Back HIV-Related Conditions Cardiovascular CROI 2015: Predicting Cardiovascular Disease in People with HIV -- Can We Do Better?

CROI 2015: Predicting Cardiovascular Disease in People with HIV -- Can We Do Better?


Four existing models used to predict risk of cardiovascular disease (CVD) underestimated that risk in large cohorts of people living with HIV, according to research presented at the recent 2015 Conference on Retroviruses and Opportunistic Infections (CROI) in Seattle. Other studies suggested that even without that underestimation being taken into account, many HIV-positive people are currently undertreated with statins, which could protect them from cardiovascular events like myocardial infarction.

These posters were presented in a themed discussion on the accuracy of models used to predict risk of developing CVD and whether they can be improved for people living with HIV. The answer may be yes, as other studies discussed in the same session suggested ways to potentially refine the estimates -- or possibly reduce the problem altogether simply by treating HIV earlier.

CVD in People with HIV

Large cohort studies comparing rates of heart disease among people with HIV and people in a matched HIV-negative population have found around a 2-fold increased risk, according to discussion leader Nina Friis-Møller of Odense University Hospital in Denmark, who provided background for the session.

This increase is at least partly explained by differences in lifestyle, most notably a higher rate of smoking among HIV-positive people. But several studies also point to an increased likelihood of CVD due to risk factors that are unique to HIV, such as inflammation or the potential metabolic and other side effects of certain antiretroviral drugs. One large study -- the Veterans Aging Cohort Study (VACS) -- found that cardiovascular risk remains about 50% higher even after adjusting for classical risk factors, comorbidities, and illicit drug use.

But how can all these risk factors be weighed in models to improve patient management? Can conventional models for the HIV-negative population be used, or do we need models tailored to HIV-positive people, in view of their specific risks?

Different existing models have taken somewhat different approaches, but the purpose of all of them is to identify people at higher risk of CVD and related events such as heart attacks and strokes, so that advice and interventions can be targeted more effectively.

  • The development of the Framingham Risk Score (FRS) in the 1950s pioneered CVD prediction, identifying individuals with traditional risk factors that increased the likelihood of cardiovascular events within 10 years. 
  • The American Heart Association and American College of Cardiology recommendations for lipid-lowering therapy are informed by the Pooled Cohort Equations (PCE), which predict the risk of a combination of cardiovascular outcomes.
  • Guidelines from the European Society of Cardiology are based on risk estimates of fatal cardiovascular events based on the Systemic Coronary Risk Evaluation (SCORE).
  • A model tailored to predict cardiovascular risk among people with HIV has been developed based on traditional and HIV-related risk factors and observed cardiovascular outcomes in the large D:A:D cohort (first published in 2010).
  • The World Health Organization has developed separate models of cardiovascular risk for populations in Africa.

Models Underestimate CVD Risk

The themed discussion at CROI included 2 studies that assessed how well different risk prediction models work for HIV-positive populations.

The first of these, presented by Kenneth Lichtenstein of the National Jewish Health Center in Denver, looked at 4 different risk prediction models -- Framingham Risk Score, PCE, SCORE, and the D:A:D model -- to see how well they predicted cardiovascular events in a large cohort of people living with HIV, namely the HIV Outpatient Study (HOPS).

The researchers analyzed longitudinal data from 2392 HIV-positive people receiving care at 10 HIV clinics in the U.S. who enrolled in HOPS no later than October 2010 and had
at least 1 year of follow-up. A majority were men, half were white, and the median age was 43 years. About 40% were smokers, half had hypertension (high blood pressure), 17% had elevated cholesterol, and 10% had diabetes. The overall median follow-up period was 6.5 years, with 725 individuals having at least 10 years of follow-up.

Cardiovascular events were defined somewhat differently for each risk model:

  • FPS: myocardial infarction, fatal coronary heart disease, and stroke;
  • PCE: myocardial infarction, stroke, and coronary artery disease;
  • SCORE for low-risk populations: fatal myocardial infarction, stroke, peripheral vascular disease, and coronary artery disease;
  • D:A:D: myocardial infarction, sudden death, coronary artery disease, stroke, and other coronary heart disease death.

In addition, the risk factors included in each equation were also somewhat different, although all included age, sex, systolic blood pressure, smoking status, and total cholesterol:

  • FPS: diabetes, high-density lipoprotein (HDL or "good") cholesterol;
  • PCE: diabetes, HDL cholesterol, low-density lipoprotein (LDL or "bad") cholesterol, race/ethnicity;
  • SCORE: no additional factors;
  • D:A:D: diabetes, family history of CVD, use of indinavir (Crixivan), lopinavir/ritonavir (Kaletra), or abacavir (Ziagen, also in Epzicom).

A total of 204 new cardiovascular events occurred during follow-up in this cohort. All equations underestimated 10-year CVD risk to varying degrees, with ratios of expected versus observed events ranging from 0.75 to 0.85.

"We found that all 4 of these models under-predicted cardiovascular events in patients with any length of time in the entire cohort, and in a sensitivity analysis looking at those with 10 years of follow-up (although this was a much smaller group)," said Lichtenstein. "So we feel that other issues and other factors need to be considered when looking into cardiovascular risk in HIV."

The second study, presented by Virginia Triant of Massachusetts General Hospital and Harvard Medical School, compared the 2013 American College of Cardiology and American Heart Association (ACC/AHA) risk prediction algorithm using the PCE, versus the FRS, for predicting CVD risk among people with HIV.

This analysis included 2270 participants in the Partners HealthCare System HIV longitudinal cohort. About 60% were men, 56% were white, 21% were black, and the median age was 46 years. Over one-third were smokers, 58% had elevated blood lipids, 35% had hypertension, and 20% had diabetes. The median follow-up period was 6.3 years.

The studies identified the proportion of individuals with high predicted 10-year CVD risk using both algorithms and evaluated the degree of discordance between them. CVD risk was considered high if >10% according to FRS or >7.5% according to ACC/AHA risk score. The study also compared observed versus predicted 5-year cardiovascular event rates (coronary heart disease for FRS or atherosclerotic CVD for ACC/AHA).

"We found that the ACC/AHA risk prediction algorithm classified a higher proportion of patients as high cardiovascular risk, with 25% classified as high risk by ACC/AHA versus 10% by [FRS]," Triant concluded. Cardiovascular risk prediction scores were discordant in 16% of patients, with ACC/AHA predicting high risk and FRS predicting low risk in 99% of these cases.

But both the FRS and the ACC/AHA algorithms actually underestimated cardiovascular risk in the HIV-positive cohort, with 5-year observed event rates being significantly higher than the predicted rates.

Statin Eligibility and Use

Even though the ACC/AHA algorithm underpredicts CVD risk among people living with HIV, following the recommendations in the 2013 ACC/AHA guidelines would result in significantly more people qualifying for statin therapy compared to the previous Adult Treatment Panel (ATP-III) guidelines, according to an analysis presented by Meredith Clement of Duke University. 

This study used data from over 13,000 people in the Veterans Affairs Clinical Case Registry of HIV-positive veterans. All were men, aged 40 to 75 (median 54), who received care during 2008-2010. More than half were on medication for hypertension, 15% had diabetes, and 16% were smokers.

First, a chart review found that even under the old guidelines, a high proportion of HIV-positive veterans were undertreated: 27% of those eligible for statin therapy under the previous guidelines were not receiving them.

Under the new guidelines, recommendations for statin therapy increased substantially, rising from 53% according to ATP-III to 65% according to ACC/AHA. This represented a 12% absolute increase in statin eligibility among veterans living with HIV, and suggests that approximately two-thirds of men with HIV may qualify for statin treatment.

"Our findings highlighted a potential gap in quality of care for HIV-infected patients, a group at increased risk for CVD," the researchers concluded. "This gap expands with the new guidelines and provides an opportunity to improve care for our HIV-positive patients," Clement said.

According to a related study of the Partners HealthCare System HIV longitudinal cohort by Triant's group, 42% were eligible for statin therapy according to the ACC/AHA guidelines compared to 26% according to ATP III guidelines. Actual statin prescription rates among eligible patients were 47% under ACC/AHA and 65% under ATP III. Among the 165 people who experienced cardiovascular events, statins were recommended for 62% by ACC/AHA and 44% by ATP III.

"While ACC/AHA recommended a higher proportion of patients to be on statin therapy and identified a higher proportion of patients with CVD outcome events compared to ATP III, approximately 40% who experienced CVD outcome events still would not have qualified for statin therapy under ACC/AHA," the researchers concluded.

Both Lichtenstein and Triant said they thought that incorporating other factors -- perhaps HIV infection itself or HIV-specific factors -- could increase the accuracy of cardiovascular risk prediction algorithms and improve long-term outcomes for people living with HIV.

Refining the Models

Two other presentations in the themed discussion session shared findings that might improve the accuracy of risk prediction models.

Daniel Drozd of the University of Washington in Seattle presented data drawn from the North American AIDS Cohort Collaboration on Research and Design (NA-ACCORD), suggesting that researchers need to be clearer about the types of myocardial infarction (MI), or heart attack, that are included in cardiovascular risk calculations, as there are physiologically distinct types.

Type-1, or primary MIs, are the results of atherosclerotic thrombotic coronary plaque rupture, while type-2, or secondary MIs, can result from a diverse array of clinical conditions.

The researchers observed 271 primary MIs and 219 secondary MIs among 25,094 NA-ACCORD participants with more than 100,000 total person-years of follow-up.

Individuals with primary MIs were more likely to be male, to have elevated total cholesterol, and to have a history of statin use and ART use. Those with secondary MIs were more likely to be female, to be black, to have injection drug use as an HIV transmission risk factor, and to be ART-naive. More than 50% of secondary MIs were due to sepsis or vasospasm related to cocaine or other illicit drug use.

Traditional CVD risk factors including age, hypertension, diabetes, and elevated cholesterol independently predicted primary MI risk, as expected. In addition, HIV-associated risk factors including detectable viral load, a history of AIDS-defining illness, and lower CD4 cell count were all independent predictors of primary atherosclerotic MIs.

"Aggressive traditional and HIV-related risk factor management, including ART, may reduce primary MI risk in HIV-infected individuals," said Drozd. But the high rate of secondary MIs emphasizes the need for greater clarity in determining outcomes in studies of the pathogenic role of HIV in CVD.

Drozd was asked about the extent to which these secondary MIs could be contributing to the excess CVD risk seen among people with HIV in some studies.

"I think it’s certainly a possibility that it’s a contributor, but I think it’s unlikely to entirely explain the excess cardiovascular risk," he said. "It should be noted that the frequency of secondary [MIs] -- which in our cohorts were 45% or so of overall events -- vary widely depending upon the population sample. And so it may be that there are other cohorts that have fewer of these secondary [MIs]."

Indeed, it is worth noting that the secondary MIs were seen more often among women who use drugs -- and there were no women in the VACS cohort study that found 50% greater CVD risk among veterans with HIV.

Finally, Jorge Salinas from Emory University in Atlanta presented a study that underlined the importance of taking into account cumulative damage caused by HIV disease over the years of infection.

Over the course of the epidemic researchers have reported conflicting findings about the relative importance of cross-sectional measures of viral load and immunological parameters as factors in the heighted CVD risk seen among people with HIV.

So Salinas and colleagues assessed whether HIV care parameters -- mainly viral load, CD4 T-cell count, and a VACS index score -- provide information about risk of myocardial infarction when measured cumulatively, not just at a single point in time. To do so, they selected more than 8000 treatment-naive individuals in the VACS virtual cohort who started ART between 2002 and 2012.

Their first model assessed baseline viremia -- a cross-sectional measure -- and found a 50% higher risk of MI among people with HIV viral load above the median level. Their second model, looking at time-updated viremia, did not show a consistent dose-dependent effect on MI incidence (as might be expected, given that viral load should fall while on treatment). Lastly, they modeled the effect of viremia copy-years -- a cumulative measure of viral load over time -- and found a consistent dose-dependent association with MI incidence. Similar models looking at CD4 measurements did not find a statistically significant effect.

The researchers also explored MI risk using the VACS index score, a composite score that incorporates age, HIV viral load, CD4 count, and measures of liver and kidney dysfunction including AST and ALT liver enzyme levels, hemoglobin, platelet count, creatinine, and hepatitis C virus infection.

The effect of baseline VACS index score did not reach statistical significance, nor did time-updated VACS index score. But a third model, which used cumulative VACS index score over time, had the most robust dose-dependent association with MI incidence.

"In summary, cumulative measures of viremia and the cumulative VACS index score provide added information about the risk of MI in people living with HIV," Salinas said. In essence, CVD risk appears to be increased among people who have had high viral load longer and who have spent a longer amount of time in poor health.


Following the presentations, the discussion then focused on how to increase risk calculation accuracy and what can be done clinically to improve patient's outcomes.

There might be other measures that can help identify poor health over time. Salinas said his model did not include nadir (lowest-ever) CD4 cell count or CD4/CD8 ratios, but suggested these are "potential HIV care measures to look at." He added that there is more and more evidence that biomarkers such as soluble CD163 or IL-6  "could potentially make it into HIV-specific risk calculators."

"In the HOPS cohort we are looking to try to see if we might develop a risk calculation that’s a little bit more consistent with the outcomes that we are seeing," said Lichtenstein. "But I think we're struggling, like everybody else, to figure out what we would be looking at."

Salinas believes it should be possible to look at viral load parameters and VACS index score in a cumulative fashion -- updated at each clinic visit -- and that this might improve accuracy.

"By looking at the cumulative VACS index score years, we are taking into account a time-updated variable that contains viral load, CD4 cell count, renal disease, etc,...and at every time point that measure is being calculated, it is taking into account all of the previous history of the patient," he said. "Ideally, I would love to see all the [calculations] that we are discussing today being included in electronic medical records, so that the research is implemented into actual clinical care."

Many discussion participants suggested that these issues might only apply to patients who have been in care for a long time, some of whom started HIV treatment late using less effective and more toxic therapy. With earlier treatment and better health status, people who initiate ART nowadays might be unlikely to have such an elevated risk for heart disease.

But what should be done for those who are already at elevated risk?

There was broad agreement that the first interventions should involve lifestyle changes -- such as diet, exercise, and smoking cessation -- which can offer a lot of benefits, and only then consider statins if these are inadequate.

However, Stephen Grinspoon of Massachusetts General Hospital, who gave a plenary talk on cardiovascular disease and statin use earlier the same day, stressed that, "this disease is happening in people with a very low [Framingham risk score] already. I just don't know if lifestyle will be enough, because it is a different sort of disease [than CVD in HIV-negative people]."

"Rather than add other factors to the risk calculation, shouldn't we just lower the threshold for when to treat?" asked one clinician in the audience. 

"In practice, effectively I do that, and many of my colleagues do," Triant replied. "Either you lower the threshold for treatment or you raise the risk...We will get direct data from REPRIEVE, etc., but there is a lot of indirect data suggesting that it is defensible to do that now."


Cardiovascular Risk Factors for HIV-Positive People

Statins May Reduce Risk of Heart Disease in People with HIV



AM Thompson-Paul, KA Lichtenstein, C Armon, et al. Cardiovascular Disease Risk Prediction in the HIV Outpatient Study (HOPS). 2015 Conference on Retroviruses and Opportunistic Infections. Seattle, February 23-24, 2015. Abstract 747.

S Regan, JB Meigs, J Massaro, V Triant, et al. Evaluation of the ACC/AHA CVD Risk Prediction Algorithm Among HIV-Infected Patients. 2015 Conference on Retroviruses and Opportunistic Infections. Seattle, February 23-24, 2015.  Abstract 751.

M Clement, L Park, A M Navar-Boggan et al. HIV-Infected Veterans and the New ACC/AHA Cholesterol Guidelines: Got Statins? 2015 Conference on Retroviruses and Opportunistic Infections. Seattle, February 23-24, 2015. Abstract 750.

S Regan, JB Meigs, S Grinspoon, et al. Application of New ACC/AHA Cholesterol Guidelines to an HIV Clinical Care Cohort. 2015 Conference on Retroviruses and Opportunistic Infections. Seattle, February 23-24, 2015. Abstract 734.

D Drozd, M Kitahata, S Heckbert, et al. Incidence and Risk of Myocardial Infarction (MI) by Type in the NA-ACCORD. 2015 Conference on Retroviruses and Opportunistic Infections. Seattle, February 23-24, 2015. Abstract 748.

J Salinas, V Marconi, D Rimland, et al. Cumulative HIV Care Measures Highly Associated With Acute Myocardial Infarction. 2015 Conference on Retroviruses and Opportunistic Infections. Seattle, February 23-24, 2015. Abstract 746.