The impact of non-alcoholic fatty liver disease and liver fibrosis on adverse clinical outcomes and mortality in patients with chronic kidney disease: a prospective cohort study using the UK Biobank

The UK Biobank (UKBB) is a national prospective cohort study aimed at improving disease prevention (http://​www.​ukbiobank.​ac.​uk/​about-biobank-uk). Over 500,000 individuals aged 40–69 agreed to participate and were recruited between 2006 and 2010. During baseline assessment visits, participants completed questionnaires about their demographics, medical history and lifestyle. Self-reported doctor-diagnosed medical conditions were verified and coded during a face-to-face interview (https://​biobank.​ctsu.​ox.​ac.​uk/​crystal/​field.​cgi?​id=​20002). Volunteers underwent a physical examination and provided blood and urine samples. All participants gave consent to be followed up through linkage to electronic health records (death and cancer records held by the Office for National Statistics and the Registrar General’s Office; hospital records held by the Department of Health’s Hospital Episode Statistics; Scottish Morbidity Records). At the time of analysis, mortality and hospital admission data were available to January 2023. Ethical approval for the UKBB study was granted by the North West Multi-Centre Research Ethics Committee (06/MRE08/65).

We identified all participants within the UKBB who had evidence of CKD at their baseline visit determined by a single eGFR value (eGFR) < 60 ml/min/1.73 m² or a random urine albumin creatinine ratio (UACR) ≥ 3 mg/mmol [20]. eGFR was calculated using single serum creatinine and cystatin C measurements, omitting race as per the most recent guidance [21]. We present the combined equation as this is a more valid measure [22] but undertake a sensitivity analysis using serum creatinine alone to calculate GFR [23], as is currently recommended for routine practice in UK [24]. Detailed information on how the urine samples were collected and the methods used to calculate the urine albumin creatinine ratio can be found in the supplementary material (Additional file 1: Supplementary methods).

Participants were excluded if they had evidence of baseline ESRD (eGFR < 15 ml/min/1.73 m² or anyone identified to have ESRD according to the UKBB algorithm (data field 42,026) [25]. Participants were also excluded if they had undergone a liver transplant and had a non-NAFLD cause of liver disease at baseline (Additional file 2: Table S1), evidence of alcohol abuse (Additional file 3: Table S2) or a baseline alcohol intake of ≥ 20 g per day for women and ≥ 30 g per day for men. Approximately 27% of participants had data on the frequency of alcohol consumption only (i.e. not weekly grams); in this case, we excluded individuals drinking daily or more than daily. Finally, we excluded all participants who did not have data available to calculate the following scores at baseline: hepatic steatosis index (HSI), fibrosis-4 score (FIB-4) and NAFLD fibrosis score (NFS). These are validated algorithms comprising both clinical and biochemical parameters (calculations shown in Table 1).

Individuals were identified as having NAFLD if they had either an ICD-9 (571.8) or ICD-10 (K75.8, K76.0) code indicating NAFLD or an HSI > 36 [26]. Individuals were defined as not having NAFLD if they had no ICD code for NAFLD and HSI < 36. The influence of advanced fibrosis was examined in patients with CKD and NAFLD using the NFS [27, 28] and FIB-4 score[28, 29]. Advanced fibrosis was defined as NFS ≥ 0.676 or FIB-4 score > 2.67 at baseline. Advanced fibrosis was excluded if there was a NFS <  − 1.455 (< 0.12 if ≥ 65 years) or FIB-4 score < 1.3 (< 2.0 if ≥ 65 years) [16]. Participants not falling into these groups were placed in an indeterminate fibrosis group. We selected NFS and FIB-4 to identify participants with liver fibrosis because they are superior to other scores that predict the presence of liver fibrosis [28] and could both be calculated from the data collected in UKBB participants. The positive predictive value of NFS ≥ 0.676 and FIB-4 > 2.67 in predicting liver fibrosis in patients with NAFLD is 90% and 80%, respectively [28].

Primary outcomes included the risk of incident CVE, progression to ESRD and ACM:

Baseline characteristics are presented as percentages, and continuous data are presented using the median and interquartile range (IQR). Univariate and multivariable Cox proportional hazards models were used to calculate the association of NAFLD and hepatic fibrosis (in people with NAFLD) on CVE, ESRD and ACM. Hazard ratios (HR) with 95% confidence intervals (CIs) are presented. Statistical significance was taken as p < 0.05. Non-cases were censored at the date of loss to follow-up, date of death or end of follow-up. Individuals were also censored for a CVE if they developed ESRD first, as it was thought ESRD would alter the course and mechanisms of cardiovascular injury and outweigh the influence of NAFLD. People were not censored for other subtypes of CVE having developed a different subtype. Multivariable adjustment was informed using direct acrylic graphs (DAGitty) [31]. Model 1 adjusted for age, sex, ethnicity and Townsend Deprivation Index (a place-based metric of socioeconomic status based on car ownership, home ownership, employment and over-crowding) [32]. Model 2 additionally adjusted for smoking status (never, previous, current), baseline eGFR (G1-5) and baseline UACR (A1-3) (Additional file 5: Table S4). Model 3 adjusted for the above factors in addition to diabetes (see Additional file 1: Supplementary methods for definitions). There was at least 80% power to detect a 15% increase in the hazard of all outcomes [33]. The linearity of the effect of each continuous variable in the adjusted models was determined using univariate Cox hazard regression with penalised splines. Where a Wald-type test using the nonlinear coefficient estimates indicated significant non-linearity, flexible splines were used in further analyses. The validity of the proportional hazards assumption for each variable was determined by examining correlations between scaled Schoenfeld residuals and time. The statistical package used was R. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines were followed in reporting this study [34].

Overall 455,260 UKBB participants had recorded baseline data for eGFR or albuminuria; 32,801 (7.2%) had evidence of CKD (Additional file 5: Table S4, Fig. 1). Following exclusions, the final study sample consisted of 18,073 participants with CKD (Fig. 1). For the analyses of CVE outcomes, a further 1458 people who had experienced a baseline CVE were excluded. Baseline demographics are presented in Table 2. The baseline characteristics of individuals excluded due to insufficient data to calculate the HSI and serum fibrosis scores did not differ significantly from those included (Additional file 6: Table S5).

In this CKD cohort, 56.2% (n = 10,152) of people were identified as having NAFLD. Those with NAFLD were more likely to be male and have a diagnosis of metabolic disease and lower baseline eGFR (Table 2). The prevalence of NAFLD risk-stratified according to CKD stage is presented in Additional file 7: Table S6. For people with CKD and NAFLD, 7.7% (n = 784) and 3.0% (n = 308) were identified as having advanced fibrosis according to NFS ≥ 0.676 or FIB-4 > 2.67, respectively. Individuals with advanced fibrosis were more likely to be male, have features of the metabolic syndrome, have experienced a prior CVE and have poorer baseline renal function (Additional file 8: Table S7).

The median follow-up time was 13.2 years for a CVE and 13.6 years for the development of ESRD and ACM. In total, 1666 individuals with baseline CKD developed a CVE, 215 progressed to ESRD and there were 1942 deaths. The event rates for incident CVE (fatal and non-fatal events), ESRD and ACM were higher for individuals with NAFLD (Additional file 9: Fig. S1). Univariate analysis of factors associated with increased HR of CVE, ESRD and ACM is shown in Additional file 10: Table S8. The event rates for all primary outcomes increased with increasing severity of CKD at baseline according to both eGFR and albuminuria (Additional file 11: Table S9).

Univariate analysis revealed that NAFLD was associated with an increased risk of all CVE (HR 1.49 [1.38–1.60], p < 0.0001), ACM (HR 1.22 [1.14–1.31], p < 0.0001) and ESRD (HR 1.26 [1.02–1.54], p = 0.0298) (Table 3). Following multivariable adjustment for age, sex, ethnicity, deprivation, alcohol, smoking, baseline renal function and diabetes, NAFLD remained an independent risk factor for CVE overall (HR 1.20 [1.11–1.30], p < 0.0001), including ACS (HR 1.22 [1.06–1.41], p = 0.0057) and HF (HR 1.29 [1.15–1.45], p < 0.0001), but not ACM (HR 0.92 [0.85–1.00]) or ESRD (HR 0.77 [0.60–0.98]) (Table 3, Fig. 2). We examined the change in direction of the association between NAFLD and ACM after adjusting for diabetes, which is a component of HSI, by assessing for potential collinearity between NAFLD and diabetes. The Phi coefficient was 0.32, indicative of a moderate positive association. Although it is conceivable that collinearity played a role in the alteration of the association between NAFLD and ACM, the strength of collinearity was not sufficiently robust to draw definitive conclusions.

NFS: Univariate analysis revealed that the NFS was associated with ACM, elevated risk of CVE and ESRD (Table 4). Following multivariable adjustment for demographics, smoking, baseline renal function and diabetes, an NFS ≥ score 0.676 remained associated with increased risk of all CVE (HR 1.19 [1.01–1.40], p = 0.0424), HF (HR 1.65 [1.36–2.01], p < 0.0001) and ACM (HR 1.31 [1.13–1.52], p = 0.0005) (Table 5, Fig. 2).

FIB-4 score: A high FIB-4 score was associated with ACM and all CVE including HF and CVA in univariate analysis, but these associations lost statistical significance following full multivariable adjustment (Table 4, Fig. 2).

In a sensitivity analysis in which eGFR was defined using creatinine alone, the NAFLD fibrosis score was no longer associated with an increased risk of CVE following multivariable adjustment; however, the findings were otherwise comparable (Additional file 12: Table S10). A further sensitivity analysis was performed where CKD was defined according to eGFR < 60 ml/min/1.73 m² alone, albuminuria ≥ 3 mg/mmol alone, or both (Additional file 13: Table S11). In all analyses, NAFLD remained associated with an increased incidence of CVE.