Mangurian--Clustering Time Series for Improving diabetes management in community mental health clinics

Mangurian--Clustering Time Series for Improving diabetes management in community mental health clinics

by Christina Mangurian -
Number of replies: 1

Describe the study design you will employ in order to determine if your intervention has had an effect on the outcome variable of interest.

  • Pre-post for the pilot study, comparing to a control clinic (year before and year after).
  • A cluster randomized trial of the CRANIUM intervention compared to treatment as usual will be conducted in 26 Community Behavioral Health (CBHS) clinics in San Francisco County. 


Define the unit-of-analysis for your main outcome evaluation, the minimum meaningful effect size, and the sample size necessary to detect this effect size.

 

The unit-of-analysis for my main outcome evaluation is the patient-level.  The two primary study outcomes are 1) evidence of glucose testing determined by glucose-specific serum test or glycated hemoglobin (A1c) and 2) evidence of treatment with metformin (among those with diabetes or positive tests).  For my sample size calculation, I will examine the proportion of patients with diabetes screening.

Sample size calculation: Sample size calculation showed that using an intracluster correlation coefficient of 0.02 and an α level of .05, a sample size of 7,800 (13 clinics, with ~300 patients per clinic) would allow for detection of a mean (SD) difference of diabetes screening of 10% (effect size 0.2) in percentage of clients receiving metabolic screening tests with 80% power.   To compensate for participant attrition and other potential threats to effect size, including clinic-to-clinic contamination, the target patient sample size was increased by 25% to 9,750.  Statistical power may be greater than 80% because analysis will be based on longitudinal methods using data at baseline, 3, 6, and 12 months.  

In reply to Christina Mangurian

Re: Mangurian--Clustering Time Series for Improving diabetes management in community mental health clinics

by Ralph Gonzales -

Nice start tackling this challenging section.

The sample size estimate seems pretty large for your effect size. Try these steps:

1. Assume for now just simple pre-post (not accounting for changes in control sites).

2. Calculate the sample size for a simple chi-square (Test of proportions).  Let's say the baseline screening rate is 10%, and you would like to introduce an intervention that would increase this to at least 20%.  And assume equally sample sizes pre and post.  Now using a sample size calculator, what do you get?

3. The next step is to account for the sample size inflation that will be necessary to account for the clustering (i.e., the design effect). What is your design effect?  This will be driven by how many eligible diabetic patients you think will be visiting the psychiatrist, on average, during the study period.  Consider looking at several design effects and sample sizes assuming ICC of 0.02, 0.2 and 0.5...

4. Multiply unclustered sample size by design effect…  is it the same as your previous estimate?