Summary of main findings
Our analysis suggests that, in patients with cirrhosis, surveillance strategies for HCC are effective, and can often be considered cost-effective. The most effective strategy for a mixed aetiology cohort of individuals with cirrhosis is AFP assay combined with US imaging on a 6-monthly basis. However, when costs are taken into account, using AFP as a triage step may be preferable. Surveillance is much more likely to be cost-effective in those with HBV-related cirrhosis, while surveillance of people with ALD-related cirrhosis appears least economically efficient.
Interpretation of findings
According to our simulation, the economic efficiency of different surveillance strategies can be expected to vary substantially according to cirrhosis aetiology. As a result, the most efficient mode of resource allocation, from a purely decision-analytic viewpoint, would be to offer surveillance of differing intensity to each subgroup. However, apart from the practical complications of establishing different recall pathways for different patient groups, concerns might also be raised about the ethical implications of this approach. For instance, one-way sensitivity analysis suggested that the particularly good value offered in the HBV group may be substantially due to the younger age of the cohort. By implication, then, there may be further subgroups of individuals with HCV and ALD, diagnosed with cirrhosis at a younger age, in which more intensive surveillance might be particularly cost-effective.
Our results also suggest that reports of the AFP test's demise (Sherman, 2001
) may be exaggerated, particularly if one adopts a cost-effectiveness perspective. We believe one reason for this is that previous authors may have failed to account fully for tumour size in their analyses. Although the AFP test has been found to be relatively insensitive for detecting HCC in general (Trevisani et al, 2001
), it has the substantial advantage that its sensitivity is only weakly correlated with tumour size, meaning that it should be a valuable aid to the detection of some of the smallest, most easily treated tumours. According to the evidence used in our model, 65% of tumours less than 2
cm in diameter secrete 20
or more of AFP (although this figure may be an overestimate, with the true proportion around 46%; Farinati et al, 2006
). In contrast, our model is configured to simulate US sensitivity of only 10.7% for the smallest category of tumour (Bennett et al, 2002
). Although this is at the pessimistic end of the range of available evidence, it should not be seen as an outlying estimate. Every study that has used an optimal reference standard (explant pathology) to investigate the diagnostic capabilities of US, in this setting, has reported disappointing sensitivity for the tumours
cm in diameter: 13.8% (Kim et al, 2001
), 22.2% (Rode et al, 2001
) and 30.0% (Liu et al, 2003
). Even if one was to adopt the most positive estimate available, all robust evidence suggests that US is less sensitive than AFP assay for the detection of the smallest tumours.
Additionally, AFP is a very cheap test and, inevitably, this is a crucial consideration from a cost-effectiveness perspective. The fact that current guidelines do not recommend using AFP screening may reflect the poor quality of current evidence or a lack of explicit attention to cost-effectiveness considerations during the development of clinical guidelines. However, a strategy led by one single diagnostic modality will always run the risk of serial false-negative findings, and this setting provides a good example (some tumours never secrete AFP and will therefore never be detected using an AFP-led approach and, equally, some tumours will infiltrate diffusely and resist US detection).
Because tumour growth rate had a clear influence on cost-effectiveness in the one-way sensitivity analysis, we investigated this factor further. We performed stratified scenario analyses, simulating three mixtures of slow-growing and fast-growing tumours, instead of applying one average growth rate throughout the model. While the combination of growth rates had an impact on the effectiveness and cost-effectiveness of surveillance (the more fast-growing HCCs, the more cost-effective surveillance becomes), the incremental relationship between surveillance strategies was preserved. The only practical implication of these additional findings is to suggest that, if the true mix of HCCs features a preponderance of slow-growing tumours, it may not be cost-effective to offer any surveillance strategy at 6-monthly intervals.
Strengths of the evaluation
This is the first analysis of the effectiveness and cost-effectiveness of surveillance of cirrhosis for HCC in the UK NHS. Comprehensive literature searches were conducted to inform the model parameters, wherever possible choosing data either derived from the UK population or most likely to be applicable to the UK population.
Extensive exploration of model structures and uncertainty suggest that our model more appropriately captures the disease and surveillance process and impacts than previous studies in this field. While we have assumed that decompensated cirrhosis is irreversible, we showed in one-way sensitivity analysis that our results are not greatly influenced by alterations in the annual rate of progression from compensated to decompensated cirrhosis. From this, it is safe to infer that allowing some regression from decompensated back to compensated cirrhosis would not significantly alter our main results. By accounting for the substantial differences in age-related incidence, natural history and response to treatment that exist according to cirrhosis aetiology, we predict that different approaches to surveillance may be justified according to different causes of cirrhosis.
Limitations of the evaluation
First, there is very little published evidence on which to base many of the parameter estimates for the model, and few data originate within the United Kingdom. This was particularly apparent for defining US performance. Second, as the primary focus of this evaluation was the effectiveness and cost-effectiveness of surveillance, we have used a simplified approach to modelling treatment in which OLT and resection are the only curative options available. We performed limited modelling of nonsurgical/ablative therapies as the evidence at the time of the analysis was inconclusive (Di Bisceglie, 2005
). Recent evidence, predominantly from Asia (Izumi et al, 2007
; Peng et al
, 2007; Zytoon et al, 2007
), suggests that such therapies may improve survival in patients with small tumours. If similarly promising findings could be shown in the United Kingdom, we would expect widespread adoption of such techniques to have beneficial cost-effectiveness implications, since a minimally invasive approach is much cheaper than OLT, and might also be expected to have less negative impact on quality of life. Third, we have assumed that entry to the surveillance programme is confined to those more likely to be considered eligible for the available curative treatment options (OLT or resection) and therefore assumes that high-risk activity (excessive alcohol consumption, intravenous drug use and so on) has ceased, and this may not be realistic. We have also assumed that the three cirrhosis aetiologies are mutually exclusive and acknowledge that many people develop cirrhosis as a result of multiple causes.
Comparison with other studies
There are no other studies published studies from Europe. We identified three comparable studies from our literature searches (Arguedas et al, 2003
; Lin et al, 2004
; Patel et al, 2005
); all were conducted in the United States and simulated HCV-related cirrhosis only. In these studies, 6-monthly surveillance using AFP and US produced utility gains of between 0.23 and 0.49 QALYs, compared with no surveillance, giving ICERs of between $24
500 and $46
600 per QALY gained. Apart from obvious dissimilarities in input values and assumptions – reflecting the different populations and health-care systems simulated – there are substantial differences in the structures of the models. Crucially, ours is the only model to have accounted for tumour size in simulating the sensitivity of surveillance.
Implications for future research
Further research is required in the following areas (i) further modelling studies using alternative modelling methods such as individual patient sampling techniques could be used to account for heterogeneity in the patient population, so that factors such as tumour growth rate, tumour characteristics and the variability in individual patients' serial test results could be assessed. Such methods could also be used to assess the optimal surveillance strategy, the optimal surveillance interval and the effects of surveillance on waiting lists for OLT; (ii) further modelling studies could also investigate innovative surveillance strategies not currently undertaken in clinical practice (e.g., alternating AFP and US investigations at 6- or 12-monthly intervals, or measuring change in AFP levels following serial tests rather than absolute levels with fixed cutoff points); (iii) further empirical and modelling analysis of the impact of age at diagnosis of cirrhosis on the effectiveness and cost-effectiveness of surveillance; (iv) empirical evaluation of newer imaging techniques (e.g., contrast-enhanced US) to detect HCCs; and (v) assessment of the effectiveness and cost-effectiveness of surveillance in other aetiologies (e.g., nonalcoholic fatty liver disease) and using other curative treatment options (e.g., ablative techniques).