9.

Precision Medicine

Everything That’s Wrong with You in Particular—a Statistical Approach

Before the nineteenth century, getting medical treatment kinda sucked. Anesthesia consisted of a shot of whiskey (or two), and medicines consisted of traditional cures, like bloodletting and hedgehog grease, with no empirical validation. In the modern world, we rely on the fact that practicing doctors have a deep alliance with scientific researchers. Researchers test treatments to see what works, and doctors weigh the evidence when deciding what to do for patients.

A good doctor is a sort of a Sherlock Holmes of your body. Whether you are sick or well, your body is constantly giving off little clues to your internal state. Some of these clues are quite obvious—if there’s a giant hole in your head, the doctor can be pretty sure of what ails you.

Other cases may be very difficult to pin down. For example, most patients diagnosed with mononucleosis only get that diagnosis after many other incorrect ones.* This isn’t really the doctor’s fault—mononucleosis is a viral disease that is quite long lasting, and its symptoms are general things like fatigue, headache, and sore throat. It’s sort of like a constant mild hangover, which makes it very hard to differentiate from baseline human existence. But if doctors first checked what molecules were prevalent in your bloodstream, they might diagnose you with mono instead of a cold.

What doctors have long called “symptoms” or “signs” more computationally inclined researchers now call “biomarkers.” A biomarker, broadly defined, is just about anything that tells us about someone’s internal state, usually in reference to whether something in there is going wrong. Most commonly, biomarkers are traditional symptoms as well as chemical cues in your body, but some researchers think the term can be broadened to encompass behavior patterns, like what Web sites you browse or what images you post online. So to speak, if you have a computer model called “How Am I Doing?” a biomarker is anything you might input into that model that would help it find an answer.

Just as the coming together of science and medical practice brought about modern medicine, the coming together of medical science with molecular analysis, data science, and machine learning may bring about a new paradigm, which is coming to be called precision medicine. In the future, you may get medical diagnoses that are determined quickly and correctly from thousands of biomarkers, followed by treatments that are tailored to you in particular. This means you will live longer, live healthier, and—if the detection systems get cheap and easy enough—you don’t spend nearly as much time wondering if that bump on your right butt cheek is cancer. Not to mention, if diagnosis and treatment of disease becomes a matter of computer power, it has potential to (for once in our lives) drive down the cost of health care.

In a single drop of your blood there is a staggering amount of information. There may be chemical biomarkers associated with looming heart failure. There may be genetic code from an undetected solid tumor. There may be hormonal biomarkers that tell us that you’re more stressed than you realize.

A deep knowledge of these subtle biomarkers not only leads to better diagnosis, but it also suggests novel treatments. If you have cancer, that cancer has certain genetic mutations we can locate. Having located those mutations, we can pick the best treatments, or go “off the shelf” to pick a treatment that might work even though it’s not designed for the disease you’re experiencing. With the very latest methods, we might even be able to create a general technique for dealing with any disease arising from genetic mutation.

As we get more data on all the variety and complexity of human bodies, and as we get better at analyzing that data, we approach a time when a computer delivers a perfect diagnosis and selects the ideal treatment method. This dream may seem far off—and at least some aspects of it are quite a ways away—but remember, the human body is of finite complexity.* Every advance brings us closer to the finish line.

Where Are We Now?

To get a sense of how the field has changed in the past fifty years, Kelly sat down with MD Anderson Cancer Center’s Dr. John Mendelsohn.

Let’s imagine, hypothetically, that you didn’t do your research on Dr. Mendelsohn because a trusted colleague told you this doctor was the perfect person to talk to about precision medicine. If you were that foolish, you would still quickly learn how important Dr. Mendelsohn is because you would look up his address after arriving at the MD Anderson Cancer Center campus and learn that his office is in the John Mendelsohn Faculty Center Building. At that point you would, hypothetically, do a panicked Google search for “John Mendelsohn,” and discover that he is, in fact, the former president of MD Anderson Cancer Center.

You would take a hypothetical deep breath, discover that your armpits were now soaked with sweat, and resolutely knock on his door.

Fortunately for hypothetical you, Dr. Mendelsohn is a friendly and welcoming man. After chatting for a while about Kelly’s research, he then gave thirty minutes of time that probably should have been spent saving the world. Sorry about that.

When Dr. Mendelsohn was born, scientists didn’t know that our genetic material was made up of DNA. When he was in med school the description in the textbooks for how proteins were made was totally wrong. When he was a young researcher, the amount of data you got from a single experiment was primitive by modern standards.

“If you sequence one human genome you’re getting five billion pieces of data. When I began doing my own research, the results could be printed out at first on a piece of paper, and then a long sheet of paper. You can’t print out five billion pieces of information and analyze it in your brain. Today at the MD Anderson Cancer Center we are sequencing the DNA of thousands of patients’ cancers each year to detect the genetic aberrations that are causing their malignancy.”

As medicine becomes more and more personalized, and individual bodies become wellsprings of data, it may be possible to find diagnoses and even cures simply (“simply”) by combing through the information for patterns. But we’ll only be able to benefit from precision medicine if we collect data from lots of people over time and find ways to store and analyze all of this data, which is going to require a lot of innovation by computer scientists and statisticians.

The National Institutes of Health recently launched the Precision Medicine Initiative Cohort Program. The program is going to collect health and environmental information from over one million participants, including data on their “-omes.”

We’re going to tell you about a bunch of “-omes” in this chapter, so here’s a quick definition—when a scientist adds “-ome” to the end of a word, she means “like . . . all of it.” So a geneticist studies particular genes while a genomicist studies all of the genes.* Like . . . all of them.

This may make it sound like the genomicist is just smarter, but it’s kind of like the difference between a psychologist and a sociologist.* The -ome suffix is just a currently fashionable naming convention.

The National Institutes of Health will get these individuals’ genomes, microbiomes, and other information, and will then follow them over time to track changes in their health. The data will be available to doctors, who can comb through them for associations among disease, genetics, and environmental factors. This is going to create a massive data set.

Unfortunately, big piles of data don’t just leap up and tell you what’s going on. This is a problem that generally worries Pfizer’s Dr. Sandeep Menon. “The amount of that data that is coming in is pretty much exponential, and . . . the people who are available with the right skill set to analyze is very minuscule. To put it bluntly—the demand is more than the supply.”

Dr. Menon is an elite biostatistician. He is vice president and head of the Biostatistics Research and Consulting Center at the largest pharmaceutical company on Earth, and even he finds it “challenging to keep up abreast of the evolving latest techniques to navigate the deluge of data.”

What concerns Dr. Menon is that there are a lot of analysts who simply don’t know how to handle their data. He thinks a lot of mistakes are being made, which slows the advance of the field and potentially harms patients.

Despite the difficulty of sorting through all this information, progress is being made on many fronts. The field of precision medicine is too enormous to capture entirely, so we’ve elected to give a bunch of specific examples that should provide you a taste of what precision medicine can do.

Genetic Disorders

The rapid advance in gene-reading technology means that getting your genome sequenced is relatively cheap—in the range of thousands of dollars, where it once costs tens of millions. But being able to diagnose a problem doesn’t equal being able to cure it.

Genetic disorders are especially hard to fix. Contrary to how it’s often stated in the media, you can’t really “edit a person’s DNA” the way you edit a document. Almost every cell in your body contains strands of DNA. To alter your genome, you’d have to alter all of your cells, or at least all of your cells that are relevant to your disease.

For example, cystic fibrosis is a genetic disorder where the body produces a lot of thick mucus in internal organs. Thick mucus in the lungs is about as fun as it sounds—it increases the risk of infection while making breathing difficult.

Another problem is mucus in the pancreas. It almost rhymes, but other than that it’s terrible—it makes nutrient absorption much harder. Diagnosis with cystic fibrosis actually does rhyme, but it once meant that you were unlikely to survive far into your twenties. Advances have gotten patients into their thirties and forties, but these advances are all ways to treat the mucus problem rather than fixing it at its genetic root.

Part of the difficulty is that there isn’t just one cystic fibrosis. The mucus buildup that identifies the condition to doctors can be the result of any of a number of genetic mutations.

A drug called ivacaftor was recently developed to deal with a specific genetic variant of cystic fibrosis. The specific mutation that the drug targets is found in only about 5% of cystic fibrosis sufferers. In medicine as usually practiced, that’s not great. But in a precision medicine paradigm, the idea is that we eventually have a treatment that targets every possible mutation. So when you are born, we identify your genetic problem and we know exactly what treatment to use. This means you not only get the right treatment, but you avoid a bunch of potentially unpleasant wrong treatments.

But this is still only palliating the problem at a deeper level. What if we could just fix the damaged code in all the right cells?

In the last chapter, we discussed a new gene-editing technique called CRISPR-Cas9. It may allow scientists to actually fix the mutations causing cystic fibrosis in patients with the disease. In case you have already managed to forget what CRISPR does, the short version is that we can now precisely snip out and replace parts of DNA in cells. In principle, this should mean we can snip and fix whatever mutations are causing cystic fibrosis in a living person.

CRISPR has already worked in slabs of intestinal tissue in the lab, which gives you some idea about how fun medical lab work is. Figuring out how to apply the technique in actual patients is still a big hurdle, and scientists are worried that we might mess up other parts of the genome while trying to fix the mutations. When you’re editing a trillion cells at once, you don’t want to have too many whoopsies.

But the cool thing about CRISPR is that it’s a general tool for fixing genetic disorders. Any disease that is caused by one or more genetic mutations should be vulnerable to this method of targeted gene edits. If CRISPR ends up being a silver bullet for gene problems, you could fire it at Huntington’s disease, sickle cell anemia, Alzheimer’s disease, and more.

Cancer Diagnosis, Treatment, and Monitoring

DIAGNOSIS

Cancer cells are hard to kill for the same reason secret androids are going to be hard to kill in the Robot Uprising of 2027: They look just like us.

A cancer cell is one of your own cells gone bad. It is a cell that should’ve served some useful bodily function, but happened to be born with or acquire a weird set of mutations that makes it generate copies of itself over and over instead of doing its job. In the case of a solid tumor, you’ve basically got a nation of bad cells living and reproducing in your body.

Mutant cells, including cancerous ones, are born in your body all the time. Typically your immune system targets and kills them. The problem is that now and then you get a very rare cell, which (1) reproduces out of control, and (2) evades the immune system, either by avoiding detection or by convincing your immune cells not to kill it.

So by the time you get a cancer diagnosis, you’ve got very dangerous cells indeed. At that point, medicine has to step in where your immune system has failed. But catching cancer can be hard.

Historically, leukemia is one of the easiest cancers to detect, because it’s blood borne and leaves a telltale buildup of white blood cells.* Solid tumors, especially small ones, can be much more stealthy. That’s why doctors ask patients to do breast exams regularly—even hard tumors can be subtle when they’re hidden in a squishy human body.

Early diagnosis is more than just a convenience. Many of the deadliest cancers are dangerous not because they are especially aggressive, but because they only cause subtle symptoms until it’s too late to stop them.

According to the National Cancer Institute, there is a 55% chance of surviving lung cancer for five years if you detect it early on. But most people don’t get a diagnosis that early. Over half of lung cancer patients aren’t diagnosed until the disease has metastasized,* at which point the five-year survival rate is about 5%.

So we want to find cancers as early as possible.* And it turns out that leukemia isn’t the only cancer that leaves a biomarker in blood. All sorts of cancers can be detected by searching for little molecules called microRNA.

In the last chapter, we gave an approximate sense of how DNA creates protein, but we mentioned that the actual process can be more complicated. One layer of complexity is added by microRNA.

MicroRNA is not yet perfectly understood, but one important role it appears to play is in what’s called gene expression. Think about it like this: Imagine you have a gene that codes for a glowing red nose. As we said, “codes for” is not a great way to say this, so let’s be more specific: Your DNA encodes a protein that automatically goes to the tip of your nose and glows red. How red your nose becomes depends on how many of these proteins are created. MicroRNA can make adjustments here. Suppose your genetic recipe for our imaginary red-nose protein normally ends with “repeat this 10 times.” That “10 times” can be adjusted up or down by microRNA, giving you either a pale, sad nose or a ruddy, vibrant nose.

Okay, so that’s neat, but so what? Conveniently for medicine, these little microRNA molecules can be found in your bloodstream. Particular microRNA bits, or altered concentrations of them, can tell us not only what cancers you may have, but also what stage those cancers are in.

For example, one study found that levels of four specific microRNAs were a strong predictor of whether someone with lung adenocarcinoma was likely to live a long time (over four years on average) or a short time (a little over nine months). Information like this can help patients and doctors decide how aggressively to approach cancer, and can help patients make decisions about how to live their remaining life spans.

In principle, by knowing what microRNA is in your blood (and maybe a couple of other fluids), we can get a sort of readout on what diseases your body carries. New proteins get created in response to just about anything your body does, so this could be an incredible source of info on exactly what your major malfunction is.

But figuring out what’s wrong with you isn’t an easy task. Next time you’re looking for some reading material, consider miRBase, the microRNA data base (mirbase.org), which as of this writing is tracking around 2,000 such molecules.

Another molecule of interest is called ctDNA, short for “circulating tumor DNA.” It’s a very new discovery, and potentially a very big deal for cancer diagnosis. In simple terms, when you have some types of solid tumor, a little of its DNA can make its way into your bloodstream.

This isn’t a great thing for you, but it’s very handy for your doctor, for two reasons: (1) It makes it very hard for solid tumors to escape detection, and (2) it means genetic analysis of a known cancer can be done without having to perform invasive surgery.

A recent study showed that if you have stage 1 non-small-cell lung carcinoma, we can already detect it in the blood via ctDNA 50% of the time. By the time you’re at stage 2, we can detect it 100% of the time. By this stage, the cancer has moved from the lungs to the lymph nodes, but has not metastasized to other organs. Usually, we don’t catch this disease until stage 3 or 4, when five-year-survival probabilities are significantly lower.

But even if you can find ctDNA and microRNA cancer signatures, it can still be hard to get the whole story on your disease.

We tend to talk about cancer in terms of where it’s found—liver cancer, bone cancer, brain cancer—but this is not the most useful way to describe it. Two types of breast cancer may come from completely different mutations. One may be deadly while the other is fairly manageable.

Making things more complex is the fact that cancer doesn’t stop mutating once it exists. As cancer cells continue to mutate, a sort of survival of the fittest happens in your body. The result can be tumors that are not only dangerous, but also genetically diverse.

The diversity of cancer genetics in a single body can render treatment extremely difficult; even if you have a chemical that makes the tumors shrink, you may be reducing only a subset of the cell types. So if you shrink a tumor, you may have killed only a certain type of cells that were vulnerable. Later, the cancer may come back more aggressive than ever.

Worse, if you receive chemotherapy or radiation therapy, you may have created further mutations in the process. And that’s not to mention that going through those treatments often really sucks, partially because they are not targeted just to cancer cells. For instance, the traditional chemotherapy treatments target cells that are dividing too fast. But some of your cells, such as stomach lining, are supposed to divide fast. It’s sort of like cutting down on citywide nefariousness by eliminating everyone with a long black mustache. Sure, you’re mostly getting rid of villains, but you’re also killing the friendly hipster who makes the best coffee in town. A worthy trade-off? Maybe. But not a pleasant experience.

To defeat cancer, you want to figure out the right treatment early on, so you can avoid giving cancer time to mutate too much. This may mean getting yearly blood tests to detect cancer early, and determine what mutations it carries. This will be important, because it means you can pick the right therapies. Picking the wrong therapy isn’t just painful; it’s dangerous. You want to know all the relevant mutations so you can make the right cocktail of drugs to treat them.

TREATMENT

One of the reasons cancer cells are so dangerous is that they evade your immune system. It’d be sort of like if there were killer robots on the loose with no conscience, but the ability to mimic humans. But what if we could train the police to recognize these creatures? Like, hey, that one guy who keeps saying “Affirmative, Human” and works as a lobbyist—maybe we should watch him more closely.

Just so, what if we could teach your immune system to target and kill sneaky cancer cells?

It works like this: You take some blood from your patient, and there you find these immune cells called T cells. The T cells we are interested in are a type that recognize structures on the surface of cells, called antigens. By looking at what type of antigens a cell has, a T cell decides if the cell needs to die.

If we know what antigens your cancer cells have, we can teach your T cells to target them.

Why T cells? We’ll let Dr. Marcela Maus of Harvard Medical School and Massachusetts General Hospital explain: “There’s two things that really make T cells special when it comes to immunotherapy. . . . They have the capacity to kill other cells . . . and . . . they have memory. They’re very long-lived cells. And once they’ve seen something once, they’re quicker to recognize it the second time and kill it even faster.”

A particularly successful genetic modification has been to teach T cells to go after a molecule called CD19, which is found on a type of white blood cell called a B cell. Leukemia and lymphoma often kill you by overproducing these white blood cells.

One problem with this approach is that the T cells often end up killing all the B cells, including the ones that aren’t cancerous. B cells are part of the immune system too, so killing all of someone’s B cells can make them temporarily immunocompromised. This isn’t awesome, but fighting off an infection is generally preferable to fighting blood cancer.

But what about tumors for which killing all of a potential cell type just isn’t an option? Defeating a brain tumor by killing all brain cells is a bit of an empty victory.

Dr. Maus has a more subtle approach. She engineers the T cells to attack an antigen that goes by the charming name epidermal growth factor receptor variant III, which we’ll refer to by the easy-to-remember acronym EGFRvIII. No normal brain cells have EGFRvIII, but some tumor cells do. So if you are “lucky” enough to have a tumor with this particular receptor, T cells can be programmed to find and kill them in particular.

Dr. Maus’s research is in an early phase, but there is hope that immunotherapy will prove to be a good method for precisely attacking multiple kinds of solid tumors in the future.

MONITORING

If the treatment works, your cancer goes into remission. But at that point you’ll still be monitored for the rest of your life. As it happens, being “cured” of cancer is a very good indicator of cancer risk in the future. Precision medicine techniques offer ways to do a better job of monitoring these patients.

For example, if your cancer is in remission, we might still watch for ctDNA, to make sure it’s not subtly returning. We might also continually genotype this ctDNA to watch for mutations or a change in prevalence of a certain type. If the ctDNA signature suddenly starts indicating more aggressive types of cancer, it may change the most desirable medical approach.

This new technology is improving rapidly, and we now find ctDNA in stool, urine, and other fluids. It may soon be the case that we can do a much better job of monitoring you for cancer, and we can do it by poking through your poop instead of your body.

The Metabolome

One hard thing in medicine is knowing how a particular patient will react to a drug. What works well for this person may work poorly for that person, and some of this is explained by the patient’s “metabolome.” The metabolome is all of your metabolites: all the small molecules that the machinery of your body operates on, like sugars and vitamins. The metabolome is not a simple system—the Human Metabolome Database (hmdb.ca) is currently tracking about 42,000 different metabolites. That’s 42,000 types of little molecules. And that doesn’t even count the proteins and other “large” molecules that operate on them, making sure you continue to convert cheeseburgers into the energy and muscle required to get more cheeseburgers.

It turns out people have a good deal of metabolomic variation. This may explain why coffee keeps you awake at night, while your friend can have a double espresso in bed, then sleep soundly. You’re processing the coffee chemicals in a different way. These differences can tell us about your internal state. For example, we’ve known for about seventy years that you can tell if someone is diabetic by checking his glucose level. If he isn’t metabolizing glucose well, you’ll find elevated levels. But with 42,000 metabolites floating around, we might be able to get more information on exactly how different people vary in their metabolic abilities.

For example, some suicidally depressed patients do not respond to any drug regimen. It’s possible that something about their metabolome is preventing or altering proper drug uptake. In a precision medicine paradigm, we might be able to determine in advance what the patient will fail to metabolize.

Metabolic information may also help human nutrition. What poisons one person may have no effect on another. This is especially important when the particular poison is tasty. For example, some people have conditions that result in hypocholesterolemia, meaning that for some reason they just don’t have as much cholesterol in their blood. This may mean that they can eat more of the delicious high-cholesterol foods that you and I are told to avoid. Maybe there are similar genes that allow people to smoke, or drink to excess, or, like . . . chew glass.* Flipping things around, a precision look at a patient may tell you that she should avoid certain foods and activities, even if they’re generally considered healthy. Maybe your mom really was wrong to make you eat broccoli.*

Knowing your metabolome may be especially important for patients who are in a dangerous medical situation. Ideally, any medicine with potential side effects should be given at the smallest possible dose that will work. But if you have a patient with an unusual metabolome, a small dose may act like a large dose, or a large dose may not work at all. If you are choosing among drugs for a patient, knowing how they’ll metabolize it may allow you to avoid painful treatments and go straight to the right stuff.

Other Things That Are Trying to Kill You

In the above sections, much of what we’ve talked about has been cancer and genetic issues. This is because these diseases are some of the most difficult to battle, and we’re just now starting to get some major wins. But precision medicine techniques should be applicable to just about anything.

For example, stress and hypertension can cause a condition called cardiac hypertrophy, which is a dangerous thickening of heart muscles. This puts you at risk for things like fatigue, headache, and suddenly dying. Given that heart muscles are made of protein, it might not surprise you that there is a microRNA signature for cardiac hypertrophy. In fact, there are a number of microRNA signatures for various types of cardiac hypertrophy. Looking at microRNA in blood gives us a noninvasive way to determine exactly what sort of cardiac hypertrophy you’re facing.

In other words, we might be able to predict that your heart is going to fail before it happens. Given that heart disease is the number one killer worldwide, a bit of a heads-up would be appreciated.

This ability to monitor blood-borne chemicals in depth may also be helpful for patients who are taking dangerous treatments. For instance, suppose you have a patient with a history of heart disease who now needs to take a chemo regimen for cancer. There is an elevated risk of heart failure, and by the time the patient is having a heart attack, it may be too late. With a microRNA analysis, you could monitor the effects of chemotherapy on heart muscles in more or less live time, then make choices accordingly.

Other recent studies have shown that there is a microRNA signature for stroke. There are also microRNA signatures that tell us what your brain is up to as you recover from stroke.

Molecular biomarkers can tell us about other diseases, like Crohn’s disease, Alzheimer’s, and even what sort of flu you may have. In fact, if you go to Google Scholar and type in “microRNA profile,” you’ll find an enormous number of papers written just in the past few years, which show that there are signatures for just about every disease from prostate cancer to depression. And not only can we detect many diseases, we can often tell what those diseases are up to.

Some researchers want to know not just the molecular picture or the clinical picture, but also the behavioral picture. Basically, they want the sort of data a creepy stalker would know about you—what TV you’re watching, what Web sites you’re browsing. This information may clue us in to whether or not you’re suffering from some types of diseases. For example, Dr. Andrew Reece of Harvard and Dr. Christopher Danforth of the University of Vermont found they could predict whether an individual was depressed based on the color and brightness of photos the individual posted on Instagram. Instagram photos posted by depressed users tended to have more blues and grays, and generally tended to be darker than photos posted by nondepressed users. Adding this large-scale information to our array of more subtle biomarkers may result in better categorization and treatment for psychiatric conditions.

It may even be the case that this sort of activity monitoring (if done on large groups of people) may find surprising new patterns that correlate with mental disorders. Maybe you think you have anxiety because you’re awkward, but actually it’s because you are constantly checking how your friends are doing on Facebook.

Social media obsession is a bit easier to diagnose than a slight uptick in a particular bit of genetic code, but it’s not necessarily information you know to share with your doctor. There may be all sorts of behaviors individuals and groups engage in that have undetected clinical significance. As wearable computing gets more popular, you may be able to turn in a more full (and more honest) picture to your doctor. Imagine a glorious future where you go to the dentist and have a friendly conversation until a wearable computer betrays the truth about your flossing frequency.

Okay, so there are some privacy issues here, but having an accurate picture of you—from what you’re reading and how much you exercise right down to a handful of molecules in your urine—may one day be the way we find what health issues you have and predict what health issues you may acquire.

Concerns

A major problem is and will continue to be cost. Ivacaftor, that cystic fibrosis treatment we mentioned before, will only be used by a few thousand people in the United States. There is probably no economy of scale to be had. So that treatment costs about $300,000 a year. Unless we arrive at extremely generalized cheap methods for drug preparation, it will be hard to drive down the cost of a product with so few consumers.

Another worry is how all this data will affect human psychology. You know your weird hypochondriac uncle? Suppose he now has an encyclopedia’s worth of data about his own body and has proof that he is suffering on forty-seven metrics. Yeah. Imagine that Awkward Thanksgiving Chat is now data driven.

And then there’s privacy. To learn more about this issue, we talked to Dr. Kirstin Matthews and Dr. Daniel Wagner at Rice University. We like them. They try to teach ethics to twenty-year-olds, which shows they have a sense of humor.

According to Dr. Matthews, “In order for precision medicine to work and be really useful, you have to connect your genetics with your records, which means all the things that happened to you through your entire life, whether it be injury, environment, or things that happened because of the genetic background. . . . Once that happens there is no anonymity anymore.”

So now potential employers or companies selling health or medical insurance can find out who you are and find out whether or not you’re likely to suffer from a mental illness or have a debilitating disease. Even if these groups don’t have access to health information, they might be able to connect some dots about your mental health simply by analyzing your social media presence. The same group that was able to see depression in your Instagram photo selections was also able to detect depression or PTSD by analyzing Twitter posts. This analysis predicted clinical outcomes months before the official diagnosis was in. Your public behavior may reveal your private health information.

Concepts like life insurance and health insurance only work because it’s hard to know in advance who will get sick or die at what time. You don’t often think about it, perhaps, but insurance is really a mathematical tool. If a thousand spouses have life insurance, each year some will unexpectedly die. Those people who lose their spouses early receive more money than they put into insurance. Those whose spouses live to old age pay more in than they get out, but they get to keep their spouses longer, which (one hopes) makes up for the loss in cash. In essence, the lucky support the unlucky.

As medicine becomes more and more personalized, this system becomes less and less tenable. There may come a time when all insured people must be genotyped. Those with favorable genotypes will pay less for insurance, while those with unfavorable genotypes will pay more. The lucky will be luckier and the unlucky will be unluckier.

To reduce this risk, the U.S. Congress passed the Genetic Information Nondiscrimination Act of 2008, which made this kind of discrimination illegal. Your employer can’t fire you because you have a genetic predisposition toward some medical issue. Nor can an insurer deny you coverage for that reason. But we’re all going to have to grapple with what it means to live in a society with extreme medical knowledge. “It’s not about protecting the data,” says Dr. Wagner, “but about protecting people from the implications of the data.” For instance, if you have a genetic predisposition for aggression, are there jobs you should not be allowed to have? And do the people around you have a right to know?

The U.S. government has not made all kinds of genetic discrimination illegal. As an example, take an app called Genetic Access Control, which was made available on GitHub. The app accessed genomic data on 23andMe (a private company through which you get your genome sequenced) and used those data to restrict users’ access to Web sites. The app’s developer suggests relatively innocuous uses for the app, including creating “safe spaces,” like Web sites that can only be accessed by females. But it’s easy to imagine how an app like this could be used for more sinister purposes. Imagine a site that only people with a certain skin color can visit, or a site that only individuals lacking genetic defects could visit. Furthermore, even the more benign uses will create problems, because identity is both genetic and cultural. Some people with a traditionally female body type carry XY chromosomes, and a group that genetically barred nonfemales would have to decide how to handle that.

23andMe quickly blocked this app’s access to their data, but we can probably expect problems like this to pop up again in the future.

And it’s not just your personal genetic information. You get half of your genome from your mom and half from your dad.* So if you make your genome public, you’re sharing half of each of your parent’s genomes. In fact, whenever you share genetic information, you are to some degree compromising the anonymity of all people in your kin group. Imagine you have a twin who is in political office, and you find out that you carry a genetic risk of schizophrenia. Do you have a social obligation to share that information? Do you have a filial obligation to hide it?

The potential benefits of precision medicine are vast, but so are the potential costs. As we move inexorably into the era of precision medicine, we should think deeply about what privacy means in an ever more technological society.

But some people believe that privacy matters are less important than the medical secrets waiting to be found in our biological data.

The Personal Genome Project* was started by Dr. Church and is a repository for open-access genomic data. Participants are explicitly not promised anonymity, and in fact the paperwork on the Web site makes it clear that when personal data are coupled with genomic data, it probably wouldn’t be hard for someone to figure out who you are.

The person who told us about the Personal Genome Project is Dr. Steven Keating, whom you may remember from the robotic construction chapter.

During the course of his research on robotic house-building trucks, Dr. Keating discovered that he had a brain tumor. He had surgery to remove it and had his tumor’s genome analyzed. For various policy reasons he found accessing the medical and research data on his tumor difficult. He has since become an advocate for making medical data open and available to both the medical community and the patients themselves.

Dr. Keating is excited about the Personal Genome Project. “If you want to contribute your genome, you have to pass a test . . . [that] ensures that you understand the liabilities you’re taking. I kind of like that approach, because the whole thing with genetics is that we don’t really know what could happen in ten years if you share your data now. You have to go through this list, and you have to sign off on, ‘I understand that, yes, a criminal could duplicate parts of my DNA and plant it on a crime scene. Yes, I understand that maybe in the future a virus could be made that’s tailored to me; it would only affect me. Yes, I understand that this information could get leaked and somehow get back to my family and show them a condition that they don’t want to see.’ You have to understand. Then, once you sign off on that, you can submit. . . . It’s important that each person understand the liabilities they’re taking because it’s their data and it’s their choice to do so, but the potential benefits in my mind are so much more powerful than those crazy potential issues, and it totally makes sense to me that this will be the future of medicine.”

How It Would Change the World

In the medium term, precision medicine techniques will be expensive. In the long term, they have the potential to drive medical costs down by detecting diseases before they become severe, by selecting the right treatments immediately, by curing genetic conditions instead of palliating them, and by tapping into the tendency of computer-related industries to deliver more product for less money over time.

Businesses like Facebook provide users a lot of services in exchange for their data about themselves. Given the value of medical data, a similar bargain may be struck between users and biotech firms. Indeed, Google’s parent company Alphabet is already making substantial investments in something called the Baseline Study, which seeks to identify disease-related biomarkers. We don’t know their long-term plans, but if the goal is to have a Google Street View of our digestive tracts, we are prepared to pony up whatever data they need.

One of the most exciting effects of precision medicine has to do with how medical trials are conducted. Suppose you had a cancer drug called explodanol. In clinical trials, five sixths of your patients are completely cured, and one sixth of your patients suddenly explode upon taking the drug. This might be a little embarrassing at your next FDA meeting.

But suppose you happen to notice that there is a difference between explodey and nonexplodey patients—the explodey patients all have a particular set of microRNA markers. Now, you can take this drug that was formerly useless because it (seemingly) randomly killed users, and you can successfully cure cancer in a large portion of the population. Also, if you want to explode one sixth of the population, you can now do it in tablet form.

Going further, by getting a genome, microbiome, metabolome, and a bunch of other -omes, you might be able to make very specific patient categories for clinical trials. This is a win on three fronts: (1) It means you can do statistically meaningful drug trials with smaller populations; (2) it means drugs don’t have to work for all or even most patients to be considered safe; and (3) it means old drugs that were shelved due to safety issues could be brought back to the market, to be used only on a particular patient category.

For example, it is a little-known fact that there exists a vaccine for Lyme disease. It was pulled off the market because a small subpopulation complained of arthritislike symptoms. So, fun fact: As of 2016, your dog, Mittens, can get a Lyme vaccine, but your son, Mittens, can’t.

Although unlikely, it’s conceivable that this subpopulation of people who react poorly to the vaccine share a set of characteristics that create the negative side effect. If we could screen out these people, the vast majority of consumers could have access to this important vaccine.

In reality, it’d probably be a bit more complex. That subpopulation may simply have been all the people who are prone to the placebo effect. But given the number of drugs that have failed to get to market over safety concerns, results like this may happen. And with better biostatistical analysis, current drugs could find new uses for difficult diseases. This may seem a little dorky, but it could be enormous. Currently, it costs over $2.5 billion to bring a new drug to market, and most drugs never make it to that point.

One of the books we read on this topic began by discussing “alternative medicine” healing practices. This raised our skeptic hackles, but the author was quickly on to such delightfully unnatural topics as “Applications of CYP2C19 in Pharmacogenetics” and “Fluorescent in Situ Hybridization.” Before he moved on, he made an interesting point—for all their lack of efficacy, one thing those old healers had in their corner was that they claimed to be tailoring treatment to the patient in particular.

The tarot cards had the minor flaw of not actually working, but the approach of saying “What’s wrong with you” instead of “What’s wrong with people like you” must be a deeply appealing one. A person offering a magical cure is willing to say “I know what’s wrong and I can fix it,” whereas in evidence-based medicine, we can’t always know what’s wrong, and even if we do it may be hopeless. Precision medicine may give us a way to bring the dreams of the age of magic and make them a reality in the age of science.