N-of-1 Trials and Analyzing Your Own Fitness Data | by Merete Lutz | Apr, 2024

N-of-1 Trials in Observe

To level out you an occasion of this system in observe, I’ll conduct my very personal analysis on quite a lot of data collected from my Whoop strap from April twenty seventh, 2018 to October fifth, 2019. Our evaluation question for this N-of-1 look at is:

Does consuming alcohol lead to poor sleep?

As an athlete and epidemiologist, I’m very aware of how detrimental alcohol will probably be in your sleep, athletic effectivity and primary wellbeing. I’ve repeatedly been suggested how athletes shouldn’t drink, however its one issue to be suggested, nonetheless one different to see the proof to your self. As quickly as I started sporting my Whoop I seen how my sleep score (a metric calculated by the Whoop app) would endure after consuming alcohol. Usually even a day later, I believed I could nonetheless see the affect. These observations made me have to do my very personal analysis, which I can lastly full now.

Notes on the Information

The two variables of curiosity in our analysis is sleep effectivity score and alcohol consumption. Sleep effectivity score ranges from 0 to 100 and is a metric calculated by the Whoop app from biometric data like respiratory value, light sleep interval, gradual wave sleep interval, and REM sleep interval.

The alcohol consumption variable is the response to the question “Did you’ve received any alcoholic drinks yesterday?” that’s responded to by Whoop clients every single day upon waking up. I always answered these questions truthfully and continually, although we’re restricted in our data in that the app doesn’t ask questions on how rather a lot alcohol was consumed. Which suggests all ranges of alcohol consumption are dealt with equally, which eliminates the prospect to research the connection on a deeper diploma. There was some missing data in our alcohol attribute, nonetheless this missing knowledge was imputed with ‘No’s as I do know from personal experience that if I had drunk the night sooner than I was sure to mark it inside the app.

Exploratory Information Analysis

The first step in any analysis is to do some exploratory data analysis (EDA). That’s merely to get a primary idea of what our data seems like, and to create a visual which will help direct our investigation.

From the above box-plots, we see that widespread sleep score appears to be bigger when no alcohol was consumed, and to have a narrower distribution. Curiously, there seems to be further outliers in sleep effectivity score when alcohol simply isn’t consumed. Possibly journey days and jet-lag can account for these outliers, as I traveled overseas 5 situations all through this sample interval.

Now that we’ve got now gotten an excellent first take a look on the info of curiosity, its time to dig into the statistical analysis.

Hypothesis Testing

To answer our evaluation question, I shall be conducting hypothesis testing. Hypothesis testing is a statistical strategy that allows us to make inferences a few inhabitants based totally on some sample data. On this case, we attempt to deduce if me consuming alcohol is said to having poor sleep that night. We don’t have data on alcohol consumption and sleep for every night I’ve been alive, so we look at our sample data as a proxy.

The first step in hypothesis testing is to formulate my hypotheses. A ‘null hypothesis’ is the idea that nothing attention-grabbing is occurring or that there isn’t a relationship or affect. In our case the null hypothesis is: There isn’t any distinction in indicate sleep effectivity between nights throughout which alcohol was consumed and was not consumed.

An ‘totally different hypothesis’ is the hypothesis that contradicts the null, and claims that the reality is there’s one thing attention-grabbing happening. In our occasion the totally different hypothesis is: There’s a distinction in indicate sleep effectivity between nights throughout which alcohol was consumed and was not consumed.

Deciding on a Statistical Test

To judge which of these hypotheses is true, we’ve got now to chose a statistical verify. We’re curious if the standard sleep score for nights throughout which I drank alcohol is totally totally different from the standard sleep score for nights throughout which I didn’t drink alcohol, and so shall be using a distinction in means to verify this. Notably, our verify statistic is: Suggest sleep effectivity with no alcohol — Suggest sleep effectivity with alcohol

Now that we’ve got now outlined our framework, we’re ready to make use of R to calculate our verify statistic and take into account our hypotheses.

Conducting our Analysis in R

From our sample data we’re in a position to calculate our seen verify statistic. The code in R is included beneath.

test_stat <- data |> 
specify(system = sleep_performance ~ alcohol) |> 
calculate(
stat = "diff in means",
order = c("No", "Positive")
)

Our verify statistic is 8.01. This amount signifies that the standard sleep score for nights throughout which I consumed no alcohol is 8.01 elements bigger than nights throughout which I did devour alcohol.

The next step inside the analysis is to generate a null distribution from our sample data. A null distribution represents all the utterly totally different values of verify statistic we’d observe if samples have been drawn repeatedly from the inhabitants. The distribution is meant to copy the variation inside the verify statistic purely as a consequence of random sampling. The null distribution is created in R beneath:

set.seed(42) #Setting seed for reproducibility
null_distribution <- data |> 
specify(system = sleep_performance ~ alcohol) |> 
hypothesize(null = "independence") |> 
generate(reps = 1000, form = "permute") |> 
calculate(
stat = "diff in means",
order = c("No", "Positive")
)

What we’re doing above is taking samples with various from our data, and calculating the excellence in means from these samples. We do this 1000 situations to generate an enormous adequate distribution so that we’re in a position to determine if our seen verify statistic is necessary.

After we’ve got now our null distribution and verify statistic, we’re in a position to calculate a two-sided p-value for an alpha of 0.05. The p-value will probably be thought-about the chance of getting a verify statistic that’s as extreme or better than our seen verify statistic if the null hypothesis is true. Put into plain phrases; it represents how most likely will probably be to see this end result if there was no true affiliation. We calculate a two-sided p-value in R beneath, as we’re taken with the chance of the verify statistic being larger or lesser than anticipated.

p_value <- null_distribution|> 
get_p_value(test_stat, route = "every")

Our p-value is 0.017 which signifies that our discovering is necessary on the alpha=0.05 diploma, which is a typically accepted diploma of significance in statistics. It signifies that the excellence in sleep score we found was very important! We’ve received the proof to reject the null hypothesis and accept the selection; there is a distinction in indicate sleep effectivity between nights throughout which alcohol was consumed and was not consumed.

I’ve included a helpful visualization of the null distribution, verify statistic, and 95% quantile range beneath. The grey bars are the quite a few doable verify statistics calculated from our 1000 samples, and the orange line represents the density of these values. The blue dashed traces signify the 97.fifth and a pair of.fifth quantiles of this distribution, previous which our verify statistic (in crimson) is confirmed to be very important.

Final Conclusions

Properly, it appears my coaches have been correct all alongside! Our analysis found that my widespread sleep score after I didn’t devour alcohol was 8.01 elements bigger than my widespread sleep score after I did devour alcohol. This distinction was found to be statistically very important, with a p-value of 0.017, which signifies that we reject the null hypothesis in favor of the selection. This statistical end result backs up my personal experience, giving me a quantitative end result that I can consider in.

Going Further

Now that I’ve this preliminary analysis beneath my belt, I can uncover further associations in my data, and even use further tough methods like forecasting and machine finding out fashions.

This analysis is a very elementary occasion of an N-of-1 look at, and isn’t with out limitations. My look at was observational comparatively than experimental, and we will’t declare causality, as there are quite a few totally different confounding variables not measured by my Whoop. If I wanted to find a causal relationship, I have to rigorously design a look at, report data on all doable confounders, and uncover a way to blind myself to the remedy. N-of-1 analysis are arduous to do exterior of a medical setting, however we’re in a position to nonetheless uncover important associations and relationships by asking straightforward questions of our data.

I hope that after this tutorial you’re taking the initiative to acquire your particular person data from regardless of well being tracker it’s possible you’ll get your palms on, and fiddle with it. I do know everyone can offer you a hypothesis about how some variable impacts their nicely being, nonetheless what most people don’t discover, is that you just simply’re nearer to getting a quantifiable reply to that question than you suppose.

References and Further Learning

[1] Davidson, Okay., Cheung, Okay., Friel, C., & Suls, J. (2022). Introducing Information Sciences to N-of-1 Designs, Statistics, Use-Cases, the Future, and the Moniker ‘N-of-1’ Trial. Harvard Information Science Analysis, (Explicit State of affairs 3). https://doi.org/10.1162/99608f92.116c43fe

[2] Lillie EO, Patay B, Diamant J, Issell B, Topol EJ, Schork NJ. The n-of-1 medical trial: the final phrase method for individualizing remedy? Per Med. 2011 Mar;8(2):161–173. doi: 10.2217/pme.11.7. PMID: 21695041; PMCID: PMC3118090.

[3] Daza EJ. Causal Analysis of Self-tracked Time Assortment Information Using a Counterfactual Framework for N-of-1 Trials. Methods Inf Med. 2018 Feb;57(1):e10-e21. doi: 10.3414/ME16–02–0044. Epub 2018 Apr 5. PMID: 29621835; PMCID: PMC6087468.

[4] Schork, N. Personalised remedy: Time for one-person trials. Nature 520, 609–611 (2015). https://doi.org/10.1038/520609a