Is N = 1 Feasible and Affordable?
In my last post, I argued that everyone is cognitively unique. Others have argued that everyone is genetically unique. Can we really tailor assistive technologies and medical care to each individual? Is it feasible? Is it affordable?
Of course, there are many examples of how we tailor technology to our personal preferences. We adjust the seats in our cars, as well as the mirrors, radio stations, etc. We tailor the desktops of our digital devices to our preferences.
What about medical and educational interventions? Can medical care and educational offerings be totally tailored to each individual? This suggests several questions. Can we totally “know” an individual? Can we “know” how to tailor interventions to individuals? Can we afford such tailoring?
Can we totally “know” an individual?
We can only know the things we choose to measure, not the things we did not measure, either because we did not realize these unmeasured attributes mattered or knew but did not know how to measure them. There will always be attributes of potential importance that we miss. Thus, we inherently have to cope with significant uncertainties.
Can we “know” how to tailor interventions to individuals?
We can empirically measure how interventions affect individuals with known characteristics using controlled studies. This process is rife with difficulties – see Rigor Mortis by Richard Harris. Beyond sloppy science, there is a fundamental problem that “control” entails excluding people who do not satisfy the study enrollment requirements. For example, if the study is focused on diabetes, it may exclude people who have any comorbidities.
For N = 1, this gets more complicated. At an extreme, we are trying to tailor a medical intervention to one person. How will we know if it works before intervening? I can imagine the clinician telling the patient, “We’ve never done this before, but we think it is exactly right for you.” The patient might ask, “What evidence do you have that this will work?”
How can we predict what will happen for an intervention – or a portfolio of interventions — that has never been tried before? We could use mathematical models to make such predictions, but the validity of these predictions would be unknown. Model-based evidence might not satisfy skeptical patients.
We need to relax the definition of N = 1. For example, we want to tailor a mix of surgery, radiation, and chemotherapy to a cancer patient with particular symptoms, comorbidities, age, gender, and genetic makeup. Each of the alternative approaches to surgery, radiation, and chemotherapy will have been extensively evaluated and their efficacy assessed relative to symptoms, comorbidities, age, gender, and genetic makeup.
There may be many combinations of approaches, perhaps even thousands, but certainly not millions. Thus, while the combination employed for this particular patient will have been tailored to him or her, it will not be a unique combination. This version of N = 1 seems readily feasible.
How might this work in education? Can we tailor the elements of a course to each individual in the classroom? Can a teacher teach many different ways simultaneously? This would be obviously difficult for a human, but likely feasible for computers. Thus, online education can tailor the presentation and discussion to the particular person and their perceptions as any point in time. For example, the next concept, principle, or fact discussed can be triggered by the specific question the student just asked. In this way, everyone would experience a lecture tailored to him or her. This version of N = 1 also seems readily feasible.
Can we afford tailoring to N = 1?
The last two examples seem affordable while the first did not. Their success will depend on continuous learning about what worked well and what did not. This will likely require capturing data far beyond the individual interactions of clinicians and patients, or teachers and students. How did the intervention affect the patient’s comorbidities? How did the intervention affect the student’s performance in other classes?
The information infrastructure needed to do this well could be expensive. However, as interoperability increases, information sharing and coordination should become easier. Then, we will be better able to know if interventions – medical or educational – are working for the whole person, a very particular N = 1.
