INTRODUCTION

Individuals infected with HIV have a 1.5-times higher risk of cardiovascular disease (CVD) compared with non-infected individuals, after controlling for traditional cardiovascular risk (CVR) factors¹-⁶. The prevalence of these traditional risk factors is higher in the HIV+ population than in the general population. However, other independent risk factors for CVD have been described that are specific for HIV individuals, such as low CD4+ T lymphocyte count and use of certain antiretroviral agents³.

CVR prediction models are mathematical functions that allow to estimate the incidence of major cardiovascular events (major adverse cardiovascular events [MACE], i.e., cardiovascular death [CVD], stroke [CVA], coronary artery disease [CAD], and peripheral artery disease). They combine risk factors such as age, sex, smoking, lipid fraction blood levels, blood pressure, and among others, into prespecified equations that ponder the weight of each item in the probability of MACE occurrence. These equations are used for the prediction of CVR of an individual who, in turn, supports decision-making on lipid fraction level goals, hypolipemiant treatment, use of other therapies (e.g., antiplatelets), intensity of follow-up, and modification of antiretroviral treatment (HAART).

Several risk prediction models have been developed; they assess the probability of MACE at 5 and 10 years in the general population⁷-¹⁴ and others that include cohorts from some Latin American countries⁹,¹⁵,¹⁶. However, few of them have been evaluated in HIV-positive populations¹⁷-¹⁹. Only the Data Collection on Adverse Events of Anti-HIV Drugs Study (D:A:D)¹⁸ includes variables that are unique to this population, such as log of CD4+ T lymphocyte count and accrual exposure to protease inhibitors, nucleoside reverse transcriptase inhibitors, and abacavir.

Information about risk prediction models in HIV patients in Colombia and Latin America is extremely limited, and even worldwide information is scarce. The objective of this study is to evaluate the performance of D:A:D score¹⁸ and other five widely known equations, in predicting MACE in a population of HIV-infected adults in Colombia.

METHODS

Hospital Universitario San Ignacio (HUSI) keeps a medical registry of its population with HIV since 2004. It collects information on socio-demographic data, immunovirological status, hematology, liver and kidney function tests, plasma lipidic fractions, HAART, comorbidities, and among other. Data were gathered from follow-up visits scheduled every 6-12 months in accordance with practice guidelines²⁰-²². For this study, subjects were included if they were followed-up for at least 5 years. It was also required that information necessary for calculation of risk scores and incidence of MACE at 5 and 10 years was available. This study was approved by the institutional review board of HUSI and Pontificia Universidad Javeriana (code: FM-CIE-0619-18). As this study used only information from routine medical care, and data were anonymized, informed consent was waived in accordance with international regulations and Colombian law.

RESULTS

A total of 1,003 patients met the inclusion criteria; mean age was 35 years, and 12% were women. Five percent were either diabetic or hypertensive and 29% were smokers. Table 1 summarizes baseline characteristics of total cohort. Of this group, 195 patients (19.4% of total cohort) were excluded due to lack of information on MACE at 5-10 years, and the remaining 808 patients had their risk calculated with all the models. Table 2 summarizes risk calculation for the whole cohort with each equation. Only eight patients presented MACE (< 1 %), two patients died due to CVD, four patients developed CAD, and two developed CVA.

After evaluating ROC curves (Fig. 1), all models presented an acceptable performance in predicting CVR in this sample, with an AUC > 0.89 and narrow confidence intervals. There were no statistically significant differences between these AUC.

Figure 1 Receiver operating characteristic curves. 

DISCUSSION

Evaluation of CVR with validated models is essential when caring for adult patients. It predicts future incidence of MACE and helps establish recommendations on lifestyle modification, institution/intensification of treatment, and frequency of follow-up. However, these models have limitations resulting from the differences (geographical, ethnical, and chronological) in the various derivation and validation cohorts that were used for their development. In spite of these limitations, numerous clinical practice guidelines on dyslipidemia and CVR agree in recommending the use of one of these equations to predict risk. Ideally, the choice of a scale over another should be based on adequate validation of reproducibility and performance in the specific population for which it is intended to be used.

There are no models developed specifically for the evaluation of CVR in Latin America and the Caribbean. Carrillo-Larco et al. found an acceptable performance for different CVR evaluation models in their systematic literature review, with some limitations derived from sample size and number of CVE during follow-up²⁴. Others with some validation in Latin America, such as GLOBORISK and INTERHEARTRISK SCORE, and risk calculators from PAHO/WHO⁹,¹⁵,¹⁶, were not analyzed because they were not used in the population object of this study, or because they did not have complete information for their estimation. They highlight that the models with best performance in Latin America are Framingham and ASCVD. Some studies with HIV populations have been reported from Brazil, although with small sample size and short follow-up²⁴-²⁷.

To the best of our knowledge, this is so far the study of this nature with the largest sample size of HIV-infected individuals in Latin-America. We demonstrate a good performance in this population for the most frequently used equations in the region. The AUC values and narrow confidence intervals without statistically significant differences between models suggest that any of these equations may be used for this specific population.

Nevertheless, one limitation of our study is the fact that it was derived from a single center, with a small size compared with that of the derivation cohort of every equation. In addition, our cohort consisted mainly of young patients (mean age 35 years), with low and very low risk, which sets a limitation for the interpretation of results in patients with higher risk. However, the HIV+ population in Colombia is generally young¹⁰. The low incidence of MACE during follow-up may be explained by this young age and a low prevalence of the other traditional risk factors at baseline and to a good adherence to guidelines. However, the small number of outcomes makes the calculation of false-negatives and false-positives limited, thus the results are of very low external validity. Subgroup analysis was not performed given the low number of outcomes (MACE).

Another important consideration is that 19% of the cohort was excluded due to lack of information on MACE and vital status at 5 and 10 years. This population might have contributed with more events. Likewise, there may be a selection bias – the sample population had a very good adherence to treatment and received a strict follow-up – which may have improved cardiovascular outcomes in the short and long term. On the other hand, our study is innovative regarding its sample size and the population included, in comparison with the information thus far available, and therefore, does not allow fully generalizing them to other populations.

In conclusion, risk models evaluated in this study had an acceptable performance for the prediction of cardiovascular events in a particular Colombian HIV cohort, especially for low and very low risk individuals. However, the limitations of the study do not allow us to give a general recommendation for the Latin American population with HIV. Therefore, it may be advisable to evaluate the validation and performance of different CVR equations with multicenter studies and larger sample sizes. Perhaps selecting the model that requires fewer variables when evaluating patients with HIV in Colombia may be reasonable.