This was a prospective study with sequential purposive sampling conducted at National Taiwan University Hospital and approved by the institutional Research Ethics Committee (202302126RINA). Adult patients (≥18 years) receiving MMF (CellCept®) or enteric-coated MPS (Myfortic®) for more than 3 days were eligible. Participants receiving TAC as part of their immunosuppressive regimen were treated with either immediate-release TAC (Prograf®) or prolonged-release TAC (Advagraf®), both of which were included in the study cohort. All participants provided written informed consent. To enhance clinical relevance, we enriched the cohort by design with patients for whom MPA monitoring had been recommended by clinical specialists because of their medical condition, suspected adverse effects, or immunosuppressive regimen. This approach included individuals at higher risk of drug-related complications and more accurately reflected the intended target population for VAMS-based TDM.

Blood samples were collected by trained nurses between June and October 2023. After MMF or MPS administration, whole blood was drawn from each patient; part of the sample was used to prepare VAMS specimens (20-µL Mitra® Tips, Neoteryx, Torrance, CA, USA), and the remainder was centrifuged to obtain plasma. Paired VAMS, plasma, and whole-blood samples were used for subsequent analyses similar to that of previous studies [27, 34, 35]. VAMS devices were dried in the dark at room temperature for at least 2 h, stored in double-layer zip bags with desiccants, and kept at −20 °C until analysis.

For sample preparation, VAMS tips were incubated with 200 µL deionized water and mixed at 1,000 rpm for 10 min (Geno/Grinder 2010, SPEX Sample Prep, Metuchen, NJ). Subsequently, 185 µL methanol containing internal standards (final concentrations: 200 ng/mL mycophenolate-d₃ and 40 ng/mL tacrolimus-¹³C-d₂) and 10 µL 0.1 M zinc sulfate were added, followed by an additional 10 min of mixing at 1,000 rpm. Whole-blood samples underwent the same extraction and centrifugation steps as VAMS but without the initial deionized water. Plasma samples were extracted with 150 µL methanol containing the same internal standards and processed under identical mixing and centrifugation conditions. All samples were centrifuged at 15,000 g for 10 min, and supernatants were filtered through 0.22-µm RC-4 filters (Sartorius, Göttingen, Germany) before LC–MS/MS analysis.

MPA and TAC working solutions, together with their isotope-labeled internal standards (MPA-d₃ and TAC-¹³C-d₂), were prepared in methanol by serial dilution of stock solutions and stored at −20 °C. Calibration ranges were 10–20,000 ng/mL for MPA (10, 50, 100, 200, 500, 1,000, 2,000, 10,000, and 20,000 ng/mL) and 0.5–500 ng/mL for TAC (0.5, 1.0, 2.5, 5.0, 10, 25, 50, 100, and 500 ng/mL). For method validation, calibration curves were prepared in plasma and whole-blood matrices by plotting analyte-to-internal-standard peak area ratios and fitted using a weighted (1/x) quadratic regression model, with correlation coefficients (R²) consistently >0.99.

The lower limits of quantification (LLOQ) were 10 ng/mL for MPA and 0.5 ng/mL for TAC. Low, medium, and high quality-control (QC) samples were prepared at 20, 200, and 2,000 ng/mL for MPA and 1, 10, and 100 ng/mL for TAC, respectively. Limits of detection and quantification were defined by signal-to-noise ratios >3 and >10. Intra-day precision and accuracy were assessed using triplicate QC samples at each level for all matrices, including VAMS. Accuracy was required to be within 85–115% (80–120% for LLOQ), and precision criteria were coefficients of variation (CV) ≤15% for QC levels and ≤20% for the LLOQ. Recoveries were determined by comparing peak area ratios of analytes spiked into blank plasma or whole blood before extraction (pre-spiked) with those spiked after extraction (post-spiked). Matrix effects were evaluated by comparing peak area ratios of post-spiked plasma with those of aqueous standards. Values close to 100% indicated negligible matrix effects, and deviations within ±20% were considered acceptable [36].

Chromatographic separation was performed on an Agilent 1290 UHPLC system coupled to an Agilent 6460 triple-quadrupole mass spectrometer (Agilent Technologies, Waldbronn, Germany) using an Acquity C18 column (1.7 µm, 2.1 × 100 mm; Waters, Milford, MA, USA). The mobile phases consisted of 2 mM ammonium acetate with 0.1% formic acid in water (A) and acetonitrile/isopropanol (4:6, v/v) (B). A gradient was applied from 50% to 100% B over 3 min, held until 4.5 min, and returned to 50% B by 5.5 min at a flow rate of 0.3 mL/min. Processed samples were injected via autosampler. The ion source was operated in multiple-reaction monitoring (MRM) mode with standard operating conditions (ion spray voltage 4,000 V, sheath gas 40 psi at 325 °C, nitrogen as collision gas). MRM transitions for each analyte and internal standard are summarized in Supplementary Table S1.

Four previously described approaches for Hct-related conversion were evaluated: two derived from DBS sampling and two from VAMS sampling [29, 30, 32, 33]. For the present analysis, all equations were applied using VAMS concentrations in place of DBS concentrations. Wenkui et al. and Kromdijk et al. proposed similar DBS-based equations, with Kromdijk et al. recommending the use of a fixed mean Hct of 0.45 L/L to facilitate clinical application [30, 33]. In this study, this model was evaluated in two variations: Formula A-fixed (A-fix), applying a constant Hct = 0.45 L/L for all patients, and Formula A-individualized (A-ind), substituting each patient’s measured Hct value to allow individualized conversion. F o r m u l a   A : V A M S a n a l y t e 1 − H c t × f b p p = p l a s m a a n a l y t e

Zimmermann et al. developed a VAMS-specific linear model with Hct normalized to 0.45(Formula B) [32]. F o r m u l a   B : C P l a s m a = A · C V A M S + B · H c t 0.45 + C · H c t 0.45 · C V A M S + I n t e r c e p t

In addition, we derived a multiple linear regression model using VAMS concentration and individual Hct as predictors (Formula C). F o r m u l a   C : C P l a s m a = A · C V A M S + B · H c t + I n t e r c e p t

Finally, based on the approach described by Jager et al., we applied a DBS-derived conversion using the mean Hct of our cohort and the blood cell-to-serum partitioning constant K_BC:Serum (Formula D) [29]. F o r m u l a   D : A n a l y t e V A M S 1 − H c t + K B C : S e r u m · H c t = A n a l y t e s e r u m

Patient demographics were summarized using descriptive statistics. To evaluate whether VAMS-derived concentrations could be used interchangeably with conventional plasma measurements and to identify the most appropriate conversion strategy, a stepwise analytical framework was applied.

Systematic bias between VAMS-derived and plasma concentrations was first assessed using Passing–Bablok regression, which estimates proportional and constant bias. Conversion formulas with regression slopes and intercepts whose 95% confidence intervals included 1 and 0, respectively, were considered to exhibit minimal systematic bias. In parallel, median percentage predictive error (MPPE) was calculated to quantify overall bias of back-calculated plasma concentrations. Dispersion of prediction errors was evaluated using median absolute percentage error (MAPE), with values <15% considered acceptable, providing a complementary measure of the magnitude of deviation between predicted and observed concentrations. Analytical agreement was subsequently assessed using Bland–Altman analysis. Predefined agreement criteria required a mean relative bias within ±20%, and the proportion of samples falling within these limits was calculated to summarize overall agreement [37]. To further differentiate the performance among candidate conversion formulas, a stricter agreement threshold of ±10% relative difference was applied in an exploratory analysis. To identify factors associated with failure to meet this stricter criterion, logistic regression analysis was performed. When a significant association was identified, receiver operating characteristic (ROC) curve analysis was conducted to assess discriminative ability and to determine an optimal threshold using Youden’s index. Clinical agreement was evaluated separately using predefined thresholds: an absolute difference of ≤100 ng/mL for concentrations <1,000 ng/mL and a relative difference of ≤10% for concentrations ≥1,000 ng/mL. The final Hct conversion method was selected based on the combined evidence from bias assessment, dispersion metrics, analytical agreement, clinical agreement, and practical applicability.

To explore the clinical implications, MPA AUC was calculated using the trapezoidal rule under several sampling scenarios: inpatient VAMS with intensive sampling (pre-dose, 0.5, 1, 2, 4, 6, 8, and 12 h), inpatient sampling at C₀, C₂, and C₁₂, and outpatient sampling at C₀, C₂, and C₆, thereby illustrating how multi-point VAMS sampling may support accurate AUC estimation for evaluating adverse effects. All statistical analyses were performed using RStudio 4.4.2 (Boston, MA) and Python 3.10.

With the optimized VAMS sample preparation and LC-MS/MS setting, MPA and TAC were eluted at 1.24 min and 2.84 min, respectively, with proper separation between the analytes. The optimized MRM chromatogram of both analytes is provided in Supplementary Figure S1. Method validation showed good linearity, with ranges of 10–20,000 ng/mL for MPA and 0.5–500 ng/mL for TAC, and both calibration curves yielding R² > 0.99 (Supplementary Table S2). Intra-day accuracy and precision were assessed using four concentration levels of quality controls. VAMS MPA showed accuracy of 83.36–99.16% and precision of 1.60–4.48%, while TAC showed accuracy of 85.61–89.37% and precision of 2.07–7.68% (Table 1). Extraction recovery was relatively consistent across QC concentrations, ranging from 79.92 to 91.02% for MPA and 98.85–114.11% for TAC. The matrix effect ranged from 94.98% to 111.52% for MPA and 95.30%–104.99% for TAC after internal standard correction, indicating that MPA-d₃ effectively corrected for matrix effects in VAMS samples. Taken together, these validation results support the analytical reliability of the established VAMS method.

Twenty-one paired samples were collected: VAMS–plasma for MPA and VAMS–whole-blood for TAC. All samples were quantified using validated methods (Supplementary Tables S2, S3), and corresponding clinical characteristics were provided in Supplementary Table S4. For MPA, four previously described Hct correction approaches (Formulas A–D, see Methods) were compared, including both fixed and individualized variations of Formula A. Each conversion model was evaluated using Passing–Bablok regression, Bland–Altman plots, and predictive performance metrics.

Distinct performance patterns were observed (Table 2). Based on Passing–Bablok regression (Figure 1) and MPPE values, Formula A-ind, Formula B, and Formula D showed the smallest proportional and constant bias, whereas Formula A-fix and Formula C exhibited marked deviation. These three models also yielded lower MAPE values, supporting their superior analytical consistency. Using predefined agreement limits of ±20% in Bland–Altman plots, more than 80% of samples for Formula A-ind, B, and D met the acceptance criteria, indicating acceptable analytical agreement. To further differentiate the performance among these three formulas, a stricter agreement criterion of ±10% was applied. Under this definition, the proportions of samples meeting the criterion were substantially lower, ranging from 57% to 71%. Greater deviations were predominantly observed at lower concentration ranges (Figure 2).

To evaluate whether concentrations influenced conversion success under the ±10% criterion, logistic regression analysis was performed, with conversion success as the dependent variable and log-transformed plasma MPA concentration as the predictor. Plasma concentration was identified as a significant determinant (odds ratio 4.41; p < 0.05), indicating that higher concentrations were substantially more likely to meet the analytical agreement. This concentration-dependent pattern was further supported by a supplemental ROC analysis (AUC = 0.73; sensitivity 0.84; specificity 0.76; p = 0.0008), which identified 488.23 ng/mL as the optimal threshold (Supplementary Figure S2).

When evaluated against clinical thresholds (Table 2), Formula B achieved the highest agreement (90%), followed by A-ind and D (86%). Despite Formula B showing slightly higher numerical agreement, Formula A-ind exhibited smaller bias dispersion (Figure 2) and required simpler calculations, suggesting better practical applicability. Overall, A-ind was selected as the most suitable method for subsequent analyses, balancing analytical robustness with clinical feasibility.

For TAC, Passing–Bablok and Bland–Altman analyses demonstrated good agreement between VAMS and whole-blood concentrations, with a regression slope close to unity and a modest negative bias (mean percent difference = −18.4%, Figure 2F). This bias remained within the predefined acceptance limits, indicating that TAC can be reliably measured by VAMS without requiring Hct correction.

To illustrate clinical applicability, the validated VAMS method was applied to two lung-transplant recipients receiving MMF + TAC who developed leukopenia. Using eight-point VAMS sampling, plasma MPA concentrations were estimated via the A-ind conversion, enabling full AUC profiling (Figure 3; Table 3). The complete eight-point sampling produced the highest AUCs (28.9 and 23.1 mg·h/L), whereas inpatient (C₀, C₂, C₁₂) and outpatient (C₀, C₂, C₆) schemes progressively underestimated exposure (−16 to −35%). This underestimation highlights the impact of sampling strategy on exposure assessment. TAC troughs from VAMS were consistent with routine whole-blood values. These findings demonstrate that VAMS facilitates minimally invasive assessment of full AUC through multi-time-point sampling and provides a practical platform for the subsequent development and evaluation of LSS.

Despite these promising results, several limitations should be acknowledged. First, our study had a relatively small sample size (n = 21), which may restrict the generalizability of the findings. Second, the patient population primarily consisted of kidney and lung transplant recipients, limiting extrapolation to other transplant populations who might also benefit from VAMS-based MPA monitoring. Third, all VAMS sampling and handling were performed by healthcare professionals, which may not fully reflect real-world scenarios where patients collect samples themselves. An additional limitation is that VAMS specimens were generated from venous whole blood rather than finger-stick capillary samples, which does not fully replicate real-world sampling conditions; however, published data suggest negligible capillary–venous bias for MPA and other immunosuppressants, indicating limited impact on clinical applicability [27, 34]. Finally, the overall Hct values of our patients were significantly lower than those of the general population, raising the need to further evaluate the applicability of these findings in individuals with normal Hct ranges. Nevertheless, because the VAMS samples in this study were collected from real transplant patients, this phenomenon further highlights the importance of carefully considering Hct effects, particularly in patients with low Hct values, and of selecting an appropriate concentration conversion strategy for this patient population.

In summary, this study demonstrates that VAMS sampling, combined with appropriate Hct-adjusted conversion, can provide reasonable and clinically interpretable MPA and TAC measurements. The method shows strong potential for conditional integration into routine transplant TDM. Future work should evaluate home-based self-sampling, validate LSS derived from VAMS, and explore broader applicability across diverse transplant populations.