Background : The safety of newly approved drugs must be assessed using postmarketing surveillance data. One of the difficulties in assessing the hazard rates of adverse events induced by the anti-cancer drug TS-1 was that the time to event was not exactly identified due to the interval censoring. Most patients were outpatients who underwent clinical laboratory tests almost periodically at 1- or 2-week intervals and therefore, the occurrence of an adverse event was confirmed at the time of testing days after the event occurrence.
Objective : The purpose of this study was to propose a new model of hazard functions for each of 4 items of adverse event induced by TS-1 using post-marketing surveillance data considering the interval censoring.
Methods : The data obtained from 3, 294 patients with gastric cancer who received an initial 4-week course of therapy with TS-1 administered orally twice a day, followed by a 4-week second course with a 2-week no-treatment period after the initial course, were used to estimate hazard functions. Four items of adverse event--hemoglobin level (HB), white blood cell (WBC), neutrophil (NEUT) and platelet counts (PLT) --were graded, respectively, using the criteria established by the Japan Society of Clinical Oncology. Slip-mixed log-logistic and slip-mixed Weibull models were proposed as candidate models for estimating hazard functions. The goodness of fit of the two candidate models was evaluated by applying them to the above-mentioned data. The hazard functions for each of 4 items were assessed using the model with the better fit.
Results : The initial occurrence of adverse event was shown to follow the slip-mixed log-logistic model for each of 4 items. Although most events occurred early on in the initial course of therapy, a small peak in HB was also observed in the second course, while no such peak appeared for the other items.

Objective : The incidence rate is used frequently in drug safety assessment. The incidence rate of adverse events is defined as the number of patients experiencing a certain adverse event divided by the number of patients administered a drug in spite of duration of administration (observation). In post-marketing surveillance, the duration of administration (observation) typically differs by patient and most of the analyses fail to take into account the differences in duration of administration (observation). Therefore, we investigated the usefulness of hazard functions in a drug safety assessment using the interim results from Clinical Experience Investigation of the oral anticancer drug, TS-1.
Methods : About three thousand patients with gastric cancer were enrolled in this Clinical Experience Investigation. TS-1 was administrated orally twice daily. One course consisted of consecutive administration for 28 days and 14 days rest. Administration was repeated in two courses. Hematological measurements, stomatitis, anorexia, nausea/vomiting, diarrhea, malaise were analyzed. Adverse events were evaluated in accordance with the criteria of the Japan Society for Cancer Therapy, which were established based on criteria established by the WHO. Time to occurrence of an adverse event was calculated from the first day of administration until the adverse event was first observed. Hazard functions were estimated by smoothing methods using kernel functions.
Results : The occurrence of adverse events using smoothed hazard functions had one peak around 10 days in the first course and decreased by administration rest. With the resumption of administration, the occurrence increased again. The occurrence in the second course were less than that of the first course.
Conclusion : The occurrence peaks of adverse events were estimated graphically by smoothed hazard functions. We conclude that hazard functions are useful as an analytical tool in drug safety assessment.