Introduction

Glucose is one of the most commonly requested analytes in clinical chemistry laboratories. Accurate and reliable measurement of glucose concentration is essential for the diagnosis and monitoring of diabetes mellitus, gestational diabetes, impaired glucose tolerance, and impaired fasting glucose [1]. The most significant preanalytical error in glucose analysis is the in vitro glycolytic degradation of glucose by blood cells, particularly erythrocytes. This can lead to falsely low glucose values and, consequently, to the underdiagnosis of a diabetic metabolic state. The rate of glycolysis is influenced by several factors, including how quickly plasma is separated from blood cells, ambient temperature, white blood cell count, and glucose concentration [2–4]. Several strategies have been proposed to overcome this critical preanalytical problem. Glycolysis can be inhibited by different additives, such as sodium fluoride, typically used in combination with the anticoagulants EDTA (NaF/EDTA) or heparin (NaF/heparin), or more effectively by acidifying the blood with a citrate buffer together with sodium fluoride and EDTA (citrate/fluoride/EDTA). Because the inhibition of the late glycolytic enzyme enolase by sodium fluoride (NaF) is insufficient to prevent glycolysis in the first few hours after sample collection [5], the recommended additive is citrate/fluoride/EDTA. This combination not only inhibits enolase but also immediately blocks the early glycolytic enzyme hexokinase, thereby effectively preventing glycolysis [2,6,7]. In the absence of glycolysis-inhibiting additives, glucose degradation can be minimized by cooling the samples and performing rapid centrifugation to remove the cells within 15–30 minutes [8,9]. Although clinical guidelines do not recommend using serum or non-acidified NaF-containing plasma for diagnosing diabetes mellitus [2,10], these sample types remain the most commonly used for diabetes screening in the outpatient setting in Germany. This is primarily due to cost and practicality, as NaF/EDTA or NaF/heparin tubes are still less expensive than the recommended citrate/NaF/EDTA tubes. An additional advantage of using serum is that, apart from glucose, numerous other laboratory parameters can be measured from a single blood sample, saving time and costs while also reducing the burden on the patient. Regarding the discrepancies in glucose levels between plasma and serum, some studies have reported that plasma glucose is higher than serum glucose, while others have found no significant difference [11]. However, most studies comparing glucose levels in plasma and serum samples have been conducted under strictly controlled laboratory conditions that do not reflect the real-world settings of routine medical practice and clinical laboratories.

Therefore, the aim of the present study was to investigate the extent to which glucose concentrations differ between plasma and serum samples collected simultaneously from outpatients under real-world conditions, and to assess whether any such differences could influence the diagnosis of diabetes mellitus. This study was not designed to investigate the impact of pre-analytical factors on differences between plasma and serum glucose concentrations under controlled laboratory conditions. Accordingly, the present study does not challenge current guideline recommendations but rather evaluates the clinical impact of routinely used sample types under real-world outpatient conditions.

Materials and methods

Results

The mean age was 58.2 ± 15.9 years (mean ± SD) for males and 58.9 ± 16.4 years for females.

Plasma and serum glucose concentrations exhibited a right-skewed distribution, with the plasma and serum curves nearly overlapping in the upper range above 115 mg/dL (Fig 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Density plot showing the distribution of glucose concentrations in 27,328 paired plasma and serum samples.

https://doi.org/10.1371/journal.pone.0344562.g001

Conversely, a leftward shift of the serum curve relative to the plasma curve is evident at lower glucose concentrations, indicating a higher frequency of low glucose values in serum compared to plasma samples. Due to the skewed distribution of glucose values, the arithmetic mean was higher than the corresponding median by 8.2 mg/dL in plasma and by 6.7 mg/dL in serum (see row M + F in Table 1). The median glucose concentration was significantly higher in males than in females by 4.0 mg/dL in plasma and 3.2 mg/dL in serum (Table 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Statistical measures of glucose concentrations (mg/dL) in paired plasma and serum samples.

https://doi.org/10.1371/journal.pone.0344562.t001

The median plasma glucose concentration was 3.4 mg/dL higher than the median serum glucose in men and 2.6 mg/dL higher in women, resulting in a sex-related difference in the plasma–serum gap of 0.8 mg/dL. Given the small difference, data from both sexes were combined for further comparisons of plasma and serum glucose in paired samples.

As shown in the scatter plot (Fig 2), there was a strong positive correlation between plasma and serum glucose concentrations.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Scatter plot showing glucose concentrations in simultaneously collected plasma and serum samples (n = 27,328).

The gray line represents the line of identity (y = x). Pearson correlation coefficient = 0.945 [95% CI: 0.943 to 0.946]; Lin’s concordance correlation coefficient = 0.931 [95% CI: 0.929 to 0.932].

https://doi.org/10.1371/journal.pone.0344562.g002

The scatter plot points were nearly symmetrically distributed around the identity line at higher glucose concentrations (>125 mg/dL). However, in the lower concentration range, the distribution was asymmetric, with serum values more frequently falling below plasma values. This slight systematic bias was reflected in a lower Lin’s concordance correlation coefficient [15] of 0.931 (95% CI: 0.929–0.932) relative to Pearson’s correlation coefficient of 0.945 (95% CI: 0.943–0.946), indicating moderate agreement (0.90–0.95) between the two measurements according to McBride [16]. The asymmetry in the deviation from the identity line is more clearly illustrated in the difference plot (Fig 3), where the absolute difference between serum and plasma glucose concentrations is plotted against the mean of the two measurements. The histogram in Fig 3 demonstrated a skewed distribution of the differences, making the median more appropriate than the arithmetic mean for statistical analysis. Across the entire concentration range, the median difference between serum and plasma glucose concentrations (serum minus plasma) was −3.1 mg/dL [95% CI: −3.2 to −2.9]. The 95% range of differences extended from −27.0 mg/dL [95% CI: −27.4 to −26.6] to 10.5 mg/dL [95% CI: 10.4 to 10.7]. In the lower concentration range (approximately <100 mg/dL), the locally estimated scatterplot smoothing (LOESS) regression curve showed an almost linear relationship between the serum–plasma glucose difference and the mean glucose concentration, with the difference increasing as the mean concentration decreased. Above 100 mg/dL, however, the LOESS regression curve was nearly identical to the zero line, indicating no substantial difference between serum and plasma glucose concentrations within this range.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Difference plot with histogram showing the absolute differences between serum glucose and plasma glucose concentrations plotted against their mean.

The median difference is indicated by the solid blue line, and the 95% limits of agreement are shown as dotted blue lines. The red solid line represents the LOESS regression line, with its 95% confidence band shown in gray. Median difference = −3.05 [95% CI: −3.23 to −2.89]; 95% lower limit of agreement = −26.98 [95% CI: −27.40 to −26.62]; 95% upper limit of agreement = 10.54 [95% CI: 10.38 to 10.71].

https://doi.org/10.1371/journal.pone.0344562.g003

For a quantitative analysis of the differences between plasma and serum glucose at varying glucose concentrations, the mean glucose values were grouped into quintiles, and the median and mean differences were calculated (Fig 4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Violin plot of serum–plasma differences in glucose concentration (y-axis) across quintiles of mean glucose concentration (x-axis).

The horizontal black line within each violin represents the median difference, with corresponding numerical values. The red dot denotes the arithmetic mean. Box plots embedded within the violins illustrate the distribution of differences by annual quarter (Q1, Q2, Q3, Q4) within each quintile.

https://doi.org/10.1371/journal.pone.0344562.g004

The largest median negative difference between serum and plasma glucose was −11.05 mg/dL [95% CI: −11.3 to −10.7] in the first quintile (41.3–77.8 mg/dL, n = 5,463), followed by −5.2 mg/dL [95% CI: −5.5 to −4.9] in the second quintile (77.8–84.5 mg/dL, n = 5,468) and −1.7 mg/dL [95% CI: −2.0 to −1.4] in the third quintile (84.5–91.5 mg/dL, n = 5,465). Therefore, at mean glucose concentrations below 100 mg/dL, the serum–plasma glucose difference increased by approximately 3.2 mg/dL for every 10 mg/dL decrease in the mean glucose concentration. In the fourth quintile (91.5–106.2 mg/dL, n = 5,466) and the fifth quintile (106.2–297.3 mg/dL, n = 5,466), the median differences were the smallest, measuring 0.5 mg/dL [95% CI: 0.3 to 0.8] and 1.1 mg/dL [95% CI: 0.9 to 1.4], respectively. As shown by the box plots for each annual quarter within the violin plots, the serum–plasma glucose difference varied seasonally, with the largest difference observed in the summer quarter (Q3) and smaller differences during the colder seasons (Q1, Q2, and Q4). The largest serum–plasma glucose differences were observed between Q1 and Q3 across all quintiles, with values of −5.1 mg/dL, −3.1 mg/dL, −1.6 mg/dL, −2.4 mg/dL, and −2.8 mg/dL for quintiles 1, 2, 3, 4, and 5, respectively (S1 Table).

The diagnostic cutoff for diabetes mellitus is a fasting glucose level of ≥126 mg/dL [2]. To determine whether differences in glucose concentrations between serum and plasma would lead to different diagnostic outcomes, glucose values <126 mg/dL were classified as negative (neg), and values ≥126 mg/dL as positive (pos) for diabetes mellitus, separately for serum and plasma. The results are presented in the contingency table (Table 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Contingency table showing the number and percentage of samples classified with a hypothetical diagnosis of diabetes mellitus based on glucose concentrations measured in plasma and serum. Glucose concentrations <126 mg/dL were classified as negative (neg) and concentrations ≥126 mg/dL as positive (pos).

https://doi.org/10.1371/journal.pone.0344562.t002

1. Point estimates with 95% CI
   1. Cohen‘s kappa                         0.884 [95% CI: 0.874 to 0.893]
   2. Correctly classified proportion                   97.8% [95% CI: 97.4 to 98.1]
   3. Agreement of positive results                   98.9% [95% CI: 98.8 to 99.1]
   4. Agreement of negative results                   88.2% [95% CI: 87.0 to 89.3]
   5. Plasma positive percent proportion for total serum negative results   1.4% [95% CI: 1.3 to 1.6]
   6. Serum positive percent proportion for total plasma negative results   1.1% [95% CI: 0.9 to 1.2]
   7. Proportion difference of positive results               0.3% [95% CI: 0.2 to 0.5]

The calculated Cohen’s kappa was 0.884 [95% CI: 0.874 to 0.893], which can be interpreted as almost perfect agreement [17]. Overall, 97.8% [95% CI: 97.4 to 98.1] of the classifications based on plasma and serum glucose values were consistent. The agreement of serum with plasma was 98.9% [95% CI: 98.8 to 99.1] for the positive classification and 88.2% [95% CI: 87.0 to 89.3] for the negative classification. A total of 1.1% [95% CI: 0.9 to 1.2] of negative plasma results were positive in serum, and 1.4% [95% CI: 1.3 to 1.6] of negative serum results were positive in plasma. This indicates that 0.3% more cases of diabetes mellitus [95% CI: 0.2 to 0.5] would have been diagnosed based on plasma rather than serum values.

Discussion

Under real-world conditions, we observed lower glucose concentrations in serum than in plasma, predominantly in the lower concentration range (<100 mg/dL). In the range above 100 mg/dL, differences between serum and plasma glucose concentrations were statistically significant but of minor clinical relevance. Based on the diagnostic cutoff value of 126 mg/dL for diabetes mellitus screening, glucose measurement in serum would detect 3 fewer pathological cases per 1,000 measurements than measurement in plasma. Given that a single laboratory value is not sufficient for the diagnosis of diabetes mellitus [2], this difference appears to be negligible. According to national guidelines for type 2 diabetes, the diagnosis of diabetes mellitus is confirmed if two laboratory results – such as fasting plasma glucose and HbA1c, fasting plasma glucose and occasional plasma glucose, two fasting plasma glucose measurements, or HbA1c and occasional plasma glucose – are within the pathological range [2]. If the results are inconsistent, a third laboratory test should be performed, and if necessary, an oral glucose tolerance test (oGTT) should be conducted. The guidelines recommend the use of venous peripheral plasma rather than serum for measuring glucose for diagnosing diabetes mellitus. Plasma is the preferred specimen because in vitro glycolysis as the most significant pre-analytical source of error can be effectively minimized by rapid centrifugation of the blood sample immediately after blood collection and/or by the addition of glycolysis inhibitors. In uncentrifuged specimens, the decrease in glucose concentration during the first two hours after blood collection is nearly identical in serum and non-acidified NaF plasma, amounting to approximately 4–5% per hour at room temperature [18]. Thereafter, serum glucose concentration continues to decline, while glucose levels in NaF plasma remain stable. An immediate and complete stop of in vitro glycolysis can be achieved by adding citrated NaF/EDTA, eliminating the need for centrifugation [19–22]. This is because in citrated NaF/EDTA plasma, in addition to the inhibition of the late glycolytic enzyme enolase, the initial enzyme of glycolysis, hexokinase, is also inhibited, thereby preventing glycolysis at a much earlier stage [23]. However, unlike conventional NaF tubes, citrated NaF/EDTA tubes are not yet widely used for glucose measurement in the outpatient setting in Germany. In our comparative study, only NaF tubes without the addition of citrate were used. Although all practices that submitted blood samples are equipped with centrifuges, the NaF plasma was not centrifuged on-site due to practical considerations. NaF tubes do not contain a separation gel. Therefore, reliable separation of blood cells from plasma cannot be guaranteed, particularly during transport to the laboratory. Alternatively, pipetting plasma into a new tube after centrifugation is not practical within the typical workflow of a medical practice. In contrast to the NaF tubes, the serum tubes used contain a separation gel that ensures reliable separation of blood cells from serum, even during transport. Therefore, the serum tubes were centrifuged on-site after the samples had fully coagulated. Although the practice staff were trained to centrifuge serum samples between 20 and 60 minutes after venipuncture, the actual time elapsed between sampling and centrifugation was not recorded. Because centrifugation of the NaF tubes did not occur in the physician’s practice but rather in the laboratory, with a delay of several hours, the glucose concentration could have decreased by up to 10 mg/dL due to the delayed inhibition of glycolysis by NaF [2]. Furthermore, inadequate mixing of the blood immediately after sampling may prevent sufficient contact between the blood and additives, potentially contributing to incomplete inhibition of glycolysis. These two effects may explain why, in approximately 39% of the sample pairs, the glucose concentration in the NaF plasma samples was lower (mean difference = 5.1 ± 3.4 mg/dL, n = 10,554) than in the corresponding serum samples, which had already been centrifuged on time at the medical practice. On the other hand, if the serum was not centrifuged promptly, this may have led to a significant decrease in serum glucose concentration, resulting in lower serum than plasma glucose levels in approximately 61% of the paired samples (mean difference = 10.2 ± 7.9 mg/dL, n = 16,774). The difference exhibited an asymmetric distribution that varied with the mean glucose concentration. At glucose concentrations above 100 mg/dL, there was virtually no difference between the median serum and plasma glucose levels. In contrast, within the lower concentration range (below 100 mg/dL), a linear increase in the serum–plasma glucose difference was observed, with an increase of approximately 3.2 mg/dL for every 10 mg/dL decrease in mean glucose concentration. Bonetti et al. [24] reported a similar observation, finding that the difference between serum and plasma glucose increased by approximately 2.8 mg/dL for every 10 mg/dL decrease in the plasma glucose reference concentration after a four-hour incubation of the serum tube at room temperature. The reasons for the increasing difference between serum and plasma glucose with decreasing mean glucose concentrations below 100 mg/dL remain speculative. Since glycolysis in NaF tubes ceases a few hours after blood collection, the observed increase in the serum–plasma glucose difference can only be explained by an increased rate of glucose degradation in serum at lower basal glucose concentrations. This phenomenon cannot be readily explained by the enzyme kinetics underlying cellular glycolysis.

The key glycolytic enzymes hexokinase and phosphofructokinase-1 have very low Km values (Michaelis-Menten constants) and operate in the saturation range even at low glucose concentrations, meaning they metabolize glucose at a constant, maximal rate regardless of the basal glucose concentration [25]. Furthermore, glucose transporter type 1 (GLUT1), the transporter responsible for transmembrane glucose uptake into erythrocytes, operates at near-maximal capacity within the physiological range (80–100 mg/dL) and is driven exclusively by the glucose concentration gradient. Therefore, a reduction in extracellular glucose would be expected to decrease the absolute rate of glucose transport into erythrocytes [25]. A hypothetical explanation is that cellular stress responses triggered by glucose deprivation and oxidative stress may facilitate increased glucose consumption to support energy and redox homeostasis [25].

Our observation that the median difference between serum and plasma glucose was 0.5 to 3 mg/dL higher during the summer months (quarter 3) compared to the colder seasons (quarters 1, 2, and 4) can be attributed to the temperature dependence of the glycolysis rate, which increases at higher temperatures [4].

In summary, our comparative analysis of over 27,000 samples under routine conditions revealed only minimal differences between plasma and serum glucose concentrations above 100 mg/dL. As a result, at the diagnostic cutoff of 126 mg/dL, glucose measurement in plasma would identify only 0.3% more cases of diabetes mellitus than measurement in serum. This marginal benefit should be carefully weighed against the practical advantages of measuring glucose in serum in real-world outpatient settings. Serum is a universal sample matrix suitable for the analysis of numerous laboratory parameters [26], thereby avoiding the need for additional blood collection. A single blood draw is less stressful for patients and significantly reduces costs. Furthermore, it results in less plastic waste, contributing to environmental sustainability. Other studies have also shown that the use of serum for glucose measurement presents no disadvantages compared to NaF plasma, provided the serum is centrifuged and separated from blood cells in a timely manner after collection [8,21,27–29]. However, this requires that medical practices be equipped with centrifuges and that staff are properly trained by the laboratory in centrifugation procedures, as was the case in this study. If centrifugation is not feasible, the use of NaF plasma for glucose measurement is justified, with the caveat that glucose concentrations in NaF plasma also decrease by approximately 10 mg/dL within the first few hours after blood collection. Despite the good agreement observed between serum and NaF plasma glucose measurements at concentrations above 100 mg/dL, both specimen types must be regarded as suboptimal for the diagnosis of diabetes mellitus, as in vitro glycolysis is not sufficiently inhibited.

To achieve full reliability in glucose results by eliminating the preanalytical issue of glycolysis, a complete transition to citrated NaF/EDTA tubes would be necessary [20,22,24]. However, these tubes are not currently widely used in Germany, and a transition would be both costly and logistically challenging to implement in outpatient settings.

One of the main strengths of this study was the large dataset, comprising over 27,000 glucose measurements from concurrently collected serum and NaF plasma samples, all analyzed under routine laboratory conditions using the same measurement method over a three-year period. This enabled an assessment of the clinical relevance of the two sample types for glucose measurement and the diagnosis of diabetes mellitus in real-world outpatient settings.

The limitations of the study include the lack of information on pre-analytical variables such as the exact time of blood collection, the storage temperature of samples at the physician’s office prior to transport, the interval between blood collection and serum centrifugation, the time elapsed from collection to measurement, and the accuracy of mixing NaF samples after collection. Furthermore, no information was available regarding the indication for glucose measurement, patients’ clinical diagnoses, medication use, or other relevant clinical data. Consequently, it was not possible to evaluate potential correlations between glucose measurements and clinical diagnoses.

Conclusions

This study, comprising more than 27,000 paired NaF plasma and serum samples collected over three years from outpatient practices, demonstrates that under real-world conditions, clinically relevant differences between serum and NaF plasma glucose concentrations were observed predominantly at glucose levels below 100 mg/dL. At concentrations above 100 mg/dL, the differences were minimal, such that at a diagnostic threshold of 126 mg/dL, only 0.3% more cases of diabetes mellitus would be identified based on NaF plasma compared with serum. Given the practical advantages of universally applicable serum tubes obtained from a single blood draw, this difference appears to be negligible for diabetes mellitus screening in routine outpatient settings. Serum may therefore be considered a pragmatic alternative to NaF plasma, provided that serum samples are centrifuged on-site in a timely manner using separator gel tubes.

Importantly, neither serum nor non-acidified NaF plasma provides complete and immediate inhibition of in vitro glycolysis. Accordingly, both specimen types must be regarded as suboptimal for the definitive biochemical diagnosis of diabetes mellitus when evaluated against current preanalytical standards. The present findings should therefore be interpreted as a comparison of two routinely used, but methodologically imperfect, sampling approaches under real-world outpatient conditions, rather than as an endorsement of their diagnostic adequacy.

Replacement of non-acidified NaF tubes with citrated NaF/EDTA tubes possibly alters these conclusions and would require re-evaluation.

Supporting information