Skip to main content
NCBI home page
PLOS One logoLink to PLOS One
. 2023 Feb 17;18(2):e0282001. doi: 10.1371/journal.pone.0282001

Prevalence and incidence of hypertension in a heavily treatment-experienced cohort of people living with HIV in Uganda

Dathan M Byonanebye 1,2,*, Mark N Polizzotto 3, Rosalind Parkes-Ratanshi 4,5, Joseph Musaazi 4, Kathy Petoumenos 1, Barbara Castelnuovo 4
Editor: Giuseppe Vittorio De Socio6
PMCID: PMC9937480  PMID: 36800379

Abstract

Introduction

The effect of long-term exposure to antiretroviral therapy (ART) on hypertension in sub-Saharan Africa remains unclear. We aimed to determine the prevalence and incidence of hypertension in people living with HIV (PLWH) with more than 10 years of ART in Uganda.

Methods

The analysis was performed within a cohort of adult PLWH with more than 10 years of ART at an HIV clinic in Kampala, Uganda. Participants were eligible for this analysis if they had ≥2 follow-up visits. Hypertension was defined as two consecutive systolic blood pressure (SBP) measures greater than 140 mmHg and/or diastolic blood pressure (DBP) greater than 90 mmHg, and/or documented diagnosis and/or the initiation of antihypertensives. We determined the proportion of PLWH with hypertension at baseline and used multivariable logistic regression to determine the factors associated with prevalent hypertension. To determine the incidence of hypertension, follow-up began from the cohort baseline date and was censored at the last clinic visit or date of the event, whichever occurred earlier. Multivariable Poisson regression was used to determine the adjusted incidence rate ratios (aIRR) of hypertension according to demographic, ART, and clinical characteristics.

Results

Of the 1000 ALT participants, 970 (97%) had ≥2 follow-up visits, and 237 (24.4%) had hypertension at baseline. The odds of prevalent hypertension were 1.18 for every 5-year increase in age (adjusted odds ratio (aOR) 1.18, 95% CI 1.10–1.34) and were higher among males (aOR 1.70, 95% CI 1.20–2.34), participants with diabetes mellitus (aOR 2.37, 95% CI 1.10–4.01), obesity (aOR 1.99, 95% CI 1.08–3.60), high cholesterol (aOR 1.47, 95% CI 1.16–2.01), and those with prior exposure to stavudine (aOR 2.10, 95% CI 1.35–3.52), or nevirapine (aOR 1.90, 95% CI 1.25–3.01). Of the 733 participants without hypertension at baseline, 116 (15.83%) developed hypertension during 4671.3 person-years of follow-up (incidence rate 24.8 per 1000 person-years; 95% CI 20.7–29.8). The factors associated with incident hypertension were obesity (adjusted incidence rate ratio (aIRR) 1.80, 95% CI 1.40–2.81), older age (aIRR 1.12 per 5-year increase in age, 95% CI 1.10,1.25), and renal insufficiency (aIRR1.80, 95% CI 1.40–2.81).

Conclusion

The prevalence and incidence of hypertension were high in this heavily treated PLWH cohort. Therefore, with increasing ART coverage, HIV programs in SSA should strengthen the screening for hypertension in heavily treated PLWH.

Introduction

The successful scale-up of antiretroviral therapy (ART) has improved the overall survival of people living with HIV (PLWH) globally and in sub-Saharan Africa (SSA) [1], although survival still lags that of people without HIV [2]. This improvement in survival has contributed to an epidemiological transition in the cause of mortality in PLWH away from AIDS-related opportunistic infections towards non-communicable diseases, including cardiovascular diseases [3, 4]. Globally, the risk of cardiovascular disease is twice as high in PLWH than in people without HIV. In some SSA countries, more than 15% of the cardiovascular burden occurs in PLWH [5]. However, it is unclear whether the increasing cardiovascular burden in PLWH is attributable to ageing and/or the increasing prevalence of cardiovascular disease risk factors, including hypertension. Hypertension is an important independent risk factor for cardiovascular disease, accounting for up to 44% of cardiovascular events in some HIV cohorts [15]. The cardiovascular risk attributable to hypertension is higher in black people than in other ethnicities [6]; however, there is little data on hypertension in heavily treated PLWH in SSA.

Despite increasing ART coverage among PLWH in SSA [7], the impact of long-term ART exposure on cardiovascular risk factors, such as hypertension, has not been fully described [8]. The relationship between ART exposure and hypertension was initially investigated in the Multicenter AIDS Cohort, in which it was demonstrated that exposure to ART for more than two years was associated with a higher risk of hypertension [9]. While this analysis was conducted before ART became widely available, subsequent studies in the modern ART era have also demonstrated a greater risk of hypertension in ART-experienced versus ART-naïve PLWH. A recent systematic review reported higher systolic and diastolic blood pressures of 4.52 mmHg and 3.17 mmHg, respectively, in ART-experienced PLWH than in treatment-naive individuals [10]. The effect of ART on blood pressure appears to be more detrimental in sub-Saharan Africa; a recent study showed that ART for at least three months is associated with an increase in systolic and diastolic blood pressure of 7.85 mm Hg and 7.45 mm Hg, respectively [11]. The prevalence of hypertension among PLWH increased by 10% between 2007 and 2017 [12], possibly due to increasing ART coverage and overall survival.

It is unclear whether the increase in blood pressure following ART initiation is sustained with prolonged exposure to ART and whether the exposure translates to higher rates of hypertension. Additionally, prior analyses have linked older antiretroviral drugs such as nevirapine, indinavir/ritonavir, and stavudine to a higher risk of hypertension [13, 14]. Until recently, these agents were the cornerstones of ART in SSA [15, 16] although it is unclear whether the risk posed by these agents persists in PLWH who switch to contemporary ART regimens, the regimens currently recommended for HIV treatment. Therefore, in this analysis, we sought to determine the prevalence and incidence of hypertension and its relationship with exposure to specific antiretroviral therapy in a Ugandan cohort of PLWH with long durations of ART.

Methods

The analysis was conducted within the antiretroviral treatment long-term (ALT) cohort (ClinicalTrials.gov #: NCT02514707), which has previously been described [17, 18]. Briefly, the ALT study is an ongoing single-centre prospective cohort of 1000 PLWH on ART for at least 10 years at baseline. Cohort participants were recruited from the Infectious Diseases Institute (IDI) HIV clinic in Kampala, Uganda, between 2014 and 2015. Because of prolonged exposure to ART at baseline (>10 years), the participants in this cohort were considered heavily treatment-experienced since HIV resistance testing is limited in resource-limited settings. At the cohort baseline, data on HIV treatment, and other covariates, including blood pressure measurements and hypertension treatment, were collected from an existing clinic health information management system. Participants in the cohorts are evaluated annually, blood pressure measurements are taken, and treatment for hypertension and other co-comorbidities, including hypertension, is documented. Blood pressure measurement is standardised at all visits, and readings are taken by qualified nurses using calibrated blood pressure machines, consistent with international guidelines [19].

Data on complications associated with HIV infection and ART is also collected. Laboratory tests (HIV RNA viral load, CD4 count, and creatinine level) are also performed at every visit, whereas lipid tests were only available at baseline. HIV treatment for participants is standardised and follows the World Health Organization (WHO) and Uganda ART guidelines [20, 21]. All PLWH receive a three-drug ART regimen consisting of two nucleos(t)ide reverse transcriptase inhibitors (NRTIs) and non-nucleoside reverse transcriptase inhibitors (NNRTI) as the first-line treatment or protease inhibitors (PIs) as the second-line treatment. In 2018, dolutegravir (DTG) was introduced in Uganda as a part of a three-drug regimen for all patients receiving ART. Patients already on ART were recommended to switch from PIs or NNRTIs to DTG-containing regimens, and the process of switching all participants is underway.

Participant selection and inclusion criteria into analysis datasets

We included all participants in the cohort with at least two blood pressure measurements recorded at two different follow-up visits. Participants with fewer than two follow-up visits were excluded because hypertension diagnosis requires at least two measurements taken at two different visits [19]. In addition, individuals with hypertension at baseline were excluded from the hypertension incidence analysis. The baseline date for all analyses was the date of enrolment in the cohort (i.e., cohort baseline).

Definitions and analysis endpoints

Hypertension was defined as systolic blood pressure (SBP) ≥ 140 mmHg and/or diastolic blood pressure (DBP) ≥ 90 mmHg on two consecutive visits or initiation of antihypertensives including angiotensin-converting enzyme inhibitors (ACEIs). This definition has been used in other analyses of HIV cohorts in SSA [22–24] and a prior analysis of the ALT cohort [18]. The definition is also consistent with international Hypertension diagnosis and management guidelines, which recommend two consecutive abnormal blood pressure measurements [19]. Therefore, prevalent hypertension, determined at baseline, was defined as two consecutive systolic blood pressure (SBP) measures greater than 140 mmHg and/or diastolic blood pressure (DBP) greater than 90 mmHg, and/or documented diagnosis and/or the initiation of antihypertensives within one year before and up to one month after the baseline date. The other covariates were also defined as consistent with a prior analysis of this cohort [18]. Diabetes mellitus was defined as a clinical diagnosis and/or random blood glucose levels ≥11.1 mmol/L based on point-of-care glucose testing and/or the initiation of antidiabetic medication, consistent with other analyses in HIV cohorts [25, 26]. Hepatitis B infection was defined as a positive hepatitis B surface antigen test result, which is the screening strategy recommended by WHO for high-burden countries [27]. The Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation was used to determine the estimated glomerular filtration rate (eGFR), and renal insufficiency was defined as eGFR <90 mL/min/1.73 m2 [28].

Statistical analyses

We determined the proportion of PLWH with hypertension at baseline and compared the characteristics of PLWH with hypertension versus those without hypertension. Descriptive statistics for ordinal and categorical variables were computed as frequencies and percentages, and continuous variables were summarised as medians (interquartile ranges, [IQR]). Multivariable logistic regression was used to determine the factors associated with hypertension at baseline. We then performed univariable analysis followed by multivariable Poisson regression to determine hypertension incidence rate ratios (aIRR) and their corresponding 95% confidence intervals (CI) according to exposure to demographic, HIV-infected, and clinical characteristics. While fitting the Poisson regression, the logarithm of person-years of follow-up was included in the model as an offset term to account for the observation time of individuals.

The overall incidence of hypertension was determined and stratified according to demographic, metabolic, and HIV-related factors. Follow-up began from the cohort baseline date and was right censored at the last available study visit or the date of the event, whichever occurred earlier. The factors that were considered in the multivariable model include sex, age, smoking and alcohol status, calendar year, prior AIDS, total cholesterol (TCHOL), triglycerides (TRIG), low-density lipid cholesterol (LDL), body mass index (BMI), diabetes, HIV RNA and CD4 counts, and eGFR and prior exposure to individual antiretroviral drugs (before baseline). These factors were selected a priori based on published data linking them to hypertension [14, 22, 24, 29, 30]. All variables, except age, were modelled as categorical variables, and the categories were selected so that clinically meaningful conclusions could be made. For example, the BMI categories followed the WHO categories [31], and blood pressure categories are based on the European Association for cardiology categorisation for blood pressure [32]. The CD4 categories are also based on the Centers for Disease Control and Prevention (CDC) categories for immunosuppression status [33] while lipids were categorised following the thresholds for cardiovascular risk [34]. All variables in this analysis were fixed at baseline, and the regression models were manually fitted. In the backward selection process, variables with a p-value <0.25 at the univariable stage were considered in the multivariable analysis. All variables dropped from the multivariate model were assessed to determine if they were confounders (i.e., causes >10% change in effects when added in the model), in which case they were retained in the adjusted model. The goodness-of-fit test was used to determine the fitness of the model with the confounder versus the one without the confounder and the one with the smallest Akaike information criterion were considered the better fitting. Specifically, BMI and age have consistently been shown to be confounders for hypertension [14, 22, 24, 29, 30]; their confounding potential was checked, and the variables were consistently included in the model. Variables, other than confounders, that were considered for multivariable analysis but dropped during backward elimination due to non-significance (p<0.05) were later added one at a time to determine adjusted estimates for these variables. We checked for multicollinearity in the variables included in the final model using a variance inflation factor (VIF) cut-off of 5.0. Odds ratios (95% confidence intervals (CI)) and incidence rate ratio (95% CI) were used to determine the association between variables and prevalent and incident hypertension, respectively, and a two-sided p-value <0.05 was considered statistically significant. The final logistic and Poisson regression models were evaluated to ensure compliance with the respective model assumptions. The Wald chi-square test was used to determine whether the final model was statistically significant.

Where data were missing for a variable, an unknown category was assigned, and the variable was fitted as a categorical variable. To determine the impact of missing data and the robustness of estimates, we used the regular Little’s test [35], to determine if missing data were missing completely at random (MCAR) and if multiple imputation was necessary. Data for this analysis were prepared using SAS Enterprise Guide software version 8.3 release 5 (SAS Institute Inc., Cary, NC, USA) and analysed with Stata version 16.0 (StataCorp, College Station, Texas, USA). Participants in the ALT cohort provided written informed consent, and the study received ethical approval from the Joint Clinical Research Council Research Ethics Committee (JCRC REC) and Uganda National Council of Science and Technology (UNCST; Approval #: UNCST Folio Number: HS 1586 24th 03 2014).

Results

Cohort description and characteristics of study participants

A total of 1000 participants were enrolled in the ALT cohort, 970 of whom had at least two follow-up visits (with at least two blood pressure measurements) and were included in the analysis to determine the prevalence of hypertension (Fig 1). The median number (interquartile range, IQR) of follow-up visits was 5 (4–6), and all visits were annual. The 30 participants who were excluded because they had fewer than two follow-up visits had comparable ages (median age: 45.0; IQR 40.0,51.0) but were mainly male (53%).

Fig 1. Participant inclusion for hypertension and weight gain analysis.

Fig 1

Open in a new tab

Prevalence of hypertension

Of the 970 participants included in the prevalence analysis, 237 (24.4%) had hypertension at cohort enrolment (baseline). Overall, the median age was 45.4 (IQR 40.4–50.5) years and there were more females (61.8%) than males. A total of 466 participants (48.0%) had experienced an AIDS event before enrolment in the cohort; 960/970 (99.0%) had HIV RNA levels <200 copies/mL, and the median CD4 cell count was 509 (IQR,365–684). The baseline smoking rate was 2.2%. The characteristics of participants with and without hypertension at enrolment are presented in Table 1.

Table 1. Characteristics of participants with versus without hypertension at cohort enrolment (n = 970).

 Variable Prevalent Hypertension No Hypertension* All
N % N % n %
237 24.4 733 75.57 970 100
Sex  Male 101 42.6 270 36.8 371 38.3
Female 136 57.4 463 63.2 599 61.8
Prior AIDS  Yes 113 47.7 353 48.2 466 48.0
No 124 52.3 380 51.8 504 52.0
Hepatitis B infection Positive 8 3.4 31 4.2 39 4.0
Negative 229 96.6 702 95.8 931 96.0
Smoking Status Current 3 1.3 18 2.5 21 2.2
Prior 50 21.1 149 20.3 199 20.5
Never 184 77.6 566 77.2 750 77.3
Alcohol Status Current 52 21.9 188 25.7 240 24.7
Prior 184 77.6 530 72.3 714 73.6
Never 1 0.4 15 2.1 16 1.7
Diabetes mellitus Yes 21 8.9 14 1.9 35 3.6
No 216 91.1 719 98.1 935 96.4
NRTI backbone TDF/3TC 67 28.3 243 33.2 310 32.0
AZT/3TC 164 69.2 486 66.3 650 67.0
ABC/3TC 4 1.7 4 0.6 8 0.8
Other 2 0.84 0 0 2 0.2
ART class NNRTIs 198 83.5 582 79.4 780 80.4
PIs 39 16.5 151 20.6 190 19.6
Variable  Median (IQR) n missing (%) Median (IQR) n missing (%) Median (IQR) n missing (%)
Age (years) 47.4 (42.4,52.7) 0 (0) 44.7 (40.3,50.3) 0 (0) 45.4 (40.4,50.5) 0 (0)
Systolic Blood Pressure 130 (120,146) 0 (0) 120 (110,125) 0 (0) 120 (110,130) 0 (0)
Diastolic Blood Pressure 80 (70,90) 0 (0) 70 (69,80) 0 (0) 70 (70,80) 0 (0)
Body mass index (Kg/M2) 23.1 (20.3,26.0) 5 (2.11) 22.1 (19.7,25.0) 8 (1.1) 22.4 (19.8,25.3) 13 (1.3)
eGFR (mL/min/1.73 m2) 116 (103,128) 86 (36.3) 122.9 (109,130) 325 (44.3) 121.4 (107,130) 411 (42.4)
HIV RNA (copies/mL) 19 (19,19) 0 (0) 19 (19,19) 0 (0) 19 (19,19) 0 (0)
Baseline CD4 (cells/μL) 518 (378,689) 0 (0) 508 (359,683) 0 (0) 509 (365,684) 0 (0)
Nadir CD4 (cells/μL) 391 (275,507) 0 (0) 401 (290,528) 0 (0) 399 (289,523) 0 (0)
TCHOL (mmol/L) 4.9 (4.3,5.7) 2 (0.8) 4.7 (4.0,5.3) 12 (1.6) 4.7 (4.1,5.4) 14 (1.4)
LDL (mmol/L) 2.8 (2.2,3.3) 2 (0.8) 2.6 (2,3.2) 11 (1.5) 2.6 (2.0,3.2) 13 (1.3)
HDL (mmol/L) 1.2 (1.0,1.5) 2 (0.8) 1.2 (1.0,1.5) 11 (1.5) 1.2 (0.98,1.5) 13 (1.3)
TRIG (mmol/L) 1.4 (1.1,1.9) 94 (39.7) 1.3 (0.9,1.8) 358 (48.8) 1.3 (1.0,1.8) 452 (46.6)
Number of Follow-up visits 5 (4,6) 0 (0) 5 (4,6) 0 (0) 5 (4,6) 0 (0)
Baseline date (mm/yy) 02/15 (08/14,05/15) 0 (0) 08/14 (05/15,06/15) 0 (0) 03/15 (08/14,06/15) 0 (0)
Exposure to ART class/NRTIS (years) Median (IQR) n exposed (%) Median (IQR) n exposed (%) Median (IQR) n exposed (%)
Cumulative exposure to NNRTIs 9.4 (7.7,9.6) 234 (98.7) 9.3 (4.9,9.6) 721 (98.4) 9.3 (5.4,9.6) 955 (98.5)
Cumulative exposure to PIs 6.3 (3.0,7.3) 37 (15.6) 4.9 (2.2,6.7) 139 (19.0) 5.4 (2.4,6.9) 176 (18.1)
Cumulative exposure to ddI 5.9 (2.7,7.1) 11 (4.6) 6.8 (5.2,7.4) 28 (3.8) 6.0 (5.1,7.4) 39 (4.0)
Cumulative exposure to d4T 2.7 (2.1,3.2) 201 (84.8) 2.6 (2.0,3.1) 570 (77.8) 2.6 (2.1,3.1) 771 (79.5)
Cumulative exposure to AZT 6.5 (4.9,7.0) 210 (88.6) 6.4 (2.4,7.0) 684 (93.3) 6.5 (3.0,7.0) 894 (92.2)
Cumulative exposure to TDF 5.1 (2.4,6.9) 64 (27.0) 3.5 (1.7,6.1) 230 (31.4) 3.6 (1.8,6.3) 294 (30.3)
Cumulative exposure to EFV 9.3 (6.2,9.6) 204 (86.1) 9.2 (4.6,9.6) 592 (80.8) 9.3 (4.9,9.6) 796 (82.1)
Cumulative exposure to NVP 2.2 (0.1,7.6) 69 (29.1) 1.6 (0.1,8.4) 278 (37.9) 1.6 (0.1,8.2) 347 (35.8)
Cumulative exposure to LPV 6.4 (5.3,7.1) 33 (13.9) 6.0 (4.0,7.2) 102 (13.9) 6.1 (4.1,7.2) 135 (13.9)
Cumulative exposure to ATV 2.1 (1.6,2.4) 8 (3.4) 1.4 (0.4,2.1) 42 (5.7) 1.6 (0.5,2.2) 50 (5.2)
Open in a new tab

Note: *733 participants without hypertension at baseline were included in the incidence analysis. n means the number of participants; AIDS-prior AIDS-defining event; NNRTIs-non nucleoside reverse transcriptase inhibitors; NRTIs-nucleos(t)ide reverse transcriptase inhibitors, ART-antiretroviral therapy, eGFR-estimated glomerular filtration rate, PI-protease inhibitors, TDF- tenofovir disoproxil fumarate; EFV-efavirenz, LPV-lopinavir; ATV-atazanavir; ddI-didanosine; d4t-stavudine; AZT-zidovudine; NVP-nevirapine; LPV-lopinavir; RTV-ritonavir; ABC-abacavir; 3TC-lamivudine; TRIG-triglycerides; HDL-high-density lipoprotein; LDL-low-density lipoprotein cholesterol; TCHOL-total cholesterol; mm/yy means month followed by the year of the baseline date.

Compared with those without hypertension, participants with hypertension at cohort enrolment were older, more likely to be male, had diabetes mellitus, had higher total cholesterol levels, and had more prolonged exposure to stavudine and nevirapine. After adjustment for confounders, the odds of prevalent hypertension were higher among males (aOR 1.70, 95% CI 1.20–2.34), as well as in participants with diabetes mellitus (2.37, 95% CI 1.10–4.01), obesity (aOR 1.99, 95% CI 1.08–3.60), higher total cholesterol (aOR 1.47, 95% CI 1.16–2.01), and prior exposure to stavudine (aOR 2.10, 95% CI 1.35–3.52), or nevirapine (aOR 1.90, 95% CI 1.25–3.01). In addition, the odds of prevalent hypertension were higher in participants with missing triglyceride levels. Finally, the odds of hypertension were 1.18 (CI 1.10–4.01) for every 5-year increment in the cohort enrolment age (Table 2).

Table 2. Factors associated with prevalent hypertension in the ALT cohort at baseline (n = 970).

Variable Variable Categories Crude OR (95% CI) p-value Adjusted OR (95% CI) p-value
Sex Female Ref Ref
Male 1.27 (0.95,1.72) 0.112 1.70 (1.20,2.34) 0.002
Alcohol History Never Ref
Current 4.15 (0.54,32.14) 0.173 1.10 (0.19,12.04) 0.856
Prior 5.21 (0.68,39.7) 0.111 1.67 (0.20,14.42) 0.698
Diabetes mellitus No DM Ref Ref
DM 2.97 (1.48,5.95) 0.002 2.37 (1.10,4.01) 0.011
Age, Per 5 years increment 1.20 (1.10,1.31) 0.000 1.18 (1.10,1.34) 0.003
BMI (Kg/M2) <18.5 0.98 (0.62,1.54) 0.922 0.91 (0.56,1.50) 0.567
18.5–24.9 Ref Ref
25–29.9 1.33 (0.92,1.92) 0.130 1.47 (0.98,2.30) 0.075
>29.9 1.66 (1.15,2.87) 0.047 1.99 (1.08,3.60) 0.031
GFR (mL/min/1.73 m2) <90 1.60 (0.85,2.99) 0.142 1.13 (0.61,2.46) 0.765
≥90 Ref Ref
Missing 0.75 (0.55,1.02) 0.063 1.43 (0.95,2.09) 0.078
HIV RNA (copies/mL) <200 1.74 (0.81,3.76) 0.159 1.78 (0.71,4.00) 0.173
≥200 Ref Ref
Baseline TRIG (mmol/L) <1.7 Ref Ref
1.7–2.2 1.07 (0.58,1.95) 0.837 0.83 (0.50,1.60) 0.680
≥2.3 1.41 (0.86,2.30) 0.169 1.17 (0.69,1.96) 0.574
Missing 0.74 (0.53,1.03) 0.071 2.94 (2.00,4.69) <0.001
Baseline LDL (mmol/L) <2.6 Ref Ref
2.6–3.3 1.11 (0.79,1.56) 0.547 0.98 (0.65,1.54) 0.861
≥3.4 1.50 (1.02,2.19) 0.037 1.10 (0.49,1.99) 0.900
Missing 0.63 (0.14,2.88) 0.549 0.60 (0.19,4.72) 0.660
Baseline TCHOL (mmol/L) <5.2 Ref
≥5.2 1.57 (1.15,2.12) 0.004 1.47 (1.16,2.01) 0.008
Missing 0.60 (0.13,2.71) 0.505 0.48 (0.46,2.80) 0.460
Prior exposure to d4T Never exposed Ref Ref
Prior Exposure 1.60 (1.08,2.37) 0.020 2.10 (1.35,3.52) <0.001
Prior exposure to TDF Never exposed Ref Ref
Prior Exposure 0.81 (0.58,1.12) 0.200 0.76 (0.61,1.34) 0.489
Prior exposure to EFV Never exposed Ref Ref
Prior Exposure 0.73 (0.51,1.07) 0.104 0.75 (0.51,1.13) 0.144
Prior exposure to NVP Never exposed Ref Ref
Prior Exposure 1.47 (0.98,2.21) 0.062 1.90 (1.25,3.01) 0.002
Prior exposure to ATV Never exposed Ref Ref
Prior Exposure 0.61 (0.28,1.31) 0.203 0.86 (0.40,2.14) 0.710
Open in a new tab

Note: OR-odds ratio, AIDS-prior AIDS-defining event; BMI-body mass index; eGFR-estimated glomerular filtration rate; TDF-tenofovir disoproxil fumarate; EFV-efavirenz; ATV-atazanavir; d4T-stavudine; NVP-nevirapine, TRIG-triglycerides, HDL-high-density lipoprotein; LDL-low-density lipoprotein cholesterol; TCHOL-total cholesterol. Covariates prior AIDS, hepatitis B infection status, smoking history, prior exposure to lopinavir (LPV), abacavir (ABC), didanosine (ddI), high-density lipoprotein, baseline year, and baseline and nadir CD4 counts were not significant and were not confounders in the univariable analysis (P>0.2) and were not considered in the multivariable analysis. The significant variables in the final multivariable model are in bold, whereas the insignificant variables are in italics.

Incidence of hypertension

After excluding 237 participants with hypertension at enrolment, 733 participants were included in the incidence analysis. Of the 733 included participants, the baseline ART regimen consisted of NNRTIs in 583 (79.5%), and the remaining 150 (20.5%) received PI-based regimens. The NRTI backbones were zidovudine/lamivudine (AZT/3TC) (486, 66.3%), tenofovir disoproxil fumarate (TDF) with emtricitabine or lamivudine (TDF/XTC) (243, 33.2%), and abacavir/lamivudine (ABC/3TC) (4, 0.6%) (Table 3).

Table 3. Characteristics of participants without hypertension at baseline (n = 733).

  Incident Hypertension No incident Hypertension All
n % n % n %
116 15.8 617 84.2 733 100
Sex Male 43 37.1 227 36.8 270 36.8
Female 73 62.9 390 63.2 463 63.2
Prior AIDS Yes 46 39.7 307 49.8 353 48.2
No 70 60.3 310 50.2 380 51.8
Hepatitis B infection Positive 4 3.5 27 4.4 31 4.2
Negative 112 96.6 590 95.6 702 95.8
Smoking Status Current 3 2.6 15 2.4 18 2.5
Prior 16 13.8 133 21.6 149 20.3
Never 97 83.6 469 76.0 566 77.2
Alcohol Status Current 24 20.7 164 26.6 188 25.7
Prior 89 76.7 441 71.5 530 72.3
Never 3 2.6 12 1.9 15 2.1
Diabetes mellitus Yes 3 2.6 11 1.8 14 1.9
No 69 59.5 358 58.0 427 58.3
Missing 44 37.9 248 40.2 292 39.8
NRTI backbone TDF/3TC 30 25.9 213 34.5 243 33.2
AZT/3TC 86 74.14 400 64.8 486 66.3
ABC/3TC 0 0 4 0.7 4 0.6
ART regimen NNRTIs 103 88.8 480 77.8 583 79.5
PIs 13 11.2 137 22.2 150 20.5
  Median (IQR) n missing (%) Median (IQR) n missing (%) Median (IQR) n missing (%)
Age (years) 45.6 (42.2,51.7) 0 (0) 44.4 (40,49.7) 0 (0) 44.7 (40.3,50.3) 0 (0)
Systolic Blood Pressure 130 (120,140) 0 (0) 112 (110,120) 0 (0) 120 (110,125) 0 (0)
Diastolic Blood Pressure 80 (70,85.5) 0 (0) 70 (67,74) 0 (0) 70 (69,80) 0 (0)
BMI (Kg/M2) 23.2 (21.1,27.1) 0 (0) 21.8 (19.5,24.9) 8 (1.30) 22.1 (19.7,25.0) 8 (1.1)
GFR (mL/min/1.73 m2) 121 (103,126) 43 (37.1) 123 (110,130) 282 (45.7) 122.9 (109,130) 325 (44.3)
HIV RNA (copies/mL) 19 (19,19) 0 (0) 19 (19,19) 0 (0) 19 (19,19) 0 (0)
Baseline CD4 (cells/μL) 534 (373.5,679) 0 (0) 506 (350,687) 0 (0) 508 (359,683) 0 (0)
Nadir CD4 (cells/μL) 411 (290,581.5) 0 (0) 399 (290,519) 0 (0) 401 (290,528) 0 (0)
TCHOL (mmol/L) 4.7 (4.2,5.4) 3 (2.6) 4.6 (4.0,5.3) 9 (1.5) 4.7 (4.0,5.3) 12 (1.6)
LDL (mmol/L) 2.7 (2.1,3.2) 3 (2.6) 2.6 (2.0,3.1) 8 (1.3) 2.6 (2.0,3.2) 11 (1.5)
HDL (mmol/L) 1.3 (1.0,1.5) 3 (2.6) 1.2 (1.0,1.5) 8 (1.3) 1.2 (1.0,1.5) 11 (1.5)
TRIG (mmol/L) 1.5 (1.0,2.2) 49 (42.2) 1.3 (0.9,1.7) 309 (50.1) 1.3 (0.9,1.8) 358 (48.8)
Number of Follow-up visits 5 (5,6) 0 (0) 5 (4,6) 0 (0) 5 (4,6) 0 (0)
Baseline date (mm/yy) 02/15 (08/14,06/15) 0 (0) 03/15 (08/14,06/15) 0 (0) 08/14 (05/15,06/15) 0 (0)
Exposure to ART class (years) Median (IQR) n exposed (%) Median (IQR) n exposed (%) Median (IQR) n exposed (%)
Cumulative exposure to NNRTIs 9.4 (9.1,9.6) 115 (99.1) 9.3 (3.9,9.6) 606 (98.2) 9.3 (4.9,9.6) 721 (98.4)
Cumulative exposure to PIs 7.9 (3.0,8.1) 11 (9.5) 4.6 (2.2,6.5) 128 (20.8) 4.9 (2.2,6.7) 139 (19.0)
Cumulative exposure to ddI 7.1 (6.0,7.6) 6 (5.2) 6.0 (5.2,7.4) 22 (3.6) 6.2 (5.2,7.4) 28 (3.8)
Cumulative exposure to d4T 2.6 (2.1,3.1) 93 (80.2) 2.6 (2.0,3.0) 477 (77.3) 2.6 (2.0,3.1) 570 (77.8)
Cumulative exposure to AZT 6.7 (4.1,7.0) 109 (94.0) 6.3 (2.0,7.0) 575 (93.2) 6.4 (2.4,7.0) 684 (93.3)
Cumulative exposure to TDF 3.1 (1.1,6.9) 25 (21.6) 3.5 (1.8,6.0) 205 (33.2) 3.5 (1.7,6.1) 230 (31.4)
Cumulative exposure to EFV 9.4 (9.1,9.6) 97 (83.6) 9.2 (3.8,9.5) 495 (80.2) 9.2 (4.6,9.6) 592 (80.8)
Cumulative exposure to NVP 0.7 (0.1,7.6) 33 (28.5) 1.6 (0.1,8.4) 245 (39.7) 1.6 (0.1,8.4) 278 (37.9)
Cumulative exposure to LPV 8.1 (7.8,8.4) 9 (7.8) 5.8 (3.9,6.8) 93 (15.1) 6.0 (4.0,7.2) 102 (13.9)
Cumulative exposure to RTV 7.9 (3.0,8.4) 11 (9.5) 4.6 (2.2,6.5) 128 (20.8) 4.9 (2.2,6.7) 139 (19.0)
Cumulative exposure to ATV 0.8 (0.4,1.3) 2 (1.7) 1.5 (0.4,2.2) 40 (6.5) 1.4 (0.4,2.1) 42 (5.7)
Open in a new tab

Note: AIDS means prior AIDS-defining event, NRTI-nucleos(t)ide reverse transcriptase inhibitors, ART-antiretroviral therapy, BMI-body mass index, eGFR-estimated glomerular filtration rate, NNRTIs-non-nucleoside reverse transcriptase inhibitors, PI-protease inhibitors, TDF-tenofovir disoproxil fumarate, EFV-efavirenz, LPV-lopinavir, ATV-atazanavir, ddI-didanosine, d4T-stavudine, AZT-zidovudine, NVP-nevirapine, LPV-lopinavir, RTV-ritonavir, ABC-abacavir, 3TC-lamivudine, TRIG-triglycerides, HDL-high-density lipoprotein, LDL-low-density lipoprotein cholesterol, TCHOL-total cholesterol, mm/yy-month/year.

A total of 116 participants (15.8%) developed hypertension during 4671.3 person-years of follow-up, with an incidence rate of 24.8 per 1000 person-years (95% CI 20.7–29.8). The risk of developing hypertension increased with increasing follow-up (S1 Fig). Of the 116 participants with incident hypertension, 93 (80.2%) initiated antihypertensive therapy. Participants with obesity were 62% more likely to develop hypertension than those with normal body mass index (BMI) (aIRR 1.80, 95% CI 1.40–2.81), while the risk of hypertension was 1.80 higher in participants with eGFR <90 mL/min/1.73 m2 compared to those with normal renal function (aIRR 1.89, 95% CI 1.20–4.56). The risk of hypertension also increased by 12% (95% CI, 10%–25%) for each 5-year increase in age. There was no evidence to suggest that the association between BMI and hypertension differed according to sex (interaction P = 0.565) or age (interaction P = 0.496). In addition, participants with diastolic blood pressure >80 mmHg and/or systolic blood pressure >120 mmHg were more likely to develop hypertension during follow-up. There was no relationship between exposure to individual antiretroviral agents and the incidence of hypertension (Table 4).

Table 4. Factors associated with of incident hypertension in the ALT cohort (n = 733).

Variable Variable categories (Events/PYFU) Incidence Rate/1000 PY (95% CI) Crude IRR (95% CI) p-value Adjusted IRR (95% CI) p-value
Prior AIDS Yes 46/2263 20.3 (15.2,27.1) 0.70 (0.48,1.01) 0.060 0.93 (0.58,1.20) 0.579
No 70/2408.3 29.1 (23.0,36.7)  Ref   Ref  
NRTI TDF/3TC 30/1538.5 19.5 (13.6,27.9) Ref   Ref  
AZT/3TC 86/3103.6 27.7 (22.4,34.2) 1.42 (0.94,2.15) 0.098 1.24 (0.83,1.90) 0.340
ABC/3TC 0/29.2 0 -    -  
ART Class NNRTIs 103/3705.6 27.8 (22.7,33.4) 2.04 (1.14,3.63) 0.016 1.50(0.90,2.70) 0.150
PIs 13/962.6 13.5 (7.8,23.3)  Ref    Ref  
Age Per 5 years 116/4671.3 24.8 (20.7,29.8) 1.19 (1.07,1.32) 0.002 1.12 (1.10,1.25) 0.009
BMI (Kg/m2) <18.5 12/638.5 18.8 (10.7,33.1) 0.82 (0.44,1.51) 0.516 0.89(0.50,1.64) 0.670
18.5–24.9 66/2864.3 23.0 (18.1,29.3)  Ref    Ref  
25–29.9 22/888.4 24.8 (16.3,37.6) 1.07 (0.66,1.74) 0.770 0.87 (0.53,1.41) 0.567
>29.9 16/280 57.1 (35.0,93.3) 2.48 (1.44,4.28) 0.001 1.80 (1.40,2.81) 0.009
GFR (mL/min/1.73 m2) <90 8/193.1 41.4 (20.7,82.8) 1.56 (0.75,3.25) 0.236 1.89 (1.20,4.56) 0.017
≥90 65/2446.8 26.6 (20.8,33.9) Ref    Ref  
Missing 43/2031.4 21.2 (15.7,28.5) 0.80 (0.54,1.17) 0.248 0.70 (0.49,1.10) 0.112
Nadir CD4 (cells/μL) <200 9/411.6 21.9 (11.4,42.0) 0.69 (0.34,1.41) 0.310 0.66 (0.36,1.38) 0.257
200–349 33/1379 23.9 (17.0,33.7) 0.76 (0.48,1.18) 0.221 0.74 (0.48,1.17) 0.170
350–500 29/1460.2 19.9 (13.8,28.6) 0.63 (0.39,1.00) 0.050 0.80(0.51,1.26) 0.250
>500 45/1420.5 31.7 (23.7,42.4)  Ref   Ref  
Baseline Year 2014 52/1829.8 28.4 (21.7,37.3) 1.26 (0.88,1.82) 0.213 1.19 (0.80,1.77) 0.277
2015 64/2841.5 22.5 (17.6,28.8)  Ref    Ref  
Systolic blood pressure (mmHg) <120 17/2399.5 7.1 (4.4,11.4) Ref    Ref  
120–129 27/1289.1 20.9 (14.4,30.5) 2.96 (1.61,5.42) <0.001 2.34 (1.31,4.33) 0.006
≥130 72/982.7 73.3 (58.2,92.3) 10.34 (6.1,17.54) <0.001 5.43(3.20,9.41) <0.001
Diastolic blood pressure (mmHg) <80 43/3512.4 12.2 (9.1,16.5) Ref    Ref  
80–84 42/705.6 59.5 (44,80.5) 4.86 (3.18,7.44) <0.001 2.83 (1.82,4.44) <0.001
≥85 31/453.3 68.4 (48.1,97.2) 5.59 (3.52,8.86) <0.001 2.39 (1.43,4.10) <0.001
Prior exposure to PIs Never exposed 105/3787.7 27.7 (22.9,33.6) Ref   Ref  
Prior Exposure 11/883.6 12.4 (6.9,22.5) 0.45 (0.24,0.84) 0.012 0.79 (0.43,1.56) 0.256
Prior exposure to TDF Never exposed 91/3227.9 28.2 (23.0,34.6)  Ref   Ref  
Prior Exposure 25/1443.4 17.3 (11.7,25.6) 0.61 (0.39,0.96) 0.031 0.69 (0.56,1.38 0.200
Prior exposure to EFV Never exposed 98/3596.8 27.2 (22.4,33.2)  Ref    Ref  
Prior Exposure 18/1074.5 16.8 (10.6,26.6) 0.61 (0.37,1.02) 0.058 0.98 (0.57,1.58) 0.771
Prior exposure to LPV Never exposed 107/4020 26.6 (22.0,32.2)  Ref    Ref  
Prior Exposure 9/651.3 13.8 (7.2,26.6) 0.52 (0.26,1.03) 0.059 0.59(0.34,1.36) 0.276
Prior exposure to ATV Never exposed 114/4415.8 25.8 (21.5,31.0)  Ref   Ref
Prior Exposure 2/255.5 7.8 (2.0,31.3) 0.30 (0.07,1.23) 0.094 1.00(0.70, 1.41) 0.936
Open in a new tab

Note: * PYFU-person-years of follow-up; AIDS-prior AIDS-defining event; NRTI-nucleos(t)ide reverse transcriptase inhibitors; ART-antiretroviral therapy; BMI-body mass index; GFR-glomerular filtration rate; PI-protease inhibitors; TDF-tenofovir disoproxil fumarate; EFV-efavirenz; LPV-lopinavir; ATV-atazanavir; ABC-abacavir; 3TC-lamivudine. "Prior exposure to PIs" was collinear with "Prior exposure to LPV" (VIF = 6.9). However, these variables were insignificant in the final model (even when included one at a time), and they were dropped. The adjusted estimates for these variables were obtained by adding the variables back into the adjusted model, one at a time. Covariates sex, prior AIDS, hepatitis B infection status, smoking history, alcohol history, diabetes mellitus, prior exposure to didanosine (ddI), stavudine (d4t), zidovudine (AZT), nevirapine (NVP), non-nucleoside reverse transcriptase inhibitors (NNRTIs), baseline lipid values (triglycerides, high-density lipoprotein, low-density lipoprotein cholesterol and total cholesterol), HIV RNA, and baseline CD4 counts were all non-significant and were not confounders in the univariable analysis (P>0.2) and were not considered in the multivariable analysis. The significant covariates are indicated in bold.

Data completeness and sensitivity analysis

Overall, there was high data completeness on most variables at baseline except eGFR (411/970, 42.4%) and triglycerides (452/970, 46.6%). The regular Little’s MCAR test gives a χ2 distance of 42.86 with 32 degrees of freedom, P = 0.095. The null hypothesis for Little’s MCAR test is that the data are missing completely at random (MCAR). Therefore, we do not reject the null hypothesis, and the test provides evidence that the missing data in the six variables of interest are not MCAR under significance level 0.05. Since data were missing completely at random, there was no justification for using imputing missing data.

Discussion

This prospective cohort study is among the few to determine the prevalence and incidence of hypertension and its association with exposure to antiretrovirals in PLWH with long ART exposure in SSA. The prevalence of hypertension at enrolment was 24.4% and the factors associated with hypertension were older age, male sex, diabetes mellitus, obesity, and elevated total cholesterol levels. Compared with PLWH without hypertension at enrolment, the odds of prevalent hypertension were also higher in individuals with prior exposure to stavudine and nevirapine. The overall cumulative incidence was 15.8%, with an incidence rate of 24.8 per 1000 person-year of follow-up. Incident hypertension was associated with older age, renal insufficiency, obesity, and blood pressure levels higher than optimal, but not with exposure to antiretroviral drugs.

The incidence of hypertension in the present analysis is higher than that in a previous analysis in the same clinic, which reported a prevalence of 15.1% and an incidence of 19 cases per 1000 person-years eight years before our analysis [36]. However, participants in the present analysis are approximately 10 years older and with longer ART exposure. Increasing rates of hypertension in PLWH have been similarly reported in another cohort of people living with HIV in Uganda [37]. Together, these results suggest increasing rates of hypertension in PLWH in SSA, probably due to ageing and increases in other hypertension risk factors such as ART exposure. While direct comparisons between cohorts cannot be made, the presented incidence of hypertension is within the range of 19 and 54 cases per 1000 person-years, as reported by two studies in Uganda and South Africa, respectively [23, 36]. However, other cohorts in SSA have reported higher incidence rates than those in our analysis [22, 24]. The difference in hypertension rates between cohorts may reflect differences in cohorts as well as differences in hypertension screening. Therefore, HIV programs in SSA should integrate hypertension screening and management into existing ART delivery innovations, as integrated models of hypertension/HIV care have been demonstrated to be feasible even in public facilities in SSA [38, 39].

Our findings are also consistent with analyses of other cohorts in SSA that reported male sex, old age, low nadir CD4 count, and obesity as risk factors for hypertension [22]. Therefore, PLWH with these characteristics should be closely monitored. Previous exposure to nevirapine and stavudine has also been associated with risk or progression in several studies [13, 22, 40, 41]. Although contemporary ART regimens are increasingly available in most SSA countries, participants with prior exposure to these toxic agents should be targeted for hypertension and cardiovascular disease screening. The exact mechanism underlying hypertension in people with prior exposure to nevirapine or thymidine analogues is unclear but may be related to metabolic changes. In the AGEhIV cohort, the association between stavudine exposure and hypertension was attenuated after adjustment for lipodystrophy [13], suggesting that lipodystrophy may have an underlying role. The mechanism underlying the elevated risk of hypertension associated with nevirapine use remains unclear and should be investigated further.

Our study analysis several limitations. First, there is potential residual confounding because we could not measure and adjust for other hypertension risk factors, such as family history, medication use, unhealthy diet, and physical inactivity. In addition, data on laboratory parameters were only available at baseline and could not be modelled as time-updated covariates. Second, we cannot rule out unmeasured biases, as the analysis is based on an observational cohort. Furthermore, blood pressure was only measured annually, and this may lead to, and this may lead to inaccurate estimation of blood pressure incident dates. However, most hypertension guidelines recommend at least one annual blood pressure measurement per year. Third, we did not determine the incidence of cardiovascular disease, an important complication of uncontrolled hypertension. However, the detection and management of cardiovascular disease remains a challenge for many HIV programs in SSA countries [8, 42]. Fourth, missing data for some variables could have resulted in bias, although the proportion of participants with missing data was consistently small across the variables, and data was completely missing at random. Finally, in our analysis, we could not determine the hypertension risk posed by contemporary ART regimens, such as INSTIs, as these agents were recently introduced in Uganda; therefore, such an analysis would be underpowered. Future studies should investigate the risk of hypertension posed by contemporary regimens in SSA cohorts since there are reports that agents are differentially associated with a greater risk of weight gain in black people [43].

Despite these limitations, our results may inform the improvement of hypertension and HIV-care programs in SSA. With the increasing survival of PLWH and the increasing burden of cardiovascular disease, screening and managing cardiovascular risk factors, such as hypertension, will become essential for preventing cardiovascular deaths, especially in SSA [6]. In the analysed cohort, the rates of other cardiovascular risk factors, such as diabetes and smoking, were low, signifying that hypertension may play an even more critical role as a cardiovascular risk factor in SSA. The low smoking rates are consistent with reports from other SSA HIV cohorts in which smoking rates are generally low in PLWH, especially in women [44]. In addition, data from longitudinal cohorts on the relationship between hypertension and antiretroviral exposure in PLWH with long ART exposure in SSA are lacking. Our cohort enrolled 1000 participants who had been on ART for at least 10 years at baseline and were subsequently followed up for almost eight years. Therefore, this analysis contributes to the evidence for screening for hypertension and its risk factors in PLWH with prolonged exposure to ART in SSA. Importantly, highlight the importance of blood pressure screening, supports lipid and glucose testing, and screening for renal insufficiency in heavily treated PLWH. Furthermore, this analysis has several strengths. This cohort had high retention, as 97% of the participants had at least two follow-up visits, thus minimising attrition bias. In addition, we used data from people with prolonged ART exposure, who reflect the future HIV population. Finally, the analysis was based on routinely collected clinical data and may represent real-life urban settings in Uganda.

In conclusion, the prevalence and incidence of hypertension in this cohort of heavily treated PLWH was higher than those previously reported in the same HIV clinic. In addition, hypertension was associated with traditional risk factors and exposure to older toxic antiretroviral agents. Collectively, our results highlight the need to monitor hypertension in PLWH with prolonged exposure to ART and the need for HIV programs in SSA to integrate hypertension screening into existing HIV treatment programs.

Supporting information

S1 Fig. Incidence rate of hypertension with increasing follow-up time.

The rates of hypertension increased with increasing follow-up duration.

(TIF)

Click here for additional data file. (80.8KB, tif)

Acknowledgments

The authors declare that they have no conflicts of interest. We thank all study participants for contributing data and all staff for coordinating the study. DBM conceived the analysis, developed the analysis plan, conducted the analysis, and wrote the first draft of the manuscript under the supervision of BC, KP, ML, and MP. BC, JM, and RPR are investigators for the ALT cohort and supervised data curation. All the authors have read and approved the final manuscript.

Data Availability

Data cannot be shared publicly as it contains potentially identifying participant information. In addition, there are restrictions on data sharing imposed by the infectious Diseases Institute and the ethics review committees, in line with the Uganda Data Protection and Privacy Act 2019. Therefore, interested researchers should send data requests jointly to: Barbara Castelnuovo, Head of Department – Research, Infectious Diseases Institute (Mulago Hospital building), P.O. Box 22418, Kampala, Uganda, email: research@idi.co.ug and to The Chair, Joint Clinical Research Council Research Ethics Committee (JCRC REC) Lubowa Hill, Plot 101 Entebbe Road, P. O. Box 10005, Kampala, Uganda. Email: jcrc@jcrc.org.ug.

Funding Statement

The ALT cohort has been partly funded through a grant from the Johnson & Johnson corporate citizenship trust. Castelnuovo B is partly funded by the Fogarty International Centre, National Institute of Health (grant# 2D43TW009771-06 “HIV and co-infections in Uganda"). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

References

  • 1.Katz I, HIV BM-B. Improved life expectancy of people living with HIV: who is left behind? Lancet HIV. 2017;4: e324–e326. doi: 10.1016/S2352-3018(17)30065-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2.Katz IT, Maughan-Brown B. Improved life expectancy of people living with HIV: who is left behind? Lancet HIV. 2017;4: e324–e326. doi: 10.1016/S2352-3018(17)30086-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.Farahani M, Mulinder H, Farahani A, Marlink R. Prevalence and distribution of non-AIDS causes of death among HIV-infected individuals receiving antiretroviral therapy: a systematic review and meta-analysis. Int J STD AIDS. 2017;28: 636–650. doi: 10.1177/0956462416632428 [DOI] [PubMed] [Google Scholar]
  • 4.Jung IY, Rupasinghe D, Woolley I, O’Connor CC, Giles M, Azwa RISR, et al. Trends in mortality among ART-treated HIV-infected adults in the Asia-Pacific region between 1999 and 2017: results from the TREAT Asia HIV Observational Database (TAHOD) and Australian HIV Observational Database (AHOD) of IeDEA Asia-Pacific. J Int AIDS Soc. 2019;22. doi: 10.1002/jia2.25219 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5.Shah ASV, Stelzle D, Ken Lee K, Beck EJ, Alam S, Clifford S, et al. Global Burden of Atherosclerotic Cardiovascular Disease in People Living With HIV: Systematic Review and Meta-Analysis. Circulation. 2018;138: 1100–1112. doi: 10.1161/CIRCULATIONAHA.117.033369 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 6.Opie LH, Mayosi BM. Cardiovascular Disease in Sub-Saharan Africa. Circulation. 2005;112: 3536–3540. doi: 10.1161/CIRCULATIONAHA.105.597765 [DOI] [PubMed] [Google Scholar]
  • 7.Joint United Nations Programme on HIV/AIDS (UNAIDS). Global HIV & AIDS statistics-2022 fact sheet. 2022. Available: https://www.unaids.org/en/resources/fact-sheet [Google Scholar]
  • 8.Okello S, Amir A, Bloomfield GS, Kentoffio K, Lugobe HM, Reynolds Z, et al. Prevention of cardiovascular disease among people living with HIV in sub-Saharan Africa. Prog Cardiovasc Dis. 2020;63: 149–159. doi: 10.1016/j.pcad.2020.02.004 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Seaberg EC, Muñoz A, Lu M, Detels R, Margolick JB, Riddler SA, et al. Association between highly active antiretroviral therapy and hypertension in a large cohort of men followed from 1984 to 2003. AIDS. 2005. doi: 10.1097/01.aids.0000171410.76607.f8 [DOI] [PubMed] [Google Scholar]
  • 10.Nduka CU, Stranges S, Sarki AM, Kimani PK, Uthman OA. Evidence of increased blood pressure and hypertension risk among people living with HIV on antiretroviral therapy: A systematic review with meta-analysis. J Hum Hypertens. 2016;30: 355–362. doi: 10.1038/jhh.2015.97 [DOI] [PubMed] [Google Scholar]
  • 11.Nduka CU, Stranges S, Bloomfield GS, Kimani PK, Achinge G, Malu AO, et al. A plausible causal link between antiretroviral therapy and increased blood pressure in a sub-Saharan African setting: A propensity score-matched analysis. Int J Cardiol. 2016;220: 400–407. doi: 10.1016/j.ijcard.2016.06.210 [DOI] [PubMed] [Google Scholar]
  • 12.Bigna JJ, Ndoadoumgue AL, Nansseu JR, Tochie JN, Nyaga UF, Nkeck JR, et al. Global burden of hypertension among people living with HIV in the era of increased life expectancy: a systematic review and meta-analysis. J Hypertens. 2020;38: 1659–1668. doi: 10.1097/HJH.0000000000002446 [DOI] [PubMed] [Google Scholar]
  • 13.van Zoest RA, Wit FW, Kooij KW, van der Valk M, Schouten J, Kootstra NA, et al. Higher Prevalence of Hypertension in HIV-1-Infected Patients on Combination Antiretroviral Therapy Is Associated With Changes in Body Composition and Prior Stavudine Exposure. Clin Infect Dis. 2016. doi: 10.1093/cid/ciw285 [DOI] [PubMed] [Google Scholar]
  • 14.Hatleberg CI, Ryom L, d’Arminio Monforte A, Fontas E, Reiss P, Kirk O, et al. Association between exposure to antiretroviral drugs and the incidence of hypertension in HIV-positive persons: the Data Collection on Adverse Events of Anti-HIV Drugs (D:A:D) study. HIV Med. 2018;19: 605–618. doi: 10.1111/hiv.12639 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Haas AD, Keiser O, Balestre E, Brown S, Bissagnene E, Chimbetete C, et al. Monitoring and switching of first-line antiretroviral therapy in adult treatment cohorts in sub-Saharan Africa: collaborative analysis. Lancet HIV. 2015;2: e271–e278. doi: 10.1016/S2352-3018(15)00087-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16.Egger M, Ekouevi DK, Williams C, Lyamuya RE, Mukumbi H, Braitstein P, et al. Cohort Profile: The international epidemiological databases to evaluate AIDS (IeDEA) in sub-Saharan Africa. Int J Epidemiol. 2012;41: 1256–1264. doi: 10.1093/ije/dyr080 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17.Castelnuovo B, Mubiru F, Kiragga AN, Musomba R, Mbabazi O, Gonza P, et al. Antiretroviral treatment Long-Term (ALT) cohort: a prospective cohort of 10 years of ART-experienced patients in Uganda. BMJ. 2018;8: 15490. doi: 10.1136/bmjopen-2016-015490 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Mubiru F, Castelnuovo B, Reynolds SJ, Kiragga A, Tibakabikoba H, Owarwo NC, et al. Comparison of different cardiovascular risk tools used in HIV patient cohorts in sub-Saharan Africa; do we need to include laboratory tests? PLoS One. 2021;16. doi: 10.1371/journal.pone.0243552 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19.Unger T, Borghi C, Charchar F, Khan NA, Poulter NR, Prabhakaran D, et al. 2020 International Society of Hypertension Global Hypertension Practice Guidelines. Hypertension. 2020;75: 1334–1357. doi: 10.1161/HYPERTENSIONAHA.120.15026 [DOI] [PubMed] [Google Scholar]
  • 20.Uganda Ministry of Health. CONSOLIDATED GUIDELINES FOR THE PREVENTION AND TREATMENT OF HIV AND AIDS IN UGANDA. Kampala; 2018. Available: http://library.health.go.ug/publications/service-delivery-diseases-control-prevention-communicable-diseases/hivaids/consolidated
  • 21.World Health Organization. Consolidated guidelines on HIV prevention, testing, treatment, service delivery and monitoring: recommendations for a public health approach. Geneva; 2021. Available: https://www.who.int/publications/i/item/9789240031593 [PubMed]
  • 22.Okello S, Kanyesigye M, Muyindike WR, Annex BH, Hunt PW, Haneuse S, et al. Incidence and predictors of hypertension in adults with HIV-initiating antiretroviral therapy in south-western Uganda. J Hypertens. 2015. doi: 10.1097/HJH.0000000000000657 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Brennan AT, Jamieson L, Crowther NJ, Fox MP, George JA, Berry KM, et al. Prevalence, incidence, predictors, treatment, and control of hypertension among HIV-positive adults on antiretroviral treatment in public sector treatment programs in South Africa. PLoS One. 2018. doi: 10.1371/journal.pone.0204020 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24.RodrõÂguez-ArbolõÂ E, Mwamelo K, Kalinjuma AV, Furrer H, Hatz C, Tanner M, et al. Incidence and risk factors for hypertension among HIV patients in rural Tanzania-A prospective cohort study. PLoS One. 2017. doi: 10.1371/journal.pone.0172089 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Byonanebye DM, Polizzotto MN, Begovac J, Grabmeier-Pfistershammer K, Abela I, Castagna A, et al. Incidence of dyslipidemia in people with HIV who are treated with integrase inhibitors versus other antiretroviral agents. Aids. 2021;35: 869–882. doi: 10.1097/QAD.0000000000002811 [DOI] [PubMed] [Google Scholar]
  • 26.Ursenbach A, Max V, Maurel M, Bani-Sadr F, Gagneux-Brunon A, Garraffo R, et al. Incidence of diabetes in HIV-infected patients treated with first-line integrase strand transfer inhibitors: A French multicentre retrospective study. Journal of Antimicrobial Chemotherapy. 2020;75: 3344–3348. doi: 10.1093/jac/dkaa330 [DOI] [PubMed] [Google Scholar]
  • 27.Hellard ME, Chou R, Easterbrook P. WHO guidelines on testing for hepatitis B and C—meeting targets for testing. BMC Infect Dis. 2017;17. doi: 10.1186/s12879-017-2765-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28.Levey AS, Stevens LA, Schmid CH, Zhang Y, Castro AF, Feldman HI, et al. A new equation to estimate glomerular filtration rate. Ann Intern Med. 2009;150: 604–612. doi: 10.7326/0003-4819-150-9-200905050-00006 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29.Okello S, Ueda P, Kanyesigye M, Byaruhanga E, Kiyimba A, Amanyire G, et al. Association between HIV and blood pressure in adults and role of body weight as a mediator: Cross-sectional study in Uganda. J Clin Hypertens. 2017;19: 1181–1191. doi: 10.1111/JCH.13092 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30.Olack B, Wabwire-Mangen F, Smeeth L, Montgomery JM, Kiwanuka N, Breiman RF. Risk factors of hypertension among adults aged 35–64 years living in an urban slum Nairobi, Kenya. BMC Public Health. 2015;15: 1–9. doi: 10.1186/s12889-015-2610-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31.Weir CB, Jan A. BMI Classification Percentile And Cut Off Points. StatPearls. Treasure Island (FL); 2019. Available: http://www.ncbi.nlm.nih.gov/pubmed/31082114 [PubMed] [Google Scholar]
  • 32.Williams B, Mancia G, Spiering W, Rosei EA, Azizi M, Burnier M, et al. 2018 practice guidelines for the management of arterial hypertension of the European society of cardiology and the European society of hypertension ESC/ESH task force for the management of arterial hypertension. J Hypertens. 2018;36: 2284–2309. doi: 10.1097/HJH.0000000000001961 [DOI] [PubMed] [Google Scholar]
  • 33.Selik RM, Mokotoff ED, Branson B, Michele Owen S, Whitmore S, Irene Hall H. Revised surveillance case definition for HIV infection—United States, 2014. MMWR Recommendations and Reports. 2014;63: 1–10. [PubMed] [Google Scholar]
  • 34.National Institute of Health. NCEP Cholesterol Guidelines. [NCEP] National Cholesterol Education Program ATP III. 2001. doi: 10.1016/j.bbrc.2005.02.046 [DOI] [Google Scholar]
  • 35.Li C. Little’s test of missing completely at random. Stata Journal. 2013;13: 795–809. doi: 10.1177/1536867x1301300407 [DOI] [Google Scholar]
  • 36.Semeere AS, Sempa J, Lwanga I, Parkes-Ratanshi R, Kambugu A. Hypertension and associated risk factors in individuals infected with HIV on antiretroviral therapy at an urban HIV clinic in Uganda. Lancet Glob Health. 2014;2: S23. doi: 10.1016/s2214-109x(15)70045-8 [DOI] [Google Scholar]
  • 37.Kalyesubula R, Kayongo A, Semitala FC, Muhanguzi A, Katantazi N, Ayers D, et al. Trends and level of control of hypertension among adults attending an ambulatory HIV clinic in Kampala, Uganda: a retrospective study. BMJ Glob Health. 2016;1: e000055. doi: 10.1136/bmjgh-2016-000055 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 38.Muddu M, Tusubira AK, Sharma SK, Akiteng AR, Ssinabulya I, Schwartz JI. Integrated Hypertension and HIV Care Cascades in an HIV Treatment Program in Eastern Uganda: a retrospective cohort study. J Acquir Immune Defic Syndr. 2019;81: 552. doi: 10.1097/QAI.0000000000002067 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 39.Muddu M, Semitala FC, Kimera I, Mbuliro M, Ssennyonjo R, Kigozi SP, et al. Improved hypertension control at six months using an adapted WHO HEARTS-based implementation strategy at a large urban HIV clinic in Uganda. BMC Health Serv Res. 2022;22: 1–14. doi: 10.1186/S12913-022-08045-8/TABLES/5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40.Menezes CN, Maskew M, Sanne I, Crowther NJ, Raal FJ. A longitudinal study of stavudine-associated toxicities in a large cohort of South African HIV infected subjects. BMC Infect Dis. 2011;11: 1–10. doi: 10.1186/1471-2334-11-244/FIGURES/5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 41.Abrahams Z, Dave JA, Maartens G, Levitt NS. Changes in blood pressure, glucose levels, insulin secretion and anthropometry after long term exposure to antiretroviral therapy in South African women. AIDS Res Ther. 2015;12: 24. doi: 10.1186/s12981-015-0065-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 42.Peck R, Mghamba J, Vanobberghen F, Kavishe B, Rugarabamu V, Smeeth L, et al. Articles Preparedness of Tanzanian health facilities for outpatient primary care of hypertension and diabetes: a cross-sectional survey. 2014; 285. doi: 10.1016/S2214-109X(14)70033-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 43.Kanters S, Renaud F, Rangaraj A, Zhang K, Limbrick-Oldfield E, Hughes M, et al. Evidence synthesis evaluating body weight gain among people treating HIV with antiretroviral therapy—a systematic literature review and network meta-analysis. EClinicalMedicine. 2022;48: 101412. doi: 10.1016/j.eclinm.2022.101412 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 44.Murphy JD, Liu B, Parascandola M. Smoking and HIV in Sub-Saharan Africa: A 25-Country Analysis of the Demographic Health Surveys. Nicotine & Tobacco Research. 2019;21: 1093. doi: 10.1093/ntr/nty176 [DOI] [PMC free article] [PubMed] [Google Scholar]

Decision Letter 0

Giuseppe Vittorio De Socio

24 Nov 2022

PONE-D-22-29381Prevalence and incidence of hypertension in a heavily treatment-experienced cohort of people living with HIVPLOS ONE

Dear Dr. Byonanebye,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

The paper that you submitted to the PLOS ONE has been seen by three referees whose reports are listed below. As you will see, the referees raised several important criticisms which make the paper unacceptable for publication in its present form. This manuscript needs substantial improvement in terms of methodology, writing and sending the message. However, if you can deal with referees' comments and modify the paper according to their suggestions the Editorial Board may reconsider your work. Please submit your revised manuscript by January 08,2023. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

  • A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

  • A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

  • An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols.

We look forward to receiving your revised manuscript.

Kind regards,

Giuseppe Vittorio De Socio, MD, PhD

Academic Editor

PLOS ONE

Journal requirements:

When submitting your revision, we need you to address these additional requirements.

1.  Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at

https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf  and

https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2. We note that you have indicated that data from this study are available upon request. PLOS only allows data to be available upon request if there are legal or ethical restrictions on sharing data publicly. For more information on unacceptable data access restrictions, please see http://journals.plos.org/plosone/s/data-availability#loc-unacceptable-data-access-restrictions.

In your revised cover letter, please address the following prompts:

a) If there are ethical or legal restrictions on sharing a de-identified data set, please explain them in detail (e.g., data contain potentially sensitive information, data are owned by a third-party organization, etc.) and who has imposed them (e.g., an ethics committee). Please also provide contact information for a data access committee, ethics committee, or other institutional body to which data requests may be sent.

b) If there are no restrictions, please upload the minimal anonymized data set necessary to replicate your study findings as either Supporting Information files or to a stable, public repository and provide us with the relevant URLs, DOIs, or accession numbers. For a list of acceptable repositories, please see http://journals.plos.org/plosone/s/data-availability#loc-recommended-repositories.

We will update your Data Availability statement on your behalf to reflect the information you provide.

3. Please amend either the title on the online submission form (via Edit Submission) or the title in the manuscript so that they are identical.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Partly

Reviewer #2: Partly

Reviewer #3: No

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: No

Reviewer #2: No

Reviewer #3: No

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: No

Reviewer #2: No

Reviewer #3: No

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

Reviewer #3: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: General comments:

This manuscript endeavors to estimating the prevalence and incidence of hypertension in a cohort of PLWH from an HIV clinic in Kampala, Uganda.

The methods for this manuscript have a few holes or confusing pieces and I've tried to make comments below to improve the manuscript.

My biggest general comment is that I don't know why both prevalence and incidence are included. More specifically, I question the model looking at incidence. The methods suggest that the only difference between the two models is that people without hypertension at baseline were included from the incidence model. I wonder how well that baseline measurement characterizes those with hypertension. Was any past history information on hypertension included into the assessment? What are the random fluctuations and measurement error for SBP and DBP? These could impact the classification of baseline hypertension. If there are misclassifications, this could lead to people being included in the analyses that already have hypertension and lead to an overestimation of hypertension in the follow-up.

In addition, since the cohort does not appear to be systematically selected, I'm not entirely sure about some of the conclusions. For instance, age has an association with hypertension but I don't know if that's true or an artifact of the selection.

Specific comments:

1. (lines 123-132) I am sure I follow how you performed your variable selection here and have three comments about this section. First, stepwise variable selection procedures, which include backward selection, usually do not do a good job of finding the most appropriate model (e.g., https://doi.org/10.1002/sim.3943). Stepwise procedures and any p-value based selection have quite a bit of evidence suggesting that they are poor at selecting the appropriate variables. For a decent summary, see the link above. Second, Sun et al. (http://dx.doi.org/10.1016/0895-4356(96)00025-X) found that bivariate screening can miss a variable that may be a confounder even when a p-value is as high as what you have (0.25). Bivariate screening can be considered a form of stepwise variable selection, which usually do not do a good job of finding the most appropriate model (e.g., https://doi.org/10.1002/sim.3943). Third, I don't understand how AIC is involved the selection. You've already used stepwise procedures to select variables, so what are you comparing? Generally it's better to perform the entire selection based on more robust criteria, especially measures which assess the fit of the model (such as AIC or BIC where you could you Burnham and Anderson (doi: 10.1177/0049124104268644) as a guide) or, better yet, a shrinkage-based estimator such as lasso or lars.

2. (lines 133-137) First off, it is possible to test to see if missing data are missing completely at random (https://doi.org/10.2307/2290157). I would do that first and, if the data are MCAR, not worry about imputation. If the MCAR test fails, then I would promote using the MICE output above the complete case analyses. Finally, for the imputation model to work best, one really needs to include as much data as possible, i.e., include auxiliary data. If you have more data, I encourage you to include that here.

3. Is follow-up time included as an offset in the Poisson model? This is never explicitly stated in the methods so I wanted to check to make sure this was done. If this was not done, this must be done; otherwise you are not calculating IRRs.

4. (Table 2) Editorial note: it's hard to read your table with the variable names centered and no lines or breaks between the variable categories.

5. (Tables 2 and 4) Is the "global p" coming from a Wald type 3 test? I strongly encourage you to include more information on the methodology in the methods section or drop this altogether. If these are Wald type 3 tests, I will admit that I have always found the results from these confusing and I think your analyses here highlight some of those. For instance in Table 2, for most binary variables in the adjusted model, the p-value of the non-reference group and global p are identical but not for Sex. For alcohol history, the two p-values are pretty high (0.945 and 0.702), but the global p is far smaller (0.182). Though, in a roughly similar situation for Baseline LDL, the global p is larger than all the individual comparisons. From my experience, most people will pay attention to the individual categories and the global tests are going to create confusion.

6. (Tables 2,4) Instead of categorizing the continuous variables in the models (e.g., age, BMI, GFR), could you analyze them as continuous, maybe with a spline or other transformation? If you are really interested in characterizing the associations, that would provide more information than categories or assuming linearity (as was done with age in 5-year increments). The categorizations of these variables are not explained in the methods and, if the categorizations don't have clinical or policy relevance, they may lack utility.

Reviewer #2: This study sought to determine the prevalence and incidence of hypertension and its relationship with exposure to specific antiretroviral therapy in a Ugandan cohort of PLWH with long durations of ART.

Minor

1. Define what is meant by “heavily treatment experience”.

2. Multivariable not multivariate.

3. The authors use rate ratios and risk ratios interchangeability yet these differ (Line 127).

4. The interpretation of odds ratios as percentages is not correct. (e.g., odds of prevalent hypertension increase by 16%).

5. Define contemporary ART regimens. Line 76.

6. What was considered as prior exposure to antiretroviral therapy in the staged model building approach line 115.

7. Provide references for definitions and analysis endpoints for example diabetes mellitus, hepatitis B, and chronic kidney disease.

Major

1. Clarify why those with more than two follow-up visits were excluded for the analysis of prevalence.

2. Was adherence to antiretroviral therapy considered in the analysis? How would it affect the estimates?

3. How was Cumulative exposure to first line and second-line ART modeled

4. For the logistic regression model how were the covariates chosen?

5. It seems counterintuitive to remove untreated viral drugs that were not independently associated with hypertension (line 117 -119) yet the objective of this study is to find the association between antiretroviral drugs and hypertension.

6. What was the rationale for choosing the non-ART confounders listed in line 121 - 122?

7. Stepwise variable selection modeling is problematic in many ways including: it yields confidence intervals for effects and predicted values that are falsely narrow (Altman and Andersen, 1989) and biased regression coefficients that need shrinkage (see Tibshirani, 1996).

8. Line 135. What was the evidence for the assumption of MAR?

9. Tamper the discussion and conclusions given that the blood pressure measurements were annual with a high possibility of measurement error and visit-visit variability.

10. Rate ratios on their own cannot be compared across populations. Instead use standardized rates for comparison of incidence rates the current study with those in other populations.

Reviewer #3: 1- The authors wrote that “all variables in this analysis were fixed at baseline”. This rules out the possibility to evaluate time-varying phenomena, such as changes in eGFR. Please, comment on this.

2- The authors raised the issue of missing variables. The proportion of missing data is reported only for some, but not for all, clinical variables, e.g. alcohol consumption. From the analysis of Table 1, there is a huge variance for HR of some variables (e.g. alcohol consumption). Please, add the number of missing data for all the variables.

3- The authors wrote: “Multivariate logistic regression was used to determine the factors associated with hypertension at baseline”. Further information is needed regarding the list of variables included in the multivariate model.

In the legend for Table 2 the authors wrote: “Covariates sex, prior AIDS, hepatitis B infection status, smoking history, prior exposure to lopinavir (LPV), abacavir (ABC), efavirenz (EFV), didanosine (ddI), high-density lipoprotein, baseline year, and baseline and nadir CD4 counts were not significant in the univariate analysis (P>0.2) and were not considered in the multivariate analysis”.

However, (1) sex is significantly associated to HTN prevalence; this should be clarified. (2) Prior exposure to EFV was reported in the table. Please, better specify the criteria adopted to design Table 2.

The authors wrote that multicollinearity was checked and a VIF >5 was used as cut-off. Considering that the whole number of variables examinated in the univariate and subsequent mutlivariate model is 26, how many varables were excluded for multicollinearity?

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

Reviewer #3: No

**********

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

Attachment

Submitted filename: Review.docx

Click here for additional data file. (16.5KB, docx)
PLoS One. 2023 Feb 17;18(2):e0282001. doi: 10.1371/journal.pone.0282001.r002

Author response to Decision Letter 0

14 Dec 2022

REVIEWER # 1

General comments:

1. General comment 1: This manuscript endeavors to estimating the prevalence and incidence of hypertension in a cohort of PLWH from an HIV clinic in Kampala, Uganda. The methods for this manuscript have a few holes or confusing pieces and I've tried to make comments below to improve the manuscript. My biggest general comment is that I don't know why both prevalence and incidence are included. More specifically, I question the model looking at incidence. The methods suggest that the only difference between the two models is that people without hypertension at baseline were included from the incidence model. I wonder how well that baseline measurement characterises those with hypertension. Was any past history information on hypertension included into the assessment? What are the random fluctuations and measurement error for SBP and DBP? These could impact the classification of baseline hypertension. If there are misclassifications, this could lead to people being included in the analyses that already have hypertension and lead to an overestimation of hypertension in the follow-up.

Response: We thank the reviewer for these comments, and we are happy to provide detailed responses below.

First, with respect to why both prevalence and incidence are reported. We would like to clarify that is common for studies to determine both the prevalence and incidence of hypertension in cohorts and we provide evidence for this approach in the publications below. There are many publications and reports that report both prevalence and incidence and some of these publications have been in prestigious journals, including PLOS ONE.

• Brennan AT, Jamieson L, Crowther NJ, Fox MP, George JA, Berry KM, Stokes A, Maskew M, Sanne I, Long L, Cassim N, Rosen S. Prevalence, incidence, predictors, treatment, and control of hypertension among HIV-positive adults on antiretroviral treatment in public sector treatment programs in South Africa. PLoS One. 2018 Oct 3;13(10):e0204020. doi: 10.1371/journal.pone.0204020. PMID: 30281618; PMCID: PMC6169897.

• Lacruz ME, Kluttig A, Hartwig S, Löer M, Tiller D, Greiser KH, Werdan K, Haerting J. Prevalence and Incidence of Hypertension in the General Adult Population: Results of the CARLA-Cohort Study. Medicine (Baltimore). 2015 Jun;94(22):e952. doi: 10.1097/MD.0000000000000952. PMID: 26039136; PMCID: PMC4616348.

• Tu K, Chen Z, Lipscombe LL; Canadian Hypertension Education Program Outcomes Research Taskforce. Prevalence and incidence of hypertension from 1995 to 2005: a population-based study. CMAJ. 2008 May 20;178(11):1429-35. doi: 10.1503/cmaj.071283. PMID: 18490638; PMCID: PMC2374870.

• Sinnott SJ, Smeeth L, Williamson E, Douglas IJ. Trends for prevalence and incidence of resistant hypertension: population based cohort study in the UK 1995-2015. BMJ. 2017 Sep 22;358:j3984. doi: 10.1136/bmj.j3984. PMID: 28939590; PMCID: PMC5609092.

In addition, reporting prevalence and incidence is complimentary and epidemiologically important since it describes the current burden of disease (prevalence) and determines whether the burden of disease is increasing or not (incidence). This approach is important in several ways. First, the HIV population in Sub-Saharan Africa is dynamic and ageing; these data have also not been described before in this setting. For example, antiretroviral therapy is changing, and coverage is increasing (and this is stated in the introduction section (line 61). Therefore, in a dynamic population, determining prevalence alone would not provide enough data to characterise the burden of hypertension and inform programming and policy. Therefore, we hope the reviewer will appreciate the rationale for determining and reporting both the prevalence and incidence of hypertension in our study.

The second issue raised by the reviewer is "how well that baseline measurement characterises those with hypertension. Was any past history information on hypertension included into the assessment? What are the random fluctuations and measurement error for SBP and DBP?". We would like to clarify that data on prior hypertension diagnosis (and treatment), and blood pressure were available, and this was the basis for defining hypertension prevalence at baseline. Prevalent hypertension, which was determined at baseline, was defined as two consecutive systolic blood pressure (SBP) measures greater than 140 mmHg and/or diastolic blood pressure (DBP) greater than 90 mmHg and/or documented diagnosis and/or the initiation of antihypertensives within one year prior and up to one month after the baseline date. This clarification has been included in the methods section (lines 111-114).

Regarding random fluctuations and measurement errors for SBP and DBP, our definitions for both prevalent and incident hypertension are based on two consecutive abnormal blood pressure measurements (≥140/90 mmHg) or evidence of antihypertensive treatment. This approach minimises measurement bias and misclassification than if one reading was used. We agree that one BP measure would falsely classify patients with subtle increments in blood pressure due to non-hypertensive events such as anxiety syndromes as having hypertension. For this reason, the definition of hypertension is based on two consecutive readings. The definition is also consistent with international Hypertension diagnosis and management guidelines that recommend two consecutive abnormal blood pressure measurements [1].

Further, hypertension treatment and blood pressure measurement data are standardised according to the clinic standard operating procedures. Blood pressure measurement is done by qualified nurses, and calibrated blood pressure machines are used. We have included this information in the manuscript (lines 90-91). These study procedures minimise misclassification and measurement bias. Again, we have made this clearer in the manuscript. As indicated in our manuscript, this definition has been used in other analyses of HIV cohorts in SSA [2–4] and a prior analysis in the ALT cohort [5]. Again, we clarify that this definition is also consistent with international Hypertension diagnosis and management guidelines that recommend two consecutive abnormal blood pressure measurements[1].

2. General comment 2: In addition, since the cohort does not appear to be systematically selected, I'm not entirely sure about some of the conclusions. For instance, age has an association with hypertension, but I don't know if that's true or an artifact of the selection.

Response: We are unsure we understand what the reviewer means by saying that this cohort is "not systematically selected". However, to clarify further, we have stated that this cohort was established in 2014, and the inclusion criteria for the cohort are clearly stated (lines 81-84). We have also indicated that data is collected in accordance with the study protocol. The cohort is also registered with clinical trials.gov (lines 81-84). We also cited a paper that clearly defines the set-up of this cohort and its objectives (line 82). Specifically, we have clarified that "The analysis was conducted within the antiretroviral treatment long-term (ALT) cohort (ClinicalTrials.gov #: NCT02514707), which has previously been described [17,18]. Briefly, the ALT study is an ongoing single-centre prospective cohort of 1000 PLWH on ART for at least 10 years at baseline. Cohort participants were recruited from the Infectious Diseases Institute (IDI) HIV clinic in Kampala, Uganda, between 2014 and 2015.".

Regarding age, we are not aware of any literature that has disputed age as a risk factor for hypertension. Ageing is one of the strongest risk factors for hypertension [6], and the evidence dates several decades. Age is associated with atherosclerosis and arterial stiffening. The paper by McEniery et al, explains the relationship between ageing and cardiovascular changes [7]. Hypertension is a common complication of ageing, and we do not believe that the finding is an artefact or a chance finding. In the methods section (lines 137-138) we have clarified that variables considered in the analysis were carefully selected based on prior published literature linking them to hypertension.

Specific comments:

1. Specific comment 1: (lines 123-132) I am sure I follow how you performed your variable selection here and have three comments about this section. First, stepwise variable selection procedures, which include backward selection, usually do not do a good job of finding the most appropriate model (e.g., https://doi.org/10.1002/sim.3943). Stepwise procedures and any p-value based selection have quite a bit of evidence suggesting that they are poor at selecting the appropriate variables. For a decent summary, see the link above. Second, Sun et al. (http://dx.doi.org/10.1016/0895-4356(96)00025-X) found that bivariate screening can miss a variable that may be a confounder even when a p-value is as high as what you have (0.25). Bivariate screening can be considered a form of stepwise variable selection, which usually do not do a good job of finding the most appropriate model (e.g., https://doi.org/10.1002/sim.3943). Third, I don't understand how AIC is involved the selection. You've already used stepwise procedures to select variables, so what are you comparing? Generally, it's better to perform the entire selection based on more robust criteria, especially measures which assess the fit of the model (such as AIC or BIC where you could you Burnham and Anderson (doi: 10.1177/0049124104268644) as a guide) or, better yet, a shrinkage-based estimator such as lasso or lars.

Response: We thank the reviewer for providing the references above. As the reviewer may already know, variable selection and model fitting is a heavily debated issues in statistical analysis. The reviewer well describes the problems of automated algorithms for model fitting, and this is well reflected In the paper by Wiegand (https://doi.org/10.1002/sim.3943) that the reviewer refers to. First, we would like to clarify that we did not use stepwise algorithms to fit the reported models since we agree that this algorithm (Stata's stepwise command) is problematic and usually erroneous[8]. We fitted our regression models manually, carefully considering both significant variables and confounders. Stata's stepwise command was not used to fit the models. The paper by Sun et al. (http://dx.doi.org/10.1016/0895-4356(96)00025-X) that the author has suggested highlights the problem of using an automated selection of variables at the bivariable level. The paper criticises the use of a lower-level threshold for variable selection, and the example in the paper uses a threshold of p=0.05 . The authors of the paper correctly state that "If the statistical p value of a risk factor in the bivariable analysis is greater than an arbitrary value (often p = 0.05), then this factor will not be allowed to compete for inclusion in multivariable analysis.). We agree that this threshold would miss important confounders and other variables with borderline significance at bivariate analysis. It is important to note that the confounding variables in the paper had a p-value <0.25. The p-values for these variables would not have been dropped at the bivariate stage. Finally, the use of p-value<0.25 has been also used in several analyses, including a hypertension prediction model[9]. In their paper Chowdhury and Turin[9], make a case for bivariate analysis and variable selection. "It has also been suggested that variable selection should start with the univariate analysis of each variable. Variables that show significance (p<0.25) in the univariate analysis, as well as those that are clinically important, should be included for multivariate analysis.6 Nevertheless, the univariate analysis ignores the fact that individual variables that are weakly associated with the outcome can contribute significantly when they are combined. This issue can be solved partially by setting a higher significance level to allow more variables to illustrate significance in the univariate analysis.". We also agree that it is important to include "variables that are clinically important". This purposeful selection of the variables was our approach and consistent with published literature, including in a publication "Hosmer DW, Lemeshow S , Sturdivant RX . Applied logistic regression. New York: John Wiley & Sons, Incorporated, 2013".

Secondly, the aim of this analysis was to identify risk (or protective) factors that are causally related to hypertension (explanatory modelling) and not to find the combination of factors that best predicts a current diagnosis or future event of hypertension (predictive modelling). The argument for explanatory modelling is that variable selection helps determine the variables that are related to the outcome, and this makes the model complete and accurate. Second, it helps select the most parsimonious model by eliminating irrelevant variables that decrease the precision and increase the complexity of the model[10]. This approach also reduces the risk of over-adjustment and unnecessary estimates [11]. Nevertheless, we agree that confounders should not be dropped from the model even when they are non-significant. All variables dropped from the multivariate model were assessed to determine if they were confounders, in which case they were retained. Specifically, BMI and age have consistently been shown to be confounders for hypertension [24,29–32]; their confounding potential was checked, and the variables were consistently included in the model (lines 145-148). Therefore, our variable selection was purposeful, and we did not exclude any confounders from the adjusted model.

In addition, based on the criticism for the staged approach for model fitting, we have dropped this approach. We carefully fitted the model manually and the steps for model fitting are described (lines 134-148). Therefore, we fitted our model manually and variable selection was based on the significance of the variables and published data linking the variables to hypertension. To check the robustness of the final model, the goodness-of-fit test was used to determine the fitness of the final models (versus the full mode, i.e. with all variables), and the models with the smallest Akaike information criterion were considered the best-fitting [8]. We agree with the observation by Greenland [8] that "Unfortunately, all model-selection methods are subject to error, and no optimal method for selecting the best model form is known." Nevertheless, we carefully selected the final model and conducted appropriate model diagnostic procedures.

Finally, we would like to clarify that the variables considered in the multivariable model were based on published data that linked these variables to hypertension (lines 137-138). Variable selection based on knowledge is a common approach in epidemiological modelling [12].

We hope that the reviewers will find our clarifications helpful.

2. Specific comment 2: (lines 133-137) First off, it is possible to test to see if missing data are missing completely at random (https://doi.org/10.2307/2290157). I would do that first and, if the data are MCAR, not worry about imputation. If the MCAR test fails, then I would promote using the MICE output above the complete case analyses. Finally, for the imputation model to work best, one really needs to include as much data as possible, i.e., include auxiliary data. If you have more data, I encourage you to include that here.

Response: We agree entirely with this suggestion by the reviewer. We have conducted the regular Little's MCAR test (in Stata) test [13] (lines 159-160), and the results are provided under the sensitivity analysis section (lines 240-244). The regular Little's MCAR test gave a χ2 distance of 42.86 with 32 degrees of freedom, P=0.095. The null hypothesis for Little's MCAR test is that the data are missing completely at random (MCAR). Therefore, we do not reject the null hypothesis and we conclude that data are missing completely at random. Therefore, as suggested by the reviewer, we have removed the results based on multiple imputations as we have evidence that data were missing at random.

3. Specific comment 3: Is follow-up time included as an offset in the Poisson model? This is never explicitly stated in the methods, so I wanted to check to make sure this was done. If this was not done, this must be done; otherwise, you are not calculating IRRs.

Response: We agree with the reviewer that the standard approach for modelling rates using Poisson is to include the logarithm of person-years of follow-up in the model as an offset term to account for the observation time of individuals. This is exactly what we did, and we have made this clearer in the manuscript (see lines 129-131).

4. Specific comment 4: (Table 2) Editorial note: it's hard to read your table with the variable names centred and no lines or breaks between the variable categories.

Response: We agree with this suggestion and have revised the tables as suggested. We hope the tables are more readable (see the revised tables).

5. Specific comment 5: (Tables 2 and 4) Is the "global p" coming from a Wald type 3 test? I strongly encourage you to include more information on the methodology in the methods section or drop this altogether. If these are Wald type 3 tests, I will admit that I have always found the results from these confusing and I think your analyses here highlight some of those. For instance, in Table 2, for most binary variables in the adjusted model, the p-value of the non-reference group and global p are identical but not for Sex. For alcohol history, the two p-values are pretty high (0.945 and 0.702), but the global p is far smaller (0.182). Though, in a roughly similar situation for Baseline LDL, the global p is larger than all the individual comparisons. From my experience, most people will pay attention to the individual categories and the global tests are going to create confusion.

Response: We understand the preferences of the reviewer and the observation that the global p-values for categorical variables may be confusing and are often misunderstood. In line with the suggestion by the reviewer, we have dropped the global p-values from the tables altogether (See tables 2 and 4).

6. Specific comment 6: (Tables 2,4) Instead of categorising the continuous variables in the models (e.g., age, BMI, GFR), could you analyse them as continuous, maybe with a spline or other transformation? If you are really interested in characterising the associations, that would provide more information than categories or assuming linearity (as was done with age in 5-year increments). The categorisations of these variables are not explained in the methods and if the categorisations don't have clinical or policy relevance, they may lack utility.

Response: We agree with the reviewer that modelling variables as continuous variables is the preferred approach because assumptions of linearity and dose-response can be assessed. However, modelling continuous variables assumes linear associations, which are not valid for many clinical parameters [14]. For example, the associations between BMI and hypertension may not be linear in some populations [15].

We agree that categorising variables intuitively to give them clinical meaning produces data that is more useful to users. All variables (except age) were categorised to assign them categories that have clinical meaning. The justification has been provided, and the necessary references made. For example, the BMI categories follow the WHO categories; blood pressure categories are based on the European Association for cardiology categorisation for blood pressure. The CD4 categories are also based on WHO categories for immunosuppression status. We have added these justifications to the methods section. We have revised the methods section to make this clearer (see lines 138-144). We are hesitant to transform continuous variables (except age) as an interpretation of estimates becomes less straightforward and again, clinical decision-making (which is the basis of our analysis) rarely depends on continuous data but on categories with known risk. For example, it is more useful for a doctor (and more so for the low-level health cadres who provide HIV care in Africa) to know the risk of hypertension is twice higher if cholesterol is above 2 mmol/L rather than saying that the risk increases by 5% for every mmol/L increase in cholesterol. Please note that age increases hypertension was modelled as a transformed continuous variable because there is strong evidence supporting the linearity between age and hypertension risk [16].

 

Reviewer #2:

This study sought to determine the prevalence and incidence of hypertension and its relationship with exposure to specific antiretroviral therapy in a Ugandan cohort of PLWH with long durations of ART.

Minor comments

1. Minor comment 1: Define what is meant by "heavily treatment experience".

Response: We have clarified that this refers to participants with prolonged exposure to ART at baseline (>10 years) (see lines 84-86)

2. Minor comment 2: Multivariable not multivariate.

Response: This change has been made throughout the entire paper.

3. Minor comment 3: The authors use rate ratios and risk ratios interchangeability yet these differ (Line 127).

Response: We have addressed and consistently used incidence rate ratio (IRR)

4. Minor comment 4: The interpretation of odds ratios as percentages is not correct. (e.g., odds of prevalent hypertension increase by 16%).

Response: We have made the suggested revision.

5. Minor comment 5: Define contemporary ART regimens. Line 76.

Response: Revised: We have clarified that these refer to antiretroviral agents that are currently recommended for HIV treatment (i.e., modern ART regimens) see line 76-77.

6. Minor comment 6: What was considered as prior exposure to antiretroviral therapy in the staged model building approach line 115.

Response: Prior exposure means any exposure (use of) to antiretroviral before baseline and this has been clarified (line 137)

7. Minor comment 7: Provide references for definitions and analysis endpoints for example diabetes mellitus, hepatitis B, and chronic kidney disease.

Response: References have been added (lines 115-120)

Major comments

1. Major comment 1: Clarify why those with more than two follow-up visits were excluded for the analysis of prevalence.

Response: We did not exclude participants with more than two follow-up visits as suggested by the reviewer but rather excluded participants with fewer than two follow-up visits (lines 102 and 171). The justification for this approach has been provided in the manuscript and is based on the definitions of our endpoint (hypertension). The definition of hypertension requires at least two measurements for hypertension diagnosis (refer to our definitions of endpoints and justification for two measures in the manuscript (lines 107-114). As noted in other responses above (reviewer 1), the definition is consistent with international guidelines for the diagnosis and management of hypertension[1].

2. Major comment 2: Was adherence to antiretroviral therapy considered in the analysis? How would it affect the estimates?

Response: We thank the reviewer for this question. Our cohort is comprised of highly treatment-experienced people living with HIV (on ART for at least ten years at baseline). Data on adherence data spanning ten years is not available in most cohorts, including ours. Therefore, we think modelling adherence at baseline would not be meaningful since it may not reflect adherence over the 10 years of antiretroviral therapy. Secondly, our focus was to determine if prior exposure (qualitatively) to antiretroviral agents was associated with higher rates of hypertension. However, we have modelled viral load which is a direct indicator of adherence. Lastly, we are unaware of any data showing a plausible causal relationship between adherence to ART and hypertension risk. Adjusting for both adherence and drug exposure would be problematic and lead to over-adjustment bias or unnecessary adjustment[11]. Finally, and possibly due to the reasons above, adherence to antiretroviral has not traditionally been included in models for cardiovascular diseases [17,18].

3. Major comment 3: How was Cumulative exposure to first-line and second-line ART modelled

Response: We modelled prior exposure and not cumulative exposure. We only provided data on cumulative exposure to antiretrovirals (table 1) as this data may have other program implications (e.g., ART resistance). Cumulative exposure has not been found to be a good predictor of cardiovascular events and this was demonstrated for abacavir in several cohorts [19,20]. Those studies reported that recent or current exposure, but not cumulative exposure to abacvir, was associated with cardiovascular risk.

4. Major comment 4: For the logistic regression model how were the covariates chosen?

Response: We have clarified that for both logistic and Poisson regressions, the variable selection was based on published data on risk factors for hypertension in HIV cohorts [4,21–24] (lines 137-138). The procedures for model fitting and selection (i.e., variable selection for inclusion in the model) have also been provided in the methods (lines 144-150). Specifically, the regression models were manually fitted based on regression diagnostic procedures (AIC) and goodness of fit. In addition, during model fitting, care was taken to ensure that confounders were included in the final adjusted model.

5. Major comment 5: It seems counterintuitive to remove untreated viral drugs that were not independently associated with hypertension (lines 117 -119), yet the objective of this study is to find the association between antiretroviral drugs and hypertension.

Response: We do not clearly understand what the reviewer refers to as "untreated viral drugs". However, we would like to inform the reviewer that we have revised the analysis based on concerns regarding the staged approach for model fitting (that we had used). This way, variables were only dropped if they were non-significant at the bivariable analysis or if they were insignificant in the adjusted model. We note further that if exposure to an antiretroviral drug was a confounder that variable was included in the adjusted model, even if it was not significant (see lines 145-146)

6. Major comment 6: What was the rationale for choosing the non-ART confounders listed in line 121 - 122?

Response: The variable selection was based on published data on risk factors for hypertension in HIV cohorts [4,21–24] and we have made this clear in the methods (lines 137-138).

7. Major comment 7: Stepwise variable selection modelling is problematic in many ways including: it yields confidence intervals for effects and predicted values that are falsely narrow (Altman and Andersen, 1989) and biased regression coefficients that need shrinkage (see Tibshirani, 1996).

Response: As discussed above (response to reviewer 1), we would like to clarify that we did not use stepwise algorithms to fit the reported models because we also agree that this method of model fitting is problematic and usually erroneous[8]. Automated stepwise algorithms (e.g. Stata's stepwise command) were not used to fit the models. We have dropped this approach based on the criticism of the staged approach for model fitting. We carefully fitted the model manually, and the steps for model fitting are described (lines 134-138). The criticism of the stepwise approach is that it often drops confounders that may be non-significant[25]. We avoided this by fitting our models manually, and all variables dropped were assessed to determine if they were confounders, in which case they were retained (see lines 145-148).

As the reviewer may already know, variable selection and model fitting is a heavily debated issues in statistical analysis. The aim of this analysis was to identify risk (or protective) factors that are causally related to hypertension (explanatory modelling) and not to find the combination of factors that best predicts a current diagnosis or future event of hypertension (predictive modelling). Variable selection helps determine the variables that are related to the outcome, and this makes the model complete and accurate. Second, it helps select the most parsimonious model by eliminating irrelevant variables that decrease the precision and increase the complexity of the model[10]. While we agree with the observation by Greenland [8] that "Unfortunately, all model-selection methods are subject to error, and no optimal method for selecting the best model form is known.", there is an increasing suggestion that the purposeful selection of variables (like we did) is perhaps a good approach [26]. Our variable selection was based on the significance of the variables and with considerations made for confounding variables.

We hope the reviewer will find our clarifications helpful.

8. Major comment 8: Line 135. What was the evidence for the assumption of MAR?

Response: We have conducted the regular Little's MCAR test (in Stata) test [13] (lines 159-160), and the results are provided under the sensitivity analysis section (lines 240-244). The regular Little's MCAR test gave a χ2 distance of 42.86 with 32 degrees of freedom, P=0.095. The null hypothesis for Little's MCAR test is that the data are missing completely at random (MCAR). Therefore, we do not reject the null hypothesis and we conclude that data are missing completely at random. Thus, as suggested by the reviewer, we have removed the results based on multiple imputations as we have evidence that data were missing at random.

9. Major comment 9: Tamper the discussion and conclusions given that the blood pressure measurements were annual with a high possibility of measurement error and visit-visit variability.

Response: We have included as a limitation that visits were annual, and therefore the precise estimation of hypertension incidence date may not be accurate (lines 281-282). However, as indicated in our methods, we have noted that blood pressure measurements are standardised and consistent with international guidelines. This approach minimises measurement bias. Regarding variability in blood pressure, we have noted in the methods that blood measurement is standardised in the clinics, and that blood pressure machines are well calibrated (lines 90-91). This minimises measurement bias. We are not aware of any data that suggests blood pressure measurements significantly fluctuate during the day. Available data suggest that blood pressure may dip during the night [27], but nocturnal blood pressure measurement is not the standard practice for hypertension screening. All study clinics are conducted during the day. What is clear is that blood pressure increases with ageing, and our results support that assertion.

10. Major comment 10: Rate ratios on their own cannot be compared across populations. Instead, use standardised rates for comparison of incidence rates in the current study with those in other populations.

Response: We agree entirely with this observation and have added a caveat that "direct comparisons with other cohorts cannot be made" (see lines 259-261). Most studies on hypertension do not often report standardised rates and therefore direct comparison would still not be possible even if we made these estimates.

 

Reviewer #3:

1. Comment 1: The authors wrote that "all variables in this analysis were fixed at baseline". This rules out the possibility to evaluate time-varying phenomena, such as changes in eGFR. Please, comment on this.

Response: We agree with the reviewer that time-updated variables would be more intuitive and would allow the testing of whether some of these factors mediate the effect between some exposures and hypertension. However, time-updated analysis for some variables calls for caution since these may lie on the causal path between the exposure variable and hypertension. For example, there is strong evidence linking the use of protease inhibitors and chronic kidney disease [28]. Therefore time-updated eGFR would potentially attenuate or even neutralise the effect we want to see since low-eGFR may be the causal link between PIs and hypertension risk. This adjustment would lead to "over-adjustment bias" [11] and would potentially underestimate the effects of antiretroviral exposure. However, modelling time-updated variables. This pathway may also be true for several variables, including obesity, and diabetes. In addition, due to HIV funding mechanisms in Africa, funding for routine laboratory measures is not currently available. In Uganda, the US President's Emergency Plan for AIDS Relief (PEPFAR) funds only viral load testing. Therefore, most patients do not have repeated measures even when they are supposed to have annual tests. However, a laboratory panel of tests is frequently run at baseline for all patients. Therefore, modelling in African cohorts tends to use baseline data. This is important if one intends to produce reproducible data that is based on programmatic practices.

Despite our arguments above, we agree that follow-up data on variables is important in causal modelling. We have noted the missing prospective data as a limitation in the discussion (line 280-281)

2. Comment 2: The authors raised the issue of missing variables. The proportion of missing data is reported only for some, but not for all, clinical variables, e.g., alcohol consumption. From the analysis of Table 1, there is a huge variance for HR of some variables (e.g., alcohol consumption). Please, add the number of missing data for all the variables.

Response: In Table 1, we have provided data on the missingness of data for all variables. Data on alcohol use was complete, and there was no missing variable on this variable (see table 1). Regarding the variance in the estimates for alcohol, this is not due to missing data as we had data on alcohol use for all included participants. The high variance is probably due to the few number of participants who had never used alcohol (n=16, 1.7%), and this is probably making estimates less precise.

3. Comment 3: The authors wrote: "Multivariate logistic regression was used to determine the factors associated with hypertension at baseline". Further information is needed regarding the list of variables included in the multivariate model.

Response: We have listed and justified the factors considered for multivariable regression (lines 134-137).

4. Comment 4: In the legend for Table 2, the authors wrote: "Covariates sex, prior AIDS, hepatitis B infection status, smoking history, prior exposure to lopinavir (LPV), abacavir (ABC), efavirenz (EFV), didanosine (ddI), high-density lipoprotein, baseline year, and baseline and nadir CD4 counts were not significant in the univariate analysis (P>0.2) and were not considered in the multivariate analysis". However, (1) sex is significantly associated to HTN prevalence; this should be clarified. (2) Prior exposure to EFV was reported in the table. Please, better specify the criteria adopted to design Table 2.

Response: We thank the reviewer for identifying this clerical error. Sex was indeed significant and was included in the final multivariate model. We have revised this (line 201). We have also revised the error regarding prior exposure to EFV.

5. Comment 5 The authors wrote that multicollinearity was checked and a VIF >5 was used as cut-off. Considering that the whole number of variables examined in the univariate and subsequent multivariate model is 26, how many variables were excluded for multicollinearity?

Response: "Prior exposure to PIs" was collinear with "Prior exposure to LPV" (VIF=6.9). However, these variables were not significant in the final model (even when included one at a time) and they were dropped from the final multivariate model. The adjusted estimates for these variables were obtained by adding the variables back into the adjusted model, one at a time. We did not find any evidence for multicollinearity between the variables in the final model (lines 231-233). Please also note that not all variables considered for multivariable adjustment were included in the final multivariable (adjusted model).

 

References

1 Unger T, Borghi C, Charchar F, Khan NA, Poulter NR, Prabhakaran D, et al. 2020 International Society of Hypertension Global Hypertension Practice Guidelines. Hypertension 2020; 75:1334–1357.

2 Okello S, Kanyesigye M, Muyindike WR, Annex BH, Hunt PW, Haneuse S, et al. Incidence and predictors of hypertension in adults with HIV-initiating antiretroviral therapy in south-western Uganda. J Hypertens Published Online First: 2015. doi:10.1097/HJH.0000000000000657

3 Brennan AT, Jamieson L, Crowther NJ, Fox MP, George JA, Berry KM, et al. Prevalence, incidence, predictors, treatment, and control of hypertension among HIV-positive adults on antiretroviral treatment in public sector treatment programs in South Africa. PLoS One Published Online First: 2018. doi:10.1371/journal.pone.0204020

4 RodrõÂguez-ArbolõÂ E, Mwamelo K, Kalinjuma AV, Furrer H, Hatz C, Tanner M, et al. Incidence and risk factors for hypertension among HIV patients in rural Tanzania-A prospective cohort study. PLoS One Published Online First: 2017. doi:10.1371/journal.pone.0172089

5 Mubiru F, Castelnuovo B, Reynolds SJ, Kiragga A, Tibakabikoba H, Owarwo NC, et al. Comparison of different cardiovascular risk tools used in HIV patient cohorts in sub-Saharan Africa; do we need to include laboratory tests? PLoS One 2021; 16. doi:10.1371/JOURNAL.PONE.0243552

6 Buford TW. Hypertension and aging. Ageing Res Rev 2016; 26:96–111.

7 McEniery CM, Wilkinson IB, Avolio AP. Age, hypertension and arterial function. Clin Exp Pharmacol Physiol 2007; 34:665–671.

8 Greenland S. Modeling and variable selection in epidemiologic analysis. Am J Public Health 1989; 79:340–349.

9 Chowdhury MZI, Turin TC. Variable selection strategies and its importance in clinical prediction modelling. Fam Med Community Health 2020; 8:e000262.

10 Chowdhury MZI, Turin TC. Variable selection strategies and its importance in clinical prediction modelling. Fam Med Community Health 2020; 8:e000262.

11 Lu H, Cole SR, Platt RW, Schisterman EF. Revisiting Overadjustment Bias. Epidemiology 2021; 32:E22–E23.

12 Walter S, Tiemeier H. Variable selection: Current practice in epidemiological studies. Eur J Epidemiol 2009; 24:733–736.

13 Li C. Little's test of missing completely at random. Stata Journal 2013; 13:795–809.

14 Dudenbostel T, Oparil S. J Curve in Hypertension. Curr Cardiovasc Risk Rep 2012; 6:281–290.

15 Tesfaye F, Nawi NG, van Minh H, Byass P, Berhane Y, Bonita R, et al. Association between body mass index and blood pressure across three populations in Africa and Asia. Journal of Human Hypertension 2007 21:1 2006; 21:28–37.

16 Lei X, Sun X, Strauss J, Zhao Y, Yang G, Hu P, et al. Health outcomes and socio-economic status among the mid-aged and elderly in China: Evidence from the CHARLS national baseline data. J Econ Ageing 2014; 4:59–73.

17 Bozzette SA, Ake CF, Tam HK, Chang SW, Louis TA, Diego S. Cardiovascular and Cerebrovascular Events in Patients Treated for Human Immunodeficiency Virus Infection. https://doi.org/101056/NEJMoa022048 2003; 348:702–710.

18 Ryom L, Lundgren JD, El-Sadr W, Reiss P, Kirk O, Law M, et al. Cardiovascular disease and use of contemporary protease inhibitors: the D: A: D international prospective multicohort study. Lancet HIV 2018; 5:e291–e300.

19 Elion RA, Althoff KN, Zhang J, Moore RD, Gange SJ, Kitahata MM, et al. Recent abacavir use increases risk for Types 1 and 2 myocardial infarctions among adults with HIV. J Acquir Immune Defic Syndr 2018; 78:62.

20 Dorjee K, Choden T, Baxi SM, Steinmaus C, Reingold AL. Risk of cardiovascular disease associated with exposure to abacavir among individuals with HIV: A systematic review and meta-analyses of results from 17 epidemiologic studies. Int J Antimicrob Agents 2018; 52:541–553.

21 Okello S, Kanyesigye M, Muyindike WR, Annex BH, Hunt PW, Haneuse S, et al. Incidence and Predictors of Hypertension in Adults with HIV Initiating Antiretroviral Therapy in Southwestern Uganda. J Hypertens 2015; 33:2039.

22 Okello S, Ueda P, Kanyesigye M, Byaruhanga E, Kiyimba A, Amanyire G, et al. Association between HIV and blood pressure in adults and role of body weight as a mediator: Cross-sectional study in Uganda. J Clin Hypertens 2017; 19:1181–1191.

23 Olack B, Wabwire-Mangen F, Smeeth L, Montgomery JM, Kiwanuka N, Breiman RF. Risk factors of hypertension among adults aged 35-64 years living in an urban slum Nairobi, Kenya. BMC Public Health 2015; 15:1–9.

24 Hatleberg CI, Ryom L, d'Arminio Monforte A, Fontas E, Reiss P, Kirk O, et al. Association between exposure to antiretroviral drugs and the incidence of hypertension in HIV-positive persons: the Data Collection on Adverse Events of Anti-HIV Drugs (D:A:D) study. HIV Med 2018; 19:605–618.

25 Sun GW, Shook TL, Kay GL. Inappropriate use of bivariable analysis to screen risk factors for use in multivariable analysis. J Clin Epidemiol 1996; 49:907–916.

26 Bursac Z, Gauss CH, Williams DK, Hosmer DW. Purposeful selection of variables in logistic regression. Source Code Biol Med 2008; 3:17.

27 Sherwood A, Steffen PR, Blumenthal JA, Kuhn C, Hinderliter AL. Nighttime blood pressure dipping: The role of the sympathetic nervous system. Am J Hypertens 2002; 15:111–118.

28 Ryom L, Dilling Lundgren J, Reiss P, Kirk O, Law M, Ross M, et al. Use of Contemporary Protease Inhibitors and Risk of Incident Chronic Kidney Disease in Persons with Human Immunodeficiency Virus: The Data Collection on Adverse Events of Anti-HIV Drugs (D:A:D) Study. Journal of Infectious Diseases Published Online First: 2019. doi:10.1093/infdis/jiz369

Attachment

Submitted filename: Response to Reviewers.docx

Click here for additional data file. (217.6KB, docx)

Decision Letter 1

Giuseppe Vittorio De Socio

16 Jan 2023

PONE-D-22-29381R1Prevalence and incidence of hypertension in a heavily treatment-experienced cohort of people living with HIV in UgandaPLOS ONE

Dear Dr. Byonanebye,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

Please submit your revised manuscript by Mar 02 2023 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

  • A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

  • A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

  • An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols.

We look forward to receiving your revised manuscript.

Kind regards,

Giuseppe Vittorio De Socio, MD, PhD

Academic Editor

PLOS ONE

Journal Requirements:

Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #1: (No Response)

Reviewer #3: All comments have been addressed

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #3: (No Response)

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #3: (No Response)

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: No

Reviewer #3: (No Response)

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #3: (No Response)

**********

6. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: Thank you very much for your detailed explanations of your choices and adding clarifications that unfortunately I missed when I read this through the first time. I still have some disagreements on a few comments that I will list below, at least partially because of poor phrasing in my initial review and the manuscript.

1. As you have stated, you are exploring causality. That means you are relying on the sampling and the regression models to provide you with unbiased estimates of the causal effects. My "not systematically selected" comment was poorly phrased, but this is what I had in mind. Regression can do ok at providing unbiased estimates of causal effects in observational studies, but it's hard to know when regression does a good job without trying out other causal designs (e.g., propensity scores, disease risk scores, etc.). Thus, if your main goal is causality, it would seem to me to make sense to me to try one of these designs.

2. In that vein, I have a hard time with this statement in your response: "…the aim of this analysis was to identify risk (or protective) factors that are causally related to hypertension…". I don't see how you can do that without having balanced groups at baseline. That's why I brought up age. I'm not disputing that age is a risk factor. I mentioned age as an example because, if the age distribution is different between the hypertension and non-hypertension groups at baseline, that could cause bias. Currently you are relying on the regression model to remove that bias, which it may or may not be able to do. In table 1, the median age is ~2.5 years higher in the hypertensive group. Unfortunately, I'm not knowledgeable enough in this area to know if that is that enough to make a difference. It might be nothing, but this is where having some sort of balancing procedure might be useful. I'm also not able to know whether the regression model is able to remove any confounding that may have (in this case, I suspect the regression model is based on the info in table 1 since there is good overlap).

3. Also, if you are not concerned with "prediction", I suggest not using the term "predictors" in the abstract, the title of Table 4, and in the discussion. Again, maybe it's just me, but this is where some of my dissonance arises.

4. Thank you for your detailed explanation of your variable selection. I agree with your general thesis that the variable selection literature is a mess, especially in epidemiology. If one goes back far enough, it is possible to find any reference to support any claim. Unfortunately, I still see what you are doing as a form of sequential variable selection based on this statement from the review: "Therefore, we fitted our model manually and variable selection was based on the significance of the variables and published data linking the variables to hypertension" and lines 143-145. You may not be using a specific algorithm and you may not be using "stepwise" where variables can exit and enter the model, but you are still using the p-value (which I presume is what you mean by "significance") to select variables in a sequential manner. From what I have seen from simulation studies, using p-values for selection with sequential model fitting does poorly at capturing the true effect unless the sample size is really large. That said, it's hard for me to assess incorporating the change-in-estimate and AIC into the selection procedure. I don't know conditional on the p-value based selection whether those reduce confounding and, if so, by how much. If we are working in an 'anything goes so long as you explain it' situation regarding variable selection, then I think what you have is permissable. But, I am not really buying p-value based selection.

Reviewer #3: (No Response)

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #3: No

**********

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

PLoS One. 2023 Feb 17;18(2):e0282001. doi: 10.1371/journal.pone.0282001.r004

Author response to Decision Letter 1

24 Jan 2023

REVIEWER # 1

General comments:

1. General comment 1: Thank you very much for your detailed explanations of your choices and adding clarifications that unfortunately I missed when I read this through the first time. I still have some disagreements on a few comments that I will list below, at least partially because of poor phrasing in my initial review and the manuscript.

Response: We thank the reviewer for carefully reviewing our responses and the revised manuscript. I particularly thank the reviewer for the scientific propositions and stimulating arguments. In our responses below, we address the specific issues raised by the reviewer.

2. Specific comment 1: As you have stated, you are exploring causality. That means you are relying on the sampling and the regression models to provide you with unbiased estimates of the causal effects. My "not systematically selected" comment was poorly phrased, but this is what I had in mind. Regression can do ok at providing unbiased estimates of causal effects in observational studies, but it's hard to know when regression does a good job without trying out other causal designs (e.g., propensity scores, disease risk scores, etc.). Thus, if your main goal is causality, it would seem to me to make sense to me to try one of these designs.

Response: We understand the concern of the reviewer (see comment 3) is that we have used the term predictor, which in their view, suggests a causal relationship. We have made revisions to the manuscript and replaced "predictor" with factors associated". We understand the reviewer suggests that if we want to maintain the term "predictor," we should explore other statistical methods (e.g., propensity scores) as these, in their view, would better prove a causal relationship. Although propensity scores are increasingly used in statistical analyses, in our opinion, we do not agree that propensity scores are always superior to multivariate regressionIn a metanalysis of 43 observational studies Shah et al., demonstrated thatpropensity score methods give similar results to the conventional regression methods in observational studies [1]. In addition, Cepeda et al. demonstrated that the performance of propensity scores in large datasets with more than seven events per variable is similar to traditional multivariable regression methods [2] Therefore, we have maintained multivariate regression. However, we agree with the reviewer that the use of the term "predictors" may be strong in this case. To this end, we have revised the paper and dropped the use of "predictor" and used "factors associated".

3. Specific comment 2: In that vein, I have a hard time with this statement in your response: "…the aim of this analysis was to identify risk (or protective) factors that are causally related to hypertension…". I don't see how you can do that without having balanced groups at baseline. That's why I brought up age. I'm not disputing that age is a risk factor. I mentioned age as an example because, if the age distribution is different between the hypertension and non-hypertension groups at baseline, that could cause bias. Currently you are relying on the regression model to remove that bias, which it may or may not be able to do. In table 1, the median age is ~2.5 years higher in the hypertensive group. Unfortunately, I'm not knowledgeable enough in this area to know if that is that enough to make a difference. It might be nothing, but this is where having some sort of balancing procedure might be useful. I'm also not able to know whether the regression model is able to remove any confounding that may have (in this case, I suspect the regression model is based on the info in table 1 since there is good overlap).

Response: We thank the reviewer for this comment. We have addressed some of the issues in this comment above. As we understand age, the reviewer's concerns are that there are baseline differences between participants who developed hypertension and those who did not. The reviewer points out (as an example) a 2.5-year difference in median age and suggests that some balancing procedure should be done. We do not agree with this suggestion. We aim to determine whether these differences are associated with a higher risk of hypertension. Balancing (e.g., matching participants by age) would remove the effect of age, which we would like to show. If matching is done for all known predictors of hypertension, then the analysis will likely not show any predictors, which is our analysis goal. As argued above we do not think that traditional multivariate regression methods are inferior to propensity scores, especially if the sample size is not small as in our case.

Specific comment 3: Also, if you are not concerned with "prediction", I suggest not using the term "predictors" in the abstract, the title of Table 4, and in the discussion. Again, maybe it's just me, but this is where some of my dissonance arises.

Response: We understand the genuine concern of the reviewer. We have revised the entire manuscript and dropped the use of "predictor" and used "factors associated". We hope the reviewer will find this change appropriate.

Specific comment 4: Thank you for your detailed explanation of your variable selection. I agree with your general thesis that the variable selection literature is a mess, especially in epidemiology. If one goes back far enough, it is possible to find any reference to support any claim. Unfortunately, I still see what you are doing as a form of sequential variable selection based on this statement from the review: "Therefore, we fitted our model manually and variable selection was based on the significance of the variables and published data linking the variables to hypertension" and lines 143-145. You may not be using a specific algorithm and you may not be using "stepwise" where variables can exit and enter the model, but you are still using the p-value (which I presume is what you mean by "significance") to select variables in a sequential manner. From what I have seen from simulation studies, using p-values for selection with sequential model fitting does poorly at capturing the true effect unless the sample size is really large. That said, it's hard for me to assess incorporating the change-in-estimate and AIC into the selection procedure. I don't know conditional on the p-value based selection whether those reduce confounding and, if so, by how much. If we are working in an 'anything goes so long as you explain it' situation regarding variable selection, then I think what you have is permissible. But I am not really buying p-value based selection.

Response: We thank the reviewer for reading our responses on this issue. The reviewer is concerned by the "sequential variable selection based" and is not sure whether "conditional on the p-value based selection whether those reduce confounding and, if so, by how much". We agree with the reviewer that variable selection solely based on p-value would be inappropriate even when done manually. However, ask the reviewer to note that our variable selection was not exclusively based on p-value alone. For example, we have clarified that we tested whether variables were confounders (cause >10% change in effect size) before dropping them from the model (lines 145-149). Specifically, we included confounders in the final model even if their p-values were insignificant. P-value-based procedures would have dropped these, and we agree that the estimates would not be robust. In each case, we also tested whether the model with the "non-significant" confounder fits the data better than the model without the confounder. In that case, we used change-in-estimate and AIC into the selection procedure. We admit this was not clear in the manuscript, and we have revised the manuscript to make this clearer in the methods. Regarding our statement in the previous responses, "Therefore, we fitted our model manually and variable selection was based on the significance of the variables and published data linking the variables to hypertension", we intended to show that confounders were fitted in the final model even when they would have been dropped based on p-value (because of non-significant p-value). To further be sure of the estimates, we have rerun a full model (with all variables stated apriori, regardless of bivariate p-value), and the results are broadly similar. Also, the reduced final model is better fitting than a full model(AIC 849.693 vs 855.7153). Therefore, we have maintained the reduced final model as this is better fitting to the analysed data.

Reviewer 3: No comments

References

1 Shah BR, Laupacis A, Hux JE, Austin PC. Propensity score methods gave similar results to traditional regression modeling in observational studies: a systematic review. J Clin Epidemiol 2005; 58:550–559.

2 Cepeda MS, Boston R, Farrar JT, Strom BL. Comparison of logistic regression versus propensity score when the number of events is low and there are multiple confounders. Am J Epidemiol 2003; 158:280–287.

Attachment

Submitted filename: Response to Reviewers.docx

Click here for additional data file. (55.9KB, docx)

Decision Letter 2

Giuseppe Vittorio De Socio

7 Feb 2023

Prevalence and incidence of hypertension in a heavily treatment-experienced cohort of people living with HIV in Uganda

PONE-D-22-29381R2

Dear Dr. Byonanebye,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

Giuseppe Vittorio De Socio, MD, PhD

Academic Editor

PLOS ONE

Additional Editor Comments (optional):

Reviewers' comments:

Acceptance letter

Giuseppe Vittorio De Socio

9 Feb 2023

PONE-D-22-29381R2

Prevalence and incidence of hypertension in a heavily treatment-experienced cohort of people living with HIV in Uganda

Dear Dr. Byonanebye:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

If we can help with anything else, please email us at plosone@plos.org.

Thank you for submitting your work to PLOS ONE and supporting open access.

Kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Dr. Giuseppe Vittorio De Socio

Academic Editor

PLOS ONE

Associated Data

    This section collects any data citations, data availability statements, or supplementary materials included in this article.

    Supplementary Materials

    S1 Fig. Incidence rate of hypertension with increasing follow-up time.

    The rates of hypertension increased with increasing follow-up duration.

    (TIF)

    Click here for additional data file. (80.8KB, tif)
    Attachment

    Submitted filename: Review.docx

    Click here for additional data file. (16.5KB, docx)
    Attachment

    Submitted filename: Response to Reviewers.docx

    Click here for additional data file. (217.6KB, docx)
    Attachment

    Submitted filename: Response to Reviewers.docx

    Click here for additional data file. (55.9KB, docx)

    Data Availability Statement

    Data cannot be shared publicly as it contains potentially identifying participant information. In addition, there are restrictions on data sharing imposed by the infectious Diseases Institute and the ethics review committees, in line with the Uganda Data Protection and Privacy Act 2019. Therefore, interested researchers should send data requests jointly to: Barbara Castelnuovo, Head of Department – Research, Infectious Diseases Institute (Mulago Hospital building), P.O. Box 22418, Kampala, Uganda, email: research@idi.co.ug and to The Chair, Joint Clinical Research Council Research Ethics Committee (JCRC REC) Lubowa Hill, Plot 101 Entebbe Road, P. O. Box 10005, Kampala, Uganda. Email: jcrc@jcrc.org.ug.

    Articles from PLOS ONE are provided here courtesy of PLOS

    RESOURCES