Companion web tool · Selim Medical Press

Signal Management in Medical Device Vigilance

Six calculators and records from the book, by Dr. Hafez Selim, MD, PhD. Book and toolkit home · Case Lab: the full process with solved cases

Educational tool. It shows the methods taught in the book. It is not validated software and not a controlled quality record. Every threshold, band, score anchor, matrix and response time here is an example of a company design, not a regulatory requirement. Book examples use fictional cases with illustrative numbers. The tool runs only in your browser. It sends nothing and stores nothing, and it never fills in a value you did not enter or load.
Chapter 8

Trend screen: p-chart limits, run rule and CUSUM

Expected count = units × baseline proportion. SD = √(n × p0 × (1 − p0)). Limits = expected + 2 or 3 × SD. The run rule flags 2 of 3 consecutive points beyond the upper 2σ limit. The standardized CUSUM is S(t) = max(0, S(t−1) + z − k), with z = (observed − expected) ÷ SD, and it signals when S exceeds h.

Fixed before monitoring starts. The rate must be above 0 and below 1,000.
Enter one value for every period, or one value per period separated by spaces or semicolons. You may write thousands with commas, for example 50,000.
Usually half the shift to detect. Must be 0 or more.
Must be above 0.
Whole numbers separated by commas, spaces, semicolons or new lines. Do not use thousands separators here.
Separate labels with commas, for example "Y3 M10, Y3 M11". If you give labels, give one for each count.
Chapter 9

Disproportionality screen and registry observed-to-expected

A ratio shows that an event is overrepresented in reports. It is not an event rate or a relative risk. An O/E flag shows more revisions than the case mix predicts. It does not show the cause.

A. Disproportionality 2×2 table (one product code)

Company screening rule (optional; fill in all three or leave all three empty)

Whole numbers. You may write thousands with commas, for example 8,910. The 95% intervals use the log method: SE of ln(PRR) = √(1/a − 1/(a+b) + 1/c − 1/(c+d)), and SE of ln(ROR) = √(1/a + 1/b + 1/c + 1/d).

B. Registry observed-to-expected with exact Poisson limits

A whole number, 0 or more.
Must be above 0.

The 95% interval for O/E is the exact Poisson interval for O divided by E. The funnel limits are exact Poisson quantiles at E: 2.5% and 97.5% for the 95% limits, and 0.1% and 99.9% for the 99.8% limits.

Chapter 10

Time to failure: life table with Kaplan–Meier survival

For each interval, enter the devices at risk at its start and the failures in it. Survival in the interval is 1 − (failures ÷ at risk), and cumulative survival is the product of these values. The Greenwood variance is S² × Σ [d ÷ (n × (n − d))], and the 95% interval is linear, S ± 1.96 × SE, truncated at 0% and 100%.

Life-table input rows
Interval label (optional)Devices at risk at startFailures in interval
Chapter 13

Signal prioritization: five-domain scale

Score each domain from 1 to 5 against the anchors, and write a rationale for each score. Scores 2 and 4 fall between the anchors shown. The example company's bands are A (20–25), B (14–19) and C (5–13). A signal with severity 5 and detectability 4 or 5 goes to Band B at least. Any urgency trigger sends it to Band A. The board may move a signal one band up or down. Either move needs a written reason, and a downward move also needs the chair's signature. Prioritization never changes a legal reporting clock.

Urgency triggers (tick any that apply)
Board override

The override is recorded on the worksheet next to the score, so the reason stays with the signal record.

Chapter 18

Risk-file update: probability band and risk region

This module uses the illustrative company 5×5 matrix from Chapter 18 (Tables 18.1 and 18.2), which is not a regulatory requirement. Enter the risk-file estimate from before the signal and the post-market data, using the exposure of the affected population only. Placing a risk in a cell does not replace the written ALARP and state-of-the-art judgment.

For example "pump-years". If you leave this empty, the tool uses "unit-years".
Before the signal (risk-file estimate)
N must be 1 or more.
After the signal (post-market estimate)
Post-control target (optional; a verification target, not a result)
Chapters 19 and 23

Decision record

This module builds a printable record from the signal-record fields (Tool T-21) and the FSCA decision-record fields (Tool T-17). The record keeps the conclusion (confirmed, refuted or indeterminate) separate from the status, which is administrative. "Closed" is a status, not a finding. Write the record on the day of the decision. Later additions are new, dated entries.

Identification
Detection
Validation, assessment and conclusion
This is a finding. It is never "closed".
Decision
Status and closure
This is a status. It records whether work continues, not what was found.
Anyone must be able to tell whether a trigger has fired without asking the author, for example "any month above 3σ". "Reopen if concerns arise" is not a trigger.

* Required.