Prediction of incident hypertension. Health implications of data mining in the ‘Big Data’ era

Hypertension is a leading cause of death and disability worldwide [1]. The number of people with hypertension is growing rapidly in part through population growth and economic advances enabling ‘Access to Excess’ calories, passive entertainment, and labor-saving devices. Hypertension prevention and control has important and growing implications for the health and economic wellness of people in virtually all nations. Novel strategies and approaches that more accurately identify individuals at risk for incident hypertension have value to the individual and to society. Using an atheoretical approach on a limited dataset, and with a thoughtful and compelling analytical approach, Völzke et al. constructed a model that predicted incident hypertension at least as well as if not slightly better than the Framingham risk equation for incident hypertension [2]. An atheoretical, data-mining approach to discovery is likely to grow rapidly in the years ahead to encompass virtually all areas of health and healthcare. With regard to previous approaches to hypertension prevention, in a hypothesis-driven model, which dominated the twentieth century, modifiable (lifestyle factors and biological systems, e.g. the renin–angiotensin or sympathetic nervous system) and nonmodifiable characteristics (e.g. age, race, and sex) variables that are plausibly linked to blood pressure and its (dys)regulation are identified. The relationship of these independent variables to the dependent variable of interest, in this case incident hypertension, is then assessed using various univariable and multivariable correlational and logistic regression models. Modifiable factors that are most strongly linked to outcomes then become the focus of intervention to change the natural history of disease with interventions. These interventions are then often focused on individuals at highest risk, based in part on nonmodifiable factors. The Trial Of Hypertension Prevention [3], TRial of Preventing HYpertension [4], and prevention of hypertension with the angiotensin-converting enzyme inhibitor ramipril in patients with high-normal blood pressure [5], were grounded in the theory based on previous studies supporting biological linkage between the intervention(s) and impact on incident hypertension. As a clinical investigator in the United States, focused on the integrative physiology of blood pressure regulation, who survived the decades of the 1980s and 1990s, I knew immediately a grant would not be funded if any reviewers perceived the application as a ‘fishing’ expedition. These review panels would never have funded Christopher Columbus to ‘discover the new world’, as he did not have any ‘new world’ gold in his application. In fact, in the 1980s and 90s, integrative physiologists rapidly became dinosaurs in the cellular and molecular biology era with its promise of simple molecular answers and treatments to efficiently prevent and manage complex physiological disease processes. Although not all of the promise of the cell and molecular approach has been realized, this era of medical science has and continues to provide important advances in the understanding, prevention, and treatment of disease. The hypothesis-driven approach to health and disease began to lose its dominance in the first decade of the third millennium Common Era as rapid and high-throughput technologies for sequencing the human genome emerged. Almost overnight, it was permissible and even highly laudable to do very large and costly single-nucleotide genetic expeditions, which had no parallels in the world of clinical integrative physiology. By this point, there was a tacit understanding that the promise of simple cell and molecular answers to complex biological disease processes was unlikely. In many cases, multiple gene–gene and gene–environment interactions were at least as complex as the physiologic disturbances these were intended to simplify. Molecular investigators quickly generated a load of single nucleotide polymorphisms, mostly in noncoding regions of the gene, and logarithm (base 10) of odds scores to correlate with various risk factors and diseases. The resulting hypothetical load provided ample speculation about how alteration of a gene product or its regulation could contribute to complex physiological disease processes. The second decade of the third millennium of the Common Era ushered in growing recognition of ‘Big Data’ and data mining. For this new science, not only are hypotheses not required, but these also are now viewed as constraining. The data mining subject of this article is not based in especially big data, but it is an example of what is likely to be a growing science that will impact hypertension and many other areas of health. This new science will have important and potentially very positive implications for health promotion and disease prevention. The era of ‘Big Data’ and data mining also raises potential challenges, for example, the rights of individuals and minority groups linked to worse health outcomes or higher costs, and the applicability of findings from one group or time period to another group or time period. An atheoretical approach magnifies existing concerns about causal rather than coincidental relationships between independent or predictor and dependent or outcome variables. ‘Big data’ in healthcare is a growing reality as personal healthcare information is aggregated at the national level in many countries. Healthcare databases will very likely grow much larger with time as high-throughput technologies provide more ‘omics’ (genomics, proteomics, lipidomics, metabolomics, etc.) information from various biological specimens. Linkage of healthcare data with social databases does or could include information on education, income, neighborhood characteristics, travel, occupation, marital status, lifestyle, spending patterns, technology preferences, and so on that is the 360-degree view, which includes social determinants of health. Private–public partnerships are emerging and will likely continue to emerge around new discoveries and insights leading to useful products and services from mining ‘Big Data’. To maximize the potential for positive and minimize risk for negative outcomes, thoughtful input from multidisciplinary teams representing four domains may be considered: ethicists, sociologists, religious leaders; informed healthcare consumers, community leaders; basic and clinical scientists, epidemiologists, behavioral scientists, clinicians, applied and theoretical biostatisticians, health economists; and policy makers and elected officials, leaders in business and finance. Guidelines developed by these teams that balance opportunities for discovery and effective use of healthcare resources to optimize health outcomes for the population with the issues of individual rights and autonomy will be important. Effective team members will understand the strengths and limitations of data mining and the need for complementary and convergent strategies. They will realize no individual or group has all the answers and will appreciate the richness and value of diverse expertise in maximizing benefits and minimizing risks. The world in which we live is ever more interconnected. ‘Big Data’, growing to an estimated 5247 gigabytes for each of earth's human inhabitants in 2020 [6], and ‘Data Mining’ [7] are now part of our world. With all advances, there is potential for positive and negative outcomes. The vast majority of individuals in medicine and science desire positive outcomes from their life's work. A broad-based, proactive and interactive approach of the medical and scientific community will help ensure that Big Data Mining is a constructive tool for advancing the health and well being of people internationally. ACKNOWLEDGEMENTS Conflicts of interest There are no conflicts of interest.