Improved statistical analysis of pre- and post-treatment patient-reported outcome measures (PROMs): the applicability of piecewise linear regression splines

Abstract

Piecewise linear regression is a straightforward approach to analyse baseline and follow-up HRQoL measures such as the EQ-5D index. It can provide a reasonable approximation of the shape of the underlying relationship where the threshold and slopes prove informative and meaningful.

Impact of prenatal screening and diagnostic testing on trends in Down syndrome births and terminations in Western Australia 1980 to 2013

“…Data on the distribution of maternal age among pregnancies and among those pregnancies with fetal DS are also presented. We used piecewise linear (segmented) regression 17 to compare the live-born rates of DS and the percentage of DS terminated over the different eras of prenatal diagnosis (1980-1993, 1994-2003, 2004+) allowing possibly different line segment slopes and intercepts within each of these three periods. Analysis was done in Excel (Microsoft Corporation, Redmond, WA, USA) and TIBCO Spotfire S+ version 8.2 (TIBCO Spotfire, Boston, MA, USA).…”

Can patient-reported outcomes predict re-operations after total hip replacement?

“…The outcome (reoperation) was measured from one year following the index surgery until the occurrence of re-operation, death, or December 31st 2015, whichever came first. The predictors for re-operation are PROMs collected from the PROMs program in SHAR [19]. The PROMs program in SHAR includes the EQ-5D health status questionnaire [20], a hip pain visual analogue scale (VAS) [21], and at follow-ups, satisfaction with the outcome using a VAS.…”

Development and validation of a shared decision‐making instrument for health‐related quality of life one year after total hip replacement based on quality registries data

“…Best predictive accuracy was observed for the multivariate adaptive regression splines. This method fits piecewise linear regression to the data and has the capacity to offer prediction and regression coefficients that confer better to the nature of the data than simple linear regression . Additionally, we observed that random forest, which is considered one of the best classification algorithms , performed worse than discriminant analysis.…”