M.Marvin Newhouse

The Markov process as a general method for nonparametric analysis of right-censored medical data

The product limit method of Kaplan and Meier for estimating survival functions and the logrank test of Mantel are widely employed for analysis of longitudinal medical data. Developed for analysis of one-time events such as death, survival analysis is also commonly adapted to more complex states such as loss of vision or cancer remission by restricting analysis to first occurrences. The nonparametric discrete time nonhomogeneous Markov process is proposed as a better model for any applications of the latter type. This simple stochastic model allows for an arbitrary number of possible states and for transitions in any direction. Maximum likelihood estimators are easily computed for the stochastic model and are identical to the product-limit estimates in the special case represented by the Kaplan-Meier model. The logrank test extends to evaluation of differences between populations with respect to any specified transition.

Time trends in personnel certification and turnover in the Collaborative Ocular Melanoma Study

Background The purpose of this analysis was to examine and quantify the coordinating center workload in regard to certification and turnover of clinical personnel who participated in the Collaborative Ocular Melanoma Study (COMS). Methods The COMS was a set of multicenter, multidisciplinary randomized clinical trials conducted to evaluate radiotherapy for initial treatment of choroidal melanoma, a primary eye cancer. Patients were screened, enrolled, treated, and followed for mortality and clinical outcomes at 43 clinical centers in the United States and Canada, beginning late in 1986. Clinical centers were added to the COMS during three calendar periods. Training and certification of clinical personnel was performed by five resource centers and coordinated at the COMS coordinating center. A personnel database was maintained at the coordinating center for tracking certification progress, recertification, transfers and resignations. Certification records were analyzed for the period 1986 through 1998 when patient accrual and treatment were completed. Results Clinical personnel (929 individuals) were certified for one or more of nine distinct roles in the COMS, resulting in 1342 certifications; 26% were certified for two or more positions. The total number certified for each role ranged from 114 to 208. Certifications per year ranged from 260 in 1987 to 35 in 1997. The certification workload at the coordinating center rose following addition of each group of new clinical centers. It also increased in 1995 as a result of recertification of current personnel mandated by the external data monitoring committee. Conclusions This analysis documented that training and certification were continuing tasks, and not limited to initiation of the study. Resignations and reassignments of certified personnel were responsible for 642 decertifications.

Feasibility of keying data from screen-displayed facsimile images in an ongoing trial: the collaborative ocular melanoma study.

As part of an ongoing clinical trial, we conducted an experiment to assess the feasibility and to determine the reliability of data entry from a computer screen display of images of data collection forms transmitted by facsimile (fax) machines directly into a computer for paper forms designed without consideration of fax or image display requirements. Feasibility was assessed on the basis of accuracy and reliability of data entry and on operator satisfaction. During a 2-week period, half of the forms received at the Collaborative Ocular Melanoma Study (COMS) coordinating center were key-entered twice, using the paper forms as the source (paper source entry). The remaining forms were entered once using paper source and were later reentered using the screen display of images of the faxed forms as the source (image source entry). The latter group of forms, or 50% of all forms received, were entered a third time, using the image source entry. Two data entry operators participated in the experiment. Discrepancy rates between and within data entry operators were calculated for both modes of entry. A total of 50,861 keystrokes (28,095 items) across 1122 records were checked for consistency. The overall discrepancy rate associated with double paper source entry was approximately 21 per 10,000 keystrokes (20 per 10,000 items). Discrepancy rates associated with paper source versus image source entry (53 per 10,000 keystrokes [57 per 10,000 items]) and double image source entry (57 per 10,000 keystrokes [47 per 10,000 items]) were similar in magnitude. Image source entry of forms received by facsimile may provide an acceptable alternative to paper entry in ongoing multicenter clinical trials where the costs of converting existing forms and systems to automated data capture may be unacceptable. This experiment confirmed the feasibility of such an alternative and suggested that improved screen displays and changes in equipment to facilitate entry of data from the screen display may enhance accuracy of entries.