This study evaluates the precision and accuracy of four pairs of handheld lactate analyzers: Lactate Plus (Nova Biomedical, USA), Lactate Pro2 (Arkray KDK, Japan), Lactate Scout 4 (SensLab GmbH, Germany), and TaiDoc TD-4289 (Taidoc Technology Corporation, Taiwan). The handheld analyzers are compared against two stationary reference analyzers: Biosen C-Line (EKF Diagnostic GmbH, Germany) and YSI Sport 1500 (Yellow Springs Instruments, USA). For simplicity, throughout the text, the analyzers are referred to as Plus, Pro, Scout, Tai, Biosen, and Ysi, respectively.

The considered analyzers differ in their methods of expressing blood lactate concentration. Lactate molecules in blood are found both in plasma and inside erythrocytes (red blood cells; RBC). All analyzers quantify blood lactate as a molar concentration, with units \(\cdot \hbox ^} = \cdot \hbox ^} = \hbox \) (millimolar). However, Biosen, Plus, Pro, and Scout hemolyze RBC and measure the total amount of lactate in plasma and RBC, expressing this relative to the volume of whole blood. In contrast, Ysi measures lactate only in plasma, expressing it relative to the volume of whole blood, whereas Tai measures only plasma lactate but expresses it relative to plasma volume. Importantly, at rest and during submaximal exercise, the lactate concentration in plasma is approximately 50% higher than in the erythrocytes (Foxdal et al. 1990). Consequently, compared to a given blood lactate concentration measured with Biosen, Ysi is expected to report a lower value, while Tai is expected to report a higher one. Due to its more common measurement technique, Biosen is used as the main reference in this study. Comparisons using Ysi as the reference are available in an external repository (Mentzoni 2024).

All lactate analyzers evaluated in this study measure the blood l-lactate concentration, [\(\hbox ^-\)], using an amperometric biosensor. The principle of this method is as follows: Utilizing an immobilized enzyme located either in the test strips (handheld analyzers) or between two membranes (stationary analyzers), lactate molecules in the blood sample are converted in an enzymatic reaction. All analyzers used in this study rely on lactate oxidase as the catalyzing enzyme, which oxidizes lactate to form pyruvate and hydrogen peroxide. Hydrogen peroxide is in turn oxidized, a process which releases electrons and creates an electrical current. The magnitude of this electrical current is directly proportional to the lactate concentration. Consequently, measuring the electrical current yields an estimate of the blood lactate concentration. More extensive reviews of this and other methods for measuring the blood lactate concentration are provided by others, cf. e.g., Nikolaus and Strehlitz (2008), Rathee et al. (2016).

The considered analyzers differ in blood sample size and measurement time. The volume of blood analyzed is 0.7, 0.3, 0.2, 0.8, 20, and 25\(}\), while the measurement time is 13, 15, 10, 5, 20–45, and 30 s for Plus, Pro, Scout, Tai, Biosen, and Ysi, respectively.

The two stationary analyzers were calibrated according to their instructions. The Biosen analyzer was calibrated twice an hour against a low (3 mM) and a high (14 mM) solution. Additionally, the Biosen analyzer calibrated itself automatically every hour against a solution of 12 mM. The Ysi analyzer was calibrated against a 5 mM solution right before starting the measurement procedure. Additionally, a 15 mM solution was used to validate the linearity of the analyzer.

A well-trained 35 year old male cyclist rode on a stationary trainer at four intensities corresponding to ratings of perceived exertion (RPE on the CR10 scale) of 1/10, 3/10, 5/10, and 6/10. For each intensity, after cycling at the set intensity for approximately 5 min, a venous blood sample was collected using a canula placed on the inside of the elbow. The sample was drawn into a vacutainer tube coated with lithium heparin. The mean power outputs in the last minute before taking the blood samples were 142 W (2.0 W \(\hbox ^\)), 224 W (3.2 W \(\hbox ^\)), 281 W (4.0W \(\hbox ^\)), and 307 W (4.4 W \(\hbox ^\)). The corresponding heart rate means were 120 bpm (63% of maximum heart rate), 149 bpm (79%), 156 bpm (83%), and 163 bpm (86%). These intensities were chosen as they cover a range of exercise intensities where measurements of blood lactate concentration are relevant (Casado et al. 2023).

Within 10 min of taking the blood sample, which was kept at room temperature (\(\approx \,^}\)) the whole time, the first measurement series of that intensity was conducted. A measurement series started by measuring the blood lactate concentration with the two stationary analyzers (Ysi first, then Biosen), referred to as pre-measurements. Blood from the sample was then measured “simultaneously” with four pairs of handheld lactate analyzers (in the order: Pro1, Plus1, Scout1, Tai1, Pro2, Plus2, Scout2, Tai2). Then, the blood lactate concentration was measured again with the two stationary analyzers (Ysi first, then Biosen), referred to as post-measurements. The mean of the pre- and post-measurements was used as reference.

For each of the four blood samples (i.e., intensities), 10 measurement series were conducted. The time between two consecutive measurement series was 4–10 min. The blood lactate concentrations (measured with Biosen) in the four series were 0.88–2.04 mM, 1.49–2.51 mM, 2.53–3.37 mM, and 3.85 to 4.89  mM. The rate of increase in blood lactate concentration with time was consistent with previous studies (Calatayud and TenÍas 2003; Zavorsky et al. 2021; Geyssant et al. 1985). Overall, this method yielded a range of blood lactate measurements from 0.88 to 4.89 mM with a median difference between consecutive measurements of 0.10 mM (measured with Biosen). A table containing all 480 measurements (four intensities, 10 measurement series, pre and post with the two stationary analyzers, and two units of each of the four handheld analyzers; \(4 \times 10 \times 12\)) is provided in an external repository (Mentzoni 2024).

To ensure consistency and an effective workflow, a single individual did all measurements with the handheld analyzers, and another did all measurements with the stationary analyzers. Both individuals are highly skilled with extensive experience in conducting lactate measurements. A video illustrating the procedure is provided in an external repository (Mentzoni 2024).

Python 3.11 was used for post-processing the measurements that were written into an Excel sheet during the measurement procedure. The pandas, numpy, statsmodels, scipy and pingouin packages were used for numerical and statistical analyses of the measurements, whereas matplotlib and seaborn were used for generating figures. The raw measurements and python scripts are provided in an external repository (Mentzoni 2024).

Residuals are calculated as the difference between the actual measured values, \(y_i\), and the predicted values, \(\hat_i\), using linear least-squares regression for two sets of measurements (e.g., Plus vs. Biosen). The residual standard error (RSE) is the square root of the mean squared error (MSE) which is the sum of the residuals squared divided by the degrees of freedom,

$$\begin \textrm = \sqrt} = \sqrt^ (y_i - \hat_i)^2}}, \end$$

(1)

with n being the number of measurements.

Bland–Altman comparisons are performed for the two units within each handheld analyzer brand to assess the within-brand agreement. The within-brand results are presented in a table in the results section; within-brand Bland–Altman plots, as well as Bland–Altman plots for each analyzer brand against the Biosen analyzer, are made available in an external repository (Mentzoni 2024). The upper and lower limits of agreement (LoA) are calculated based on the t-score,

$$\begin \text = \overline \pm t_} \times \textrm, \end$$

(2)

here \(\overline\) is the mean of the differences between the two units and SD is the standard deviation of the differences. Setting the confidence level to \(\%\) and number of degrees of freedom to 39 (\(n-1\) with \(n=40\) measurements on each analyzer unit) yield

$$\begin \text = \overline \pm t_} \times \textrm = \overline \pm 2.02 \ \textrm, \end$$

(3)

which is slightly larger than if using the z-score (1.96 SD).