Sample size calculation for multiple regression / GEE

[josient]

Hi,

I'm studying the determinants of recovery of Quality of Life (QoL) in lung cancer patients following surgery.
We will have 4 different measurement occasions in time: 1 month before surgery, 1, 3, and 6 months post-surgery.

Our aim is to answer the question: what are the determinants of recovery of QoL in lung cancer patients treated with lung resection?
We will include physical and psychological variables (measured at all 4 occasions), as well as descriptive variables such as age, gender, and information about the surgery.
We expect that there will be 6 predictors (although literature is not conclusive about this).
Next to that, QoL measures are expected to be non-linear: i.e., after surgery a sharp drop in scores is expected with a rapid recovery in the first months, while recovery in the 3rd - 6th months will be more and more gradual.

So, I'm wondering whether I can answer my question using multiple regression or GEE (i'm not familiar with GEE, but knows it exists in SPSS and it seems promising) or that I will need another statistical program that takes into account the multiple levels of the data (i.e. repeated measures in the same person).

Next to that, I'm struggeling to calculate my sample size needed to answer the research question. There is no research available that uses the same predictors as I am to estimate QoL in this population. So how do I decide what effect size I need to use for sample size calculation???

Hopefully someone can help me with these questions!

With kind regards,

Josien Timmerman

[Bruce Weaver]

A multilevel model (via MIXED) is another option for analyzing occasions clustered within individuals.  Jos Twisk's nice introductory book (Applied Multilevel Analysis) includes a chapter on sample size estimation.  In a nutshell, he says to compute a "standard" sample size estimate (i.e., for a comparable model, ignoring the clustering), and then apply an adjustment that takes into account the clustering.  (There are actually two possible adjustments, but I'll let you read about that.)

HTH.

josient wrote

Hi,

I'm studying the determinants of recovery of Quality of Life (QoL) in lung cancer patients following surgery.
We will have 4 different measurement occasions in time: 1 month before surgery, 1, 3, and 6 months post-surgery.

Our aim is to answer the question: what are the determinants of recovery of QoL in lung cancer patients treated with lung resection?
We will include physical and psychological variables (measured at all 4 occasions), as well as descriptive variables such as age, gender, and information about the surgery.
We expect that there will be 6 predictors (although literature is not conclusive about this).
Next to that, QoL measures are expected to be non-linear: i.e., after surgery a sharp drop in scores is expected with a rapid recovery in the first months, while recovery in the 3rd - 6th months will be more and more gradual.

So, I'm wondering whether I can answer my question using multiple regression or GEE (i'm not familiar with GEE, but knows it exists in SPSS and it seems promising) or that I will need another statistical program that takes into account the multiple levels of the data (i.e. repeated measures in the same person).

Next to that, I'm struggeling to calculate my sample size needed to answer the research question. There is no research available that uses the same predictors as I am to estimate QoL in this population. So how do I decide what effect size I need to use for sample size calculation???

Hopefully someone can help me with these questions!

With kind regards,

Josien Timmerman