2

I need help in finding which statistical approach I need to use.

I have two groups: G1 (disease) and control.

For both groups we measured their blood level every 10 minutes: 0 to 650 min.

We are trying to find the differences between G1 (disease) and control and the significance of the difference.

The variables in factors are Age, Diagnosis (disease or control), and time. The dependent variable is the blood level.

I do not know which approach I need to use to find this significance (difference).

I tried using Multiple Regression Analysis:

P = constant + B1(age) + B2(Diagnosis) + B3(Time)

Diagnosis 0:constant 1:Patient

If I also want to include the Interaction (Diagnosis x Time) how would I calculate that to put into Excel for analysis?

Nick Cox
  • 56,404
  • 8
  • 127
  • 185
user26705
  • 39
  • Finding “the” significance is not always a fruitful approach. There might be a number of differences (difference in the overall/average blood level, more changes/variance for one group, different shape/profiles, etc.) Why did you take so many measures in the first place? You might want to think about how you expect blood level to change over time and what specific difference you are interested in. Otherwise, you might just as well compute the mean level for each participant/observation unit and do a test on that. – Gala Jun 12 '13 at 09:07

2 Answers2

2

I second @user21240's answer and @zbicyclist's comment but would like to offer a few additions:

First, the model @user21240 proposes models a linear course of the blood level over 11 hours time. We don't know what substance you are measuring, but if there could be diurnal cycles (e.g., for cortisol), you should account for these. My recommendation: transform your time variable using restricted cubic splines or natural splines.

Second, you are measuring each participant multiple times. It stands to reason that every participant's blood level will correlate strongly over time. You should really take that correlation into account in the model and use mixed models (also known as "repeated measures models").

Third, an elaboration on my second point above: the first and the second measurement of each participant will probably correlate more strongly than the first and the thirtieth measurement. In your mixed models, you can account for this effect by choosing an appropriate correlation structure for your random effects. In the statistical package R, you can use the corCAR1 correlation structure in the nlme package to model a time-dependent autoregressive (AR) of order 1 correlation structure. Unfortunately, I don't know what SPSS offers in this line, so you may want to consult the manual or help pages.

Fourth, as @zbicyclist notes, you may want to look at interactions between Age and Diagnosis... but you may also want to investigate interactions between Diagnosis and Time (and spline-transformed Time, see above), since the Diagnosis may have an impact on the time course.

Fifth, by now you have a model selection problem on your hands (spline transform Time or not, how many spline knots, log transform Age or not, which interaction to use, which correlation structure to use...). For the fixed effects, you can use a variant of AIC; this reference may help. Unfortunately, I am not aware of a formalized way to choose the correlation structure - here I would simply use what makes sense and what your software offers.

Good luck!

Community
  • 1
Stephan Kolassa
  • 123,354
  • Hey Stephan. yes we are measuring cortisol levels in the blood. that is why we took multiple measurement in time~. How would i set up an equation to look at the interaction between age and diagnosis too (Age x Diagnosis)? And also would Multiple Regression Analysis work for this case too? since... I am not familiar with (how) to model a time-dependent autoregressive(AR). – user26705 Jun 13 '13 at 17:03
  • 1
    OK, so for cortisol, you definitely want some way to account for diurnal profiles, e.g., with splines. Unfortunately, I have no idea at all about SPSS. Would some example code in R help you? (I'd rather not spend a lot of time on this unless it would really be helpful, and if you are stuck with SPSS, you would probably better look for a SPSS guru.) – Stephan Kolassa Jun 13 '13 at 19:55
1

You could treat the diagnosis as a dummy variable, fit a linear model: $Blood Level= \alpha + \beta_1 Age +\beta_2 Diagnosis + \beta_3 Time + \epsilon$ and look at the significance of the dummy variable, $Diagnosis$.

fredrikhs
  • 1,254
  • This model is a good place to start. Might want to see whether Age works as-is (the effect on BloodLevel is linear) or whether some transformation is needed. Similarly, after inspection you might decide you need an interaction term (Age x Diagnosis) – zbicyclist Jun 12 '13 at 20:54
  • Wouldn't the lack of independence between observations (measures from the same person at different times) be an issue if you want to run tests? Wouldn't the interaction between time and diagnosis be the most interesting thing in this design? – Gala Jun 13 '13 at 11:07
  • How would i set it up to do the interaction term (Age x Diagnosis)? – user26705 Jun 13 '13 at 16:56