Dolutegravir-based therapy, diet, physical activity and weight gain: a 48-week prospective cohort in South Africa

Introduction

The transition to tenofovir disoproxil fumarate, lamivudine, and dolutegravir (TLD)-based antiretroviral therapy (ART) has been associated with excess weight gain and clinical obesity, particularly in Black women.¹ For people with HIV (PWH) in South Africa, the risk of weight gain associated with TLD-based ART is compounded by a growing obesity epidemic due to changes in dietary patterns, urbanization, and aging.^2,3 The resulting weight gain also increases an already elevated risk of cardiovascular disease (CVD).⁴ Understanding the role of diet and physical activity in this weight gain is crucial as is trialing strategies to mitigate weight gain and its downstream risk of CVD.

Given the lack of behavioral data for PWH on TLD-based regimens, our study aimed to describe diet and physical activity behaviors and their relationship to weight gain among PWH in South Africa over 48 weeks post-transition to TLD.

Methods

Study population, design, and data collection

The Population Effectiveness of Dolutegravir Implementation in sub-Saharan Africa: a Prospective Observational Cohort Study (DISCO) followed 499 PWH in KwaZulu-Natal, South Africa as they transitioned to TLD-based ART. Participants ≥18 years old and on efavirenz-based ART for ≥6 months were enrolled from 2019 to 2022 and followed for 48 weeks. Full methods have been published previously.¹ At the enrollment, 24-week, and 48-week timepoints, study staff collected anthropomorphic measurements, including body weight (kg), height (cm), and waist circumference (cm).

Our outcome of interest was the occurrence of clinically significant weight gain, defined as a ≥10% increase in body weight over the 48-week period.⁵ Participants also completed questionnaires on dietary behavior and physical activity at each timepoint.

Our primary behaviors of interest over 48 weeks were: (1) change in fruit intake (servings/week), (2) change in vegetable intake (servings/week), (3) change in frequency of fast food intake (no change or became less frequent vs. more frequent), (4) change in frequency of fried food intake (no change or became less frequent vs. more frequent), (5) change in frequency of sugar-sweetened beverage (SSB) intake (no change or became less frequent vs. became more frequent), and (6) continuous change in physical activity (MET-minutes/week).

Ethical approval

Ethical approval for the study was obtained from Massachusetts General Brigham (2019P000898), the University of KwaZulu-Natal Biomedical Research Ethics Committee (No. BE485/19), and the South Africa Department of Health. All participants provided written informed consent.

Statistical analysis

We analyzed data from all non-pregnant DISCO participants with weight data available at enrollment and 48 weeks. We compared changes in each behavioral exposure between enrollment and 48 weeks among participants with and without clinically significant weight gain. T-tests were used to assess the mean change in continuous variables (fruit intake, vegetable intake, and physical activity), and Chi-squared tests were used to assess the change in proportions (fast food, fried food, and SSB intake frequency).

We fit logistic regression models to evaluate the relationship between each behavior and the occurrence of clinically significant weight gain, adjusting for age at enrollment (categorized as greater versus less than the median cohort age of 42 years), sex, highest completed education level (no schooling, primary, secondary, or tertiary school), and ART duration prior to TLD transition (in years).

Results

We analyzed data from 367 participants from the DISCO cohort. Participants with missing weight measurements at the first and/or last timepoint were excluded from all analyses (n = 102, 20.4%). Women who were or became pregnant at any point during the study were also excluded (n = 37, 7.4%). Overall, 132 (26.5%) of participants were excluded. The mean age of participants was 43.1 years old (standard deviation [SD] = 12.1 years), with no significant difference between those with and without clinically significant weight gain (42.4 years vs. 43.3 years) (p = 0.59). In total, 287 (78%) of all participants were female, including 56 out of 66 (85%) of those with clinically significant weight gain (Table 1). At the time of transition to TLD, 360 (98.1%) participants were virally suppressed, five (1.36%) participants were not virally suppressed, and viral load data was missing for two (0.54%) participants.

Table 1.

Cohort characteristics overall and stratified by occurrence of clinically significant weight gain.

Characteristic	Overall (n = 367)	Without Clinically Significant Weight Gain (n = 301)	With Clinically Significant Weight Gain (n = 66)	P-value
Age (years) (n, %)				0.62
≤42	190 (52)	154 (51)	36 (55)
42+	177 (48)	147 (49)	30 (45)
Mean, (SD)	43 (12)
Sex (n, %)				0.15
Male	80 (22)	70 (23)	10 (15)
Female	287 (78)	231 (77)	56 (85)
Education (n, %)				0.61
No schooling	45 (12)	38 (13)	7 (11)
Primary school	73 (20)	59 (20)	14 (21)
Secondary school	232 (63)	192 (64)	40 (61)
Tertiary	17 (5)	12 (4)	5 (8)
Weight (kg) (n, %)				0.011
≤55	57 (16)	41 (14)	16 (24)
56–70	155 (42)	123 (41)	32 (48)
71+	155 (42)	137 (46)	18 (27)
BMI (kg/m2) (n, %)				0.058
<18.5	19 (5)	13 (4)	6 (9)
18.5–24.9	166 (45)	131 (44)	35 (53)
≥25	182 (50)	157 (52)	25 (38)
ART duration (years) (Median, IQR)	6.73, (4.41, 10.03)	6.79, (4.57, 9.70)	6.50, (3.59, 10.68)	0.80

Eighteen percent of participants in the DISCO study experienced clinically significant weight gain. These participants had increased fruit intake compared to those who did not have significant weight gain (0.21 servings/week [95% CI: −0.9 to 1.3] versus −1.07 servings/week [95% CI: −1.5 to −0.6], respectively) (p = 0.015). However, we found no differences in change in physical activity (1390 MET-min/week [95% CI: −122.9 to 2904] versus 2009 MET-min/week [95% CI: 1138 to 2881], respectively) (p = 0.54), change in fast food (8.51% [95% CI: 0.53% to 16.5%] versus 15.1% [95% CI: 10.8% to 19.4%], respectively) (p = 0.23), fried food (29.8% [95% CI: 16.7% to 42.9%] versus 27.2% [95% CI: 21.8% to 32.5%], respectively) (p = 0.71), or SSB intake frequency (24.2% [95% CI: 13.9% to 34.6%] versus 16.3% [95% CI: 12.1% to 20.4%], respectively) (p = 0.12). Similarly, no differences were observed in vegetable intake (−0.32 servings/week [95% CI: −1.2 to 0.58] versus −0.89 servings/week [95% CI: −1.2 to −0.55], respectively) (p = 0.17) (Figure 1).

Figure 1.

Fruit, vegetable, fast food, fried food, and sugar-sweetened beverage intake and physical activity across 48 weeks for participants with and without clinically significant weight gain.

In a univariable logistic regression model with clinically significant weight gain as the outcome, change in fruit intake, vegetable intake, and physical activity were significantly associated with weight gain (p < 0.001). However, in a multivariable logistic regression model adjusted for age, sex, education level, ART duration, and behaviors, none of the behavioral factors were significantly associated with clinically significant weight gain (Supplemental Digital Content S1). In sex-stratified models, these results were preserved for both women and men (Supplemental Digital Content S2 & S3).

Discussion

Our findings show little relationship between self-reported diet and physical activity on clinically significant weight gain as defined by ≥10% increase in body weight over 48 weeks after transitioning to TLD. The only exception to this was fruit intake which differed modestly between those who gained weight and those who did not. These findings are important, as changes in lifestyle behaviors, possibly driven by differences in appetite, have been hypothesized to contribute to overweight and obesity in people who are switched to integrase inhibitors. Our results suggest these behaviors may not be primary factors affecting weight gain in this population. Of note is that one study from a high-income context found greater weight gain among people with low physical activity which was not reproduced in our population; this may be due to differences in context or because our study considers physical activity among a greater suite of measured behaviors that can all influence weight gain.⁶ Consequently, interventions targeting diet and exercise alone are unlikely to prevent weight gain or clinical obesity in PWH, though they still offer benefits for cardiovascular and metabolic health. Alternative strategies may be needed to manage weight gain and obesity in this population, potentially including anti-obesity medications in those at highest risk.

Our study’s strengths include the use of contextually validated behavior reporting scales, highly accurate, clinic-based anthropometric measurements, and follow-up of a large group of PWH observed as part of routine HIV care in rural South Africa. However, several limitations should be noted. The use of self-reported behaviors may introduce recall bias or socially desirability bias. Additionally, we did not collect biological markers such as blood glucose or blood pressure, limiting our ability to assess cardiovascular and metabolic health. Nevertheless, changes in diet and activity behaviors are still beneficial in protecting against adverse cardiometabolic outcomes, regardless of weight loss.⁷ The setting and population of our study also invite certain limitations. The rural setting limits the generalizability of our findings to urban populations or other regions, and the weight change associated with TLD initiation for African women, our main study population, has not been reproduced to the same magnitude in different populations or geographical areas. However, the risk of weight gain for Black women upon TLD initiation is preserved across continents.¹ Lastly, data collection for this study overlapped with the COVID-19 pandemic. Due to social distancing restrictions during the time, 20.4% of participants had missing body weight data, which may have affected our analysis (Supplemental Digital Content S4). Additionally, the pandemic may have impacted changes in dietary and activity behaviors.

While a change in fruit intake was modestly associated with clinically significant weight gain, no other behavioral factors were significant. Other strategies additional to lifestyle changes, such as pharmacologic intervention, may be required to successfully combat and mitigate weight gain in PWH.