390 Abstracts

A32 IMPROVING THE CLASSIFICATION OF STAGE: A POTENTIAL PROBLEM

FOR COMPARATlVE STUDIES

Linda Ward, for the Brltlsh Stomach Cancer Group CRC Trials Unit

University of Birmingham Birmingham, England

Changes designed to improve stage as a measure of prognosis may have implications for comparative studies (Feinstein, 1985). Precise classification of gastric cancers requires detailed surgical and pathological reporting, plus considerable expertise. We compared 249 cases entered to an adjuvant RCT during 1976- 80 vs 1242 concurrent, similar patients identified at the W. Midlands Cancer Registry. Cases were reviewed to identify suitable groups for comparison with the randomized patients. Three similar stage systems were applied: 1) basic (palliative or curative surgery, node and serosal involvement), 2) our trial system (adding histological involvement of resection lines) and 3) the registry criteria (using depth of penetration). The overall effect on numbers and proportion surviving to 5 years is shown:

Total available

Stage 1 Stage 2 Stage 3 Curative (NK pah) Stage 4 (Pall. op.) *Not evaluable

Basic Trial (n = 1425) (n = 1378)

No. (% at 5 yrs) No. (% at 5 yrs)

123 (46) 79 (48) 223 (29) 136 (29) 616 (9) 639 (11)

61 (23) 122 (32) 402 (3) 402 (3)

66 (11) 113 (12)

Registry (n = 1446)

No. (% at 5 yrs)

32 (66) 257 (32) 606 (10) 152 (22) 339 (2)

45 (22)

*Excluded from statistics

The registry provided the best prognostic indicator by going back to pathologists to confirm cases of very early stage 1 disease. The trial system was the least effective as more non-trial cases were unevaluable or assigned to the unspecified group due to missing resection line data. Stages 1 to 3 in registry and trial were not directly comparable. The BASIC system represented the best reclassification possible with the available data. We found systematic differences in classification and completeness of assessment between patients in the RCT and normal cancer registrations. This could introduce bias and caution may be needed when comparing reported survival rates by stage of disease.

A33 CLASSIFICATION OF PROTOCOL DEVlATlONS IN A COMPLEX STUDY

Claudla S. Moy for the Collaborative Ocular Melanoma Study Assurance Committee The Johns Hopkins University

Baltimore, Maryland

In any clinical trial, adherence to the protocol is central to the scientific integrity and validity of the study. Serious bias in the analysis and interpretation of treatment comparisons may ensue if a substantial proportion . of enrolled patients should have been excluded based on elrgrbrlrty criteria, if randomized patients did not receive the assigned treatment, or if they received treatment in a fashion that deviated seriously from the expected regimen. Thus, a system for identifying, documenting, and reporting deviations is essential.

In multicenter clinical trials, protocol monitoring is complex and there are many opportunities for protocol deviations to occur. The protocol monitoring experience of the Collaborative Ocular Melanoma Study (COMS), a set of two multicenter, randomized clinical trials encompassing 42 clinical centers and 5 resource centers, will be discussed. Protocol deviations have been classified into two main categories: ma/or, indicating a substantial departure from the protocol that might affect treatment comparisonti or other analyses of study objectives; and minor, a deviation that is not likely to affect the analyses. Deviations were further categorized according to whether they were related to enrollment or treatment and whether or not they could have been prevented by the clinical investigator. Overall, very few deviations have been detected. For example, for the nearly 1100 patients enrolled in the first five years of data collection, only 12 (1.1%) major deviations related