To assess the diagnostic accuracy of reported physical examination findings for the diagnosis of cirrhosis, and to determine the effects of study characteristics on diagnostic accuracy.

MEDLINE was searched from 1966 to August 2000; the search terms were reported. Additional articles were sought by scanning the bibliographies of standard reference texts, proceedings of meetings or conferences on cirrhosis or liver disease, and the reference sections of selected articles or review articles on liver disease, cirrhosis, or physical examination. The authors of the primary studies were contacted for data not reported in the published article and for additional published or unpublished studies.

Study designs of evaluations included in the review

Diagnostic accuracy studies were eligible for inclusion. Six prospective and five retrospective diagnostic cohort studies and one case-control study were included.

Specific interventions included in the review

Studies that evaluated items obtained by physical examination were eligible for inclusion. Studies that evaluated only biochemical or radiological examinations were excluded. The items evaluated by the included studies were: hepatomegaly, splenomegaly, palmar erythema, ascites, collateral circulation, encephalopathy, firm liver, jaundice, spider angiomata, facial telangectasia, gynaecomastia, irregular liver surface, parotidomegaly, peripheral oedema, sharp liver edges, tender liver, testicular atrophy, thick liver edge and white nails.

Reference standard test against which the new test was compared

The included studies were required to report liver biopsy results. The diagnosis of cirrhosis was defined by histological evidence of irreversible chronic injury to liver parenchyma with extensive fibrosis and formation of regenerative nodules.

Participants included in the review

Studies were eligible for inclusion if they evaluated patients with known or suspected liver disease or cirrhosis. The participants in the included studies were aged from 3 to 90 years.

Outcomes assessed in the review

The included studies were required to report sufficient data for the calculation of the sensitivity and specificity.

How were decisions on the relevance of primary studies made?

The titles of citations were scanned and abstracts of potentially relevant articles were retrieved. If the information in the abstract indicated that the study was likely to satisfy the inclusion criteria, or if no abstract was available, the full text article was evaluated. The authors did not state how many reviewers screened for relevance.

The validity of the included studies was assessed using four criteria: independent, blind comparison with a reference standard; evaluation in an appropriate spectrum of patients; reference standard applied regardless of the result of the diagnostic test; the reference standard was applied before starting any intervention based on the findings of the examination. One author assessed the validity of the included studies.

One author extracted the primary data using data extraction forms. The results, in the form of 2x2 contingency tables, were extracted and used to calculate outcome measures.

How were the studies combined?

Where sensitivity and specificity were not significantly correlated (i.e. no evidence of a threshold effect), and where there was no evidence of statistically significant between-study heterogeneity in either sensitivity or specificity, the summary estimates of sensitivity and specificity were derived using a random-effects model. Where sensitivity and specificity were significantly correlated, a summary receiver operating characteristic (ROC) curve was used to summarise the data. If sensitivity and specificity were heterogeneous, no summary estimate was derived.

How were differences between studies investigated?

Spearman's rank correlation coefficient was used to assess the correlation between sensitivity and specificity, while the chi-squared test was used to assess between-study heterogeneity.

Where a minimum of 3 studies per subgroup were present, subgroup analyses were planned to determine the effects of the following covariates on diagnostic accuracy: sample size; independence of examination and reference standard (measured on a 4-point scale); study design (prospective, retrospective, or experimental); primary aetiology of cirrhosis (alcohol, viral hepatitis, or other).

For summary ROC curves, sensitivity analyses were performed to evaluate the effect of including those studies considered most liable to various types of bias.

Twelve studies, with a total of 1,895 participants, were included.

Eleven of the 12 included studies reported that either clinical examination was performed prior to biopsy and blinded pathologists read the biopsy, or that examination was performed prior to biopsy. In all studies the examiners were not blinded to other clinical information. Biopsy was performed independently of the findings at examination in 9 studies, the control group did not undergo biopsy in the case-control study, and the independence of the decision to perform biopsy was not clear in 2 studies.

Ten physical signs or symptoms were reported in 3 or more studies.

The maximum joint sensitivity and specificity (derived from the summary ROC curve) was 0.64 for splenomegaly and 0.75 for hepatomegaly.

For ascites (7 studies), the sensitivity was 0.34 (95% confidence interval, CI: 0.22, 0.49) and the specificity was 0.95 (95% CI: 0.89, 0.98).

For collateral circulation (3 studies), the sensitivity was 0.42 (95% CI: 0.26, 0.61) and the specificity was 0.94 (95% CI: 0.71, 0.99).

For encephalopathy (3 studies), the sensitivity was 0.15 (95% CI: 0.06, 0.33) and the specificity was 0.98 (95% CI: 0.97, 0.99).

For firm liver (3 studies), the sensitivity was 0.68 (95% CI: 0.55, 0.79) and the specificity was 0.75 (95% CI: 0.62, 0.85).

For jaundice (3 studies), the sensitivity was 0.36 (95% CI: 0.25, 0.48) and the specificity was 0.85 (95% CI: 0.80, 0.89).

For spider angiomata (8 studies), the sensitivity was 0.50 (95% CI: 0.39, 0.61) and the specificity was 0.88 (95% CI: 0.75, 0.95).

The data for palmar erythema were heterogeneous and unsuitable for pooling; the sensitivity ranged from 0.12 to 0.63 and the specificity from 0.49 to 0.98.

The sensitivity and specificity values for a further ten physical findings were reported in single studies. These values were tabulated in the review article.

Exclusion of the case-control study from the analysis of hepatomegaly did not significantly change the summary ROC curve. Exclusion of the study in which clinical and pathological assessments were not made independently from the splenomegaly dataset resulted in a change in summary estimates; the sensitivity was 0.82 (95% CI: 0.52, 0.95) and the specificity was 0.60 (95% CI: 0.21, 0.89).

Subgroup analyses were possible for splenomegaly and ascites. None of the covariates examined resulted in a significant change in diagnostic accuracy.

The majority of studies assessing the diagnostic accuracy of physical examination were undertaken in highly selected populations. Physical signs generally had low sensitivity for the diagnosis of cirrhosis and, therefore, were not useful for excluding disease. Signs with higher specificity, which may be useful to confirm cirrhosis in those with moderate to high pre-test probability of disease, tended to be associated with clinically decompensated disease.

Physical signs need to be used alongside additional criteria in order to be of use in identifying patients with cirrhosis.

This review addressed a clearly stated question, which was further defined by appropriate inclusion criteria. The search strategy was limited to one electronic database and follow-up bibliographic searches; this might have resulted in incomplete retrieval of the available data. Although some measures to identify unpublished studies were reported, publication bias was not formally assessed. The details of the included studies were clearly and comprehensively reported, and an assessment of their methodological quality (using criteria appropriate to diagnostic accuracy studies) was reported. It should be noted, however, that the conduct of the data extraction and quality assessment by a single author might have left the review vulnerable to the introduction of error and/or bias.

The methods used to pool the data from the included studies were clearly described and appropriately applied. The authors discussed the limitations of both the primary studies and the review in detail. The authors' conclusions follow from the data and are presented in an appropriately cautious manner.

Practice: The authors stated that physical signs are not useful for excluding cirrhosis, but may be useful to confirm cirrhosis in those with moderate to high pre-test probability of disease. They further stated that physical signs need to be used with additional criteria to increase the probability of identifying patients with cirrhosis.

Research: The authors stated that future studies of the diagnostic accuracy of clinical findings should address the appropriateness of the study population (studies have primarily been conducted in referral centres, by clinicians experienced in liver disease), and should report clear descriptions of their populations and methods. Consensus on the standardisation of liver biopsy is needed. Future studies will also need to consider how to measure the independent contributions of individual covariates or clinical findings, as well as which combinations produce the best diagnostic accuracy. Attention should be given to identifying and evaluating other ways to detect cirrhosis (e.g. noninvasive laboratory testing, imaging techniques, improved biopsy techniques).

This is a critical abstract of a systematic review that meets the criteria for inclusion on DARE. Each critical abstract contains a brief summary of the review methods, results and conclusions followed by a detailed critical assessment on the reliability of the review and the conclusions drawn.