Consider this: You’re a patient with the symptoms of early stage coronary heart disease (CHD). You visit your doctor for a checkup and, staring at the list of medications you’ve been prescribed, you ask your doctor about the benefits of taking a particular statin to prevent a heart attack.
“Well,” says the doc, “if roughly 120 people with CHD similar to yours take this particular statin for seven years, it’ll prevent one heart attack.”
“What!” you exclaim. “But I just asked you if it would help me!”
“We just can’t be sure about you,” says the doc. “All we can be sure about is that it will help someone with CHD. It’s the best we’ve got.”
This might sound hyperbolized, but a measure of drug effectiveness called number needed to treat (NNT) suggests it’s not. NNT measures how many people with a certain diagnosis need to take the drug in order for one person to derive benefit. Perhaps surprisingly, this number is astoundingly high for many common drugs used to treat chronic disease, often above 50 or even 100.
Why isn’t NNT a number you hear from your doctor? Pharmaceutical developers take issue with high NNTs because they don’t accurately capture the value delivered to the people helped by the drug. If four people out a 100 were going to have a myocardial infarction, the drug (hypothetically) lowers that to two. Most doctors will tell you how the drug works comparatively – a 50% reduction in risk. But we don’t know in advance if you’re one of those four. And it doesn’t capture the reality that the drug will most likely not do anything at all for your CHD – and consequently, the unknown side effects might harm you for nothing.
With hypertension, diabetes, obesity and many others on the rise, chronic disease is a critical problem facing the US healthcare system. Dr. Gordon Williams, a distinguished endocrinologist and professor at Harvard Medical School, believes the uncertainly in treating chronic disease lies in the approaches we’ve historically taken to solving them. Instead of focusing on populations, Dr. Williams suggests there is greater value in solving the mechanism of disease using integrated, pre-clinical studies of a single person. He calls this person-oriented medicine. The assumption here is that syndromes are actually less variable and converge on a fewer number of discrete mechanisms than may be widely believed. Therefore, it is an economic argument – making the upfront investment to answer cause and effect in one person first, will lead to answers for the many.
Scientists will by and large claim that their goal is to identify causes. But current attempts to solve chronic disease are almost entirely population-oriented. Researchers average the clinical responses of groups who express similar symptoms, thus reinforcing the “spray and pray” practice of medicine where outcomes are at best probabilistic. With this type of approach, we cannot answer the question of whether something will help a given patient; instead, we throw nearly smart guesses at the wall until something sticks. Even if we do eventually converge on causes, developing effective diagnostics to parse out subpopulations is arduous work based on symptoms and – once again – resorts to population level treatment.
Person-oriented approach is methodologically different from (and we posit, may be disruptive to) population medicine in a number of key ways:
– Precision. The goal is for a disease to be precisely, mechanistically defined and for the cause to be understood in one person. This allows for the development of individualized strategies to diagnose, prevent and treat the disease. Precisely understanding something requires lots of rapid iterations on small scale, so precision is dependent on a smaller sample size. Put simply: we actually solve one problem because we take the right approach.
– Generalizable. Identifying the root cause of a disease for an individual allows this understanding to be linked to a unique biomarker, and is therefore more generalizable across individuals with the same causal pathology. Population data includes too many patients for whom the treatment should not be generalized. Most common chronic diseases are actually gene-environment interactions that, under new pressures, have deleterious effects over time. They are highly conserved within given populations, often associated with origin or ethnicity.
– Compliance. Person-based strategies to diagnose and treat the disease are more specific and reduce secondary effects of treatment. This is much more effective at symptom alleviation than trying to control symptoms because it targets the core pathway – instead of amplifying, hard-to-pin-down contributors or feedback loops. Motivation to comply is also higher when patients are shown causal, instead of probabilistic data (including about side effects), because patients can then make a rational decision about how and where the therapy is worth integrating into their lifestyle.
– Affordable. This approach is more cost effective, both on the therapeutic development and healthcare delivery fronts. It requires a lower user base in preclinical and early clinical phase trials because, instead of relying on sufficiently large Ns to generate statically significant results, the targeted therapeutics targets a cause/effect relationship linked absolutely to a biomarker and can in theory, have near 100% efficacy. On the delivery end, it reduces the use of expensive prescriptions with deleterious or useless effect.
– The right data. Data is becoming ubiquitous and cheap. But whether this deluge of data is useful in solving chronic disease at a causal level is questionable. “Big data” relies largely on static information about populations in uncontrolled environments, introducing high degrees of baseline variability. The causal mechanism often cannot be deduced from such approaches because it isn’t present in the data. Population metabolomic or epigenetic data is static, and while it can point researchers towards useful correlations, it is not feasible to control for environmental factors or “stress” systems in the way that is necessary to differentiate the disease mechanism from averages, or other time-dependent changes in gene expression because of the large population size. This is possible with person-oriented medicine, which Dr. Williams describes as drawing heavily on the methods of endocrinology, where an established a baseline for hormones, for instance, is not useful as an average. These techniques are intensive and therefore only cost effective in the individual. This acute focus and environmental control can pay off where “big data” cannot.
Certainly, person-oriented medicine will not settle all controversies or answer all cause and effect questions about human disease. But if it is effective and disruptive, it will certainly lead lead to lower costs, more rapid treatment, and better access for patients. Every patient with a chronic illness asks: “How can you help me?” With Dr. Williams’ brand of person-oriented medicine, biomedical science may have found an economical way to answer.