Statistics from Altmetric.com
Roger Chapman (Gut 2002;51:9–10) commenting on the recent important study from Nottingham1 concluded that there is a strong case for liver biopsy in most asymptomatic patients with persistently abnormal liver tests, even when diagnostic serology is negative. This conclusion is reasonable, particularly if diagnostic accuracy is paramount. However, accuracy is not the only consideration and other equally valid conclusions can be made from different viewpoints.
Unfortunately, liver biopsy is often painful,2 requires bed rest for at least six hours,3 and is associated with a small but definite mortality.4 We need to appraise our patients of these factors and the likely benefits so that they can make an informed choice. Standard methods of evidence based medicine can greatly assist us in doing this.
As the predominant finding on biopsy is non-alcoholic fatty liver disease (NAFLD), the first question is: can any other test reliably predict fatty liver in this situation? There are three imaging techniques which can detect fatty liver—ultrasound, computerised tomography, and magnetic resonance imaging. Ultrasound is the most patient friendly, cheapest, safest, and most readily available. Furthermore, it is the only imaging technique for which we have sensitivities and specificities for fatty liver.5–8 The most recent study8 gives a sensitivity of 89% and a specificity of 93%.
However, to obtain the predictive value of a positive or negative test one needs to know not just sensitivity and specificity but the prevalence (the pretest probability) of the condition being tested for in the population being studied. The Nottingham study provides precisely that and we now know that in England the prevalence of fatty liver in “well” patients with abnormal liver tests and negative serology is 66%. The easiest way of obtaining the post test probability of a positive or negative ultrasound scan (the positive and negative predictive values, respectively) is to calculate the likelihood ratios and apply them to the nomogram devised by Fagan.9
The likelihood ratio for a positive test (LR+) is sensitivity/100–specificity which, using the latest data,8 is 12. The likelihood ratio for a negative test (LR−) is 100–sensitivity/specificity, which is 0.12. From the nomogram9 it can be shown that a positive scan for fatty liver has a positive predictive value of 96%. Many would consider this degree of certainty sufficient to diagnose fatty liver and not biopsy. If the scan is negative it can be shown that there is still a 20% probability of fatty liver and one would therefore be more likely to favour biopsy. However, the patient might well want to know what the biopsy might reveal. We can readily provide them with this information by recalculating the percentage of patients having the various liver conditions when the number expected to have a positive scan is subtracted.
Column 2 of table 1 shows the prevalence of the various liver conditions found in the Nottingham study. Column 3 shows the likely prevalence of the various conditions in those with a negative scan. This clearly informs the patient’s and our decision making. Initially, we can consider how important it is to detect these various conditions and how they might be managed.
To take account of the remaining 19% patients with fatty liver (that is, those not detected by ultrasound and shown in column 3 of table 1) one might simply rely on diet and exercise for all those with a raised body mass index, and good control in diabetics. Currently, what else can be done for such patients outside clinical trials? Bearing in mind the likelihood of unsuspected drug damage and alcohol excess, taking a more careful history may be appropriate. One wonders whether knowledge of cryptogenic hepatitis, granuloma, sarcoid, amyloid, and glycogen storage disease would significantly change management. One might be happy to miss the diagnosis of primary biliary cirrhosis and primary sclerosing cholangitis until jaundice or other symptoms supervenes. Perhaps the only two conditions it would be important not to miss are haemochromatosis and autoimmune hepatitis. Therefore, it would be possible to appraise patients with a normal scan that there is a 6.6% chance of missing a condition which may benefit from treatment (prednisolone or venesection). That is, 15 patients would need to be subjected to biopsy to detect one requiring important treatment. Such patients could be said to have informed choice.
Imaging has no place in staging NAFLD.10 Currently, there is no established treatment for NAFLD apart from weight reduction and good diabetic control, and there is good reason for recommending this in all such patients. However, for patients entering clinical trials, staging biopsy is likely to be necessary.
In conclusion, we do not believe that most “well” patients with abnormal liver tests and normal serology need biopsy, but we are now able to give patients an informed choice by applying simple evidence based medicine to the findings of the Nottingham study.