Extrahepatic disease clusters and mortality in people with steatotic liver diseases: a prospective analysis of 64,749 females and 113,587 males in the UK Biobank | BMC Medicine

We included 178,336 individuals with SLD in analysis (36.3% females, mean age 57.3 years; 73.5% MASLD, 19.0% MetALD, 6.4% ALD) (Fig. 1). About one in five (20.2%) of SLD participants had MM, higher in females than in males (24.4% vs. 17.8%). The prevalence of MM was 21.3%, 15.9% and 17.4% in people with MASLD, MetALD and ALD, respectively. Overall, compared to people without MM, people with MM were more likely to be older, living in socioeconomically deprived area, less educated, smokers and drink less. They were also more likely to have diabetes and low HDL cholesterol, but less likely to have hypertension and high TG (Table 1).

Disease clusters and their characteristics

Latent class analysis derived five distinct disease clusters for males and females, respectively, with some overlapping and some sex-specific patterns (Additional file 1: Fig. S1(A), Table S1). For males, although the clustering evaluation metrics indicated that an 8-cluster solution was statistically optimal, we opted against this model due to concerns on clinical redundancy and potential overfitting. Upon review, the 8-cluster solution included two highly similar respiratory clusters (one characterised by asthma and COPD, the other by asthma and other chronic respiratory disease), as well as two overlapping cardiovascular clusters (one with ischaemic heart disease, heart failure and arrhythmia, and another with ischaemic heart disease, heart failure and heart valve disease). A similar issue of cluster overlap was also observed in the 6-cluster solution. Therefore, to balance clinical interpretability and parsimony, we selected the 5-cluster model for subsequent analyses.

Table 2 summarises the characteristic diseases within each cluster. In both sexes, a respiratory cluster (characterised by asthma, COPD and other chronic respiratory diseases), a mental health cluster (dominated by depression and anxiety, with substance use disorder additionally present in males) and a cancer/osteoarthritis cluster (including solid organ cancers, chronic respiratory diseases and osteoarthritis) were observed. Among males, a heart cluster (characterised by ischaemic heart disease, arrhythmia and heart failure) and a stroke cluster (characterised by stroke and paralysis) were identified. In females, these two clusters merged into a single heart/stroke cluster, which included ischaemic heart disease, arrhythmia and stroke. Additionally, females had a unique thyroid cluster, consisting of thyroid disorders and connective tissue diseases. All these characteristic diseases showed high exclusivity (> 20%) and most showed high E/O ratio (> 2.0). Additional file 1: Table S2 shows more details on the LTCs in each cluster.

In males, the most common disease cluster to which individuals were assigned was the cancer/osteoarthritis cluster (30.0%), followed by respiratory cluster (27.7%), heart cluster (20.3%) and mental cluster (13.9%) while stroke cluster (8.1%) was the least common. In females, the most common cluster was respiratory cluster (26.0%), followed by mental cluster (21.9%), cancer/osteoarthritis cluster (21.5%) and thyroid cluster (19.6%), and the least common was heart/stroke cluster (11.0%) (Table 2). Across SLD subtypes, mental health cluster was more prevalent in ALD in both males and females, while heart/stroke and thyroid clusters were more prevalent in MASLD in females (Additional file 1: Fig. S2).

Additional file 1: Tables S3 and S4 present the baseline characteristics of males and females assigned to each disease cluster. In both males and females, clustering was associated with age, socioeconomic status, physical activity and CMRFs. Individuals in the respiratory and mental health clusters were younger, while those in the heart and/or stroke cluster were the oldest. Both sex-specific mental health and stroke clusters were more socioeconomically deprived, and the stroke clusters had lower level of physical activity. Both the stroke and heart clusters had higher prevalence of diabetes and low HDL than overall males. The heart cluster and stroke cluster also demonstrated differences in socioeconomic deprivation, physical activity level, diabetes, hypertension and low HDL cholesterol.

In females, the mental health and heart/stroke clusters were more likely to live in socioeconomically deprived area, with the heart/stroke cluster also less likely to have higher education, be never smokers or engage in high physical activity level. While diabetes was less common in the mental health cluster, it was more prevalent in the heart/stroke cluster, which also had higher TG and lower HDL levels. Mental health cluster was also more likely to be White ethnicity in females.

Disease clusters and all-cause mortality

During a median follow-up of 13.8 years, 14,595 (12.8%) and 6171 (9.5%) deaths were captured in males and females, respectively. Compared to males with SLD but without MM, males with MM had an excess mortality rate of 12.3/1000 person-years (20.0 vs. 7.7), yielding an HR estimate of 2.00 (95%CI 1.93, 2.08). Compared to females without MM, females with MM had an excess mortality rate of 6.0/1000 person-years (11.7 vs. 5.7), responding to an HR estimate of 1.80 (95%CI 1.71, 1.90).

People in all disease clusters showed higher all-cause mortality than people without MM. In males, heart and stroke clusters were associated with the highest mortality, with HR of 2.63 (2.48, 2.78) and 2.36 (2.16, 2.58); respiratory, mental health and cancer/osteoarthritis clusters showed HR of 1.62 (1.51, 1.73), 1.84 (1.69, 2.00) and 1.85 (1.75, 1.96), respectively. In females, heart/stroke cluster was associated with the highest mortality, with HR of 2.90 (2.64, 3.20). Respiratory, mental health, cancer/osteoarthritis and thyroid clusters showed HR of 1.73 (1.58, 1.89), 1.57 (1.42, 1.74), 1.85 (1.69, 2.02) and 1.42 (1.28, 1.58), respectively (Fig. 2, Additional file 1: Table S5). Sensitivity analyses by additional adjustment for CMRFs and removing the first 2 years of follow-up generated similar results (Additional file 1: Table S6).

Disease clusters and cause-specific mortality

Examining causes of death, 65% of all deaths were due to extrahepatic cancers, CVD and liver-related diseases, and the remaining 35% were due to other causes. More specifically, in males, 31.9% and 31.8% deaths were attributed to extrahepatic cancers and CVD, while only 1.1% to HCC and 4.4% to liver-related diseases. Similarly in females, 39.9%, 21.9%, 0.3% and 3.3% deaths were attributed to extrahepatic cancers, CVD, HCC and liver-related diseases, respectively. Extrahepatic cancer and CVD were the major causes of death across all clusters, contributing 58.1% to 67.3% of all deaths in males and 59.1% to 66.7% in females. CVD was the biggest cause of death in the stroke and heart clusters in males and in the stroke/heart cluster in females, while extrahepatic cancers remained the biggest cause of death for all other clusters (Additional file 1: Fig. S3).

Compared to people without MM, having MM was associated with higher mortality from extrahepatic cancers by 45% (1.45 (1.37, 1.54)), CVD by 173% (2.73 (2.56, 2.92)), HCC by 67% (1.67 (1.20, 2.32)) and liver-related diseases by 84% (1.84 (1.55, 2.18)) in males and extrahepatic cancers by 40% (1.40 (1.30, 1.52)), CVD by 130% (2.30 (2.04, 2.58)) and liver-related death by 46% (1.41 (1.06, 1.86)) in females. All five clusters were associated with increased mortality of extrahepatic cancers, with the highest risk in cancer clusters in males (1.77 (1.62, 1.92)) and females (1.74 (1.53, 1.98)). All five clusters were associated with mortality from CVD, with the highest risk in stroke and heart clusters in males (3.34 (2.86, 3.90) and 5.28 (4.83, 5.77)) and stroke/heart cluster (5.66 (4.75, 6.75)) in females. The disease clusters also showed general trend of positive associations with mortality of HCC and liver-related diseases, although non-significantly, likely due to the small number of events in these clusters (Table 3).

Stroke cluster males also had a lower mortality rate (26.9 vs. 30.1 per 1000 person-year) than heart cluster (Additional file 1: Table S5), with lower mortality of CVD but higher mortality of cancer compared to the heart cluster (Additional file 1: Fig. S3).

Cluster validation

To assess the clustering stability across subsamples, we applied LCA to randomly selected 80% and 50% subsets of the full sample and compared the results with the primary full-sample analysis. We observed that the aBIC plots demonstrated similar trends across all samples, and the optimal number of clusters remains unchanged at 5 when applying all three model selection criteria (Additional file 1: Fig. S1(B, C)). The disease profiles of the derived clusters were highly consistent with those identified in the full-sample analysis (Additional file 1: Table S7). The distributions of posterior probability for each cluster were also similar across the three sample analyses (Additional file 1: Table S8). Comparing the full and 80% sample analyses, 99.0% males and 98.2% females were assigned to the same clusters; compared the full and 50% sample analyses, 96.3% males and 96.2% females remained in the same clusters. These results supported the robustness and reproducibility of the clustering solution.

To evaluate the impact of individuals with low posterior probability on the association estimates, we conducted a sensitivity analysis excluding participants with a maximum posterior probability < 70% for cluster assignment. This resulted in the exclusion of 5995 (29.7%) males and 3511 (22.2%) females (Additional file 1: Table S9). Additional file 1: Table S10 shows the associations between disease clusters and all-cause and cause-specific mortality outcomes, which were largely consistent with those observed in the primary analysis, further supporting the stability of clustering assignments.