ORIGINAL ARTICLE The McNemar Change Index worked better than the Minimal Detectable Change in demonstrating the change at a single subject level Antonio Caronni a, * , Michela Picardi b , Giulia Gilardone b , Massimo Corbo b a IRCCS Fondazione Don Carlo Gnocchi Onlus, via Alfonso Capecelatro 66, 20148 Milano, Italy b Department of Neurorehabilitation Sciences, Casa di Cura del Policlinico, Via Dezza 48, 20144 Milano, Italy Accepted 18 November 2020; Published online 24 November 2020 Abstract Background and Objective: To assess the agreement between the Rasch Change Index (RCI), minimal detectable change (MDC), and McNemar Change Index (McCI), three statistics for demonstrating the patient’s improvement/deterioration. Methods: The Mini-Balance Evaluation Systems Test (Mini-BESTest (MB)) (a balance scale developed with the Rasch analysis) was administered before and after rehabilitation to 315 neurological patients. The MB RCI was chosen as the criterion standard for detecting the patient’s improvement. Positive likelihood ratios and negative likelihood ratios (PLRs and NLRs, respectively) were used to evaluate the MDC and McCI accuracy in identifying the patient’s improvement. Three different MB MDCs were assessed. Results: One-hundred patients improved their MB in accordance with the RCI. All three MDCs and the McCI were solid in ruling out the patient’s improvement (NLR !0.2). The McCI and the largest MDC were also good in detecting the patient’s improvement (PLRO5), whereas the smaller MDCs were not. Of the four indices, McCI was the most robust in case of missing items. Conclusion: A patient stable in accordance with the MDCs or McCI is actually stable as per the criterion standard. To be reasonably sure that the patient is actually improved, larger MDC values or the McCI should be preferred, and the McCI is preferable if there are missing items. Ó 2020 Elsevier Inc. All rights reserved. Keywords: Minimal detectable change; Assessing the change; Assessing treatment outcome; Disability evaluation; Psychometrics; Rehabilitation 1. Introduction Demonstrating that the patient is changed, such as improved after treatments or worsened because of the dis- ease progression, is of obvious importance. This is not a trivial problem. Every measure is corrupted by error [1], and two measures of the same individual are likely different even when there has been no modification of the patient’s status. Conversely, in front of a modification of a patient’s measure, the clinician wonders if this modification indi- cates a genuine modification of the patient or just reflects the measurement error. There is a wealth of literature dedicated to the minimal detectable change (MDC) [2,3], the smallest modification of a patient’s measure exceeding the measurement error. Thanks to the MDC, the clinician can easily identify the patient who benefitted from therapies and the patient who did not. In addition, he/she can find as well the patient who significantly deteriorated, for example, because of a progressive disease. Applications of the MDC in clinical trials are also important. As an example, the MDC can be used for calculating the sample size [4] and the number needed to treat [5]. The MDC is commonly used with ordinal scales, which are most of the outcome measures in rehabilitation. Howev- er, in a strict mathematical sense, this practice should be avoided given that the MDC is grounded in mean and stan- dard deviation calculation, and so its use should be limited to interval or ratio level measurements [6]. Moreover, the MDC should be calculated for each of the different out- comes measures while, at the present moment, the MDC is known for only a portion of them. The MDC could also be different in different groups of individuals (e.g., controls vs. patients [7]) and a specific MDC should be calculated for each of them. Finally, the MDC should only be used with no missing items (i.e., full questionnaire, no unscored items). Declarations of interest: none. * Corresponding author. IRCCS Fondazione Don Carlo Gnocchi On- lus, via Alfonso Capecelatro 66, 20148, Milano, Italy. Tel.: þ39 02 403081. E-mail address: acaronni@dongnocchi.it (A. Caronni). https://doi.org/10.1016/j.jclinepi.2020.11.015 0895-4356/Ó 2020 Elsevier Inc. All rights reserved. Journal of Clinical Epidemiology 131 (2021) 79e88