Skip to main content
. 2025 Feb 20;11(1):e004844. doi: 10.1136/rmdopen-2024-004844

Table 3. Summary of the main findings for each imputation method in data missing at random.

Method Performance measure* ASDAS ASDAS<2.1
Baseline 6 months 12 months Baseline 6 months 12 months
Unconditional M-SI Bias Strong over-estimation Slight under-estimation Slight under-estimation Strong under-estimation Strong over-estimation Strong over-estimation
Conditional M-SI Bias Slight under-estimation Moderate over-estimation
Deterministic LR-SI Bias Moderate under-estimation Close to unbiased Slight over-estimation Moderate under-estimation Moderate over-estimation Moderate over-estimation
Random LR-SI Bias Moderate under-estimation Close to unbiased Slight over-estimation Strong over-estimation Moderate under-estimation Moderate under-estimation
Unconditional HD-SI Bias Strong over-estimation Slight under-estimation Slight under-estimation Strong under-estimation Slight over-estimation Moderate over-estimation
Conditional HD-SI Bias
Predictive mean deterministic HD-SI Bias Slight under-estimation Close to unbiased Close to unbiased Strong over-estimation Slight over-estimation Close to unbiased
Predictive mean random HD-SI Bias Slight under-estimation Close to unbiased Close to unbiased Strong over-estimation Slight over-estimation Slight over-estimation
Weighted predictive mean random HD-SI Bias Close to unbiased Close to unbiased Slight under-estimation Strong over-estimation Close to unbiased Slight over-estimation
Random LR-MI Bias Moderate under-estimation Close to unbiased Slight over-estimation Strong over-estimation Moderate under-estimation Moderate under-estimation
Coverage Strong under-coverage Slight under-coverage Slight under-coverage Strong under-coverage Moderate under-coverage Moderate under-coverage
Unconditional HD-MI Bias Strong over-estimation Slight under-estimation Slight under-estimation Strong under-estimation Slight over-estimation Moderate over-estimation
Coverage Strong under-coverage Slight under-coverage Strong under-coverage Strong under-coverage Slight under-coverage Strong under-coverage
Conditional HD-MI Bias
Coverage
Predictive mean random HD-MI Bias Slight under-estimation Slight under-estimation Close to unbiased Strong over-estimation Slight over-estimation Slight over-estimation
Coverage Strong under-coverage Slight under-coverage Moderate under-coverage Strong under-coverage Slight under-coverage Moderate under-coverage
Weighted predictive mean random HD-MI Bias Close to unbiased Close to unbiased Slight under-estimation Moderate over-estimation Slight over-estimation Slight over-estimation
Coverage Moderate under-coverage Slight under-coverage Slight under-coverage Moderate under-coverage Slight under-coverage Moderate under-coverage

‘–’ indicates that results were not produced due to insufficient number of individuals in classes in the simulated datasets.

Assessment of bias: close to unbiased (<1%), slight bias (1%–5%), moderate bias (5%–10%) and strong bias (>10%). Assessment of coverage: correct coverage (Monte Carlo 95% CIs of coverage include 95%), slight under-coverage (90%–95%), moderate under-coverage (80%–90%), strong under-coverage (<80%) and over-coverage (95%–100%). Monte Carlo 95% CIs of coverage were calculated based on corresponding 1.96 Monte Carlo SEs.

*

Strong under-coverage was observed for all single imputation methods.

ASDASAxial Spondyloarthritis Disease Activity ScoreHD-MIhot deck multiple imputationHD-SIhot deck single imputationLR-MIlinear regression multiple imputationLR-SIlinear regression single imputationM-SImean single imputation