Disproportionality analysis asks one question of a spontaneous reporting database: is this event reported more often with this drug than with everything else? The counts go into a 2x2 table:
|
The event |
All other events |
| The drug |
a |
b |
| All other products |
c |
d |
- a: reports with the drug and the event;
- b: reports with the drug and any other event;
- c: reports with the event for all other products;
- d: reports for all other products and all other events.
The trap is in b and d: databases give you totals, and the table needs the totals minus the event. If a drug has 250 reports and 6 include the event, b is 244, not 250.
Each statistic compares a reporting proportion, not a rate of the event in patients: the database knows how many reports were sent, not how many people took the drug.