1
296 Abstracts P-45 THE DATAFAX PROJECT D. Wayne Taylor and Eric G. Bosch Chmcal Ep~demlologyand B~ostat~st~cs McMaster Umverszty Hamilton, Canada The DATAFAX project was designed to assess the feas~bd~ty of (1) transmitting original data collections forms by ordinary FAX machines d~rectly ~nto a computer at the coordinating center, (2) creating software to read both check boxes and handwritten numbers from the FAX ~mages and to enter the results d~rectly ~nto a study database, and (3) presenting the study data clerk w~th a synchromzed spht screen v~ew of the FAX ~mages and corresponding data entry screens for data vahdat~on, w~th unreadable fields h~ghhghted m color Potential advantages of such a system ~nclude 1 Rap~d return and central review of study data 2 Reduced burden on central data management staff w~th a sh~ft m focus from data entry to vahdat~on 3 Abd~ty to retain original source documents at the centers for later reference during the study The key advantages of th~s approach compared to a PC-based d~stnbuted data entry system ~nclude 1 Improved quahty control and data security 2 Respons~b~hty for data entry and computer sk~lls are not placed on participating chnlcal personnel 3 Reduced start up t~me and ~ncreased flex~b~hty to add and modify data collection forms Progress to date ~n a large cross-Canada study ~nd~cates that th~s approach ~s feasible Accuracy ~n reading check boxes ~s over 99% Handwntten numbers can be correctly ~dent~fied 95% of the ttme by following three s~mple rules and 85% of the t~me w~th no ~nstruct~on at all We demonstrate the current version of the DATAFAX software running on a SUN SPARCstat~on 1 P-46 ESTIMATING RATES OF CHANGE IN RANDOMIZED STUDIES N. Laird and F. Wang Syntex Laboratones Palo Alto, Cahforma Th~s paper deals w~th the extension of the pretest posttest experimental design to the longitudinal data and chn~cal trial setting We assume a basehne (or pretest) measurement ~s taken on all subjects, who are then randomized, w~thout regard to baseline values, to a treatment group Repeated measurements are taken postrandom~zat~on at specified t~mes. Our objective ~s to estimate the average rate of change (slope) m the experimental groups and the differences ~n the slopes We contrast two d~fferent approaches a multivariate one that regards the entire vector of responses 0nclud~ng the basehne) as random outcomes, or a un~vanate one that uses each subject's least squares slope as an outcome. Our mult~vanate approach ~s essentially a generahzat~on of Stanek's (1987) Seemingly Unrelated Regression (SUR) estimator for the pretest-posttest design The multivariate approach ~s natural to apply ~n th~s setting, and optimal ~f the assumed model ~s correct However, the most efficient estimator requires assuming that the basehne mean parameters are the same for all experimental groups Whde th~s assumption ~s reasonable ~n the randomized setting, the resulting multivariate estimator uses post-randomization data as a covanate, ~f the assumed hnear model ~s not correct, th~s can lead to d~stort~ons ~n the estimated treatment effect We propose ~nstead a reduced-form multivariate estimator that may be somewhat less efficient, but protects against model m~sspec~ficat~on

·P-45 The datafax project

Embed Size (px)

Citation preview

Page 1: ·P-45 The datafax project

296 Abstracts

P-45 THE DATAFAX PROJECT

D. Wayne Taylor and Eric G. Bosch Chmcal Ep~demlology and B~ostat~st~cs

McMaster Umverszty Hamilton, Canada

The DATAFAX project was designed to assess the feas~bd~ty of (1) transmitting original data collections forms by ordinary FAX machines d~rectly ~nto a computer at the coordinating center, (2) creating software to read both check boxes and handwritten numbers from the FAX ~mages and to enter the results d~rectly ~nto a study database, and (3) presenting the study data clerk w~th a synchromzed spht screen v~ew of the FAX ~mages and corresponding data entry screens for data vahdat~on, w~th unreadable fields h~ghhghted m color

Potential advantages of such a system ~nclude 1 Rap~d return and central review of study data 2 Reduced burden on central data management staff w~th a sh~ft m focus from data entry to vahdat~on 3 Abd~ty to retain original source documents at the centers for later reference during the study

The key advantages of th~s approach compared to a PC-based d~stnbuted data entry system ~nclude 1 Improved quahty control and data security 2 Respons~b~hty for data entry and computer sk~lls are not placed on participating chnlcal personnel 3 Reduced start up t~me and ~ncreased flex~b~hty to add and modify data collection forms

Progress to date ~n a large cross-Canada study ~nd~cates that th~s approach ~s feasible Accuracy ~n reading check boxes ~s over 99% Handwntten numbers can be correctly ~dent~fied 95% of the ttme by following three s~mple rules and 85% of the t~me w~th no ~nstruct~on at all We demonstrate the current version of the DATAFAX software running on a SUN SPARCstat~on 1

P-46 ESTIMATING RATES OF CHANGE IN RANDOMIZED STUDIES

N. Laird and F. Wang Syntex Laboratones Palo Alto, Cahforma

Th~s paper deals w~th the extension of the pretest posttest experimental design to the longitudinal data and chn~cal trial setting We assume a basehne (or pretest) measurement ~s taken on all subjects, who are then randomized, w~thout regard to baseline values, to a treatment group Repeated measurements are taken postrandom~zat~on at specified t~mes. Our objective ~s to estimate the average rate of change (slope) m the experimental groups and the differences ~n the slopes

We contrast two d~fferent approaches a multivariate one that regards the entire vector of responses 0nclud~ng the basehne) as random outcomes, or a un~vanate one that uses each subject's least squares slope as an outcome. Our mult~vanate approach ~s essentially a generahzat~on of Stanek's (1987) Seemingly Unrelated Regression (SUR) estimator for the pretest-posttest design

The multivariate approach ~s natural to apply ~n th~s setting, and optimal ~f the assumed model ~s correct However, the most efficient estimator requires assuming that the basehne mean parameters are the same for all experimental groups Whde th~s assumption ~s reasonable ~n the randomized setting, the resulting multivariate estimator uses post-randomization data as a covanate, ~f the assumed hnear model ~s not correct, th~s can lead to d~stort~ons ~n the estimated treatment effect We propose ~nstead a reduced-form multivariate estimator that may be somewhat less efficient, but protects against model m~sspec~ficat~on