Statins by Numbers; The problem with numerically driven guidelines for cardiovascular disease

November 29, 2013

Statins by Numbers

By JASON KARLAWISH

PHILADELPHIA — MEDICINE is having its moneyball moment.

In his book “Moneyball,” Michael Lewis chronicled how the Oakland A’s, in order to identify the best predictors of a winning baseball team, used a highly formulaic, statistics-driven approach in place of the traditional assessments of coaches and managers. This month, in a similar spirit, the American Heart Association and the American College of Cardiology issued new, numerically driven guidelines for the treatment of cardiovascular disease.These guidelines recommend that doctors no longer use a patient’s LDL cholesterol level to decide whether to prescribe a cholesterol-lowering statin, and instead rely on the results of a web-based “risk calculator” — the Omnibus Risk Estimator — that determines a person’s chances of suffering atherosclerotic cardiovascular disease in 10 years.

Into the Omnibus Risk Estimator you enter nine variables, including age, sex, total cholesterol and systolic blood pressure, and the estimator returns your 10-year and lifetime risks of stroke, heart attack or death from cardiovascular disease. With these data, you and your doctor decide whether to invest in a lifetime of daily therapy with a statin pill.

This is a revolutionary shift. Once upon a time, medicine was a discipline based on the nuanced diagnosis and treatment of sick patients. Now, Big Data, networked computers and a culture obsessed with knowing its numbers have moved medicine from the bedside to the desktop (or laptop). The art of medicine is becoming the science of an insurance actuary.

It is also becoming big business. The developers of the World Health Organization’s FRAX calculator, which calculates your 10-year risk of major osteoporotic fracture, licensed it to General Electric, a manufacturer of bone-density measurement devices.

What is the problem with grounding medical practice in the cold logic of numbers? In theory, nothing. But in practice, as decades of work in fields like behavioral economics have shown, people — patients and doctors alike — often have a hard time making sense of quantified risks. Douglas B. White, a researcher at the University of Pittsburgh, has shown that the family members of seriously ill patients, when presented with dire prognoses, typically offer quite variable understandings not only of qualitative terms such as “extremely likely” but also of quantitative terms such as “5 percent.” We like our numbers, but despite our desire for better information and an ethic of “informed consent,” we don’t know how to use them.

Far more worrisome is where the numbers come from. Until the last decade or so, estimates of risk came from a doctor’s head. Now the numbers often come from a machine, which makes them seem objective and credible. But like Dorothy confronting the Wizard of Oz, we need to look behind the curtain. It seems that anyone with a Big Data set and a statistics software package can develop an algorithm, give it a user-friendly interface, and behold: Your future is foretold. It’s fast. It’s simple. But it’s opaque, and it may be wrong.

The Omnibus Risk Estimator is one of many available cardiovascular disease risk calculators. When you enter a patient’s data into them, you get a disturbingly wide range of results. Depending on which algorithm you use, you may need a lifetime of statin therapy. Or not.

Why the variation? Because the data sets used to develop the calculators themselves vary widely, and often are derived from populations that do not resemble that of the patient in question. This flaw is most pronounced when a calculator developed in one country is applied to patients in another country, where habits, health care and genetics can substantially differ. But it can be a problem even within a country. Over time, people and their habits change. The Framingham, Mass., of 1980 is not the Framingham of 2010. In general, people these days smoke less, gain more weight and take more medications. Numbers themselves may be precise, but the information they convey can be erroneous.

Calculators like the Omnibus Risk Estimator are simply tools, or devices, akin to the hip prostheses and pacemakers doctors implant in their patients. They are designed well, or they aren’t. But unlike other medical devices, which must undergo standardized testing and unbiased review and monitoring by the Food and Drug Administration, these risk calculators are developed with little regulatory authority over their design and use. There needs to be better oversight.

But even if professional and public action ultimately set higher standards, and even if we can get the numbers right, we must be mindful that these are calculators whose results prescribe patented, expensive drugs to millions of people. We have only to recall the successes of the Oakland A’s to know that whoever controls the numbers wins.

Jason Karlawish is a professor of medicine, medical ethics and health policy at the University of Pennsylvania.