The study design, eligibility and protocol of the trial is described in full in the original publication[20]. The trial was registered on the Australian and New Zealand Clinical Trials registry (trial number NCT00613782). This was a 40 wk, randomised, double-blind placebo-controlled trial conducted at single tertiary referral centre in Australia. Ethics approval for the study was granted by the Human Research Ethics Committee, Austin Health.

Eligible participants were men aged 35-70 years with a history of type 2 diabetes and a fasting, early morning TT concentration of ≤ 12.0 nmol/L (346 ng/dL), as measured by electrochemiluminescence immunoassay (ECLIA) and averaged across two readings.

Exclusion criteria for the trial were testosterone therapy within five years of randomisation, screening TT concentrations < 5.0 nmol/L (144 ng/dL), established pituitary or testicular disorder, luteinising hormone level > 1.5x upper limit of normal, prostate-specific antigen level > 4 µg/L, a history of urinary obstruction, prostate cancer or breast cancer, haematocrit > 0.50, uncontrolled hypertension (> 160/90 mmHg despite treatment), untreated obstructive sleep apnoea, estimated glomerular filtration rate < 30 mL/min, cardiac insufficiency, active malignancy, unstable psychiatric disease, weight > 135 kg, use of glucagon-like peptide-1 agonist therapy or very low-calorie diet, or an HbA1c level > 8.5% (69 mmol/mol).

Eligible participants were randomly assigned in a concealed 1:1 allocation to either testosterone or placebo therapy. Intramuscular testosterone undecanoate 1000 mg or a visually identical placebo was administered at 0, 6, 18 and 30 wk. Participants were followed up for a total of 40 wk.

In this post-hoc analysis we assessed as the primary outcome the change in liver fat fraction, as measured using MRI liver in-phase (IP) and opposed-phase (OP) T1 sequences and expressed as percentage fat. The primary outcome measure for the index study was the change across groups and time from baseline in the homeostasis model assessment index of insulin resistance using a computer-based calculation rather than the original linear equation.

Clinical and biochemical variables that were assessed in participants at baseline and 40 wk included body weight, body mass index (BMI), waist circumference, alanine aminotransferase (ALT), gamma-glutamyl transferase, alkaline phosphatase, bilirubin, albumin, international normalised ratio, TT, sex hormone-binding globulin, calculated free testosterone (cFT), fasting glucose, HbA1c and lipid profile. Free testosterone levels were calculated using Vermeulen’s formula, as previously described[21]. Additional measurements of total body mass, lean mass and fat mass were assessed by DXA scan (DXA Prodigy, Version 10.51; GE Lunar, Madison, WI) at 0 and 40 wk.

MRI scans of the abdomen were obtained at enrolment and after 40 wk of therapy using a 3 Tesla MRI scanner (Siemens, Erlangen, Germany). As part of the initial study, subcutaneous and visceral adipose tissue volume was calculated for each patient by analysis of five 10-mm slices around the L4 vertebral superior endplate using the SliceOmatic program software (version 4.2; Tomovision, Montreal, Canada) as previously described[20].

MRI images obtained before and after therapy were reanalysed for the present study by an expert liver radiologist who was blinded to treatment allocation. Liver fat fraction before and after therapy was calculated using conventional chemical shift imaging with IP and OP T1 sequences. An averaged signal intensity for IP and OP was obtained from three 5 to 10 cm² regions within hepatic parenchyma of each scan. The fat fraction was calculated using the formula (IP - OP)/2 × IP. This technique has been previously validated to accurately estimate liver fat fraction to levels of up to 50%[22-27].

Continuous data are displayed as mean (standard deviation) for normally distributed data or median [interquartile range] for skewed data. Categorical data are presented as number (percentage). The Student t-test and Mann-Whitney test were used for normal and non-normal data, respectively. Exploratory data analysis included pairwise examination for correlation that may introduce multicollinearity into the final regression model. Linear regression modelling was used to identify variables associated with week 40 Liver fat proportion, which was the primary outcome variable, whilst controlling for baseline liver fat proportion (analysis of covariance). The outcome variable was natural log-transformed for these analyses as use of the untransformed values violated model assumptions. Univariate regression with all clinically relevant variables was performed and those with P values < 0.20 were selected for inclusion in the multivariable model. A manual elimination process was undertaken to arrive at the final model. The coefficients in the final model were back transformed to report relative change in liver fat proportion between groups (geometric mean). Standard regression diagnostics were performed to ensure non-violation of the underlying model assumptions. Analyses were performed in R v4.02[28].

A total of 39 men with type 2 diabetes and low serum testosterone were included in our analysis, of whom 20 received testosterone therapy and 19 received placebo. They represent a subset of 88 patients from the original trial who underwent MRI liver scanning both at the start and end of treatment. Remaining study participants did not undertake MRI scans either due to incompatible or unverifiable metal implants, inability to fit in the MRI machine or claustrophobia.

The baseline characteristics of our study participants are shown in Table 1 and are similar between the testosterone and placebo therapy group with the exception of lean mass which was lower in the testosterone group and HDL which was higher in the testosterone group compared to placebo. Compared to the participants in the index study who did not have MRI scans, our cohort had lower baseline BMI, visceral adiposity, waist circumference and higher cholesterol levels, but otherwise had comparable baseline characteristics as shown in Table 2. All patients had hereditary haemochromatosis and other causes of low testosterone excluded and no patients were taking medications were taking steatogenic medications during the study period. Five participants in the testosterone group and three in the placebo group were taking glucagon-like peptide-1 analogues or thiazolidinediones at baseline which continued throughout the study at stable doses. No other patients were taking medications known to directly influence liver fat (Supplementary Table 1). Median alcohol consumption was 3.5 (IQR 0-8.5) standard drinks per week in the testosterone group and 4 (IQR 0.5-7) standard drinks per week in the placebo group. One patient in the testosterone group had heavy alcohol consumption (defined as > 14 units per week), reporting 28 standard drinks per week during the study period. No patients in the placebo group had heavy alcohol consumption.

All patients in the study had significant hepatic steatosis, defined as a liver fat fraction ≥ 5%. All patients but the one patient with heavy alcohol consumption had secondary causes of hepatic steatosis excluded and hence met diagnostic criteria for NAFLD as defined by the American Association for the Study of Liver Diseases[1]. Median liver fat fraction at baseline was 15.0% (IQR 11.5%-21.1%) in the testosterone group and 18.4% (IQR 15.0%-28.9%) in the placebo group (P = 0.14). The median values of all liver function tests were within normal limits for both groups at baseline (Table 1).

In the testosterone group, the median absolute reduction in liver fat fraction was 3.5% (IQR 2.9%-6.4%) and median relative reduction was 27.3% (IQR 18.0%-37.6%). Liver fat fraction increased in the placebo group, with a median absolute increase in liver fat fraction of 1.2% (IQR -2.6-3.0%) and median relative increase of 6.8% (IQR -7.3-15.3%). At week 40, the median liver fat fraction was 9.8% (IQR 8.6%-13.5%) in the testosterone group and 19.6% (IQR 17.8%-29.1%) in the placebo group.

The change in absolute liver fat ranged from -15.3% to +3.6% in the testosterone group, with 18 of 20 individuals achieving a reduction in liver fat (Figure 1). The change in absolute liver fat ranged from -5.9% to +27.5% in the placebo group, with eight of 19 individuals achieving a reduction in liver fat (Figure 1). One patient from the placebo group was an outlier who had a 27.5% absolute increase in liver fat associated with a 32% increase in visceral adipose tissue (VAT) and a 7% increase in body weight. At week 40, there were no significant changes in liver function tests in either treatment group.

Open in New Tab   Full Size Figure   Download Figure

Figure 1 Absolute change in liver fat in all patients receiving placebo and testosterone therapy from baseline to 40 wk follow up, reported as proportion of fat. Lines connect observations from the same participant. Red indicates an increase, green a decrease in liver fat proportion.

Univariate regression analysis found that testosterone therapy, BMI, baseline liver fat and week 40 TT and cFT concetrations were all significantly associated with changes in liver fat. Week 40 cFT and TT concentrations were, however, highly correlated ( hence a decision to use only cFT in the development of the multivariate model.

The multivariate model was significant (F_(2,35)(54.13), R² = 0.807, P < 0.001) and included treatment arm controlled for baseline liver fat proportion. In detail, those in the testosterone treatment group had a significantly lower follow-up liver fat proportion than those in the placebo group (P < 0.001). The Beta co-efficient of -0.48 (-0.67, -0.29) is on the log-transformed scale. Back-transformation of the co-efficient and associated 95% confidence interval returned values of -0.383 (-0.490, -0.254). Thus, in this model, testosterone therapy was associated with a 38.3% relative reduction in liver fat proportion (95% confidence interval 25.4% to 49.0% reduction) after controlling for baseline liver fat proportion compared to those receiving placebo. No other covariates reduced the unexplained residual variance in the model significantly to this estimate of between-group difference.

Baseline median TT concentrations as measured by ECLIA were 9.8 [7.2-12.00] nmol/L and 7.8 [5.9-10.8] nmol/L in the testosterone and placebo groups, respectively. After 40 wk of therapy, trough TT concentrations increased to 12.9 [11.9-15.0] nmol/L in the testosterone group and 8.8 [7.3-10.8] nmol/L in the placebo group (P < 0.001). Baseline median cFT concentrations were 216 [151-272] pmol/L and 176 [146-224] pmol/L in the testosterone and placebo groups respectively. After 40 wk of therapy cFT increased to 322 [231-392] pmol/L in the testosterone group and 193 [167-250] pmol/L in the placebo group (P < 0.001).

At 40 wk patients receiving testosterone therapy had no statistically significant changes in overall body weight, BMI, waist circumference, fasting glucose or HbA1c compared to placebo (Table 3).

At week 40, compared to placebo, patients receiving testosterone therapy had significantly increased lean mass (mean change 2111 g (1148-3073) relative to placebo, P < 0.001) and significantly decreased fat mass (mean change -2969 g (-3998 to -1941) relative to placebo, P < 0.001). All patients receiving testosterone therapy had reductions in fat mass and 19/20 patients receiving testosterone therapy had increases in lean mass (Supplementary Figure 1). Subcutaneous adipose tissue (SAT) was significantly decreased at week 40 in patients receiving testosterone therapy compared to placebo (mean change -359 cm³ (-570 to -147) relative to placebo, P = 0.002) but VAT was unchanged (Supplementary Figure 2).

Testosterone therapy was well-tolerated with rare serious adverse events that were not significantly different compared to placebo as outlined in Table 4 of the index study[20]. Patients receiving testosterone therapy had significant increases in haemoglobin and haematocrit at week 40 compared to placebo (P < 0.001 for both).

Testosterone levels are commonly low in men with type 2 diabetes and most men with type 2 diabetes have hepatic steatosis from non-alcoholic fatty liver disease (NAFLD). Animal models show that low testosterone states from castration results in hepatic steatosis and that testosterone replacement improves hepatic steatosis.

Hepatic steatosis occurs in NAFLD which currently has no readily available effective treatment. This study was conducted to provide rationale for future prospective studies of testosterone therapy for NAFLD.

To evaluate the effect of testosterone therapy on liver fat fraction as measured by magnetic resonance imaging (MRI) in a cohort of diabetic men with lowered testosterone levels. We further aimed to determine other factors associated with changes in liver fat in this population.

We performed a secondary analysis of a previous 40 wk, randomised, double-blinded, placebo-controlled trial of intramuscular testosterone undecanoate in men with type 2 diabetes and lowered serum testosterone levels. Liver fat as determined by MRI scan before and after therapy was analyzed in addition to blood tests and body composition scans.

Patients who received testosterone therapy had an absolute reduction of liver fat fraction by 3.5% and patients who received placebo had an absolute increase in liver fat fraction by 1.2%, with a between group difference of 4.7%, P < 0.001. After controlling for baseline liver fat, testosterone therapy was associated with a relative reduction in liver fat of 38.3% (P < 0.001).

Testosterone therapy was associated with a reduction in hepatic steatosis in a cohort of men with type 2 diabetes and lowered serum testosterone levels.

This study provides rationale for future prospective clinical trials of testosterone therapy for the treatment of NAFLD focusing on liver related endpoints.