Minimally important changes do not always reflect minimally important change; moreover, there is no need for them

Excerpt

In response to the recently published letter: “Likely change indexes do not always index likely change; moreover, there is no need for them” [1], we suggest instead that this sentiment applies to “minimally important change.” The author correctly points out that the likely change index (LCI) [2] is similar to the reliable change index (RCI), a significance test for within individual change. The RCI is calculated as: \(({X}_{1}- {X}_{2})/\sqrt{2} SEM\), where the X₁ and X₂ are PRO scores for an individual patient at baseline and a follow-up, the SEM (standard error of measurement) is the \({SD}_{1}\sqrt{1-reliability}\) (SD₁ = standard deviation at baseline, we note that the SD of the change score can be used instead). The LCI was proposed because many patients who report that they have changed (gotten better or worse) will be classified as not changed using the conventional p < 0.05 threshold with the RCI. The amount of change (critical value) needed to be significant on the RCI is known as the coefficient of repeatability: \(critical value*\sqrt{2} SEM\). For p < 0.05, the critical value is 1.96, but this value decreases as the p-value increases. As we demonstrate in our article, LCI thresholds for 68% and 50% confidence tended to align more with anchor-based estimates of meaningful change and, we suggest, may produce more accurate individual classification overall [2]. The first assertion in the letter is that LCI interpretation relies upon a statistical fallacy: that the significance level used for the LCI will indicate the probability of real change–for example, that the LCI using a p-value of p = 0.32 would indicate at 68% probability that the patient experienced true change. While it is correct that this interpretation would support a statistical fallacy, we did not use this interpretation for the LCI. We also do not agree that using other p-values than p < 0.05 is “awkward.” It is increasingly recognized in scientific and statistical communities that the dogma of adhering to p < 0.05 can be more harmful than beneficial. …