SUMMARY
Background:
Integrase strand transfer inhibitor (INSTI) initiation has been associated with diabetes among antiretroviral therapy (ART)-naïve people with HIV (PWH). We examined the effect of switching to INSTIs on incident diabetes among ART-experienced PWH.
Methods:
We emulated a target trial within longitudinal cohorts of PWH in the US and Canada. Participants without diabetes who had used non-nucleoside reverse transcriptase inhibitors (NNRTIs) or protease inhibitor (PIs) for ≥180 days (2016–2022) were followed from encounters where they continued an NNRTI or PI vs switched to an INSTI, for up to 5 years. The effect of switching on incident diabetes was estimated using weighted Cox regression with robust variance. We further assessed whether the effect (i) varied by time since switch and (ii) was explained by weight gain in the first year.
Findings:
13,071 participants were followed from 2,702 encounters where they switched to an INSTI from an NNRTI, 54,766 encounters where they continued an NNRTI, 1,714 encounters where they switched to an INSTI from a PI, and 26,599 encounters where they continued a PI. Switching from PIs to INSTIs conferred an adjusted hazard ratio (HR) of 1.38 (95% confidence interval [CI], 1.06–1.80) for incident diabetes, whereas switching from NNRTIs to INSTIs conferred an HR of 1.10 (95% CI, 0.87–1.39). The diabetes risk was higher during the first two years after switching from PIs to INSTIs (HR, 1.67; 95% CI, 1.21–2.30), but not thereafter (HR, 1.08; 95% CI, 0.75–1.57; interaction P = 0.06). The effect of switching from PIs to INSTIs on diabetes did not appear to be explained by weight gain.
Interpretation:
The increased diabetes risk after switching from PIs to INSTIs highlights a metabolic implication of regimen change and may warrant close monitoring early after switch, regardless of weight gain.
INTRODUCTION
Diabetes mellitus affects more than 10% of people with HIV (PWH), and incidence is rising as this population ages.1 Antiretroviral therapy (ART) contributes to metabolic complications, with various agents associated with weight gain, lipohypertrophy, and insulin resistance.2 Since 2015, integrase strand transfer inhibitor (INSTI)-based regimens have been recommended as first-line for most PWH due to their virologic efficacy.3 Although multiple real-world studies have reported an increased diabetes risk among ART-naïve PWH initiating INSTI-based regimens compared with other ART regimens,4–6 this association has not been consistent across all studies.7 Furthermore, INSTI-based regimens have been associated with greater weight gain compared to non-nucleoside reverse transcriptase inhibitor (NNRTI)-based regimens and, to a lesser extent, protease inhibitor (PI)-based regimens in both randomized and observational studies,8–10 and weight gain on ART is a major risk factor for type 2 diabetes.11
Evidence on the diabetes risk associated with prevalent INSTI use among ART-experienced PWH has been mixed. Some observational studies have identified INSTI use as a risk factor in this population, whereas others have not observed such an association.12–14 Using a target trial emulation approach, we recently reported an 11% increased risk of diabetes—though not statistically significant— after switching from an NNRTI- or PI- to an INSTI-based regimen in a single HIV clinic cohort.15 Here, we apply similar methods to a larger and more generalizable multi-site cohorts of ART-experienced PWH across the US and Canada.
METHODS
Study design
In this multi-site longitudinal cohort study of PWH, we emulated a target trial to assess the effect of switching from an NNRTI or PI-based regimen to an INSTI-based regimen on incident diabetes risk.
Data on study participants were obtained from the North American AIDS Cohort Collaboration on Research and Design (NA-ACCORD), a consortium of 27 clinical and interval cohorts in the US and Canada.16 NA-ACCORD is part of the International Epidemiologic Database to Evaluate AIDS (IeDEA) global network. Annual standardized data submissions (demographics, prescriptions, comorbidities, laboratory results) undergo quality assurance at the Data Management Core (University of Washington) and are then processed by the Epidemiology/Biostatistics Core (Johns Hopkins Bloomberg School of Public Health). The activities of the NA-ACCORD were approved by the Johns Hopkins University institutional review board (JHU IRB 23533), and each participating NA-ACCORD cohort was granted ethics approval by its respective local institutional review board to share its data with NA-ACCORD.
The target trial included adults with HIV without diabetes history who had received NNRTIs or PIs without prior INSTI exposure. In our study, we included PWH aged ≥18 years from participating cohorts with ≥1 HIV clinical encounter between January 1, 2016, and December 31, 2022, who (1) had no diabetes history, (2) were on an NNRTI or PI for ≥180 days, (3) had never used an INSTI, and (4) had a weight measurement within the year before the encounter. The study period (2016–2022) aligns with INSTI-based regimens being recommended as first-line ART,3 minimizing confounding by indication from earlier periods when PWH with metabolic risk factors could have preferentially switched from older regimens with known metabolic toxicity.
Procedures
We used a nested trials approach to emulate a target trial where each time participants meet the inclusion criteria at an HIV clinical encounter, they would be randomized to either continue an NNRTI or PI versus switch to an INSTI.17 In our study, each time participants treated with an NNRTI or PI had a clinical encounter, they were eligible to switch to an INSTI. Antiretroviral drug exposure was ascertained from prescription records. We classified each eligible clinical encounter (the “index encounter” and unit of observation) as a switch to INSTI if the participant had a new INSTI prescription within ±14 days of the index encounter. Otherwise, an observation was classified as continuing an NNRTI or PI. Participants could contribute multiple observations where they continued an NNRTI or PI, but at most one observation where they switched to an INSTI. We excluded encounters occurring within 30 days of a prior qualifying encounter, ensuring at least a 1-month interval between observations.15
In both the target trial and our study, follow-up began on the date of the index encounter and continued until the first occurrence of incident diabetes (the outcome) or a censoring event. In our study, the censoring events included (1) loss to clinical follow-up, defined as the last encounter before a gap between encounters ≥1 year, (2) death, (3) 5 years after the index encounter, (4) administrative closure or end of observation window for diabetes for the corresponding cohort site,18 or (5) the end of administrative follow-up on December 31, 2022. Following the index encounter, observations were not censored at the end of study antiretroviral prescription per the intention-to-treat principle. However, those continuing an NNRTI or PI at the index encounter were censored upon a later switch to an INSTI.
Outcome
Incident diabetes was defined as the first occurrence of: (1) laboratory hemoglobin A1c value ≥6.5%, (2) initiation of a diabetes-specific medication, or (3) new diabetes diagnosis plus initiation of a diabetes-related medication. This is a commonly used definition by the NA-ACCORD.4,19 Diabetes-specific and diabetes-related medications are detailed in appendix p 1. Type 1 diabetes and gestational diabetes were excluded from the diagnosis-based definition.
Statistical analysis
In the target trial, treatment assignment—switching to an INSTI or continuing an NNRTI or PI—would be random. In our emulated study, treatment assignment was nonrandom; therefore, identification of causal effects relied on the assumption of no unmeasured confounding after covariate adjustments (exchangeability assumption). To account for confounding from nonrandom treatment assignment, we used inverse probability of treatment weights (IPTW).20 Two separate IPTW were derived: one from a logistic model predicting the probability of switching to an INSTI versus continuing an NNRTI, and another from a logistic model predicting the probability of switching to an INSTI versus continuing a PI, conditional on all covariates described below. Additionally, we used inverse probability of censoring weights (IPCW) to mitigate potential bias from informative censoring. We estimated the weights separately for censoring due to (i) loss to clinical follow-up and (ii) a subsequent switch to an INSTI following index encounters where an NNRTI or PI was continued. Both censoring weights were estimated using a multivariable logistic model for the probability of censoring in each 90-day block after the index encounter, conditional on the covariates at the index encounter, exposure, time since the index encounter, and time-updated covariates. The final weights were the cumulative product of the censoring and treatment weights.
Covariates were selected based on their relationship with treatment exposure and/or outcome. Age, sex at birth, and race (Black, White, Other, or Unknown) were measured via the electronic medical record. BMI—underweight (<18.5 kg/m2), normal weight (18.5–24.9 kg/m2), overweight (25.0–29.9 kg/m2), or obese (≥30 kg/m2)—, viral suppression (HIV viral load ≤200 copies per ml), and CD4 count ≤200 cells per μL were ascertained based on the most recent measurement within the year leading up to the index encounter. We assessed whether the participants had a history of hypertension, hyperlipidemia, and chronic kidney disease stage 3 or greater (definitions detailed in appendix p 2) prior to the index encounter.19 We assessed the presence of prescriptions for antithrombotics, antidepressants, antipsychotics, and opioid analgesics at the index encounter. Switching from tenofovir disoproxil fumarate (TDF) to tenofovir alafenamide fumarate (TAF) is associated with weight gain and often coincides with an INSTI switch.21 To isolate INSTI effects, we classified TAF changes as: (1) initiation (no TAF in the 180–14 days before, but prescribed 14–180 days after the index encounter); (2) discontinuation (TAF prescribed 180–14 days before, but not 14–180 days after the index encounter); (3) continuation (TAF prescribed in both periods); and (4) no TAF (never prescribed in both periods). TDF changes were classified similarly.15
We assessed between-group differences in covariates at the index encounter using pairwise standardized mean differences for switching to INSTI vs continuing NNRTI and switching to INSTI vs continuing PI.20 An absolute standardized mean difference >0.10 was considered indicative of meaningful imbalance.
In our main analysis, we estimated hazard ratios (HRs) and 95% confidence intervals (CIs) for incident diabetes associated with switching to any INSTI versus continuing an NNRTI or PI using weighted Cox regression with robust variance estimators to account for repeated observations on participants and uncertainty in the estimation of the weights.
We conducted three secondary analyses to further investigate the relationship between INSTI switch and diabetes. First, we assessed the diabetes risk associated with specific agents—bictegravir, dolutegravir, elvitegravir, or raltegravir— compared to continuing an NNRTI or PI. Using a multinomial logistic model, we derived IPTW based on the predicted probability of switching to specific INSTI agents.
Second, to assess whether diabetes risk varied by time since switching to INSTIs, we included an interaction term for follow-up period (first two years vs subsequent years) and INSTI switch, informed by findings from a prior single-site study that reported a significant interaction between INSTI switch and time since switching on diabetes risk.15
Third, to assess whether the risk of diabetes differed by BMI prior to switching to INSTI, we included an interaction term for BMI (≥25 kg/m2 versus <25 kg/m2) based on the most recent measurement within the year leading up to the index encounter. This threshold was selected to distinguish overweight or obese individuals from those with normal or underweight BMI.
To evaluate the extent to which weight gain may explain the relationship between INSTI switch and diabetes, we conducted a sensitivity analysis estimating the effect of switching under a hypothetical intervention restricting weight gain to ≤5% within the first year.22,23 This represents the controlled direct effect of INSTI switch independent of weight gain. To operationalize this, we censored follow-up at the first time a participant’s weight exceeded 105% of their weight at the index encounter. We then applied another set of IPCW to account for the potential selection bias introduced by this censoring.
In a second sensitivity analysis, we explored the main effect over an extended period (January 1, 2007 to December 31, 2022), during which switching decisions may have varied as NNRTI- and PI-based regimens were phased out and INSTI-based regimens became the recommended first-line therapy. Similar censoring criteria were applied for follow-up, except the maximum follow-up time was expanded from 5 to 10 years.
We performed all analyses using R version 4.3.1 and used the Generalized Estimating Equation Package (“geepack”) and Survival Package (“survival”).
Role of the funding source
The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report.
RESULTS
Between 2016 and 2022, 13,071 participants who were followed from 2,702 encounters where they switched to an INSTI from an NNRTI, 54,766 encounters where they continued an NNRTI, 1,714 encounters where they switched to an INSTI from a PI, and 26,599 encounters where they continued a PI (see Appendix p 6 for study sample derivation). Efavirenz accounted for 70% of NNRTI use at the index encounters. Participants were followed for a median (interquartile interval [IQI]) of 2.5 (1.1–4.2) years.
In the main analysis, switching from NNRTIs to INSTIs was associated with an adjusted HR of 1.10 (95% CI, 0.87–1.39; Table 2). Switching from PIs to INSTIs conferred an adjusted HR of 1.38 (95% CI, 1.06–1.80). The adjusted cumulative incidence of diabetes for both comparisons is illustrated in Figure 1.
Table 2.
The risk of incident diabetes associated with switching to INSTIs between January 2016 and December 2022
| N of PWH who ever contributed observations | N of events/N of encounters (%) | Adjusted incidence rate per 1,000 person-years (95% CI) | Adjusted HR (95% CI) a | |
|---|---|---|---|---|
| Switching from NNRTI to INSTI | ||||
| Continued NNRTI | 7,895 | 3,459/54,766 (6.3) | 24.5 (23.7-25.3) | Reference |
| Switched to INSTI | 2,702 | 250/2,702 (9.3) | 27.7 (23.9-32.0) | 1.10 (0.87-1.39) |
| Switching from PI to INSTI | ||||
| Continued PI | 4,459 | 1,509/26,599 (5.7) | 22.7 (21.6-23.9) | Reference |
| Switched to INSTI | 1,714 | 150/1,714 (8.8) | 30.4 (25.7-35.7) | 1.38 (1.06-1.80) |
| Specific INSTI | ||||
| NNRTI or PI | 12,237 | 5041/81,365 (6.2) | 23.9 (23.2-24.5) | Reference |
| Bictegravir | 1,097 | 87/1,097 (7.9) | 29.4 (18.6-44.1) | 1.17 (0.81-1.70) |
| Dolutegravir | 1,953 | 182/1,953 (9.3) | 27.8 (23.5-32.6) | 1.10 (0.87-1.40) |
| Elvitegravir | 1,322 | 127/1,322 (9.6) | 30.1 (22.8-39.0) | 1.21 (0.83-1.75) |
| Raltegravir | 44 | 4/44 (9.1) | 28.7 (9.3-67.0) | 1.14 (0.30-4.41) |
Abbreviations: CI, confidence interval; HR, hazard ratio; INSTI, integrase strand transfer inhibitor; NNRTI, non-nucleoside reverse transcriptase inhibitor; PI, protease inhibitor
Adjusted HRs were derived from weighted Cox regression that accounted for demographic and HIV-related characteristics, comorbidities, co-medications, and laboratory measurements (see Statistical analysis section for the complete list of covariates and weighting method).
Figure 1.
Adjusted cumulative incidence of diabetes after (a) switching from NNRTIs to INSTIs and (b) switching from PIs to INSTIs.
N at risk and N of events represent weighted values.
Abbreviations: INSTI, integrase strand transfer inhibitor; NNRTI, non-nucleoside reverse transcriptase inhibitor; PI, protease inhibitor
Secondary analyses comparing switches to individual INSTI agents versus continuing NNRTIs or PIs are presented in Table 2. Adjusted HRs for diabetes were broadly similar across individual INSTIs, ranging from 1.10 (95% CI, 0.87–1.40) for dolutegravir to 1.21 (95% CI, 0.83–1.75) for elvitegravir.
The diabetes risk was concentrated in the first two years after switching to INSTIs. Compared with continuing NNRTIs, the hazard of diabetes with switching to INSTIs was 1.30 (95% CI, 0.98–1.72) during the first two years post-switch and 0.89 (95% CI, 0.62–1.26; interaction P = 0.08) thereafter. At 2 years, the cumulative risk of diabetes was 5.7% among those who switched to an INSTI versus 4.4% among those who continued an NNRTI (adjusted relative risk, 1.30; 95% CI, 1.01–1.69). Similarly, compared with continuing PIs, the hazard of diabetes with switching to INSTIs was 1.67 (95% CI, 1.21–2.30) during the first two years and 1.08 (95% CI, 0.75–1.57; interaction P = 0.06) thereafter. At 2 years, the cumulative risk was 6.0% among those who switched an INSTI versus 4.0% among those who continued a PI (adjusted relative risk, 1.51; 95% CI, 1.12–2.05).
We did not find significant effect modification by BMI at the index encounter. Among participants switching from NNRTIs to INSTIs, the adjusted HR for incident diabetes was 1.02 (95% CI, 0.79–1.32) for those with BMI ≥25 kg/m2 and 1.47 (95% CI, 0.88–2.45) for those with BMI <25 kg/m2 (interaction P = 0.17). Similarly, among participants switching from PIs to INSTIs, the adjusted HR was 1.40 (95% CI, 1.04–1.88) for those with BMI ≥25 kg/m2 and 1.30 (95% CI, 0.73–2.33) for those with BMI <25 kg/m2 (interaction P = 0.82).
In the sensitivity analysis estimating diabetes risk under a hypothetical scenario in which weight gain did not exceed 5% during the first year, the hazard of diabetes after switching from NNRTIs to INSTIs was 1.03 (95% CI, 0.79–1.36), and from PIs to INSTIs was 1.37 (95% CI, 1.02–1.84), findings that were similar to those observed in the main analysis.
The sample characteristics for the expanded study period from January 2007 through December 2022 are presented in the appendix p 3–4. In this analysis, switching from NNRTIs to INSTIs was associated with an adjusted HR of 1.28 (95% CI, 1.10–1.49), while switching from PIs to INSTIs was associated with an adjusted HR of 1.27 (95% CI, 1.12–1.45). Switches to individual INSTI agents, compared with continuing an NNRTI or PI, were associated with an increased risk of diabetes, with qualitatively similar risk estimates observed across agents during the expanded study period (appendix p 5).
DISCUSSION
In this longitudinal, large-scale cohort of PWH in the US and Canada, switching from PIs to INSTIs was associated with a 38% higher risk of incident diabetes among ART-experienced individuals. In contrast, switching from NNRTIs to INSTIs was associated with a modest but non-significant increase in risk. The excess diabetes risk following a switch from a PI to an INSTI was concentrated in the first two years after the regimen change, but not thereafter. Importantly, the diabetes risk following switching to INSTIs did not appear to be explained by weight gain in the first year after switching.
Evidence on the association between prevalent INSTI use and diabetes risk among ART-experienced PWH has been mixed.12–14 Using target trial emulation, we followed ART-experienced PWH from the decision point to either continue an NNRTI or PI or switch to an INSTI. The higher hazard for diabetes observed when switching from PIs compared with NNRTIs may reflect the greater metabolic toxicity associated with PI use, rendering these individuals more vulnerable to additional effects of INSTI exposure, particularly in the period shortly after the switch. An alternative explanation would be the contribution of residual confounding, whereby those with higher baseline diabetes risk were preferentially switched from PIs to INSTIs. While direct head-to-head comparisons of INSTI versus PI for diabetes risk have been limited in the real-world setting, the observed hazard ratio of 1.10 for diabetes after switching to INSTI versus continuing NNRTI in our study was similar to the hazard ratio of 1.17 for diabetes reported in a study of ART-naïve PWH in NA-ACCORD initiating INSTI versus NNRTI.4 Another cohort study comparing the initiation of INSTI with non-INSTI (NNRTI or PI) in ART-naïve PWH also found approximately 30% increased risk of diabetes or hyperglycemia, defined using diagnosis codes at 6 months of follow-up.5
Switching to individual INSTI agents produced qualitatively similar effects on new-onset diabetes. Although none of the individual risk estimates reached statistical significance in our main analysis spanning 2016–2022, in a sensitivity analysis extending the study period to 2007–2022 yielded similar but statistically significant estimates with the larger sample. The ordering of point estimates across agents differed by period, and prior studies in ART-naïve PWH have also reported inconsistent rankings.4,5 Overall, our findings suggest a consistent diabetes risk across INSTI agents, supporting a potential class effect that merits consideration in clinical decision-making.
Importantly, accounting for weight gain during the first year after switching to INSTIs only modestly attenuated this association, consistent with prior findings in ART-naïve populations showing a 5% reduction after adjusting for 1-year weight change.4 Together, these findings imply that mechanisms beyond weight gain alone may contribute to the increased risk of diabetes observed with INSTI use. In particular, insulin resistance has been shown to develop soon after INSTI initiation, preceding measurable changes in body weight or fat distribution, raising the possibility of direct drug effects on glucose metabolism.24 Inflammatory and immune activation pathways may also contribute, and recent translational studies show that INSTIs can directly affect adipocyte biology, leading to oxidative stress, mitochondrial dysfunction, and insulin resistance.25 Thus, the metabolic consequences of INSTIs are likely multifactorial, with several potential pathways contributing to the elevated diabetes risk.
Our analysis indicates that PWH switching to INSTIs face a significantly increased risk of diabetes within the first two years, suggesting a possible period of heightened vulnerability. Prior studies have also reported early metabolic changes, including increased insulin resistance, soon after INSTI initiation.24,26 These findings underscore the need for close monitoring shortly after switching, though they warrant cautious interpretation given potential residual confounding. Notably, this temporal pattern parallels observations for cardiovascular disease risk in other cohorts. In the RESPOND cohort (Europe and Australia), cardiovascular disease risk was increased during the first two years after INSTI exposure.27 Similarly, the HIV-CAUSAL Collaboration (North American and Europe) reported an early separation of cardiovascular disease risk curves, favoring higher risk with INSTI exposure compared with non-exposure, which later converged.28 While our study did not evaluate cardiovascular disease risk, these parallels raise the possibility that cardiometabolic effects of INSTIs may be most pronounced early in the exposure period, a hypothesis that warrants further investigation. Interestingly, diabetes risk did not differ by BMI in our study, consistent with a study from the RESPOND cohort, where BMI adjustment only modestly attenuated diabetes risk.14
Our study is a real-world investigation of diabetes risk after switching to INSTIs, using large-scale longitudinal data representative of PWH in the US, which supports external validity.29 We emulated a target trial with nested trials approach, comparing switching to INSTIs versus continuing NNRTIs or PIs, a rigorous design to mitigate biases and strengthen internal validity.17 We used inverse probability weighting to address potential confounding from nonrandom treatment assignment and informative censoring. We also accounted for concurrent changes and use of TDF and TAF, further enhancing the robustness of our findings. Nonetheless, the effect of switching from TDF to TAF on diabetes risk remains an important question that warrants dedicated investigation.
Our study also has limitations. First, as an observational analysis, residual confounding is possible despite our rigorous methods. Confounding by indication may exist if clinicians preferentially switched PWH with greater metabolic risk to INSTIs. We addressed this by adjusting for measured risk factors and restricting the main analysis to 2016–2022, when INSTIs were recommended as first-line ART.3 Nonetheless, the higher diabetes risk observed when switching from PIs—agents with known metabolic side effects—than from NNRTIs may still reflect residual confounding, even after adjustment for cardiometabolic risk factors. Unmeasured factors such as diet or physical activity may also have influenced results. Additionally, the attenuation of the association after the first two years may be inconsistent with a typical dose–response. This could reflect a short-term induction effect—where risk increases only initially—or residual confounding despite rigorous methods. Second, as prescription records were used to ascertain drug exposure, their actual intake by the patient could not be guaranteed. Third, hemoglobin A1c measurements may underestimate glycemia in PWH, which could have led to underestimation of diabetes cases.30 Fourth, our study setting was the US and Canada, which may limit generalizability to other regions. Lastly, long-acting cabotegravir was not assessed.
In conclusion, switching to INSTIs from PIs was associated with a significantly increased risk of incident diabetes among ART-experienced PWH in diverse, longitudinal HIV cohorts in the US and Canada. As cardiometabolic diseases increasingly drive morbidity and mortality among PWH, understanding the potential metabolic impact of ART is increasingly important. Our analysis is not intended to discourage switching to INSTIs, but rather to inform PWH and their treating clinicians of the potential metabolic risks that should be weighed in making individualized treatment decisions in such common clinical scenarios. For example, our findings may help guide shared decision making when switching to INSTIs is considered and highlight the importance of diabetes screening, particularly during the first two years after the switch. Future studies could compare diabetes risk among PWH switching to INSTIs vs switching to other antiretroviral classes, and to identify subpopulations who may be at higher risk after switching.
Supplementary Material
Table 1.
Characteristics of participants at HIV clinical encounters where they switched to INSTIs compared versus continued NNRTIs or PIs between January 2016 and December 2022
| Clinical encounters where participants switched to INSTI versus continued NNRTI | Clinical encounters where participants switched to INSTI versus continued PI | |||
|---|---|---|---|---|
| Switched to INSTI | Continued NNRTI | Switched to INSTI | Continued PI | |
| N of participants who contributed at least one observation | 2,702 | 7,895 | 1,714 | 4,459 |
| N of clinical encounters | 2,702 | 54,766 | 1,714 | 26,599 |
| Age, median (IQI) | 55 (46-62) | 56 (47-64) | 55 (47-62) | 55 (47-63) |
| Male sex at birth | 2,452 (91%) | 50,551 (92%) | 1,446 (84%) | 23,218 (87%) |
| Race | ||||
| Black | 1,052 (39%) | 21,305 (39%) | 753 (44%) | 11,432 (43%) |
| White | 1,382 (51%) | 27,625 (50%) | 806 (47%) | 13,170 (50%) |
| Other | 198 (7%) | 4,494 (8%) | 118 (7%) | 1,524 (6%) |
| Unknown | 70 (3%) | 1,342 (2%) | 37 (2%) | 473 (2%) |
| Year of encounter | ||||
| 2016 | 973 (36%) | 17,281 (32%) | 776 (45%) c | 9,650 (36%) |
| 2017–2018 | 1,105 (41%) | 22,412 (41%) | 651 (38%) | 10,898 (41%) |
| 2019–2020 | 548 (20%) | 10,582 (19%) | 244 (14%) | 4,379 (16%) |
| 2021–2022 | 76 (3%) b | 4,491 (8%) | 43 (3%) c | 1,672 (6%) |
| Duration of NNRTI or PI use prior to clinical encounter in years, median (IQI) | 6 (3-10) | 8 (4-11) | 5 (3-9) | 7 (4-11) |
| Hemoglobin A1c, median (IQI) a | 5.5 (5.2-5.8) | 5.5 (5.2-5.8) | 5.4 (5.1-5.7) | 5.4 (5.1-5.7) |
| BMI in kg/m2, median (IQI) | 26 (23-30) | 27 (24-30) | 26 (23-30) | 26 (23-30) |
| BMI categories | ||||
| Underweight | 65 (2%) | 1,198 (2%) | 40 (2%) | 721 (3%) |
| Normal weight | 984 (36%) | 18,944 (35%) | 609 (36%) | 9,146 (34%) |
| Overweight | 1,004 (37%) | 20,311 (37%) | 641 (37%) | 9,773 (37%) |
| Obese | 649 (24%) | 14,313 (26%) | 424 (25%) | 6,959 (26%) |
| Viral suppression | 2,556 (95%) | 52,380 (96%) | 1,578 (92%) | 24,532 (92%) |
| CD4 count ≤200 cells per μL | 44 (2%) | 732 (1%) | 73 (4%) | 953 (4%) |
| Comorbidities | ||||
| Hypertension | 1,082 (40%) | 23,761 (43%) | 678 (40%) | 11,440 (43%) |
| Hyperlipidemia | 1,220 (45%) | 26,444 (48%) | 758 (44%) | 12,927 (49%) |
| Chronic kidney disease | 349 (13%) | 7,108 (13%) | 347 (20%) | 5,002 (19%) |
| Co-medications | ||||
| Antithrombotic | 61 (2%) | 1,212 (2%) | 46 (3%) | 794 (3%) |
| Antidepressant | 584 (22%) | 12,115 (22%) | 421 (25%) | 7,371 (28%) |
| Antipsychotic | 98 (4%) | 1,585 (3%) | 88 (5%) | 1,834 (7%) |
| Opioid | 237 (9%) | 5,019 (9%) | 195 (11%) | 3,927 (15%) |
| Concomitant TAF use at clinical encounter | ||||
| Initiated TAF | 1,532 (57%) b | 5,594 (10%) | 696 (41%) c | 3,202 (12%) |
| Continued TAF | 263 (10%) b | 16,058 (29%) | 290 (17%) c | 7,796 (29%) |
| Discontinued TAF | 30 (1%) | 188 (0%) | 15 (1%) | 152 (1%) |
| None | 877 (32%) b | 32,926 (60%) | 713 (42%) c | 15,449 (58%) |
| Concomitant TDF use at clinical encounter | ||||
| Initiated TDF | 14 (1%) | 103 (0%) | 26 (2%) | 161 (1%) |
| Continued TDF | 1,101 (41%) b | 32,913 (60%) | 526 (31%) c | 12,452 (47%) |
| Discontinued TDF | 1,011 (37%) b | 2,857 (5%) | 468 (27%) c | 1,877 (7%) |
| None | 576 (21%) b | 18,893 (34%) | 694 (40%) | 12,109 (46%) |
Abbreviations: BMI, body mass index; INSTI, integrase strand transfer inhibitor; IQI, interquartile interval; NNRTI, non-nucleoside reverse transcriptase inhibitor; PI, protease inhibitor; TAF, tenofovir alafenamide; TDF, tenofovir disoproxil fumarate
The most recent hemoglobin A1c measurement within the year leading up to the clinical encounter. Such measurement was available for 1,201 (44%) of encounters where a switch from an NNRTI to an INSTI occurred and 26,842 (49%) of encounters where an NNRTI was continued. Such measurement was available for 721 (42%) of encounters where a switch from a PI to an INSTI occurred and 12,556 (47%) of encounters where a PI was continued.
Indicates absolute standardized mean differences >0.10 when comparing clinical encounters in which participants switched to an INSTI with those in which participants continued an NNRTI.
Indicates absolute standardized mean differences >0.10 when comparing clinical encounters in which participants switched to an INSTI with those in which participants continued a PI.
RESEARCH IN CONTEXT.
Evidence before this study
We searched PubMed for observational and randomized studies evaluating the risk of incident diabetes associated with integrase strand transfer inhibitors (INSTIs) among people with HIV (PWH). Search terms included “integrase strand transfer inhibitor,” “INSTI”, “bictegravir”, “dolutegravir”, “elvitegravir”, “raltegravir”, and “diabetes,” covering the period from January 1, 2007 to June 30, 2025. We identified observational studies of heterogeneous scale that evaluated diabetes risk associated with initiation or prevalent use of INSTIs among people with HIV (PWH). Four studies that investigated the diabetes risk associated with initiating INSTIs: one study found a higher diabetes risk when compared with initiating non-nucleoside reverse transcriptase inhibitors (NNRTIs) and two studies finding higher risk of diabetes when compared with starting non-INSTI-based regimens; however, this trend was not observed across all studies; for example, one analysis reported similar diabetes incidence rates between dolutegravir- and darunavir-based regimens. Among PWH with prevalent use of antiretroviral therapy (ART), two studies found an association between INSTI use and higher diabetes risk, whereas another study did not observe such a relationship.
Added value of this study
While previous studies have examined the relationship between INSTI initiation or prevalent INSTI use and diabetes risk, they provide limited information for ART-experienced PWH who are considering a switch to INSTI-based therapy. This study adds value by directly addressing this gap. Using a large multi-site cohort of PWH receiving NNRTI- or protease inhibitor (PI)–based regimens, we estimated the effect of switching to an INSTI on subsequent diabetes risk. Additionally, we evaluated whether diabetes risk varies according to time since switching and whether early weight gain explains the observed association. These analyses offer greater clarity on the timing and potential mechanisms of risk. Together, this study refines understanding of the potential metabolic implications of switching to INSTI-based regimens and informs clinical decision-making around such regimen changes.”
Implications of all the available evidence
Together with prior studies that raised concern regarding the increased risk of diabetes associated with INSTI initiation in ART-naïve PWH, our findings indicate that ART-experienced individuals who switch from a PI to an INSTI also face higher risk. A growing body of evidence implicates INSTIs with diabetes risk across both ART initiation and switching contexts, underscoring the need to weigh metabolic risk when selecting ART regimens.
Acknowledgments
Dr. Hwang was supported by T32DK062707 from the National Institute of Diabetes and Digestive and Kidney Diseases for this work. Dr. Lesko was supported by K01AA028193 from the National Institute on Alcohol Abuse and Alcoholism for this work. Dr. Brown was supported by K24AI120834 from the National Institute of Allergy and Infectious Diseases for this work. Dr. Zalla was supported by K99AI181608 from the National Institute of Allergy and Infectious Diseases for this work. Dr. Pytell was supported by K23DA060358 from the National Institute on Drug Abuse for this work. Dr. Tseng was supported by R03DK135898 from the National Institute of Diabetes and Digestive and Kidney Diseases for this work. Dr. Moore was supported by U01DA036935 from the National Institute on Drug Abuse and P30AI094189 from the National Institute of Allergy and Infectious Diseases for this work. Dr. Fojo was supported by R01MD018539 from the National Institute on Minority Health and Health Disparities and K08MH118094 from the National Institute of Mental Health for this work.
The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health (NIH). This manuscript is the result of funding in whole or in part by the NIH. It is subject to the NIH Public Access Policy. Through acceptance of this federal funding, NIH has been given a right to make this manuscript publicly available in PubMed Central upon the Official Date of Publication, as defined by NIH. This work was supported by National Institutes of Health grants U01AI069918, F31AI124794, F31DA037788, G12MD007583, K01AI093197, K01AI131895, K23EY013707, K24AI065298, K24AI118591, K24DA000432, KL2TR000421, N01CP01004, N02CP055504, N02CP91027, P30AI027757, P30AI027763, P30AI027767, P30AI036219, P30AI050409, P30AI050410, P30AI094189, P30AI110527, P30MH62246, R01AA016893, R01DA011602, R01DA012568, R01 AG053100, R24AI067039, U01AA013566, U01AA020790, U01AI038855, U01AI038858, U01AI068634, U01AI068636, U01AI069432, U01AI069434, U01DA03629, U01DA036935, U10EY008057, U10EY008052, U10EY008067, U01HL146192, U01HL146193, U01HL146194, U01HL146201, U01HL146202, U01HL146203, U01HL146204, U01HL146205, U01HL146208, U01HL146240, U01HL146241, U01HL146242, U01HL146245, U01HL146333, U24AA020794,U54MD007587, UL1RR024131, UL1TR000004, UL1TR000083, Z01CP010214 and Z01CP010176; contracts CDC-200-2006-18797 and CDC-200-2015-63931 from the Centers for Disease Control and Prevention, USA; contract 90047713 from the Agency for Healthcare Research and Quality, USA; contract 90051652 from the Health Resources and Services Administration, USA; grants CBR-86906, CBR-94036, HCP-97105 and TGF-96118 from the Canadian Institutes of Health Research, Canada; Ontario Ministry of Health and Long Term Care; and the Government of Alberta, Canada. Additional support was provided by the National Institute Of Allergy And Infectious Diseases (NIAID), National Cancer Institute (NCI), National Heart, Lung, and Blood Institute (NHLBI), Eunice Kennedy Shriver National Institute Of Child Health & Human Development (NICHD), National Human Genome Research Institute (NHGRI), National Institute for Mental Health (NIMH) and National Institute on Drug Abuse (NIDA), National Institute On Aging (NIA), National Institute Of Dental & Craniofacial Research (NIDCR), National Institute Of Neurological Disorders And Stroke (NINDS), National Institute Of Nursing Research (NINR), National Institute on Alcohol Abuse and Alcoholism (NIAAA), National Institute on Deafness and Other Communication Disorders (NIDCD), and National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK).
Funding:
US National Institutes of Health.
Footnotes
Declaration of interests
Dr. Brown has served as a consultant to ViiV Healthcare, Gilead Sciences, Janssen, and EMD Serono. Dr. Alexander is past Chair of FDA’s Peripheral and Central Nervous System Advisory Committee and a co-founding Principal and equity holder in Stage Analytics. Dr. Alexander’s arrangements have been reviewed and approved by Johns Hopkins University in accordance with its conflict-of-interest policies. Dr. Althoff received grants from NIH, royalties from a Coursera specialization that she directs, and travel and lodging support for her role on the Scientific Steering Committee of the International Workshop on HIV and Hepatitis C virus Observational Databases. Dr. Pytell received grants from NIH. Dr. Falade-Nwulia received grants from NIH and AbbVie, consulting fees from Gilead Sciences, and served a leadership role in the International Network on Health and Hepatitis in Substance Users. Dr. Moore received grants from NIH. Dr. Marconi received grants from U.S. Department of Veterans Affairs (VA), Centers for Disease Control and Prevention (CDC), NIH, Gilead Sciences, ViiV Healthcare, and Merck & Co; served on the advisory board for the following studies: IL-1b inhibitor study, OuTSMART, CLEAR HIV, ECLIPSE, and VB201; and served as a study section chair for NIH. Dr. Koethe served as an advisor to, and received grants from, Merck & Co. and Gilead Sciences. Dr. Horberg received grants from NIH. Dr. Sterling received grants from NIH. Other authors have no potential conflicts of interest to report.
Data sharing statements
The data used in this study were obtained from the North American AIDS Cohort Collaboration on Research and Design (NA-ACCORD). These data are subject to data use agreements and governance policies established by the NA-ACCORD. As such, individual investigators are not permitted to share the data directly. Researchers interested in accessing NA-ACCORD data may submit a formal request through the NA-ACCORD (naaccord.org), subject to review and approval in accordance with established procedures.
REFERENCES
- 1.Althoff KN, Stewart C, Humes E, et al. The forecasted prevalence of comorbidities and multimorbidity in people with HIV in the United States through the year 2030: A modeling study. PLoS Med 2024; 21: e1004325. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Koethe JR, Lagathu C, Lake JE, et al. HIV and antiretroviral therapy-related fat alterations. Nat Rev Dis Primers 2020; 6: 48. [DOI] [PubMed] [Google Scholar]
- 3.Günthard HF, Saag MS, Benson CA, et al. Antiretroviral Drugs for Treatment and Prevention of HIV Infection in Adults: 2016 Recommendations of the International Antiviral Society–USA Panel. JAMA 2016; 316: 191–210. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Rebeiro PF, Jenkins CA, Bian A, et al. Risk of Incident Diabetes Mellitus, Weight Gain, and Their Relationships With Integrase Inhibitor–Based Initial Antiretroviral Therapy Among Persons With Human Immunodeficiency Virus in the United States and Canada. Clin Infect Dis 2021; 73: e2234–42. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.O’Halloran JA, Sahrmann J, Parra-Rodriguez L, et al. Integrase Strand Transfer Inhibitors Are Associated With Incident Diabetes Mellitus in People With Human Immunodeficiency Virus. Clin Infect Dis 2022; 75: 2060–5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Ursenbach A, Sireyjol A, Delpierre C, et al. Increased incidence of diabetes in people living with HIV treated with first-line integrase strand transfer inhibitors: A French multicentre retrospective study. HIV Med 2025; 26: 166–72. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Hsu R, Brunet L, Fusco JS, et al. Incident type 2 diabetes mellitus after initiation of common HIV antiretroviral drugs. AIDS 2021; 35: 81–90. [DOI] [PubMed] [Google Scholar]
- 8.Venter WDF, Moorhouse M, Sokhela S, et al. Dolutegravir plus Two Different Prodrugs of Tenofovir to Treat HIV. N Engl J Med 2019; 381: 803–15. [DOI] [PubMed] [Google Scholar]
- 9.Bansi-Matharu L, Phillips A, Oprea C, et al. Contemporary antiretrovirals and body-mass index: a prospective study of the RESPOND cohort consortium. Lancet HIV 2021; 8: e711–22. [DOI] [PubMed] [Google Scholar]
- 10.Pantazis N, Sabin CA, Grabar S, et al. Changes in bodyweight after initiating antiretroviral therapy close to HIV-1 seroconversion: an international cohort collaboration. Lancet HIV 2024; 11: e660–9. [DOI] [PubMed] [Google Scholar]
- 11.Herrin M, Tate JP, Akgün KM, et al. Weight Gain and Incident Diabetes Among HIV-Infected Veterans Initiating Antiretroviral Therapy Compared With Uninfected Individuals. J Acquir Immune Defic Syndr 2016; 73: 228–36. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Spieler G, Westfall AO, Long DM, et al. Trends in diabetes incidence and associated risk factors among people with HIV in the current treatment era. AIDS 2022; 36: 1811–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Taramasso L, Squillace N, Ricci E, et al. Incident diabetes in course of antiretroviral therapy. AIDS 2023; 37: 1269–76. [DOI] [PubMed] [Google Scholar]
- 14.Rupasinghe D, Bansi-Matharu L, Law M, et al. Integrase Strand Transfer Inhibitor–Related Changes in Body Mass Index and Risk of Diabetes: A Prospective Study From the RESPOND Cohort Consortium. Clin Infect Dis 2025; 80: 404–16. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Hwang YJ, Lesko CR, Brown TT, et al. Association between switching to integrase strand transfer inhibitors and incident diabetes in people with HIV. AIDS 2024; 38: 1696–702. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Gange SJ, Kitahata MM, Saag MS, et al. Cohort Profile: The North American AIDS Cohort Collaboration on Research and Design (NA-ACCORD). Int J Epidemiol 2007; 36: 294–301. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Danaei G, Rodríguez LAG, Cantero OF, Logan R, Hernán MA. Observational data for comparative effectiveness research: An emulation of randomised trials of statins and primary prevention of coronary heart disease. Stat Methods Med Res 2013; 22: 70–96. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Althoff KN, Wong C, Hogan B, et al. Mind the gap: observation windows to define periods of event ascertainment as a quality control method for longitudinal electronic health record data. Ann Epidemiol 2019; 33: 54–63. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Wong C, Gange SJ, Moore RD, et al. Multimorbidity Among Persons Living with Human Immunodeficiency Virus in the United States. Clin Infect Dis 2018; 66: 1230–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Austin PC, Stuart EA. Moving towards best practice when using inverse probability of treatment weighting (IPTW) using the propensity score to estimate causal treatment effects in observational studies. Stat Med 2015; 34: 3661–79. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Surial B, Mugglin C, Calmy A, et al. Weight and Metabolic Changes After Switching From Tenofovir Disoproxil Fumarate to Tenofovir Alafenamide in People Living With HIV: A Cohort Study. Ann Intern Med 2021; 174: 758–67. [DOI] [PubMed] [Google Scholar]
- 22.VanderWeele TJ. Marginal Structural Models for the Estimation of Direct and Indirect Effects. Epidemiology 2009; 20: 18–26. [DOI] [PubMed] [Google Scholar]
- 23.Hernán MA, Hernández-Díaz S. Beyond the intention-to-treat in comparative effectiveness research. Clin Trials 2012; 9: 48–55. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Dirajlal-Fargo S, Moser C, Brown TT, et al. Changes in Insulin Resistance After Initiation of Raltegravir or Protease Inhibitors With Tenofovir-Emtricitabine: AIDS Clinical Trials Group A5260s. Open Forum Infect Dis 2016; 3: ofw174. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Gorwood J, Bourgeois C, Pourcher V, et al. The Integrase Inhibitors Dolutegravir and Raltegravir Exert Proadipogenic and Profibrotic Effects and Induce Insulin Resistance in Human/Simian Adipose Tissue and Human Adipocytes. Clin Infect Dis 2020; 71: e549–60. [DOI] [PubMed] [Google Scholar]
- 26.Summers NA, Lahiri CD, Angert CD, et al. Metabolic Changes Associated With the Use of Integrase Strand Transfer Inhibitors Among Virally Controlled Women. J Acquir Immune Defic Syndr 2020; 85: 355–62. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Neesgaard B, Greenberg L, Miró JM, et al. Associations between integrase strand-transfer inhibitors and cardiovascular disease in people living with HIV: a multicentre prospective study from the RESPOND cohort consortium. Lancet HIV 2022; 9: e474–85. [DOI] [PubMed] [Google Scholar]
- 28.Rein SM, Lodi S, Logan RW, et al. Integrase strand-transfer inhibitor use and cardiovascular events in adults with HIV: an emulation of target trials in the HIV-CAUSAL Collaboration and the Antiretroviral Therapy Cohort Collaboration. Lancet HIV 2023; 10: e723–32. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Characteristics of NA-ACCORD participants and people with HIV in the US as of 2021. The North American AIDS Cohort Collaboration on Research and Design (NA-ACCORD). 2021. https://naaccord.org/vital-statistics (accessed Sept 25, 2025).
- 30.Slama L, Palella FJ, Abraham AG, et al. Inaccuracy of haemoglobin A1c among HIV-infected men: effects of CD4 cell count, antiretroviral therapies and haematological parameters. J Antimicrob Chemother 2014; 69: 3360–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The data used in this study were obtained from the North American AIDS Cohort Collaboration on Research and Design (NA-ACCORD). These data are subject to data use agreements and governance policies established by the NA-ACCORD. As such, individual investigators are not permitted to share the data directly. Researchers interested in accessing NA-ACCORD data may submit a formal request through the NA-ACCORD (naaccord.org), subject to review and approval in accordance with established procedures.