The Promise of Personalized Medicine
By Prof. Dan Merenstein MD
While the existence of responders and non-responders are evident in any clinical trial, today’s forward thinkers envision the next logical step: personalized medicine. Here, I’m a skeptic in a sea of enthusiasts. Although a few ISAPP experts have cautioned that the claims for personalized approaches outpace their scientific substantiation (see here and here), approaches that promise a personalized approach abound. Anyone from their home can send in their stools to companies to get personalized advice based on their microbiome analysis. And in the scientific literature, some studies are showing promising data when it comes to predicting responses based on diet or the baseline gut microbiota (see here and here).
The appeal of these approaches is great. Who doesn’t want a personalized approach from a physician or nutritionist? Even a casual observer knows that a weight-loss plan that works great for one person often fails spectacularly for another. The range of different statins and blood pressure meds exist because not all work the same in everyone. Why would we expect our body to react to foods and medicines exactly the same as someone else’s?
Personalized medicine is not just exciting, it makes intuitive sense and the buzz around it is leading us to expect it is just around the corner or even already here. But if we expect our nutritionist to base her recommendations on our microbiome or blood test, or if we expect our doctor to run genetic or other tests before putting us on medications, we will most often be gravely disappointed.
Knowing the likelihood that a medication will work
The key concept in personalization is figuring out the likelihood that a medication will work for a specific individual. Let’s take a step back and talk about how this concept is typically captured in medicine – with the number needed to treat (NNT). This is a clinical statistical concept that is derived from absolute, not relative, risk difference. Often recommendations are communicated as relative risk – the effects sound so much more dramatic! “Get this shot, it reduces your risk of getting this disease by 50%.” This is so much more convincing than communicating absolute risk difference. “Get this shot, it reduces your risk of getting a disease from 2 in 1000 to 1 in 1000.” You realize what sounded like such an amazing effect in reality reflects a close-to-zero absolute reduction in risk. The number needed to treat is based on the absolute difference in the probability of an outcome and reflects the number of people needed to be treated to get a benefit in one person. For the above example, 1000 people need to get the shot for one person to benefit, so the NNT=1000. The lower the NNT, the more certainty that the intervention will benefit a specific person.
The NNT conveys an often-misunderstood concept in medicine – that medical interventions not only don’t always work, but might only work for a very small percentage of people. If one steps back and thinks about that, it also makes intuitive sense. Most people were not dying from their urinary tract or sinusitis infection before antibiotics, the body does a pretty good job of improving on its own.
The little secret in medicine is that in most cases we are nowhere close to understanding how to individualize treatment in an evidence-based manner. If physicians have a treatment with a NNT of 10 or less, we are pretty ecstatic. Even in such a case, we’re a long way off from knowing exactly which person will respond out of a group of ten people, let alone the personalized treatments that would suit each one of those people.
One example of a very low NNT in medicine today is GLP-1 medications. I am not sure in my near 30 years in family medicine that we have witnessed one single medication change lives and our practices as quickly as GLP-1 medications. While undoubtedly there have been lifesaving oncology and cardiovascular breakthroughs, in primary care I believe the GLP-1s have had the most significant impact. We all have patients and friends that have lost weight they have struggled with their whole lives. So how do we decide if even drugs as effective as GLP-1s will work before prescribing? Well, we don’t. We give them out and cross our fingers. They seem to work in nearly everyone, but the data says you need to treat 2 people for it to work in 1 (NNT=2). That is quite amazing, but GLP-1s are not a personalized medicine nor do we know how to predict who will respond.
Can we personalize biotics and nutrition based on microbiome data?
Getting back to microbiome research and companies that promise personalized biotics…Most of biome research is done by bench researchers not pharma companies. Pharma very much understands NNT and has based their entire business model on it. They are more than happy to sell a drug to treat sinusitis that 15 patients (NNT=15) need to take for one to show benefit. The bench researcher though thinks (and was trained) differently. The bench researcher keeps working in his lab until he figures out what improves the relative probability of an outcome and that is what gets published. Not exactly the same, but this can be thought of in the lab as similar to figuring out who is a responder vs a non responder. I think this training and focus has seeped into the entire microbiome field and thus the obsession with, and even expectation of, personalized biotic interventions. Let me be very clear – there is nothing wrong with this as a goal. We all deserve personalized treatment. But the sad news is that we are very far away from that. In 2025 there are a few things in medicine that are very personalized – some blood thinners, some cancer treatments and some psychiatric meds, but in many more cases we are all essentially treated the same and just hoping we are the one it works for.
We will likely have greater impact and higher uptake of biome interventions once researchers and companies realize the goal does not need to be personalized or responders vs non-responders but a societal impact of a reasonable NNT. Yes one day you may be able to individualize your biome intervention but doing large enough studies to see population level changes is really where the research is currently.
So yes, let’s keep striving to understand how our microbiome is connected to health and disease, how food and medicine can change our microbiome for the better, and how our own individual microbiome patterns may predict our responses to interventions and diets. Let’s also retain our expectations to have evidence-based, tailored medical and nutritional recommendations in the future. But don’t get tricked into thinking the science is really there for most of us to be treated with biotics or other interventions in a personalized fashion. While some companies can tell you what bacteria you may have at lower levels than the average person, they cannot tell you what happens if you try to supplement that bacteria. Maybe we will never know the full answer, but it seems intuitive that before we can personalize microbiome treatments, we should at least be able to tell someone what a healthy microbiome is. Such an understanding might be the basis of true personalized interventions, should they be possible in the future.
