A spontaneous reporting database holds 100,000 reports. For Drug D and event E the counts are:
| Event E | All other events | Total |
|---|
| Drug D | a = 30 | b = 970 | 1,000 |
| All other drugs | c = 120 | d = 98,880 | 99,000 |
| Total | 150 | 99,850 | 100,000 |
Step 1: proportional reporting ratio. PRR = [a ÷ (a + b)] ÷ [c ÷ (c + d)] = (30 ÷ 1,000) ÷ (120 ÷ 99,000) = 0.03 ÷ 0.001212 = 24.75. Event E makes up 3% of Drug D's reports but only about 0.12% of reports for other drugs.
Step 2: reporting odds ratio. ROR = (a × d) ÷ (b × c) = (30 × 98,880) ÷ (970 × 120) = 2,966,400 ÷ 116,400 = 25.48.
Step 3: 95% confidence interval for the ROR. Standard error of ln(ROR) = square root of (1/30 + 1/970 + 1/120 + 1/98,880) = 0.2067. ln(25.48) = 3.238, so the interval is e to the power of (3.238 ± 1.96 × 0.2067), giving 17.0 to 38.2. The lower limit is well above 1.
Step 4: expected count. If Drug D were reported with event E at the background rate, we would expect 1,000 × 150 ÷ 100,000 = 1.5 reports. We have 30.
Step 5: interpret. By common screening thresholds (for example a PRR of at least 2 with at least 3 cases and a chi-squared of at least 4, criteria often attributed to Evans and colleagues), this is a signal of disproportionate reporting. It is a reason to review the cases, not proof that Drug D causes event E. Reporting bias, publicity, confounding by indication and duplicate reports can all inflate the numbers, so the next step is clinical review of the individual cases.