Matching multiple groups of subjects by covariates.
R (and RStudio). No packages needed.
Open MatchUp.R
Enter getwd() in the console to find which folder you are in.
Put Matching.csv in this folder.
This contains some sample data shown below - in this instance with 3 disease groups.
| Unique ID | Group | Sequence number | AGE | SBP | BSA | SEX |
|---|---|---|---|---|---|---|
| 0114/0014 | 1 | 1 | 45 | 134 | 1.23 | 1 |
| 0115/0061 | 1 | 2 | 43 | 123 | 1.31 | 1 |
| 0115/0086 | 1 | 3 | 56 | 143 | 1.02 | 0 |
| 0117/0014 | 2 | 4 | 23 | 123 | 1.34 | 1 |
| 0117/0016 | 2 | 5 | 33 | 132 | 1.33 | 1 |
| 0117/0018 | 2 | 6 | 78 | 142 | 1.56 | 0 |
| 0117/0022 | 3 | 7 | 44 | 111 | 1.44 | 0 |
| 0117/0023 | 3 | 8 | 54 | 99 | 1.33 | 0 |
| 0217-/006 | 3 | 9 | 33 | 166 | 1.09 | 1 |
Run the code then in the console type: MatchUp("Matching.csv", final.group.size = ...) For this test data you can set final.group.size = 1
The output is the same CSV file with one extra column appended which denotes whether that subject should be included or not in the matched dataset.
Edit Matching.csv as necessary and change final.group.size as required. Final group size must not be larger than any of the group sizes.
Matching.csv must be (a) .csv format, (b) column names for each column, (c) column names in capitals will be used for matching, column names not entirely in capitals will not be used for matching, (d) The column name for subjects IDs must contain the letters, in capitals, "ID".
Final.group.size decides the number in each final group. Small samples may lead to non-convergence in the GLM, although the warnings for this have been switched off, so you won't know. Bigger samples will lead to better results...
-
There are three or more non-overlapping clinical groups which need to be matched on the basis of covariates
-
The initial groups may be of unequal sizes
-
You want to match the groups 1:1:1
-
There are no missing data
i. Propensity matching approach
ii. Sequential comparison of each group ('target' group) with the remaining groups ('pooled' groups)
iii. Subject in the target group with lowest propensity score is discarded, provided the number of subjects in the target group exceeds a minimum threshold