Supplemental Appendix
Flow diagram to clarify participation
Details of how the 12,990 patients were selected from the ICNARC case mix programme (Fig. 1S, Supplemental Digital Content 2,
Complete case analyses
The analyses reported in the main document use the dataset that includeslactate values that resulted from multiple imputation to avoid excluding those patients without missing values for peak lactate. Here we present analyses excluding patients without data for peak lactate.
Continuous analysis
Identical analyses were conducted, and the same patterns and statistical relationships were seen. With regard to in hospital mortality for lactate <12mol/L (Fig. 2S, Supplemental Digital Content 3, and lactate <2mmol/L (Fig. 3S, Supplemental Digital Content 4, and with regard to ICU LOS (Fig. 4S, Supplemental Digital Content 5, and Hospital LOS (Fig. 5S, Supplemental Digital Content 6,
Categorical analysis
Identical analyses were conducted, and the same patterns and statistical relationships were seen (Table 1S, Supplemental Digital Content 7,
In addition to establishing that the removal of lactates values that resulted from MI from the analysis doesn’t alter the relationship between lactate and outcome, we also tested to see if the surgical status (elective versus emergency) affects these relationships. The test for interaction is non-significant using the MI values (p=0.298), and if only actual recorded lactates are used (p=0.217)
Sensitivity analyses
Sensitivity analyses were performed to evaluate if the relationship between lactate and mortality across elective and emergency surgery was consistent after excluding unplanned admissions to the critical care unit following elective surgery (defined as those where the decision to admit to critical care was taken after the induction of anaesthesia). As the main analysis, we used multiple imputation to address the potential bias and loss of precision that could result from complete case analysis.
Continuous analysis
Identical analyses were conducted, and the same patterns and statistical relationships were seen. With regard to in hospital mortality for lactate <12mol/L (Fig.6S, Supplemental Digital Content 8, and lactate <2mmol/L (Fig.7S, Supplemental Digital Content 9, and with regard to ICU LOS (Fig.8S, Supplemental Digital Content 10, and Hospital LOS (Fig.9S, Supplemental Digital Content 11,
Categorical analysis
Identical analyses were conducted, and the same patterns and statistical relationships were seen (Table 2S, Supplemental Digital Content 12, The test for interaction is non-significant (p=0.564)
Types of surgery
The top 20 most common surgical procedures for the emergency and elective cohorts are tabulated in Table 3S (Supplemental Digital Content 13,
Figure Legends
Figure 1S: Flow diagram illustrating patient selection.
Figure 2S: Plot of unadjusted (left panel) and adjusted (right panel) elective versus emergency blood lactate level interaction (within blood lactate levels are <12)using a linear scale on the x-axis for lactate and the log scale on the y-axis for odds of in hospital mortality. The vertical axis on the right is labelled on the probability scale (response scale). Notice from the unequal spacing of the tick marks that although the axis is labelled on the scale of the response (i.e., the probability scale), the effects are plotted on the scale of the linear predictor (the logit scale). Lactate expanded by restricted cubic splines using 4 knots(0.7, 1.3,2.2, 5.7). Right panel: adjusted for the ICNARC mortality score, age and reason for surgery.
Figure 3S: Plot of unadjusted and adjusted emergency and blood lactate levels interaction (within blood lactate levels <2) using a linear scale on the x-axis for lactate and the log scale on the y-axis for oddsof in hospital mortality.The vertical axis on the right is labelled on the probability scale (response scale). Notice from the unequal spacing of the tick marks that although the axis is labelled on the scale of the response (i.e., the probability scale), the effects are plotted on the scale of the linear predictor (the logit scale). Lactate was modelled as linear. Right panel: adjusted for the ICNARC mortality score, age and reason for surgery.
Figure 4S:Effect display for the interaction between lactate levels and type of surgery with length of ICU stay (logarithm-transformed). Predicted mean (on the log scale) of ICU length of stay, adjusted for confounding variables.
Figure 5S:Effect display for the interaction between lactate levels and type of surgery with length of hospital stay (logarithm-transformed). Predicted mean (on the log scale) of hospital length of stay (logarithm-transformed), adjusted for confounding variables.
Figure 6S: Plot of unadjusted (left panel) and adjusted (right panel) elective versus emergency blood lactate level interaction (within blood lactate levels are <12)using a linear scale on the x-axis for lactate and the log scale on the y-axis for odds of in hospital mortality. The vertical axis on the right is labelled on the probability scale (response scale). Notice from the unequal spacing of the tick marks that although the axis is labelled on the scale of the response (i.e., the probability scale), the effects are plotted on the scale of the linear predictor (the logit scale). Lactate expanded by restricted cubic splines using 4 knots (0.7, 1.3, 2.2, 5.7). Right panel: adjusted for the ICNARC mortality score, age and reason for surgery.
Figure 7S: Plot of unadjusted and adjusted emergency and blood lactate levels interaction (within blood lactate levels <2) using a linear scale on the x-axis for lactate and the log scale on the y-axis for odds of in hospital mortality. The vertical axis on the right is labelled on the probability scale (response scale). Notice from the unequal spacing of the tick marks that although the axis is labelled on the scale of the response (i.e., the probability scale), the effects are plotted on the scale of the linear predictor (the logit scale). Lactate was modelled as linear. Right panel: adjusted for the ICNARC mortality score, age and reason for surgery.
Figure 8S: Effect display for the interaction between lactate levels and type of surgery with length of ICU stay (logarithm-transformed). Predicted mean (on the log scale) of ICU length of stay, adjusted for confounding variables.
Figure 9S: Effect display for the interaction between lactate levels and type of surgery with length of hospital stay (logarithm-transformed). Predicted mean (on the log scale) of hospital length of stay (logarithm-transformed), adjusted for confounding variables.