Disease state transition probabilities across the spectrum of NAFLD: A systematic review and meta-analysis of paired biopsy or imaging studies

Abstract

Background & Aims:

We conducted a meta-analysis to summarize rates of progression to and regression of nonalcoholic fatty liver (NAFL), nonalcoholic steatohepatitis (NASH), and fibrosis in adults with nonalcoholic fatty liver disease (NAFLD).

Methods:

We searched PubMed/Medline and 4 other databases from 1985 through 2020. We included observational studies and randomized controlled trials (RCTs) in any language which used liver biopsy or imaging to diagnose NAFLD in adults with a follow-up period ≥48 weeks. Rates were calculated as incident cases/100 person-years and pooled using the random-effects Poisson distribution model. Heterogeneity was assessed using the I² statistic.

Results:

We screened 9,744 articles and included 54 studies involving 26,738 patients. Among observational studies, 20% of healthy adults developed NAFL (incidence rate=4.8/100 person-years) while 21% of people with fatty liver had resolution of NAFL (incidence rate=2.4/100 person-years) after a median of about 4.5 years. In addition, 31% of patients developed NASH after 4.7 years (incidence rate=7.4/100 person-years) whereas in 29% of those with NASH, resolution occurred after a median of 3.5 years (incidence rate=5.1/100 person-years). Time to progress by 1 fibrosis stage was 9.9, 10.3, 13.3 and 22.2 years for F0, F1, F2, and F3, respectively. Time to regress by 1 stage was 21.3, 12.5, 20.4, and 40.0 years for F4, F3, F2, and F1, respectively. Rates estimated from RCTs were higher than those from observational studies.

Conclusions:

In our meta-analysis, progression to NASH was more common than regression from NASH. Rates of fibrosis progression were similar across baseline stage, but patients with advanced fibrosis are more likely to regress than those with mild fibrosis.

Graphical Abstract

INTRODUCTION

Globally, one in four adults has non-alcoholic fatty liver disease (NAFLD), defined as the presence of ≥5% hepatic fat content without secondary causes such as excessive alcohol use, viral liver diseases, or drug use.¹ Most patients have simple steatosis, or non-alcoholic fatty liver (NAFL). About 20% have non-alcoholic steatohepatitis (NASH), a more progressive condition that can lead to cirrhosis and advanced liver-related complications.² Compared to patients with NAFL, those with NASH have more than 10 times higher incidence of hepatocellular carcinoma and mortality.¹ With the rising prevalence of obesity and diabetes—both risk factors for NAFLD—NAFLD/NASH is expected to be the leading cause of liver transplant in the US in the coming decade.³

Understanding the natural process of progression and regression of NAFLD is important for early identification of high-risk patients and appropriate tailoring of disease prevention and control interventions. Longitudinal cohort studies have provided initial insights regarding the evolution of NAFLD over time.^4–6 Previous systematic reviews demonstrated that patients with NASH progress more rapidly to advanced fibrosis than do those without.^2,7 Also, little is known about the incidence of NAFL in otherwise healthy individuals or the rate of development of NASH from NAFL. Recently, a number of observational studies and randomized controlled trials (RCTs) of NAFLD treatments have been conducted and provide valuable information on the natural history of NAFLD.

Mathematical models including decision modeling are used to simulate the impact of different interventions on disease development and health outcomes to evaluate the comparative long-term benefits and cost-effectiveness of screening and treatment. Estimates of progression and regression rates for each stage must be incorporated to ensure accurate model projections.^8,9 However, previous meta-analyses calculated fibrosis progression rates by subtracting the number of stages progressed by the number of stages regressed.^2,7 To address these limitations, we conducted a systematic review and meta-analysis to describe the natural history of NAFLD, including both the progression to and regression of NAFL, NASH, and fibrosis. These data will help patient counseling, disease monitoring, and microsimulation modeling.

MATERIALS AND METHODS

We conducted this meta-analysis following the updated Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement.¹⁰

Search strategy

A comprehensive search strategy with no language restriction was constructed to capture the relevant literature describing the natural history of NAFLD. The following databases were searched from 1985 through December 9, 2020: Ovid MEDLINE®, Ovid Embase, Scopus, Wiley’s Cochrane Central Register of Controlled Trials, and Web of Science’s Science Citation Index. We also manually searched for additional studies using reference lists from eligible studies and systematic reviews. The complete search strategies are provided in the Supplementary Methods and Appendix Table 1.

Selection criteria

We included studies that: 1) were RCTs or cohort studies; 2) included adult populations (≥18 years); 3) used liver biopsy and/or imaging to diagnose NAFLD at baseline and follow-up; 4) contained data to examine the progression to and/or regression of NAFL, NASH, or fibrosis; and 5) had a follow-up period of at least 48 weeks. We excluded editorials, letters to the editor, conference abstracts, cross-sectional or case-control studies, studies that combined adult and pediatric cohorts, used noninvasive means other than imaging to diagnose NAFLD, had insufficient data to estimate outcomes, only reported mean change of fibrosis stage for the entire cohort, and studies in which patients with other potential causes of fatty liver such as viral hepatitis and alcohol-related liver disease were not excluded. For RCTs, we included only patients in the placebo arm. We also excluded patients if they received intensive lifestyle modifications that could improve hepatic steatosis and other histological findings of NAFLD. When there were multiple studies using the same patient cohort, we included those reporting the most comprehensive data or having the longest follow-up period.

Data extraction

Titles and abstracts of identified studies were reviewed independently by J.Y.P. and L.Z. Full-texts were examined and pre-determined inclusion and exclusion criteria applied. Disagreement was resolved by discussion with a third investigator (P.L.).

Data extraction was also performed independently by J.Y.P and L.Z. using standardized form; P.L. resolved discrepancies. The following was extracted: 1) study characteristics: first author’s name, year of publication, geographic location of the study population, funding source, study design, data collection period, and data source; 2) patient characteristics: sample size, mean age, percent female, mean body mass index, percent with obesity, diabetes, hypertension, metabolic syndrome, and hyperlipidemia; 3) disease progression characteristics: duration of follow-up, diagnostic methods (liver biopsy or imaging), histologic staging criteria (for studies using liver biopsy) or cut-off values (for studies using imaging), attrition rate, proportion of patients having NAFL, NASH, and fibrosis by stage at baseline and follow-up, and proportion of patients whose disease progressed, regressed, or remained unchanged.

Outcomes

Outcomes included rates of progression to or regression of NAFL, NASH, and fibrosis, measured by pooled incidence rates. Progression to or regression of NAFL was based on a priori author-defined criteria (Appendix Table 2). Progression to NASH was defined as developing definite NASH for those with no or borderline NASH at baseline. Regression of NASH was defined as resolution of NASH to borderline or no NASH. Borderline NASH and NASH were based on author-defined criteria (Appendix Table 3). We considered patients to have NASH progression or regression only if development or resolution of NASH was explicitly stated. Fibrosis progression was defined as an increase of ≥1 point and regression as a reduction of ≥1 point on fibrosis score. Among studies using liver biopsies, we used studyspecific fibrosis scores, based on either NASH Clinical Research Network (CRN) or Brunt scoring systems.¹¹ For one study that used METAVIR scoring system¹² we converted its 7 stages into 5— equivalent to NASH CRN and Brunt systems for uniform inclusion in the meta-analysis—as follows: METAVIR stage 0 was stage 0, METAVIR stage 1 or 2 was stage 1, METAVIR stage 3 was stage 2, METAVIR stage 4 or 5 was stage 3, and METAVIR stage 6 was stage 4. For imaging studies, progression/regression were based on author-defined cut-offs.

Assessment of risk of bias

Risk of bias of in the studies that were included was identified using the most recent Cochrane tools for non-randomized studies¹³ and randomized trials.¹⁴ Details are described in Supplementary Methods.

Statistical analysis

We reported incidences for RCTs and observational studies separately. Incidences of progression and regression were calculated as number of incident cases divided by person-years. If studies did not report person-years, we calculated them as the total number of patients multiplied by mean follow-up duration (in years), accounting only for patients who had outcome information at follow-up. For fibrosis, we reported incidences based on both cases and stages, stratified by baseline fibrosis stage and NASH versus NAFL. Incident cases equaled number of patients with fibrosis progressing/regressing at follow-up. Incident stages were calculated as number of stages changed at follow-up. Incidences and 95% confidence intervals (CI) were then presented as cases (or stages) per 100 person-years and pooled using the random-effects Poisson distribution model.¹⁵ The years required to progress/regress by one stage was calculated as the reciprocal of incidence. Heterogeneity between studies was assessed using the I² statistic, with <33%, 33–66%, ≥67% representing low, moderate, and high heterogeneity, respectively. We also conducted pre-planned subgroup analyses by geographic location (country- or continent-specific), diagnostic method (liver biopsy vs. imaging), and length of follow-up (<5 years vs. ≥5 years). We constructed funnel plots to assess publication bias. A p-value of <0.05 was considered statistically significant. All analyses were conducted in R v.4.1.2 (Vienna, Austria) using metafor package.

RESULTS

Study identification

A total of 14,387 publications were imported into Covidence^™ software (Melbourne, Australia); 4,644 exact duplicates were removed. We identified one additional study through manual search,¹⁶ making a total of 9,744 studies for title and abstract screening. Full-texts of 213 studies were retrieved and assessed for eligibility. Of included studies, we contacted the corresponding authors of 5 studies that reported the overall distribution of fibrosis by stage at baseline and/or follow-up without details on the individual patient change (Appendix Table 4). Two authors responded; one provided usable data.¹⁷ The remaining four studies were, therefore, excluded. We contacted the corresponding authors of 11 additional studies for further information on the progression and regression of fibrosis but received either no response or no usable data.^18–28 However, these 11 studies were eligible due to having information on NAFL and/or NASH. Therefore, full-text review resulted in 60 eligible studies. We further excluded six studies due to repeated use of the same data (Appendix Table 5). Ultimately, 54 studies were included with 18 being RCTs and 36 observational studies. One study was in Spanish;¹² the rest were in English. Details on the identification process and exclusions are in Figure 1.

Figure 1.

PRISMA Flow Diagram of study selection

Study characteristics

There were 18 RCTs enrolling 1,844 patients in placebo groups with a median duration of 1.0 year (range: 0.9–2.0 years) and 36 observational studies following 24,894 patients over a median of 4.7 years (range: 1.7–13.7 years) (Table 1). Four-fifths were published after 2010. Seven studies (13%) were conducted in a multinational setting, 11 (20%) in North America (all in USA), 2 (4%) in South America, 11 (20%) in Europe, 22 (41%) in Asia, and 1 (2%) in Israel. All but one RCT and 70% of observational studies used liver biopsy. Sixteen studies (31%) contained data on progression to/regression of NAFL, 28 studies (52%) on NASH, and 31 (57%) studies on fibrosis. Four studies (7%) had data on all 3 conditions. The median age of the patients was 51 years (range: 40–68), 50% (range: 17–100%) were male, and median BMI was 29.5 kg/m² (range: 21.9–37.7 kg/m²) (Appendix Table 6). Nearly half the patients had hypertension or diabetes mellitus (median: 50% and 48%, respectively).

Table 1.

Characteristics of included studies

Author, Year	Location	Study Period	Sample Size	Diagnostic Method (scoring system)	Duration of follow-up (mean; total person-years)	Reported data for NAFLD-related condition (Y, N)
						Simple Steatosis	NASH	Fibrosis
A. Randomized Controlled Trials
Armstrong, 201654	UK	2010–2014	22	Biopsy (Brunt)	0.9; 20.3	N	Y	Y
Bril, 201955	USA	2010–2016	32	Biopsy (Brunt)	1.5; 48	N	Y	N
Chan, 201756	Malaysia	2012–2015	50	Biopsy (NASH CRN)	0.9; 46.2	N	Y	Y
Friedman, 201819	Multinational	2014–2016	126	Biopsy (NASH CRN)	1; 126	N	Y	Y
Harrison, 201829	Multinational	2012–2016	159	Biopsy (NASH CRN)	1.8; 293.5	N	Y	Y
Harrison, 201920	USA	2016–2019	74	Biopsy (NASH CRN)	1; 74	N	Y	N
Harrison, 202017	Multinational	2016–2019	105	Biopsy (NASH CRN)	1.4; 145.4	N	Y	Y
Harrison, 202030	Multinational	2017–2019	331	Biopsy (NASH CRN) & FibroScan	0.9; 305.5	N	Y	Y
Loguercio, 201231	Italy	2005–2008	53	Biopsy (Brunt)	1; 53	N	N	Y
Neuschwander-Tetri, 201532	USA	2011–2014	142	Biopsy (NASH CRN)	1.4; 196.6	Y	Y	Y
Newsome, 202057	UK	2016–2020	80	Biopsy (Brunt)	1.4; 110.4	N	Y	Y
Ratziu, 201622	Multinational	2012–2015	92	Biopsy (NASH CRN)	1; 92	N	Y	N
Ratziu, 202023	Multinational	2014–2017	72	Biopsy (NASH CRN)	2; 144	N	N	Y
Sanyal, 201025	USA	2005–2009	83	Biopsy (NASH CRN)	1.8; 132.9	N	Y	N
VanWagner, 201158	USA	2005–2009	9	Biopsy (Brunt)	1; 9	N	Y	N
Wong, 201327	HK, China	2009–2012	77	H-MRS	1; 77	Y	N	N
Younossi, 201928	Multinational	2015–2018	311	Biopsy (NASH CRN)	1.5; 466.5	N	Y	N
Zein, 201159	USA	2006–2010	26	Biopsy (Brunt)	1; 26	N	Y	N
B. Observational Studies
Ajmera, 201818	USA	2009–2017	95	Biopsy (NASH CRN)	1.7; 161.5	N	Y	N
Akuta, 201652	Japan	1980–2016	36	Biopsy (Brunt)	4.6; 165.6	N	N	Y
Castro, 201933	Brazil	2005–2011	39	Biopsy (NASH CRN) & Fibroscan	10; 390	N	N	Y
Chan, 201460	Malaysia	2003–2010	35	Biopsy (NASH CRN)	6.4; 224	N	Y	Y
Cho, 201961	Korea	2000–2010	2726	Ultrasound	5.2; 14129.8	Y	N	N
Ekstedt, 20065	Sweden	1988–2005	68	Biopsy (Brunt)	13.7; 938.4	N	N	Y
Evans, 200234	UK	1985–1999	46	Biopsy (Brunt)	8.2; 377.2	N	N	Y
Fassio, 200435	Argentina	1986–2002	22	Biopsy (Brunt)	4.3; 94.6	N	N	Y
Hamaguchi, 201062	Japan	1997–2008	39	Biopsy (Brunt)	2.4; 93.6	Y	Y	Y
Harrison, 200321	USA	1985–2001	22	Biopsy (Brunt)	5.7; 125.4	N	N	Y
Hui, 200536	HK, China	1996–2004	17	Biopsy (Brunt)	5.8; 97.8	N	N	Y
Kamarajah, 201844	Malaysia	2012–2017	113	Biopsy (NASH CRN) & Fibroscan	3.1; 348.4	N	N	Y
Kim, 200937	South Korea	2000–2005	2895	Ultrasound	5; 14475	Y	N	N
Kim, 201842	South Korea	2007–2013	956	Ultrasound	4.6; 4397.6	Y	N	N
Kleiner, 201946	USA	2004–2013	446	Biopsy (NASH CRN)	4.9; 2185.4	Y	Y	Y
Lee, 201043	South Korea	2004–2007	1705	Ultrasound	2.2; 3716	Y	N	N
Lin, 201938	Taiwan	2001–2010	10	Biopsy (NASH CRN)	2.4; 23.9	N	N	Y
McPherson, 201563	UK	1991–2011	108	Biopsy (NASH CRN)	6.6; 712.8	N	Y	Y
Moreno Sanchez, 199112	Spain	1980–1985	10	Biopsy (METAVIR)	4.8; 48.3	N	N	Y
Nasr, 201839	Sweden	1988–2015	32	Biopsy (NASH CRN)	9.1; 291.3	N	N	Y
Nogami, 201964	Japan	2006–2017	34	Biopsy (NASH CRN)	10; 340	N	N	Y
Pais, 201140	France	1998–2009	6	Biopsy (Kleiner)	4.8; 28.8	N	Y	Y
Pais, 201365	France	1998–2009	70	Biopsy (Brunt)	3.4; 238	N	Y	Y
Reddy, 202024	USA	2006–2016	36	Biopsy (NASH CRN)	3.8; 136.8	N	Y	N
Seko, 201566	Japan	1999–2014	52	Biopsy (NASH CRN)	2.7; 137.8	N	Y	Y
Seo, 201567	South Korea	2006–2009	1916	Ultrasound	4.2; 8047.2	Y	N	N
Shen, 201226	HK, China	2004–2010	51	Biopsy (NASH CRN)	3; 153	N	Y	N
Teli, 199541	UK	1978–1994	40	Biopsy (defined in study)	8.6; 345.7	Y	Y	Y
Vilar-Gomez, 202068	USA	2004–2016	90	Biopsy (NASH CRN)	5.6; 504	N	Y	N
Wang, 201869	China	2006–2014	6310	Ultrasound	7; 44170	N	N	N
Wang, 202070	China	2010–2015	4273	Ultrasound	4.4; 18801.2	Y	N	N
Wong, 201071	HK, China	2006–2009	52	Biopsy (Brunt)	3; 156	N	Y	Y
Wong, 201516	HK, China	2008–2013	565	H-MRS & FibroScan	3.9; 2212.9	Y	N	N
Zelber-Sagi, 201472	Israel	2003–2011	147	Ultrasound	6.8; 999.6	Y	N	N
Zhang, 201973	China	2011–2014	1325	Ultrasound	3; 3975	Y	N	N
Zhou, 201274	China	2005–2009	507	Ultrasound	4; 2028	Y	N	N

Note: HK, Hong Kong; H-MRS, Proton Magnetic Resonance Spectroscopy; NAFLD, Non-alcoholic Fatty Liver Disease; NASH, Non-alcoholic Steatohepatitis; NASH CRN, Non-alcoholic Steatohepatitis Clinical Research Network; N, no; Y, yes

Assessment of study’s risk of bias

Among RCTs, most or all had “low” risk of bias in 4 of 5 domains, the exception being bias due to missing outcomes (Appendix Figure 1). Five studies were at “high” risk of bias because their attrition rate exceeded 20%.^28–32

Among observational studies, none had “critical” risk of bias in any domain, but a large percentage of studies (~70%–80%) had “moderate” or “serious” risk for confounding, participant selection, or outcome measurement (Appendix Figure 2). Among 15 studies at “serious” risk of bias in ≥1 domain, twelve were in the confounding domain due to not stratifying their analysis by patient characteristics or not conducting multivariable analysis.^{12,21,24,33–41} Six were in the participant domain for including <25 patients.^{12,21,35,36,38,40} One was in the missing data domain because authors did not clearly state the number (and reasons) for loss-to-follow up, had 20% or more of patients lost to follow-up, or missing duration of follow-up.⁴² Finally, five were in the outcome measurement domain because mean follow-up duration was <2 years or identification of outcomes by pathologists or radiologists were not clearly stated.^{12,18,33,38,43} All studies had “low” risk of bias in selection of reported results and “low” or “moderate” risk in classification of study population.

Progression to and regression of NAFL

Sixteen studies reported rates of progression to or regression from NAFL. One-fourth used liver biopsy involving 667 patients and the others used imaging (ultrasound, H-MRS) involving 24,727 patients. No RCTs reported progression while two reported regression (pooled incidence=18.4/100 person-years). Among observational studies, 3,871/19,323 (20%) patients with no NAFL at baseline developed NAFL after a median of 4.4 years. In contrast, 892/4,295 patients (21%) with NAFL regressed after a median of 4.6 years. Pooled incidence was 4.8/100 person-years (95%CI: 3.7–6.3) for progression (I² = 98.43%) and 2.4/100 person-years (95%CI: 1.2–5.0) for regression (I² = 98.64%) (Figure 2). Heterogeneity was high for observational studies and zero for RCTs.

Figure 2:

Pooled incidence rate (cases per 100 person-years) of a) progression to and b) regression of nonalcoholic fatty liver (NAFL) or simple steatosis

Progression to and regression of NASH

Among 28 studies in patients with NASH, 12 provided data on progression, and 25 on regression. All studies used liver biopsy. Two RCTs reported progression; pooled incidence was 37.5/100 person-years. In 15 RCTs reporting regression, 165/1,511 (11%) patients resolved NASH after a median of 1.4 years; pooled incidence was 9.2/100 persons-years (95% CI: 6.6–12.9). Among observational studies, 116/372 (31%) patients developed NASH over a median of 4.7 years (range: 2.4–11.6 years). In addition, 189/661 (29%) patients with NASH resolved at follow-up. Pooled incidence was 7.4 and 5.1/100 person-years for progression and regression, respectively (Figure 3). Heterogeneity was high (I² >67%).

Figure 3:

Pooled incidence rate (cases per 100 person-years) of a) progression to and b) regression of nonalcoholic steatohepatitis (NASH)

Progression to and regression of fibrosis

Twenty-eight studies used liver biopsy alone; 3 used both biopsy and FibroScan® to stage fibrosis. Of the three studies that used both Fibroscan and liver biopsy, two used biopsy and FibroScan® at both baseline and follow-up, but only biopsy data were included.^30,44 The third study reported only FibroScan® at follow-up, so we included FibroScan data.³³ We excluded one study which used FibroScan® to identify advanced fibrosis or cirrhosis but did not define the fibrosis stage.¹⁶

In RCTs, 19/104 (18%) patients with no fibrosis progressed: 84% to F1 and 16% to F2. Among 89 patients with F1, 28 (31%) progressed, 14 (16%) regressed, and 47 (53%) remained stable. From F2, nearly half progressed or regressed, while the others remained stable. From F3, 22% regressed while 13% progressed to cirrhosis. Finally, from F4, 13% of patients regressed to bridging fibrosis. Pooled incidence of progression from F0, F1, F2, and F3 were 19.1, 25.4, 18.5, and 8.0/100 person-years, respectively (Table 2; Appendix Figure 3). Pooled incidence for regression from F1, F2, F3, and F4 were 12.7, 21.2, 17.2, and 9.0/100 person-years, respectively (Appendix Figure 4). Heterogeneity was low or moderate. No RCTs reported data separately for patients with NAFL.

Table 2:

Pooled incidences of fibrosis progression and regression among randomized controlled trials

Baseline fibrosis stage	No. of studies/patients	Person-years of follow up	Progression					Regression
			No. of cases	Incidence rate (cases/100 person-years)	No. of stages	Incidence rate (stages/100 person-years)	Time to progress by 1 stage (years)	No. of cases	Incidence rate (cases/100 person-years)	No. of stages	Incidence rate (stages/100 person-years)	Time to regress by 1 stage (years)
NAFLD
F0	4/104	108.9	19	19.1 (9.4–39.1)	22	23.4 (10.9–50.3)	4.3 (2.0–9.2)	NA	NA	NA	NA	NA
F1	5/89	109.9	28	25.4 (13.4–48.1)	37	34.0 (19.8–58.4)	2.9 (1.7–5.1)	14	12.7 (7.5–21.5)	14	12.7 (7.5–21.5)	7.9 (4.7–13.3)
F2	5/111	146.3	27	18.5 (12.7–26.9)	30	20.3 (13.4–30.7)	4.9 (3.3–7.5)	31	21.2 (14.9–30.1)	36	24.6 (17.8–34.1)	4.1 (2.9–5.6)
F3	10/423	576.8	54	8.0 (4.5–14.0)	54	8.0 (4.5–14.0)	12.5 (7.1–22.2)	94	17.2 (12.9–22.9)	132	24.0 (17.4–33.0)	4.2 (3.0–5.7)
F4	4/252	370.2	NA	NA	NA	NA	NA	34	9.0 (4.6–17.9)	34	9.0 (4.6–17.9)	11.1 (5.6–21.7)
NASH
F0	3/29	33.9	10	29.4 (14.2–60.8)	13	38.4 (22.3–66.1)	2.6 (1.5–4.5)	NA	NA	NA	NA	NA
F1	5/89	109.9	28	25.4 (13.4–48.1)	37	34.0 (19.8–58.4)	2.9 (1.7–5.1)	14	12.7 (7.5–21.5)	14	12.7 (7.5–21.5)	7.9 (4.7–13.3)
F2	4/94	129.3	27	20.9 (14.3–30.5)	30	23.2 (16.2–33.2)	4.3 (3.0–6.2)	23	17.8 (11.8–26.8)	28	21.7 (15.0–31.4)	4.6 (3.2–6.7)
F3	9/415	568.8	54	8.4 (4.9–14.3)	54	8.4 (4.9–14.3)	11.9 (7.0–20.4)	94	17.7 (13.2–23.6)	132	25.0 (18.3–34.3)	4.0 (2.9–5.5)
F4	4/252	370.2	NA	NA	NA	NA	NA	34	9.0 (4.6–17.9)	34	9.0 (4.6–17.9)	11.1 (5.6–21.7)

Note: No meta-analysis was possible for patients with NAFL because no randomized controlled trials provided data separately for patients with versus without NASH. NAFLD, Non-alcoholic Fatty Liver Disease; NAFL, Non-alcoholic Fatty Liver; NASH, Non-alcoholic Steatohepatitis.

In observational studies, 123/297 (41%) patients progressed from F0 after a median of 5.8 years: 58% to F1, 28% to F2, 12% to F3, and 2% to F4. Among patients with F1, 23% regressed to F0 while 37% progressed to higher stages (59% to F2, 34% to F3, and 7% to F4). In patients with F2, a similar proportion progressed, regressed, or remained stable (32%, 37%, and 31%, respectively) after 5 years. Patients with F3 regressed more often than progressed, 32% vs. 22%, while a larger proportion of patients in the observational studies with stage F4 regressed than noted in RCTs (29% vs. 13%). Pooled incidence of progression from F0, F1, F2, and F3 were 6.5, 6.9, 5.8, and 4.5/100 person-years, respectively (Table 3). Pooled incidence for regression from F1, F2, F3, and F4 were 2.5, 4.6, 6.0, and 4.7/100 person-years, respectively. In addition, patients with NAFL tended to have lower incidence of progression than those with NASH although no tests for subgroup difference were statistically significant. Forest plots of incidence calculated as stages/100 person-years are in Supplementary Results and Appendix Figures 5–6.

Table 3:

Pooled incidences of fibrosis progression and regression among observational studies

Baseline fibrosis stage	No. of studies/patients	Person-years of follow up	Progression					Regression
			No. of cases	Incidence rate (cases/100 person-years)	No. of stages	Incidence rate (stages/100 person-years)	Time to progress by 1 stage (years)	No. of cases	Incidence rate (cases/100 person-years)	No. of stages	Incidence rate (stages/100 person-years)	Time to regress by 1 stage (years)
NAFLD
F0	18/297	1965.5	123	6.5 (4.6–9.0)	196	10.0 (7.7–13.2)	9.9 (7.6–13.0)	NA	NA	NA	NA	NA
F1	19/385	2110.0	143	6.9 (4.8–10.0)	212	9.7 (6.8–13.7)	10.3 (7.3–14.7)	90	2.5 (1.4–4.6)	90	2.5 (1.4–4.6)	40.0 (21.7–71.4)
F2	19/219	1224.4	71	5.8 (4.6–7.3)	80	7.5 (6.1–9.2)	13.3 (10.9–16.4)	92	4.6 (2.4–8.8)	106	4.9 (2.3–10.5)	20.4 (9.5–43.5)
F3	14/202	1009.3	45	4.5 (3.3–6.0)	45	4.5 (3.3–6.0)	22.2 (16.7–30.3)	64	6.0 (3.7–9.6)	101	8.0 (4.6–13.8)	12.5 (7.2–21.7)
F4	9/41	230.7	NA	NA	NA	NA	NA	12	4.7 (2.1–10.8)	12	4.7 (2.1–10.8)	21.3 (9.3–47.6)
NAFL
F0	7/72	418.8	23	6.1 (3.1–11.80	34	8.6 (4.8–15.5)	11.6 (6.5–20.8)	NA	NA	NA	NA	NA
F1	5/58	216.2	27	12.2 (7.6–19.7)	39	17.6 (11.3–27.4)	5.7 (3.6–8.8)	6	2.6 (0.9–7.3)	6	2.6 (0.9–7.3)	38.5 (13.7–111.1)
F2	4/18	63.8	3	4.7 (1.5–14.6)	4	6.3 (2.4–16.7)	15.9 (6.0–41.7)	8	12.5 (6.0–25.9)	10	14.2 (5.1–39.7)	7.0 (12.5–19.6)
F3*	1/11	NA	NA	NA	NA	NA	NA	NA	NA	NA	NA	NA
F4*	1/11	NA	NA	NA	NA	NA	NA	NA	NA	NA	NA	NA
NASH
F0	9/41	255.7	20	8.2 (4.9–13.7)	30	11.9 (8.0–17.8)	8.4 (5.6–12.5)	NA	NA	NA	NA	NA
F1	12/120	645.3	50	7.8 (5.7–10.7)	72	10.9 (8.1–14.8)	9.2 (6.8–12.3)	19	1.5 (0.4–6.2)	19	1.5 (0.4–6.2)	66.7 (16.1–250.0)
F2	12/92	522.6	35	6.7 (4.8–9.3)	44	8.4 (6.3–11.3)	11.9 (8.8–15.9)	20	3.7 (1.0–8.8)	29	4.5 (1.6–12.6)	22.2 (7.9–62.5)
F3	8/52	256.8	13	5.1 (2.9–8.7)	13	5.1 (2.9–8.7)	19.6 (11.5–34.5)	18	5.9 (1.8–19.0)	23	6.3 (1.6–24.6)	15.9 (4.1–62.5)
F4	4/11	69	NA	NA	NA	NA	NA	1	1.2 (0.1–24.0)	1	1.2 (0.1–24.0)	83.3 (4.2–1000.0)

Note: NAFLD, Non-alcoholic Fatty Liver Disease; NAFL, Non-alcoholic Fatty Liver; NASH, Non-alcoholic Steatohepatitis.

^*No meta-analysis was possible because there was only one study providing data on patients with NAFL and F3/F4 at baseline.⁵²

Sensitivity analyses

Rates of NASH progression/regression were lower in observational studies with ≥5 years of follow-up or conducted in USA/Europe than studies with <5 years or conducted in Asia, respectively (Appendix Tables 9&10). Fibrosis progression rates in patients with F0/F1 and regression rates in all patients were also lower in studies with longer than shorter follow-up (Appendix Tables 11&12). Detailed subgroup analyses are included in Supplementary Results and Appendix Tables 7–18. Publication bias was not observed (Appendix Figures 7–12).

DISCUSSION

In this systematic review and meta-analysis of 54 studies, we provide a comprehensive assessment of the natural history of NAFLD, ranging from NAFL to NASH and fibrosis. We found that among observational studies, one in every 20 otherwise healthy adults develop NAFL in a year. Interestingly, 10 of 11 studies were conducted in Asia with a pooled rate of 5.1/100 person-years (95%CI: 3.9–6.6) which was similar to that reported in an earlier meta-analysis.¹ Our pooled incidence was >10 times higher than that reported by a population-based study of Olmsted County, Minnesota, USA,⁴⁵ but that study used ICD codes to identify NAFL instead of ultrasound, as was used in the studies we reviewed.

Novel findings from the present study include the meta-analysis based estimate that after developing fatty liver, 2.4% of patients in observational studies have resolution of NAFL each year. These patients were considered to receive no conventional treatment for NAFLD although they might undergo partial lifestyle modifications or change their behaviors during the observational period. Similar estimates for RCTs was not possible due to the limited sample size. Of note, regression rate in observational studies using liver biopsy was nearly half of that in studies using imaging. Patients who had a liver biopsy are likely to have more severe NAFLD-related conditions than those receiving imaging with a lower probability of regression in biopsied patients. Of the three studies using liver biopsy, one was a registry of patients from eight large US medical centers.⁴⁶ In addition, because ultrasound is only able to reliably detect >33% hepatic steatosis, patients with a reduction between 5–33% steatosis instead of <5% might have been classified as improved, overestimating the regression rate.^47,48 At the same time, patients without NAFL who developed 5–33% steatosis might have been missed by ultrasound, underestimating the progression rate. Ultrasound is less sensitive in obese patients, reducing the accuracy of the diagnosis and severity of NAFLD in this population.⁴⁹ Also, ultrasound provides strata (<5, 5–33, 33–66, >66%) but given the interobserver variability and the challenge to accurately quantify changes, studies using ultrasound alone may have limitations. Therefore, better quality data are needed on development and resolution of NAFL, especially outside of Asia where the majority of published studies were conducted.

Another important finding was that annually, about 7% of NAFL patients developed NASH and 5% of those with NASH had resolution in observational studies. Because the NAFLD activity score used to diagnose NASH does not include fibrosis as a component, and the included studies did not report NASH stratified by presence of fibrosis, we could not determine if baseline fibrosis stage affects NASH progression. Incidence and regression of NASH did not differ much by region but were 4–6 times higher in studies with shorter than longer follow-up. Compared to observational studies, rates of NASH progression and regression were higher in RCTs. One RCT enrolled patients with fibrosis²⁰ and both RCTs required a NAFLD activity score (NAS) ≥4 with a score ≥1 in steatosis, ballooning, and lobular inflammation.^20,32 Patients in these studies had NASH and may therefore experience a higher rate of progression than patients in observational studies. A recent meta-analysis estimated ~12% of placebo patients with NASH achieved NASH resolution without worsening of fibrosis, which differ only by older age.⁵⁰

To estimate fibrosis progression and regression, we included 3 times as many studies as a previous meta-analysis.² We found that rates of fibrosis progression and regression were much higher in RCTs than in observational studies. In observational studies, 59% of F0 patients remained stable and 6% progressed to bridging fibrosis or cirrhosis at follow-up compared to 82% and 0% in RCTs, respectively. On average, 7% of F0 progressed each year, primarily to stage 1 or 2. Patients with NASH progressed faster than those without, progressing from stage 0 in 8.4 years for NASH versus 11.6 years for NAFL, estimates that are similar to those reported by Singh et al.² Because NASH patients can develop fibrosis at a more rapid rate than NAFL patients) and advanced fibrosis is associated with increased mortality,⁵¹ interventions to prevent patients with NASH from developing fibrosis are important. However, current RCTs have focused mostly on patients with fibrosis and high quality data are needed to determine the prevalence and natural course of NASH patients without fibrosis.

In observational studies, 22% of patients with F1/F2 progressed to F3/F4 with a higher proportion in those with F2 than F1. However, rates in patients with NASH were higher for F2 but lower for F1 than those without NASH. These findings suggest that baseline stage is important in fibrosis development, while the role of active NASH could not be confirmed in patients with F≥F1. Furthermore, 71% of F4 patients remained stable. Since we did not evaluate decompensated cirrhosis or other complications of cirrhosis, we may have overestimated the proportion of stable cirrhotic patients. We also could not estimate incidence among patients with NAFL and F3/F4 because only one study reported data separately for non-NASH patients.⁵² The overall incidence of fibrosis progression/regression for all NAFLD patients was comparable to that for NASH patients. It is, therefore, possible that most patients with significant fibrosis already had NASH at evaluation. In the NASH CRN registry, ~95% of patients with F3/F4 had definite or borderline NASH.⁶ Finally, we could not estimate fibrosis progression or regression for patients with NAFL in RCTs because no studies published to date provide these data.

Heterogeneity was moderate to high in most estimates for NAFL and NASH, consistent with previous meta-analyses.^1,2,53 Potential reasons include differences in patient characteristics, variability in imaging or biopsy reading, and study design. Subgroup analyses helped minimize the heterogeneity. Importantly, RCTs had little or no heterogeneity compared to observational studies, especially for fibrosis estimates in patients with advanced stages at baseline. RCTs often have shorter duration and smaller sample size than observational studies. In our study, observational studies had a median follow-up that was 5 times longer and sample size that was 13 times larger than those in the RCTs. Also, patients enrolled in RCTs have more severe disease and are evaluated more frequently, with better outcome assessment than in observational studies.

Strengths of our analyses include the large number of studies evaluated that allows us to provide robust estimates of progression and regression of NAFL and NASH without treatment. In addition to examining fibrosis in patients with NAFL vs. NASH and by baseline fibrosis stage separately, we also distinguished rates of progression from those of regression using both cases and stages per 100 person-years. These data provide detailed information on the likelihood of progression/regression and magnitude of the potential change and allows us to estimate progression in people with F3/F4 that could not be identified by previous analyses.^2,7

Limitations include our estimates being confounded by the differences among studies in patient characteristics, modality and quality of diagnosis, healthcare settings, or outcome assessment. We mitigated the confounding effect to some extent by including all high quality published data. Second, without access to patient-level data we could not estimate disease progression in high-risk groups such as patients with Hispanic ethnicity, obesity, diabetes, or metabolic syndrome, who might have a faster disease progression. Such analyses will require potentially depositing raw data into a repository to permit investigators to include patient-level data. Third, we used author-defined criteria for determining outcomes, and these varied across studies, leading to potentially inconsistent patient classification. These could also be overcome by the availability of raw data. Finally, because not all studies provided fibrosis data for patients with NAFL vs. NASH, our estimates for subgroups could be biased. However, despite these limitations, the inclusion of a large number of high quality studies allowed for robust analyses.

In conclusion, our meta-analysis provides the most comprehensive, up-to-date data on the natural history of NAFLD in adults. Understanding disease progression is important for stratifying high-risk patients for intervention versus monitoring. Our approach can help identify patients in need of aggressive and early treatment to prevent advanced complications, target them for RCTs, and allow evidence based prognosis with patients. Microsimulation studies will benefit from the detailed estimates provided in our study. Future studies on the natural history of NAFLD in children and other high-risk adults will complete our epidemiological understanding of the disease, which will effectively guide our clinical care efforts and improve patients’ quality-of-life.

WHAT YOU NEED TO KNOW.

Background:

Nonalcoholic fatty liver disease affects one-fourth of adults, but its natural history is not well characterized. Our study meta-analyzed rates of progression to and regression of simple steatosis, NASH, and fibrosis.

Findings:

One in 20 adults developed simple steatosis but only 2.4% regressed. Progression to NASH was more common than regression from it, whereas rates of fibrosis progression were similar across baseline stage.

Implications for Patient Care:

Because NASH patients can develop fibrosis more quickly than those with simple steatosis, interventions to prevent fibrosis progression in these patients are in urgent need.