Skip to main content

Table 2 CEM results

From: Integrating data from randomized controlled trials and observational studies to predict the response to pregabalin in patients with painful diabetic peripheral neuropathy

Cluster

Patients by cluster, n

Global imbalancea

Reduction in global imbalancea After CEM (%)

 

Observational Study Alone

Observational Study + RCT in matched dataset After CEM, n

   
 

Before CEM

After CEM

 

Before CEM

After CEM

 

1

696

332

343

0.72

0.68

6

2

777

279

306

0.70

0.26

63

3

542

201

245

0.72

0.27

63

4

556

162

195

0.84

0.33

61

5

287

105

237

0.74

0.30

59

6

301

125

202

0.71

0.30

58

Total

3159

1204b

1528

   
  1. Abbreviations: CEM coarsened exact matching, RCT randomized controlled trial
  2. aThe degree of imbalance represents level of bias in the covariates’ distributions for a given sample. According to Iacus et al. (2008) [56], ‘the key goal of matching is to prune observations from the data so that the remaining data have better balance between the control and the treated groups’ (e.g., the observational study dataset of each cluster and RCT data). ‘Exactly balanced data [i.e., global imbalance score = 0] means that controlling further for X is unnecessary (since it is unrelated to the treatment variable), and so a simple difference in means on matched data can estimate the causal effect; approximately balanced data requires controlling for X with a model (e.g., the same model that would have been used without matching), but the only inferences necessary are those relatively close to the data, leading to less model dependence and reduced statistical bias than without matching.’ (See: 1) Imbens GW, Rubin DB. Causal inference in statistics, social, and biomedical sciences. Cambridge, UK: Cambridge University Press; 2015 [57]. 2) King G, Lucas C, Nielsen R. The balance-sample size frontier in matching methods for causal inference. Am J Poli Sci. doi:10.1111/ajps.12272 [58]. 3) Stuart EA. Matching methods for causal inference: a review and a look forward. Stat Sci. 2010;25:1–21 [59].) Therefore, in our case, an imbalance of 0 means that the empirical distribution of the covariates of the Observational Study dataset in a given cluster is equivalent to in the RCT data; an imbalance of 1 means that the empirical distribution is completely different
  3. bOne hundred sixty-two of them were excluded from the ARMAX model calibration because they lacked pain and sleep interference data for the full six weeks; hence there were 1042 observational study patients in the calibration dataset for developing the ARMAX models
Back to article page

BMC Medical Research Methodology

ISSN: 1471-2288

Contact us