The databases used were PubMed / Medline, Scielo, and Science direct, and the following Mesh terms and names of the main AAS used in the search: “anabolic androgenic steroids”; “testosterone congeners”; “testosterone”; “resistance training”; “strength training”; “resistance exercise”; “strength exercise”; “HDL”; “LDL”; “lipoprotein”; “triglycerides”, using the Boolean operator “and”. In addition, all terms were translated into Portuguese and used for searches in Portuguese language databases. The searches ended in June 2018.

The inclusion criteria for the articles were: randomized clinical trials that evaluated the effects of AAS in healthy adults who did resistance training and were evaluated for the following outcomes: HDL, LDL, TG and muscle hypertrophy. Articles that did not use control groups, that did not present a detailed description of the resistance training program, and studies including patients that used any medication that could influence the lipid profile were excluded.

The Kappa coefficient was calculated to identify the level of agreement between researchers.

Two independent researchers (M.C.C.T and C.L.S.L.P) selected the articles by titles, excluding those that did not address the proposed theme. Then, the abstracts were analyzed to identify the articles that met the inclusion criteria.

Subsequently, studies that met the inclusion criteria were selected for systematic review and subsequent meta-analysis. In case of divergence, a third researcher (C.K.C.S) was consulted. The relevant data were extracted from the included studies by two independent reviewers (CLSLP and MCCT) – mean ± standard deviation (SD) of pre- and post-intervention period, and changes from baseline (post-test - pre-test) and pooled SD. Data presented in mmoL/L were converted to mg/dL.

The studies that did not present grouped SD of one of the interventions had their values estimated through p-value, confidence interval, or correlation coefficient, and a correlation of R = 0.80 was adopted in the calculations.¹⁷

The risk of bias was assessed by the Cochrane Collaboration's Risk of Bias Tool 1.0. The tool was used in its full version, with no insertions or changes in any of its seven domains,¹⁸ which were assessed independently by two reviewers (M.C.C.T and C.L.S.L.P).

The following domains were considered for bias assessment: (a) selection bias due to the generation of random sequence, (b) selection bias due to allocation concealment, (c) performance bias, (d) detection bias, (e) attrition bias, (f) reporting bias and (g) other bias. The supporting material of the RoB 1.0 tool (no internal assumptions were made) was used as support for the evaluation of the domains. Discrepancies were resolved by consensus between the reviewers. The data are shown in Figure 6.

The meta-analyses were conducted using the Review Manager software (RevMan 5.3. Copenhagen: The Nordic Cochrane Center, The Cochrane Collaboration, 2014). Random effect models with inverse variance method for continuous data were adopted, and data were expressed as difference of means and confidence intervals (95%CI). The heterogeneity between the studies was calculated using Cochrane's Q analysis with a statistical significance level of 0.10 for heterogeneity, and the I² for inconsistencies in treatment effect sizes (I²> 50% accepted for substantial heterogeneity).¹⁹ For exploration of heterogeneity among studies, each study was removed one at a time to verify whether it was a potential cause of heterogeneity. Kappa statistics were performed using SPSS software version 21 to identify the agreement in the selection of studies, and analysis of the methodological quality evaluated by the Risk of Bias Tool 1.0 tool from the Cochrane Collaboration. Accuracy of searches was analyzed by dividing the number of articles included by the number of articles selected was divided after duplicates were removed. Subsequently, the NNR (Number needed to read) = 1 / precision was calculated, according to Lee et al.²⁰.

Based on the Cochrane collaboration tool for assessment of controlled clinical trials, no study was judged as low risk of bias (Figure 6).

As for the selection bias (randomization and allocation concealment), all studies were classified as “risk of uncertain bias”. For performance (accreditation of participants and professionals), approximately 33.3% of the studies^11,21 were classified as “low risk”.

In the assessment of detection bias (blinding of evaluators), only one study was classified as low risk.¹¹

Approximately 50% of the studies were classified as “high risk” for “attrition bias” (incomplete outcomes).^11,21,22 Four studies (66.7%) were classified as low risk for reporting bias (selective outcome report)^12,21,23,24 and two were classified as risk of uncertain bias.^11,22 Only two studies were classified as low risk for other biases;^21,23 the others were classified as risk of uncertain bias.

It should be noted that the studies included had a small sample size and did not describe the evaluation of statistical power, allowing random errors due to low power.

In domain assessment, the Kappa coefficient for agreement between the two reviewers was 0.70.

General description of the included studies is summarized in Table 1. In total, 170 individuals completed the studies with RT, mean sample size 22.3 ± 12.5, and age range between 18 and 65 years. The highest mean age reported was 47 years.²² Four studies evaluated trained individuals,^11,21,23,24 only one study included sedentary individuals,¹² and one study did not describe the physical activity status of participants, whether experienced practitioners or beginners.²² In the meta-analysis, the number of subjects evaluated was 134, 54, 67 and 133 for the outcomes HDL, LDL, triglycerides and muscle hypertrophy, respectively. Of the included studies, one study evaluated lipid profile as the primary objective,²³ one study did not describe dietary control/guidance,²⁴ two studies reported adverse effects related to acne, nipple tenderness,¹¹ headaches, muscle cramps, dehydration and mood swings.²¹ One study presented the calculation of sampling power a posteriori for its main outcome,²² but no study presented a sample calculation for the analyzed outcomes; all studies showed multiple comparisons. The study by Bhasin et al.¹¹ used different samples for the analysis of lipid profile and anthropometric outcomes.

Interventions lasted from 4 to 12 weeks, with a weekly frequency of 3-4 sessions per week. Regarding the dose-response of strength training, the volume varied from 3 to 5 sets per exercise of 3-15 repetitions at an intensity of 65 to 95% of a maximum repetition. Only Broeder et al.²² used a non-linear (wave) periodization.

AAS doses were applied intramuscularly weekly^11,23,24 or as capsules consumed two to three times a day, every day.^12,21,22 The doses of injectable steroids ranged from 200mg/week of testosterone decanoate^23,24 to 600mg/week of testosterone enanthate.¹¹ Only the study by Bashin et al.¹¹ used supraphysiological doses. Control groups received weekly intramuscular administration of sesame oil¹¹ or peanut oil,²³ or consumed rice flour-,¹² maize starch-²² or maltodextrin- derived placebos.²¹ The study by Kuipers et al.²⁴ did not report the substance used as a placebo.

Six studies were included in the statistical analysis, which revealed substantial heterogeneity (I² = 97%) among the studies. Three of the six studies showed no effects on HDL,^11,23,24 and three ^12,21,22 showed changes in HDL levels, with only one reporting a great reduction.²¹ All these studies that showed changes in HDL levels used oral prohormone drugs, whereas studies that used injectable drugs did not show changes in HDL. In the subgroup analysis, no heterogeneity was found between the studies that used injection drugs (I² = 0%, p = 0.46); however, considerable heterogeneity was found in the analysis of the studies that used oral drugs (I² = 97%, p <0.001).

No changes in HDL were identified with the use of oral or injectable substances (-9.33 mg/dL, 95%CI −18.30, −0.36, p<0.00001; I²=97% and 0.75 mg/dL, 95%CI −0.93, 2.42, p=0.46; I² = 0 %, respectively).

As shown in Figure 2, in the total analysis, no significant difference was found in HDL-c reduction for AAS groups (-5.62mg/dL, 95%CI −12.10, 0.86, p=0.09; I² =97%). In general, an average decrease of 6.2±6.1mg/dL was observed in the AAS group after the intervention, while a decrease by 0.8±3.0mg/dL was observed in the control group. No great methodological heterogeneity was identified in the studies.

Three studies that evaluated LDL-c levels were included in the statistical analysis. The meta-analysis revealed considerable heterogeneity among the studies (I²=95%). Only one study reported a reduction in LDL-c levels,¹¹ and only the study by Granados et al.²¹ showed a significant increase. Due to the limited number of studies evaluating changes in this outcome, it was not possible to perform a sensitivity analysis of the substances used.

As shown in Figure 3, no significant difference was found in LDL values between AAS and control groups (7.76mg/dL, 95%CI −9.70, 25.23, p=0.57; I² =95%). In general, an average change of 9.1±18.2mg and 0.9±4.2mg in LDL was found in the AAS group and control groups, respectively, after the interventions.

Four studies were included in the statistical analysis. It was not possible to perform a sensitivity analysis of the substances used. In the total analysis, as shown in Figure 4, there was a statistically significant difference in the reduction of triglycerides for the steroid group (-11.68mg/dL, 95%CI −23.07,-0.26, p=0.04; I² = 42%). In general an average decrease of 5.6±24.2mg/dL and 0.8±14.4mg/dL in triglyceride levels was observed in AAS group and control group, respectively, after the interventions.

Considering the effects of AAS on muscle hypertrophy, considerable heterogeneity was demonstrated (I²=91%, p<0.00001) among the studies analyzed. Of the studies that evaluated the effects of AAS on lipid profile, four evaluated muscle hypertrophy, of which, two did not demonstrate hypertrophic changes.^12,22 As shown in Figure 5, there was no statistical difference in muscle hypertrophy between AAS and placebo groups (±2.44kg; 95CI% 0.02; 4.86, p=0.05). The experimental group obtained an average hypertrophic increase of 3.7 ± 2.0 kg after the intervention, while the control group obtained an increase of 1.4±1.1kg.

We conclude that current evidence does not support an effect of low-to-moderate doses of AAS on HDL-c and LDL-c levels. In addition, the use of AAS, in the doses used in clinical trials, had no beneficial effect on muscle hypertrophy.