signal is missed in many models in Table 8b. A second design was analyzed where there was only one follow-up time (m = 1). This is typical of designs that incorporate time but want independent measurements. In comparing Tables 8e to 8h with their counterparts Tables 8a to 8d, respectively, the signals detected are much weaker and, of course, there is no information about a even though estimates were calculated from the regression coefficients for the baseline term.

Clinical Trials

Clinical trials generally have much larger sample sizes, especially phase III trials. In this section the same designs and models are used, but the sample size is increased. In Tables 9a and 9b the response is 50% in the experimental treatment group and n = 200. In Tables 9c and 9d , the response is only 10% in the experimental group and n = 500. The latter case is more typical of clinical safety data. With larger sample sizes, the properties of the models would be expected to improve. The p-values may get smaller and the MSEs, at least the variance component, should be reduced because more samples should produce more information. Here the bias seems to be unaffected for both response categories. This generally means that if the wrong model is chosen, more measurements will not make it better.

Tables 9a and 9b look very similar to those above, but Tables 9c and 9d show some notable features. First because the models do not estimate the proportion p of responders, the models for the experimental treatment group are a weighted average of 10% quadratic response and 90% no response. This should cause the bias to increase, which it generally does. When autocorrelation is present and the response rate is low (Table 9d), the signal is lost completely (i.e., no information is available). It remains to be seen if better statistical procedures can find this signal.

DISCUSSION Overview of Results

Biomarker experiments with repeated measures over time do not ensure that additional information will be obtained, even though, theoretically, it is guaranteed. The experimental design has to be correct, the biomathematical model of time response has to be correct, and the statistical modeling procedure must be an efficient estimator of that model. If any one of these parts is broken, information can be lost or destroyed completely. For efficacy biomarkers, this means wasted money or missed opportunities. For safety biomarkers, this leads to late attrition or a market recall.



Error df





jö-Value Quad





Project Management Made Easy

Project Management Made Easy

What you need to know about… Project Management Made Easy! Project management consists of more than just a large building project and can encompass small projects as well. No matter what the size of your project, you need to have some sort of project management. How you manage your project has everything to do with its outcome.

Get My Free Ebook

Post a comment