Using Phosphatidylethanol as an Adjunct to Self-Reported Alcohol Use Improves Identification of Liver Fibrosis Risk

Abstract

INTRODUCTION:

Accurate assessment of alcohol use informs prevention and management of liver disease. We examined whether phosphatidylethanol (PEth, an alcohol metabolite) blood concentrations are associated with liver fibrosis risk independently of self-reported alcohol use, among persons with and without HIV.

METHODS:

We pooled individual-level data from 12 studies from the United States, Russia, Uganda, and South Africa with PEth, Alcohol Use Disorders Identification Test-Consumption (AUDIT-C), and fibrosis-4 (FIB-4) measurements. We conducted mixed-effects logistic regression of the relationship between PEth and AUDIT-C as continuous variables (after checking linearity), with high FIB-4 (≥2.67). We divided PEth (range 0–1,000) by 83.3 to put it on the same scale as AUDIT-C (0–12) to directly compare odds ratios. Adjusted models included sex, race/ethnicity, age, body mass index, HIV, and virologic suppression status.

RESULTS:

Among 4,644 adults, the median age was 49 years (interquartile range [IQR]: 40–55), 998 (21%) were female, and 3,520 (76%) were living with HIV, among whom 2,386 (68%) were virologically suppressed. Median PEth was 13 ng/mL (IQR: <8–132.0) and median AUDIT-C was 3 (IQR: 1–6); 554 (12%) had high FIB-4. The adjusted odds ratios per 83.3 ng/mL difference in PEth and one-unit difference in AUDIT-C with high FIB-4 were 1.15 (95%CI: 1.08–1.22) and 1.03 (95%CI: 1.00–1.07), respectively. Findings were similar when PEth and AUDIT-C were treated as categorical variables.

DISCUSSION:

PEth was independently associated with high FIB-4, with a larger odds ratio than that of the association of AUDIT-C. The use of PEth may improve identification of alcohol use and liver fibrosis prevention and management.

INTRODUCTION

Alcohol use is a leading cause of liver disease onset, morbidity, and mortality (1). Fibrosis, a marker of liver disease severity and a strong predictor of clinical outcomes (2), is exacerbated by alcohol use in most forms of liver disease (1,3–5), including in alcohol-associated liver disease (ALD) (1), metabolic dysfunction-associated steatotic liver disease (MASLD) (6–9), and viral hepatitis (5). Though fibrosis can be reversed if detected early, it is typically asymptomatic before advanced stages (2). Fibrosis screening in asymptomatic persons is generally limited to those considered high risk, with alcohol use a key indicator of risk (10,11). Thus, accurate assessment of alcohol use is critical to prevention, diagnosis, and management of liver disease.

Self-report is key to clinical assessment of alcohol use given its low cost and wide accessibility, yet alcohol consumption is often under-reported due to interviewer skill (12), recall bias, or social desirability bias (13). In primary care, alcohol use under-reporting can result in a missed opportunity for early liver fibrosis screening or referral to alcohol use disorder treatment. Among persons with liver disease, underreporting could result in misclassification of disease as alcohol-associated or not, which in turn could influence liver disease management, including choice of prognostic scoring system, linkage to alcohol use disorder care, and ongoing alcohol use monitoring (14,15). Phosphatidylethanol (PEth), an ethanol metabolite that is only formed in the presence of alcohol, is highly sensitive and specific in detecting unhealthy alcohol use, including among persons with liver disease (16–19). PEth concentrations are highly correlated with levels of alcohol intake in the previous month (20). Furthermore, because PEth concentrations also reflect individual differences in alcohol metabolism, concentrations may indicate alcohol-related harm for an individual. Given these properties, PEth is emerging as a promising adjunct to self-report in many clinical settings (21).

We sought to determine whether PEth concentrations are associated with liver fibrosis risk independently of self-reported alcohol consumption in a large, diverse, international sample of adults. While histological assessment of liver biopsy is the gold standard measure of fibrosis stage, non-invasive measures, including transient elastography and low-cost blood tests, are increasingly recommended for screening and longitudinal monitoring (2,11,22). We used the fibrosis-4 (FIB-4) index because it can be calculated from routinely collected clinical blood tests, is strongly associated with biopsy-confirmed liver fibrosis (23,24), and is a stronger predictor of fibrosis than other non-invasive blood tests in patients with ALD (6,25), MASLD (26,27) hepatitis B virus (28), and hepatitis C virus (24). High FIB-4 also predicts liver disease progression (29) and mortality (29–31). In a cohort of persons living with HIV in Louisiana, PEth concentrations were more strongly associated with high FIB-4 than self-reported alcohol use (32). We now examine this association in a highly diverse and international sample of persons with and without HIV from 12 studies contributing to a previous individual participant data meta-analysis (33). We aimed to determine whether PEth concentrations could provide an advantage to self-report alone in identifying persons at risk of liver fibrosis.

METHODS

This study is a cross-sectional analysis using data from our previous individual participant data meta-analysis of 21 studies that examined the sensitivity of PEth for detecting unhealthy alcohol use (33). Study inclusion criteria for the meta-analysis were: PEth concentrations (16:0/18:1 homologue with ≤8 ng/mL as the lower limit of quantification) and self-reported alcohol use collected at the same study visit; self-report measured by the Alcohol Use Disorders Identification Test-Consumption (AUDIT-C) (34,35) or an alternative method that allowed for calculation of AUDIT-C scores; and at least 30 observations for which AUDIT-C scores were ≥3 for women or ≥4 for men. We excluded studies that focused on populations for whom there may be reasons to misreport alcohol, such as studies of prisoners, persons in alcohol rehabilitation, or clinical trials in which eligibility depended on self-reported alcohol use without biomarker confirmation. For this analysis, we included the 12 studies that collected aspartate transaminase (AST), alanine transaminase (ALT), and platelet counts (19,36–46). If, by design, these tests were only collected on a subset of participants within a study, we only included the subset with the relevant data. Only baseline visit data were included, as most studies did not conduct these tests longitudinally. Since the time of the previous meta-analysis, 2 studies completed additional PEth testing and we have added these observations to the compiled dataset. The 12 included studies were conducted in the United States, Russia, Uganda, and South Africa and encompassed adults living with and without HIV (see eTable 1, Supplementary Digital Content 1, http://links.lww.com/AJG/D452).

Study measures

FIB-4 scores were calculated as (age [years] * AST [U/L])/(platelet count [10⁹/L] * √ALT [U/L]) (23,24). The primary study outcome was FIB-4 ≥2.67. This threshold optimizes specificity in predicting significant fibrosis for MASLD (27) and was strongly associated with all-cause mortality in a large, representative sample from Korea (30) and adult patients with or at risk of MASLD in the United States (29).

The primary exposure was alcohol use, measured with PEth and AUDIT-C. We used AUDIT-C as a continuous variable from 0 to 12. For PEth, we set values > 1,000 ng/mL (3% of the sample) to 1,000 because PEth is only considered linear up to 1,000 (47). We then divided PEth values by 83.3 to achieve the same range of values (0–12) as the AUDIT-C. This scaling of PEth retained it as a continuous variable but enabled a comparison of the magnitudes of odds ratios for AUDIT-C and PEth. We examined the linearity assumptions for associations of PEth and AUDIT-C with the log-odds of high FIB4 as described under Statistical Analysis. In all studies, blood samples were analyzed for PEth levels by the United States Drug Testing Laboratories using previously published methods (47).

Potential effect modifiers or confounders included age in years, sex assigned at birth (only one included study collected gender identity), race/ethnicity (defined as African for studies based in Africa, African American, White, or Other), body mass index (BMI; kilograms/meters²), and HIV status (negative, living with HIV and virologically suppressed, or living with HIV and not suppressed or viral load unknown).

As a secondary analysis, we also considered categorical measures of low-, medium-, and high-risk alcohol use by PEth and AUDIT-C. Using recently published PEth thresholds, we classified alcohol use as low-risk (<20 ng/mL), medium-risk (20–199 ng/mL), and high-risk (≥200 ng/mL) (48,49). For AUDIT-C, low-risk scores are established as <3 for women and <4 for men (35). Because the threshold between medium- and high-risk AUDIT-C is uncertain, we used a data-driven approach to set the AUDIT-C threshold, by first determining the proportion of the sample with PEth ≥200, then determining the AUDIT-C score that classified the closest proportion of the sample as high risk. Using this approach, the appropriate threshold for high-risk drinking for AUDIT-C was a score of ≥7.

Statistical analysis

Demographic characteristics were summarized overall and by FIB-4 level. We assessed the correlation between PEth and AUDIT-C as continuous measures with the Spearman correlation coefficient (rho). In this multistudy pooled analysis, we estimated associations with the log odds of high FIB-4 using mixed effects logistic regression with a random intercept for study. We first assessed whether the relationship between each continuous variable (age, BMI, PEth, AUDIT-C) and the log odds of high FIB-4 was linear by examining lowess curves. If any curvature was suggested, we compared logistic regression models with linear vs quadratic predictors, and determined linear forms had the best fit according to Akaike's Information Criteria. We then estimated associations of PEth and AUDIT-C as linear predictors of the log-odds of high FIB-4 in separate models, followed by models that included both variables. Adjusted models included prespecified covariates of age, sex, HIV status, race/ethnicity, and BMI (35,48,49). We speculated that possible differences in alcohol metabolism by BMI, race/ethnicity, or sex could modify the association between PEth and FIB-4 and therefore tested for effect modification with interaction terms between PEth and these variables, in models adjusted for age. We conducted stratified analyses for any global P-values for interactions <0.05. We also assessed whether HIV status or age modified the association between PEth and FIB-4 in exploratory analyses. We included covariates that were not found to be effect modifiers in adjusted models. We conducted similar analyses using PEth and AUDIT-C as categorical variables, as described above.

Sensitivity analyses

(1) To examine the potential bias in complete case analyses we conducted unadjusted analyses in the sample with complete data. (2) We modeled the log odds of FIB-4 ≥3.25 as high risk, a threshold used for persons with viral hepatitis (23), and to predict cirrhosis. (3) Because age is a component of FIB-4 and adjusted models included age, we considered the AST to platelet ratio (APRI) as another validated serum marker of liver fibrosis, calculated as AST (U/L)/(upper limit of the normal range) × 100/platelet count (10⁹/L). We used 30 as the upper limit of the normal for AST and classified APRI values ≥ 1.5 as having a high risk of significant fibrosis (50). (4) The New Orleans Alcohol Use in HIV (NOAH) Study, among the 12 studies in this analysis, previously examined the association between PEth and FIB-4 (32). To ensure these data did not drive our overall findings, we also estimated our multivariable model excluding this study. (5) Because the time frame for alcohol use in AUDIT-C varied by study (eTable 1 [see Supplementary Digital Content 1, http://links.lww.com/AJG/D452]: 3 months, 1 year, unspecified), we conducted analyses stratified by AUDIT-C time frame.

All analyses were conducted in STATA version 18.0. The analysis protocol was preregistered (51).

Ethics

All studies were reviewed and approved by their respective ethical review boards, and only de-identified data were provided to the individuals conducting the analysis (P.M.M., J.H.) at the University of California San Francisco. The University of California San Francisco ethical review board certified this secondary data analysis as exempt according to the Common Rule.

RESULTS

Among 4,708 individuals, 64 (1%) were missing either PEth or AUDIT-C and were excluded from this analysis. Of the 4,644 included, the median age was 49 years (interquartile range [IQR]: 40–55), 998 (21%) were female, and 3,520 (76%) were living with HIV among whom 2,386 (68%) were virologically suppressed (Table 1). Overall, 229 (5%), 1,261 (27%) and 3,154 (68%) were from Russia, Africa, and the United States, respectively. Median PEth was 13 ng/mL (IQR: <8.0–132.0) and median AUDIT-C was 3 (IQR: 1–6). PEth and AUDIT-C were moderately correlated (Spearman's rho 0.61, P-value <0.001). Participant characteristics within each of the 12 studies are reported in Table 2, Supplementary Digital Content 1, http://links.lww.com/AJG/D452. Overall, 554 (12%) had high FIB-4 scores (≥2.67).

Table 1.

Participant characteristics overall and distributions across FIB-4 levels

Unadjusted associations between alcohol use and high FIB-4 (≥2.67)

On a continuous scale, each 83.3 ng/mL increase in PEth concentration (analogous to a 1-point increase in AUDIT-C) was associated with 1.15 times the odds of high FIB-4 (95%CI: 1.10–1.21; Table 2). Each unit increase in AUDIT-C was associated with 1.06 times the odds of high FIB-4 (95%CI: 1.03–1.09). When categorized, high alcohol consumption was associated with increased odds of high FIB-4; the magnitude of the association was greater when assessed with PEth ≥200 ng/mL (odds ratio [OR] 2.59, 95%CI: 1.96–3.42) than with AUDIT-C ≥7 (OR 1.67, 95%CI: 1.38–2.02).

Table 2.

Unadjusted associations of alcohol use and covariates with high FIB-4 (≥2.67)

Adjusted associations between alcohol use and high FIB-4 (≥2.67)

After adjusting for age, sex, race/ethnicity, BMI, and HIV status, each 83.3 ng/mL increase in PEth was associated with higher odds of high FIB-4 (aOR 1.16, 95%CI: 1.09–1.23; Table 3). Adjusting for the same covariates in a separate model, a single unit increase in AUDIT-C was also associated with higher odds of high FIB-4 (aOR 1.07, 95%CI: 1.04, 1.09). In the model including both PEth and AUDIT-C, the association between PEth and high FIB-4 was similar to the model with PEth alone (aOR 1.15, 95%CI: 1.08–1.22), while the association between AUDIT-C and high FIB-4 was attenuated (aOR 1.03, 95%CI 1.00–1.07).

Table 3.

Adjusted associations between continuous measures of alcohol use and high FIB-4 (≥2.67) overall and stratified by race/ethnicity

We found evidence of effect modification by race/ethnicity for the association of PEth with FIB-4 (P-value for interaction = 0.0005) but not by any other a priori effect modifiers (P-values for interaction with sex = 0.56 and BMI > 0.99). In the adjusted model, the association between PEth and high FIB-4 was attenuated in Africans (OR 1.06, 95%CI: 1.01–1.11), but did not significantly differ by any other race/ethnicity category (White OR: 1.29, 95%CI: 1.24–1.35; African American OR 1.17, 95%CI: 1.13–1.21; Other OR 1.33, 95%CI: 1.24–1.43). Africans were more likely to be living with HIV due to study design (92% vs 71%), more likely to be underweight (BMI<18.5, 20% vs 3%), younger (median 39 years vs 51), and less likely to have high FIB-4 (5% vs 15%) than non-Africans. We conducted post hoc analyses to explore age and HIV status as potential effect modifiers of the association between PEth and high FIB-4, but did not find effect modification.

In analyses of PEth and AUDIT-C categorized (Table 4), PEth >200 ng/mL was associated with increased odds of high FIB-4 (aOR 2.60, 95%CI 1.74, 3.89) in the full sample, while AUDIT-C ≥7 was not (aOR 1.27, 95%CI: 0.98–1.66). Among Africans, the association between PEth ≥200 ng/mL and high FIB-4 was attenuated (aOR 1.05, 95%CI: 0.35–3.13).

Table 4.

Adjusted associations between categorical measures of alcohol use and high FIB-4 (≥2.67) overall and stratified by race/ethnicity

Sensitivity analyses

First, BMI and HIV status were missing in 3.5% and 1.4%, respectively, and, in adjusted models including only studies from Africa, HIV-negative persons were omitted from the model due to perfect prediction (none with high FIB-4). To assess potential bias in complete case analyses, we estimated unadjusted associations between measures of alcohol use and high FIB-4 among those with complete data and in Africans among those living with HIV and did not observe meaningful differences compared with analyses in the full dataset. Next, we considered alternative outcomes (FIB-4 ≥3.25, APRI >1.5), and all results were consistent with the primary analysis (see APRI results in eTables 3 and 4, Supplementary Digital Content 1, http://links.lww.com/AJG/D452). Furthermore, when we excluded the NOAH study, which previously examined PEth and FIB-4, results did not notably differ. Finally, in analyses stratified by AUDIT-C time frame (3 months vs 1 year), results were comparable.

DISCUSSION

In a diverse international sample of over 4,000 adults, including 3,520 (76%) living with HIV, we found a strong association between PEth concentrations and high FIB-4 levels. The association persisted after adjusting for self-reported alcohol use and potential confounders. Self-report, on the other hand, had only a modest association with FIB-4, which was attenuated and no longer statistically significant after adjustment for PEth. These findings are likely conservative compared with clinical care, because self-report was obtained in research settings using the standardized AUDIT-C and assurance to participants regarding data confidentiality and may be less prone to under-report. These data suggest that PEth offers a significant clinical advantage to self-reported alcohol use, especially in populations vulnerable to liver disease, such as those with HIV, tuberculosis, or substance use.

Our results are consistent with recent findings from a cohort included in our analysis, the NOAH Study, which found a stronger association between PEth and FIB-4 than self-reported alcohol use (32). In addition, research conducted in Sweden has demonstrated an association between PEth concentrations and biopsy confirmed liver fibrosis among persons with or at risk of MASLD (7–9). Our research extends this work by using a large sample size and by examining the independent effects of PEth compared with self-report.

Biomarkers of alcohol use are increasingly recommended in liver disease management (52); however, incorporating PEth into clinical care is still uncommon. Recent evaluations of PEth implementation in liver transplantation clinics have noted significant disparities in detection of alcohol use between PEth and self-report in this high-stakes setting where transplant eligibility may depend on abstinence (16,21,53–55). Importantly, communication around discordant PEth and self-report may be challenging. Addiction experts suggest that biomarkers can support constructive discussions and improve disease management (56). In a qualitative study of PEth use in hypertension management, providers found that while communication around discordant results required sensitivity and care, the use of PEth generally improved open dialogue with patients about alcohol (57). Because of individual differences in alcohol metabolism, PEth concentrations may also signify alcohol-related harm for an individual; this concept may alleviate provider concerns about the challenge of communicating discordant PEth and self-report. Recommendations from a liver disease management team regarding discordant PEth and self-report emphasize the importance of consistent care team messaging, being aware of personal biases, and remaining compassionate to provide a safe space for open discussion (58).

The reasons for the attenuated associations between PEth and FIB-4 among Africans are unclear. Participants in African studies were younger, had lower BMI, and HIV was more prevalent (by design) compared with non-African studies, yet adjustment or stratification by these variables did not explain the attenuation. Metabolic factors may be more prevalent among US participants though we lacked these measures. A recent meta-analysis found lower FIB-4 and APRI thresholds may be needed to classify fibrosis risk among those with hepatitis B virus in sub-Saharan Africa (59), and authors suggested environmental factors or endemic infections may explain differences in the performance of FIB-4 in Africa. Hepatitis B and C are highly prevalent in sub-Saharan Africa, but we did not have consistent measures of viral hepatitis to assess the potential influence of this factor in our analysis. Notably, persons with viral hepatitis may receive more counseling to reduce alcohol use, which may increase social desirability bias in self-report.

Our study has some limitations. First, we did not have histology-confirmed liver fibrosis. We considered alternative measures of liver injury, such as markers of liver inflammation, representing a key alcohol-related pathway to fibrosis (60), transient elastography, and other blood-based noninvasive tests for fibrosis screening. The components of the FIB-4 index were the most widely available across multiple datasets, reflecting the wide accessibility of this measure. Still, though FIB-4 is a rigorous scoring system of fibrosis risk, it can be elevated in the absence of fibrosis. Heavy alcohol use can lower the platelet count (61), which would elevate both FIB-4 and APRI. Of note, FIB-4 predicts mortality (30,31,62) and cardiovascular events (62–64), potentially expanding the clinical significance of our findings beyond the risk to the liver. Second, we did not have cohorts representing Asia or Latin America, and within the United States, we had limited representation of Asian, Native American, and Latinx populations. Importantly, some Latinx populations have a higher prevalence of polymorphisms associated with an increased risk of MASLD and ALD (65). This was not a general population sample, rather, a compilation of cohorts with a variety of comorbidities such as trauma, HIV, and tuberculosis. Therefore, results may be most relevant to populations with existing comorbidities. Third, we did not have consistent metrics of viral hepatitis, which was likely higher in cohorts with HIV, liver disease, and substance use. We also did not have a measure of diabetes, an important risk factor for liver disease, which may influence FIB-4 performance (66). However, we would not expect these to change our conclusion about the value of adding PEth to self-reported alcohol consumption to enhance detection of fibrosis risk. Fourth, a potential limitation is the cross-sectional nature of our analysis.

Strengths of our study include the large sample size derived from an individual participant data meta-analysis. The sample is highly diverse in terms of geography, race/ethnicity, and comorbidities, including persons with HIV, tuberculosis, liver disease, traumatic injury, and US Veterans. Additional strengths include our consistent results across multiple sensitivity analyses and a rigorous assessment of multiple potential effect modifiers.

In conclusion, use of PEth as an adjunct to self-reported alcohol use has the potential to flag an otherwise unknown indication to screen for liver fibrosis at stages amenable to reversal, to facilitate appropriate diagnoses of liver disease accounting for alcohol use, and to inform liver disease management and transplantation care strategies. More work is needed to learn how to best incorporate PEth into clinical care.

CONFLICTS OF INTEREST

Guarantor of the article: Pamela M. Murnane, PhD and Judith A. Hahn, PhD.

Specific author contributions: P.M.M. and J.A.H. conceptualized the study and wrote the first draft of the statistical analysis plan and manuscript. P.M.M. conducted the analyses. J.A.H., M.A., G.C., T.F., K.R.C., A.C.J., T.W.K., E.K., K.A.M., P.M., W.R.M., B.M., V.L.R., K.S., S.S., and M.S.S. contributed previously collected data to this study. All co-authors substantially and critically reviewed and edited the concept, analysis plan, and manuscript.

Financial support: This work was supported by the following grants: NIH U01 AA026223 (J.A.H.), NIH U01 AA020776 (J.A.H.), NIH R01 AA018631 (J.A.H.), NIH K24 AA022586 (J.A.H.), NIH K01MH119910 (P.M.M.), NIH U01 AA026226 (G.C.), NIH R01 AI119037 (K.R.J.), NIH U01 AA020780 (K.S.-A.), NIH R01 DA016065 (M.S.S.), NIH U01 AA020797 (R.L.C.), NIH T32 DA017629 (V.L.R., MPIs: Lanza and Maggs), NIH P60 AA009803 (P.M.), NIH R01 AA017911 (S.S.), NIH K23 AA024503 (M.A.), NIH R24AA019661 (M.A.), NIH R01 DA051464 (M.A.), NIH P01 AA029545 (A.C.J.), NIH U24 AA020794 (A.C.J.), NIH U01 AA020790 (A.C.J.), NIH U01 AA020795 (A.C.J.), NIH U01 AA022001 (A.C.J.), NIH U10 AA013566 (A.C.J.), NIH K24 AA022523 (M.K.), NIH R01 AA029312 (M.K.), NIH R34MH122268 (Magidson/B.M.); NIH K23 AA031334 (L.Y.H.), NIH U01AA020784 (MPIs: Heeren and Kim) NIH U01 HL146242 (P.C.T.), NIH K24 AI108516 (P.C.T., NIH R01 DK109823 (P.C.T.).

Potential competing interests: None to report.

Study Highlights.

WHAT IS KNOWN

• ✓ Excessive drinking causes liver disease and worsens existing liver disease.

• ✓ Clinical assessment of alcohol use typically relies on self-report.

• ✓ Phosphatidylethanol (PEth) is a promising biomarker to more accurately assess alcohol use.

WHAT IS NEW HERE

• ✓ PEth concentrations were directly compared to self-reported alcohol use to identify liver fibrosis risk.

• ✓ PEth was independently and more strongly associated with high fibrosis risk than self-report.

• ✓ This study included over 4000 individuals from diverse cohorts from across the globe.

Supplementary Material

acg-120-1567-s001.docx ^{(54.6KB, docx)}

acg-120-1567-s002.pdf ^{(375.2KB, pdf)}