From N-of-3 to Stats-of-1: Quick Questions and Answers from our Editors
Last updated: Oct 29, 2021
Tell us briefly about your training background and current work.
Julio. I have a background in software engineering, and I tinkered for a while with Geographical Information Systems and simulations. For the past seven years, I have been working in mobile sensing and digital phenotyping, first in the context of Parkinson’s Disease during my Ph.D. at the University of Manchester in the UK, then in the domain of binge eating during a short postdoc at the University of Bristol.
At my current research position at the University of Pittsburgh, we design and execute observational and interventional studies that aim to support cancer patients and survivors and other patient groups through different collaborations. I am also a supporter of open science and software, leading the development of the Reproducible Analysis Pipeline for Data Streams (RAPIDS), an open-source pipeline to standardize the data processing of mobile sensing projects.
Clair. I am a clinical psychologist. I specialize in providing treatment to adolescents and adults who find that strong emotions are getting in the way of their day to day lives. I teach people skills to help manage these strong emotions. My background is in treatment development research and I spent my post-doc doing research on misophonia, which is a condition in which people are bothered by certain sounds (e.g., chewing).
I currently work as a staff psychologist at TAP clinic in Durham, NC, where I focus on providing evidence-based psychological treatments to patients and training new therapists.
Eric J. I’m a health data scientist, formerly a biostatistician. Having started off as a neurobiology major, I shifted into statistical programming (SAS) in pharma for five years, followed by public health biostatistics during my doctoral program. I trained as a postdoc in health behavior change, and finally found my way to digital health in industry. I am obsessed with developing statistical designs and analyses for n-of-1 or single-individual settings using digital health data—particularly using causal inference and machine learning methods.
I currently work at Evidation Health. There, my team and I (alongside other teams) help clients design, run, and analyze experimental and observational studies involving app-delivered surveys and wearable sensor data. Evidation’s larger mission is “to enable and empower everyone to participate in better health outcomes”—a mission I enthusiastically support.
How does N-of-1, single-case experimental design, etc. fit into your work?
Julio. Several chronic health conditions can manifest in very different ways from person to person. For example, although Parkinson’s Disease has a core set of symptoms like tremor, its severity and presence vary across participants, and other symptoms like fatigue or constipation can have an even more meaningful impact in their lives.
This variability can make it difficult to find commonalities across a group of people, and so it makes sense to try to understand disease treatment or interventions in a personalized way. In these cases, N-of-1 studies are a great alternative to collecting and analyzing data about someone’s health and behavior over time.
Here is where smartphone and wearable devices come into play because they are gadgets that many patients are already familiar with and allow us to collect data from multiple integrated sensors. These are data sources that researchers in the community have shown the potential to quantify people’s routines and activities that work as a proxy to their health.
All this led me into one of my main lines of work; N-of-1 studies to analyze the link between disease symptoms and mobile sensor data.
Clair. My grad school advisor was a huge proponent of single-case experimental design (SCED) and it’s a great methodology for psychologists. Because each person is so different, it allows us to do a rigorous evaluation of the treatments we’re testing in order to understand if they’re working. I used SCED for most of my research in graduate school and on post-doc. Now that I’m a full time clinician, SCED is still really helpful to me because it allows me to make sure I can determine whether an intervention is helping my patients.
Eric J. I’ll echo what Julio said. I got interested in n-of-1 studies because I wanted to help a loved one figure out how to manage their irritable bowel syndrome (IBS) symptoms. Echoing Clair’s thoughts, I learned about their methodological cousins, SCEDs, along the way. I saw that there was a scientific and clinical need to connect these two approaches (along with mobile sensing and digital phenotyping as Julio mentioned earlier, and other fields) in a more fundamental way. This emerging unified field could inspire new interest and accelerate methods development for making digital health really work for each patient individually. So I created Stats-of-1!
What tools (software, statistical, etc.) do you use when conducting N-of-1, SCED, etc. work?
Julio. I work primarily on the computational analysis of data. For data analysis I use R and Python through Visual Studio Code (it is snappy, stable, and comes with a rich ecosystem of extensions). I also use git and Github for keeping track and publishing our code. For productivity, Google Docs for collaborative writing, Zotero for reference management (free and open-source, plus their plugins for Word and Google Docs are amazing) and Notion to keep my notes and to-do’s. For collecting and processing mobile sensor data I go for the AWARE Framework plus RAPIDS.
Clair. I use pretty simple tools! I do most of my analyses in Excel and I use a lot of free online tools for statistical analyses. Rumen Manolov in particular creates lots of helpful free tools that I use.
Eric J. I use many of the same tools as Julio’s: R for data management and analysis, Github for version control, and Google Docs for collaborative writing. An excellent grad student I’ve been working with on an n-of-1 quantified self paper, Igor Matias, also recently introduced me to Overleaf for collaborating on LaTeX documents—and it’s amazing! Clair will be happy to note that Igor and I also work a good amount in Excel, as his students and “clients” find it most accessible.
What is a misconception about small sample studies you wish you could correct?
Julio. That you cannot get any meaningful or clinically relevant insights from the data of a single participant. Traditional cross-sectional approaches were a great option when it was easier to ask people to participate in a research study once rather than multiple times intensively over weeks or months. However, the advent of mobile sensors and electronic momentary assessments (aka digital surveys) has made individual longitudinal approaches easier to execute. It will be a slow shift, but over time more and more people have come to recognize the potential of small but time-intensive studies to get a step closer to personalized medicine.
Clair. That SCEDs are not rigorous designs! Randomized controlled trials (RCTs) are often considered the gold-standard for research design but they have their own weaknesses. SCEDs have a lot of strengths and allow for really rigorous research that is very meaningful. And small sample studies are especially valuable in the early stages of research. Before spending a lot of money on a RCT an SCED can help test and refine a treatment.
Eric J. I fully echo Clair’s and Julio’s sentiments: That SCEDs, n-of-1 studies, and other idiographic (i.e., individual-focused) designs are not rigorous or clinically useful. In fact, clinicians often end up combining and tailoring treatment regimens “off-label” from their prescribed uses, indicating a natural urge to focus on each patient’s unique health needs. A few organizations have even formalized this process, offering n-of-1 trial designs as a clinical service. It just makes sense to understand and help each person first and foremost; use averages as a guide, but tailor your guidance and intervention from there.
To reiterate Julio’s point, group-based approaches were the most feasible for a long time in clinical health research. But wouldn’t it be better to pivot into prioritizing each individual first, now that we have the digital means to do so? And Clair’s point is an important one: In psychology, achieving an “average treatment response” may not be a useful goal in clinical practice because each person responds differently to therapy in important, recurring ways that impact their daily quality of life.
Todd Rose puts it nicely in The End of Average: Traditional study design and analysis first aggregates everyone’s data, then analyzes it for average associations or effects (“aggregate-then-analyze”). Even so-called “precision medicine” and “personalized health” takes this approach—albeit with very small groups. But with today’s ever-present digital health tools, we are finally empowered to truly analyze individuals first, and then subsequently aggregate to find patterns across groups (“analyze-then-aggregate”).
What development would you like to see in the next 5-10 years of research in this field?
Julio. I would like to see a boom in the statistical approaches to analyze these kinds of study designs. The technology to collect personal longitudinal sensor and self-reported data with minimal participant disruption is maturing. The tools and methods to summarise and process such data are consolidating. However, there is still much open space for innovation for methods that can identify and model the temporal dynamics of the data we collect to diagnose, manage, or intervene in a disease.
Clair. I would also like to see more statistical approaches and I would like to see more journals (especially high impact journals) accepting SCED papers so they can get more recognition!
Eric J. My co-editors again perfectly summarize my own thoughts. I’ll add that I’d love to see these researchers and citizen scientists from these different fields come together to align on concepts, language, and key fundamental approaches. This alignment will more quickly innovate statistical methods by opening up new lines of thought. This means better health coaching, diagnoses, and interventions will get to participants faster and sooner—all without sacrificing methodological rigor. If I could name this unified field, I’d probably call it esametry.