We applied competing risks modelling to a large cohort of individuals with a histological diagnosis of NAFLD that was prospectively followed for a median follow-up time of 84 months and found that the probability of developing LRE was negligible in F0-F1, low in F2, and clinically relevant in F3-F4 patients. Furthermore, the risk of incident EHE persisted among all three groups and increased according to the baseline severity of liver fibrosis. Notably, these results were replicated in a large cohort of patients with NAFLD diagnosed by non-invasive assessment of liver fibrosis.
Patients with NAFLD are characterised by a high risk of developing not only hepatic but also EHE[3–13] that are associated with the severity of liver fibrosis.[15,16] In the context of the heterogeneous burden of hepatic and non-hepatic outcomes and the potential differential effects of fibrosis on their development, available evidence on the natural history of NAFLD is limited by the lack of conclusive data when LRE and EHE are considered as competing risks. In the present study, the application of competitive risk modelling on the full spectrum of NAFLD patients confirmed that the severity of liver fibrosis is an independent risk factor for both LRE and EHE. Notably, after stratifying for liver fibrosis severity, we observed that in F0-F1 patients the probability of developing LRE was negligible over time. However, these patients experienced a low but progressive increase in the risk of EHE (Table 3). Patients with F2 fibrosis were characterised by an LRE risk that was low but higher than in the F0-F1 group, risk of HE (Table 3), mostly due to the incidence of HCC; and also exhibited an increasing incidence of EHE (Table 3). Finally, patients with F3-F4 fibrosis had a higher and clinically relevant probability of developing both LRE and EHE (Table 3). Subgroup analyses in patients without previous CVE and/or previous extrahepatic cancers, as expected, demonstrated a slightly lower risk of developing EHE. Our data indicate that at a horizon of 60 months, the probability of developing LRE is negligible in F0-F1 patients, low in F2 patients, and clinically relevant in F3-F4 patients, while the risk of EHE is present in all NAFLD patients but increased according to the severity of liver fibrosis. Notably, our data from F2 patients add further insights regarding the risk of developing HCC in patients with NASH but without cirrhosis,[30–33] even if systematic screening in these patients cannot be suggested.[34–36] A recent multicentre study used a competitive risks model to evaluate the occurrence of hepatic and extrahepatic events in a cohort of 458 patients with histological diagnosis of NAFLD with bridging fibrosis or cirrhosis, including patients with Child A6 cirrhosis. The probability of LRE increased progressively from F3 to F4 Child-Pugh A5, and further to F4 Child-Pugh A6 patients. In contrast, an inverse trend for EHE was observed. A subgroup analysis of our cohort that split F3 from F4 patients showed a high incidence of LRE and EHE in both subgroups, and further confirmed the higher incidence of EHE in respect to LRE in F3 patients and an inverse relationship in F4 patients. Our analysis also showed a lower HCC incidence in F3 compared to F4 patients, thus raising doubts regarding the cost-effectiveness of HCC screening in F3 patients, in whom clusters at higher risk for HCC should be identified.
Our competitive cause-specific risks model in F0-F1 and F2 patients identified older age and previous history of EHE as risk factors for incident EHE, and older age as an independent predictor of LRE. On the other hand, in patients with F3-F4 fibrosis, older age, obesity, thrombocytopenia, and higher log(GGT) levels were risk factors for LRE, while older age and previous CVE were predictors of EHE. Notably, when combining these risk factors (presence or absence of all the independent predictors), we identified classes of patients with high or low probabilities of developing events. Specifically, from the low- to the high-risk class, the 60-month probability of LRE increased between F2 and F3-F4 patients. Similarly, the 60-month risk of EHE increased between F0-F1, F2, F3-F4 patients (Figure 2, Table 4, Figure S2). Collectively, these data could inform risk stratification in NAFLD patients to further refine follow-up and resource allocation. It is especially notable that all results were confirmed when replacing the CSC with the FG model.
The present study addresses another unresolved question of NAFLD natural history, especially in patients with advanced fibrosis; that is whether first LRE or EHE in NAFLD patients are definitively committed to the liver or not liver-related way. We showed that among F0-F1 or F2 patients developing a first EHE, the risk for a second LRE remains low, especially in the F0-F1 group. Most notably, F3-F4 patients who had a first LRE maintained a clinically relevant risk of EHE, while the inverse was observed for those experiencing a first EHE. These findings suggest that patients with NAFLD and advanced fibrosis should receive follow-up for both hepatic and extrahepatic complications after developing a first event.
Our findings in the histological cohort were replicated in a large multicentre cohort of NAFLD patients in whom liver fibrosis was assessed non-invasively by using LSM. This is a crucial point because, despite the limited accuracy of LSM for staging fibrosis in NAFLD,[37,38] we demonstrated that similar results regarding natural history and risk stratification were yielded by histological and LSM assessments. Consequently, the reproducibility of our results will allow us to confidently apply our conclusions to large clinical populations in whom NAFLD severity may be evaluated non-invasively, even if our tertiary cohort is different from NAFLD individuals from the general population.
The clinical applicability of our findings is further strengthened because competitive risk modelling is a more appropriate statistical method to assess the impact of clinical decision making in complex epidemiological and clinical contexts such as NAFLD, that feature independent major clinical endpoints. A Competing risk is an event that either precludes or modifies the risk of an event of interest. Time-to-event analysis using either the naive Kaplan-Meier estimator or the PH Cox model to estimate the probability of the event of interest in the presence of competing risks could provide biased estimates, as also partially observed in the present study. Consequently, CSC and/or FG models should be applied and the predicted CIFs reported instead. Although CSC and FG often yield similar results, the interpretation and the conclusions drawn from these models may differ. The CSC model estimates risk factor effects on the hazard of a specific event. Although CSC estimates may be identical to those obtained from a PH Cox model, the probability of the event of interest from the Cox model could be overestimated. However, caution should be exercised when interpreting the effects of risk factors on the CIFs from a CSC model, as an increase in the hazard due to increased predictor values does not necessarily correspond to an increased CIF. The FG model overcomes this problem as it can be viewed as a model for the CIF. Unlike in the CSC, where a predictor acts on the hazard (instantaneous rate of occurrence of the event of primary interest is estimated in subjects who are event-free), in the FG the predictor acts on the subdistribution hazard, that is the instantaneous rate of occurrence of the event for subjects who are event-free or who have experienced a competing event, which in turn acts on the predicted CIF proportionally. In the FG model, an increase in the covariate is related to an increase in the corresponding CIF, but the magnitude of the CIF increase cannot be quantified. A drawback of the FG model is that the sum of the CIFs of the two events may exceed 1 for some covariate patterns. This drawback is obviated by the CSC model.
The primary limitation of this study is the lack of time-dependent variables and data capturing liver disease progression over time. Another notable limitation is the lack of data regarding common risk factors for EHE such as smoking and family histories of CVE and extrahepatic cancers. The multicentre study design could also affect our results; however, a GEE approach was used, and robust standard errors were obtained. Another potential limitation is that the competing risks and classical PH Cox models may yield similar predicted probabilities when the proportion of events is low (<10%). Such a low proportion was present in our F0-F1 and F2 strata, but not in F3-F4. Consequently, the results of the competing risks model can be viewed as a refinement of the PH Cox analysis in the F0-F1 and F2 strata, while providing optimal results in F3-F4, as it produces unbiased predicted CIFs. Other limitations are the use of BMI ≥ 30 kg/m2 instead of ethnic-specific BMIs to diagnose obesity, the potentially hidden alcohol abuse during follow-up, the lack of baseline and follow-up data on high-risk oesophageal varices and of time-dependent variables, and selection bias in the liver biopsy cohort. Regarding this last point, we observed a lower LRE incidence rates in all fibrosis strata compared to the findings of a recent meta-analysis. This difference, probably related to the heterogeneous follow-up among studies, could potentially affect the applicability of our results. Finally, the aggregation of HCC and LD together as LRE and cardiovascular events and extrahepatic cancers as EHE further limits the interpretation of our results.
In conclusion, this study of NAFLD patients stratified for baseline severity of liver fibrosis showed that over a 60-month horizon the probability of EHE is consistent and increases according to the severity of fibrosis, while the risk of LRE is negligible in F0-F1 patients, low in F2 patients, and significant in F3-F4 patients. The latter require ongoing follow-up for both LRE and EHE. These data can inform personalised prognosis and follow-up in patients with NAFLD.
Guarantor of the article
Aliment Pharmacol Ther. 2022;55(5):604-615. © 2022 Blackwell Publishing