In a recent issue of HEPATOLOGY, Lai et al. report results of a cohort study in Taiwan showing a 2.1-fold higher frequency of subsequent hepatocellular carcinoma (HCC) an average of 2.8 years after enrollment among men and women identified with type 2 diabetes.1 This study joins 10 case-control2–15 and 5 prior cohort studies16–23 from 7 countries since 1986 to report on a relationship between diabetes and HCC, all of which show a higher than expected association between these 2 conditions (mean RR = 2.4 with a range of 1.27 to 4.88). All prior studies concluded that diabetes is a risk factor, if not actually etiologic, for HCC even though the studies have not established that diabetes precedes HCC nor considered the possibility that both HCC and diabetes might be the product of other factors, particularly chronic liver disease.
The study by Lai et al. is of particular interest, not so much because it is the first study in a Chinese population, but because of methodological strengths allowing firmer conclusions than are justified from prior studies, notably (1) the substantial effort of these investigators made to limit their evaluation to HCC cases arising subsequent to population enrollment, thus making it clear that diabetes preceded the cancer; (2) diagnosis of diabetes based on fasting blood sugars at enrollment; and (3) the excellence of the Taiwan national health insurance system giving high ascertainment and accuracy of death by cause. The first point is worthy of brief elaboration. Some of the other cohort studies have excluded clinical HCC cases at enrollment, but Lai et al. are the first to eliminate early preclinical cases. They used a 2-stage process [primary screening with alpha-fetoprotein (AFP) and other blood tests associated with increased HCC risk followed by hepatic ultrasound of those with positive results] to their blood test algorithm to find and exclude small preclinical tumors. Thus, their final analysis included primarily HCC cases occurring after the initial screening. While it is likely that some early HCC tumors were missed, the exclusion of prevalent cases go a long way toward establishing that diabetes preceded HCC. Other study strengths include reduction of study self-selection on the part of the subjects, because they consented to participate without knowing of the hypothesized connection between diabetes and liver cancer and high-outcome ascertainment since follow-up was based on the Taiwan's universal health insurance system. These methods, particularly the efforts to eliminate HCC existent at enrollment justify the conclusions. But, as discussed below, this careful methodology does not establish diabetes as a cause of HCC.
There are also some unexplained oddities about the Taiwan study data: (1) the almost 2-fold greater participation of women than men; (2) the proportion of hepatitis B surface antigen (HBsAg) positivity among HCC cases is much lower than has been reported in any of the many previous studies of HCC in Taiwan; (3) the frequency of HCC cases which are positive for both HBsAg and hepatitis C virus (HCV) antibody (anti-HCV) is about half what would be expected if these agents were acting independently; and (4) the frequency of AFP elevations in the HCC cases is much lower than is generally recognized from other studies in Taiwan or elsewhere. The meaning of these observations is unclear and suggests that the study population may not be representative of patients with HCC in Taiwan. Our further comments are based on the assumption that resolution of these issues will not influence the basic observation of an increased frequency of subsequent HCC in subjects shown to have diabetes at enrollment compared with those without diabetes.
Strangely, many of the previous investigators of these epidemiological studies and many members of the medical community have viewed the observed association as suggesting diabetes to be a risk factor for HCC without realizing that the reverse might be the case. A third possibility, and the one I believe is most likely, is that an increased risk of both HCC and diabetes results from underlying liver disease from some or all of the established causes of HCC. Diabetes is an important risk factor for nonalcoholic fatty liver disease which has been reported to progress to HCC. Many studies reported a higher frequency of diabetes among patients with chronic HCV infection. Furthermore, cirrhosis which is diabetogenic often precedes HCC. Because not all the studies mentioned these possibilities, it is worth restating that temporal direction must be established before any factor can be identified as causing another. And, by their nature case-control studies estimate the magnitude of association but do not provide information about the temporal direction of cause and which might be the effect.
The only undisputed hard-and-fast criteria required to establish causation are association (significant correlation between putative cause and effect) and temporal direction such that the cause(s) precede the effect(s). Temporal direction is often obvious, at least when the association of interest is between an external and an internal factor, e.g., alcohol drinking and macronodular cirrhosis. It is historically noteworthy that many biologists resisted the very strong evidence that hepatitis B virus (HBV) causes HCC because they believed that persons with HCC might be at increased risk of contracting chronic HBV infections, i.e., reversed temporality.
Prospective studies are a common method of establishing the temporal sequence between 2 factors, but they serve to do so only when the study is constructed to differentiate outcomes occurring before and after exposure to the putative risk factor under study. Even though some of the studies prior to Lai eliminated HCC diagnosed before enrollment, none made the effort to find and eliminate clinically inapparent cases at enrollment or excluding all cases for at least 5 years so they could be excluded when they become large enough to become clinically manifest and not be confused with incident cases.
Other important observations of the Lai study are statistically increased risk of HCC observed in diabetics with each of the established HCC etiologic agents they studied: HBsAg positivity, anti-HCV positivity, heavy alcohol consumption, and/or heavy smoking. These observations will contribute to the overall body of knowledge about the relationship of liver disease, diabetes, and HCC.
The Lai study also attempted to identify confounders, another source of confusion in epidemiological studies. Of interest is their observation that obesity (expressed as BMI ≥ 30) does not appear to be a source of confounding. Age and gender, as usual, were found to be confounders as were total cholesterol and LDL.
Confounding as used in epidemiology has a specific, narrower meaning. Confounders are noncausal factors whose occurrence vary with the etiologic agent, and therefore with the outcome(s), under study. They are confusing, notoriously tricky, sometimes difficult to find, may go unidentified for years, and there is never certainty that all have been eliminated. They are particularly difficult in relation to internal (host-based) putative causal factors such as diabetes and liver disease.
The past several decades have seen great advances in the understanding of the etiologies of HCC. Established causes of HCC are many, including aflatoxin consumption, alcohol drinking, alpha radiation (thorium-232), anabolic steroid abnormalities, HBV chronic infection, HCV chronic infection, hemochromatosis, tobacco smoking, as well as several metabolic disorders. It is not known to what extent these etiologies share mechanistic pathways, but where data is available, the effects appear to be additive rather than multiplicative when they occur together. There is striking variation in the relative frequencies of the etiologies in different populations. HBV is the dominant etiology in all Chinese populations and throughout east Asia and the Pacific. In the United States, the dominant etiologies in non-Asian Pacific populations are HCV and alcohol.
Growing understanding of mechanisms of action of proposed etiological agents usually accompany acceptance of agents as etiologic. Thus, it is interesting that relatively little is known about the mechanisms by which most accepted agents cause HCC. A notable exception is alpha radiation, which damages DNA, from Thorotrast dyes used in the past as diagnostic contrast media. Frustration in identifying mechanisms for other well-established etiologies has led many scientists to suspect nonspecific mechanisms related to liver inflammation, hepatocyte damage, and repair.
Lai et al. and several other authors of studies have suggested mechanisms by which diabetes might induce cancer, e.g., insulin has insulin growth factor–like activity. But it is not difficult to imagine that liver disease might alter glucose metabolism, perhaps through its effect on sex hormone binding globulin, which also transports insulin.
Sorting out the causal sequence and relationships between glucose/insulin metabolism and liver disease, including HCC, is likely to take considerable time and effort. Fundamental problems inherently exist with the insidious onset and long cryptic nature and fluctuating courses of liver disease and diabetes, making their identification and long-term evaluation difficult, irrespective of whether they are risk factors or outcomes. A starting point might be to determine whether the liver disease of various etiologies is as frequent and severe in patients with HCC who are also with and without diabetes. If they can be found, it may be particularly useful to determine whether diabetes is unusually frequent in patients with HCC who are also without liver disease or exposure to the known etiologic agents.
Definitive determination of the causal relationships between diabetes and liver disease will not be easy and is likely to require considerable work by a variety of scientists. Whatever studies are undertaken, they will need to concentrate on establishing sequences in the causal chain. No single study or single type of study will work. All studies will need to be attuned to methods to the effects of known and potential confounders of which age, gender, and race are the most obvious, over and beyond those which directly affect liver disease and glucose metabolism.
The Taiwan population is ideal for some of these studies because of the national insurance system which makes follow-up much easier than in most other populations. Dr. Lai's Keelung population should be kept under study and reevaluated for HCC incidence 5 or more years after elimination of enrollment cases. The investigators can also look at evolving glucose and insulin changes while comparing HBsAg carriers with noncarriers.
More prospective epidemiologic studies are needed starting earlier/younger in carefully defined diabetes patients, and should look for indications of increased frequency and severity of liver disease over long-term periods, with liver biopsies if possible. Similar studies starting with early liver disease to document the frequency and evolution of diabetes and concomitant metabolic changes should be conducted. Of particular interest may be populations with high exposure to known HCC carcinogens before the occurrence of liver disease, particularly because their known HCC risk may justify biopsies to accompany glucose and insulin studies. It will also be of interest to know whether alterations in glucose insulin metabolism are the same or different for different types of liver disease.
Animal studies will be of interest. Carefully considered primate studies might be of interest if sufficient numbers are available. Of greater interest might be glucose/insulin studies in animals susceptible to hepadnavirus infections, such as woodchucks, pekin ducks, and Beechey ground squirrels, that would look for differences in infected versus noninfected animals in different stages of liver disease.
In conclusion, HCC and type 2 diabetes have repeatedly been shown to occur together about twice as frequently as would be expected by chance alone, too consistently from these diverse studies to be due to chance or bias. Lai et al. and the authors of every case-control and cohort study to date have suggested that diabetes must be considered a possible etiology, or at least risk factor for HCC, and I believe it is at least as likely that the diabetes is induced by the liver disease which usually precedes HCC. Possible confounding such as obesity must also be considered. The fundamental issues for now are to confirm the Lai study observation that diabetes precedes HCC and to determine whether diabetes (abnormal glucose and insulin metabolism) precedes the liver disease that leads to HCC or whether the liver disease precedes the diabetes.