This is the first comprehensive clinical validation study of capillary microsampling for venetoclax. We demonstrated that capillary microsampling can reliably quantify venetoclax concentrations when appropriate whole blood-to-plasma conversion is applied. Without correction, venetoclax concentrations from both DBS and VAMS underestimated plasma (mean bias − 21% and − 14%, respectively). This bias was strongly linked to Hct, with higher levels causing greater underestimation of venetoclax in capillary whole blood, due to the smaller plasma fraction relative to erythrocytes. Importantly, despite both being volumetric capillary microsampling techniques, DBS (with HemaXis DB10) and VAMS (with Mitra Clamshell) differed in their Hct dependence: the venetoclax plasma/DBS concentration ratio rose steeply with increasing Hct (slope 3.52; R2 = 0.71), whereas the plasma/VAMS ratio showed a more modest Hct dependence (slope 1.51; R2 = 0.37). Consequently, Hct inflated the plasma/DBS ratio more than twice as strongly as plasma/VAMS, making DBS more susceptible to Hct-related bias. Using the observed relationship between Hct and the plasma-to-microsample ratio, we derived a linear correction model that achieved excellent agreement with venous plasma (AR 95% DBS, 91% VAMS), with Bland–Altman limits within the ± 25% predefined clinical threshold, no proportional or constant bias, and strong predictive performance (MPPE ≈ 0, MAPE < 15%). This fulfills international validation criteria for microsampling in PK monitoring [19, 40]. A key limitation, however, is that this approach remains dependent on contemporaneous Hct values, which are not routinely available in home-based self-sampling. This was illustrated by the few out-of-range cases in both DBS and VAMS, all traced to a single patient whose Hct was measured ≥ 20 days from sampling. Even with a low CV, stale values undermined correction accuracy. Although Hct variability was summarized descriptively using a 1-year CV to reflect real-world Hct fluctuations during treatment, our findings suggest that the timing of the Hct measurement relative to sampling is more critical for correction accuracy than long-term variability. Reliable correction to plasma therefore requires Hct measured closest to sampling. Although volumetric microsampling is often proposed to nullify technical Hct bias by standardizing sample volume [45], our findings demonstrate that physiological Hct bias persists across volumetric devices, showing that Hct remains a critical determinant. Encouragingly, recent advances in nondestructive Hct measurement for DBS and VAMS, using reflectance, near-infrared spectroscopy, or image analysis, now enable direct Hct determination from the microsample itself [46,47,48]. These developments may overcome one of the main barriers to routine and home-based implementation.

While our empirical Hct-ratio models (Eqs. 1a–1d) achieved excellent agreement and predictive performance, literature-based capillary blood-to-plasma correction strategies achieved only moderate-to-poor agreement, with acceptance rates typically ≤ 67% within ± 20% of plasma values and in some cases performing worse than no correction. Although venetoclax is confined to the plasma compartment [5], theory-driven Hct formulas systematically mis-corrected plasma concentrations. In DBS, dividing by (1-Hct) or applying fixed/mean Hct values consistently overestimated plasma levels, with AR values < 53% and biases exceeding 20–30%. Adding partition parameters did not improve performance. In VAMS, Hct formulas again under- or over-corrected concentrations, with only one method meeting the 67%/20% criterion, but performing suboptimally in DBS. Together, these findings demonstrate that Hct must be modeled empirically rather than through generic corrections, and consistent with prior studies, confirm that Hct alone does not explain all capillary-venous variability [25, 49].

Despite both being volumetric devices, VAMS achieved acceptable agreement and predictive performance with a broader range of literature-based strategies than DBS. In VAMS, several regression-based calibrations and conversion factors crossed the 67% AR threshold (69–80%) with low-to-moderate bias and MAPE around 10–13%. However, their 95% LoA still extended up to − 40% to 57%, exceeding the ± 25% clinical target. In contrast, no literature-based correction strategy in DBS met the 67%/20% acceptability criterion, with the best-performing approaches plateauing at 61–62% AR and MAPE 15–17%. Even when mean bias was small, variability dominated. Overall, wide 95% LoA for both devices limited the clinical acceptability of all literature-based correction strategies.

Reviewing the outcomes of literature-based correction strategies summarized in Table 1 reveals substantial heterogeneity that limits comparability across studies. Many did not meet IATDMCT recommendations of ≥ 40 paired samples spanning the full therapeutic range [19, 40], with several relying on small cohorts or restricted to trough levels, which can inflate apparent accuracy while masking error at higher concentrations, as would be relevant for peak or AUC-based TDM. Definitions of acceptability also varied, with AR thresholds ranging from 15 to 25% and reference values differing (plasma versus mean), while predictive metrics were often not reported. Methodological diversity was considerable: some studies used venous rather than capillary blood, non-volumetric punch DBS instead of volumetric devices, and applied different regression approaches (slope-only factors, ordinary least squares, Deming, Passing–Bablok). Hct handling was likewise inconsistent, often relying on fixed or mean values or population-level partition parameters instead of contemporaneous, sample-level Hct. Another often overlooked factor is the timing of paired sample collection, which should occur within 5–10 min [19] and has been reported as challenging in clinical practice [49]. Even small deviations can introduce variability in the post-dose window until Tmax or at trough depending on the drug’s half-life. In our study, the 10-min window for paired sampling was not feasible in four cases. Together, these differences may help explain why strategies developed for other plasma-confined drugs did not yield similar results for venetoclax, and more broadly confirm that performance does not generalize across drugs, devices, or cohorts.

As an alternative for plasma-confined drugs, dried plasma spots have been explored to bypass whole blood-to-plasma conversion, supported by the emergence of devices that generate dried plasma spots directly from a capillary drop. However, they often require relatively large blood volumes and show inconsistent analyte recovery, including reduced plasma protein recovery, which is a significant limitation for highly protein-bound compounds [40]. Systematic errors have also been reported when comparing dried plasma spots with liquid plasma for oncolytic drugs [50], underscoring the need for further bioanalytical and clinical validation before DPS can replace DBS or VAMS in this setting.

In addition to clinical validation, we evaluated feasibility, as clinical implementation requires methods suitable for routine care. We demonstrated that home-based self-sampling was feasible: nearly all patients collected microsamples independently, most returned samples were analytically adequate (76%), and usability ratings favored VAMS over DBS as the more patient-friendly option for decentralized venetoclax monitoring. Comparable findings have been reported in kidney transplant recipients, where both DBS and VAMS proved feasible, with a consistent end-user preference for VAMS [51]. From a laboratory perspective, however, DBS cards such as the HemaXis DB10 integrate readily into automated workflows, whereas VAMS automation remains less developed [52]. Similar feasibility has been reported in other oncology cohorts using VAMS, with successful home-sampling rates ranging from 70 to 93% across studies involving oral oncolytics [49, 53]. In contrast, self-sampling feasibility data for DBS using the HemaXis DB10 device remain limited in oncology. However, a real-world study in kidney transplant recipients reported that 91% of returned samples were suitable for analysis [51]. It is important to interpret home feasibility outcomes within the clinical context, as populations differ markedly. For example, venetoclax-treated patients with AML face a limited life expectancy and are generally more fragile [54], whereas kidney transplant recipients often achieve long-term survival [55]. Training conditions also differed between the VAMS home-sampling studies: in one study, patients relied mainly on written instructions after in-clinic demonstration [53], whereas in another, initial sampling was supervised and instructions reinforced [49]. In our study, initial sampling was also unsupervised, further suggesting that structured training could improve outcomes. Broader comparisons of microsampling devices confirm device-dependent feasibility, with success rates of 65–89% for various microsampling devices, but only 12% for the high-volume device Minicollect tube [20]. It is also important to acknowledge that remote monitoring with these devices will not be feasible for all patients. In our study, SUS scores ranged as low as 23 for DBS and 65 for VAMS, indicating variability in user acceptance. Although sample adequacy was analytically acceptable, the 24% failure rate in our study remains substantial. This may have operational implications: delayed throughput times, repeat sampling, and additional patient contacts may offset some of the logistical advantages of home-based microsampling and influence overall financial feasibility. Therefore, microsampling should not be considered a one-to-one replacement for venipuncture in all patients. Instead, structured training and careful patient selection, including identifying individuals unlikely to perform microsampling independently, will be critical for successful implementation.

Microsampling may be particularly attractive in outpatient settings such as CLL, where venetoclax treatment often occurs in the ambulatory setting and long-term monitoring is required. This reflects the more favorable prognosis and longer survival observed in CLL compared with AML [56]. In contrast, venetoclax-treated patients with AML require intensive in-hospital treatment and monitoring and are at higher risk for infection-related complications [3]. Although these patients already visit the hospital frequently, minimizing avoidable exposure remains clinically relevant [57]. In this context, home-based microsampling could enable more intensive PK profiling, such as AUC assessment, without additional clinic visits. Importantly, although this study provides clinical validation and demonstrates feasibility, the clinical utility of venetoclax microsampling remains to be studied.

This study has some limitations. First, although we included 25 participants and 64 paired samples, meeting the IATDMCT threshold of ≥ 25 patients and ≥ 40 samples when patient numbers are limited, this remains below the general recommendation of ≥ 40 participants with single paired samples [19]. Nonetheless, our sample size was higher than most previous oncolytic drug studies using microsampling (Table S1). Second, we did not validate the conversion formula in an independent dataset. Only one prior study (n = 12 patients) among those whose correction strategies we applied in this study reported validation in an independent sample set [35]. This limitation is particularly relevant for general correction factors and regression-based calibrations, which are more prone to overfitting [25]. In contrast, our primary approach was individualized, using the empirically observed Hct-plasma/microsample ratio, which may be less sensitive to this issue. Third, while patients successfully performed home-based sampling, they were not instructed to collect trough samples specifically; feasibility for clinically relevant timepoints such as troughs therefore remains to be demonstrated. Finally, patients were not randomly selected but represented venetoclax-treated individuals at our center who were clinically able and willing to participate. This introduces potential self-selection bias, as fitter or more functionally independent patients may have been overrepresented, which could have led to an overestimation of feasibility outcomes. At the same time, the study has notable strengths. Sampling covered both trough and a wide range of post-dose intervals, capturing pharmacokinetic variability across the therapeutic range. This design allowed for evaluation of suitability not only for trough-based monitoring, but also for peak- or AUC-based TDM. Two volumetric devices were evaluated, which are better suited for home use and circumvent key limitations of non-volumetric microsampling [19, 45]. Analyzing both also revealed device-specific differences often overlooked when microsampling techniques are considered interchangeable. Finally, we systematically evaluated a broad set of correction strategies, including those previously proposed for plasma-confined drugs, which provided important insights into the conditions under which correction approaches succeed or fail.