The findings of our meta-analysis of randomized trials evaluating insulin-sensitizing drugs in HALS differed according to the drug studied. We found that rosiglitazone modestly improved fasting insulin, but worsened fasting lipid levels and had no favorable impact on measures of central adiposity. Conversely while pioglitazone had no effect on fasting insulin or glucose, it improved fasting HDL-cholesterol without negative effects on other lipids. It too had no favorable impact on measures of central adiposity. In contrast, metformin favorably impacted outcomes across all three areas of interest including statistically significant reductions in fasting insulin, fasting triglycerides, body mass index and waist-to-hip ratio. Head-to head trials reinforced the findings of placebo- and no treatment-controlled trials with metformin demonstrating more favorable impacts on lipids and body fat redistribution, compared to rosiglitazone. Severe side effects were uncommon with all three drugs.
Overall completeness and applicability of evidence
Though sixteen trials involving over 920 subjects met our inclusion criteria, the completeness of the data differed according to intervention. While rosiglitazone and metformin were both evaluated in multiple trials, the findings for pioglitazone were based on only two trials, one of which had just 14 subjects. With regard to generalizability, our inclusion criteria allowed for a wide range of HALS populations, which varied by both the character and severity of their lipodystrophy manifestations and the composition of their ART regimens. Subjects were predominantly male but females comprised 25-50% of several trials. The trial settings and drug dosing were consistent with those found in clinical practice. Important groups that were not represented in the study populations include adolescents and children with HALS.
Quality of the evidence
In assessing the methodological quality of the included studies, we found that two trials used "no treatment" groups rather than placebo controls [46, 47, 49] and thus could have been susceptible to performance bias. Both of these unblinded trials favored the intervention group more strongly than comparable blinded trials for the fasting insulin and fasting glucose outcomes, thereby possibly inflating these summary estimates. Additionally, several trials did not adequately describe their method of randomization or allocation concealment thus precluding formal assessment of the likelihood of selection bias.
Potential biases in the review process
It is possible that we missed relevant trials, although we believe this is unlikely based on our systematic search efforts and no evidence of publication bias. We are aware of several ongoing trials [54–56] evaluating pioglitazone which, when available, should allow for a more robust analysis of the treatment's efficacy and safety.
In regard to our inclusion criteria, it is possible that allowing populations with diverse manifestations of HALS and continued use of certain antiretroviral agents may have contributed to the failure of several studies to find statistically significant differences for particular clinical outcomes. For example, subjects included based on isolated derangements in one element of HALS (i.e. central adiposity) might have been normal or only mildly affected with regard to other elements (i.e. insulin sensitivity) and thus would have had little if any room to benefit with respect to the latter measure. Similarly, because some antiretroviral agents including specific NRTIs and PIs, have been shown to increase central adiposity  and down-regulate PPAR receptors [35–37], their continued use could have diluted or negated potential beneficial effects of the interventions. Nonetheless, despite variation in ART usage across trials, nearly all summary estimates passed their tests for heterogeneity.
In regards to our analysis, our summary estimates for the body morphology outcomes may be less robust or generalizable than other outcomes due to reporting in only a small subset of studies. Furthermore, while the most consistently reported morphology-related variables--BMI, WHR and VAT--capture elements of central adiposity, they do not specifically address peripheral wasting, which may be particularly important in the decision to use metformin in patients with lipoatrophy [43, 58].
Agreements and disagreements with other studies or reviews
This is the first quantitative meta-analysis evaluating the effect of insulin-sensitizers in HALS. In 2004 Benavides  sought to summarize the efficacy and safety any pharmacological therapy for the treatment of HALS, in particular the effects on fat redistribution. While the review concluded that "no drug exists to fully ameliorate or correct the cosmetic changes of HALS", the review was limited by studies with non-randomized designs and the inability to quantitatively assess the effects of specific agents. In 2007 McGoldrick  reviewed randomized trials of statins, fibrates, and insulin-sensitizers for managing dyslipidemias in HIV-infected subjects taking ART. While concluding that "most studies suggested beneficial effects and satisfactory safety profiles", the review also warned that "rosiglitazone appeared to have some detrimental effects on lipid profiles". Due to limitations of data reporting, again no quantitative summarizations were performed.
When placed in the context of studies evaluating insulin-sensitizers in non-HIV populations, our findings are consistent with growing safety concerns surrounding the use of thiazolidinediones, most notably rosiglitazone, due to increased cardiovascular events. The American Diabetes Association and the European Association for the Study of Diabetes now specifically recommend against the use of rosiglitazone, and while pioglitazone is still included in the management pathway, is considered "less validated". Additionally our positive findings for metformin are consistent with a recent review of the medication in persons at risk for diabetes mellitus which showed favorable impacts on lipid profiles, insulin resistance, and BMI over a mean study period of 1.8 years in 31 study trials.
Implications for research
Ongoing efforts to identify and compare potential interventions for HALS would benefit from several actions. First, standardization of the methods used to evaluate insulin sensitivity and body morphological changes, including measures of both central adiposity and peripheral wasting, would greatly facilitate comparisons across studies and interventions. In addition, the inclusion of additional measures such as quality of life and ART compliance could capture important effects not revealed by changes in laboratory and imaging studies alone. Similarly, longer term studies of the more promising interventions are needed to fully assess whether the short-term changes in surrogate markers of cardiovascular risk translate into long-term benefits in reduced cardiovascular events and mortality.
Implications for practice
The findings of our meta-analysis suggest several implications for practice. With regard to rosiglitazone, we believe there is adequate evidence suggesting that the drug should not be given to patients with HALS. While pioglitazone did not appear to cause the adverse lipid problems seen with rosiglitazone, the number of completed clinical studies is small and more data is needed to determine whether it is an effective treatment option for patients with HALS. Lastly, while the evidence showed that metformin had favorable effects across all three components of HALS, whether these short-term, surrogate changes will translate to long-term clinical benefits is not known.