This was a prospective observational cohort study.

A Dutch Emergency Medical Service (EMS), referring to three university-affiliated hospitals, dispatched ambulances that were staffed by certified EMS nurses with two years of postgraduate training in a critical care setting (ICU, cardiac care unit, anaesthesiology or ED) and one year of training in EMS-specific procedures.

A convenience sample of patients were enrolled who required urgent ambulance dispatching and had a systolic blood pressure below 100 mmHg, respiratory rate of less than 10 or more than 29 breaths/minute or a Glasgow Coma Scale (GCS) of less than 14 on arrival of the ambulance. Exclusion criteria were the unavailability of a first lactate measurement or epileptic seizures, in which case hyperlactataemia is prognostically less sensitive 15. The study was approved by the Medical Ethics Committee, which waived the need for obtaining informed consent.

Pre-hospital treatment was provided by EMS nurses according to Dutch national ambulance protocols (Landelijk Protocol Ambulancezorg (LPA)). These protocols are in accordance with the pre-hospital and advanced trauma life support guidelines of the National Association of Emergency Medical Technicians, based on the Advanced Trauma Life Support standard of the American College of Surgeons. During the study period from June 1997 to November 1998, LPA version 4 (1996 to 1999) was used. When compared with the current version 7 (2007 to 2010), most protocols were similar.

The first lactate measurement (T1) was performed by EMS nurses as soon as possible after arrival at the scene (before any pre-hospital treatment); the second measurement (T2) was obtained just before or on arrival at the ED (after pre-hospital treatment). The lactate level was measured in venous or capillary blood immediately after blood was drawn (at T1 or T2) using a point-of-care hand-held lactate analyser (Accutrend, Roche Diagnostics, Mannheim, Germany). This is a small, battery-powered, reflectance photometer with a turnaround time of 60 seconds that uses chemistry test strips on which a drop of blood is applied. Hospital physicians were not informed about the lactate levels collected by the EMS nurses. Other obtained data at both T1 and T2 included heart rate, diastolic and systolic blood pressures, peripheral oxygen saturation obtained by pulse oxymeter (SpO₂) and GCS. SpO₂ was regarded as a binary variable, which was defined as abnormal if it was lower than 92% or if the pulse oxymeter signal could not be retrieved because of inadequate peripheral perfusion (n = 25). If heart rate and blood pressure readings could not be obtained because of cardiac arrest at T1 (asystole or ventricular fibrillation, n = 11), we considered these values as 0 (this was only done at T1, not at T2).

The primary outcome measured was in-hospital mortality.

Because lactate levels were not normally distributed, they were logarithmically transformed before analysis. To evaluate the prognostic accuracy of the lactate levels, receiver operating characteristic (ROC) curves for in-hospital mortality were constructed and area under the ROC curves (AUROC) were calculated. Using ROC-curve analysis, we defined appropriate cut-off values (which are not available for the pre-hospital setting) and calculated sensitivity, specificity, positive predictive values (PPV) and negative predictive values (NPV). In order to identify patients who were likely to die, the test had to be sensitive while remaining specific 16 and had to have an acceptable PPV 17. Mortality rates of patients with high or low lactate levels were compared using a chi squared test or Fisher's exact test if necessary, based on sample size.

In order to identify independent predictors of in-hospital death, adjusted for standard variables available in the pre-hospital setting, a multivariable Cox proportional hazards (PH) model was constructed. The variables systolic blood pressure, heart rate, GCS and the change in lactate level from T1 to T2 were simultaneously entered in this model (the number of variables was restricted to four to reduce the possibility of overfitting). A backward elimination method was used, in which each step removed the variable with the highest p-value above 0.10 according to the likelihood ratio test. Interaction between all variables was not tested because of the risk of overfitting. The PH assumption was confirmed by entering variable-by-time interaction terms one by one, with time on the log scale. By choosing Cox PH instead of logistic regression analysis, we took account of the time of death, rather than just dead (yes or no) in the analysis. Statistical analyses were performed using SPSS version 11.0.1/12.0.1 (SPSS, Inc., Chicago, IL, USA).

We enrolled 135 patients. Three patients were excluded because of a missing first lactate measurement and eight patients had epileptic seizures. The baseline characteristics of the remaining 124 patients are described in Table 1. The mean (standard deviation) time at the scene from arrival to departure of the ambulance was 16 (8) minutes. Mean duration of the subsequent transfer to the ED was 11 (6) minutes. The total time from arrival at the scene to arrival in the ED was 27 (10) minutes.

Of the 124 patients who were included in the study on arrival of the ambulance at the scene, 92 survived and 32 died. Compared with the survivors, the non-survivors had a lower systolic blood pressure, lower GCS, more often an abnormal SpO₂ and an older age (Table 2). Heart rates were not significantly different. Lactate levels were higher in the non-survivors (Figure 1).

At T1, the AUROC of lactate for in-hospital death was 0.69 (95% confidence interval (CI) = 0.58 to 0.80, p = 0.001). We established a lactate level of 3.5 mmol/L as the best cut-off point for T1. A lactate level of 3.5 mmol/L or more was 75% sensitive (95% CI = 60 to 90%) and 63% specific (95% CI = 53 to 73%) for prediction of death, with a PPV of 41% (95% CI = 29 to 54%) and a NPV of 88% (95% CI = 80 to 96%). Mortality in patients with a high lactate level was 41% (95% CI = 29 to 54%), compared with 12% (95% CI = 4 to 20%) for those with a lower level (Figure 2). Patients with high lactate levels also had lower systolic blood pressures (100 vs 137 mmHg, p < 0.001), lower GCS (10 vs 14, p < 0.001), more often an abnormal SpO₂ (74 vs 21%, p < 0.001) and were more often admitted to the ICU (57 vs 36%, p = 0.022).

At T1, 33 patients had a systolic blood pressure below 100 mmHg. To adjust for the presence of a systolic blood pressure below 100 mmHg 18, a stratified analysis was performed, which showed that lactate was still significantly associated with mortality (Figure 3).

Follow-up lactate measurements were available for 106 patients. Of these patients, 78 survived and 28 died in the hospital. Compared with the survivors, the non-survivors had a lower GCS (9 vs 13, p < 0.001) and a higher lactate level (Figure 2). Systolic blood pressure, heart rate and SpO₂ did not significantly differ between the two groups.

At T2, the AUROC was 0.72 (95% CI = 0.60 to 0.84, p = 0.001). Here, 3.5 mmol/L was again considered as the most appropriate cut-off point with a sensitivity for death of 64% (95% CI = 47 to 82%), specificity of 74% (95% CI = 65 to 84%), PPV of 47% (95% CI = 31 to 63%) and a NPV of 85% (95% CI = 77 to 94%). In the high lactate group, 47% (95% CI = 31 to 63%) of the patients died, while only 15% (95% CI = 6 to 23%) of those with a lower lactate level died (Figure 2). Additionally, patients in the high lactate group had a lower systolic blood pressure (125 vs 140 mmHg, p = 0.017), lower GCS (10 vs 13, p = 0.002), more often an abnormal SpO₂ (50 vs 22%, p = 0.003) and were more often admitted to the ICU (71 vs 35%, p < 0.001).

Eleven patients had a systolic blood pressure below 100 mmHg at T2. In the other patients with a systolic blood pressure of 100 mmHg or above (n = 95), mortality rates remained significantly higher in those with high (47%, 14 out of 30) versus low lactate levels (15%, 10 out of 65, p = 0.001).

When examining the evolution of lactate during the pre-hospital phase, the lactate level, on average, increased 0.1 mmol/L in non-survivors, whereas in survivors it decreased 0.6 mmol/L (p = 0.044). This evolution of lactate from T1 to T2 had prognostic significance even after the effect of the other parameters (systolic blood pressure, heart rate and GCS) had been taken into account in the multivariable Cox PH model. Of the variables, only the change in lactate level and the GCS were independently associated with in-hospital mortality (Table 3). The hazard of death decreased by 80% (95% CI = 24 to 95%) for every 63% decrease of the lactate level at T2 relative to the level at T1 (i.e. a larger decrease in lactate during pre-hospital treatment was associated with decreased mortality). Although a model with six instead of four entered variables is a possible overfit, adding age and SpO₂ to the start model resulted in a final model in which delta lactate remained independently associated with in-hospital mortality (with equal hazard ratio, 95% CI and p value, data not shown).

To test the hypothesis that blood lactate levels remained predictive for outcome in a population that is not obviously in circulatory shock, we repeated the analyses in the subgroup of patients without cardiac arrest. In addition, this would correct for possible negation of the association of tachycardia with mortality because of the coding of heart rate as 0 in cases of cardiac arrest.

Twelve patients had cardiac arrest at T1. Four patients died out-of-hospital (before T2). Of the eight patients with return of spontaneous circulation at T2, four died during hospital admission and four survived. In the subgroup excluding the 12 patients with cardiac arrest (n = 112, in-hospital mortality 21%), lactate level remained a prognostic marker for in-hospital death. The AUROC was 0.66 (95% CI = 0.52 to 0.80, p = 0.015) at T1 and 0.69 (95% CI = 0.55 to 0.82, p = 0.007) at T2. A lactate level of 3.5 mmol/L remained the most appropriate cut-off value at both time points. Using this value at T1, mortality was 35% (95% CI = 21 to 49%) in the group with high lactate levels compared with 12% (95% CI = 4 to 20%) in the group with low lactate levels (p = 0.005). At T2, this was 43% (95% CI = 26 to 61%) compared with 15% (95% CI = 11 to 19%)(p = 0.002). In the final model of multivariable Cox PH analysis performed in the non-cardiac arrest patients, the effect of the change in lactate levels from T1 to T2 remained equally strong with a hazard ratio of 0.22 (95% CI = 0.04 to 1.11), but it was not statistically significant (p = 0.067).