Dealing with indeterminate outcomes in antimalarial drug efficacy trials: a comparison between complete case analysis, multiple imputation and inverse probability weighting

Table 6 Performance measures of complete case and maximum likelihood estimator for handling 45% missingness in recurrences for individuals treated with artemether-lumefantrine in estimating day 28 cured proportion

Full data estimate of day 28 cure proportion (SE) = 0.9637 (0.1561) ^a
	Complete Case analysis	Maximum Likelihood Estimator
Mechanism 1
Bias	1.3724 (0.0145)	0.0000 (0.0075)
Relative Bias	1.42%	-0.00%
Model based SE	0.2153 (0.0007)	0.2100 (0.0008)
Empirical SE	0.2177 (0.005)	0.2159 (0.0048)
Coverage	21.4% (1.3%)	93.1% (0.8%)
RMSE ^b	0.5503 (0.0077)	0.2169 (0.0022)
Mechanism 2a (Weak scenario)
Bias	1.4093 (0.0149)	-0.0004 (0.0077)
Relative Bias	1.46%	-0.04%
Model based SE	0.2179(0.0008)	0.2123 (0.0008)
Empirical SE	0.2227 (0.0049)	0.2185 (0.0049)
Coverage	19.2% (1.2%)	93.4% (0.8%)
RMSE ^b	0.5686 (0.0082)	0.2186 (0.0023)
Mechanism 2b (Strong scenario)
Bias	1.4437 (0.0152)	-0.0009 (0.0079)
Relative Bias	1.50%	-0.09%
Model based SE	0.2202 (0.0008)	0.2143 (0.0008)
Empirical SE	0.2248 (0.0050)	0.2204 (0.0049)
Coverage	17.2% (1.2%)	93.8% (0.8%)
RMSE ^b	0.5844 (0.0084)	0.2203 (0.0023)

RMSE Root Mean Squared Error, AL artemether-lumefantrine, AS-AQ artesunate-amodiaquine, DP dihydroartemisinin-piperaquine
^a The “true” estimates of cured proportion (total cured/total number of patients treated) before missingness was induced. Those with new infections were counted as cured. The variance for cured proportion (p) for a total number of patients (n) was calculated as p(1 − p)/n. The variance was converted to the cloglog scale using the equation presented in Additional file 1, Section C.
^b Monte Carlo error for the RMSE presented on mean squared error scale. Monte Carlo Standard Errors shown in parentheses

ISSN: 1471-2288