(a) Scores plot and (b) loadings line plot for the predictive component ([1]) of the OPLS model created using 24 hours postoperative spectral data normalised to the preoperative data regressed against the AT values of the patients (1 predictive component and 3 orthogonal components, , , and ). OPLS is an extension of PLS(-DA) where the PLS components are rotated so as to remove any noise in the dataset, producing predictive components, which only account for the variation in the dataset that is related to the response variable. Any variation not related to the response variable is placed into orthogonal components (). This makes interpretation of the scores plots easier, since the influence of noise is removed. Black triangles represent patients with AT < 11 mL min⁻¹ kg⁻¹¹ (for display only, the model was not built to discriminate according to this cut-off value of AT), and red squares represent patients with AT > 11 mL min⁻¹ kg⁻¹¹. The AT values of the samples are displayed on the plot. The main solid black ellipse surrounding the majority of the scores in each plot is Hotelling’s T² elliptical tolerance region, which indicates the 95% confidence limits. A clear trend can be seen between the plasma metabolic profile and AT score at 24 hours, with AT values increasing from the left hand side of the plot to the right in a continuous fashion. (b) The loadings plot reveals the chemical shift regions responsible for this trend, and as previously shown, lipids, acetate, and ketones are influential.