The effect of hypothetical interventions to limit weight gain on incident type 2 diabetes mellitus among people with HIV initiating antiretroviral therapy in the United States, 2012–2021

Abstract

Among people with HIV (PWH), initiating antiretroviral therapy (ART) is associated with weight gain that might increase the risk of cardiometabolic outcomes. We estimated the effect of hypothetical interventions to restrict increases in body mass index (BMI) in the two years after ART initiation on the 7-year risk of incident type-2 diabetes mellitus (T2DM) among PWH in the North American AIDS Cohort Collaboration on Research and Design (NA-ACCORD) clinical cohorts using the parametric extended g-formula. Among 13,822 normal weight or overweight PWH (BMI ≥18.5 and ≤29.9 kg/m²) who initiated ART between 2012–2021, the 7-year risk of T2DM was 5.6% (95% confidence interval [CI]: 4.8%, 6.3%). Compared to no intervention on BMI, if BMI increase during the first 2 years was capped at 5% of baseline BMI, the risk ratio (RR) for T2DM risk was 0.94 (95% CI: 0.86, 1.03); if BMI was capped at 29.9 kg/m², RR=0.88 (95% CI: 0.80, 0.97); and if PWH were prevented from gaining any weight over the first 2 years, RR=0.81 (95% CI: 0.72, 0.92). These results provide preliminary evidence to promote studies of real weight maintenance interventions among PWH initiating ART.

INTRODUCTION

In the current treat-all era of the human immunodeficiency virus (HIV) endemic, people with HIV (PWH) are initiating antiretroviral therapy (ART) earlier in the course of their infection and staying on ART for decades. Untreated HIV infection is associated with weight loss and metabolic disruptions, but ART initiation is also increasingly recognized as being associated with weight gain exceeding a return to health or aging-related weight gain trajectory¹ and exceeding weight gain that would be expected in people without HIV.^2,3 There are ongoing investigations into which specific antiretroviral medications (ARVs) may be weight-increasing (e.g., integrase strand transfer inhibitors [INSTI], protease inhibitors, and tenofovir alafenamide fumarate [TAF]), weight-neutral (e.g., doravirine), or weight-suppressing (e.g., efavirenz or tenofovir disoproxil fumarate [TDF]).^4–6 ART-associated weight gain that results in obesity—about a fifth of PWH having a normal body mass index (BMI) at ART initiation develop obesity within 3 years¹—may have implications for the risk of cardiometabolic outcomes and other health complications. HIV infection may or may not be independently related to type 2 diabetes mellitus (T2DM) risk, however PWH with HCV coinfection and PWH prescribed certain ART regimens may be at increased risk of T2DM.^7–10 Over 1-in-4 PWH are projected to have T2DM by 2030.¹¹

The clinical implications of specific ARVs on cardiometabolic outcomes is being investigated with sometimes conflicting results.^8,9,12–16 However, given that PWH will take ART for the lifetime,^17,18 an equally or more clinically relevant question is, ‘can mitigating the effects of early weight gain after ART initiation reduce health complications among PWH?’ Before, or in parallel with, investigations into specific weight-maintenance interventions for PWH, we sought to estimate the degree to which modest interventions to prevent or reduce weight gain in the first 2 years after ART initiation could impact the overall burden of T2DM in PWH. The first 2 years on ART is a period with more frequent clinical and laboratory monitoring, which presents an opportunity to monitor weight gain and early intervention, such as healthy diet, exercise, etc. Additionally, there is some evidence that excess weight gain among PWH relative to the general population is concentrated in the first 2 years on ART.^2,3 Estimating the effect of hypothetical weight-maintenance interventions can calibrate expectations regarding the potential effects of a real intervention.

Herein, we estimated the effect of hypothetical interventions that would restrict weight gain in the two years following ART initiation on the 7-year risk of incident T2DM in a cohort of PWH who initiated ART as part of routine clinical care in the modern ART era.

METHODS

Study Sample

We used data from the North American AIDS Cohort Collaboration on Research and Design (NA-ACCORD), the North American region of the International epidemiology Databases to Evaluate AIDS (IeDEA) project.¹⁹ The NA-ACCORD has been described elsewhere, but briefly, single- and multi-site clinical and interval cohorts prospectively collect data on adults with HIV living in the United States (US) and Canada, which are then combined and harmonized. Data collection and analyses have been approved by local institutional review boards and the [anonymized for review]. Clinical cohorts include adults successfully linked to and engaged in care (defined as attending ≥2 clinic visits in 12 months) who consented to share their data. Patient demographics, HIV acquisition risk factor(s), prescribed medications, laboratory results, and dates of attended clinical visits were curated from patient medical records.

Our study sample consisted of adults enrolled in NA-ACCORD-contributing clinical cohorts in the US who were ART-naïve (HIV RNA >400 copies/mL and no prior record of ART treatment) when they initiated ART between January 1, 2012 and December 1, 2021. To ensure clinical equipoise, we excluded people who were underweight (BMI <18.5 kg/m²) or obese (BMI ≥30 kg/m²) at ART initiation, as different weight-related guidance would apply to those groups; our study sample was restricted to PWH who had a BMI in the normal (18.5 to 24.9 kg/m²) or overweight (25.0 to 29.9 kg/m²) category at ART initiation. We excluded people with diabetes (any type) prior to ART initiation.

We followed people from ART initiation until the first of T2DM, death, loss to follow-up (the last attended clinic visit, CD4 cell count, or viral load test prior to a 2-year gap in monitoring),²⁰ or administrative censoring at 7 years of follow-up or the site-specific administrative end of follow-up (December 31, 2021 for most sites). Deaths were ascertained from clinical sources and regular matches to the National Death Index.

Overview of Approach

We estimated exposure contrasts that have been called “intervention effects,”^21–23 which compared the predicted risk of T2DM had we intervened to prevent weight gain according to different intervention rules (the counterfactual risk) to the risk under the natural course. We first estimated 7-year risk of T2DM under the natural course²⁴—the risk of T2DM in the absence of any interventions on weight (but accounting for loss to follow-up that may depend on covariates).^24,25 We then used the extended parametric g-formula to estimate the risk of T2DM that would have been observed under a number of hypothetical interventions to restrict weight gain during the first 2 years after ART initiation. The designation of “extended” refers to the fact that we modeled our hypothetical interventions by changing (capping) the value of BMI that would have been observed in the absence of the intervention.²⁵

Body Mass Index

Weight was collected at clinic visits as part of routine care. BMI at ART initiation was defined as the BMI most proximal to ART initiation measured in the year prior to ART initiation, or if no BMI measurement was available prior to ART initiation, the first BMI measurement up to 3 months after ART initiation was used. Over follow-up, BMI was defined as the most recent measurement of weight in kilograms divided by the square of their height in meters (kg/m²). Pregnancy is not routinely captured in the NA-ACCORD data, and it is possible that we misclassified weight gain due to pregnancy as unhealthy weight gain; however, we anticipate pregnancies are rare in this sample (only 10% female and median age = 39 years).

Hypothetical Interventions

We considered three targets of a hypothetical intervention on BMI over the first two years post-ART initiation among individuals who would otherwise experience an increase in their BMI. 1) We first modeled a hypothetical intervention to cap individuals’ BMI in the first 2 years at 29.9 kg/m² (e.g., someone who started ART with BMI = 20 kg/m² could gain weight until BMI = 29.9 kg/m² and then would be restricted from gaining any more weight). This cap aligns with the hypothetical intervention “do not allow people to develop obesity.” 2) We next modeled a hypothetical intervention to cap individuals’ BMI increase at a 5% increase relative to baseline BMI (e.g., someone who started ART with BMI = 20 kg/m² could gain weight until BMI = 20 kg/m² × 1.05 = 21 kg/m² and then would be restricted from gaining any more weight). 3) In our most restrictive scenario, we modeled a hypothetical intervention to prevent any weight gain over the first two years on ART (e.g., someone who started ART with BMI = 20 kg/m² would be restricted from gaining any weight). Under all hypothetical interventions, individuals would be allowed to lose weight and re-gain weight that they lost, as long as their weight did not exceed the cap for that hypothetical intervention.

Outcome: Type 2 Diabetes Mellitus

We defined an incident T2DM diagnosis as occurring in the month in which an individual first had: a measurement of hemoglobin A1c (HgA1c) >6.5%; a prescription for a diabetes-specific medication; or a clinical diagnosis of type 2 or unspecified diabetes and a prescription for a diabetes-related medication (Appendix A includes lists of diabetes-specific and diabetes-related medications). This operationalized definition of diabetes was developed based on extensive data quality and completeness checks, with input from the NA-ACCORD Scientific Working Groups.

Covariates

For this analysis, we considered the following time-fixed and time-varying covariates to be confounders; confounders were selected based on a directed acyclic graph (Supplemental Figure 1).²⁶ Time-fixed confounders, defined at baseline (ART initiation), included age, sex at birth, race and ethnicity, HIV acquisition risk factors reported at enrollment into care (men who have sex with men (MSM), history of injection drug use (IDU), and heterosexual sex) coded as indicator variables (not mutually exclusive), smoking status (ever/never), ever infected with hepatitis C virus (HCV), and history of anxiety or depression (based on prior clinical diagnoses). We discretized time to the person-month for analysis; if there were multiple values of a time-varying covariate in the same month, we took the median of the values. Time-varying confounders included blood glucose level, total cholesterol, treated hypertension (a diagnosis of hypertension and a prescription of anti-hypertension medication—see Appendix A), CD4 cell count, and estimated glomerular filtration rate (eGFR). The baseline values of time-varying confounders (e.g., CD4 cell count at ART initiation) were also considered for inclusion as independent variables in the models described below. Laboratory values were all drawn from routine laboratory measurements taken as part of routine clinical care; median time between observations within individuals ranged from every 3.7 months (BMI) to 9.2 months (cholesterol). We used the last value carried forward for time-varying confounders. Baseline values of all laboratory measurements were defined as those drawn most proximal to ART initiation in the year prior to ART initiation, or if unavailable, the first measurement drawn up to 3 months after ART initiation. If no value was available in the specified time frame, we multiply imputed the missing value (k=20) and combined estimates across the 20 imputed datasets using Rubin’s rules.²⁷

Statistical Analysis

We used the parametric extended g-formula to estimate the 7-year cumulative incidence (risk) of T2DM under the natural course and three different hypothetical intervention scenarios defined above.^25,28 Descriptions of the parametric g-formula and parametric extended g-formula have been published elsewhere.^25,28–30 Briefly, we first fit parametric (pooled logistic or linear) models for:

1. BMI in month $j=1\dots J$, conditional on all time-fixed covariates and time-varying covariates in month $j-1$;

2. blood glucose in month $j$, conditional on time-fixed covariates, time-varying covariates other than BMI in month $j-1$, and BMI in month $j$;

3. total cholesterol in month $j$, conditional on time-fixed covariates, time-varying covariates other than BMI and blood glucose in month $j-1$, and BMI and blood glucose in month $j$;

4. incident treated hypertension in month $j$ among those not having treated hypertension, conditional on time-fixed covariates, time-varying covariates other than BMI, blood glucose, and total cholesterol in month $j-1$, and BMI, blood glucose, and total cholesterol in month $j$;

5. CD4 cell count in month $j$, conditional on time-fixed covariates, eGFR in month $j-1$, and BMI, blood glucose, total cholesterol, and hypertension status in month $j$;

6. eGFR in month $j$, conditional on time-fixed covariates and the other time-varying covariates in month $j$;

7. incident T2DM in month $j$ (the outcome), conditional on time-fixed covariates and the other time-varying covariates in month $j$; and

8. death in month $j+1$, conditional on time-fixed covariates, time-varying covariates in month $j$, and not having had incident T2DM in month $j$.

Full details on covariate specification for the final models are included in Appendix B. We saved parameter estimates from each model for use in the Monte Carlo simulation.

We used Monte Carlo simulation to estimate the risk (complement of the Kaplan-Meier survival function) of incident T2DM approximated by the joint distribution of the models above. To do this, we resampled people from our study sample at baseline with replacement 50,000 times (to minimize simulation error), discarding follow-up data. We simulated follow-up data for each of the 50,000 resamples based on the parameters estimated from the models above to estimate the natural course.²⁴ To check the model specification (i.e., parameterization of variables), we compared the observed data (weighted with inverse probability of censoring weights to account for non-ignorable censoring)^31,32 with the simulated follow-up data under the natural course; we compared cumulative incidence curves for T2DM and mortality, and the proportion of the cohort over time with hypertension, and mean variable values over time for BMI, blood glucose, total cholesterol, CD4 cell count, and eGFR.

To obtain estimates of the risk of T2DM under each of the three hypothetical interventions of interest, we first predicted the natural value of BMI in month $j$ (that is, the value of BMI that would have been observed if the intervention were discontinued right before $j$) and then, if the BMI in the natural course exceeded the threshold set by the intervention scenario and $j\le 24$, we set the BMI value to the threshold.²⁵ The thresholds for each of the interventions were: 1) 29.9 kg/m²; 2) baseline BMI × 1.05; and 3) baseline BMI. BMI was allowed to take the value predicted under the natural course in all months $j=25$ to $84$ following ART initiation.

We estimated the standard error of estimates by calculating the standard deviation from 200 unrestricted random bootstrap resamples.^33,34 Models described above were fit and Monte Carlo simulation was run within each bootstrap resample. All analyses were carried out using SAS 9.4 (SAS Institute Inc., Cary, NC).

Sensitivity analyses

We estimated effects as above within strata defined by baseline ART regimen and race. We estimated risks of T2DM accounting for death as a competing event.^35,36

Role of the funding source

Funding for this study came from the National Institutes of Health. The funders had no role in the design, analysis, or interpretation of this study.

RESULTS

The study sample included 13,822 PWH with a BMI between 18.5 and 29.9 kg/m² when they initiated ART between 2012–2021. The median age at ART initiation was 39 years (interquartile range [IQR]: 29, 51). The majority were male (90%) and 71% had MSM as an HIV acquisition risk. A third (35%) were White, 42% were Black, and 14% were Hispanic (Table 1).

Table 1.

Characteristics of 13,822 NA-ACCORD participants with a normal (18.5 to 24.9 kg/m²) or overweight (25.0 to 29.9 kg/m²) BMI and no history of diagnosed type 2 diabetes mellitus when they were observed to initiate ART, 2012–2021

	N=13,822
Age, yearsa	39 (29, 51)
Male sex at birthb	12,494 (90)
Race/ethnicity
Non-Hispanic White	4,845 (35)
Non-Hispanic Black	5,830 (42)
Hispanic	1,933 (14)
Non-Hispanic, non-White/non-Black	1,214 (9)
HIV acquisition risk factorc
MSM	6,176 (71)
IDU	638 (7)
Heterosexual sex	2,636 (30)
Missing	5,082
Year of ART initiationa	2015 (2013, 2017)
First ART regimen classd
INSTI	8230 (60)
PI	1984 (14)
NNRTI	3608 (26)
HCV infection	1,497 (11)
Baseline anxiety diagnosis	2,220 (16)
Baseline depression diagnosis	2,745 (20)
Laboratory values
Blood glucose, mg/dLa	90 (82, 98)
Missing	1482
Total cholesterol, mg/dLa	160 (137, 187)
Missing	3940
CD4 cell count, cells/mm3, a	382 (208, 574)
Missing	743
eGFR, mL/min/1.73m2, a	103 (85, 117)
Missing	588
Prior hypertension diagnosis	2,286 (17)
Smoking
Ever	7376 (53)
Never	3605 (26)
Not collected	2841 (21)
BMI, kg/m2, a	24.3 (22.1, 26.6)
Normal (18.5 to 24.9 kg/m2)	7,856 (57)
Overweight (25.0 to 29.9 kg/m2)	5,966 (43)

Abbreviations: ART, antiretroviral treatment; BMI, body mass index; eGFR, estimated glomerular filtration rate; HCV, hepatitis C virus; HIV, human immunodeficiency virus; IDU, injection drug use; INSTI, integrase strand transfer inhibitor; MSM, men who have sex with men; NA-ACCORD, North American-AIDS Cohort Collaboration on Research and Design; NNRTI, non-nucleoside reverse transcriptase inhibitor; PI, protease inhibitor

^aMedian (IQR)

^bN (%)

^cNot mutually exclusive

^dMutually exclusive, where INSTI group included 383 people on a dual PI/INSTI, 112 on a dual NNRTI/INSTI, and 74 on a PI/INSTI/NNRTI regimen; PI group included 58 people on a dual PI/NNRTI

Under the natural course, the mean BMI was predicted to rise from just under 24.5 kg/m² at ART initiation to about 26.5 kg/m² around 7 years after ART initiation (Figure 1); 18% of people would have had a BMI >29.9 in the first 2 years of follow-up (and been eligible for the first hypothetical intervention); 73% of people would have gained >5% of their baseline BMI at some point in the first 2 years of follow-up (and been eligible for the second hypothetical intervention); and 95% of people had a predicted increase in BMI in the first 2 years of follow-up (and been eligible for the third hypothetical intervention). Under the natural course, the 7-year risk of T2DM was 5.6% (95% confidence interval [CI]: 4.8%, 6.3%) (Table 2). The predicted 7-year risk of death was 6.1% (95% CI: 5.4%, 6.8%). The models provided a good fit to the data (Supplemental Figure 2).

Figure 1.

Predicted mean body mass index (BMI, kg/m²) over the 7 years after ART initiation under the natural course and interventions on BMI over 2 years after ART initiation: BMI capped at 29.9; BMI increase capped at 5%; no BMI increases

Table 2.

Estimates of the effects^a of hypothetical interventions on weight change in the first 2 years after ART initiation on 7-year risk of type 2 diabetes mellitus estimated among 13,822 NA-ACCORD participants with a normal (18.5 to 24.9 kg/m²) or overweight (25.0 to 29.9 kg/m²) BMI and no prior diagnosis of type 2 diabetes mellitus at ART initiation, 2012–2021

	Risk, % (95% CI)	RD (95% CI)	RR (95% CI)
Natural course, no intervention on BMI	5.6 (4.8, 6.3)	00. 00	1. 00
BMI capped at 29.9	5.2 (4.5, 5.9)	−0.3 (−0.8, 0.2)	0.94 (0.86, 1.03)
BMI increase capped at 5%	4.9 (4.2, 5.6)	−0.6 (−1.2, −0.1)	0.88 (0.80, 0.97)
No BMI increases	4.5 (3.8, 5.2)	−1.0 (−1.7, −0.4)	0.81 (0.72, 0.92)

Abbreviations: ART, antiretroviral treatment; BMI, body mass index; CI, confidence interval; eGFR, estimated glomerular filtration rate; HCV, hepatitis C virus; HIV, human immunodeficiency virus; IDU, injection drug use; MSM, men who have sex with men; NA-ACCORD, North American-AIDS Combined Cohort On Research and Design; RD, risk difference; RR, risk ratio

^aAdjusted for age at ART initiation, male sex at birth, black race, HIV acquisition risk factors (MSM, IDU, heterosexual sex), HCV infection, history of diagnosed anxiety or depression at ART initiation, and time-updated blood glucose, total cholesterol, CD4 cell count, eGFR, and hypertension

Under a hypothetical intervention to cap BMI at 29.9 kg/m² for the first two years after ART initiation, the projected mean BMI followed a similar trend as under the natural course but remained slightly lower through 7 years (Figure 1). Despite the modest change in the population mean BMI, preventing a transition to obesity was estimated to lead to a small decrease in T2DM risk: risk difference (RD) = −0.6% (95% confidence interval [CI]: −1.2%, −0.1%) and risk ratio (RR) = 0.88 (95% CI: 0.80, 0.97) (Table 2). Under a hypothetical intervention in which BMI increases were capped at 5%, population average BMI was effectively flat for the first 2 years. Capping BMI increases at 5% did not appreciably change T2DM risk at 7 years after ART initiation relative to the natural course: RD = −0.3% (95% CI: −0.8%, 0.2%) and RR = 0.94 (95% CI: 0.86, 1.03). Under a hypothetical intervention in which BMI was not allowed to increase at all over the first 2 years on ART, projected mean BMI sharply decreased over the first 2 years on ART by a little more than 1 kg/m² then rose to around 1 kg/m² higher than baseline by 7 years—about 1.5–2 kg/m² lower than the mean BMI under the natural course. Under this most restrictive hypothetical intervention, the risk of T2DM at 7 years after ART initiation was 19% lower (RR = 0.81 (95% CI: 0.72, 0.92); RD = −1.0% (95% CI: −1.7%, −0.4%)). The predicted risk of T2DM over the 7 years of follow-up under the natural course and each hypothetical intervention is plotted in Figure 2.

Figure 2.

Cumulative incidence of type 2 diabetes mellitus under the natural course and hypothetical interventions on BMI over 2 years after ART initiation: BMI increase capped at 5%; BMI capped at 29.9; no BMI increases

The predicted evolution of time-varying covariates under the natural course and each hypothetical intervention are shown in Supplemental Figure 3. The associations between hypothetical interventions on BMI after ART initiation and other covariates are not interpretable as effects unless the causal structure linking changes to BMI and each of the covariates matches exactly the causal structure linking changes to BMI and T2DM (e.g., assuming no unmeasured confounders of BMI and changes in total cholesterol). Therefore, any inference from these figures should be interpreted as exploratory. However, interventions restricting BMI increases in the first 2 years after ART initiation appeared to lower mean glucose, slow increases in total cholesterol, and reduce the proportion of people who were diagnosed with incident hypertension over 7 years. The most restrictive hypothetical intervention to prevent any increase in BMI over the first 2 years after ART initiation was associated with an increase in mortality risk.

Stratified analyses and analyses in which death was treated as a competing event all produced results that were consistent with the primary results (Supplemental Table 1).

DISCUSSION

We estimated the potential population-level effects of a hypothetical weight maintenance intervention applied over the first 2 years after ART initiation on the 7-year incidence of T2DM would be modest (a 6%−19% relative reduction in risk). Given concerns over ART-related weight gain for PWH,³⁷ balanced against the benefits of ART for PWH, this analysis provides potential estimates of how much an intervention to cap or prevent weight gain in the 2 years after ART initiation^4,38,39 might reduce risk of T2DM. Ideally, it serves to prompt the development of specific interventions on weight management for PWH.

Some mitigation of weight gain among PWH will be needed as PWH live longer and remain on ART regimens with metabolic effects.^4,8,9,16 In prior studies, approximately a fifth of PWH who initiated ART with a normal BMI became overweight after 3 years on ART and a fifth of PWH who initiated ART with a BMI in the overweight range developed obesity after 3 years on ART (these proportions were estimated prior to the widespread use of INSTIs and TAF, which are thought to be most weight-increasing).^1,37 In our data, about 18% of PWH with normal or overweight BMI at ART initiation were classified as having obesity by 2 years on ART, and about 40% were classified as having obesity by 7 years on ART. In the general population, adults’ BMI changes by about +1% per year.

A meta-analysis of 34 behavioral weight-management interventions found that, on average, over 2 years, the interventions were associated with weight change of −1.8 kg (about 4 pounds),⁴⁰ which, for a man of average height (5’9”) and 180 pounds (BMI = 26.6 kg/m²) would equate to a 2% reduction in BMI. We did not test specific weight-management interventions in this study and the specific intervention that is adopted may have different effects (weaker or stronger) than those predicted here. Indeed, a prior trial of a weight management intervention in the general population found larger absolute effects on remission of T2DM than it did on weight loss.⁴¹ The many strategies or lifestyle factors (e.g., diet, exercise, medications) through which people in our data lost or maintained weight do not map to a well-defined intervention, and our results represent the average of the effects of different sets of actions that brought about the observed changes in BMI.⁴² However, our results showing that 6–19% of incident T2DM cases over 7 years could be prevented by intervening on weight gain over just 2 years, might further motivate the development of such interventions tailored to PWH initiating ART. For context, in the Diabetes Prevention Program trial, after 10 years, an intensive lifestyle intervention reduced diabetes incidence by 34% and metformin reduced diabetes incidence by 18%.⁴³

We also did not estimate the effect of weight loss among PWH who were overweight or obese at ART initiation because weight loss that we might have observed in our study sample was more likely to represent unintentional disease-related weight loss than intentional weight management.⁴⁴ Few weight loss or weight maintenance interventions have been tested for PWH, who may have unique clinical considerations related to their medical history (e.g., history of lipoatrophy and chronic inflammation) and who may struggle with traditional lifestyle advice due to higher prevalence of adverse social determinants of health (e.g., access to healthy foods or safe places to exercise).^45,46 However, options for PWH with obesity are emerging in the form of new weight loss pharmaceuticals (GLP1s and SGLT2s),^47–49 although their use was not widespread during our study period and particularly not for weight management. The effects of these weight loss pharmaceuticals and their interaction with ART (current and historical) in people with HIV should be estimated explicitly.

While we estimated that a hypothetical intervention to completely restrict weight gain for the first 2 years of follow-up was associated with the largest reduction in T2DM risk, it was also associated with an increased risk of death. The observation that preventing any increase in BMI over the first 2 years was associated with an increase in mortality is an indication that the goal should not be no BMI increase for anyone: people may gain weight as a result of a return to health phenomenon, and weight loss may be healthy (e.g., as a result of improved diet or exercise) or unhealthy (e.g., due to illness such as cancer). This finding may also signal that the means by which people lost weight in our data do not map well to an intervention that would be considered for clinical implementation (i.e., that weight loss was more likely to result from an undesirable “intervention” on health, such as illness onset). Finally, it may signal that there were additional unmeasured confounders (maybe for the effect of BMI on T2DM, but definitely for the effect of BMI on death) that were not accounted for in our study.

We modeled the effects of hypothetical interventions on weight using the parametric extended g-formula,²⁵ which does require assumptions in order to interpret our results causally—although all epidemiologic studies with a causal goal require similar if not more stringent assumptions. We assumed that we have sufficiently measured and modeled time-varying confounders of the effect of weight change in the first two years after ART initiation and risk of incident T2DM. We did not include potential confounders such as steroid use or statin use (because of difficulties with ensuring we could calibrate the models well and concerns about data completeness). One might also be concerned about lack of information on diet and exercise leading to uncontrolled confounding. The e-value (the risk ratio for the association between the unmeasured confounder and the exposure and between the confounder and the outcome that would shift the observed RR to 1.00) to explain away the observed effect of a 5% cap on BMI increase over the first 2 years after ART initiation is 1.53,^50,51 which is well within what is plausible for the effect of either diet or exercise on weight maintenance and T2DM risk. Thus, if one considered diet and exercise confounders, our results could be easily explained by uncontrolled confounding. However, diet and exercise are the primary levers that might be part of a weight maintenance intervention, in which case our failure to account for them would be more of a problem of causal consistency as discussed above.⁵² Given that causal consistency is unlikely to be entirely true, we should interpret our results cautiously, as quantifying what might be possible, rather than what is inevitable. In addition to the standard causal identification assumptions, the g-formula relies on more parametric models being correctly fit than other methods; we assumed we correctly specified all those 8 parametric models described in the Methods.

In conclusion, we have provided evidence that even modest interventions to prevent or reduce weight gain in the first 2 years on ART may modestly reduce the overall burden of T2DM among PWH. Such modest reductions can translate to large numbers of cases of T2DM averted. For example, in our sample of just over 13,000 PWH, capping BMI to 29.9 over just two years was estimated to reduce T2DM risk by 0.6%, equivalent to about 83 cases of T2DM prevented. Given the potential impact of interventions that lead to restrictions in weight gain after ART initiation, it is critical to develop and implement real-world interventions to limit weight gain and obesity onset among PWH initiating ART.

Appendix A. Medications used in determining clinical covariates (see text)

Diabetes-specific medications

ALOGLIPTIN	DPP-4
LINAGLIPTIN	DPP-4
SAXAGLIPTIN + METFORMIN	DPP-4 + biguanide
LINAGLIPTIN + METFORMIN	DPP-4 + biguanide
INSULIN	Insulin
INSULIN + LIRAGLUTIDE	Insulin + Miscellaneous anti-diabetic (incretin mimetic)
INSULIN + LIXISENATIDE	Insulin Combo
NATEGLINIDE	Meglitinide
REPAGLINIDE	Meglitinide
REPAGLINIDE + METFORMIN	Meglitinide + Biguanide
LIRAGLUTIDE	Miscellaneous anti-diabetic
SAXAGLIPTIN	Miscellaneous anti-diabetic
SITAGLIPTIN	Miscellaneous anti-diabetic
PRAMLINTIDE	Miscellaneous anti-diabetic (Amylin analog)
EXENATIDE	Miscellaneous anti-diabetic (incretin mimetic)
SITAGLIPTIN + METFORMIN	Miscellaneous anti-diabetic + Biguanide
CANAGLIFLOZIN	SGLT2
DAPAGLIFLOZIN	SGLT2
EMPAGLIFLOZIN	SGLT2
ERTUGLIFLOZIN	SGLT2
CANAGLIFLOZIN + METFORMIN	SGLT2 + Biguanide
DAPAGLIFLOZIN + METFORMIN	SGLT2 + Biguanide
EMPAGLIFLOZIN + METFORMIN	SGLT2 + Biguanide
ERTUGLIFLOZIN + METFORMIN	SGLT2 + Biguanide
CHLORPROPAMIDE	Sulfonylurea
GLICLAZIDE	Sulfonylurea
GLIMEPIRIDE	Sulfonylurea
GLIPIZIDE	Sulfonylurea
GLIPIZIDE EXTENDED-RELEASE	Sulfonylurea
GLYBURIDE	Sulfonylurea
TOLAZAMIDE	Sulfonylurea
TOLBUTAMIDE	Sulfonylurea
GLIPIZIDE + METFORMIN	Sulfonylurea + Biguanide
GLYBURIDE + METFORMIN	Sulfonylurea + Biguanide
PIOGLITAZONE + GLIMEPIRIDE	Thiazolidinedione + Sulfonylurea
ROSIGLITAZONE + GLIMEPIRIDE	Thiazolidinedione + Sulfonylurea

Diabetes-related medications

ACARBOSE	Alpha-glucosidase inhibitor
MIGLITOL	Alpha-glucosidase inhibitor
METFORMIN	Biguanide
METFORMIN + ROSIGLITAZONE	Biguanide + Thiazolidinedione
ALBIGLUTIDE	GLP1 agonist/incretin memetic
DULAGLUTIDE	GLP1 agonist/incretin memetic
LIXISENATIDE	GLP1 agonist/incretin memetic
SEMAGLUTIDE INJECT	GLP1 agonist/incretin memetic
SEMAGLUTIDE UNSPECIFIED	GLP1 agonist/incretin memetic
PIOGLITAZONE	Thiazolidinedione
ROSIGLITAZONE	Thiazolidinedione
TROGLITAZONE	Thiazolidinedione
PIOGLITAZONE + METFORMIN	Thiazolidinedione + Biguanide
DIABETES MEDICATION UNSPECIFIED	Diabetes

Anti-hypertensive medications

BENAZEPRIL	ACE Inhibitor (ACEI)
CAPTOPRIL	ACE Inhibitor (ACEI)
CILAZAPRIL	ACE Inhibitor (ACEI)
ENALAPRIL INJECT	ACE Inhibitor (ACEI)
ENALAPRIL ORAL	ACE Inhibitor (ACEI)
ENALAPRIL UNSPECIFIED	ACE Inhibitor (ACEI)
ENALAPRILAT INJECT	ACE Inhibitor (ACEI)
FOSINOPRIL	ACE Inhibitor (ACEI)
LISINOPRIL	ACE Inhibitor (ACEI)
MOEXIPRIL	ACE Inhibitor (ACEI)
PERINDOPRIL	ACE Inhibitor (ACEI)
QUINAPRIL	ACE Inhibitor (ACEI)
RAMIPRIL	ACE Inhibitor (ACEI)
TRANDOLAPRIL	ACE Inhibitor (ACEI)
ENALAPRIL + FELODIPINE	ACEI + Calcium channel blocker
TRANDOLAPRIL + VERAPAMIL	ACEI + Calcium channel blocker
BENAZEPRIL + HYDROCHLOROTHIAZIDE	ACEI + Diuretic
CAPTOPRIL + HYDROCHLOROTHIAZIDE	ACEI + Diuretic
CILAZAPRIL + HYDROCHLOROTHIAZIDE	ACEI + Diuretic
ENALAPRIL + HYDROCHLOROTHIAZIDE	ACEI + Diuretic
FOSINOPRIL + HYDROCHLOROTHIAZIDE	ACEI + Diuretic
LISINOPRIL + HYDROCHLOROTHIAZIDE	ACEI + Diuretic
MOEXIPRIL + HYDROCHLOROTHIAZIDE	ACEI + Diuretic
PERINDOPRIL + INDAPAMIDE	ACEI + Diuretic
QUINAPRIL + HYDROCHLOROTHIAZIDE	ACEI + Diuretic
RAMIPRIL + HYDROCHLOROTHIAZIDE	ACEI + Diuretic
DOXAZOSIN	Alpha blocker
PRAZOSIN	Alpha blocker
TERAZOSIN	Alpha blocker
PRAZOSIN + POLYTHIAZIDE	Alpha blocker + Diuretic
AZILSARTAN	Angiotensin receptor blocker (ARB)
CANDESARTAN	Angiotensin receptor blocker (ARB)
EPROSARTAN	Angiotensin receptor blocker (ARB)
IRBESARTAN	Angiotensin receptor blocker (ARB)
LOSARTAN	Angiotensin receptor blocker (ARB)
OLMESARTAN	Angiotensin receptor blocker (ARB)
TELMISARTAN	Angiotensin receptor blocker (ARB)
VALSARTAN	Angiotensin receptor blocker (ARB)
VALSARTAN + SACUBITRIL	Angiotensin receptor blocker (ARB)
CLONIDINE	Antiadrenergic
GUANABENZ	Antiadrenergic
GUANADREL	Antiadrenergic
GUANFACINE	Antiadrenergic
METHYLDOPA	Antiadrenergic
PHENOXYBENZAMINE	Antiadrenergic
CLONIDINE + CHLORTHALIDONE	Antiadrenergic + Diuretic
METHYLDOPA + CHLOROTHIAZIDE	Antiadrenergic + Diuretic
METHYLDOPA + HYDROCHLOROTHIAZIDE	Antiadrenergic + Diuretic
CANDESARTAN + HYDROCHLOROTHIAZIDE	ARB + Diuretic
EPROSARTAN + HYDROCHLOROTHIAZIDE	ARB + Diuretic
IRBESARTAN + HYDROCHLOROTHIAZIDE	ARB + Diuretic
LOSARTAN + HYDROCHLOROTHIAZIDE	ARB + Diuretic
OLMESARTAN + HYDROCHLOROTHIAZIDE	ARB + Diuretic
TELMISARTAN + HYDROCHLOROTHIAZIDE	ARB + Diuretic
VALSARTAN + HYDROCHLOROTHIAZIDE	ARB + Diuretic
ACEBUTOLOL	Beta blocker
ATENOLOL	Beta blocker
BETAXOLOL	Beta blocker
BISOPROLOL	Beta blocker
CARTEOLOL	Beta blocker
CARVEDILOL	Beta blocker
ESMOLOL	Beta blocker
LABETALOL	Beta blocker
METOPROLOL	Beta blocker
NADOLOL	Beta blocker
NEBIVOLOL	Beta blocker
OXPRENOLOL	Beta blocker
PENBUTOLOL	Beta blocker
PINDOLOL	Beta blocker
PROPRANOLOL	Beta blocker
TIMOLOL	Beta blocker
ATENOLOL + CHLORTHALIDONE	Beta blocker + Diuretic
BISOPROLOL + HYDROCHLOROTHIAZIDE	Beta blocker + Diuretic
METOPROLOL + HYDROCHLOROTHIAZIDE	Beta blocker + Diuretic
NADOLOL + BENDROFLUMETHIAZIDE	Beta blocker + Diuretic
PROPRANOLOL + HYDROCHLOROTHIAZIDE	Beta blocker + Diuretic
TIMOLOL + HYDROCHLOROTHIAZIDE	Beta blocker + Diuretic
AMLODIPINE	Calcium channel blocker
BEPRIDIL	Calcium channel blocker
CLEVIDIPINE	Calcium channel blocker
DILTIAZEM	Calcium channel blocker
FELODIPINE	Calcium channel blocker
ISRADIPINE	Calcium channel blocker
NICARDIPINE	Calcium channel blocker
NIFEDIPINE	Calcium channel blocker
NIMODIPINE	Calcium channel blocker
NISOLDIPINE	Calcium channel blocker
VERAPAMIL	Calcium channel blocker
AMLODIPINE + BENAZEPRIL	Calcium channel blocker + ACEI
AMLODIPINE + PERINDOPRIL	Calcium channel blocker + ACEI
AMLODIPINE + OLMESARTAN	Calcium channel blocker + ARB
AMLODIPINE + TELMISARTAN	Calcium channel blocker + ARB
AMLODIPINE + VALSARTAN	Calcium channel blocker + ARB
AMLODIPINE + OLMESARTAN + HYDROCHLOROTHIAZIDE	Calcium channel blocker + ARB + Diuretic
AMLODIPINE + VALSARTAN + HYDROCHLOROTHIAZIDE	Calcium channel blocker + ARB + Diuretic
ACETAZOLAMIDE	Diuretic
AMILORIDE	Diuretic
AMILORIDE + HYDROCHLOROTHIAZIDE	Diuretic
BENDROFLUMETHIAZIDE	Diuretic
BENDROFLUMETHIAZIDE + RAUWOLFIA	Diuretic
BUMETANIDE	Diuretic
CHLOROTHIAZIDE	Diuretic
CHLOROTHIAZIDE + RESERPINE	Diuretic
CHLORTHALIDONE	Diuretic
EPLERENONE	Diuretic
ETHACRYNIC ACID	Diuretic
FUROSEMIDE	Diuretic
HYDROCHLOROTHIAZIDE	Diuretic
HYDROFLUMETHIAZIDE	Diuretic
INDAPAMIDE	Diuretic
METHYCLOTHIAZIDE	Diuretic
METHYCLOTHIAZIDE + RESERPINE	Diuretic
METOLAZONE	Diuretic
POLYTHIAZIDE	Diuretic
QUINETHAZONE	Diuretic
SPIRONOLACTONE	Diuretic
SPIRONOLACTONE + HYDROCHLOROTHIAZIDE	Diuretic
THIAZIDE	Diuretic
TORSEMIDE	Diuretic
TRIAMTERENE	Diuretic
TRIAMTERENE + HYDROCHLOROTHIAZIDE	Diuretic
TRICHLORMETHIAZIDE	Diuretic
HYDRALAZINE INJECT	Peripheral vasodilator
HYDRALAZINE ORAL	Peripheral vasodilator
HYDRALAZINE UNSPECIFIED	Peripheral vasodilator
HYDRALAZINE + ISOSORBIDE DINITRATE	Peripheral vasodilator + Anti-anginal
HYDRALAZINE + HYDROCHLOROTHIAZIDE	Peripheral vasodilator + Diuretic
HYDRALAZINE + HYDROCHLOROTHIAZIDE + RESERPINE	Peripheral vasodilator + Diuretic
HYDRALAZINE INJECT	Peripheral vasodilator
HYDRALAZINE UNSPECIFIED	Peripheral vasodilator
ALISKIREN	Renin inhibitor
ALISKIREN + VALSARTAN	Renin inhibitor + ARB
ALISKIREN + HYDROCHLOROTHIAZIDE	Renin inhibitor + Diuretic
MINOXIDIL	Vasodilator
HYPERTENSION MEDICATION UNSPECIFIED	Anti-hypertensives unspecified

Appendix B. Specification of covariate functional form in all models used to fit the g-formula

1. $E\left[{BMI}_j\right]={\beta }_{{BMI}_0}+{\beta }_{{BMI}_1}\times {in}_j+{\beta }_{{BMI}_2}\times {in1}_j+{\beta }_{{BMI}_3}\times {in2}_j+{\beta }_{{BMI}_4}\times {in3}_j+{\beta }_{{BMI}_5}\times \mathit{\text{BMIARTcat}}+{\beta }_{BM{I}_6}\times \mathit{\text{AgeARTcat}}2+{\beta }_{BM{I}_7}\times \mathit{\text{AgeARTcat}}3+{\beta }_{BM{I}_8}\times \mathit{\text{AgeARTcat}}4+{\beta }_{BM{I}_9}\times \mathit{\text{female}}+{\beta }_{BM{I}_{10}}\times \mathit{\text{White}}+{\beta }_{BM{I}_{11}}\times \mathit{\text{Hispanic}}+{\beta }_{BM{I}_{12}}\times \mathit{\text{OtherRace}}+{\beta }_{BM{I}_{13}}\times \mathit{\text{MSMrisk}}+{\beta }_{BM{I}_{14}}\times \mathit{\text{IDUrisk}}+{\beta }_{BM{I}_{15}}\times HCV+{\beta }_{BM{I}_{16}}\times \mathit{\text{SmokeNever}}+{\beta }_{BM{I}_{17}}\times \mathit{\text{AnxDep}}+{\beta }_{{BMI}_{19}}\times {BMI}_{j-1}+{\beta }_{{BMI}_{20}}\times {\mathit{\text{eGFRcat}}1}_{j-1}+{\beta }_{{BMI}_{21}}\times {\mathit{\text{eGFRcat}}2}_{j-1}+{\beta }_{{BMI}_{22}}\times {CD4}_{j-1}+{\beta }_{{BMI}_{23}}\times {HTN}_{j-1}+{\beta }_{{BMI}_{24}}\times {\mathit{\text{TotCholcat}}1}_{j-1}+{\beta }_{{BMI}_{25}}\times {\mathit{\text{TotCholcat}}2}_{j-1}+{\beta }_{{BMI}_{26}}\times {\mathit{\text{Glucosecat}}1}_{j-1}+{\beta }_{{BMI}_{27}}\times {\mathit{\text{Glucosecat}}2}_{j-1}+\epsilon$;

2. $E\left[{\mathit{\text{Glucose}}}_j\right]={\beta }_{{\mathit{\text{Glucose}}}_0}+{\beta }_{{\mathit{\text{Glucose}}}_1}\times {in}_j+{\beta }_{{\mathit{\text{Glucose}}}_2}\times {in1}_j+{\beta }_{{\mathit{\text{Glucose}}}_3}\times {in2}_j+{\beta }_{{\mathit{\text{Glucose}}}_4}\times {in3}_j+{\beta }_{{\mathit{\text{Glucose}}}_5}\times {in4}_j+{\beta }_{{\mathit{\text{Glucose}}}_6}\times \mathit{\text{BMIARTcat}}1+{\beta }_{{\mathit{\text{Glucose}}}_7}\times \mathit{\text{BMIARTcat}}2+{\beta }_{{\mathit{\text{Glucose}}}_8}\times \mathit{\text{AgeART}}2cat2+{\beta }_{{\mathit{\text{Glucose}}}_9}\times \mathit{\text{female}}+{\beta }_{{\mathit{\text{Glucose}}}_{10}}\times \mathit{\text{White}}+{\beta }_{{\mathit{\text{Glucose}}}_{11}}\times \mathit{\text{Hispanic}}+{\beta }_{{\mathit{\text{Glucose}}}_{12}}\times \mathit{\text{OtherRace}}+{\beta }_{{\mathit{\text{Glucose}}}_{13}}\times \mathit{\text{MSMrisk}}+{\beta }_{{\mathit{\text{Glucose}}}_{14}}\times \mathit{\text{IDUrisk}}+{\beta }_{{\mathit{\text{Glucose}}}_{15}}\times HCV+{\beta }_{{\mathit{\text{Glucose}}}_{16}}\times \mathit{\text{SmokeNeve}}r+{\beta }_{{\mathit{\text{Glucose}}}_{17}}\times \mathit{\text{AnxDep}}+{\beta }_{{\mathit{\text{Glucose}}}_{18}}\times {Age}_j+{\beta }_{{\mathit{\text{Glucose}}}_{19}}\times {\mathit{\text{BMIcat}}1}_j+{\beta }_{{\mathit{\text{Glucose}}}_{20}}\times {\mathit{\text{BMIcat}}2}_j+{\beta }_{{\mathit{\text{Glucose}}}_{21}}\times {\mathit{\text{BMIcat}}3}_j+{\beta }_{{\mathit{\text{Glucose}}}_{22}}\times {\mathit{\text{eGFRcat}}1}_{j-1}+{\beta }_{{\mathit{\text{Glucose}}}_{23}}\times {\mathit{\text{eGFRcat}}2}_{j-1}+{\beta }_{{\mathit{\text{Glucose}}}_{24}}\times {CD4}_{j-1}+{\beta }_{{\mathit{\text{Glucose}}}_{25}}\times {HTN}_{j-1}+{\beta }_{{\mathit{\text{Glucose}}}_{26}}\times {\mathit{\text{TotChol}}}_{j-1}+{\beta }_{{\mathit{\text{Glucose}}}_{27}}\times \mathit{\text{GlucoseART}}+{\beta }_{{\mathit{\text{Glucose}}}_{28}}\times {\mathit{\text{Glucose}}}_{j-1}+\epsilon$;

3. $E\left[{\mathit{\text{TotChol}}}_j\right]={\beta }_{{\mathit{\text{TotChol}}}_0}+{\beta }_{{\mathit{\text{TotChol}}}_1}\times {in}_j+{\beta }_{{\mathit{\text{TotChol}}}_2}\times {in1}_j+{\beta }_{{\mathit{\text{TotChol}}}_3}\times {in2}_j+{\beta }_{{\mathit{\text{TotChol}}}_4}\times {in3}_j+{\beta }_{{\mathit{\text{TotChol}}}_5}\times {in4}_j+{\beta }_{{\mathit{\text{TotChol}}}_6}\times \mathit{\text{female}}+{\beta }_{{\mathit{\text{TotChol}}}_7}\times \mathit{\text{White}}+{\beta }_{{\mathit{\text{TotChol}}}_8}\times \mathit{\text{Hispanic}}+{\beta }_{{\mathit{\text{TotChol}}}_9}\times \mathit{\text{OtherRace}}+{\beta }_{{\mathit{\text{TotChol}}}_{10}}\times \mathit{\text{MSMrisk}}+{\beta }_{{\mathit{\text{TotChol}}}_{11}}\times \mathit{\text{IDUrisk}}+{\beta }_{{\mathit{\text{TotChol}}}_{12}}\times HCV+{\beta }_{{\mathit{\text{TotChol}}}_{13}}\times \mathit{\text{SmokeNeve}}r+{\beta }_{{\mathit{\text{TotChol}}}_{14}}\times \mathit{\text{AnxDep}}+{\beta }_{{\mathit{\text{TotChol}}}_{15}}\times {\mathit{\text{BMIcat}}1}_j+{\beta }_{{\mathit{\text{TotChol}}}_{16}}\times {\mathit{\text{BMIcat}}2}_j+{\beta }_{{\mathit{\text{TotChol}}}_{17}}\times {\mathit{\text{BMIcat}}3}_j+{\beta }_{{\mathit{\text{TotChol}}}_{18}}\times {\mathit{\text{eGFRcat}}1}_{j-1}+{\beta }_{{\mathit{\text{TotChol}}}_{19}}\times {\mathit{\text{eGFRcat}}2}_{j-1}+{\beta }_{{\mathit{\text{TotChol}}}_{20}}\times {CD4}_{j-1}+{\beta }_{{\mathit{\text{TotChol}}}_{21}}\times {HTN}_{j-1}+{\beta }_{{\mathit{\text{TotChol}}}_{22}}\times \mathit{\text{TotCholART}}+{\beta }_{{\mathit{\text{TotChol}}}_{23}}\times {\mathit{\text{TotChol}}}_{j-1}+{\beta }_{{\mathit{\text{TotChol}}}_{24}}\times {\mathit{\text{Glucose}}}_j+\epsilon$;

4. $\mathit{\text{logit}}\left[{HTN}_j=1|{HTN}_{j-1}=0\right]={\beta }_{{HTN}_0}+{\beta }_{{HTN}_1}\times {in}_j+{\beta }_{{HTN}_2}\times {in1}_j+{\beta }_{{HTN}_3}\times {in2}_j+{\beta }_{{HTN}_4}\times {in3}_j+{\beta }_{{HTN}_5}\times {in4}_j+{\beta }_{{HTN}_6}\times \mathit{\text{female}}+{\beta }_{{HTN}_7}\times \mathit{\text{White}}+{\beta }_{{HTN}_8}\times \mathit{\text{Hispanic}}+{\beta }_{{HTN}_9}\times \mathit{\text{OtherRace}}+{\beta }_{{HTN}_{10}}\times \mathit{\text{MSMrisk}}+{\beta }_{{HTN}_{11}}\times \mathit{\text{IDUrisk}}+{\beta }_{{HTN}_{12}}\times HCV+{\beta }_{{HTN}_{13}}\times \mathit{\text{SmokeNeve}}r+{\beta }_{{HTN}_{14}}\times \mathit{\text{AnxDep}}+{\beta }_{{HTN}_{15}}\times \mathit{\text{AgeARTcat}}2+{\beta }_{{HTN}_{16}}\times \mathit{\text{AgeARTcat}}3+{\beta }_{{HTN}_{17}}\times \mathit{\text{AgeARTcat}}4+{\beta }_{{HTN}_{18}}\times {Age}_j+{\beta }_{{HTN}_{19}}\times {\mathit{\text{BMIcat}}1}_j+{\beta }_{{HTN}_{20}}\times {\mathit{\text{BMIcat}}2}_j+{\beta }_{{HTN}_{21}}\times {\mathit{\text{BMIcat}}3}_j+{\beta }_{{HTN}_{22}}\times {\mathit{\text{eGFRcat}}1}_{j-1}+{\beta }_{{HTN}_{23}}\times {\mathit{\text{eGFRcat}}2}_{j-1}+{\beta }_{{HTN}_{24}}\times {CD4}_{j-1}+{\beta }_{{HTN}_{25}}\times {\mathit{\text{TotChol}}}_j+{\beta }_{{HTN}_{26}}\times {\mathit{\text{Glucose}}}_j+\epsilon$;

5. E$\left[{CD4}_j\right]={\beta }_{{CD4}_0}+{\beta }_{{CD4}_1}\times {in}_j+{\beta }_{{CD4}_2}\times {in1}_j+{\beta }_{{CD4}_3}\times {in2}_j+{\beta }_{{CD4}_4}\times {in3}_j+{\beta }_{{CD4}_5}\times {in4}_j+{\beta }_{{CD4}_6}\times \mathit{\text{female}}+{\beta }_{{CD4}_7}\times \mathit{\text{White}}+{\beta }_{{CD4}_8}\times \mathit{\text{MSMrisk}}+{\beta }_{{CD4}_9}\times \mathit{\text{IDUrisk}}+{\beta }_{{CD4}_{10}}\times HCV+{\beta }_{{CD4}_{11}}\times \mathit{\text{SmokeNeve}}r+{\beta }_{{CD4}_{12}}\times \mathit{\text{AnxDep}}+{\beta }_{{CD4}_{13}}\times \mathit{\text{AgeARTcat}}2+{\beta }_{{CD4}_{14}}\times \mathit{\text{AgeARTcat}}3+{\beta }_{{CD4}_{15}}\times \mathit{\text{AgeARTcat}}4+{\beta }_{{CD4}_{16}}\times CD4ART+{\beta }_{{CD4}_{17}}\times {\mathit{\text{BMIcatC}}1}_j+{\beta }_{{CD4}_{18}}\times {\mathit{\text{BMIcatC}}2}_j+{\beta }_{{CD4}_{19}}\times {CD4}_{j-1}+{\beta }_{{CD4}_{20}}\times CD4\mathit{\text{ARTtime}}+{\beta }_{{CD4}_{21}}\times {HTN}_j+{\beta }_{{CD4}_{22}}\times {\mathit{\text{TotCholcat}}1}_{j-1}+{\beta }_{{CD4}_{23}}\times {\mathit{\text{TotCholcat}}2}_{j-1}+\epsilon$;

6. $E\left[{\mathit{\text{eGFR}}}_j\right]={\beta }_{{\mathit{\text{eGFR}}}_0}+{\beta }_{{\mathit{\text{eGFR}}}_1}\times {in}_j+{\beta }_{{\mathit{\text{eGFR}}}_2}\times {in1}_j+{\beta }_{{\mathit{\text{eGFR}}}_3}\times {in2}_j+{\beta }_{{\mathit{\text{eGFR}}}_4}\times {in3}_j+{\beta }_{{\mathit{\text{eGFR}}}_5}\times {in4}_j+{\beta }_{{\mathit{\text{eGFR}}}_6}\times \mathit{\text{female}}+{\beta }_{{\mathit{\text{eGFR}}}_7}\times \mathit{\text{White}}+{\beta }_{{\mathit{\text{eGFR}}}_8}\times \mathit{\text{Hispanic}}+{\beta }_{{\mathit{\text{eGFR}}}_9}\times \mathit{\text{OtherRace}}+{\beta }_{{\mathit{\text{eGFR}}}_{10}}\times \mathit{\text{MSMrisk}}+{\beta }_{{\mathit{\text{eGFR}}}_{11}}\times \mathit{\text{IDUrisk}}+{\beta }_{{\mathit{\text{eGFR}}}_{12}}\times HCV+{\beta }_{{\mathit{\text{eGFR}}}_{13}}\times \mathit{\text{SmokeNeve}}r+{\beta }_{{\mathit{\text{eGFR}}}_{14}}\times \mathit{\text{AnxDep}}+{\beta }_{{\mathit{\text{eGFR}}}_{15}}\times \mathit{\text{AgeARTcat}}2+{\beta }_{{\mathit{\text{eGFR}}}_{16}}\times \mathit{\text{AgeARTcat}}3+{\beta }_{{\mathit{\text{eGFR}}}_{17}}\times \mathit{\text{AgeARTcat}}4+{\beta }_{{\mathit{\text{eGFR}}}_{18}}\times {\mathit{\text{eGFRART}}+{\beta }_{{\mathit{\text{eGFR}}}_{19}}\times \mathit{\text{BMIcat}}1}_j+{\beta }_{{\mathit{\text{eGFR}}}_{20}}\times {\mathit{\text{BMIcat}}2}_j+{\beta }_{{\mathit{\text{eGFR}}}_{21}}\times {\mathit{\text{BMIcat}}3}_j+{\beta }_{{\mathit{\text{eGFR}}}_{22}}\times {\mathit{\text{eGFRcat}}1}_{j-1}+{\beta }_{{\mathit{\text{eGFR}}}_{23}}\times {\mathit{\text{eGFRcat}}2}_{j-1}+{\beta }_{{\mathit{\text{eGFR}}}_{24}}\times {CD4}_j+{\beta }_{{\mathit{\text{eGFR}}}_{25}}\times {HTN}_j+{\beta }_{{\mathit{\text{eGFR}}}_{26}}\times {\mathit{\text{TotCholcat}}1}_j+{\beta }_{{\mathit{\text{eGFR}}}_{27}}\times {\mathit{\text{TotCholcat}}2}_j+{\beta }_{{\mathit{\text{eGFR}}}_{28}}\times {\mathit{\text{Glucose}}}_j+\epsilon$;

7. $\mathit{\text{logit}}\left[{T2DM}_j=1|{T2DM}_{j-1}=0\right]={\beta }_{{T2DM}_0}+{\beta }_{{T2DM}_1}\times {in}_j+{\beta }_{{T2DM}_2}\times {in1}_j+{\beta }_{{T2DM}_3}\times {in2}_j+{\beta }_{{T2DM}_4}\times {in3}_j+{\beta }_{{T2DM}_5}\times {in4}_j+{\beta }_{{T2DM}_6}\times \mathit{\text{female}}+{\beta }_{{T2DM}_7}\times \mathit{\text{White}}+{\beta }_{{T2DM}_8}\times \mathit{\text{Hispanic}}+{\beta }_{{T2DM}_9}\times \mathit{\text{OtherRace}}+{\beta }_{{T2DM}_{10}}\times \mathit{\text{MSMrisk}}+{\beta }_{{T2DM}_{11}}\times \mathit{\text{IDUrisk}}+{\beta }_{{T2DM}_{12}}\times \mathit{\text{SmokeNeve}}r+{\beta }_{{T2DM}_{13}}\times \mathit{\text{AnxDep}}+{\beta }_{{T2DM}_{14}}\times \mathit{\text{AgeARTcat}}2+{\beta }_{{T2DM}_{15}}\times \mathit{\text{AgeARTcat}}3+{\beta }_{{T2DM}_{16}}\times \mathit{\text{AgeARTcat}}4+{\beta }_{{T2DM}_{17}}\times {\mathit{\text{BMIcat}}1}_j+{\beta }_{{T2DM}_{18}}\times {\mathit{\text{BMIcat}}2}_j+{\beta }_{{T2DM}_{19}}\times {\mathit{\text{BMIcat}}3}_j+{\beta }_{{T2DM}_{20}}\times {\mathit{\text{eGFRcat}}1}_j+{\beta }_{{T2DM}_{21}}\times {\mathit{\text{eGFRcat}}2}_j+{\beta }_{{T2DM}_{22}}\times {CD4}_j+{\beta }_{{T2DM}_{22}}\times {HTN}_j+{\beta }_{{T2DM}_{23}}\times {\mathit{\text{TotCholcat}}1}_j+{\beta }_{{T2DM}_{24}}\times {\mathit{\text{TotCholcat}}2}_j+{\beta }_{{T2DM}_{25}}\times \mathit{\text{TotChol}}+{\beta }_{{T2DM}_{26}}\times {\mathit{\text{Glucose}}}_j+\epsilon$;

8. $\mathit{\text{logit}}\left[{\mathit{\text{Death}}}_j=1|{\mathit{\text{Death}}}_{j-1}=0\right]={\beta }_{{\mathit{\text{Death}}}_0}+{\beta }_{{\mathit{\text{Death}}}_1}\times {in}_j+{\beta }_{{\mathit{\text{Death}}}_2}\times {in1}_j+{\beta }_{{\mathit{\text{Death}}}_3}\times {in2}_j+{\beta }_{{\mathit{\text{Death}}}_4}\times {in3}_j+{\beta }_{{\mathit{\text{Death}}}_5}\times {in4}_j+{\beta }_{{\mathit{\text{Death}}}_6}\times \mathit{\text{female}}+{\beta }_{{\mathit{\text{Death}}}_7}\times \mathit{\text{White}}+{\beta }_{{\mathit{\text{Death}}}_8}\times \mathit{\text{Hispanic}}+{\beta }_{{\mathit{\text{Death}}}_9}\times \mathit{\text{OtherRace}}+{\beta }_{{\mathit{\text{Death}}}_{10}}\times \mathit{\text{MSMrisk}}+{\beta }_{{\mathit{\text{Death}}}_{11}}\times \mathit{\text{IDUrisk}}+{\beta }_{{\mathit{\text{Death}}}_{12}}\times \mathit{\text{SmokeNeve}}r+{\beta }_{{\mathit{\text{Death}}}_{13}}\times \mathit{\text{AnxDep}}+{\beta }_{{\mathit{\text{Death}}}_{14}}\times CD4ART+{\beta }_{{\mathit{\text{Death}}}_{15}}\times {Age}_j+{\beta }_{{\mathit{\text{Death}}}_{16}}\times {Age1}_j+{\beta }_{{\mathit{\text{Death}}}_{17}}\times {Age2}_j+{{\beta }_{{\mathit{\text{Death}}}_{18}}\times Age3}_j+{\beta }_{{\mathit{\text{Death}}}_{19}}\times {\mathit{\text{BMIcat}}1}_j+{\beta }_{{\mathit{\text{Death}}}_{20}}\times {\mathit{\text{BMIcat}}2}_j+{\beta }_{{\mathit{\text{Death}}}_{21}}\times {\mathit{\text{BMIcat}}3}_j+{{\beta }_{{\mathit{\text{Death}}}_{22}}\times \mathit{\text{eGFRcat}}1}_j+{\beta }_{{\mathit{\text{Death}}}_{23}}\times {\mathit{\text{eGFRcat}}2}_j+{\beta }_{{\mathit{\text{Death}}}_{24}}\times {CD4}_j+{\beta }_{{\mathit{\text{Death}}}_{25}}\times {HTN}_j+{\beta }_{{\mathit{\text{Death}}}_{26}}\times {\mathit{\text{TotCholcat}}1}_j+{\beta }_{{\mathit{\text{Death}}}_{27}}\times {\mathit{\text{TotCholcat}}2}_j+{\beta }_{{\mathit{\text{Death}}}_{28}}\times \mathit{\text{TotChol}}+{\beta }_{{\mathit{\text{Death}}}_{29}}\times {\mathit{\text{GlucoseCat}}1}_j+{\beta }_{{\mathit{\text{Death}}}_{30}}\times {\mathit{\text{GlucoseCat}}2}_j+\epsilon$.

Where:

in	person-months since ART initiation
in1	(max(0,in-12)2-max(0,in-76)2)/(76–12)2
in2	(max(0,in-24)2-max(0,in-76)2)/(76–12)2
in3	(max(0,in-36)2-max(0,in-76)2)/(76–12)2
in4	(max(0,in-52)2-max(0,in-76)2)/(76–12)2
female	1 if sex=female; 0 if sex=male
White	1 if race=White and ethnicity=(non-Hispanic or missing); 0 otherwise
Hispanic	1 if ethnicity=Hispanic, 0 otherwise
OtherRace	1 if race NE (White or Black) and ethnicity NE Hispanic; 0 otherwise
MSMrisk	1 if MSM listed as an HIV acquisition risk factor; 0 otherwise
IDUrisk	1 if IDU listed as an HIV acquisition risk factor; 0 otherwise
HCV	1 if ever evidence of an HCV infection; 0 otherwise
SmokeNever	1 if patient asked about smoking and was not a current or former smoker; 0 otherwise
AnxDep	1 if patient had a diagnosis of anxiety or depression prior to ART initiation; 0 otherwise
BMIARTcat	1 if BMI at ART initiation <25 kg/m2; 0 otherwise
BMIARTcat1	1 if BMI at ART initiation <18.5 kg/m2; 0 otherwise
BMIARTcat2	1 if BMI at ART initiation ≥30 kg/m2; 0 otherwise
AgeARTcat2	1 if 40 years ≤ age at ART initiation <50 years; 0 otherwise
AgeARTcat3	1 if 50 years ≤ age at ART initiation <60 years; 0 otherwise
AgeARTcat4	1 if 60 years ≤ age at ART initiation; 0 otherwise
AgeART2cat2	1 if age at ART initiation ≥46 years; 0 otherwise
GlucoseART	glucose (mg/dL) at ART initiation
TotCholART	total cholesterol (mg/dL) at ART initiation
CD4ART	CD4 cell count (cells/mm3) at ART initiation
eGFRART	eGFR (mL/min/1.73m2) at ART initiation
Agej	Age in month j (ageART+in/12)
Age1j	(max(0,age-23)2-max(0,age-65)2)/(65–23)2
Age2j	(max(0,age-34)2-max(0,age-65)2)/(65–23)2
Age3j	(max(0,age-48)2-max(0,age-65)2)/(65–23)2
BMIj	BMI measured in kg/m2 in month j
BMIcat1j	1 if BMIj <18.5 kg/m2; 0 otherwise
BMIcat2j	1 if 18.5 kg/m2 ≤ BMIj <25 kg/m2; 0 otherwise
BMIcat3j	1 if 25 kg/m2 ≤ BMIj <30 kg/m2; 0 otherwise
BMIcatC1j	1 if BMIj <18.5 kg/m2; 0 otherwise
BMIcatC2j	1 if BMIj ≥30 kg/m2; 0 otherwise
eGFRj	eGFR measured in mL/min/1.73m2 in month j
eGFRcat1j	1 if eGFRj <45 mL/min/1.73m2; 0 otherwise
eGFRcat2j	1 if 45 mL/min/1.73m2 ≤ eGFRj <90 mL/min/1.73m2
CD4j	CD4 measured in cells/mm3 in month j
CD4ARTtime	product term between CD4ART and follow-up time (in)
HTNj	1 if person had treated hypertension any time prior to month j; 0 otherwise
TotCholj	Total cholesterol measured in … in month j
TotCholcat1	1 if TotCholj <200 mg/dL; 0 otherwise
TotCholcat2	1 if 200 mg/dL ≤ TotCholj <240 mg/dL; 0 otherwise
Glucosej	Glucose measured in … in month j
Glucosecat1j	1 if glucosej<100 mg/dL; 0 otherwise
Glucosecat2j	1 if 100 mg/dL ≤ glucosej < 126 mg/dL; 0 otherwise
Diabetesj	1 if person meets the NA-ACCORD definition of diabetes (see text); 0 otherwise