"5 and 20 Rule" (Sackett et al, 2000): While easy to remember, it may oversimplify the problem if the selected outcome is infrequent.
|
Loss to Follow-Up
|
Interpretation
|
|
< 5%
|
conclusions probably acceptable
|
|
5 - 20%
|
intermediate
|
|
> 20%
|
validity of conclusions seriously open to question
|
Rule of Thumb of Schulz and Grimes: The loss to follow-up should not exceed the outcome event rate.
Estimating the possible impact of missing patients lost to follow-up (Sackett et al, 1991):
(1) Calculate the fraction of patients who developed a particular outcome:
fraction as measured =
= (patients with outcome) / (all patients not lost to follow-up)
(2) Calculate the possible outcome range if follow-up patients were available:
best case scenario =
= fraction of patients if none of the follow-up patients had the outcome =
= (patients with outcome) / ((all patients not lost to follow-up) + (patients lost to follow-up)) =
= (patients with outcome) / (patients initially enrolled)
worst case scenario =
= fraction of patients if all of the follow-up patients had the outcome =
= ((patients with outcome) + (patients lost to follow-up)) / ((all patients not loss to follow-up) + (patients lost to follow-up))
= ((patients with outcome) + (patients lost to follow-up)) / (patients initially enrolled)
(3) If the fraction as measured is similar to the two assumed fractions, then impact of the patients lost to follow-up is probably not great.
(4) If the there is a "significant" difference between the two fractions including patients lost to follow-up, then the outcome reported may not be valid.
NOTE: "Significant" is not defined, but a ratio of the fractions from the worst case to best case scenarios may be a simple measure. The acceptable example had a ratio of 1.1 while the unacceptable example had a ratio of 7.1. Probably a ratio < 1.5 is good and a ratio > 4 is high.
