Arana A, Arellano FM, Sánchez-Matienzo D, Perez-Gutthann S. Choosing the measure of relative proportional reporting in quantitative analyses of spontaneous reports. Presented at the 20th ICPE International Conference on Pharmacoepidemiology & Therapeutic Risk Management; August 22, 2004. Bordeaux, France. [abstract] Pharmacoepidemiol Drug Saf. 2004 Sep 23; 13(Suppl 1):S324.

BACKGROUND: The primary purpose of spontaneous adverse drug reaction reporting is to provide early warnings of hazards, which have not been recognized prior to marketing of a drug. One of the signal detection techniques consists of measuring the relative proportional reporting (RPR)

METHODS: This tool’s ‘null hypothesis’ holds that drugs have no side effects and it is assumed that all adverse events (AE) are random events and that all drugs should have similar AE profiles. It follows that each drug’s contribution to the number of a certain type of AE should be proportional to its overall contribution to the AE database. A is the number of reports with an AE of interest for a given drug, B is the same number for all the other drugs in the database, C is the number of all other AE reported with the drug of interest, and D is all other AE for all other drugs. The measures of RPR are expressed in three ways. There is not consensus on the names The proportional reporting ratio PRR¼(A/(AþC))/ ((AþB) / (AþBþCþD)) The proportions ratio (relative risk) RR¼(A/(AþC))/ (B/(BþD)) The odds ratio OR¼(A/C)/(B/D) The PRR is systematically biased to the null as the average against which a drug of interest is compared to contains the drug of interest. As in proportional mortality studies (Miettinen, OS. AJE 1981; 114: 144–148), the RR depends on the number of ‘other AE’ and how common they are in relation to the AE of interest. The OR is free from this arbitrary element and is independent of the number and characteristics of ‘all other AE’. The OR can be interpreted as the observed to expected ratio or the standardized risk ratio in the assumption that the reporting rate for the ‘other AE’ is unrelated to exposure.

RESULTS: The following simplified example shows these effects in RPR measures: A fictitious database contains for Drug A, one AE of Rash, and 10 ‘other AE’ (9%). For all other drugs there is one AE of Rash, and 100 ‘other AE’ (1%). The PRR is 5.1, the RR 9.2, and the OR 10. In one year, for Drug A there are 10 AE of Rash, and 100 ‘other AE’ (9%). For all other drugs there are 1000 AE of Rash and 100 000 ‘other AE’ (1%). The PRR is 9.1, the RR 9.2, and the OR 10. Imagine an antidote is developed that prevented 50% of the AE other than Rash. Now, for Drug A there are 10 AE of Rash, and 50 ‘other AE’. For all other drugs there are 1000 AE of Rash, and 50 000 ‘other AE’. The PRR would be 8.4, the RR 8.5, and the OR 10.

CONCLUSION: The reporting ORis the measure of choice. It is not influenced by the size and characteristics of the ‘other AE’ like the PRR and the RR.

Share on: