Desktop version

Home arrow Economics

  • Increase font
  • Decrease font

<<   CONTENTS   >>

Five Clinical Trials in Schizophrenia

The combined data of five clinical trials in schizophrenia (for details, see Section 2.2.2) are considered here. It will be examined whether the Brief Psychiatric Rating Scale (BPRS) is an appropriate surrogate for the Positive and Negative Syndrome Scale (PANSS) using the MI-based approach detailed in Section 15.3 to obtain “balanced” datasets. Treating physician is used as the clustering unit in the analyses. The largest cluster size in the dataset was 52, of whom 9 patients received an active control and 43 patients received the experimental treatment. Thus, in all cluster-by-treatment groups that had less than 43 patients, data were imputed to achieve balance. Clusters which contained less than 10 patients were discarded from the analyses (to avoid problems during the MI phase). The imputations were conducted for each of the clusters separately, using S =BPRS, T = PANSS, and Z = treatment in the imputation model. A total of 1000 imputations were conducted, and for each of the imputed datasets, Model (15.1) was fitted (using both the FA0(4) and UN covariance parameterizations for D).


When Model (15.1) was fitted to the non-imputed data, convergence issues occurred. Both the models that used the UN and FA0(4) parameterization for the D matrix converged to a non-PD D/H matrix.

When the MI-based approaches were used, there was 100% proper convergence in both the MI UN and MI FA0(4) scenarios. The mean Rriai of the properly converged results and their CI95% for the MI UN and FA0(4) models both equaled 0.920 [0.875; 0.955]. The mean Rfndiv and their CI95% for the MI UN and FA0(4) models were identical and equaled 0.923 [0.913; 0.933].

To establish a frame of reference against which these estimates can be compared, the non-imputed dataset was analyzed using a two-stage approach (a full bivariate weighted fixed-effect model was used; for details see Chapter 4). This analysis yielded R'2rial = 0.913 with CI95% = [0.877; 0.949] and Rndiv = 0.920 with CI95% = [0.912; 0.927].

Overall, the results indicate that there was a good agreement between the trial- and individual-level surrogacy estimates that were obtained in the MI-based and non-MI-based approaches. Both analyses lead to the conclusion that S =BPRS is a good surrogate for T =PANSS at both the level of the trial (treating physician) and the individual. Thus, the treatment effect on T can be predicted with high accuracy based on the treatment effect on S (high R2rial), and T can be predicted with high accuracy based on S (high R2ndiv).

<<   CONTENTS   >>

Related topics