Gut reaction

I came across a 2015 paper written by David Zeevi and his colleagues this week.

The paper is titled Personalized Nutrition by Prediction of Glycemic Responses. Their graphical abstract is:

I had a profound, positive gut reaction to a paper about the gut. I think there is a lot to admire in this paper and I believe it has an important role to play in how we might present our research in sport as we explore how we might use analytics to transform performance.

Their brief abstract is:

People eating identical meals present high variability in post-meal blood glucose response. Personalized diets created with the help of an accurate predictor of blood glucose response that integrates parameters such as dietary habits, physical activity, and gut microbiota may successfully lower post-meal blood glucose and its long-term metabolic consequences.

I was particularly interested in how the research team analysed the data from their participant cohort:

We asked whether clinical and microbiome factors could be integrated into an algorithm that predicts individualized postprandial (post-meal) glycemic responses (PPGRs). To this end, we employed a two-phase approach. In the first, discovery phase, the algorithm was developed on the main cohort of 800 participants, and performance was evaluated using a standard leave-one-out cross validation scheme, whereby PPGRs of each participant were predicted using a model trained on the data of all other participants. In the second, validation phase, an independent cohort of 100 participants was recruited and profiled, and their PPGRs were predicted using the model trained only on the main cohort.

In their discussion, David and his colleagues note:

In this work we measured 46,898 PPGRs to meals in a population-based cohort of 800 participants. We demonstrate that PPGRs are highly variable across individuals even when they consume the same standardized meals. We further show that an algorithm that integrates clinical and microbiome features can accurately predict personalized PPGRs to complex, real-life meals even in a second independently collected validation cohort of 100 participants. Finally, personalized dietary interventions based on this algorithm induced lower PPGRs and were accompanied by consistent gut microbiota alterations.

The conclusion to the paper includes the following observation:

Dietary interventions based on our predictor showed significant improvements in multiple aspects of glucose metabolism, including lower PPGRs and lower fluctuations in blood glucose levels within a short 1-week intervention period. It will be interesting to evaluate the utility of such personalized intervention over prolonged periods of several months and even years. (My emphasis)

I do think this is an exemplary paper. I found the language and transparency of their narrative very engaging. There were some powerful visualisations to add to my engagement. Their use of machine learning has sent me off thinking about the generalisability of their approach.

Above all, I was left thinking about “the utility of such personalized intervention over prolonged periods of several months and even years” in a wider sense of personal learning and performance journeys. David and his colleagues have shone a microbiota light on how I might do this.

Michael, Picking and Predicting

Somewhere on my book shelves I have a worn copy of Michael Oakeshott’s edited edition of Thomas Hobbes’ Leviathan (1946).

It may had been even more worn had I realised that he was the co-author (with Guy Griffith) of A Guide to the Classics: or How to Pick the Derby Winner (1936). Michael and Guy were Cambridge fellows at the time. The publisher of a new edition of the book noted:

The book takes the abstraction out of the Derby by attacking the systems which had been developed by generations of ‘form’ experts. It exposes theoretical solutions as fraudulent – instead it applies hard-headed empirical and historical analysis.

Michael Oakeshott was an influential political philosopher in the twentieth century and I found it fascinating that he applied his approach to picking a Derby winner subsequently “to his analysis of rationalism in politics”.

I like Paul Franco’s (1990:1) view of Michael. He was, Paul suggests:

a traditionalist with few traditional beliefs, an “idealist” who is more sceptical than many positivists, a lover of liberty who repudiates liberalism, an individiual who prefers Hehel to locke, a philosopher who disapproves of philosophisme, a romantic and a marvelous stylist.

I think this makes him perfectly suited to a role as analyst, particularly with his interest in second order questions.

Ed Smith (2017) says of the Derby book (despite some of its assumptions dating):

Although the specifics have dated, the ­intellectual disposition is more relevant than ever, especially as sport is experiencing a revolution driven by data analytics. All decision-making in sport (not just gambling, but also recruitment and selection by coaches) hinges on probability. Oakeshott’s second chapter – to what extent does past form determine future performance? – now preoccupies sport’s cleverest thinkers and mathematicians.

Michael’s approach was to use historical judgement. Ed notes:

The more we know about data in sport, the more the Oakeshott position – confidence in good judgement rather than scientific “proof” – gains strength.

He concludes:

Oakeshott’s ideas on racing provide a case for the value and usefulness of the humanities – inexact but wise, sceptical but informed by deep knowledge.

I think this is excellent advice and comes in a week when Alan McCall, Maurizio Fanchini and Aaron Coutts (2017) urge caution about prediction in sport.

Their invited commentary in the International Journal of Sports Physiology and Performance:


  • Highlights the common misinterpretation of studies investigating association to those actually analysing prediction
  • Provides practitioners with simple recommendations to quickly distinguish between methods pertaining to association and those of prediction.

I do believe that the quest for prediction can be undertaken with humility and the humanities.

Ed’s alert to Michael Oakeshott’s work is very timely. It speaks to the possibilities of disciplined historical insights in conjunction with the remarkable innovations in data capture and analysis. It should encourage us to think about we construct analyses of performance that entangle past present and future.

Photo Credit

Epsom Derby 2010 (Monkeywise, CC BY 2.0)

Going for Home (Monkeywise, CC BY 2.0)

Winning and Losing in the Regular Super Netball Season 2017

The regular season in the inaugural Suncorp Super Netball Competition concluded last weekend. The Vixens, Lightning, Giants and Magpies have progressed to the playoffs.

I have been following Champion Data’s coverage of the games played. There were two drawn games in the regular season (Firebirds v Lightning, week 1; Vixens v Swifts, week 3).

My median profiles for winning and losing teams in the remaining 54 games over 14 rounds were:

I used BoxPlotR to visualise some of the data too.

A comparison of Winners and Losers

The data for this visualisation:

These data are available in this GitHub Repository.

Photo Credit

Twitter (RSN927am)


About BoxPloR

“This application was developed with Nature Methods as described in this editorial and this blog entry. Nature methods also dedicated a Points of View and a Points of Significance column to box plots.

This application allows users to generate customized box plots in a number of variants based on their data. A data matrix can be uploaded as a file or pasted into the application. Basic box plots are generated based on the data and can be modified to include additional information. Additional features become available when checking that option. Information about sample sizes can be represented by the width of each box where the widths are proportional to the square roots of the number of observations n. Notches can be added to the boxes. These are defined as +/-1.58*IQR/sqrt(n) which gives roughly 95% confidence that two medians are different. It is also possible to define the whiskers based on the ideas of Spear and Tukey. Additional options of data visualization (violin and bean plots) reveal more information about the underlying data distribution. Plots can be labeled, customized (colors, dimensions, orientation) and exported as eps, pdf and svg files.