If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Corresponding author. Address: Department of Hepatology and Gastroenterology, Aarhus University Hospital, Palle Juul-Jensens Boulevard 99, DK-8200, Aarhus N, Denmark. Tel.: +45 2425 2944.
Department of Hepatology and Gastroenterology, Aarhus University Hospital, Aarhus, DenmarkDepartment of Clinical Epidemiology, Aarhus University Hospital, Aarhus, DenmarkDivision of Epidemiology and Public Health, School of Medicine, University of Nottingham, Nottingham, United Kingdom
Division of Epidemiology and Public Health, School of Medicine, University of Nottingham, Nottingham, United KingdomNIHR Nottingham Biomedical Research Centre (BRC), Nottingham University Hospitals NHS Trust and the University of Nottingham, Nottingham, United Kingdom
Current guidelines from EASL recommend that most patients with cirrhosis are offered surveillance for hepatocellular carcinoma (HCC), but fewer patients than expected actually receive it. The recommendation is based on observational studies and simulations, not randomised trials. In this opinion piece we argue that a randomised trial of HCC surveillance vs. no surveillance is necessary and feasible, and we believe that clinician and patient participation in HCC surveillance would be better if it were based on trial results demonstrating its value.
We read with interest the expert opinion on surveillance for hepatocellular carcinoma by Jepsen and West.1 The authors propose that a randomized control trial of HCC surveillance vs. no surveillance is necessary and feasible; they postulate that clinician and patient involvement would be improved if trial results proved its benefit.
For decades, clinicians faced by a patient with advanced hepatocellular carcinoma (HCC) have felt helpless. The cancer had developed ‘silently’, causing no symptoms until it had spread widely or overwhelmed liver function and was no longer amenable to potentially curative therapy. Despite the recent ‘revolution’ in systemic therapies, leading investigators opined in 2019 that ‘although [new systemic agents] have the potential to improve outcomes, a survival increase of 2–5 months remains poor’.1 To clinicians, the obvious answer was to detect the cancer at a treatable stage by undertaking surveillance of patients perceived to be at highest risk, namely those with chronic liver disease at the stage of cirrhosis.
I (PJ) clearly remember my meeting with John, a 57-year-old man, in the outpatient clinic. He had been diagnosed with alcohol-related cirrhosis 5 years earlier and was doing very poorly back then, but he had managed to quit alcohol and live a stable life, and his ascites was now a thing of the past. I would have discharged him from our outpatient clinic, had he not said that he was losing weight despite maintaining his usual diet. His wife confirmed this, and so we agreed to do an ultrasound examination to “make sure that nothing was amiss”. Of course, he had a hepatocellular carcinoma (HCC). It turned out to be multifocal and not amenable to any treatment with curative intent. I talked to the couple again, and they were understandably frustrated that we had been following him for years without noticing this cancer that was now going to kill him. Oh yes, I remember this conversation. And I am convinced that many of you reading this remember similar conversations from your practice. I am equally convinced that you do not remember quite so well the many patients you have seen whose ultrasound examination did not show an HCC, just as you forget the times when the ultrasound showed something that required additional CT scans before the patient was cleared of cancer suspicion. Studies from behavioural psychology have shown that we overestimate the occurrence of events that come easily to mind, such as an HCC that we might have diagnosed earlier.
This pattern may explain why many hepatologists are in favour of HCC surveillance, and equally why hepatologists may be poorly positioned to judge the value of HCC surveillance.
Current guidelines from EASL recommend that patients with cirrhosis are offered surveillance for HCC, with some exceptions.
It is well recognised that this recommendation rests on observational studies and simulations, not randomised trials. Fewer than expected patients with cirrhosis actually receive surveillance.
One solution has been to develop models to predict the individual patient’s risk of HCC, expecting that low-risk patients can be exempted from surveillance. We believe that there is a need for a randomised trial of HCC surveillance for the reasons we expound in this opinion piece.
The purpose of HCC surveillance
HCC surveillance aims to prolong survival by reducing deaths from HCC whilst simultaneously improving (or at least not decreasing) quality of life; it is hoped that surveillance allows an earlier HCC diagnosis and curative treatment. Yet, a United States study found that 27.5% of cirrhosis patients experienced harms during 3 years of HCC survival.
While none of them died from those harms, they may have suffered a decrement in their quality of life.
Leaving aside quality of life and focusing on survival, HCC surveillance will provide the greatest benefit to those whose rate of HCC development (rate 1→2 in Fig. 1) is high compared with their rate of death without HCC (rate 1→4), such as a 50-year-old man with compensated cirrhosis due to hepatitis C who is otherwise healthy. By extension, HCC surveillance provides the least benefit to those whose rate of death without HCC is much higher than their rate of HCC development, or whose rate of death from other causes after HCC diagnosis (rate 2→4) far outweighs their rate of death from HCC (rate 2→3). This view is recognised by the EASL guidelines which state that patients’ life expectancy and ability to tolerate potentially curative treatment should be taken into consideration when offering surveillance.
There are 4 states and 4 transitions between them indicated by arrows. The intensity with which a transition occurs is expressed as a rate, and the probability of making a specific transition is expressed as a risk. A transition from state 2 to state 1 is possible with curative treatment (or spontaneous regression) of HCC, but it is not necessary for the discussion and therefore not in the figure. HCC, hepatocellular carcinoma.
Problems with current evidence for HCC surveillance
We rely on observational studies
A well-conducted randomised trial provides an unbiased and unconfounded estimate of the average effect that we can expect from offering HCC surveillance to our patients. As a rule, regulatory bodies insist that clinical interventions are shown to be net beneficial in at least 1 well-conducted randomised clinical trial before they are adopted into clinical practice. HCC surveillance comes with benefits, harms, and costs – just like any other intervention. We are wrong to think that HCC surveillance is the default, and that the burden of proof is on those who question the benefits of surveillance.
Fortunately, randomised trials typically reach the same conclusions as the observational studies that have gone before them, but there have been notable examples where a medical practice established without a randomised trial was stopped or even reversed when a randomised trial was finally conducted.
but we find the statement problematic for 2 reasons. First, the 1.5% per year statement has been repeated so many times that its perceived validity is much higher than its true validity; studies that find an HCC risk above 1.5% per year often claim that HCC surveillance is therefore beneficial in their population.
Second, we have not been able to determine from the 1996 simulation study or its references whether simulations assumed an HCC risk of 1.5% per year or an HCC rate of 1.5 per 100 person-years. The latter is often presented as a “rate of 1.5%”, and because many authors use “rate” and “risk” interchangeably,
the true meaning is often lost. It is of crucial importance whether researchers mean “rate” (events per person-years of observation) or “risk” (probability of experiencing event), as the risk of HCC depends on the rate of HCC development and the rate of death without HCC (Fig. 1).
Table 1 uses simulations to show how the risk of HCC varies in response to changes in the rate of death without HCC when the HCC rate is held constant at 1.5 per 100 person-years.
The rate of death without HCC among patients with cirrhosis was between 10 and 18 per 100 person-years in studies from the United States, England, and Sweden.
We are wrong to think that an HCC rate of 1.5 per 100 person-years therefore means that the risk of HCC is 1.5% after 1 year, 5∗1.5% = 7.5% after 5 years, etc. We must incorporate the rate of death without HCC into our computation of HCC risk, particularly when the rate of death without HCC is high (Table 1). The solution is simple: be accurate when it comes to rate or risk; do not express a rate as a percentage; and use the cumulative incidence function to compute HCC risk, not the Kaplan-Meier function.
Table 1Results of simulation studies to demonstrate how the incidence rate of death without HCC affects the risk of HCC, assuming a constant incidence rate of HCC.
If we use the Kaplan-Meier method instead of the cumulative incidence function to compute the cumulative risk of HCC, we are by definition assuming that the incidence rate of death without HCC is zero. HCC, hepatocellular carcinoma.
1.5%
7.3%
14.0%
1.5
1.5
1.5%
6.9%
13.0%
1.5
6
1.5%
6.3%
10.5%
1.5
15
1.4%
5.0%
7.2%
1.5
24
1.3%
4.2%
5.4%
Incidence rates are per 100 person-years. Note that the risk of HCC goes down as the incidence rate of death without HCC goes up. If we mistakenly assume that an incidence rate of 1.5 per 100 person-years equals a risk of 1.5% per year, so that the 5-year risk is 5∗1.5% = 7.5% and the 10-year risk is 15%, we will overestimate the true risk substantially when the incidence rate of death without HCC is high.
† If we use the Kaplan-Meier method instead of the cumulative incidence function to compute the cumulative risk of HCC, we are by definition assuming that the incidence rate of death without HCC is zero. HCC, hepatocellular carcinoma.
We do not know how HCC surveillance affects quality of life
It is possible that HCC surveillance reduces patients’ quality of life. Maybe surveillance burdens patients with frequent reminders of a looming cancer, the need to travel to the hospital, false-positive screening tests, or futile screening examinations offered to those who would not benefit from an early HCC diagnosis. Alternatively, maybe they feel reassured by the screening examinations, and maybe the biannual hospital visits are a welcome opportunity to get help with non-HCC health issues. At present we do not know how HCC surveillance affects quality of life, and the decision to recommend HCC surveillance should be based on the effects on survival and on quality of life.
A randomised trial could tell us both and indeed is being attempted in another situation that is similar to HCC surveillance, namely Barrett’s oesophagus.
Feasibility of conducting a randomized control trial for liver cancer screening: is a randomized controlled trial for liver cancer screening feasible or still needed?.
The participants in this study were given verbal information and a written decision aid to guide their decision to participate in a trial of HCC surveillance vs. no surveillance. As many as 204 (99.5%) said they would not participate, the vast majority because they preferred surveillance. We have 2 concerns. First, the decision aid was biased in favour of HCC surveillance. There is no mention of harm except that a figure on the second-last page mentions the possibility of “inconvenience of clinical visits, ultrasounds and other tests every 6 months” and of “anxiety from false positive result” (without explaining what a “false positive result” is). The figure is preceded by pages stating, in our view, that a survival benefit from HCC surveillance is plausible although not guaranteed.
Feasibility of conducting a randomized control trial for liver cancer screening: is a randomized controlled trial for liver cancer screening feasible or still needed?.
but they cannot answer the question of whether the resulting high-risk group (or the low-risk group) benefits from HCC surveillance. Of course, the studies have value nonetheless, and they are indispensable in the planning stages of a randomised trial.
Prediction models have identified several correlates of cirrhosis severity as risk factors for HCC, along with male gender and increasing age. However, most of the studies used Cox regression to identify risk factors, so they have in fact identified predictors of the rate of HCC development,
What matters for decisions about HCC surveillance is the risk of HCC development. Again, this risk depends on the rate of HCC development and the rate of death without HCC, and prediction studies must consider both rates in their analysis.
The ideal patients for HCC surveillance are those who possess a characteristic with a strong positive effect on the rate of HCC development and a negative or small effect on the rate of death without HCC. Male gender is such a characteristic.
Table 2 shows results of simulation studies illustrating the effects of hypothetical risk factors on the rate of HCC development and on the rate of death without HCC.
others resemble increasing age or decompensation with their increasing effects on both the rate of HCC and the rate of death without HCC. Clearly, the effect of an HCC risk factor on the risk of HCC depends on the factor’s effects on both possible outcomes (HCC and death without HCC). One suggestion to facilitate the interpretation of prediction models is to use Fine & Gray regression instead of Cox regression.
Table 2Results of simulation studies to illustrate the effects of hypothetical risk factors that affect the hazard ratio of both competing events on the 5- and 10-year relative risks of HCC.
Hypothetical risk factor
Hazard ratio of HCC development for patients with vs. without risk factor
Hazard ratio of death without HCC for patients with vs. without risk factor
5-year relative risk of HCC for patients with vs. without risk factor
10-year relative risk of HCC for patients with vs. without risk factor
#1
3
1
2.79
2.72
#2
3
2
2.30
1.91
#3
3
3
1.88
1.42
#4
2
1
1.90
1.87
#5
2
2
1.51
1.28
#6
2
3
1.27
0.96
Risk factor #1 could be male gender, and risk factor #5 could be clinically significant portal hypertension. In these simulations the incidence rate of death without HCC is 10 per 100 person-years, and the incidence rate of HCC is 1 per 100 person-years. Note that the effect of a risk factor on the relative risk of HCC depends on that risk factor’s effect on both competing events. The relative risks are only weakly affected by the incidence rate of HCC: If it is increased from 1 to 2 per 100 person-years, the 10-year relative risk for risk factor #5 increases from 1.28 to 1.30. They are more strongly affected by a change of the incidence rate of death without HCC: If that rate is increased from 10 to 20 per 100 person-years, the 10-year relative risk for risk factor #5 decreases from 1.28 to 1.08. HCC, hepatocellular carcinoma.
It is likely that some patients with cirrhosis can expect substantial benefit from HCC surveillance, while others can expect substantial harm; we just do not know which patients fall into which group. Clinical equipoise exists somewhere between these extremes, i.e. there are patients whom we would expect to have a fifty-fifty chance of experiencing net harm or net benefit. It is our hope that we can reach a consensus on who these equipoised patients are in terms of, for example, underlying chronic liver disease, severity of cirrhosis, gender, age, comorbidities, and frailty. Then we should include these patients in a randomised trial of HCC surveillance vs. no surveillance. Such a trial could give us all the information we need about benefits, harms, and costs so that an informed decision can be made by patients, clinicians and policy makers on whether or not to take up, deliver or invest in a surveillance programme. It would also avoid the complexities of rates and risks and competing outcomes, as we would simply be comparing all-cause mortality between those under surveillance and those not. There are, however, obstacles to such a trial.
Sample size
Recently, a randomised trial of CT-based surveillance for lung cancer in men aged 50 to 74 years with a smoking history found no effect on all-cause mortality: After 10 years of follow-up, the 13,195 participants randomised to surveillance vs. no surveillance had all-cause mortality rates of 13.93 vs. 13.76 per 1,000 person-years, for a rate ratio of 1.01 (95% CI 0.92 to 1.11). The authors’ conclusion, however, focused exclusively on the rate ratio for lung cancer-related mortality (0.76, 95% CI 0.61 to 0.94), which was the primary outcome.
A randomised trial of HCC surveillance must study all-cause mortality as the primary outcome; it is far more relevant to the patient than HCC-related mortality. Moreover, it can be difficult to determine whether a patient died from HCC or from cirrhosis. Algorithms to define HCC-related death exist,
but they could never be good enough for a trial of an intervention that possibly has a small effect.
An HCC surveillance trial designed to compare all-cause mortality after, say, 5 years between patients randomised to HCC surveillance or no surveillance will need to include several thousand patients (Table 3). Note that if we argue that a trial would require an unrealistically large number of patients, we are at the same time arguing that the effect of surveillance will be very small.
Table 3Sample size estimations depending on all-cause mortality and its components and the hypothesised effect of surveillance on deaths from HCC-related causes.
5-year mortality from non-HCC causes (Fig. 1: state 4)
Proportion of deaths from HCC averted by HCC surveillance
All-cause 5-year mortality
30%
5%
1.5%
28.5%
50%
29.3%
90,730
40%
5%
2.0%
38.0%
50%
39.0%
57,878
50%
5%
2.5%
47.5%
50%
48.8%
38,074
40%
10%
4.0%
36.0%
50%
38.0%
14,418
40%
15%
6.0%
34.0%
50%
37.0%
6,386
50%
20%
10.0%
40.0%
50%
45.0%
2,390
40%
25%
10.0%
30.0%
50%
35.0%
2,282
30%
25%
7.5%
22.5%
50%
26.2%
3,530
30%
25%
7.5%
22.5%
25%
28.1%
14,368
40%
5%
2.0%
38.0%
25%
39.5%
231,912
40%
5%
2.0%
38.0%
75%
38.5%
25,678
40%
5%
2.0%
38.0%
100%
38.0%
14,418
All-cause mortality comprises HCC-related causes (Fig. 1: state 3) and other causes (Fig. 1: state 4). The hypothetical trial has all-cause 5-year mortality as the primary outcome and employs 1:1 randomisation, a power of 80%, a significance level of 5%, and a one-sided test of statistical significance (null hypothesis: surveillance has no effect or reduces survival; alternative hypothesis: surveillance improves survival). HCC, hepatocellular carcinoma.
In a randomised trial of HCC surveillance, the patients randomised to ‘no surveillance’ should receive the standard-of-care except HCC surveillance. That standard may differ between countries. For example, in some places those patients would be followed as outpatients, in others they would not. It will be important to have prespecified subgroup analyses by standard-of-care, because some of the benefit of HCC surveillance may come from the outpatient visits that accompany ultrasound examinations, not from a reduction in HCC-related mortality.
Another source of regional variation is the access to liver transplantation. The benefit of HCC surveillance is likely greater if all patients with unresectable HCC within the Milan criteria can be offered a liver transplantation. That will not be possible in all countries,
and that is a threat to the generalisability of trial results. One solution is to conduct prespecified subgroup analyses by ‘standard practice for liver transplantation for HCC’.
Screening tools
Abdominal ultrasound (with or without alpha-fetoprotein) is the standard screening tool and has been for many years. Newer imaging tools have been examined, but none are in widespread use and neither are biomarker panels. It has been argued that “the promise of HCC screening in reducing HCC-related mortality cannot be fulfilled with currently available tests.”
It is true that more sensitive and specific screening tests could greatly increase the benefits and reduce the harms of HCC surveillance, but for now ultrasound is the standard that we think must be tested in a randomised trial.
Conclusion
Currently there is insufficient evidence to recommend HCC surveillance in clinical practice. We believe that healthcare systems, clinicians, and patients would be much more likely to participate in HCC surveillance if it were based on randomised trial results showing it is of value.
Financial support
PJ was supported by a grant from the Novo Nordisk Foundation (NNF18OC0054612). The funding organization was not involved in the conception, the writing, or the decision to submit the manuscript for publication.
Authors’ contributions
Peter Jepsen drafted the manuscript. Both authors revised and edited the manuscript for content.
Conflict of interest
The authors declare no conflicts of interest that pertain to this work.
Please refer to the accompanying ICMJE disclosure forms for further details.
Feasibility of conducting a randomized control trial for liver cancer screening: is a randomized controlled trial for liver cancer screening feasible or still needed?.