Risks and clinical predictors of cirrhosis and hepatocellular carcinoma diagnoses in adults with diagnosed NAFLD: real-world study of 18 million patients in four European cohorts

Non-alcoholic fatty liver disease (NAFLD) is the most common cause of liver disease worldwide. NAFLD represents a spectrum of disease that includes simple steatosis, non-alcoholic steatohepatitis (NASH) and fibrosis [1]. The numbers of individuals presenting with end-stage complications of NASH, namely decompensated cirrhosis and hepatocellular carcinoma (HCC), are rising [2, 3], and NASH is rapidly becoming the most common indication for liver transplantation [4]. Yet not all patients within the NAFLD spectrum progress, and for the majority, NAFLD is a benign condition [1]. A key clinical challenge is to identify the proportion of patients who are at high risk of developing advanced liver disease, so that interventions, including the many novel therapies in development, can be targeted to those at greatest need.

Our current understanding of NAFLD epidemiology and progression largely derives from single-centre studies of small- or medium-sized cohorts and meta-analyses of these [5‐7]. These studies, together with emerging data from placebo arms of therapeutic trials [8], have taught us that patients with existing evidence of progressive disease (e.g., fibrosis) are at risk of further progression to HCC and decompensated cirrhosis, albeit this may reflect a degree of lead-time bias. Such studies often involve formal assessment of well-phenotyped patients at inclusion but are, by design, selective and may not represent the ‘real-world’ situation for the majority of patients with NAFLD. Paired biopsy data have been reported, although the second biopsy is often performed because of clinical suspicion and not per study protocol, which may bias estimates of progression [9]. Real-world patients are socially and ethnically diverse, have comorbidities and concomitant medications or simply cannot commit to long-term studies or trials and therefore may not be represented by any of these study designs.

Increasingly, real-world data derived from primary care electronic health records (EHR) of a sizeable proportion of the general population [10, 11] are being used to address these issues. In many European countries, where healthcare is largely state-funded and there are low or absent primary care co-payments, the population has unrestricted access to healthcare via primary care physicians who act as gatekeepers for referral to secondary care [12]. People register with primary care centres at birth or when they move to an area in order to access healthcare; therefore, primary care EHR represent data that are as close to the ‘general’ population as possible. If a practice joins the database, all the patients at that practice are registered in the database and, although there is an option for individual patients to opt out, this is minimal (< 1%).

In order to gain insights into the NAFLD spectrum of diseases in real-world patients, we extracted data from four large European primary care databases and identified a cohort of patients with a diagnosis of NAFLD or of NASH. Our aim in this study was to estimate the risk for patients with diagnoses of NAFLD or NASH to acquire a new diagnosis of cirrhosis and HCC and to understand the main predictors for this.

Out of 18,782,281 eligible individuals in the four databases, we identified 136,703 (0.7%) who had a recorded diagnosis of either NAFLD or NASH (coded NAFLD/NASH) and who met the inclusion criteria (Additional file 1: Table S1). The Spanish (SIDIAP) and UK (THIN) databases contributed 71% of all cases; the remaining 29% of coded NAFLD/NASH cases were from the Dutch (IPCI) and Italian (HSD) databases. In SIDIAP, 2.5% of all coded NAFLD/NASH patients (n = 1880) had NASH, and in THIN, this was 4.7% (n = 1212). Due to the coding, NAFLD and NASH could not be distinguished in IPCI and HSD. Therefore, in the initial phase of analysis, we combined all NAFLD and NASH codes from all four databases as coded NAFLD/NASH.

Comparing coded NAFLD/NASH patients across the four databases, there were minor differences between databases in mean age, BMI and proportion with diabetes (Table 1 and Additional file 1: Table S2). BMI data were available in 64.6% of patients with coded NAFLD/NASH and in 45.9% of matched controls (Additional file 1: Table S3). In the subset of patients for whom data were available, ALT and AST values were highest in THIN, and the proportion of obese patients highest in SIDIAP. Sufficient data were available to calculate the non-invasive fibrosis Fib-4 score (age, AST, ALT and platelets) in 46.7% of patients (range 12.6–62.6%, Table 2). THIN (UK) had the smallest proportion of patients with Fib-4 data (12.6%), in whom the proportion of patients with high-risk scores was 10.5%, highest among the four databases.

Patients with a coded diagnosis of NAFLD/NASH had comparable age and sex distribution, smoking rates and duration of follow-up as matched controls (Table 1). As expected, however, controls had lower BMI; lower rates of obesity, hypertension or diabetes; and lower serum levels of ALT and AST.

To our knowledge, this is the largest study to date that has used EHR data to investigate rates of new diagnoses of advanced liver disease in patients with NAFLD. Our patients were well-matched to a very large number of controls according to sex, age, GP practice and most recent visit, thus limiting bias due to geographical and socioeconomic diversity and behaviours relating to health service utilisation. Patients with coded NAFLD/NASH are at significantly increased risk of acquiring a diagnosis of cirrhosis or HCC, compared to matched controls. The risk is greater in patients with a coded diagnosis of NASH compared to NAFLD and in those with high-risk Fib-4 fibrosis scores compared to indeterminate or low-risk scores. Diabetes is an independent risk factor for progression to either HCC or cirrhosis diagnoses in both coded NAFLD/NASH patients and matched controls.

We applied minimal selection criteria and therefore were able to include over 78% of all adults registered in the databases, hence the ‘real-world’ nature of the study. The overall proportion of people with coded NAFLD/NASH diagnoses is lower than expected as reported previously [10], is in keeping with other primary care work [19] and may reflect levels of awareness of NAFLD/NASH in primary care [20, 21]. Hence, our data, by definition, can only represent the visible part of the clinical iceberg. Despite this, we find that patients with coded NAFLD/NASH acquire diagnoses of life-threatening liver disease within a relatively short follow-up period (median 3.3 years).

It is not feasible that the short time intervals between coded diagnosis of NAFLD/NASH and advanced liver disease reflect true rates of disease progression, estimated to be one fibrosis stage per 7 years [22]. The acquisition of a new code in the healthcare record does not necessarily mean that pathological progression has occurred at that time, nor that the stage did not exist at baseline. Our interpretation of these data is that patients in Europe are being diagnosed at the later stages of disease, which are associated with greater risk of liver-related mortality [23‐25].

Less than 50% of patients had sufficient data to calculate Fib-4, the components of which are also needed to calculate many other non-invasive fibrosis scores [26]. There was marked national variation in fibrosis assessment; 73.1% of patients in whom we could calculate Fib-4 were from the Spanish database. We have no way of determining whether these scores were actually calculated by clinicians and whether they influenced decision-making. This is despite the fact that such risk stratification is central to most guidelines [27‐29], used to determine clinical management, select patients for clinical trials and probably triage patients for future therapy.

In the databases where NAFLD/NASH codes could not be distinguished (HSD and IPCI), even those with low-risk Fib-4 scores were at increased risk of cirrhosis and HCC compared to controls. This further suggests that primary care records under-estimate disease severity and that some patients with NAFLD/NASH diagnoses actually have advanced fibrosis or cirrhosis already. Apart from a diagnosis of NAFLD/NASH, diabetes was the strongest independent risk factor for acquiring a diagnosis of cirrhosis or HCC. In the matched control population, the HR for diabetes was even higher than the coded NAFLD/NASH cohort, which may reflect a significant number of individuals with undiagnosed NAFLD/NASH among the controls. The importance of diabetes is consistent with a review of patients who had undergone more than one biopsy in the course of their routine clinical care in the UK, which showed that diabetes was a risk factor for progression of fibrosis [9]. Obesity is an important risk factor for many cancers including HCC [30], but we did not find that in our study. If patients are diagnosed late in the disease spectrum, it is unlikely that patients will have undergone surveillance and HCC may be diagnosed at late stages when symptoms including weight loss are manifest. Taken together, these findings emphasise the need to recognise risk factors for progressive disease and to detect disease at early stages when interventions can be more effective.

This study is subject to limitations. The nature of real-world data is such that we cannot ascertain the origin of codes nor the motivation for adding diagnoses to the patient record. Although the study is based in primary care, it is likely that a large proportion of diagnoses will have been made with some involvement of secondary care. It would be inaccurate to assume that all patients who carry the code ‘NASH’ have had a liver biopsy and histological assessment and it might be that the diagnosis was assumed and recorded based on, for example, ultrasound evidence of fatty liver and elevated serum transaminases or increased stiffness on transient elastography. Similarly, it was not possible to confirm that the matched controls did not have NAFLD/NASH. However, the clinical features of patients with coded NAFLD/NASH are consistent with the diagnostic codes, although if patients with NAFLD/NASH do exist in the control group then the effect sizes reported here are underestimates of the real risk. This means that there are individuals living with diabetes in primary care who have not been diagnosed with NAFLD/NASH but are at significantly increased risk of developing liver cirrhosis and cancer.

The estimated size of the NAFLD problem has raised fears of large unmanageable patient numbers who are not at immediate threat of disease. Notwithstanding our expectation that many cases have not been identified in this study, we have shown that 0.6% of patients with an existing coded diagnosis of NAFLD/NASH acquire a diagnosis of cirrhosis and/or HCC within a 3-year follow-up period. This gives us insight into the rate at which advanced disease is discovered, even if this is not the natural history in the general population. The clinical impact of our data is that they highlight the large gaps in diagnosis and risk assessment of NAFLD and NASH with variable rates of risk stratification, staging of disease and seemingly late diagnosis.

Our knowledge of NAFLD/NASH is being based on small, highly selected cohort studies. These have been accurate in telling us the potential scale of the prevalence and progression of disease, but the reality for many in the general population is some way from that. In order to affect population health and make an impact on the overall health burden of advanced liver disease, we cannot simply rely on introducing effective therapies to the small number of people with established diagnoses. The current approach to opportunistically investigate those in whom abnormalities in liver tests arise is clearly not working. While better biomarkers are needed that identify those at risk more precisely, the current tools are not being used, leaving many patients unclear as to the stage of their disease and its significance to their health. Therefore, making an impact on advanced liver disease will need co-ordinated efforts to identify those with NAFLD, to stage their disease and target those at risk of progression.