From: Identifying clinical course patterns in SMS data using cluster analysis
SMS data in original format (all SMS time points used for clustering) | SMS data transformed into regression coefficients by spline analysis | |
---|---|---|
Simple and intuitive | ✓ | |
Copes with data that do not show a time trend | ✓ | |
Copes with data from clinical course patterns that are fluctuating | ✓ | |
Copes with clinical course data that are all zero values or the same value at all time points | ✓ | |
Preserves all the original information in the data | ✓ | |
With imputation of missing data, all cases can be included, regardless of clinical course patterns | ✓ | |
Copes better with missing data | ✓ | |
A data reduction technique (reduces the likelihood of overfitting the data) | ✓ | |
Reduces the collinearity (autocorrelation) of the data | ✓ | |
Requires pre-hoc assumptions about which spline characteristics are clinically important. This may improve interpretability but also may introduce bias, and require the exclusion of cases that do not meet those assumptions | ✓ |