There is growing demand for human pathology test services in Australia and around the world, driven by the ageing global population and increasing incidence of chronic diseases [13]. Dried blood spot sampling is an alternative to traditional blood sampling, and has been used in clinical and epidemiological studies for several decades [6, 8, 14]. This method provides results that are comparable to those obtained through traditional venipucture [2, 12], without its logistical obstacles regarding sample collection, processing, transportation and storage. For the measurement of HbA1c, DBS has also been shown to produce results that are comparable to those obtained through venous sampling [15–25]. In our study, we showed that HbA1c levels from DBS samples collected via finger pricking from volunteers with and without diabetes were comparable to those measured from venous samples, when measured up to seven days after collection.
In our study, DBS samples collected from finger pricking (capDBS) were analysed on the same day (D0), four (D4), seven (D7) and fourteen (D14) days after collection. High correlation and agreement between capDBS results on D0 and venous blood HbA1c values showed that the analysis of samples collected on matrix paper and analysed immediately provides results that are similar to those obtained and processed by traditional methods.
In a real-life scenario, DBS samples are mailed or shipped to the pathology laboratory that performs the assays. Therefore, DBS samples are not analysed immediately. To assess whether this gap between collection and analysis may interfere with the results, we performed analyses also four, seven and fourteen days after collection. We observed that, over time, the correlation between DBS and venous blood results becomes weaker, and the 95 % limits of agreement become wider, especially for D14 results, which may be clinically unacceptable. It is noteworthy that WB samples also degrade over time when not analysed immediately, particularly if not kept refrigerated. Haemoglobin degradation products may show up in samples that have coagulated and aged. These products may co-elute with, or be incompletely separated from, HbA1c. In these cases, the HbA1c value obtained may be reported as higher than it actually is [26]. This effect is particularly evident for venDBS samples, which were collected without anticoagulant.
We used a linear regression model for cross-validation, to ensure unbiased measures of goodness-of-fit and prediction intervals for WB. In that model, capD0-DBS results were closer to the predicted WB values for all evaluated HbA1c steps (4 %, 7 %, 7.5 %, and 10 %), and the prediction intervals were narrower. On the remaining days, capDBS results were further away from the predicted WB values, and the prediction intervals broadened over time. In the clinics, capD0-DBS samples would provide the most accurate HbA1c results, closer to the predicted WB results and with a narrower prediction interval. However, the difference between predicted WB and both capD4- and capD7-DBS results may be clinically acceptable, as well as their prediction intervals. In the case of capD14-DBS results, their prediction intervals may be too wide to be clinically acceptable. We applied four different capDBS values to the model, but any result can be applied to it (R program available upon request), providing similar behaviour.
In some cases, patients may have difficulty in collecting sufficient amount of blood samples from finger pricking on the matrix paper. That difficulty was evidenced by the fact that the sample size for each day was not equal to the total number of recruited participants. Therefore, we assessed whether venous blood collected through standard methods and spotted on the matrix paper would produce similar results. In those analyses, venous DBS samples were correlated to traditionally-processed venous blood samples in a similar way as capillary DBS. Also, there was high correlation and agreement between capDBS and venDBS results.
In our study, we recruited 67 diabetic patients. Most of them had type 2 diabetes, and had HbA1c levels that are considered adequate (particularly among participants with type 2 diabetes). Only three participants had HbA1c levels higher than 9 %. Therefore, results might have been different should more participants with decompensated diabetes had been recruited. Samples were measured at least in duplicates, and the median intra-assay coefficients of variation were clinically acceptable, lower than 3 % at all times. However, some participants had heterogeneous results. It is unclear why results using the same sample and assay method may vary in some participants.
One of the key issues to be considered for the employment of DBS sampling is the standardization of the analysis of the DBS measurements. It is essential to predict, with the highest possible level of accuracy, the concentration of HbA1c in WB from the values measured in the DBS tests. In a recent study, a meta-analysis of seventeen heterogeneous studies (employing different methods for measuring HbA1c) was performed by Affan et al., and a correction formula to approximate the DBS results to the WB values was published (12). We employed their correction formula in our current studies (results not presented), but the outcomes of the corrected DBS values resulted in a poorer approximation to the WB values. It appears that the time elapsed between sample collection and processing is a key component of the variability observed in the DBS sampling. Indeed, we found that there are significant differences among the Bland-Altman plots constructed from data on each particular day (e.g. D0, D4, D7 and D14) between DBS vs. WB. Thus, in our current analytical method, we analysed the data of each processed day independently by proposing mean values and prediction intervals for each particular processing day. Additionally, we corrected capillary DBS results by applying a correction formula that derived from regression analyses for each particular day, to approximate the regression line to the equality line. Thus, we obtained a different formula for each day, and observed that corrected capillary DBS results were closer to the predicted WB ones on all days except D14, when MSE was higher (i.e., WB results were less precisely predicted).
We acknowledge that our study is limited by the fact that participants were evaluated in a controlled research setting, and results may be different when capillary HbA1c is evaluated in a real-life scenario. Future studies need to evaluate samples from participants who collect their capillary DBS samples on their own, and mail them to the testing laboratory via standard postal services (subjected to confounding factors such as delays and temperature variations). Furthermore, future studies should evaluate the prediction intervals for other elapsed times such as D1, D2, and D3, and also determine how these prediction intervals can be applied in the management of diabetes. To answer those questions, future studies should evaluate healthy individuals and those with diabetes who are treatment-naïve, and compare their DBS values and their prediction intervals with their WB HbA1c outcomes. By considering their WB values as the gold-standard, a more accurate clinical interpretation of the prediction intervals, as proposed here, could be established.