In the general population, previously unexplained elevated liver enzymes (ELE) are frequent findings and often cause a subsequent consultation of a hepatologist or presentation to a referral center, when first diagnosis of hepatopathy is made in a primary care doctor setting. The prevalence of ELE differs slightly in large cohort studies depending on the region or country data extracted from: While up to 9.8 % of participants have been diagnosed to have unexplained ELE in a North American cohort, a recent German cohort study including >14,000 inhabitants of the Rhine-Main area, reports 19.9 % with elevated alanine aminotransferase (ALT) and 14 % with increased gamma‐glutamyltransferase (gGT) [1,2]. Next to this high number of patients, the prevalence of unexplained ELE in the general population further increases: While Clark et al. reported a rate of elevated aminotransferases of 7.9 % in North American adults, analysed in the third National Health and Nutrition Examination Survey (NHANES) from 1988 to 1994, the rate increased to 9.8 % in subsequent NHANES study, including 6,823 participants, analysed between 1999 and 2002 [2,3]. But abnormal ELE are not only a frequent finding in the western civilization: up to 12.8 % of Iranian adults and 11.4 % of patients in a cohort of Chinese adults living in Taiwan were found with abnormal liver function tests, underlining the global burden of undiagnosed liver disease [4–6].
Although some patients with ELE have only transient abnormal liver function tests, the diagnostic work up in the vast majority of these patients is often cost intensive as it includes serological testing for both, common and rare diseases and often requires further imaging diagnostics or ultrasound [7–9]. Regarding the aetiology of underlying liver disease, non-alcoholic fatty liver disease, recently renamed as Metabolic dysfunction-associated steatotic liver disease (MASLD) is by far the most frequent undiagnosed disease, with a prevalence of up to 40 % in a recent study [10–12].
MASLD, previously termed non-alcoholic fatty liver disease (NAFLD), is indeed the most prevalent chronic liver disease worldwide, affecting more than one-third-of the adult population [13]. Its prevalence continues to rise in parallel with the global epidemics of obesity and type 2 diabetes and is projected to exceed 55 % of adults by 2040 [14,15]. Beyond its high prevalence, MASLD is clinically significant due to its association with both hepatic and extrahepatic complications, requiring not only a timely diagnosis, but also subsequently lifestyle interventions or medical treatment. Of note, MASLD-related HCC is currently the fastest rising cause of liver cancer and the most rapidly growing indication for orthotopic liver transplantation [16].
In terms of diagnosis, liver biopsy remains the diagnostic gold standard for assessing hepatic steatosis and fibrosis [17]. However, due to its invasive nature and limited feasibility in routine care, non-invasive imaging techniques such as conventional ultrasound are widely used, as routine ultrasound enables semi-quantitative grading of steatosis [18–21]. While its sensitivity and specificity for detecting moderate to severe steatosis are high, accuracy of ultrasound guided grading is markedly reduced for mild steatosis [21,22].
Beyond conventional B-mode ultrasound, recent advances in quantitative ultrasound techniques, including attenuation imaging (ATI) and the ultrasound-guided attenuation parameter (UGAP), enable a more objective assessment of hepatic steatosis. These attenuation-based methods quantify physical properties of liver tissue and have demonstrated strong correlations with MRI-derived proton density fat fraction, with improved sensitivity for the detection of mild steatosis. However, despite increasing clinical interest, availability and standardization of quantitative ultrasound techniques remain heterogeneous across clinical settings [23].
More advanced modalities such as magnetic resonance spectroscopy (MRS) and controlled attenuation parameter (CAP) offer superior diagnostic accuracy, but require additional devices and are cost intensive [24–27].
Due to these limitations, several non-invasive scores have been developed to estimate steatosis using clinical and laboratory data —such as the Hepatic Steatosis Index (HSI), the Fatty Liver Index (FLI), and the NAFLD Liver Fat Score (NAFLD-LFS) [26,28–30]. However, their diagnostic performance varies considerably between populations, underlining the need for external validation and potential recalibration in independent clinical settings to ensure reliable application before broader clinical use [31].
The aim of this study was to develop a practical, laboratory-based score for estimating the presence of hepatic steatosis, and to validate its diagnostic performance in a European real-world clinical cohort of patients with elevated liver enzymes, who presented to a University referral center for further diagnosis. Furthermore, the newly established score was tested against the established Hepatic Steatosis Index (HSI).
2Patients and Methods2.1Derivation cohortFor data acquisition and identification of significant variables, a retrospective cohort of n = 206 patients who had undergone liver biopsy for various clinical indications between 2019 and 2020 was analysed. The cohort included patients with a range of liver diseases, including non-alcoholic fatty liver disease (NAFLD; n = 50), autoimmune hepatitis (AIH; n = 25), alcohol-related liver injury (ALD; n = 19), drug-induced liver injury (DILI; n = 15), primary biliary cholangitis (PBC; n = 6), and idiopathic liver enzyme elevations (n = 26). The cohort was intentionally selected to represent a wide spectrum of steatosis severity, allowing for robust score calibration.
All biopsies were obtained via sonographically guided or laparoscopic liver puncture as previously described [32]. The NAFLD Activity Score (NAS) was used to assess degree of steatosis, hepatocellular ballooning and inflammation on hematoxylin and eosin (H&E)–stained liver biopsy sections as described by Kleiner et al. [33]. Only patients with a NAFLD Activity Score (NAS) ≥ 4 were classified as "steatosis-positive" for score derivation [34,35]. To ensure reliability and minimize interobserver variability, liver biopsies were independently assessed by three hepatopathologists, and consensus was reached in cases of disagreement.
Demographic, anthropometric and selected routine laboratory parameters were used as independent variables.
2.2Validation cohortFor validation of the score, patients with unexplained ELE who presented to the hepatology outpatient clinic of the First Department of Medicine of the University Medical Center Hamburg-Eppendorf (UKE) between January 1, 2019, and December 31, 2020 were analysed.
As part of the routine work-up procedure, all patients underwent a standardized laboratory and non-invasive diagnostic assessment, including whole blood count, blood chemistry including AST, ALT, γGT, alkaline phosphatase, glutamate dehydrogenase (GLDH), albumin, total protein concentration, bilirubin, creatinine, c-reactive protein (CRP) and coagulation test (international normalized ratio (INR), quick value). Testing for autoimmune liver disease was carried out as described in recent guidelines [36].
Testing for viral hepatitis (A, B, C and E), as well as screening for hemochromatosis, Wilson´s disease and celiac disease (anti-transglutaminase antibodies), and alpha-1 antitrypsin deficiency, was performed. A blood lipid profile, as well as glycated hemoglobin (HbA1c), was obtained.
A routine ultrasound of the abdomen was carried out in every patient. Steatosis was graded according to established criteria: grade 0, normal echogenicity of the right hepatic lobe compared with the renal cortex; grade 1, mild diffuse increase in hepatic echogenicity with preserved visualization of the diaphragm and intrahepatic vessels; grade 2, moderate diffuse echogenicity with slightly impaired visualization of intrahepatic vessels and diaphragm; and grade 3, marked echogenicity with poor or absent visualization of intrahepatic vessels, diaphragm, and the posterior aspect of the right hepatic lobe. Sonographic assessments were performed by experienced examiners.
In addition, liver stiffness measurement (fibroscan ®) including assessment of the controlled attenuation parameter (CAP) was carried out in every patient.
For screening of alcohol misuse, urinary ethyl glucuronide and carbohydrate-deficient transferrin (CDT) in the serum were obtained. Patients’ drinking habits, medical history and co-medication were also assessed on first presentation to the outpatient clinic.
Patients’ data were retrospectively collected by reviewing the electronic patient record system and were analysed post hoc.
2.3Score development and validationFor score development, identification of potential predictors was performed using univariate analysis. Student’s t-test was applied for continuous variables and the chi-square test for categorical variables. Variables with a p-value < 0.10 in the univariate analysis were subsequently included in the multivariate logistic regression model. Based on the identified significant variables, a regression equation was calculated. The exponential function of the equation was modified and adapted, the regression coefficients were linearly scaled by dividing by the smallest coefficient.
Diagnostic performance for the detection of hepatic steatosis was assessed using the area under the receiver operating characteristic curve (AUROC), as well as sensitivity, specificity, and positive and negative likelihood ratios. Optimal cut-off values were determined via ROC analysis.
2.4Comparison of the HABIT score with the Hepatic Steatosis IndexThe Hepatic Steatosis Index (HSI) was first described by Lee et al. in 2010, established in a South Korean cohort of over 10,000 individuals who underwent routine health check-ups [20]. HSI was designed as a simple, non-invasive screening tool for NAFLD, incorporating the following variables: body mass index (BMI), the ALT/AST ratio, sex, and the presence of diabetes mellitus . The Hepatic Steatosis Index (HSI) is calculated using the formula: HSI = 8 × (ALT/AST) + BMI, with an additional 2 points added if the patient is female and/or has type 2 diabetes [37].
In the initial report, a HSI score < 30 ruled out NAFLD with high sensitivity (92.5 %), while values > 36 indicated NAFLD with high specificity (92.4 %). Values between 30 and 36 were considered diagnostically indeterminate. Using these thresholds, 86.3 % of individuals in the validation cohort were correctly classified.
2.5Statistical analysisFor all cohorts, patient demographic data and clinical course were obtained by reviewing digital medical files obtained from the electronic hospital data system. Categorical variables were described in terms of percentages and frequencies; continuous variables were described in terms of median with minimum-to-maximum range. To compare non-parametric data, the Kruskal-Wallis test was used to compare non-normal distributed continuous variables between several groups.
For pairwise group comparison, the Bonferroni-corrected Mann-Whitney test was used. Variables with a p-value < 0.10 in the univariate analysis were subsequently included in the multivariate logistic regression model.
All statistical analyses were performed using IBM SPSS Statistics (Version 29.0.2.0) and GraphPad Prism (Version 10.4; GraphPad Software).
2.6Ethical approvalAll patients data were retrospectively recruited via reviewing electronic medical charts. All data were subsequently analysed anonymously according to local law (Hamburgisches Krankenhausgesetz (HmbKHG, established April 17th 1991, last revision December 17th 2018).
3Results3.1Derivation of the HABIT scoreIn the derivation cohort, a total of n = 206 patients were analysed, median age was 40 years (range: 18–80), with 42 % of patients being male. The median body mass index (BMI) was 25.2 kg/m² (range: 15.8–58.6), with the highest median observed in the NAFLD subgroup (28.4 kg/m²). Non-invasive assessment of fibrosis and steatosis using FibroScan® measurements yielded a median liver stiffness measurement (LSM) of 7.9 kPa (range: 2.9–75.0) and a CAP of 243.5 dB/m (range: 100–400). Demographic characteristics and baseline laboratory findings are depicted in Table 1.
Baseline characteristics of the derivation cohort.
Abbreviations: ALD, alcoholic liver disease; AIH, autoimmune hepatitis; ALT, alanine aminotransferase; AP, alkaline phosphatase; AST, aspartate aminotransferase; BMI, body mass index; CAP, controlled attenuation parameter; CRP, C-reactive protein; DILI, drug-induced liver injury; GGT, gamma-glutamyltransferase; Hb, hemoglobin; INR, international normalized ratio; LSM, liver stiffness measurement; MCV, mean corpuscular volume; NAFLD, non-alcoholic fatty liver disease; PBC, primary biliary cholangitis. Subgroup ‘Others’, including hemochromatosis, cirrhosis, hemangioma, focal nodular hyperplasia, hepatic adenoma, hepatitis C virus infection, hepatitis B virus infection, and Wilson's disease.
Among the cohort, n = 40 patients had liver biopsies with NAS ≥ 4 and were classified as having clinically significant hepatic steatosis, while 166 patients with NAS < 4 or absence of liver steatosis served as the control group.
Univariate analysis revealed significant differences between both groups with respect to BMI (p = .023), HbA1c (p = .001), triglyceride levels (p = .018), LDL/HDL ratio (p = .042), and the presence of arterial hypertension (p = .014). Multivariate logistic regression analysis was conducted, including all variables with p < 0.10 in the prior univariate analysis. In this model, HbA1c value (odds ratio (OR) 2.398; 95 % confidence interval (CI) 1.244–4.623; p = .009), triglyceride levels (OR 1.015; 95 % CI 1.005–1.025; p = .004), and BMI (OR 1.044; 95 % CI 0.997–1.093; p = .069) emerged as independent predictors of steatosis (see Table 2). Variables with p ≥ 0.10 were excluded from the final model.
Multivariate logistic regression analysis to identify independent predictors of hepatic steatosis.
Abbreviations: ALT, alanine aminotransferase; ALT/AST ratio, ratio of alanine aminotransferase to aspartateaminotransferase; AST, aspartate aminotransferase; BMI, body mass index; γ-GT, gamma-glutamyltransferase; HbA1c,glycated hemoglobin (hemoglobin A1c); HDL, high-density lipoprotein; LDL, low-density lipoprotein; LDL/HDL ratio,ratio of low-density lipoprotein to high-density lipoprotein; MCV, mean corpuscular volume; OR, odds ratio; * indicates final model variables.
The following logistic regression equation was used to calculate the probability of steatosis:
(HbA1c in %, triglycerides in mg/dL, BMI in kg/m²)
To provide practicability in daily clinical practice, the exponential function was omitted, and regression coefficients were linearly scaled by dividing by the smallest coefficient (0.0144), yielding the following integer values:
After the adaptation of the integer values, a simplified score based on the three variables HbA1c, BMI and triglycerides (HABIT) was calculated as:
Diagnostic performance of the adapted score was evaluated in the derivation dataset with an AUROC of 0.83, comparable to the original logistic regression model (seeFig. 1A). To further assess the diagnostic performance of the HABIT score using ultrasound as reference within the derivation cohort, sonographic data were analyzed where available. Out of n = 206 patients of the derivation cohort, n = 5 individuals were excluded due to non-evaluable ultrasound examinations. In the remaining 201 patients, the HABIT score yielded an AUROC of 0.82 (95 % CI 0.76–0.88) for the detection of ultrasound-defined hepatic steatosis which was comparable to the histology-based performance observed in the full derivation cohort.
Receiver operating characteristic (ROC) curve of the HABIT Steatosis Score in the training cohort (A). The score was evaluated against histologically confirmed hepatic steatosis (NAS≥4), AUROC was 0.83. The diagonal line represents a non-discriminative model (AUROC = 0.5). Comparison of diagnostic performance of the HABIT Steatosis Score and the Hepatic Steatosis Index (HSI) in the validation cohort (B). Ultrasound-based detection of hepatic steatosis was used as the reference. The HABIT Score (blue line) achieved an AUROC of 0.81, while the HSI (red line) yielded an AUROC of 0.77.
For external validation, a total of n = 1,461 patients with enzymes were analysed retrospectively, n = 813 were excluded because of missing data, inconclusive diagnosis or other reasons than ELE for presentation to the center, resulting in a cohort of n = 648 patients. Characteristics of the validation cohort are depicted in Table 3. Among the cohort, a total of n = 328 patients (51 %) were diagnosed with liver steatosis based on sonographic findings.
Baseline characteristics of the validation cohort.
Abbreviations: ALD, alcoholic liver disease; AIH, autoimmune hepatitis; ALT, alanine aminotransferase; AP, alkaline phosphatase; AST, aspartate aminotransferase; BMI, body mass index;, CAP, controlled attenuation parameter; CRP, C-reactive protein; DILI, drug induced liver injury; GGT, gamma-glutamyltransferase; Hb, hemoglobin; INR, international normalized ratio; LSM, liver stiffness measurement; MCV, mean corpuscular volume, NAFLD, non-alcohlic fatty liver disease; PBC, primary biliary cholangitis.
After applying the HABIT score in the validation cohort, the median score values differed markedly between steatosis positive and negative patients: 608.7 (range: 428.4–1580.7) in steatosis-positive and 506.0 (range: 361.5–1205.5) in steatosis negative patients. The AUROC for detecting hepatic steatosis was 0.81 (95 % CI: 0.78–0.84), see Fig. 1B When adjusting cut-off thresholds, a value of < 490 ruled out hepatic steatosis with a sensitivity of 93.6 % (95 % CI: 90.4–95.8) and a negative likelihood ratio of 0.156 (95 % CI: 0.101–0.241). Conversely, a score of > 630 identified steatosis with a specificity of 92.5 % (95 % CI: 89.1–94.9) and a positive likelihood ratio of 5.732 (95 % CI: 3.825–8.589; see Table 4)
Diagnostic performance parameters of the HABIT Steatosis Score.
Data are presented as percentages or means with 95 % confidence intervals in brackets.
According to these thresholds, 317 patients (49 %) had scores either below 490 or above 630. Among these, 85.8 % (n = 272) were correctly classified.
3.3Correlation of HABIT score and grade of liver steatosisA grading of steatosis according to sonographic findings was available for n = 612 patients in the validation cohort. Differences in score values across steatosis grades were assessed by using non-parametric statistical tests. A significant correlation between median score values and degree of steatosis was observed: median grade 0: 506; grade 1: 573; grade 2: 631; grade 3: 677; p < 0.001. Subsequent pairwise comparisons using the Bonferroni-corrected Mann-Whitney U test demonstrated significant differences between grade 0 and 1, grade 0 and 2, grade 0 and 3, grade 1 and 2 (all p < 0.001), as well as between grade 1 and 3 (p = 0.0019). No significant difference was observed between grades 2 and 3 (p = 1.0), indicating a limited discriminatory capacity of the score at more advanced stages of steatosis (Fig. 2A). In addition, among the 850 patients with available CAP measurements, linear regression demonstrated a significant positive association between CAP and the HABIT Score (B = 1.02; SE = 0.096; β = 0.345; p < 0.001; Fig. 2B).
Correlation between HABIT Steatosis Score and degree of sonographically graded steatosis (A). Value of the HABIT increases significantly in patients with advanced hepatic steatosis (median: grade 0 = 506; grade 1 = 573; grade 2 = 631; grade 3 = 677). The Kruskal-Wallis test showed a significant overall difference (p < 0.001). Post hoc tests confirmed significant differences between all grades (all p < 0.001), except between grade 2 and 3 (p = 1.0). The HABIT score significantly correlates with the CAP value (B). Linear regression demonstrated a significant positive association between CAP and the HABIT Steatosis Score (B = 1.02; SE = 0.096; β = 0.345; p < 0.001). The model was statistically significant (F(1847)=114.64; p < 0.001) and explained 11.9 % of the variance in the HABIT Steatosis Score (R2=0.119). On average, each 1 dB/m increase in CAP was associated with a 1.02-point increase in the HABIT Steatosis Score.
The HSI score was calculated for patients in the validation cohort as previously described [20,22]. Within the validation cohort, n = 85 patients (13.6 %) had diabetes, mean ALT and AST values were 88.5 ± 84.3 U/L and 53.0 ± 58.4 U/L, respectively, corresponding to a mean ALT/AST ratio of 1.67. (seeTable 3).
The AUROC for detecting steatosis was 0.77 (95 % CI: 0.73–0.80) for HSI and 0.81 for the HABIT score (seeFig. 1B)
When applying the HSI, n = 23 (3.5 %) had a value < 30, n = 91 (13.7 %) were classified at an intermediate range between 30- 36 points while n = 548 (82.8 %) had a score of > 36. At the lower cut-off point (< 30.0), the sensitivity for detecting steatosis was 98.2 % [95 % CI: 96.1–99.2] and specificity was 5.1 % [95 % CI: 3.2–8.0]. The negative predictive value for patients with score value < 30 was 73.9 % [95 % CI: 53.5–87.5], and the negative likelihood ratio was 0.357 [95 % CI: 0.143–0.895]. At the upper cut-off point (> 36.0), the sensitivity remained high at 93.0 % [95 % CI: 89.7–95.3] with a specificity of 27.3 % [95 % CI: 22.8–32.4]. The positive predictive value was 55.8 % [95 % CI: 51.7–59.9], with a positive likelihood ratio of 1.280 [95 % CI: 1.191–1.376].
4DiscussionOver the past years, the prevalence of unexplained ELE in the general population has increased markedly worldwide. The rising number of patients reflects both an increasing incidence of underlying liver diseases and, in parallel, a growth in the demand for comprehensive diagnostic approaches. Despite the broad spectrum of primary and secondary liver diseases, in the vast majority of cases, the underlying liver disease is related to NAFLD or MASLD, respectively [10,38–40]. Although MASLD is a reversible disease and potent medical therapies are already available, missing the diagnosis may lead to disease progression with development of fibrosis, liver cirrhosis or even hepatocellular carcinoma, elucidating the urgent need for an early, reliable and cost-effective diagnosis of liver steatosis in the primary care setting [41,42].
In this larger cohort study of patients with previously unexplained ELE presenting at our referral center, NAFLD was the most prevalent underlying liver disease. As most patients were referred from primary care for further diagnostic evaluation, these findings suggest both insufficient awareness of fatty liver disease and a diagnostic gap in the primary care setting.
To meet this need, we developed the HABIT Score, a simple, laboratory-based score relying on HbA1c, body mass index (BMI), and serum triglycerides. All these variables are easy to obtain in the outpatient care setting and well-known surrogates for patients’ metabolic status: HbA1c reflects long-term glycemic status and is strongly correlated with insulin resistance—a core mechanism in MASLD pathogenesis [43]. Although HbA1c can be modified over relatively short periods through lifestyle interventions or pharmacological treatment, it reflects longer-term glycaemic exposure and chronic metabolic dysfunction. In the context of MASLD, HbA1c therefore represents a clinically meaningful marker of metabolic risk rather than an acute disease parameter. Nevertheless, dynamic changes in metabolic control may influence score values over time and should be considered when interpreting HABIT results, particularly in longitudinal settings.
Triglycerides reflect dysregulated lipid metabolism—a hallmark of the metabolic syndrome [44]. BMI, while nonspecific, remains a pragmatic anthropometric index of adiposity and is commonly included in other steatosis scores such as the Fatty Liver Index (FLI) and the Hepatic Steatosis Index (HSI) [29,37,45].
The diagnostic performance of the HABIT Score is comparable to that of existing non-invasive models: In the biopsy-confirmed training cohort, the score achieved an AUROC of 0.83, and in the external validation cohort, using ultrasonography as a reference, it retained a solid AUROC of 0.81. These results are in line with those reported for the NAFLD-Liver Fat Score (AUROC ∼0.86), HSI (∼0.81), and FLI (∼0.84) in their respective derivation studies [29,30,37].
However, several studies have demonstrated that these scores perform less accurately in unselected, population-based cohorts [46–48]. In a pivotal external validation study by Jung et al. (2020), five steatosis scores were evaluated in a community-based Korean cohort (n = 1301) using MRI as the reference standard . While NAFLD-LFS showed the best performance (AUROC 0.72), both FLI and HSI performed worse than in their original studies (AUROCs 0.68 and 0.69, respectively) [49]. These findings highlight the importance of external validation and underscore how score performance may vary depending on population characteristics, disease prevalence, and the reference standard.
In our study, the Hepatic Steatosis Index (HSI) showed limited diagnostic utility. Although originally developed in a large East Asian screening cohort, its performance in our European validation cohort was markedly reduced. All patients in our cohort presented with elevated liver enzymes, which directly impacts the HSI formula that incorporates the ALT/AST ratio. The mean ALT/AST ratio in our population was 1.67, consistent with a typical MASLD profile. Since the HSI scales proportionally with this ratio, the vast majority of patients exceeded the upper diagnostic threshold (> 36), regardless of actual steatosis status. As a result, the score distribution was heavily skewed and specificity substantially impaired. Importantly, we deliberately evaluated the HSI in this cohort to assess whether a score developed in a general screening population retains diagnostic accuracy in a more clinically relevant setting—namely, patients with unexplained elevation of liver enzymes. This reflects a common scenario in hepatology and primary care, yet one in which many established scores have not been externally validated. Our findings suggest that transaminase-based indices such as the HSI may have limited applicability in such populations, and underscore the need for tools specifically adapted to these clinical conditions.
In contrast, the HABIT Score was specifically developed for this clinical scenario. Its exclusive reliance on metabolic parameters—HbA1c, BMI, and triglycerides—makes it independent of liver enzyme levels which can remain within normal limits in a large proportion of patients with steatosis [50] and thus robust to the type of diagnostic inflation seen with the HSI. In our study, the HABIT Score showed a more balanced distribution across the scoring range and achieved good diagnostic accuracy in both derivation and validation cohorts. Additional analyses within the derivation cohort demonstrated comparable HABIT discrimination when ultrasound rather than histology was used as the reference standard.
As with other models, a diagnostic gray zone remains a limitation of the HABIT Score. In the validation cohort, 51 % of patients fell within the intermediate range between the two proposed cut-offs (490–630) and could not be definitively categorized. However, 23.5 % of patients were below the lower cut-off (<490), with a low probability of steatosis (NPV 86.2 %), and 25.5 % exceeded the upper cut-off (>630), indicating a high likelihood of steatosis (PPV 85.5 %).
While overall discrimination as assessed by AUROC was modestly higher for the HABIT score compared with HSI (0.81 vs. 0.77), clinically relevant differences emerged when applying predefined cut-off thresholds. For HSI, only 3.5 % of patients had values below the lower cut-off (<30); consequently, the vast majority of patients (82.8 %) were classified as positive using the upper cut-off (>36), resulting in very limited specificity (27.3 %) and a low positive predictive value of 55.8 %. From a clinical perspective, this effectively translates into near-universal classification as steatosis-positive and offers little potential to reduce downstream imaging. In contrast, application of the HABIT Score allowed approximately one-quarter of patients to be confidently ruled out at the lower cut-off. Using predefined thresholds, 49 % of patients were classified either below 490 or above 630, with 85.8 % of these classifications being correct. Notably, patients below the lower HABIT cut-off could be excluded with high sensitivity, enabling avoidance of sonographic assessment in a clinically meaningful proportion of cases.
Taken together, despite similar AUROC values, HABIT provides superior triage utility compared with HSI by enabling both effective rule-out and rule-in strategies. In this context, the HABIT score may serve as a practical decision-support tool in routine clinical work-up of patients with unexplained elevated liver enzymes, facilitating more targeted use of imaging resources.
Moreover, the HABIT Score was able to significantly distinguish between different grades of hepatic steatosis as detected by ultrasound- an important feature not routinely achieved by non-invasive scores. Although the score was unable to differentiate between grades 2 and 3, this distinction is less clinically critical for early detection and population-level screening.
There are several limitations of the study that need to be addressed: While liver biopsy served as the reference in the training cohort, ultrasonography was used in the external cohort. Conventional ultrasound has limited sensitivity for detecting mild steatosis, and a certain degree of misclassification may occur. This should be taken into account when interpreting the reported AUROC values of the HABIT score. The training cohort was composed of patients with elevated transaminases referred for biopsy, potentially introducing selection bias. Both the derivation and validation cohorts were derived from a European population and exclusively included patients with unexplained elevated liver enzymes. Therefore, the applicability of the HABIT score to population-based screening settings, non-ELE cohorts, and other ethnicities remains uncertain and requires further external validation.
In addition, the retrospective nature of the study limits causal inference and may be subject to information bias or incomplete data capture. Moreover, although BMI did not reach statistical significance in the multivariate model (p = 0.069), it was retained for clinical and metabolic consistency in our score calculation.
5ConclusionsIn summary, the HABIT Score offers a metabolically coherent, pragmatic, and accessible tool for detecting hepatic steatosis. It uses only three easy-to-obtain standard clinical parameters, distinguishes between early steatosis grades and performs better than existing scores in a cohort of European patients with elevated liver enzymes.
Its development was based on a biopsy-confirmed cohort using histology as the reference standard, further supporting the robustness of the score. Its balanced distribution and alignment with the MASLD framework support its use in both clinical practice and research.
The implementation of this score is expected to improve the detection of hepatic steatosis in the primary care setting.
FundingThis research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.
Author contributionsE.H., V.W., S.B. and K.H: data collection. E.H., T.F., statistical and data analysis. T.F.,S.P.: idea of the study, data analysis. E.H., T.F.: writing the manuscript, S.P., S.H., A.W.: critical review, proofreading.
All authors declare no conflict of interests.
