Modeling the impact of cardiovascular prevention strategies

Cardiovascular events remain a highly prevalent cause of premature death and disability. Among the various strategies aimed at decreasing the risk of these events, blood pressure (BP) lowering is among the most efficient. The level of evidence for the benefit to risk ratio of the various drugs is growing, and guidelines need frequently to be updated to take into account new evidence in the field. Guidelines do not take into account cost aspects explicitly, even though the availability of efficient low-cost drugs and the implicit consideration of economic aspects are likely to explain some differences in trans-Atlantic positions, for example, regarding the choice of first-line drugs. One of the reasons is the dependency of the cost estimates to country-specific features. Most economical modeling works for these drugs are ordered by pharmaceutical companies to complete the files for prize negotiation with health authorities once the marketing authorization is obtained [1,2]. The aim is then to illustrate some economic advantage of choosing the new drug at a given stage of the management of hypertension. A second important use of economic models is to explore the advantages and pitfalls of different strategies. For example, Marshall [3] explored the impact and the cost-effectiveness of different drug strategies to reduce coronary events for a given profile of risk; Bibbins-Domingo et al.[4] illustrated how modest reductions in dietary salt could reduce cardiovascular event rates in the United States at lesser costs than BP-lowering drugs; Chamnan et al.[5] advocated a risk-based approach compared to a global one for screening people in cardiovascular prevention; and Marchant et al.[6] computed the changes in the number of patients targeted according to successive issues of guidelines regarding hypertension and the resulting changes in the public health impact, they also compared the size of the population to be treated to the actual size of the treated population by age and sex in France. In this issue of the Journal of Hypertension, Dodhia et al.[7] propose a method to compare different evidence-based interventions when scaled-up to the best quality implementation compared with the current situation with the purpose of informing health policy and research. They explore various interventions such as salt reduction through food industry collaboration, salt and diet modifications and improvement of BP-lowering drug use through the application of current guidelines, as well as some changes in the perimeter of the drug target population. They illustrate the public health impact of different scenarios on the burden of disease, computed from the number of prevented events weighted by an average score of disability. Costs were computed from the healthcare perspective. The scenarios simulation was applied to a few age groups summary statistics, representative of the English population over 16 years of age. They explore the sensitivity of their results to some hypotheses regarding drug costs, level of detection, treatment and control of BP or the efficacy of diet sodium lowering. They conclude that important health gains are possible and, most of all illustrate, how their approach could serve in choosing which health strategies should be promoted. A particularity of such modeling approaches is, among other obvious advantages, their capacity to be nurtured with data from various origins: here Dodhia et al. explored the disability weight estimated in the Netherlands, applied to summary characteristics from an English epidemiological survey completed with national death statistics. An important choice has to be made in the simulation of the public health impact of cardiovascular prevention strategies, regarding the estimate of drug efficacy. This efficacy could be predicted either through the reduction of the target risk factor [e.g. BP, HbA1c or low-density-lipoprotein (LDL) cholesterol] or through the risk reduction observed under treatment in clinical trials and meta-analyses. The first option is logical when the simulation deals with a new drug without available estimate of its impact on cardiovascular morbidity and mortality. However, it relies on strong assumptions regarding the risk factor to risk relationship, within a therapeutic model. This association could be derived from epidemiological data, which is a simple method but certainly wrong, assuming that the epidemiological relationship between BP and subsequent risk, observed between individuals, is equivalent to the therapeutic relationship, within individuals. Such an assumption would lead to define BP target as less than 120 mmHg for everybody, as individuals with a spontaneous BP less than 120 mmHg have the lowest risk, whatever the age or sex. A more satisfying approach for estimating treatment efficacy according to the first option is to predict risk reduction from the BP reduction in applying the relationship observed in clinical trials meta-regressions, such as that proposed by Staessen et al.[8]. However, this first option relies on the paradigm that all risk reduction from antihypertensive drugs is derived from BP lowering, which is not true, from a systematic review on individual patient data [9]. Here, Dodhia et al. made the second option choice, taking the average event reduction adopted by the National Institute for Health and Clinical Excellence (NICE) guidelines, without including BP reduction in the computation. This solution is undoubtedly robust, but does not take into account the variability of responses between individuals regarding BP or cardiovascular events. In addition, they limit the modeling process to the first-line single drug and assume that BP control is required to obtain the benefit from drugs, which is certainly partially wrong. In some settings, it might be difficult to translate treatment effects from randomized controlled trials into clinical practice. For example, in the studies showing significant effects of salt restriction on BP, low-salt food and snacks were cooked and distributed to the participants. Such an intervention is very expensive and not possible to apply in clinical practice. The effect of less intensive (and cheaper) interventions is, thus, much smaller. Another important choice in modeling public health impact is to work at summarized profile levels, capturing all the information from a category of the population (e.g. age and sex categories) or at the individual level. Some tools have been developed, applied in the modeling of therapeutic strategies in hypertension [6], or in diabetes [10], which offers some advantage. These tools take into account the diversity of a given population, reproducing specific features such as the demographic characteristics, the distribution of the risk factors of interest and their covariations [11]. This realistic approach offers the best fit of a simulation to a given real population. As a natural counterpart, its results are more specific and less transposable to other settings. However, the costs and effects of health policy interventions such as educating general practitioners and impinge on the food industry will have to be further explored. The development of computer performances allows mixing systems biology approaches and simulations at realistic population levels. This will make it possible if not easy in a near future to test in silico different strategies, using physiological, pathophysiological and pharmacological complex but relevant features, taking into account social determinants of patients’ compliance and medical behaviors, the impact of health authority decisions or social security reimbursement measures. Achieving such ambitious goals would need to develop platforms to share data related to complex physiological models, pharmacokinetic and pharmacodynamics information and clinical trial results at the individual patient level, as well as demographic, social and behavioral information. The development of interfaces between databases of real prescriptions, high level of evidence information provided by randomized clinical trials, will offer dramatically important solutions in the validation process. This validation will have to make these simulations not only realistic but also reliable. These achievements would give their full dimensions to the purpose of Dodhia et al. of informing health policy and research. Acknowledgements This work was supported by the AgenceNationale de la Recherche through a grant for the BIMBO project, SYSCOMM Nr 002, ANR http://www.agence-nationale-recherche.fr/. Conflicts of interest There are no conflicts of interest.