Diabetes and Data. What more can we learn? Part 2.

Diabetes and Data. What more can we learn? Part 2.
Diabetes and Data. What more can we learn? Part 2.

In the first part of this evaluation, I looked at Steps, Sleep and Total Daily Dose (TDD) to see if I could find any relationships between these things. The data suggested that step patterns had a greater impact on daily dose than other metrics that we might use.

In this part, I introduce Time in Range (TiR) to see whether any of the factors we looked at last time might influence that.

This time, we have four sets of data.

  • Sleep hours
  • Steps
  • Step patterns
  • Total Daily Dose
  • Time in range

Box and Whisker plots for sleep, steps, total daily dose and time in range of these are shown below:

Box and whisker plot showing sleep hours by day of the week
Daily sleep hours box and whisker chart
Daily Step Count Box and Whisker Chart
Total daily dose box and whisker plot by day of the week
Daily Total Daily Dose Box and Whisker chart
Daily time in range Box and Whisker Chart

If you were to look across these plots, you might think that that there’s a relationship between a number of these factors. And yet, when you start to dig into the data, once again, the story appears a little different.

Factor effects on Time In Range

It looks, initially at least, as though step count or sleep may have an effect on TIR and that there appears to be an inverse effect between TDD and TIR. When we undertake statistical analysis on these pairs of data, it looks much less interesting.

Unlike the relationships with total daily dose from the previous part of this article, the correlation values for sleep hours to TIR was moderate (0.5) and held significance (p-value 0.001), step count’s correlation with TBR and TIR was consistently low (around 0.2) with very little significance (p-values all greater than 0.05). Hours of sleep also held very little correlation or significance with average glucose levels, and a loosely moderate correlation with standard deviation of glucose levels.

In terms of steps and sleep, and their relation to Time in Range, only the step pattern really seemed to show anything, when looking at it as a box and whisker chart.

Step pattern/Time in Range Box and Whisker Chart

This chart suggests that the distributed step pattern should result in a greater time in range. However, it’s worth noting that the depth of the 25th-75th percentile box is greater in the distributed step count. This is reflected in the correlation data, which shows a moderate correlation (0.5) but with a reasonable level of significance (p-value 0.001) which once again seems to concur with the data that Diabetes UK established with activity snacking.

Is there a relationship between TDD and TIR?

Plotting a scatter chart of TDD and TIR, it does suggest that there is some form of relationship.

In fact, there appears to be quite a strong relationship, and the correlation between TDD and TIR turned out to be -0.64, which is edging into the high-moderate zone. The p-value was 0.000001, which is also very low, suggesting that there’s quite a lot of significance in the relationship between TDD and TIR.

If we then cut the data slightly differently, and look at TDD relative to average TDD, we see that as I fell below the average TDD, in general my TIR increased.

It’s fair to say that there is still a fair amount of variance in this data, and this is reflected in the R-squared value, which was 0.41.

What does this data tell us?

As we learned in part one of this article, the pattern of steps over the course of the day seems to have the greatest effect on Total Daily Dose. Similarly, it also seems to have a reasonable effect on Time in Range. Perhaps more surprising is that while sleep does show correlation with time in range, neither step count nor sleep seem to be strongly correlated with time below range. Similarly, Time Below Range has little correlation with the step count pattern.

If I go back and review my time in range and TDD data, it becomes fairly obvious why the relationship shown in the previous section appears to exist. Generally, the days where my TDD was highest relate to days with either “sticky” meals or where a cannula has given up, and it’s been challenging to manage diabetes. So while it might be a “true” relationship, it isn’t necessarily a useful one.

This also casts doubt on the relationships highlighted between step pattern, TDD and TIR. Were they just a fluke? Was 6 weeks of data enough to randomise the occurrences linked to cannulas being less effective on day 3? It’s a difficult question to answer, and highlights the need to capture a lot more data and do some proper research into biometrics and diabetes data.

Regardless of the above, I am convinced that there are relationships between various biometrics and type one management, and it’s an area that I think needs significantly more investigation and research. The type of data I’ve reviewed here would very quickly enable a broader population view of how frequently “feet to the floor” effects happen in the morning, and in theory would allow us to potentially identify population characteristics that are associated with it.

If we’re looking for a way to involve more people with diabetes in research projects then this seems like a very simple way forward!


  1. Food is surely the missing factor here, as you said. Are you eating more through the working week to Friday, with consequently higher TDD and lower TIR? Are you eating the same amount at weekends?
    I find the step patterns interesting. Perhaps you could chart effects on TDD and TIR of longer weekend dog walking outings?

    • So there’s the interesting point in this. The weekday Vs weekend data provides a red herring as we also have a couple of holidays thrown into the mix. In general, I eat just as much food when not working as when working and still see the difference linked to step patterns, which suggests that is the key factor.

      What’s perhaps also interesting is the different effects of constant movement compared to bidaily in the effects on blood glucose and the resultant need for insulin/time in range.

      Looking back through my data, the more sedentary days result in higher post prandial outcomes with slower drops than those where movement is constant, which also helps to explain the different insulin needed and time in range.

Leave a Reply

Your email address will not be published.