Though much attention has been paid - here at Epidemix and elsewhere - on the power of genomics as a predictive tool for disease, there are other approaches to forecasting risk that are potentially more helpful, equally bold, if somewhat less sexy. I had the chance a couple weeks ago to learn about one: a new predictive test for diabetes risk developed by Tethys Bioscience. It is a cool tool, and I think it represents a new breed of diagnostics and predictive testing.
The idea behind the Tethys test, called PreDx, is to create a tool that can accurately identify those at increased risk for developing type II diabetes. Diabetes, we know, is one of the fastest growing diseases in the country (and world), accelerated by the upsurge in obesity. 24 million Americans have diabetes, with another 2 million cases diagnosed annually. 60 million Americans are at high risk of developing diabetes, many of these obese or overweight.
The traditional tool for diagnosing the disease, as well as for diagnosing a *risk* for developing the disease, is a blood glucose test. The so-called "gold standard" test, a fasting blood sugar value of 140 mg/dl constitutes diabetes, while normal levels run between 70-110 mg/dl. You can see the issue here: What does a value of between 109 and 139 mean? This is the problem with these firm cut-offs - their you-have-it-or-you-don't nature means that you're failing to capture people until they have a disease. We're missing the opportunity to get ahead of illness and maintain health.
In the last couple years several genome-wide association studies have identified certain genetic variants with diabetes, to great fanfare. But the problem with these associations is that the rest of the puzzle is, to mix metaphors, blank. We don't yet know what context these associations exist in, so the seven or eight markers that have been identified may be the complete span of genetic influence, or they may be seven or eight of 1,000 markers out there. In other words, there's lots of work to do there still.
A more traditional attempt at early detection has been the diagnosis of metabolic syndrome, which I've written about lots. In a nutshell, metabolic syndrome is an attempt to establish some cutoffs - from glucose, blood pressure, waist circumference - to define a disease that's a precursor to other disease (namely diabetes and heart disease). It's an ambitious extrapolation of our ability to quantify certain biological markers, but it's inexact and, the argument goes, hasn't proven any better at actually identifying those at risk than blood glucose alone. In other words, it has defined a pre-disease state without actually changing the outcome (at least, that's the argument; it is a subject of great debate).
OK, so that's the backdrop: a single conventional test that identifies disease better than risks, an emerging but incomplete measure of potential risk, and a measure of pre-disease that has ambiguous impact on the disease. So how about something that identifies risk accurately enough and early enough and strongly enough that it actually impacts the progression towards disease?
That's the idea behind PreDx. The test itself is an ELISA test, which for you microarray junkies may seem disappointing. ELISAs are nothing fancy, they've been used for nearly 40 years to detect proteins. The cool part, though, is what goes into that test. Tethys scanned through thousands of potential biomarkers that have been associated with components of diabetes - obesity, metabolic disorder, inflammation, heart disease - and settled on a handful that all closely correlate with diabetes. That's the ELISA part, testing for levels of those five or so biomarkers. Then second stage of the test is the algorithm: a statistical crunching of the various levels and presence of those markers, to arrive at a Diabetes Risk Score.
The DRS is a number between 1 and 10, shown to the tenth of a point, that equals a risk for developing type II diabetes over the next five years. a 7.5 equals a 30% risk of developing diabetes within 5 years, a 9 equals a 60% risk. (the risk for the general population is about 12%, equal to a 5.5 on the PreDx scale)
So what's cool here is the algorithm. Unlike many new diagnostic tests, the smarts aren't necessarily in the chemistry or the complexity of the technology (it's not quantum dots or microfluidics or stuff like that). The smarts are in the algorithm, the number crunching. Basically, the test lets the numbers do the work, not the chemistry.
At $750 a pop, the PreDx test is too expensive to be used as a general screening test - it's best used by physicians who've already determined their patient is at an increased risk, through conventional means. Tethys says that'll save $10,000 in healthcare costs on the other end. In other words, it's a way to pull people out of that pre-diabetes pool, spot their trajectory towards disease, intervene, and avoid onset. In other other words, it's a tool to change fate. Which is kinda impressive.
Another interesting thing here is the simplicity of the 1 to 10 scale. Obviously, this is the work of the algorithm; the actual data doesn't neatly drop into a 4.5 or a 7.3 figure, it must be converted into those terms. That in/of itself is a complicated bit of biostatistics, and it's beyond me to assess how they do it. But the fact that an individual will be presented with a Diabetes Risk Score of, say, 8.1, and then shown a chart that very clearly puts this at about a 40% risk of developing diabetes - well, that's a lot easier for a lay person to make sense of than a blood glucose level of 129 miligrams-per-deciliter. Heck, it's a lot easier for a *physician* to make sense of.
What's more, this is a quantification of *risk*, not a straight read of a biological level. That's a very different thing, much closer to what we want to know. We don't want to know our blood glucose level, we want to know what our blood glucose level says about our health and our risk for disease. The closest physicians can usually get us to *that* number is to go to general population figures - in the case of diabetes, that 12% risk figure for the general population.
What PreDx represents, then, is how we are moving from a general risk to a personalized number. This isn't the abstract application of population studies to an individual, it's the distillation of *your* markers, using statistical analysis to arrive at an individual risk factor.
So I find this pretty compelling. Tethys is developing other predictive diagnostic tests for cardiovascular disease and bone diseases, but far as I know, there are not many similar predictive diagnostic tests out there, for any disease. As mentioned, we have the genomic assocations, which are coming along.
Anything else somebody can clue me in on? Lemme know...