A new test statistic for the Kodlin-Standish drug surveillance model

✍ Scribed by Hans K. Ury

• Publisher: Elsevier Science
• Year: 1972
• Tongue: English
• Weight: 832 KB
• Volume: 5
• Category: Article
• ISSN: 0010-4809
• DOI: 10.1016/0010-4809(72)90061-4

✦ Synopsis

For the purpose of drug surveillance, i.e., for continuous evaluation of possible associations between drugs and pathological manifestations (events), Kodlin and Standish (I) proposed a response time model involving the time of first drug administration and the event appearance time; by means of explicit parameter estimates, computed from samples of such times which are extracted from routine medical records, they then derived a test statistic A. Using the same mathematical model, we propose a related test statistic x which appears to be as robust as A, more stable for small sample sizes, and somewhat more powerful, on the average, over linear models. The performance of both test statistics is evaluated by Monte Carlo methods and x is calculated for the numerical example (Penicillin and "Rash") given by Kodlin and Standish (I). Like d, x might be generally useful in testing for independence of two phenomena whose occurrence times are observable. Another test statistic is also investigated.