Design and Performance of a Low-Cost, Portable Platelet Counter for Point-of-Care and at-Home Use

Background

Patients with recurrent or chronic moderate-to-severe thrombocytopenia can experience a marked reduction in quality of life due to the need for regular blood draws, frequent contact with the health care system, and potential for bleeding complications requiring emergency department visits and/or hospital admissions. To improve quality of life while maintaining effective platelet monitoring, we introduce a portable, point-of-care platelet counter that can be used in any health care setting or at home.

Methods

Obtaining a platelet count requires three components: a counter, staining modality, and erythrocyte lysing agent. The counting device described here is the Athelas One, a small cylindrical device containing a miniaturized microscope embedded with a control board. A strip containing a stained microfluidic channel approximately 10µm tall and thus suitable for a monolayer of blood cells houses the sample. An ammonium oxalate solution is used to lyse the erythrocytes in the capillary sample. Images of the blood sample are acquired and then analyzed using a convolutional neural network to yield platelet counts. To use the device, a patient pricks his or her finger and collects 20µL of blood in a K2EDTA-coated minivette. The blood is mixed with the lysing agent and inserted into the strip, which in turn is inserted into the device. A platelet count is obtained within 5 minutes.

In this observational study, we enrolled 233 adults at least 21 years of age with normal platelet counts at a single teaching hospital. Non-hematological illness was allowed. Each patient's platelet count was obtained using both capillary samples processed on Athelas One and paired anticoagulated venous samples processed on a standard hematology analyzer, the Sysmex XS-1000i (SX). Platelet counts from Athelas One and SX were compared using a Passing-Bablok regression model with a Pearson correlation coefficient. Bland-Altman analysis was also performed. Confidence intervals (CI) were calculated using a Student's t-distribution. Least squares regression modeled the relationship between time necessary to collect capillary samples and bias between Athelas One and SX. Analyses were conducted in R.

Results

A strong linear relationship was observed between Athelas One and SX: Passing-Bablok regression revealed a slope of 0.98 and a y-intercept estimate of -16.45 (Figure), and the Pearson correlation-coefficient was 0.70. Bland-Altman statistics revealed an average percent difference of -8.66% (95% CI: -6.55% to -10.77%), reflecting a bias of Athelas One to undercount as compared with SX.

In an effort to explain the predilection for undercounting, device performance was analyzed as a function of sample collection time. Least squares regression revealed a slope of -1.08 (95% CI: -1.55 to -0.61) and a y-intercept estimate of 8.16 (95% CI: -12.15 to 28.47). The Pearson correlation-coefficient was 0.54.

Conclusion

Athelas One compared favorably with SX. The bias for undercounting may be a consequence of a pseudothrombocytopenia phenomenon due to platelet activation and clot formation prior to sample collection, which sequesters platelets out of solution. This hypothesis is supported by the fact that longer time needed to collect the capillary sample correlated with larger negative bias between Athelas One and SX. To improve agreement, a number of methods in the counting procedure, namely sample collection, could be altered. A potential solution involves wiping the finger with an anticoagulant prior to sample collection to discourage clot formation.

This proof-of-concept analysis demonstrates the plausibility of point-of-care platelet monitoring. Future work will focus on validating Athelas One in thrombocytopenic and thrombocytotic samples.