Impact of a Mediterranean Diet on Hepatic and Metabolic Outcomes in Non-alcoholic Fatty Liver Disease

The Medina Randomised Controlled Trial

Elena S. George; Anjana Reddy; Amanda J. Nicoll; Marno C. Ryan; Catherine Itsiopoulos; Gavin Abbott; Nathan A. Johnson; Siddharth Sood; Stuart K. Roberts; Audrey C. Tierney


Liver International. 2022;42(6):1308-1322. 

In This Article

Materials and Methods

Study Design and Participant Recruitment

This study is a parallel multicentre randomised controlled trial registered with the Australian New Zealand Clinical Trials Registry (ANZCTR) Trial ID: ACTRN12615001010583 and was conducted at three large academic liver centres in Melbourne, Australia with recruitment from January 2015 until March 2018, the final appointment was completed in April 2019, 12-month sustainability data was collected but is not reported herein. Recruitment ceased when the number of participants needed for the intrahepatic lipid's primary outcome was reached. The published protocol details the study design and methods.[25] Participants deemed eligible for the study following a screening and baseline visit conducted by a researcher were enrolled in the study. Participant eligibility included those aged >18 years who had a body mass index (BMI) between 20 and 39.9 kg/m2, with a diagnosis of NAFLD (without any evidence of other forms of liver disease), determined by routine ultrasound or biopsy; insulin resistance based on a HOMA IR score of >2; and at least one elevated serum ALT level (>20 U/L female, >30 U/L male) during the past 6 months. Exclusion criteria are detailed elsewhere.[25] Following informed consent and screening, participants were randomised 1:1, stratified by gender and presence of diabetes by computer-generated methodology. The random allocation sequence was completed by the statistician who assigned participants to one of the two groups. Participants had to meet the inclusion criteria defined in Figure 1. Eligible participants identified from liver clinic appointment lists who were interested in participating were provided with participant information and consent forms and written consent was obtained. As a result of the nature of the study which prescribed a diet to both groups the participants and dietitians administering the diets could not be blinded. However, all analysis was conducted by a statistician who was blinded to the study conditions. Human research ethics committee approval was obtained from all hospital sites. Approval was also obtained from La Trobe University human research ethics committee.

Figure 1.

MEDINA study inclusion and exclusion criteria

Diet Intervention

The LFD was based on dietary recommendations as set out by the Australian Dietary Guidelines and The Heart Foundation, advising dietary patterns that equated to approximately 30% of total energy from fat, 50% from carbohydrate and 20% from protein.[21,22] The MedDiet arm was based on a traditional Cretan diet composed of approximately 44% of energy from fat (>50% monounsaturated fatty acids), 33% from carbohydrates, 15%–20% from protein and up to 5% from alcohol, as described by George et al. where qualitative food group targets and example resources have also been described in detail.[26] Advice regarding alcohol consumption was to remain below 20 g per day as recommended by the National Health and Medical Research Council (NHMRC).[27] Participants in both dietary intervention groups had three face to face appointments, at baseline and weeks 6 and 12, and an additional three phone call follow up reviews at weeks 2, 4 and 9. Participants randomised to the LFD arm received $20 AUD Coles supermarket vouchers at each face-to-face appointment to be spent on key staple foods outlined in the dietary recommendations. Participants randomised to ad libitum MedDiet were given hampers including extra virgin olive oil and nuts for the intervention duration, as well as canned fish and legumes, to model the diet and showcase examples of appropriate staple foods. MedDiet participants were also supplied a Mediterranean diet cookbook.[28] Dietary data were collected using 3-day food diaries to determine habitual diet and to assess adherence to dietary prescription. These were collected at baseline, 6 weeks and 12 weeks. Food diaries were self-reported written records of food and beverage intake. Independent dietitians delivered the respective interventions. Three-day food diaries were entered and analysed using the software FoodWorks 8™. Dietary interventions overlapped with some of the recommendations provided such as increasing wholegrains, vegetables, fruits and reducing discretionary items.

Demographic and Anthropometric Data

Demographic information including age, gender, smoking status, education, employment and ethnicity was collected by self-report at baseline. Anthropometric measures including height (m), weight (kg), waist, hip and neck circumference (cm) were taken at baseline, mid 6 weeks and 12 weeks. Body composition measurements were taken using Bioelectrical Impendence Analysis (BIA) with the Seca mBCA 515. As participants arrived in a fasted state for pathology these always occurred at about the same time, in the morning, participants were asked to void their bladder and wear light clothing, and the same clothing, at all appointments. Any jumpers or jackets were removed as were shoes and socks for the analysis. These data points were collected by a researcher at all face-to-face time points; baseline, 6 weeks and 12 weeks. All data collection was in accordance with the research protocol outlined elsewhere.[25] Height and weight were used to calculate BMI (kg/m2).

Primary Outcome

Intrahepatic lipids (IHL) were assessed at baseline and at 12 weeks. IHL was measured using non-invasive proton magnetic resonance spectroscopy (1HMRS) on an Avanto 1.5 T system (Siemens) by an independent qualified radiographer, using a protocol developed for the MEDINA trial and detailed previously.[14] Briefly, hepatic spectra were acquired using point resolved spectroscopy (TR = 3000 ms, TE = 35 ms, 16 measurements, 1024 sample points) with a 3.0 × 3.0 × 3.0 cm volume of interest, during free breathing. IHL was quantified as the percentage of the methylene resonance to water (jMRUI version 5.2, EU Project) by a blinded researcher. This replaced insulin resistance as the primary outcome (as per trial registration) as a result of slow participant recruitment.

Secondary Outcomes

Liver stiffness measurement (LSM), in kPa using Fibroscan™ (Echosens) was determined at baseline and 12 weeks by experienced hepatologists at The Alfred who were blinded to the study intervention. All other secondary outcomes were collected at baseline, 6 weeks and 12 weeks and included Homeostatic Model Assessment-Insulin Resistance (HOMA-IR),[29] liver function tests (alanine aminotransferase (ALT), aspartate aminotransferase (AST), gamma-glutamyl transferase (GGT) and alkaline phosphatase (ALP)), systolic and diastolic blood pressure (mmHg), and serum lipids (total cholesterol (TC), low-density lipoprotein (LDL), high-density lipoprotein (HDL) and triglycerides (TGs) (mmol/L)). The inflammatory marker high sensitivity C-reactive protein (hs-CRP) (mg/L) was also measured. Reference ranges for these biomarkers were derived from Alfred Health pathology.


The MEDAS score is a 14 point checklist developed and validated by researchers for the PREDIMED trial, a study which assessed MedDiet in participants with cardiovascular disease.[30] This score is designed to assess MedDiet adherence where participants received a score of one for adherence and zero for non-adherence for each of the 14 criteria where a higher score is indicative of greater adherence.[31] A nine-point score for adherence to LFD was also developed for the PREDIMED study and used to assess compliance in this study. These scores were used to determine participant adherence to the respective diets and were assessed at baseline and 12 weeks.

Statistical Analysis

Descriptive statistics are presented as mean ± standard deviation for continuous data, and frequency and percentage for categorical data. Between-group intervention effects were estimated using restricted maximum-likelihood linear mixed models using outcome data from all available time points. These models included terms to estimate group effects at the mid- (where available) and post-intervention time points and were adjusted for pre-intervention diabetes status and baseline visceral fat level by including their interactions with time point. Specification of an unstructured residual error variance–covariance matrix ensured that models implicitly adjusted for the baseline level of the outcome.[32] All participants with post-baseline outcome data contributed to the estimation of intervention effects. A sensitivity analysis was also conducted in which observations that had disproportionate effects on model estimates (as determined by inspection of DFBETA values) were checked for and, where present, the model was refitted with the influential observation(s) omitted. Within-group changes from baseline to post intervention were estimated separately for each intervention arm using restricted maximum-likelihood linear mixed models with random intercepts for participants and a fixed effect of time. Inspection of preliminary model diagnostics (e.g., normality of residuals, homogeneity of variance) led to all but six outcomes (fat mass, systolic blood pressure, diastolic blood pressure, total protein, albumin and globulin) being log-transformed for use in inferential analyses. For models with log-transformed outcomes, unstandardised coefficients and their confidence intervals can be exponentiated and subsequently interpreted as estimated multiplicated effects on geometric means.[33] All statistical analyses were conducted using Stata/SE 16 (StataCorp, TX) and statistical significance was set at p < .05. For biomarkers including liver enzymes and lipids, there were gender-specific reference ranges, however because of the small numbers included in this trial these values were combined for statistical analysis.

Power Calculation

The sample size calculation was based on IHL summary data in Table 2 of Ryan et al.,[14] with the assumption of a 25% change of IHL in the MedDiet group and 5% in the LFD group, resulting in difference of 20% in change of IHL. The inputs required power of at least 80% with type I error = 5%. The required sample size for each group was 17. Allowing for a potential 20% dropout, the required sample size was 21 participants per group.