Association between integrase strand transfer inhibitor use with insulin resistance and incident diabetes mellitus in persons living with HIV: a systematic review and meta-analysis

Whether integrase strand transfer inhibitors (INSTIs) are associated with a higher risk of incident type 2 diabetes mellitus (DM) than other antiretroviral therapies (ART) needs to be established. MEDLINE, Embase, Web of Science, and ClinicalTrials.gov registries were searched for studies published between 1 January 2000 and 15 June 2022. Eligible studies reported incident DM or mean changes in insulin resistance measured by Homeostatic Model for Insulin Resistance (HOMA-IR) in patients on INSTIs compared with other ARTs. We performed random-effects meta-analyses to obtain pooled relative risks (RRs) with 95% CIs. A total of 16 studies were pooled: 13 studies meta-analyzed for incident diabetes with a patient population of 72 404 and 3 for changes in HOMA-IR. INSTI therapy was associated with a lower risk of incident diabetes in 13 studies (RR 0.80, 95% CI 0.67 to 0.96, I2=29%), of which 8 randomized controlled trials demonstrated a 22% reduced risk (RR 0.88, 95% CI 0.81 to 0.96, I2=0%). INSTIs had a lower risk compared with non-nucleoside reverse transcriptase inhibitors (RR 0.75, 95% CI 0.63 to 0.89, I2=0%) but similar to protease inhibitor-based therapy (RR 0.78, 95% CI 0.61 to 1.01, I2=27%). The risk was lower in studies with longer follow-up (RR 0.70, 95% CI 0.53 to 0.94, I2=24%) and among ART-naïve patients (RR 0.78, 95% CI 0.65 to 0.94, I2=3%) but increased in African populations (RR 2.99, 95% CI 2.53 to 3.54, I2=0%). In conclusion, exposure to INSTIs was not associated with increased risk of DM, except in the African population. Stratified analyses suggested reduced risk among ART-naïve patients and studies with longer follow-up. International Prospective Register of Systematic Reviews (PROSPERO) registration number: CRD42021273040.


INTRODUCTION
Antiretroviral therapy (ART) has revolutionized HIV treatment and significantly reduced AIDS-associated mortality globally, particularly in sub-Saharan Africa. 1 People living with HIV (PLHIV) have more prevalent insulin resistance and diabetes mellitus (DM) than HIV-negative populations due to a combination of demographic and socioeconomic factors, in addition to HIV-related factors. 2 3 WHAT IS ALREADY KNOWN ON THIS TOPIC ⇒ People living with HIV (PLHIV) have a higher prevalence of metabolic perturbations compared with HIVnegative populations, and integrase strand transfer inhibitors (INSTIs) are currently the preferred firstline and second-line antiretroviral therapy (ART).⇒ Some studies suggested more weight gain among INSTIs users compared with other ART regimens, while others reported accelerated hyperglycemia preceded by weight loss, weeks to a few months after initiating INSTIs.

WHAT THIS STUDY ADDS
⇒ This systematic review and meta-analysis comprising ~75 000 PLHIV on different ART regimens is the first to examine the risk of insulin resistance and type 2 diabetes mellitus (DM) in INSTIs compared to other ART regimens.⇒ Analyses showed that compared to protease inhibitors and non-nucleoside reverse transcriptase inhibitors, INSTI exposure was not associated with increased risk of insulin resistance and/or DM. ⇒ We also identified in multiple analyses that INSTIs might be associated with a reduced risk of type 2 DM in certain subpopulations of PLHIV.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY
⇒ Our findings contribute to the evidence of metabolic safety of INSTI therapy, which might implicate the choice of therapy for millions of PLHIV.⇒ We demonstrated that exposure to INSTI therapy did not pose higher risk of insulin resistance and/ or DM compared to other ART regimens.Initiating or switching to INSTIs is safe; nevertheless, monitoring is warranted in certain high-risk groups.

Metabolism
[9][10][11][12] In the early ART era, nucleoside reverse transcriptase inhibitors (NRTIs) were coupled in combinations of predominantly stavudine, didanosine, zidovudine, lamivudine and zalcitabine with non-nucleoside reverse transcriptase inhibitors (NNRTIs). 13In 2018, the WHO recommended the use of INSTIs, particularly dolutegravir (DTG) as first-line ART, and since then, the use of INSTIs has largely overtaken NNRTIs and PIs. 18This was after multiple countries reported primary resistance to NNRTIs above the recommended threshold of 10%. 19 202][23][24][25] Despite their favorable side-effect profiles, INSTIs have consistently been associated with weight gain. 23 26Whether the weight gain in PLHIV translates to disorders in glucose metabolism in the long term remains to be demonstrated. 27ultiple case series on ART-experienced patients presenting with diabetic ketoacidosis with preceding weight loss a few weeks to months after starting INSTIs have been published. 24 25 28 29However, large population cohort studies have yielded conflicting results about the risk of diabetes among INSTI users. 30iven the inconsistent literature and to better quantify the risk, we performed a comprehensive literature review and meta-analyses aiming to summarize the current evidence on the association of INSTI therapy with insulin resistance, hyperglycemia, and incident DM versus PIs and NNRTI-based ART.We also explored the effect of other HIV-related factors and potential confounders on this association.

RESEARCH DESIGN AND METHODS
The protocol for this systematic review is registered on International Prospective Register of Systematic Reviews database (CRD42021273040) and published. 31This study is being reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses checklist. 32The link to the study dataset is listed in the online supplemental material (SD).

Search strategy and selection criteria
We searched PubMed/MEDLINE, Embase, and Web of Science (Clarivate) databases without language or geographical restrictions for randomized controlled trials (RCTs), cohort studies and case-control studies for eligible studies (online supplemental material, Emethods 1).Additionally, we searched Cochrane and clinicaltrials.org registries for eligible RCTs.Our search limit was fixed to the year 2000 to capture phase III clinical trial safety data, given that the first INSTIs, raltegravir, was approved by the Food and Drug Administration in 2007, and the search was last updated on 15 June 2022.We also searched abstracts of HIV conference meetings (International AIDS Society's Conference on Retroviruses and Opportunistic Infections) for the same themes seeking studies that were eventually published.To identify relevant publications, two authors (FM and HK) independently screened all potential abstracts and reference lists in review articles.For published studies with desired outcomes but without data to calculate relative risk (RR) of diabetes, we reached out to authors for raw data.Studies eligible for full review were agreed on through consensus.A senior investigator (NB) was referred to in case of disagreement between the authors.
Studies were eligible if they reported risk of incident diabetes (or reported the required data to calculate incidence) with or without metabolic syndrome and/ or insulin resistance, had exposure to INSTIs for ≥12 weeks, and had comparative arms of either NNRTI or PI anchored ART.Studies with cross-sectional design and studies including pregnant or breastfeeding mothers were excluded.Since we aimed to compare INSTIs versus PIs and/or NNRTIs as anchor agents, we also excluded studies where INSTIs were administered with PIs or NNRTIs in the same regimen.For studies with multiple publications, we included the publication with the most extended follow-up.

Data analysis
We evaluated two outcomes: incident hyperglycemia and type 2 DM (new cases) as a discrete outcome or as part of metabolic syndrome (online supplemental table S1).A separate analysis was performed for mean changes in insulin resistance measured by the Homeostatic Model for Insulin Resistance (HOMA-IR) index, a factor of fasting blood glucose and insulin.We extracted variable study and population characteristics into excel forms (online supplemental table S3).Adjusted effect estimates were sought whenever reported; otherwise, raw data were retrieved.
The quality of the studies was assessed using the Newcastle-Ottawa Scale for cohort or case-control studies 33 and the Revised Cochrane Risk-of-Bias tool for randomized trials 34 (online supplemental tables S4 and S5).
Statistical analysis was done using meta-R package V.4.0.5 with R package Metaphor and Stata V.15 to generate forest plots of pooled effects with 95% CIs.We Metabolism performed a random-effects meta-analysis adjusting for in-between-study heterogeneity to pool the risk (new cases/overall population at risk) of DM with or without metabolic syndrome (as discrete outcomes).The populations of interest were HIV patients exposed to INSTIs compared with patients on NNRTI or PI-based ART regimens.We assessed in-between study heterogeneity using the I 2 statistic with DerSimonian and Laird's method, using values <50%, 51%-74%, and ≥75% to represent low, moderate, and high heterogeneity, respectively. 35We sought evidence for publication bias by applying Egger's test and visually inspecting funnel plots for asymmetry (if ≥10 studies). 36We also performed several subgroup analyses to explore if the risk of DM was affected by longevity on INSTIs, particular types of INSTIs, geographical region of study participants, and past exposure to ART, as some ART drugs were associated with abnormal glucose metabolism. 14To further explore sources of heterogeneity, we also carried out subanalyses by study design, type of caring facility, and type of non-INSTIs in the control group to compare pooled effects and heterogeneity.A p value of <0.1 was considered a statistically significant subgroup effect.We considered sensitivity analyses to test the robustness of our findings by including only studies reporting adjusted risk estimates, excluding studies with comorbidities like viral hepatitis B and C, studies where primary outcome was a metabolic endpoint and studies with no apparent conflict of interest.Studies reporting changes in mean HOMA-IR were separately analyzed to pool mean changes (on a continuous scale) of HOMA-IR pre-INSTI and post-INSTIs exposure compared with PIs and/or NNRTIs.Additionally, we performed a univariable metaregression to explore the effect of the following variables on the outcome: the effect of year of publication, follow-up duration, average age, CD4 count, body mass index (BMI) of participants and the proportion of male participants if at least 10 studies reported sufficient data.In this systematic review and meta-analysis, sex was defined as biological sex at birth.

Literature search and study selection
Out of the 124 studies identified for full-text review, 16 studies were deemed eligible for inclusion in the metaanalysis, 27 30 37-50 and 3 studies included in the systematic review could not be pooled in the quantitative synthesis [51][52][53] (figure 1).Excluded studies and reasons for exclusion are presented in online supplemental table S6.
Overall, the quality of the studies was rated as high (online supplemental table S4 and S5).Common to most RCTs was a lack of blinding in the assessment of the outcome.

Effect of weight gain
We sought to analyze the effect of baseline weight, weight gain or changes in BMI on the incidence of diabetes in the study populations.Eleven studies 27 30 38-41 43-49 provided estimates of weight and/ or BMI at baseline, yet changes were not presented per type of ARTs nor stratified per persons who developed diabetes and/or hyperglycemia, making it difficult to analyze.
For studies reporting incident insulin resistance and/ or diabetes across different meta-analyses when the number of studies was ≥10, no publication bias or small study effect was detected by funnel plot asymmetry and by Egger's test (online supplemental figure S6).

Metaregression analysis
We further explored the influence of specific study and HIV-related factors on the pooled risk of developing insulin resistance and/or type 2 DM between INSTIs and non-INSTI comparators.Neither the proportions of male, black population, or publication year were associated with the pooled risk in univariable meta-regression analysis.However, studies with longer follow-up duration were significantly associated with lower risk of type 2 DM in INSTIs compared with non-INSTIs (online supplemental table S7 and figure S9).

Influence analysis
We conducted influence analysis by the leave-one-out method to investigate the individual impact of each study (online supplemental figure S10).There was no significant change in the pooled effect estimates.Baujat plot pointed to one study with the most impact on overall

Metabolism
The common presentation was diabetic ketoacidosis preceded by weight loss, weeks to months after initiating therapy, which might represent a typical phenotype of insulin deficiency. 25 29ome of the postulated mechanisms for the accelerated hyperglycemia included intracellular magnesium chelation induced by INSTIs leading to altered hepatic and skeletal muscle insulin signaling, mitochondrial dysfunction from previous exposure to more toxic NRTIs and possible genetic predisposition (online supplemental file 1, ref. 57). 56Interestingly, at the population level, INSTIs particularly DTG have been consistently associated with weight gain (online supplemental file 1, ref. 58).A recent systematic review concluded that INSTIs have a higher risk of DM compared with alternative backbone ART regimens. 56Most of the conclusions in that narrative review were premised on the consistent association of INSTIs with weight gain, a known precursor for metabolic syndrome or DM. 56We could not conclusively ascertain the effects of weight gain on the incidence of diabetes as data were lacking to perform a subanalysis for BMI changes.In the current analysis, studies with follow-up more than 12 months showed a 30% lower risk of type 2 DM among INSTIs versus non-INSTIs.This reduced risk tended to attenuate when restricted to studies with shorter follow-up.We observed a trend toward more insulin resistance (increase in HOMA-IR) rather than overt type 2 DM among INSTIs compared with non-INSTIs (online supplemental figure S7).It is unclear whether this trend is induced by the increased weight accompanying the 'return-to-health phenomenon' with possible metabolic perturbations in some susceptible individuals or could lead to overt type 2 DM and metabolic syndrome in the long term (online supplemental file 1, ref. 59).Considering the small sample size of this analysis (three studies

Metabolism
with 766 patients) and the heterogeneity of PLHIV populations, long-term follow-up studies are therefore warranted, particularly accounting for sex, the presence of malnutrition, obesity, and/or metabolic syndrome at treatment initiation.
A threefold increased risk of diabetes in African patients was observed in the subanalysis by geographical origin.These two pooled studies 46 48 were high-quality RCTs involving ART-naïve adults with primarily virological outcomes.They included ART-naïve patients of mean baseline age 32-38 years with unsuppressed viral loads and mean baseline CD4s of 280 and 336 cells/mm 3 .The baseline BMI for both studies did not significantly differ from the mean BMI from other meta-analyzed studies with BMI data.Exposure groups had patients on DTG and comparator groups, efavirenz.Estimates of metabolic syndrome prevalence among PLHIV in SSA range from 13% to 58%, with a higher proportion among ART-experienced than among ART-naïve (online supplemental file 1, ref. 60).It is likely that the increased risk of type 2 DM observed is driven by the higher prevalence of metabolic syndrome in this population (online supplemental file 1, ref. 60).On metaregression for age, baseline CD4, and viral load, we found a pattern of an increased risk of diabetes with higher baseline viral loads and low CD4 cell counts (online supplemental file 1, figure S8).This is in tandem with the known literature suggesting that chronically heightened inflammation in patients with high viral loads is a driver of insulin resistance and hence a precursor of type 2 DM (online supplemental file 1, ref. 61).These factors could have been drivers of this risk in this African population with more likely late presentation compared with PLHIV in resource-affluent settings.These results should, however, be interpreted with caution, given there were only two studies meta-analyzed with a small patient population; hence, these findings may not be extrapolated to the general African population.Studies suggested that women living with HIV have higher risk of ART-related weight gain compared with men (online supplemental file 1, ref. 62); moreover, women with HIV have higher odds of type 2 DM compared with women without HIV infection(online supplemental file 1, ref. 63).This might be attributed to higher weight gain, and possibly more prevalent cardiometabolic risk factors in women population with HIV.Whether African women living with HIV have heightened risk for type 2 DM compared with male peers is debated.In a meta-analysis of 20 studies from Africa, the prevalence of type 2 DM was similar in HIV and non-HIV populations regardless of sex, and similar prevalence was noted between treated and untreated PLHIV, though in between-studies heterogeneity was high (online supplemental file 1, ref. 64).In our analysis, sex was not associated with the pooled risk of type 2 DM in metaregression analysis.
INSTI exposure was associated with a low risk of diabetes, noted in ART-naïve populations compared with ART-experienced patients.This is in line with collection of reports on lower prevalence of metabolic syndrome in ART-naïve versus ART-experienced patients (online supplemental file 1, ref. 61).Another potential explanation might be thatbe clinicians tended not to start INSTIs in patients at high risk of diabetes, which could not be applied to ART-exposed patients being switched to INSTIs due to virological failure with less consideration for metabolic risk (online supplemental file 1, ref. 65 and 66).
We encountered certain limitations such as insufficient data on possible factors affecting glucose metabolism, which are potential confounders such as changes in BMI, family history of diabetes, lifestyle, concurrent drugs such as steroids and gender-affirming hormonal therapy in transgender patients.In the ART-experienced populations, we could not adjust for prior exposure to drugs like stavudine, didanosine, and zidovudine, known to cause lipodystrophy, insulin resistance and dyslipidaemia due to lack of patient-level data.There was variation in the criteria used to define diabetes in the different studies, with most retrospective cohort studies using multiple criteria: HBA1C, fasting blood glucose, oral glucose tolerance tests, and prescriptions for diabetes medication, while most RCTs used division of AIDS grading of fasting blood glucose (online supplemental file 1, ref. 67).To partially account for these limitations, we conducted influence and stratified analyses by study design, primary metabolic outcome, and ART status.There was minimal heterogeneity and an absence of publication bias across several subgroups and sensitivity analyses.

CONCLUSION
In conclusion, this meta-analysis demonstrated that INSTI use was not associated with an increased risk of DM compared with PIs and NNRTIs except in African PLHIV.There is a need for long-term follow-up studies with primarily metabolic outcomes to ascertain these results further and delineate the contribution of weight gain in PLHIV exposed to INSTIs on glucose dysmetabolism.Additionally, the increased risk of DM in African PLHIV merits more targeted research as this population in the meta-analysis was largely under-represented.

Figure 1
Figure 1 PRISMA flowchart for study selection.INSTI, integrase strand transfer inhibitor; NNRTI, non-nucleoside reverse transcriptase inhibitor; PI, protease inhibitor; PRISMA, Preferred Reporting Items for Systematic Reviews and Meta-Analyses.

Figure 2
Figure 2 Forest plot of the association of INSTI exposure to incident hyperglycemia and diabetes mellitus compared with other art regimens.The crude numbers of events are based on the longest follow-up reported in the studies.INSTI, integrase strand transfer inhibitor; RR, relative risk.

Table 1
Study characteristics of the 13 studies included in the meta-analyses for incident type 2 DM and the three studies for insulin resistance *The numbers represent patients without DM at baseline enrolled in the metabolic analyses in each study.NBEron et al 51 not included in the metanalyses.† ACTG, AIDS Clinical Trials Group; ART, antiretroviral therapy; DM, diabetes mellitus; FPG, Fasting Plasma Glucose; HbA1c, Glycated Hemoglobin; HOMA-IR, Homeostatic model of Insulin Resistance; INSTI, integrase strand transfer inhibitor; N/A, not available; OGTT, Oral Glucose Tolerance Test; RCT, randomized controlled trial; RR, relative risk.Table 1 Continued Metabolism demonstrated by area of origin (p<0.01).Further interpretation of subgroup analyses is reported in table 2.

Table 2
Subanalysis for the risk of diabetes mellitus with exposure to INSTIs in people living with HIV INSTI, integrase strand transfer inhibitor; NNRTI, non-nucleoside reverse transcriptase inhibitor; PI, protease inhibitor; RCT, randomized controlled trial; RR, relative risk.