1. Introduction
There have been significant changes in the epidemiology of complications of type 2 diabetes (T2D) over the past few decades. Studies from the US, Europe, and Asia have shown declines in cardiovascular disease (CVD) and microangiopathy reflecting, in large part, improved CVD risk factor management [
1,
2,
3]. Recently published Australian data collected over 25 years from our representative community-based longitudinal Fremantle Diabetes Study Phase I (FDS1) and II (FDS2) show the same trends [
4]. This means that people with T2D are now living long enough to develop nonvascular complications with major consequences. One of the most prominent among these is likely to be hepatobiliary disease (HBD), especially non-alcoholic fatty liver disease (NAFLD).
Diabetes is associated with an increased risk of a variety of HBDs. The prevalence of NAFLD in T2D is estimated to be approximately double that in the general population [
5], and there have been concerns that a ‘tsunami’ of diabetes-related NAFLD complications including non-alcoholic steatohepatitis (NASH), cirrhosis, and hepatocellular carcinoma (HCC) is imminent in part because of the increasing incidence of T2D fueled by the obesity epidemic [
6]. Gallbladder disease is also associated with diabetes independent of obesity in most relevant studies [
7]. There have been reports of a high risk of death due to alcohol-related liver disease in men with diabetes [
8], while viral hepatitis [
9,
10] and genetic hemochromatosis [
11] have also been considered to increase the risk of liver disease complicating diabetes.
Specific evidence of a link between T2D and HBD comes from the FDS1 [
12] and from a Canadian administrative database study conducted between 1994 and 2006 [
13]. In the latter study, there was a hazard ratio of 1.77 for the combined endpoint of liver cirrhosis, liver failure, or liver transplantation in newly diagnosed people with diabetes of unspecified type. The findings in the FDS1 from data collected between 1993 and 2010 showed there was a significantly increased incidence rate ratio (IRR) of 1.66 for hospitalizations and deaths for/from a wider range of HBD in 1294 participants with T2D compared to 5156 matched individuals without diabetes followed for a mean of 11.5 years from recruitment in 1993–1996 [
12]. However, only 7.6% of incident HBD events in the FDS1 participants with T2D were attributable to NAFLD.
Given the shift in complications away from vascular disease over recent decades [
1,
2,
3,
4] and the increasing obesity in developed countries such as Australia [
14], we postulated that the incidence of HBD, and particularly NAFLD, would have increased since the FDS1 was conducted, in line with global predictions [
6]. The aim of the present study was, therefore, to determine the incidence of hospitalizations and deaths for/from HBD in participants with T2D in the FDS2 cohort and in matched individuals without diabetes recruited 15 years after the equivalent groups in the FDS1.
2. Materials and Methods
2.1. Participants, Epidemiologic Setting, and Approvals
The FDS2 is an observational, longitudinal study of people with known diabetes conducted in a zip code-defined geographic area surrounding the port city of Fremantle in the state of Western Australia (WA) [
15]. The FDS2 utilized the same basic design as FDS1 for identification and recruitment of participants with diabetes and matched residents from the same catchment area who did not have diabetes. During a three-year recruitment period between 2008 and 2011, individuals with diabetes (except gestational diabetes) were identified from hospital, clinic, and primary care patient lists; widespread advertising through local media, pharmacies, optometrists, and networks of health care professionals; and, in the case of FDS2 but not FDS1, third-party mail-outs to registrants of the Australian National Diabetes Services Scheme and the National Diabetes Register. Full details of recruitment, the final FDS2 sample characteristics including classification of diabetes types, and nonrecruited people identified with diabetes in the catchment area have been published [
15]. The FDS2 protocol was approved by the Human Research Ethics Committee of the Southern Metropolitan Area Health Service (07/397 18 October 2007). All participants gave written, informed consent.
In FDS2, 4639 people with diabetes were identified from a population of approximately 157,000, and 1668 (36%) recruited of whom 1509 (90%) had T2D. Four age-, sex-, and postcode-matched residents (n = 6036) with no coding of diabetes before study entry on any WA health database were randomly selected from the study catchment area for each FDS2 participant at the time of their enrolment using the WA Electoral Roll and the WA Registry for Births, Deaths, and Marriages as a source of all residents. The present analyses excluded types of diabetes other than T2D. The follow-up of the matched residents was censored if and when they developed diabetes.
2.2. Clinical Assessment and Laboratory Tests
At baseline and three biennial face-to-face reviews up to 2017, a medical questionnaire was completed, a physical examination was performed, and fasting biochemical tests were carried out in a single, nationally accredited laboratory. Self-reported alcohol consumption was recorded, as were details of prior illnesses including HBD. Ethnic background was assessed as Anglo-Celt, Southern European, Other European, Asian, Aboriginal, or Other. Microvascular and macrovascular complications of diabetes were identified using standard criteria [
16]. Face-to-face assessments were interspersed with three biennial postal questionnaires to supplement data collection.
2.3. Hospitalizations, Mortality, and Cancer Ascertainment
All hospitalizations, deaths, and cancer registrations in WA are recorded in the WA Data Linkage System (WADLS) [
17], which was used to provide FDS2 participant outcomes from January 1980 until the end of December 2016. Incident HBD was taken as hospitalization/cancer registration for a diagnosis (principal or secondary) of (1) liver disease, (2) viral hepatitis, (3) malignant primary neoplasm of liver and intra-hepatic bile ducts, and (4) disorders of gall bladder and biliary tract as per relevant International Classification of Disease (ICD)-9 Clinical Modification (CM) and ICD-10 Australian Modification (AM) codes (see
Table 1). NAFLD (ICD-10: K76.0) and NASH (ICD-10: K75.8) are not specifically defined within ICD-9 and so consideration of these conditions was limited to post-introduction of ICD-10 coding in Australia (on 1 July 1999). Causes of death were reviewed independently by two FDS physician investigators and classified under the system used in the UK Prospective Diabetes Study [
18]. In the case of discrepant coding, case notes were consulted and a consensus obtained. Deaths attributable to NAFLD-associated cirrhosis were those in which (1) NAFLD or associated terminology was part of the reported causes of death, (2) another cause for cirrhosis such as alcohol or viral hepatitis was not reported, and/or (3) NAFLD was considered likely based on longitudinal FDS2 and other premorbid data. Secondary liver cancer and its sequelae and liver failure not attributable to HBD were excluded from the analyses.
2.4. Statistical Analysis
The computer packages IBM SPSS Statistics 25 (IBM Corporation, Armonk, NY, USA) and StataSE 15 (StataCorp LP, College Station, TX, USA) were used for statistical analysis. Data are presented as proportions, mean ± SD, geometric mean (SD range), or, in the case of variables that did not conform to a normal or log-normal distribution, median [interquartile range, IQR]. Two-sample comparisons were by Fisher’s exact test for proportions, by Student’s t-test for normally distributed variables, and Mann–Whitney U-test for non-normally distributed variables. Overall and 10-year age- and sex-specific incident rates for HBD were compared (1) in those with T2D in FDS1 and FDS2, and (2) for FDS2, in those with T2D and no diabetes, and respective IRRs derived. Overall IRRs for those aged >25 years were also estimated.
For the FDS2 type 2 diabetes cohort, Cox regression with age as the timeline and backward conditional variable entry (p < 0.05) and removal (p ≥ 0.05) was used to identify independent determinants of the first episode of HBD during follow-up from clinically plausible baseline variables at p < 0.20 in bivariable analyses. Log-normally distributed data were log (ln) transformed before analysis. The validity of the proportional hazards’ assumption was assessed by the examination of time-dependent covariates. Since some variables of interest were missing up to 20 values (1.5%), missing values were multiply imputed (×20), defining imputation models that included incident HBD. HBD component diagnoses were not included in similar Cox regression models because of their limited numbers.
4. Discussion
The present detailed epidemiologic data show that the burden of HBD has increased over the last few decades in community-based Australian adults with T2D. Participants with T2D in the FDS2 were 30% more likely to be hospitalized for HBD, registered with hepatobiliary cancer, or to have died from/with HBD compared with those from the FDS1 recruited on average 15 years earlier. In addition, HBD was over twice as frequent during follow-up in FDS2 participants with T2D relative to a matched group of people without diabetes from the same geographic area, more than double the excess HBD risk in T2D versus no diabetes seen with the same comparison in FDS1 [
12]. These findings, and the significantly greater BMI and waist circumference in FDS2 participants with T2D versus those in FDS1, support the notion that an increase in HBD complicating T2D is one of the legacies of the obesity epidemic, albeit that the average annual increase in HBD events over the 15 years between FDS1 and FDS2 was 2%/year. This is of concern but, notwithstanding that obesity-related liver disease is one part of HBD as a whole, it appears to fall short of the predicted ‘tsunami’ [
6].
The age-specific IRRs indicated that the greatest disparity in HBD events for participants with T2D in FDS2 versus FDS1 was in those aged <65 years. This finding is in accord with US epidemiologic data showing that middle-aged adults have accounted for an increasing proportion of diabetes-related, largely vascular, complications over recent decades [
20]. The same pattern was evident in the present study for the comparison between the two matched cohorts without diabetes in FDS2. This suggests that increasing obesity at younger age groups in the general population in Australia [
14] and other developed countries such as the US [
21] may have parallel implications for HBD rates for those without diabetes.
There were a number of independent baseline predictors of incident HBD events in FDS2 participants in a Cox proportional hazards model with age as the time scale. These included Aboriginal ethnicity, consistent with previous reports of an increased risk of hospitalization for cirrhosis [
22] and of hepatobiliary cancer [
23,
24] in Australian indigenous communities. Both BMI and ABSI were in the model, confirming the strong link between obesity and HBD, with the ABSI as an indicator of more central concentration of body mass than BMI [
19] having a comparatively stronger association. Although hypertension is a feature of the Metabolic Syndrome, which includes central adiposity and an increased risk of NAFLD [
25], it has long been recognized that chronic liver disease can be associated with low blood pressure [
26], consistent with the inverse relationship between systolic blood pressure and HBD events in the present study. The association with severe renal impairment can occur in the absence of significant alterations in renal histology (pre-renal), but intrinsic renal abnormalities can also complicate chronic liver disease [
27]. A raised serum gamma glutamyl transferase and low platelet count and are found commonly in chronic liver disease, and their ratio has been suggested as a prognostic index for the development of fibrosis and cirrhosis [
28].
Within the limitations of temporal changes in the ICD system that complicated direct comparisons between FDS phases, the percentage of HBD events attributable to NAFLD/NASH in FDS2 was 54% greater than in FDS1 and the IRR versus no diabetes in FDS2 was greater than 4. These observations are consistent with the increase in central adiposity between FDS phases and the relatively high prevalence of NAFLD in T2D compared with that in the general population [
5]. However, NAFLD/NASH was coded in only one in 11 HBD events in FDS2 and represented a similar percentage (10%) of the low rate of deaths from HBD (<4% of total mortality). As with HBD events as a whole, the between-phase increase in NAFLD/NASH in T2D is concerning but appears to fall short of the predicted surge in diabetes-related NAFLD complications [
6].
There are a number of possible reasons why our epidemiologic data do not align with predictions. First, the glitazones and newer therapies for T2D including the glucagon-like peptide 1 receptor agonists (GLP-1RA) and sodium–glucose co-transporter-2 inhibitors (SGLT2i) have evidence of benefit in NAFLD/NASH [
29]. However, few of our participants (<1%) in FDS1 and FDS2 were taking a glitazone, and the use of GLP-1RA and SGLT2i use in FDS2 was similarly low given that they were marketed in Australia relatively late in the period of follow-up. Renin-angiotensin system (RAS) blocking drugs may have a role in attenuating the deleterious effects of RAS activation on the liver [
30]. The use of these agents increased between FDS phases and may have limited an increase in NAFLD/NASH events over time. Lipid-modifying agents including statins and fibrates, which increased considerably between FDS phases, have also been suggested as influencing hepatic fat accumulation and fibrosis but the data are inconsistent [
31]. There is evidence that improved glycemic control, as seen in FDS2 relative to FDS1, might itself reduce the progression of NAFLD/NASH [
32]. An examination of the probably complex relationship between these factors and temporal changes in HBD, including NAFLD/NASH, was beyond the scope of the present study.
Current guidelines suggest that liver enzymes, steatosis biomarkers, and liver ultrasonography be considered for all people with T2D based on the need to identify individuals with NAFLD who are at risk of progression to NASH and its adverse consequences [
33,
34]. Given the present data, which suggest that these people form a relatively small subgroup of the total T2D population, as well as the high direct and indirect costs of such testing, the relatively low predictive value of non-invasive tests, the risks of liver biopsy, and the current paucity of effective treatments [
35], future detailed analyses of longitudinal population-based studies of representative samples such as FDS2 could be useful in generating algorithms for simple cost-effective NALFD/NASH screening in usual care. Such algorithms could identify those people with T2D who are at risk of adverse outcomes and thus warrant further, more intensive investigation including specialized biochemical tests and imaging.
The present study had limitations. Although the use of administrative data to identify true cases of HBD has had variable reliability in previously published studies [
36,
37], the WADLS is regularly validated [
17]. In addition, the known relationship between T2D and HBD including NAFLD/NASH may have meant that coding errors favored greater HBD ascertainment in individuals with diabetes. This would mean that the true difference between participants with T2D and without diabetes was less than we observed. We performed restricted multivariable analyses as there were limited data available for people without diabetes in both cohorts, and the number of outcomes in individual components of HBD including NAFLD/NASH was also limited. The strengths of the study include the representative, community-based participant samples and detailed participant-level data for those with T2D.