sensitivity 72% (90% CI 57-83%), specificity 51% (90% CI 39-63%), +LR 2.06 (90% CI 1.90-3.04) and -LR 0.18 (90% CI 0.07-0.47); per-protocol analysis: sensitivity 88% (90% CI 74-95%), specificity 58% (90% CI 45-71%), +LR 2.12 (90% CI 1.5-3.0) and -LR 0.20 (90% CI 0.08-0.50).
LLS could be evaluated in all cases. The finding of nodular LLS had a+LR of 8.06 (90% CI 3.61-18.01) for ruling in cirrhosis, while the finding of smooth LLS had a -LR of 0.06 (90% CI 0.01-0.33) for ruling out cirrhosis, and 34% of patients were indeterminate. qLLS was measured in all patients and had an AUROC of 0.851 (90% CI 0.758-0.945; p<0.0001) for the diagnosis of cirrhosis.
The combination of LLS and TE in these patients maintained good accuracy: +LR 8.74 (90% CI 3.96-19.31), -LR 0.09 (90% CI 0.03-0.29); after combination of the two non-invasive techniques, 15% of patients were indeterminate.
|
|
| |
| |
|
|
|
