210 likes | 483 Views
PMA P030016 Specular Microscopy Sub-study. Gerry Gray, Ph.D. Cardiovascular and Ophthalmic Team Leader, Division of Biostatistics Office of Surveillance and Biometrics Center for Devices and Radiological Health October 3, 2003. Specular microscopy sub-study design.
E N D
PMA P030016 Specular Microscopy Sub-study Gerry Gray, Ph.D. Cardiovascular and Ophthalmic Team Leader, Division of Biostatistics Office of Surveillance and Biometrics Center for Devices and Radiological Health October 3, 2003
Specular microscopy sub-study design • Endothelial cell counts • Specular microscope photographs • Multiple images per eye • Images read at a core center (Emory) • Follow up • Original design 3 months, 1 and 2 years. • Study modified to add 3, 4 year visits. • Purpose: investigate effects on endothelial cells through time • 306 eyes enrolled & have at least one count
Estimates of cell loss over duration of the study • Estimates are fairly stable regardless of method of calculation • Range of estimates at 3 years: 8.5% to 8.9% • Approximately 225 – 235 cells/mm^2 • Includes both initial operational loss and normal loss due to ageing • Range of estimates at 4 years: 8.4% to 9.7% • Approximately 220 – 260 cells/mm^2
“Steady state” long term loss • Estimate depends largely on: • Cohort that is used • Whether we use all the data or only the 3 & 4 year data • Sponsor’s analysis: • Percent change between 3 & 4 years • Using only 3 & 4 year observations from cohort with both 3 & 4 year visits • Recall from the previous Table that this 57-patient cohort has a relatively low 3-year count. • Estimated percent change = 0.07% (i.e. a slight gain) • 95% CI [-1.44%, 1.58%] • Other cohorts (e.g. BL, 2+ visits) have relatively higher 3-year count. • Various analyses using these cohorts produce a loss of about 2% per year.
Test for change in slope • In all cohorts, the loss does appear to “level off” after 3 years. • But there’s no strong statistical evidence that the “leveling off” is real (versus random chance and/or a small 4-year sample). • Piecewise linear model: • Initial (operative) loss from baseline to 3 months • Linear loss from 3 months to 3 years • Linear loss (possibly different slope) from 3 years to 4 years • Test for different 3-4 year slope: p = 0.37 • Implication: “steady state” loss should be estimated using all data after 3 months.
Extrapolation caveats • ALWAYS a questionable exercise to extrapolate beyond the range of available data, especially to the degree we want here. • Highly dependent on the model we use & the assumptions we make. • BOTH of the previous extrapolations assume that loss will continue linearly for 30 years. • Probably much more important to think about: • If it’s necessary to obtain good long-term data. • If so, how to go about it.
Phase IV study possibilities • Continuation of phase III study • Higher quality data • Fewer patients • More costly • “Registry” approach • Simpler & cheaper • More patients • Less information (specular microscopy not generally available) • Choice depends on goals
How do individual patients fare? • Perhaps more important than “average” cell loss through time. • What proportion of patients will have major operative loss? • What proportion of the patients will show cell loss greater than some critical amount?
Regression on individual eyes Piecewise linear fits to 206 individual eyes (BL, 2+ cohort).
Regression on individual eyes • Mean baseline = 2654 cells/mm^2 • Mean initial loss = 204 cells/mm^2/yr (absolute loss 51 cells/mm^2, or 1.9%) • Mean rate after 3 months = 53 cells/mm^2/yr, or 2%/yr • Tolerance interval for long term rate: • 95% confident that • 60% of patients have a loss no worse than 82 cells/mm^2^yr (~3.1%) • 90% of patients have a loss no worse than 163 cells/mm^2/yr (~6.1%) • 99% of patients have a loss no worse than 247 cells/mm^2/yr (~9.3%)
Predictors of endothelial cell loss • There appear to be several statistically significant predictors of endothelial cell loss (baseline measurements): • This includes anterior chamber depth, MRSE, diopter, ECD coefficient of variation • Dependent variable (annual loss) taken from per-eye regression. • Using continuous measurements for covariates. • Sponsor presented analyses using “binned” data (cut at 15.0D or at 7D & 10D) that do not show MRSE effect. • Age, gender, IOP, cylinder, axis, etc. did not appear to be significant predictors of cell loss.
Clinically significant ACD effect? Annual cell loss estimated using per-eye regressions.
Sponsor’s ACD summary Estimates are percent loss relative to index group.