When evaluating a clinical trial, readers often jump to the P value of the primary endpoint to determine whether the results of a trial are “statistically significant” or not. Although the P value is truly a continuous variable, the scientific community has been conditioned to disregard all results with P values ≥ 0.05, but to fully endorse any trials with a “statistically significant” P value less than 0.05.
Putting the debate and controversy about P values aside for the moment, as a reader, would you be less impressed with a study that changed from being statistically significant to insignificant if one single patient changed from not having the primary endpoint to having the primary endpoint? Especially in an era with a blind reliance on P values, the knowledge of the “fragility” or “robustness” of a study’s P value is another useful data point for readers to critically understand and analyze the results of a clinical trial.
The Concept of the “Fragility Index” for Clinical Trials
The “fragility index” is the number of patients that would need to convert from a “non-event” to an “event” outcome in order to make the study lose statistical significance (p ≥ 0.05). The larger the fragility index, the more robust the results of a trial are.
As an example, consider the PROTECT study (N Engl J Med. 2011 Apr 7;364(14):1305-14). In this trial comparing dalteparin versus heparin for VTE prophylaxis in critically ill patients, the authors concluded in a secondary analysis that patients receiving dalteparin were less likely to experience pulmonary embolism than patients in the heparin group:
|Heparin Group (N=1862)||Dalteparin Group (N=1862)||P Value|
At first glance, the results of this trial seem “very significant”. Because the P value isn’t even close to 0.05, one might falsely conclude that this analysis is very robust and that surely dalteparin reduces the risk of pulmonary embolism compared with heparin prophylaxis.
Using ClinCalc.com’s new Fragility Index Calculator, the fragility index of this analysis is “2”, meaning that if two patients in the dalteparin group were “counted” as having a pulmonary embolism, the analysis would lose statistical significance (calculating out to P=0.051).
Calculation of Fragility Index
The fragility index is calculated by analyzing the results of a binary endpoint using Fisher’s exact test. If the P value is significant (p < 0.05), one patient in the group with the lowest incidence rate of the endpoint is converted from a “non-event” (such as “no pulmonary embolism”) to an “event” (such as “having a pulmonary embolism”). This process is continued until the Fisher’s exact test results in a non-significant P value (≥ 0.05):
ClinCalc’s Fragility Index Calculator
ClinCalc.com is excited to announce our newest calculator — the Fragility Index Calculator. As described above, the calculator will take the input of a clinical trial and calculate the number of patients required to create a non-significant (p ≥ 0.05) result. An example of the calculator output is included below: