Hypotheses Testing versus Confidence Interval
Shein-Chung Chow in Innovative Statistics in Regulatory Science, 2019
In comparative clinical trials, it is not uncommon to have multiple controls (or references). For example, for assessment of biosimilarity between a proposed biosimilar product (test product) and an innovative biological product (reference product), there may be multiple references, e.g., a US-licensed reference product and an EU-approved reference version of the same product. In this case, the method of pairwise comparisons is often applied. When two reference products (e.g., a US-licensed reference and an EU-approved reference) are considered, the method of pairwise comparisons includes three comparisons (i.e., the proposed biosimilar product versus the US-licensed reference product, the proposed biosimilar product versus the EU-approved reference product, and the US-licensed reference product versus the EU-approved reference product).
Comparing Means
Shein-Chung Chow, Jun Shao, Hansheng Wang, Yuliya Lokhnygina in Sample Size Calculations in Clinical Research: Third Edition, 2017
In practice, a question concerning when pairwise comparisons or a simultaneous comparison should be used often arises. To address this question, consider the following example. Suppose a sponsor is investigating a pharmaceutical compound for the treatment of patients with cancer. The investigator is not only interested in showing the efficacy of the test drug but also in establishing dose-response curve. To achieve this study objective, a four-group parallel trial is designed with four treatments: P (Placebo), A (10 mg), B (20 mg), and C (30 mg). Let μP, μA, μB, and μC represent the true mean of the clinical response under the four treatments, respectively. Since the primary objective of the trial is to demonstrate the efficacy of the test drug, the following hypotheses for pairwise comparison with the placebo are useful for the demonstration of the efficacy of the test drug.
Statistical Methods for Assessment of Biosimilars
Wei Zhang, Fangrong Yan, Feng Chen, Shein-Chung Chow in Advanced Statistics in Regulatory Critical Clinical Initiatives, 2022
The United States FDA, on the other hand, suggests pairwise comparisons should be performed. FDA's recommendation three comparisons, namely, (i) BP vs US, (ii) BP vs EU, and (iii) US vs EU. As indicated by the ODAC (held on 13 July 2017), FDA's recommended method of pairwise comparison suffers from the following limitations: (i) pairwise comparison does not use the same reference product in the comparison (i.e., these comparisons use different similarity margins), and (ii) pairwise comparison does not utilize all data collected from the test and two reference groups. In addition, pairwise comparison may not be able to detect the following possible relationship (pattern) among BP, US, and EU under the three-arm parallel-group design: (i) US > BP > EU, (ii) US > EU > BP, (iii) BP > US > EU, (iv) BP > EU > US, (v) EU > BP > US, and (vi) EU > US > BP.
Assessment of safety culture among job positions in high-rise construction: a hybrid fuzzy multi criteria decision-making (FMCDM) approach
Published in International Journal of Injury Control and Safety Promotion, 2018
In both the AHP and ANP techniques, integration of the mathematical model with human judgment is necessary. A pairwise comparison (PC) should always be carried out between all pairs of criteria as well as of alternatives to prioritize them. These PCs are conducted by a group of experts with sufficient knowledge in the concerned area. However, an inherent uncertainty and vagueness with the mapping of the DMs’ judgments into crisp values exists. This uncertainty in the initial stages of the decision-making process limits the chances of achieving satisfactory results. For this reason, the fuzzy AHP/ANP approach was introduced in which fuzzy linguistic numbers, e.g. ‘between four and six times more important’ are used for uncertain comparison ratios instead of crisp values (Shafiee, 2015).
Differences in metacognitive beliefs among patients with bipolar disorder with or without previous suicide attempts
Published in Nordic Journal of Psychiatry, 2019
Ender Cesur, Özge Şahmelikoğlu Onur, Evrim Erten
The assumption of equivalence of covariance matrices was not met (Box’s M = 81.952, F (30. 1571) = 2.645, p < .001). For this reason, Pillai’s trace was used as a multivariate test, because it is more resistant to violations of MANOVA’s covariance assumption. There were statistically significant differences among the groups in terms of both the scale scores (V (s) = 0.657, F (10.450) = 21.997, p < .001, partial eta2 = 0.33). Partial eta2 was used because the magnitude of eta for each particular effect depends to some degree on the significance and number of other effects in the design. Partial eta minimizes the effects of this issue. As the test results of the MANOVA were statistically significant, univariate test results were examined. There were statistically significant differences among the groups in terms of all variables. Pairwise comparisons were made in order to clarify the differences.
Motor demands of cognitive testing may artificially reduce executive function scores in individuals with spinal cord injury
Published in The Journal of Spinal Cord Medicine, 2021
Jinhyun Lee, Shauna Dudley-Javoroski, Richard K. Shields
Pilot testing of participants with SCI indicated that an SCI cohort of n > 18 would provide >80% power to detect significant differences from expected NON values (U.S. normative values). Differences between SCI and NON groups and between Proctored and Un-proctored tests were examined via the Kruskal-Wallis H Test for non-normally distributed data, followed by Dunn’s post-hoc tests of pairwise differences with Benjamini-Hochberg false discovery rate adjustment for multiple comparisons. Based on the study aims, we a priori identified the key pairwise comparisons to be used during Benjamini-Hochberg adjustment. For DCCS and Flanker (timed tests that included calculation of the Adjusted score), pairwise comparisons of interest were: 1) NON Proctored vs NON Un-Proctored; 2) NON Proctored vs SCI Proctored; 3) NON Un-Proctored vs. SCI Un-Proctored; 4) NON Un-proctored vs SCI Adjusted (to determine whether apparent NON vs SCI differences persist after applying the mathematical offset); and 5) SCI Un-proctored vs SCI Adjusted (to determine whether application of the offset yields SCI values that are different from a group of people with SCI who have minimal arm/trunk impairment). For List Sorting and Picture Sequence (untimed tests with no calculation of an Adjusted score), pairwise comparisons of interest were: 1) NON Proctored vs NON Un-Proctored; 2) SCI Proctored vs SCI Un-Proctored; 3) NON Proctored vs SCI Proctored; 3) NON Un-Proctored vs SCI Un-Proctored. Significance was P < 0.05 for all between-group analyses.
Related Knowledge Centers
- Psychometrics
- Psychology
- Law of Comparative Judgment
- Thurstone Scale
- Rasch Model
- Just-Noticeable Difference
- Potentially All Pairwise Rankings of All Possible Alternatives