Comparative performance of cardiovascular risk prediction models in people living with HIV

Background Current cardiovascular risk assessment in people living with HIV is based on general risk assessment tools; however, whether these tools can be applied in sub-Saharan African populations has been questioned. Objectives The study aimed to assess cardiovascular risk classification of common cardiovascular disease (CVD) risk prediction models compared to the Data Collection on Adverse Events of Anti-HIV Drugs (D:A:D) 2010 and 2016 models in people living with HIV. Method Cardiovascular disease risk was estimated by Framingham Cardiovascular and Heart Disease (FHS-CVD, FHS-CHD), Atherosclerotic Cardiovascular Disease (ASCVD) and D:A:D 2010 and 2016 risk prediction models for HIV-infected participants of the Ndlovu Cohort Study, Limpopo, rural South Africa. Participants were classified to be at low (< 10%), moderate (10% – 20%), or high-risk (> 20%) of CVD within 10 years for general CVD and five years for D:A:D models. Kappa statistics were used to determine agreement between CVD risk prediction models. Subgroup analysis was performed according to age. Results The analysis comprised 735 HIV-infected individuals, predominantly women (56.7%), average age 43.9 (8.8) years. The median predicted CVD risk for D:A:D 2010 and FHS-CVD was 4% and for ASCVD and FHS-CHD models, 3%. For the D:A:D 2016 risk prediction model, the figure was 5%. High 10-year CVD risk was predicted for 2.9%, 0.5%, 0.7%, 3.1% and 6.6% of the study participants by FHS-CVD, FHS-CHD, ASCVD, and D:A:D 2010 and 2016. Kappa statistics ranged from 0.34 for ASCVD to 0.60 for FHS-CVD as compared to the D:A:D 2010 risk prediction model. Conclusion Overall, predicted CVD risk is low in this population. Compared to D:A:D 2010, CVD risk estimated by the FHS-CVD model showed similar overall results for risk classification. With the exception of the D:A:D model, all other risk prediction models classified fewer people to be at high estimated CVD risk. Prospective studies are needed to develop and validate CVD risk algorithms in people living with HIV in sub-Saharan Africa.


Introduction
Sub-Saharan Africa (SSA) has the largest HIV-infected population in the world comprising around 70% of the global HIV population of 37.7 million. 1 While antiretroviral therapy (ART) has reduced morbidity and mortality in people living with HIV (PLHIV), 2 complications associated with ageing such as cardiovascular disease (CVD), cancer, osteoporosis, and other end-organ diseases are increasing. 3,4 Cerebrovascular disease and CVD were ranked as the fourth and fifth leading causes of death in Africa in 2015. 5 In PLHIV early onset of CVD has been observed. 6,7,8,9 Risk of CVD is increased up to two times in PLHIV. 10 It is likely that inflammation from HIV infection, which only partially resolves on ART contributes to CVD development. Moreover, there is an association between HIV and traditional CVD risk factors, 11,12 and adverse reactions of the ART medication may also play a role 13 . http://www.sajhivmed.org.za Open Access To identify PLHIV at high-risk of CVD, the Data Collection on Adverse Events of Anti-HIV Drugs (D:A:D) 14 risk algorithm can be used. However, the D:A:D prediction algorithm was developed based on data from a predominantly high-risk male population residing in high-income countries and did not include clinical outcomes such as cardiac failure, transient ischemic attack (TIA) and peripheral artery disease (PAD). Therefore, its application for CVD risk prediction in PLHIV in SSA has been questioned. Currently, there is no validated CVD risk prediction tool for PLHIV in low-and middle-income settings. 15,16,17 We compare CVD risk estimation by common CVD risk prediction models. Reported angina, heart failure, or heart attack by a parent or sibling before the age of 60 years was considered as family history of CVD (the Framingham Heart Study-Cardiovascular Disease [FHS-CVD], 18 the Framingham Heart Study-Coronary Heart Disease [FHS-CHD], 19 the Atherosclerotic Cardiovascular Disease Risk Score [ASCVD] 20 and D:A:D 2016 relative) with the D:A:D 2010 model 14,21 in PLHIV in rural South Africa.

Study design and study population
This study was conducted using baseline data from the Ndlovu Cohort Study (NCS), 21 a which is investigating the association of HIV infection, HIV treatment, and conventional CVD risk factors on CVD risk in rural South Africa. 8,22 Baseline assessment of study participants of the NCS was undertaken between November 2014 and August 2017 in Elandsdoorn, Limpopo, South Africa. Overall, 1927 people (male and female) participated in the study, among them 887 infected with HIV. Participants for the NCS were recruited through community campaigns, at local events and shopping centres, as well as at the Ndlovu Medical Centre (NMC). The NMC included a large rural HIV treatment facility, contracted by the South African Department of Health, providing freeof-charge HIV treatment and follow-up to ≈3700 HIV-positive patients. Details on the design and the methods of the NCS have been described previously. 21 The current analysis is based on 735 HIV-infected participants aged 30 to 75 years. The age cut-off points were chosen according to the age range for both Framingham and ASCVD prediction models. Established CVD was defined as selfreported angina pectoris, stroke, myocardial infarction and heart failure. Participants with established CVD were excluded from the current sub-study. Figure 1 provides details on the inclusion and exclusion criteria.

General characteristics
Information on age, sex, income, civil status, demographics, socio-economic status (SES), medical history, and medication use (both HIV-related and for other medical issues), family history, and lifestyle were collected by use of standardised questionnaires. 23 Detailed information on past and current HIV treatment was recorded including the time between diagnosis and treatment initiation, medication prescription and treatment response (latest plasma HIV-1 viraemia, latest CD4+ cells). ART status was assessed by self-reporting and complemented with information from an electronic HIV registry (TIER.net). 24 Physical activity was evaluated by use of the international physical activity questionnaire (IPAQ). 25 This questionnaire assesses the intensity of physical activity and sitting time in order to estimate total physical activity in metabolic equivalents (MET)-min/week. Information was collected on smoking and alcohol status (i.e. ever or never use of cigarettes or other tobacco products, and alcoholic beverages, and the participants' current use of cigarettes and alcoholic beverages). History of raised blood pressure, any drugs taken for blood pressure (previously or currently), history of diabetes, and other CVD risks were also assessed. Participants with a body mass index (BMI) > 30 kg/m 2 were considered obese. Diabetes mellitus (DM) was defined based on haemoglobin A1c (HbA1c) > 6.5 mmol/L or being on diabetes medication. Income in this low-resource setting was categorised as low (< R648.00/month [South African rand]), medium (R648.00 -R992.00/month) and high (> R992.00/ month). 26 Reported angina, heart failure, or heart attack by a parent or sibling before the age of 60 years was considered as family history of CVD.

HIV testing procedure
Participants underwent HIV testing unless their status was known to be HIV-positive and on ART.  ABON Biopharma, Hangzhou, China). 27 Participants who tested positive for HIV upon enrolment in our study were referred to the NMC, or any other local HIV treatment facility, to initiate ART. Participants with confirmed HIV-positive status were recruited from the NMC or outreach testing programmes. Documentation of a positive HIV test result was needed for all patients to be eligible for enrolment as HIV-infected participants in the study.

Physical measurements
Anthropometric measurements included weight, height, and waist and hip circumference. Blood pressure was measured in all participants in a seated position after 5 min of rest on both arms and repeated on the arm with the highest values. The average of the second and third reading was used for analysis. Hypertension was defined as systolic or diastolic blood pressure of ≥ 140/90 mmHg or use of blood pressure lowering medication.

Biological measurements
Total cholesterol, high-density lipoprotein (HDL) cholesterol, low-density lipoprotein (LDL) cholesterol, triglycerides, random glucose and HbA1 were measured in all individuals. For HIV-positive participants viral load and CD4 count was measured. 21

Cardiovascular risk estimation
Cardiovascular disease risk was estimated using the FHS-CVD, 18 FHS-CHD, 17 ASCVD 20 and HIV-specific D:A:D prediction equations. 14,28 Details on the risk prediction equations and information regarding the different cohorts and inclusion criteria for the development of the risk scores are shown in Table 1. The cohorts for FHS and the ASCVD risk prediction models have a follow-up of over 10 years, while the D:A:D cohort has a median follow-up of five years. For the current analysis, we assumed that the CVD risk was constant over time, allowing us to predict 10-year CVD risk. From the FHS, we chose to include both the CVD and the CHD prediction model in our analysis, as these are widely used for risk estimation in clinical practice. Generally, the D:A:D 2010 algorithm is applied in clinical practice for risk estimation in PLHIV, motivating our choice to use this algorithm as reference for our comparative analysis of general CVD risk prediction scores. 29 Details of the endpoints used by these different risk prediction models are described in Appendix Table 1-A1. All risk prediction models included CHD-related deaths, myocardial infarction, stroke, and stroke death as outcome, except for the FHS-CHD score, which only assessed CHD death and myocardial infarction. Details on equations used to calculate the individualised CVD risk by different CVD risk prediction tools are shown in Appendix Figure 1-A1.

Data management and statistical analysis
Baseline characteristics of the study participants are presented as mean and standard deviation (s.d.) for normally distributed continuous variables, median and interquartile range (IQR) for non-normally distributed continuous variables and for categorical variables as count and percentage. There was a high percentage of blood pressure data obtained by a non-validated blood pressure device (61.6%) and hence discarded. Missing values on blood pressure were assumed to be missing at random (MAR). Multiple imputations were  29 All valid blood pressure measurements, gender, age, body mass index (BMI), smoking status, glucose level, total cholesterol, HIV medication use, employment, and education were included in the imputation model. Each parameter of interest in each imputed data set was estimated separately and later combined with Rubin's rule for analysis. Convergence plots were used to confirm that the MICE algorithm was converted. Complete case analysis was performed with the mean of the imputed data sets.
Following individual CVD risk calculation, the overall cumulative cardiovascular risk for each model was then compared using a cumulative density plot. Hence, HIVinfected study participants were categorised as having low risk (< 10%), moderate risk (10% -20%), and high-risk (> 20%) of a CVD event within 10 years as categorised in previous similar studies. 30,31 These categories were plotted to compare the predictions of the general CVD risk prediction models (ASCVD and FHS-CVD and FHS-CHD) with the HIVspecific D:A:D models. Furthermore, we performed subgroup analysis according to age. Since age is the most important predictor in the prediction models, it can be expected that predicted risks are higher in the older age groups. In

Study population characteristics
The majority of the 735 participants were women (56.7%). A third (32%) of the study participants were classified as overweight and obese, and the prevalence of diabetes was 3.5%. Current smoking was reported by 23.1% of the participants. Further details on study population characteristics are provided in Table 2.
In subgroup analysis, we analysed CVD risk within different age groups. For the age category 30-39-years, all were categorised as low risk by all the prediction models; for the age category 40-49-years, 97.0% were categorised as low risk, 3.0% as moderate risk, and ≤ 1.0% as high risk. For those aged 50-59-years, 68.3% were categorised as low risk, 25.4% as moderate risk, and 6.3% as higher risk; for those aged 60 years or older, 30.0% were categorised as low risk, 42.0% as moderate risk, and 28.0% as higher risk.  Table 3.

Discussion
We showed that FHS-CVD and D:A:D 2010 risk equations predict relatively similar 10-year CVD and 5-year CVD risk in PLHIV in a rural community in South Africa. Comparison  which could also lead to higher estimates. Moreover, it is considered that ethnic or racial status is a significant predictor of CVD risk. 34 To date, however, most CVD prediction models 35 are not taking into consideration ethnic or racial status, which could be a possible reason for the lower estimates observed in the NCS, a study undertaken in a Black African population in rural SSA. Ethnic or racial status should be considered as a potentially important factor when developing prediction models. Currently it is not known what risk prediction model more accurately estimates CVD risk in a SSA setting.
A previous study in PLHIV in Portugal, 31  Overall, in the current study, few participants were categorised to be at high CVD risk. This is most likely related to the young age of the study population and the relatively low prevalence of traditional CVD risk factors, such as smoking, abnormal lipid levels and alcohol consumption. Conventional CVD risk factors have been shown in several studies to be lower for PLHIV compared to the general population in low-and middle-income settings. 7,26,36,37 Underlying reasons for this observation, however, remain unclear. 37,38 One possibility is increased access to healthcare and regular health checks for PLHIV. 39 There are numerous risk prediction algorithms to calculate CVD risk. 40,41,42 The FHS-CVD, FHS-CHD and ASCVD are among the most frequently used and well-known CVD risk algorithms. These prediction models are recommended by many guidelines for the prevention and management of CVD. 43 Our study shows that these risk prediction models classified PLHIV at the NCS to be at similarly low, medium or high CVD risk as the HIV-specific D:A:D 2010 model. These risk algorithms are all based on cohorts with a long follow-up and are generally available in the form of easy-to-use electronic calculators. 45,46 The results of our study show that all models predicted relatively similarly in the low-risk category; therefore, overestimation in a low-risk population is unlikely by use of any of the risk prediction models. For PLHIV with a moderate   and high risk of CVD under-treatment, and for those with a low risk, over-treatment could be avoided, potentially resulting in drug-related adverse reactions, drug resistance, and drug-drug interactions. Risk stratification could guide preventive measures and efficient use of resources until a CVD SSA prediction model has been developed and validated.
To identify the CVD risk prediction models most applicable to PLHIV in the general population in different settings over time, longitudinal studies collecting clinical outcome data are recommended. The D:A:D model has been calibrated for a 5-year risk prediction. Therefore, extrapolation is required to predict a 10-year CVD risk. The assumption in this study is that risk is stable over time, thus extrapolation of the estimates of the D:A:D models leads to greater uncertainty and a higher risk of producing less relevant results. This study is one of few studies that have added the recalibrated D:A:D 2016 risk prediction model in their analysis to provide a better picture of estimates and results in SSA populations.
In conclusion, CVD risk estimation using Framingham CVD as compared to the HIV-specific D:A:D 2010 model showed similar results for risk classification of PLHIV. This study demonstrates that currently used CVD risk prediction models show good agreement in the low CVD risk category, whereas the observed agreement is moderate for the medium-risk and high-risk categories. To enable the development or validation of CVD risk algorithms that can be used broadly in PLHIV in SSA populations, prospective studies are recommended. Information on newer antiretrovirals 44 along with other CVD risk factors, and clinical outcomes, 7,26,47 could lead to refined CVD prediction models for use in the SSA setting. 48  Note: Patients were divided into three categories: < 10%, 10% -20% and > 20% risk of CVD. 'Agreement' indicates the degree to which the model categorises the patients in the same way as the D:A:D 2010 model. The kappa statistics are included. CVD, cardiovascular disease; FHS, Framingham heart study; CHD, coronary heart disease; ASCVD, atherosclerotic cardiovascular disease.