sensitivity analysis

Sensitivity analysis refers to repeating primary analysis calculations after modifying certain criteria or variable ranges; the goal is to determine whether such modifications significantly affect the results initially obtained. With any study, a number of decisions are made based on certain assumptions or by consensus (eg, choosing 5-year rolling windows as opposed to 2or 8-year rolling windows). If results of sensitivity analyses conducted with slight changes in certain variables (eg, changing rolling window period, changing cutoffs for average glucose levels by ±10%) are similar to the initial results, the investigators can be more confident of the robustness of their results.

In this case, patients with diabetes who had high fluctuations in glucose levels were excluded. If, on repeat analysis excluding those patients, the p-value for the association between average glucose levels and dementia remains statistically significant among the remaining patients with diabetes, this would further confirm the investigators' conclusions as it would make it less likely that high fluctuations in glucose levels explain the different hazard ratios obtained.