Clinical course of non-alcoholic fatty liver disease and the implications for clinical trial design

Abstract

Background and Aims:

The predicted risk and timeline to progression to liver outcomes in NAFLD are not well-characterized in the population. We aimed to examine the risk and time to progression to cirrhosis, hepatic decompensation and death in a contemporary population with long follow-up, in order to inform effectiveness endpoints and sample calculations in clinical trials of cirrhosis.

Methods:

This is a retrospective study of prospectively collected data in a medical record linkage system including all adults diagnosed with NAFLD between 1996-2016 by clinical, biochemical and radiological criteria in Olmsted County, Minnesota and followed until 2019. Liver-related outcomes and death were ascertained and validated by individual medical record review. Time and risk of progression from NAFLD to cirrhosis to decompensation and death were assessed using multistate modeling.

Results:

A total of 5123 NAFLD individuals (median age 52 years, 53% women) were followed for a median of 6.4 (range 1-23) years. The risk of progression was as follows: from NAFLD to cirrhosis: 3% in 15 years; compensated cirrhosis to first decompensation: 33% in 4 years (8%/year); first decompensation to two or more: 47% in 2 years. Albumin, bilirubin, non-bleeding esophageal varices and diabetes were independent predictors of decompensation. Among the 575 deaths, 6% were liver-related. Therapeutic trials in compensated cirrhosis would require minimum enrollment criteria of 2,886 subjects followed over 2 years to detect at least 15% relative decrease in liver endpoints.

Conclusion:

In this population-based cohort with 23 years of longitudinal follow-up, NAFLD was slowly progressive, with liver-related outcomes affecting only a small proportion of people. Reduction in liver related endpoints in clinical trials require large sample sizes and long follow-up.

Graphical Abstract

LAY SUMMARY

For patients with compensated NASH cirrhosis, the time spent in this state and the risk of progression to decompensation are not well-known in the population. We examined the clinical course of a large population-based cohort over 23 years of follow-up. We identified that adults with compensated cirrhosis spend a mean time of 4 years in this state and have a 10% per year risk of progression to cirrhosis or death. The risk is three-fold higher in adults with cirrhosis and one decompensation. These results are reflective of placebo arm risks in drug clinical trials and are essential in the estimation of adequate sample sizes.

INTRODUCTION

Due to the slowly progressive nature of NAFLD, studies designed to explore clinically meaningful liver outcomes of natural history are scarce; consequently, our understanding of how the disease unfolds longitudinally in the population remains limited. More commonly, research studies have used surrogate endpoints of disease progression, such as fibrosis stage or presence of NASH, usually collected through clinical trials or dedicated registries (1). However, there are limited data to link longitudinal changes in discrete histologic parameters or equivocal biomarker values to main liver outcomes with true impact on population health, such as development of cirrhosis, decompensation, liver transplantation or liver-related death. More recently, consensus on endpoints in precirrhotic and cirrhotic NASH has recognized the importance of liver related endpoints in phase III studies and all-cause mortality in phase IV trials (2). However, robust data on hepatic decompensation events which inform power calculations applicable to real-world scenarios are sparse. Studies of treatment effectiveness in the real world will be crucial and these will require inputs from real-world data on patient outcomes.

While efforts to examine the contemporary prevalence of the disease are increasing (3), the risk of disease progression and incidence of hard outcomes remain to be characterized in the population at large. A recent analysis of disease progression risk in 475 individuals with NAFLD (4) reported liver-related events in those with cirrhosis (19% at 2 years) and provided a more detailed exploration of longitudinal changes in histologic parameters and biomarker. Whereas the data provide further detailed clarification of disease progression, the study is limited by short follow-up of 2 years and lack of generalizability to individuals outside of a narrow disease spectrum (patients with F3 and compensated cirrhosis, enrolled in a clinical trial of simtuzumab vs placebo). In a registry-based cohort of NAFLD participants followed over 4 years, the incidence of liver-related events and mortality was higher in those with advanced disease (5) compared to those with stage 0-2 fibrosis. Whether the results are generalizable to those with NAFLD in the general population remains to be established. The expected risk and timing of transition from non-cirrhotic NAFLD to each of these distinct states represent highly relevant knowledge gaps, which would inform clinical management and trial design using therapies aimed to prevent the progression to cirrhosis or decompensating events(6).

Hence, in this work, we use a well-characterized cohort of individuals with NAFLD in the community with sufficient longitudinal follow-up to allow development of liver-related outcomes across the entire disease spectrum, with complete ascertainment of all lifetime medical events. We used a multistate modeling approach to characterize the clinical course of NAFLD by assessing the risk and time to progression from non-cirrhotic NAFLD to compensated and decompensated cirrhosis, using mortality as competing risk in each state along the clinical course. This work provides real-world estimates of disease progression which are critical inputs in power calculations for late phase clinical trials with hard endpoints.

METHODS

Study population

We reconstructed a cohort study of incident NAFLD cases between 1996-2016 in Olmsted County MN adult residents followed until 2019. The detection of incident cases was done using prospectively collected data in a medical record linkage system of the Rochester Epidemiology Project (REP). REP is an integrated electronic infrastructure corroborating the health data of all residents in Olmsted County, Minnesota, which includes all the medical providers in the county including: Mayo Clinic and its affiliated hospitals, Olmsted Medical Center and its affiliated hospital, and the Rochester Family Clinic(7, 8). All medical information from each of these providers for each individual resident of Olmsted County is linked to a single identification number allowing linkage across health care systems and facilitating the retrieval of corresponding patient record from each site. All data recorded during each medical encounter are available, including patient demographics, medical diagnoses and procedures, vitals, laboratory data, images, medication orders, as well as birth and death certificates.

Adults diagnosed with NAFLD were identified using a 3-step algorithm previously described(9): a code-based algorithm previously described(10), followed by natural language processing of abdominal imaging reports confirming presence of steatosis or cirrhosis, then internal validation by in-depth medical record review of a large subset of the cohort (Supplemental Methods and Figure 1). After exclusion of the false positive NAFLD cases, this rigorous multistep algorithm had an accuracy of >92%. Participants who were not Olmsted County residents on the date of NAFLD diagnosis and those without research authorization were also excluded.

Figure 1. Study flowchart and algorithm to validate NAFLD cases.

Participants were followed from study entry date (first NAFLD diagnosis date) until death, last date of follow up or study end-date (2019). Those who were diagnosed with another liver disease from the list of competing etiologies after study entry were accounted as NAFLD until the date of that diagnosis, when they were transitioned into a separate state of “other liver disease”. This was done to account for those who were followed as NAFLD for a period of time (years) and were later diagnosed with a superimposed liver disease, which is reflective of occasional scenarios in real-world clinical practice.

Study definitions

NAFLD was defined as: 1. presence of steatosis on biopsy/imaging in the absence of competing etiologies OR 2. cirrhosis with coexistent metabolic comorbidities (obesity or diabetes mellitus) in the absence of competing etiologies. The primary outcomes were development of cirrhosis, hepatic decompensation, liver transplantation and death.

Cirrhosis was identified using at least one code for cirrhosis or a decompensation event (i.e., ascites, encephalopathy, bleeding/non-bleeding esophageal varices and jaundice (Supplementary Table 2) or natural language processing of radiology reports containing keywords for cirrhosis (e.g., ‘cirrho’, ‘cirh’, ‘nodular’, and ‘nod’) at any point during the follow up. Additionally, those who did not meet the above criteria, but developed FIB-4≥2.67 during follow-up were identified and their medical records were reviewed for confirmation of possible unrecognized cirrhosis. All medical records meeting at least one of these criteria were individually reviewed and participants who did not have evidence of compensated or decompensated cirrhosis (even if FIB-4≥2.67 but no additional criteria) in the medical record were not ascertained as “cirrhosis”.

Compensated cirrhosis was confirmed if at least one of the following criteria were met: histologic evidence of stage 4 fibrosis; imaging evidence of cirrhotic liver morphology and/or portal hypertension (splenomegaly, portosystemic shunting, nonbleeding esophageal varices); increased liver stiffness concordant with cirrhosis (>5kPa on MRE(11) or >12kPa on transient elastography) in the absence of decompensation, which is defined below.

Decompensating events were defined as ascites (evidenced on imaging), esophageal variceal (EV) bleeding (proven by esophagogastroduodenoscopy), hepatic encephalopathy based on physician notes and jaundice (based on physician notes and total bilirubin >1.5 mg/dl). We did not include MELD-based criteria in this list due to frequent intermittent increases of MELD>15 due to events other than decompensation (infection, cholelithiasis, etc). Liver transplantation (LT) was identified during medical record review. Hepatocellular carcinoma was identified by ICD 9/10 codes and validated by chart review of imaging reports, clinical notes and, when available, histology.

Deaths and causes of death are recorded in the REP Death Browser on a quarterly basis using Electronic Death Certificate Date from the State of Minnesota. The data includes the full death certificate information plus ICD 9/10 coded causes of death, along with the coded underlying cause of death.

Statistical analysis

We used a multistate model (which is a type of competing risk analysis) to determine the risk of progression to the next advanced clinical stage, liver transplantation or death. In this 6-state model (Supplemental Figure 1), each NAFLD individual transitions unidirectionally from a state NAFLD without cirrhosis (state 1) to a state of compensated cirrhosis (state 2), first decompensation, HCC or LT (state 3) and two or more decompensations (different types), HCC or LT (state 4); transitions to a state of death (state 5) or other liver disease (state 6) can occur from each of the other four states. Transition to LT occurred from either state 2 or 3. At study entry, each subject starts in state 1, 2 or 3, according to the liver disease stage at NAFLD diagnosis. The model allowed transitions between non-consecutive states, based on the clinical status at the time of diagnosis of progression (1 to 3, 2 to 4, etc).The restricted mean survival time (the mean time in each state) was computed as the mean number of years the participants spent between the time of entry in the respective state until they transitioned out to another state (progression to any state of disease progression, death or other liver disease). This represents the number of years until diagnosis of the progressive state, not the actual time of onset (i.e. compensated cirrhosis can be clinically recognized sometime after the actual disease onset, which is asymptomatic). The risk of transition between states represents the probability of leaving the state into a progressive disease state, other liver disease or death after the mean time spent in that state. State 6 (other liver disease) represents all participants who were diagnosed with a competing liver disease etiology with impact on progression (eg. alcohol-related liver disease, etc) after being followed as NAFLD for a period of time. Transition to death from this state was not analyzed separately since it no longer reflects the natural history of NAFLD alone. Secondary analyses of the association between FIB-4 and elastography with progression to cirrhosis were performed using the same multistate modeling approach. The resulting progression probabilities were used to estimate the minimal sample size necessary to demonstrate a relative decrease in liver-related events and mortality in randomized clinical trials with investigative agents, using the population-based results as the placebo/control estimates. We used the online calculator for comparative clinical trials for time-to-event endpoints from the National Cancer Institute www.prevention.cancer.gov/power. The program computes required sample size, trial duration or statistical power for an accrual trial design based on the method by Rubinstein (12). The hazard rates were calculated using the formula [-In (1-probability)] / time, using the probabilities obtained from the multistate model described above.

Lastly, the NAFLD mortality risk in reference to the general population was examined by calculating the ratio of observed death rates in NAFLD to the expected death rates of the Minnesota general population age 50, using the %expsurv SAS.macro.

Statistical analyses were performed in SAS v9.4 (SAS Institute; Cary, NC) and R survival package, version 3.2.1 (R Foundation for Statistical Computing, Vienna). The study was approved by the Institutional Review Boards of Mayo Clinic and Olmsted Medical Center.

RESULTS

A total of 5123 individuals were diagnosed with NAFLD during the study timeframe. The median age at diagnosis was 52 years, 53% were women and 85% were Caucasian (Table 1). The median time of follow-up for the overall cohort was 6.4 (range 1-23) years. At the time of initial diagnosis, 96 (1.8%) had cirrhosis. The median follow-up of this group was 2.3 years.

Table 1.

Cohort characteristics at the time of NAFLD diagnosis

Characteristics	NAFLD cohort N= 5123
Age, range	52 (18-95)
Female sex, n (%)	2731 (53%)
Race, n (%)
Caucasian	4374 (85.3%)
African American	172 (3.4%)
Asian American	219 (4.3%)
Other	358 (6.9%)
Hispanic	323 (6.3%)
Body mass index, median (IQR)	33 (29-39)
Obese	69.4%
Cirrhosis	96 (1.8%)
Diabetes	1729 (34%)
Hypertension	2044(40%)
Dyslipidemia	3338 (65%)
Current smoking	570 (14%)
Total bilirubin (mg/dl), median (IQR)	0.5 (0.4, 0.7)
Creatinine (mg/dl), median (IQR)	1.0 (0.8, 1.1)
INR, median (IQR)	1.0 (0.9, 1.1)
Albumin (md/dl), median (IQR)	4.2 (3.9, 4.4)
AST (IU/ml), median (IQR)	31 (23, 44)
ALT (IU/ml), median (IQR)	40 (26, 66)
Platelets x109/ml, median (IQR)	244 (202, 291)

Overall, a total of 194 individuals were in a state of compensated (state 2) or decompensated (states 3 or 4) cirrhosis during the study timeframe, either at baseline/diagnosis or developed during follow-up. The most frequent decompensation event was ascites (n=100), followed by jaundice (n=60), hepatic encephalopathy (n=60) and variceal bleeding (n=22). A total of 575 individuals died.

NAFLD progression risk and time spent in each state

The risk of progression through NAFLD states or death during the study is shown in Figure 2.

Figure 2. Time-in-state and transition risk between the following disease states: NAFLD, compensated cirrhosis, first decompensation, two or more decompensating events/liver transplantation and death.

The number of years on the top bar represents the mean time spent in each state (15 years in pre cirrhotic NAFLD, 4 years in compensated cirrhosis, 2 years in cirrhosis with 1 decompensation, and 2 years in a state of 2 or more the compensating events. Individuals entered the pre cirrhotic NAFLD state at the time of diagnosis. Entry into the state of cirrhosis and decompensation occurred either as a transition from previous states or at the time of new diagnosis (top “New diagnosis” arrows) for those who were diagnosed in late stage NAFLD. The horizontal arrows represent cumulative risk of transition to any subsequent disease state after the mean time spent in that state (eg. 3% transitioned from NAFLD to a subsequent state of cirrhosis or decompensation/HCC/liver transplantation (either sequentially or by skipping an intermediate state) after 15 years. Lower vertical arrows illustrate cumulative risk of transition from each disease state directly to death. The results above results are from the multistate model.

NAFLD without cirrhosis (state 1):

The 5086 NAFLD individuals without cirrhosis at the time of diagnosis spent a mean time of 15 years in this state. Cumulatively, 96 (3.29%, 95%CI 2.57, 21%) individuals progressed towards subsequent states of cirrhosis/decompensation/LT and 408 (14.4%, 95%CI 12.82, 16.07%) died without progression to cirrhosis.

The risk of progression to cirrhosis and decompensation increased by FIB-4 strata at NAFLD diagnosis. In reference to those with FIB-4<1.3, the risk of NAFLD progression was higher for those with FIB-4 of 1.3-2.67 (HR=3.67, 95% CI 1.65-8.15, p=0.0014) and FIB-4>2.67 (HR=56.26, 95% CI 25.77-122.83, p<0.001). The risk of death was higher in those with FIB-4>2.67 (HR=3.26, p<0.001), but not in those in intermediate FIB-4 strata (HR=1.08, 95%CI 0.79-1.45, p=0.62). In multivariate analysis, predictors of progression to cirrhosis included age (HR=1.40 per decade, p<0.0009), DM (HR=3.0, p=0.003), BMI (HR=1.04, p=0.003), platelet count (HR=0.97, p<0.001) and albumin (HR=0.42, p=0.002).

Compensated NASH cirrhosis (state 2):

The 104 people diagnosed with compensated NASH cirrhosis at any time during the study (at study entry and during follow-up) spent a mean time of 4.1 years in this state. The 4-year cumulative incidence of progression to decompensation/HCC/LT was 32.7% (95%CI 24.44, 43.83%), whereas that of death was 8.5% (95% CI 4.30, 16.53%).

NASH cirrhosis with one decompensation (state 3):

The 87 individuals with NASH cirrhosis with one decompensation at any time during the study spent a mean of 2.3 years in this state. The 2-year cumulative incidence of progression towards more than 1 decompensation/HCC/LT was 48.1% (95%CI 27.15-59.32%), whereas the incidence of death was 16.9% (95% CI 11.23, 29.12%).

NASH cirrhosis with more than one decompensation (state 4):

The mean time in state of the 82 participants was 2.1 years. The 2-year cumulative incidence of death was 45.8% (95%CI 35.62, 58.79%).

Predictors of decompensation events in NASH cirrhosis

Univariate Cox regression analysis using age, sex and clinically relevant time-dependent covariates (Table 2) indicated that albumin, total bilirubin and presence of non-bleeding esophageal varices were independently associated with future risk of decompensation. A total of 44 incident decompensation events occurred in individuals with compensated cirrhosis with complete data available for all covariates. Presence of diabetes mellitus (DM) was associated with decompensation risk but did not meet statistical significance (p=0.08). However, due to known clinically significant association with risk of NAFLD progression, diabetes was included in the multivariate Cox regression model. In the final multivariable model, the levels of albumin and bilirubin, as well as the presence of non-bleeding EVs and DM were independent predictors of decompensation, with a predictive performance (c-statistic) of 0.75.

Table 2.

Predictors of decompensation in NASH cirrhosis

	Univariate regression model		Multivariable regression model*
Co-variate (unit)	HR (95%CI)	p-value	HR (95%CI)	p-value
Age (decade)	1.19 (0.95, 1.49)	0.12
Female sex	1.37 (0.77, 2.45)	0.28
BMI (1kg/m2)	1.02 (0.99, 1.06)	0.22
Albumin (1mg/dl)	0.27 (0.18, 0.43)	<0.001	0.26 (0.16, 0.42)	<0.001
Platelets (109)	1.02 (0.82, 1.27)	0.83
Total bilirubin (1 mg/dl)	1.35 (1.12, 1.63)	0.002	1.27 (1.00, 1.61)	0.04
INR	1.07 (0.65, 1.75)	0.78
Non-bleeding varices	3.52 (2.02, 6.14)	<0.001	3.4 (1.96, 5.94)	<0.001
Creatinine (0.1 mg/dl)	1.06 (0.99, 1.15)	0.10
Diabetes mellitus	2.32 (0.92, 5.84)	0.08	3.2 (1.06, 9.60)	0.03

^*When testing the independent role of MELD-Na in decompensation prediction in the above multivariate model replacing “total bilirubin” with “MELD-Na”, the coefficient was 1.02 (95% CI 0.96-1.09), p value=0.44 (log rank test).

Probability of disease states and death 20 years after NAFLD diagnosis

Figure 3 illustrates the proportion of individuals in each of the following states every year following NAFLD diagnosis: alive with non-cirrhosis NAFLD, alive with NASH cirrhosis/LT, death after passing through cirrhosis state, death without passing through cirrhosis state and other liver disease diagnosed after NAFLD (includes those who died after entering this state).

Figure 3. Distribution of outcomes up to 20 years after NAFLD diagnosis.

The colored areas represent the “fate” of participants at each two-year intervals after diagnosis: the proportion of individuals in each of the states of NAFLD alive (orange), alive with cirrhosis/liver transplant (purple), dead after cirrhosis (green), dead without cirrhosis (pink) and those who were diagnosed with other liver disease after NAFLD (including deaths in this state). The proportions of individuals in each state add to 100% at each time interval. At any time frame the proportion of those alive with cirrhosis in the population is 1%. Cumulatively, at 20 years, 50% were alive with NAFLD, 1.1% alive with NASH cirrhosis/LT, 26% were alive or dead after subsequent diagnosis with another liver disease in addition to NAFLD, 19.8% died without cirrhosis and 2.4% died after NASH cirrhosis/LT. The results above are from the multistate model.

Several important points can be observed based on the results of this multi-state model. First, at any time point during the lifetime of individuals with NAFLD, those who were alive with cirrhosis/LT represent approximately 1% of the entire NAFLD population (in purple). With time, the proportion of those alive with NAFLD decreased, while those who died increased, but the probabilities of transitions into cirrhosis/LT (from NAFLD) and out of cirrhosis/LT (into death) were small and balanced throughout the follow-up.

Second, 20 years after NAFLD diagnosis, the overall probability of death in this population was 22%. The majority of deaths (20% of individuals) occurred without going through a state of cirrhosis (pink). The most common causes of death were cancer (26%) and cardiovascular disease (20%), whereas liver-related death represented 6% of cases (S Figure 3). In reference to the expected (E) Minnesota death rates for a population age 50, those with NAFLD had a higher observed (O) mortality: O/E (95% CI): 1.41 (1.31-1.52), p<0.001 (Figure 4A). In reference to the general population, the observed mortality was proportionally higher in women: O/E (95% CI): 1.52 (1.37-1.69) than men: O/E (95% CI): 1.30 (1.16-1.45), p-value women versus men=0.042 (Figure 4B).

Figure 4. Survival probability in NAFLD (black) compared to expected survival for the Minnesota age-matched population (red).

A. Overall. In reference to the expected (E) Minnesota death rates for a population age 50, those with NAFLD had a higher observed (O) mortality: O/E (95% CI): 1.41 (1.31-1.52), p<0.001 (log-rank test)

B. Men and women. In reference to the general population, the observed mortality was proportionally higher in women: O/E (95% CI): 1.52 (1.37-1.69) than men: O/E (95%CI): 1.30 (1.16-1.45), p-value women versus men=0.042 (log-rank test)

Lastly, 20 years after NAFLD diagnosis, the predicted states of the individuals in this cohort were as follows: 50% alive with NAFLD, 1.1% alive with NASH cirrhosis/LT, 26% were alive or dead after subsequent diagnosis with another liver disease in addition to NAFLD, 20% died without cirrhosis and 2.4% died after NASH cirrhosis/LT.

Sample size estimation for randomized clinical trials with liver-related and mortality endpoints

Table 3 shows the estimated sample size for randomized clinical trials with investigative agents using the progression risk to liver outcomes or death. For illustration purposes, we chose an arbitrary study time of 5 years, consistent with late phase studies, with 3 years accrual time and 2 years of follow-up time. However, the study duration and loss to follow-up estimates can be modified in the online calculator, while the hazard ratios remain constant. Studies including compensated cirrhosis would require enrollment of at least 3,522 subjects followed over at least 2 years to detect relative decrease in liver endpoints of at least 15% (Table 3A). When the study design includes mortality as a composite endpoint (Table 3B), the sample size decreases accordingly, with at least 2,886 participants for the same relative therapeutic effect.

Table 3. Sample size estimation for randomized clinical trials in NAFLD for specific outcomes.

The numbers were obtained by inserting the progression risk from Figure 2 into the online calculator http: prevention.cancer.gov/power. The program assumes that trial follow-up data will be analyzed with the Mantel-Haenszel (log rank) test for comparing survival curves. For all cases we used a hypothetical trial with 3 years accrual plus 2 years of follow-up, 0.80 power, with no loss to follow-up.

A. Liver-related events only.
NAFLD state		Compensated cirrhosis	Decompensated cirrhosis
Outcomes (any of)		Hepatic decompensation, liver transplantation	Liver transplantation
Probability of events		33% in 4 years	48% in 2 years
Sample size based on anticipated relative decrease in incidence in the intervention group versus placebo	15%	3522	1602
	25%	1074	498
	35%	498	234
B. Composite outcome of liver-related events or death
NAFLD state		Compensated cirrhosis	Decompensated cirrhosis
Outcomes (any of)		Hepatic decompensation, liver transplantation, death	Liver transplantation, death
Probability of events		42% in 4 years	65% in 2 years
Sample size based on anticipated decrease in incidence in the intervention group versus placebo	15%	2886	1356
	25%	882	420
	35%	408	198

DISCUSSION

In this large real-world NAFLD cohort study with longitudinally collected health care data over 23 years of follow-up, we use multistate modeling to make several key observations in the clinical course of NAFLD, related to disease progression risk, mortality, time spent in main disease states and predictors of decompensation. These results are critical inputs of the outcomes in the placebo group in estimating the sample size of late phase clinical trials in NASH cirrhosis.

In NAFLD without cirrhosis, progression to cirrhosis and associated complications is slow, with only 3% of patients progressing over 15 years. In this timeframe, the risk of death from non-liver related causes is much higher (14%). It is important to note that these estimates are likely to underestimate progression to compensated cirrhosis due to lack of symptoms and systematic screening in the community. Moreover, these estimates apply to all encompassing pre-cirrhotic stages of fibrosis because real-world clinical practice lacks routine systematic assessment of liver fibrosis at pre-defined time intervals to detect the exact time of progression from stage to stage. Hence, these results should be interpreted in the appropriate context and applied to the general state of “NAFLD before cirrhosis”, with the understanding that most of these individuals have minimal to no fibrosis, typical of a general population outside of clinical trials and registries. These results provide population-based estimates that are critical for modeling the benefit of cirrhosis screening, and for patient education and counseling at the time of NAFLD diagnosis.

Once compensated cirrhosis stage is reached, individuals with NAFLD spend approximately 4 years in this state. During this timeframe, the risk of further progression to liver-related events is higher than that of death without decompensation (8% versus 2% per year). These data are of critical importance when designing clinical trials targeting patients with compensated cirrhosis. As treatment endpoints for patients with advanced fibrosis/cirrhosis evolve towards clinical outcomes of decompensation(13), these real-world results can serve as estimates of placebo arm outcomes (Table 3). These results are in line with the published literature (14) (4, 5) and provide the complementary perspective of the clinical course of the disease in the population when followed for up to 23 years, more enriched with outcomes. Unlike registry trials(5), ascites was the most common decompensation. It is worth noting that the actual time spent in compensated cirrhosis state is likely longer than 4 years, when considering the additional time from the actual onset of cirrhosis until diagnosis, which remains unknown. Robust analyses of the subset of patients who developed incident cirrhosis during study timeframe (after excluding those with compensated cirrhosis at the time of study entry) was limited by the small number of patients in this subset (60 adults, of whom 21 progressed to decompensation and 4 died). However, in the absence of universal systematic screening for cirrhosis in NAFLD, this is reflective of both real-world scenarios and clinical trials, and the results are applicable to the prognostication of their clinical trajectory from that point forward.

The main risk factors for decompensation are presence of diabetes mellitus and nonbleeding esophageal varices, each increasing the risk 3-fold. Hence, these two parameters should be used for stratification of expected outcomes in clinical trial design. Although the development of a risk score was not the major goal of this study, the model including the main variables associated with decompensation (albumin, bilirubin, presence of esophageal varices and diabetes) showed a comparable performance (c-statistic 0.75) to other predictive models using similar parameters, further validating the impact of these variables in NAFLD prognostication and supporting their use in participant stratification clinical trials (4) (15-19).

After the first decompensating event, progression to further decompensation or need for LT occurs in almost half of individuals within 2 years. Hence, first decompensation should prompt enhanced vigilance and close surveillance to prevent the high morbidity and health care burden associated with hospitalizations and emergency visits. This is a critical practice gap and ripe zone for clinical practice optimization models. Onset of two or more decompensating events is associated with a 46% probability of death in the subsequent 2 years.

These population-based estimates are vital for the design of late phase clinical trials in NASH cirrhosis, as they provide placebo arm outcome risk in the real-world, as opposed to those assessed from registry-based data, and it would be illustrative of expected impact when used in the hepatology practice after approval. Trials targeting compensated cirrhosis should include between 498 and 3,522 participants over 3 years to show effectiveness in reducing liver-related outcomes (using death as competing risk) in 5 years, depending on drug potency (Table 3A). Including mortality in the composite endpoints lowers the necessary sample size (Table 3B).

The multi state model illustrated in Figure 3, which is a more informative competing risk analysis model for NAFLD progression, provides an informative cross-sectional overview of the different disease states at various times throughout the natural history of NAFLD in the population, starting at the time of diagnosis and following them for up to 20 years afterwards. At any point in time, the proportion of NAFLD cases that were alive and carried a diagnosis of cirrhosis (compensated or decompensated) is ~1% (Supplementary Figure 3). This proportion is maintained relatively constant by the equal entry-exit swing of the new cirrhosis diagnoses and deaths occurring in this state. It is likely that implementation of systematic screening for cirrhosis in NAFLD would increase the proportion of known cirrhosis cases. Whether this would result in decreased mortality risk and/or morbidity and healthcare expenditure is an area of critical importance for future research.

The 20-year cumulative incidence of death was approximately 22% in this NAFLD population (20% without cirrhosis and 2% with cirrhosis). Among those who died, the major causes were related to malignancy (26%) and cardiovascular disease (20%), whereas liver-related causes represented 6% of the total. These findings are in-line with several previous epidemiological studies and underline the importance of cancer screening and cardiovascular disease prevention in this population(20-22). Despite increasing awareness and management insight in the past few decades, NAFLD continues to remain associated with a higher mortality than the general population, with a disproportionally higher mortality in women(23).

Strengths and Limitations

The strengths include the large population-based cohort with complete enumeration of all individuals diagnosed with NAFLD, which avoids sampling bias of volunteers for registry-based studies or spectrum bias of patients with advanced disease from referral centers. The follow-up was sufficiently long to capture hard outcomes such as disease progression and mortality in a database that captures all events occurring during the life-time of a population. Case ascertainment was confirmed by thorough abstraction of medical records, using pre-specified definition criteria, both in the non-cirrhotic NAFLD and 100% of cases identified as possible cirrhosis. All decompensation outcomes were individually reviewed and confirmed. Hence, although the initial search used administrative codes, the reliability and validity were ensured by individual chart review, minimizing the inherent limitations of code-based algorithms. The multistate modeling is the ideal approach to capture transitions through clinically-relevant states and account for competing scenarios such as death or subsequent diagnosis of an alternative liver disease. We did not separate the state of compensated cirrhosis into further subclasses such as mild portal hypertension or varices without bleeding. These states have different risks of bleeding and decompensation (24). However, outside of a prospective clinical trial in which all patients undergo screening or surveillance under strict schedule, it was not feasible to apply this to a large population database, in which the practice and screening schedule varies among different healthcare providers, akin to real-world practice.

Inherent to retrospective cohort studies in the population, there are several limitations to consider in data interpretation. There is a risk of selection bias in all studies of NAFLD that is clinically identified rather than by systematic screening. However, these data are still valuable to the treating physicians who need to assess a patient’s risk of progression once diagnosed with NAFLD. Measurement bias of diagnostic suspicion may lead to overestimation of the risk of cirrhosis or death if identification of NAFLD was due to apparent clinical features. Similarly, diagnosis of cirrhosis may be underestimated in light of absent systematic screening in this population during the study timeframe. However, a large proportion of NAFLD epidemiological research in large populations is affected by this bias (those enrolling in registries such as NHANES are a sample of the entire population, big data mining of large databases relying on ICD code case-ascertainment such as the Veterans Affairs or insurance-based registries) because systematic screening strategies in the community are lacking. We believe that this bias risk is minimized in a cohort using a medical-record linkage system because the database uses the full medical history including multiple providers for care and outcome assessment. There was a limited number of elastography tests or liver biopsies to allow robust and detailed stratification of risk progression to cirrhosis based on liver stiffness or histology. The data should be generalized to the population reflective of the Midwest, which has a higher proportion of Whites than the US population (84% vs 76%) and a lower proportion of Hispanics, but similar to the majority of current clinical trials. However, a large dataset with granular health history amenable to chart review and internal validation, such as REP, but also perfectly representative of the entire US population is improbable. The power of observational data to inform public health research lies in its effective use of results on the population group that is studied, if that group is well represented in the disease(25). Whites continue to have a large representation among those affected by NAFLD. Hence, understanding disease progression from large observational data in this group should not be dismissed because randomized trials are not always feasible or timely. In a recent meta-analysis included six studies and noted that racial disparities overall played a smaller role in NAFLD prognosis compared with prevalence(26).

In summary, NAFLD is a slowly progressive disease with liver-related outcomes affecting a small proportion of people but with mortality rates than remain higher than the general population despite several advances in the awareness, diagnosis, and management of this disease. Randomized clinical trials should be designed with an adequately estimated sample size and study duration to enhance the opportunity to demonstrate drug efficacy.

• Population-derived estimates of progression risk to cirrhosis and decompensation in NAFLD are essential in estimating the sample size for clinical trials of therapeutic interventions

• Among adults with compensated NASH cirrhosis, the risk of progression to decompensation or death is 10% per year

• Among adults with first decompensation, the risk of progression to subsequent liver events of death is 32% per year

• Therapeutic trials in compensated cirrhosis would require a minimum of 2,886 participants followed for at least 2 years to detect a minimum of 15% relative decrease in liver-related endpoints

Abbreviations:

DM

diabetes mellitus

EV

esophageal varices

HCC

hepatocellular carcinoma

LT

liver transplantation

NAFLD

nonalcoholic fatty liver disease

REP

Rochester Epidemiology Project

Data availability Statement:

Data can be available upon reasonable request to the Corresponding Author.