Practical issues encountered while determining Minimal Clinically Important Difference in Patient-Reported Outcomes

Table 1 MCID estimations (with their 95% confidence interval) for different distribution-based methods applied to the LIGALONGO dataset using different imputation methods

	Distribution-based methods
	0.5 SD_b	1/3 SD_b	0.2 SD_b	SEM	MDC	0.5 SD_ch	1/3 SD_ch	0.2 SD_ch
Complete cases	10 [10; 11]	7 [6; 7]	4 [4; 4]	9 [9; 10]	26 [24; 28]	8 [7; 9]	5 [4; 6]	3 [3; 4]
Imputation by the mean^a	10 [9; 11]	7 [6; 7]	4 [4; 4]	9 [8; 10]	25 [23; 27]	8 [7; 9]	5 [5; 6]	3 [3; 4]
*Simple MICE imputation**^b	10 [10; 11]	7 [7; 7]	4 [4; 4]	9 [9; 10]	26 [25; 26]	8 [8; 8]	5 [5; 5]	3 [3; 3]
Complex MICE imputation^c	11 [10; 11]	7 [7; 7]	4 [4; 4]	9 [9; 10]	26 [25; 27]	8 [8; 8]	6 [5; 6]	3 [3; 3]
Simple MICE imputation (available information)^d	10 [10; 11]	7 [7; 7]	4 [4; 4]	9 [9; 10]	26 [25; 26]	8 [8; 8]	5 [5; 6]	3 [3; 3]
Complex MICE imputation (available information)^d	10 [10; 11]	7 [7; 7]	4 [4; 4]	9 [9; 10]	26 [25; 26]	8 [8; 8]	5 [5; 6]	3 [3; 3]

Values in bracket are 95% Confidence Interval
MICE Multivariate Imputation with Chained Equations, SD_b Standard deviation at baseline score (visit 1), SEM Standard Error of Measurement, MDC Minimal Detectable Change, SD_ch Standard deviation of the difference score (score at Visit 5 – score at Visit 1)
^aMissing scores were imputed by the mean-score, and missing anchors were imputed on the base of a weighted-probability
^bMissing scores were imputed using personal mean matching, anchor was imputed using a polytomous regression, both using clinical and demographic variables, and GH scores
^cMissing scores were imputed using personal mean matching, anchor was imputed using a polytomous regression, both using clinical and demographic variables, and all SF-36 scores
^dThe same MICE methods were applied, using only available information

ISSN: 1477-7525