Accuracy of controlled attenuation parameter for liver steatosis in patients at risk for metabolic dysfunction-associated steatotic liver disease using magnetic resonance imaging: a systematic review and meta-analysis

Introduction

The prevalence of metabolic dysfunction-associated steatotic liver disease (MASLD) is greatly increasing: MASLD affects approximately 25-30% of the adult population [1]. Its advanced form, metabolic dysfunction-associated steatohepatitis (MASH), is among the leading causes of hepatocellular carcinoma and liver transplantation in the United States [2,3]. Despite its recognition as a significant healthcare burden, MASLD remains largely underdiagnosed [4,5].

Liver biopsy is considered the reference standard for MASLD diagnosis [1,2]. However, it is an invasive procedure with inherent limitations, including significant intra- and inter-observer variability and sampling error [6]. Moreover, histologic assessment is impractical for screening at the population level, hampers patient recruitment in clinical trials, while it can potentially lead to life-threatening complications. Therefore, noninvasive biomarkers are needed in both clinical practice and clinical research as alternatives to liver biopsy. Magnetic resonance imaging-derived proton-density-fat fraction (MRI-PDFF) is an accurate and reproducible imaging modality for the assessment of liver steatosis (LS) that strongly correlates with biopsy results [7,8]. Consequently, MRI-PDFF is widely used in MASH clinical trials, either sequentially or as main criterion for patient selection. Nevertheless, its use in the everyday clinical setting is limited because of its high cost and limited availability.

The controlled attenuation parameter (CAP) is an ultrasound modality that allows for the rapid, noninvasive evaluation of LS [9]. However, consensual CAP thresholds for the diagnosis of steatosis in the context of MASLD are lacking [10]. Several trials and meta-analyses have addressed this issue by evaluating the diagnostic accuracy of CAP against liver biopsy [11-14]. While histology remains the gold standard, a shift is evident in emerging studies favoring MRI-PDFF over biopsy as a reliable alternative for steatosis assessment. The accuracy of CAP in identifying various degrees of steatosis, as defined by MRI-PDFF, has not been fully elucidated. We conducted a systematic review and meta-analysis to assess the diagnostic accuracy of CAP for the detection of hepatic steatosis in patients at risk for MASLD, using MRI-PDFF as the reference standard.

Materials and methods

We report our systematic review and meta-analysis, carried out in accordance with the Preferred Reporting Items for a Systematic Review and Meta-analysis of Diagnostic Test Accuracy Studies (PRISMA-DTA) statement (Supplementary Table 1) [15]. Our review is based on a pre-specified protocol registered in PROSPERO (CRD42023464466).

Results

After duplicate removal, we screened 3171 records, from which we selected 10, describing 8 studies with 1116 patients (Supplementary Fig. 1) [29-36]. The baseline characteristics of these studies are presented in Supplementary Table 8. All studies assessed the diagnostic accuracy of CAP against MRI-PDFF with a threshold of 5% [29-36], while 6 of them provided CAP diagnostic accuracy estimates for MRI-PDFF ≥10% [29-31,34-36]. One study was available solely as a conference abstract [29]. Most studies were single-center, following a prospective design (Supplementary Table 9). Three studies had a case-control design [30,33,36]. Seven studies acquired CAP measurements using both the M and XL probe with a Fibroscan device (Echosens, France) [29-35]. One study acquired CAP measurements using the iLivTouch device (Hisky Med, China) [36]. Most studies employed the region-of-interest (ROI) approach to analyze PDFF maps with a varying number of ROIs. The sample size of the included studies ranged from 69-248 participants. The prevalence of MASLD (defined as MRI-PDFF ≥5%) ranged from 65.2-93.9%. Across all participants, 45% were male and 16.6% had type 2 diabetes (T2D). Mean BMI and age were 29.9 kg/m² and 49.6 years, respectively. The average mean alanine aminotransferase and aspartate aminotransferase levels across studies were 37.1 U/L and 29.4 U/L, respectively.

Diagnostic accuracy

In our main analysis, CAP sensitivity and specificity for MRI-PDFF ≥5% ranged from 0.74-0.91 and from 0.57-0.92, respectively (Fig. 1). Respective values for the diagnosis of MRI-PDFF ≥10% ranged from 0.79-0.87 and from 0.51-0.87 (Fig. 2). The pooled sensitivity and specificity of CAP for the diagnosis of MRI-PDFF ≥5% was 0.84 (95%CI 0.79-0.88, I²=72.2%) and 0.77 (95%CI 0.68-0.84, I²=52.6%), yielding an LRp of 3.67 (95%CI 2.55-5.27), LRn of 0.21 (95%CI 0.16-0.30) and a DOR equal to 17.4 (95%CI 9.6-31.6). The pooled sensitivity, specificity, LRp, LRn and DOR for MRI-PDFF ≥10% were 0.83 (95%CI 0.80-0.87, I²=14.1%), 0.72 (95%CI 0.59-0.82, I²=86.4%), 2.97 (95%CI 1.94-4.55), 0.23 (95%CI 0.17-0.31), and 12.8 (95%CI 6.4-25.9) respectively. The AUROCs for MRI-PDFF ≥5% and MRI-PDFF ≥10% were 0.88 (95%CI 0.85-0.90) and 0.85 (95%CI 0.82-0.88), respectively (Fig. 3,4).

Figure 1 Forest plot of individual study and pooled estimates of sensitivity and specificity of CAP for MRI-PDFF ≥5%

CAP, controlled attenuation parameter; MRI-PDFF, magnetic resonance imaging proton density fat fraction

Figure 2 Forest plot of individual study and pooled estimates of sensitivity and specificity of CAP for MRI-PDFF ≥10%

CAP, controlled attenuation parameter; MRI-PDFF, magnetic resonance imaging proton density fat fraction

Figure 3 Hierarchical summary receiver operating curve (HSROC) plot of sensitivity versus specificity of CAP for MRI-PDFF ≥5%

Each circle represents a study, with the size proportional to the study size. The curve represents the summary receiver operating characteristic curve for CAP. The square represents the summary estimate of test performance. The bigger-sized dashed outline represents the 95% prediction region and the smaller-sized dashed outline the 95% confidence region

CAP, controlled attenuation parameter; MRI-PDFF, magnetic resonance imaging proton density fat fraction

Figure 4 Hierarchical summary receiver operating curve (HSROC) plot of sensitivity vs. specificity of CAP for MRI-PDFF ≥10%

Each circle represents a study, with the size proportional to the study size. The curve represents the summary receiver operating characteristic curve for CAP. The square represents the summary estimate of test performance. The bigger-sized dashed outline represents the 95% prediction region and the smaller-sized dashed outline the 95% confidence region

CAP, controlled attenuation parameter; MRI-PDFF, magnetic resonance imaging proton density fat fraction

Subgroup and sensitivity analyses

Results from subgroup analyses are presented in Table 1. Studies conducted in Europe or the USA had pooled sensitivity and specificity of 0.79 and 0.75, respectively, while studies conducted in Asia had pooled sensitivity and specificity both of 0.91. Studies applying ≥2 quality criteria to determine a valid CAP examination resulted in higher specificity estimates (0.85, 95%CI 0.75-0.91) than those using <2 criteria (0.67, 95%CI 0.49-0.81). Similarly, studies with lower BMI produced higher specificity estimates (0.85, 95%CI 0.73-0.93) compared to studies with a mean BMI ≥30 kg/m² (0.72, 95%CI 0.57-0.83). A CAP positivity threshold of ≥288 dB/m (AASLD recommended threshold) resulted in a pooled sensitivity, specificity LRp, LRn and DOR of 0.81, 0.68, 2.55, 0.28 and 9.2 respectively. These estimates were similar for the CAP positivity threshold of ≥275 dB/m recommended by the EASL. Results from sensitivity analyses yielded similar estimates to our main analysis (Supplementary Table 10).

Table 1 Results from subgroup analyses for MRI-PDFF≥5%

Clinical utility

We explored the clinical utility of CAP in diagnosing MRI-PDFF ≥5% using Fagan nomograms across different prevalence scenarios. In the first scenario, reflecting MASLD prevalence in the general population, we assumed a pre-test probability (prevalence) of 30%. A negative CAP result, with an LRn of 0.2, reduced the post-test probability for MASLD to 8% (Supplementary Fig. 4). In the second scenario, reflecting MASLD prevalence among individuals with obesity and/or T2D, a pre-test probability of 70% was applied. Here, a positive CAP result led to a post-test probability of 90% (Supplementary Fig. 5). Fig. 5 presents the post-test probability for liver steatosis based on pre-test probability and results of CAP testing.

Figure 5 Conditional probability plot for MRI-PDFF ≥5%

The upper dashed curve represents a positive test, and the lower dashed curve a negative test

MRI-PDFF, magnetic resonance imaging proton density fat fraction; LR, likelihood ratio; NPV, negative predictive value; PPV, positive predictive value

Discussion

In this systematic review and meta-analysis, we assessed the accuracy of CAP for diagnosing liver steatosis in patients at risk for MASLD, using MRI-PDFF as the reference standard. In addition, we assessed the accuracy of CAP for detecting liver steatosis ≥10%, a threshold commonly employed during screening in MASH clinical trials. CAP demonstrated a good diagnostic accuracy for the detection of both liver steatosis ≥5% (AUROC, 0.88) and liver steatosis ≥10% (AUROC, 0.85). Pooled sensitivity (84% and 83% respectively) and specificity (77% and 72% respectively) estimates were similar for both target conditions. For MRI-PDFF ≥5%, subgroup analyses based on study origin (Asia), BMI (<30 kg/m²) and a number of criteria used to determine a successful CAP examination (≥2) resulted in higher diagnostic accuracy estimates.

To our knowledge, this is the first meta-analysis assessing the diagnostic accuracy of CAP in patients at risk for MASLD, using MRI-PDFF as the reference standard. We performed a thorough literature search of major electronic databases and gray literature sources, and contacted authors for additional data in order to produce a comprehensive summary of all the available evidence to date. We performed our analyses using robust methodology, following guidance from the recently published Cochrane Handbook for Systematic Reviews of Diagnostic Test Accuracy [25]. Moreover, we performed several sensitivity and subgroup analyses to address heterogeneity and assessed the certainty of evidence using the GRADE methodological framework.

Certain limitations should be acknowledged. Despite a thorough literature search, we identified only a small number of studies that met our eligibility criteria. Consequently, our analyses included a small number of studies and participants, and therefore should be interpreted with caution. Most of the studies included in our meta-analysis recruited participants both with and without T2D (proportion of participants with T2D ranging from 14.5-56%). However, data regarding the accuracy of CAP based on the presence of diabetes were sparse, prohibiting subgroup analysis based on T2D status. Furthermore, most of the included studies were at high risk of bias, owing to concerns regarding the positivity thresholds of the index test. More specifically, the positivity threshold for CAP was based on data from analyses (Youden index), rather than being pre-specified. Notably, the studies included in our meta-analysis were conducted prior to the transition from NAFLD to MASLD terminology. Consequently, only 2 studies explicitly outlined the inclusion of metabolic syndrome and/or cardiometabolic comorbidities as criteria [32, 33]. Nevertheless, considering the mean BMI of 29.9 kg/m² (range: 28.3-31.7 kg/m²) across included studies, we are confident that they all align with the MASLD definition, regardless of the initial terminology used. Several studies have examined the overlap between MASLD and NAFLD definitions, indicating high concordance rates of up to 96% [37]. Finally, there was high heterogeneity among studies in both sensitivity and specificity estimates.

Previously published meta-analyses used liver biopsy as the reference standard for MASLD diagnosis [38-40]. In line with our findings, these meta-analyses support a good diagnostic accuracy of CAP for the detection of steatosis ≥S1 (Stage 0 vs. Stage 1-3) and steatosis ≥S2 (Stage 0-1 vs. Stage 2-3), with AUROCs ranging from 0.85-0.96 and from 0.79-0.83, respectively [38-40]. Discrepancies in our results in terms of performance estimates can be primarily attributed to the different reference standards employed in each meta-analysis (MRI-PDFF vs. liver biopsy). Despite using the same metric (%), histology and PDFF assess steatosis using different approaches. MRI-PDFF offers a continuous measurement directly derived from the properties of the liver, while biopsy results reflect the proportion of hepatocytes with macrovesicular lipid droplets, expressed on an ordinal scale. A recent study evaluated the agreement between the 2 methods, suggesting that above the 5% steatosis threshold, histology estimates tend to exceed PDFF results by up to approximately 3-fold [41]. Furthermore, variations in the prevalence of the target condition and CAP positivity thresholds among primary studies included in each meta-analysis should be taken into consideration. In our study, the median prevalence of MASLD was 76%, while in previous meta-analyses the median prevalence ranged from 90-97%. Meta-epidemiological data suggest an association between higher sensitivity/lower specificity and higher prevalence settings [42]. In our meta-analysis, the average median CAP positivity threshold for MASLD diagnosis was 280 dB/m. In previous meta-analyses, CAP positivity thresholds ranged from 254-273.6 dB/m for steatosis ≥S1 [38-40]. Lower CAP positivity thresholds are likely to favor sensitivity estimates, whereas higher positivity thresholds would enhance specificity results.

In line with our results, previous meta-analyses identified geographic region and BMI as potential sources of heterogeneity [38,39]. Studies conducted in Asia tended to produce higher diagnostic accuracy estimates. Whether or not these findings are related through the lower prevalence of obesity in Asia compared to western countries, and the higher prevalence of lean MASLD in Asia compared to Europe and the USA, is a matter of controversy [43]. Other factors, including dietary and exercise habits, should be taken into consideration before any conclusions are drawn. In addition, there are currently no CAP-specific quality criteria to determine a valid CAP examination. Clinicians commonly apply the already established criteria for liver stiffness measurements. Based on our findings, implementing at least 2 of these criteria (preferably a combination of more than 10 measurements and an IQR/median <30%) can lead to better diagnostic accuracy estimates. Towards this direction, Caussy et al suggest that the accuracy of CAP for steatosis detection is more reliable when the IQR of CAP is <30 dB/m [35]. Nevertheless, these CAP-specific quality indicators need further validation before implementation in clinical practice. Our results corroborate findings from a well conducted individual patient meta-analysis assessing CAP using biopsy as reference standard [14]. In contrast to this meta-analysis, we did not identify major differences in the diagnostic performance of CAP between steatosis grades (i.e., ≥5% and ≥10%). Even though there was a trend towards lower AUROCs between the 2 groups, we consider the absolute difference of 5% between the MRI-PDFF thresholds too small to detect major differences.

Based on our findings, the positive and negative LRs of CAP for MRI-PDFF ≥5% were 3.67 and 0.21, respectively. An LRp >10 suggests that a positive test is good at ruling in a diagnosis, whereas an LRn <0.1 is considered clinically meaningful for ruling out a diagnosis [44]. Consequently, the performance of CAP does not endorse its standalone use for ruling in liver steatosis in the context of MASLD. However, in high prevalence settings (i.e., diabetes/obesity/metabolic disease outpatients’ clinics) available evidence supports the sequential use of CAP as a triage strategy for ruling out liver steatosis, alongside other noninvasive scores for steatosis (Fatty Liver Index, Hepatic Steatosis Index) and fibrosis (FIB-4 index) [45-47]. CAP positivity thresholds over ≥275 or ≥288 dB/m might be suitable for this purpose [1,10]. An MRI-PDFF value ≥10% is commonly employed for screening participants in MASH clinical trials. Based on our findings, CAP’s pooled LRn was 0.23 for MRI-PDFF ≥10%. Therefore, CAP could be used in clinical trials as an initial test to exclude patients with lower degrees of steatosis, thus identifying those patients who are likely to meet other noninvasive inclusion criteria (i.e., MRI-related thresholds) before eventually undergoing liver biopsy. Finally, it should be noted that our results derive mainly from cohorts in tertiary care centers. Currently, there is a paucity of data on the diagnostic performance of CAP using PDFF as reference standard in a primary care setting. To this end, the EPSONIP study (NCT03864510) aims to assess the prevalence and predictors of MASLD in patients with T2D in primary care.

CAP’s pooled sensitivity and specificity for MRI-PDFF ≥5% were 0.84 and 0.77, respectively. In a hypothetical cohort of 100 at-risk patients, this implies that 16 individuals would be misclassified as false negatives, indicating the test’s failure to identify underlying liver steatosis, and 23 patients would be labeled as false positives, suggesting a misdiagnosis of the target condition. These results highlight the importance of adopting a sequential noninvasive approach to accurately stratify patients at risk for liver steatosis through the implementation of additional risk scores or circulating biomarkers. In the context of a mass screening program, a different diagnostic threshold for CAP with improved sensitivity might be more appropriate, to effectively rule out the target condition and identify the majority of patients who would require further diagnostic workup.

In conclusion, CAP has acceptable accuracy for the diagnosis of MRI-defined steatosis in patients at risk for MASLD. A pooled sensitivity over 0.80 with an LRn of 0.21 suggests that CAP could be used to effectively rule out steatosis in high-prevalence settings. Further research with adequately powered and rigorously conducted studies is warranted to identify the optimal diagnostic threshold for CAP and establish its place among screening algorithms.