Session V: Development of Appropriate CHD Risk Prediction Models

Introduction

Prediction equations for coronary heart disease (CHD) risk are useful tools that inform clinicians and patients about the absolute risk for developing CHD. A basic principle in CHD prevention is that the intensity of risk-reducing interventions should be based on the individual patient’s absolute CHD risk. In the current era of HIV infection and highly active antiretroviral therapy (HAART), knowing one’s CHD risk and acting to reduce it has become imperative to long-term survival. Given the increased life expectancy due to HAART, more HIV infected persons will experience complications not related to HIV per se, and will reach an age where they are at increased risk for developing CHD.

However, existing CHD risk prediction equations were not developed in HIV-infected adults or children. In the general population, CHD risk prediction models derived from the Framingham Heart Study estimate the risk of total CHD (angina pectoris, myocardial infarction (MI), coronary heart disease death)¹, or estimate the risk for hard CHD endpoints (MI, coronary heart disease)². The traditional risk factors used to predict CHD risk and how risk factor alterations affect CHD outcomes in HIV-infected and HIV-seronegative people are summarized in Table 1. The estimates of the relative effects of traditional risk factors on CHD outcomes appear similar between HIV and non-HIV-infected patients. However, they are based on only two studies in HIV-infected patients. While traditional CHD risk factors may operate in the same manner in HIV patients as in the general population, there may still be a need to identify and evaluate HIV-specific CHD risk factors and equations, refine existing CHD prediction equations, and develop new HIV-specific CHD prediction equations for adults, adolescents and children. To date, Framingham CHD risk predictions have performed reasonably well when applied to HIV-infected patients. We need to evaluate whether new HIV-specific CHD risk prediction approaches perform better than existing algorithms. This represents a formidable challenge, but is crucial to improve care and reduce health care costs for people living with HIV.

Table 1.

Do traditional main cardiovascular risk factors predict the risk of CAD/CVD in HIV-infected (HIV+) similarly to HIV-uninfected (HIV−) persons?

		% Increase in risk per unit for each study
CV risk factor	Unit	HIV+(14,36)		HIV− (49–54)
Age	per 1 year older	9%	6%	6%–9% (7)
Sex	Male vs female	n.s.	110%	110%–160% (2)
Diabetes	Yes vs. no	260%	90%	140%–252% (3)
Smoking	Yes vs. no.	140%	290%	70%–290% (3)
Hypertension	Yes vs no	30%	80%	80%–90% (3)
Total cholesterol	per 1 mmol/L# ↑	-	26%	25%–33% (3)
HDL-cholesterol	per 1 mmol/L# ↑	-	−28%	−52% (1)

Adapted from J.Lundgren (Chicago June 2007).

( ) =number of studies considered.

^#1 mmol/L = 39 mg/dL

Existing CHD Prediction Equations

Historically, CHD prediction equations have been derived to estimate risk for an initial coronary event in a population with minimal coronary heart disease risk at the beginning of the observation period ^1,3,4, generally from data obtained in middle-aged adults. Such equations may not provide suitable predictions for young people living with HIV. Of concern, adolescents have been and continue to be infected with HIV, and despite substantial reductions in mother-to-child transmission in the US, infants in resource-limited regions of the world continue to be infected with HIV and adolescents continue to experiment with intravenous drugs, share needles, and engage in unprotected sex. Existing equations predict the 10-yr CHD risk based upon longitudinal observational data. Observational data are limited in their ability to predict adverse events ⁵. HIV was identified in the 1980’s, and HAART has only been available for ~10yrs, so predicting 10-yr CHD risk with limited HAART experience will challenge existing CHD prediction equations.

Despite these limitations, existing CHD prediction algorithms have been applied to HIV-infected people ^6–13 and have performed reasonably well ^14,15,16,17. Of note, there appears to be a significant, but modest concordance among 3 popular CHD risk stratification algorithms when applied to HIV ¹⁰. For the time being, it appears the Framingham Risk Score calculator (http://hp2010.nhlbihin.net/atpiii/calculator.asp) can be used effectively to rank CHD risk in HIV-infected people; the only information required is age, smoking status, and fasting lipid profile. Most ^6–9, but not all studies ^11,13 found a slightly greater CHD risk in HIV-infected people taking HAART and experiencing abnormal adipose tissue distribution, when compared to ‘matched’ HIV-seronegative controls. The contribution of type and duration of individual anti-HIV drugs, and their differential effects on serum lipids/lipoproteins to CHD risk are still debated ^{14,17–20,21,22,23}. When actual CHD events (acute MI, cardiac death) are documented, the limited data suggest that CHD risk factors were more prevalent, and acute MI rates were higher in HIV compared to age matched non-HIV patients ^12,14,20,24 (Fig. 1). In these reports, however, the absolute risk of MI was low compared with the known benefits of HAART in terms of reducing AIDS-related mortality ^21,25,26. Given the limitations of existing CHD prediction equations, the evolving nature of the cardiometabolic syndrome in HIV, and the relative infancy of this area of inquiry, it is important to grasp this window of opportunity and move forward with the refinement of existing or development of new HIV-specific CHD prediction equations.

Fig 1. (From Triant et al., ²⁴).

Acute MI rates (events/1000 patient years) were increased with advancing age in HIV compared with non-HIV patients. Grey line = HIV infected. Black line = HIV-seronegative control group. Both genders included.

HIV-Specific CHD Risk Factors

Whether HIV infection per se is an independent CHD risk factor, or the increased CHD risk is solely due to HAART, remains controversial. CHD risk is increased in autoimmune diseases like rheumatoid arthritis and systemic lupus erythematosus ^27,28, so HIV-related chronic proinflammatory processes and altered immune function may impart greater CHD risk than in the general population. Even prior to HAART availability, HIV infection was associated with a proatherogenic lipid profile; low HDL, elevated small dense LDL particles and triglyceride levels ^29,30. However, it may be difficult to delineate among CHD risk associated with HIV per se, HAART, and traditional risk factors, because the vast majority of the HIV-infected patients will receive HAART in the course of their disease.

Among traditional CHD risk factors, the prevalence and intensity of substance abuse (esp. tobacco and recreational drugs) are greater in patients with HIV than in the general population ^{6–8,11–13,31–33}. In HIV, HAART, other medications, and poor lifestyle/behavioral choices (diet, physical inactivity, and substance abuse) may adversely affect (directly and indirectly) traditional coronary risk factors: lipid/lipoprotein levels, regional adipose tissue distribution, and insulin sensitivity. The increased CHD risk seen in some patients on HAART may primarily be related to the effects of HAART on traditional CHD risk factors. But other, yet unidentified mechanisms, may also contribute to the increased CHD risk (e.g., direct effect of HAART on the vasculature)³⁴. Evidence from the D:A:D and other studies indicates that increased duration of protease inhibitor (PI) therapy leads to an increase in the rate of cardiovascular disease events ^20,35,36. HAART regimens are changed frequently and each drug within a class (nucleoside reverse transcriptase inhibitors, non-nucleoside reverse transcriptase inhibitors, protease inhibitors) is associated with very different effects on traditional CHD risks factors ^8,37–39. New anti-HIV drugs and drug classes (entry, fusion, integrase and maturation inhibitors) are continuously introduced. HAART prescribing practices are dynamic, variable and complicated, and it is not certain if they should be included in HIV-specific CHD risk prediction models. Such models are under development ^8,16,22. Conversely, we may need several, separate CHD risk algorithms for people living with HIV (i.e., HAART-naïve, history of HAART exposure, discontinued HAART, children, and adolescents). Finally, HIV is a global epidemic and CHD prediction models will need to be adapted to these markedly different geographical, ethnic and racial variations ^14,40.

Refining Existing Prediction Equations

The starting point for the development of a CHD risk prediction equation in patients with HIV is a comprehensive medical history and clinical examination with standardized collection of key predictor (independent) risk factors: age, sex, fasting lipids (total-, LDL-, HDL-cholesterol, total/HDL ratio), systolic blood pressure, history of diabetes mellitus treatment, fasting or postprandial glucose levels, use of tobacco and other substances ^1,4,28. Most existing equations predict coronary heart disease morbidity (new MI, ischemic events including stroke, coronary artery angioplasty, bypass surgery, carotid endarterectomy) and mortality (cardiac death). Such endpoints need to be defined a priori and captured in data collection (dependent variables). Baseline risk factors and treatments for hypertension, dyslipidemia, hyperglycemia and other conditions need to be captured. In particular, specialized models have been developed for persons with type 2 diabetes that consider additional potential predictor variables; hemoglobin A1c_c, microalbuminuria, and duration of diabetes ^41–43. In HIV, a challenge will be to allow for a higher prevalence of lipid- and glucose-lowering, and antihypertensive therapies, as well as HIV-specific factors (i.e. HIV RNA, CD4 cell count, duration of disease and therapy).

In prospective studies, selecting risk factors, and developing and assessing the utility of a new CHD prediction model typically involves a proportional hazards regression model. The utility of the model can be assessed using several different performance measures. Relative Risk. For each risk factor, proportional hazards modeling yields regression coefficients for a study cohort. The relative risk of a variable is computed by exponentiation of the regression coefficient, and is a measure of the increased risk for someone with a given risk factor (e.g., tobacco use) compared with the risk for someone who doesn’t use tobacco. Discrimination. Discrimination is the ability of a prediction model to separate those who experience hard CHD events from those who do not. This is quantified by calculating the c statistic, analogous to the area under a receiver-operator characteristic curve ⁴⁴. This value represents an estimate of the probability that a model assigns a higher risk to those who develop CHD within a specified follow-up than to those who do not. Calibration. Calibration measures how closely predicted outcomes agree with actual outcomes. For this, a version of the Hosmer-Lemeshow chi-square statistic is used ⁴⁵. The chi-square statistic is used to compare the differences between predicted and actual event rates. Small values indicate good calibration and values >20 indicate significant lack of calibration. Recalibration. An existing CHD prediction model can be recalibrated if it systematically over- or under-estimates CHD risk in a new population, but correctly ranks subjects in terms of CHD risk. For example, recalibrating the Framingham risk prediction equation would involve inserting the mean risk factor values and average incidence rate for the new population into the Framingham equation. Kaplan-Meier estimates can be used to determine average incidence rates.

New CHD Prediction Equations for HIV

In HIV, estimates of CHD risk have been reported and calibrated by the D:A:D investigators ^8,14,15,17 and others ^20,22,24, and new HIV-specific CHD prediction algorithms have recently been proposed ^16,21,31. The D:A:D-derived HIV-specific CHD risk equations incorporated protease inhibitor exposure and traditional CHD risk factors (e.g., age, tobacco use), and captured CHD events (incl. MI) as the dependent variable. Among 33,594 person-years, 157 CHD events occurred; the D:A:D equation predicted 153, while Framingham predicted 187 events. Framingham predicted remarkably well, but tended to underestimate CHD events in HIV+ tobacco users (Fig 2–3). The area under the D:A:D receiver-operator characteristic curve discrimination (ROC) statistic was 0.78 (95%CI = 0.75–0.82), and the D:A:D prediction equations were accurate in men vs women, and tobacco users vs non-users. Several lessons have been learned that can be applied going forward. The ROC discrimination statistic (0.78) was excellent and whether this can be improved upon, by including additional traditional or HIV-specific risk factors, needs to be evaluated. External validation of the D:A:D prediction model is warranted to determine whether it is generalizable among people living with HIV.

Fig 2. (From D:A:D ¹⁵).

Based on duration of HAART and the Framingham CHD risk prediction, CHD events were underpredicted in comparison to events observed in D:A:D. CHD events were MI according to the WHO-MONICA classification, monitored and verified by two independent physicians.

Fig 3. (From D:A:D¹⁶).

When subdivided on the basis of gender and tobacco use, Framingham CHD risk prediction appeared to under estimate observed CHD events in D:A:D for tobacco users, and over estimate CHD events for HIV-infected patients who reported never using tobacco. CHD events represent a composite endpoint including MI, invasive coronary artery procedure including coronary artery bypass, or angioplasty/stenting, or death from other coronary heart disease.

Finally, non-invasive, direct measures of carotid intima media thickness or brachial artery reactivity/dilatation have also been used to evaluate cardiovascular disease risk ^46–48 in HIV-infected adults and children. In theory, these measures may provide more predictive value for premature atherosclerosis and CHD in HIV. In support of this, one study using multivariate models has reported that Framingham risk stratification independently predicted carotid intima media thickness in HIV-infected adults ⁴⁶. This suggests that additional research should focus on the utility, sensitivity, and specificity of these and other non-invasive surrogate measures (coronary vessel imaging) of premature CHD in HIV, rather than waiting for the somewhat rare, but hard CHD outcomes like MI, stroke, endarterectomy, or cardiac death to occur.

Controversial Issues, Gaps in Knowledge and Future Research Priorities

Given limited data, the estimation of the relative effects of traditional risk factors on CHD outcomes appears similar between HIV-infected and non HIV-infected patients. To date, Framingham CHD risk predictions have performed reasonably well when applied to HIV-infected patients. There is an important need to identify HIV-specific risk factors, and test whether new HIV-specific CHD risk prediction approaches perform better than existing algorithms. In this regard, the following research needs and priorities were identified:

• Perform a large, multi-center, multi-national prospective study that captures a core set of traditional and HIV-specific variables that are all collected at standard intervals in HAART-naïve and treated HIV-infected children, adolescents, and adults to determine with certainty the optimal HIV specific risk prediction algorithm and the relative contribution of HIV-infection and HAART to CHD in this population. Ideally such a study or related studies would obtain fundamental data on (1). The relative contribution of HIV-infection (replication, altered immunity, and inflammation), HAART component(s), central adiposity, peripheral lipoatrophy, metabolic dysregulation to CHD risk in HIV. (2). The safety and efficacy of standard treatments (diet, weight loss, physical activity), and more intensive therapies (anti-inflammatory agents, glucose-, lipid-, and blood pressure-lowering agents, switching HAART regimens) for CHD in HIV.

• Consider the use of large administrative databases/registries as interim data sources to develop and refine HIV-specific CHD risk prediction equations, as a prospective study would require many years of follow-up. Participants should be enrolled from developed and resource-limited regions of the world where HAART regimens are more standardized.

• Develop novel database integration techniques to combine ongoing cohorts and re-calibrate existing models (D:A:D, SMART, MACS, VA, ART, ACTG). A focus of current HIV observational and cohort studies should be the implementation of standardized, prospective, validated CHD endpoints together with core HIV markers and treatment, and traditional CHD risk factors to allow such integration to be undertaken.