Premium
Plateletpheresis efficiency and mathematical correction of software‐derived platelet yield prediction: A linear regression and ROC modeling approach
Author(s) -
JaimePérez José Carlos,
JiménezCastillo Raúl Alberto,
VázquezHernández Karina Elizabeth,
SalazarRiojas Rosario,
MéndezRamírez Nereida,
GómezAlmaguer David
Publication year - 2017
Publication title -
journal of clinical apheresis
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.697
H-Index - 46
eISSN - 1098-1101
pISSN - 0733-2459
DOI - 10.1002/jca.21518
Subject(s) - plateletpheresis , receiver operating characteristic , medicine , apheresis , linear regression , statistics , software , nuclear medicine , regression analysis , mathematics , platelet , computer science , programming language
Background Advances in automated cell separators have improved the efficiency of plateletpheresis and the possibility of obtaining double products (DP). We assessed cell processor accuracy of predicted platelet (PLT) yields with the goal of a better prediction of DP collections. Study design and methods This retrospective proof‐of‐concept study included 302 plateletpheresis procedures performed on a Trima Accel v6.0 at the apheresis unit of a hematology department. Donor variables, software predicted yield and actual PLT yield were statistically evaluated. Software prediction was optimized by linear regression analysis and its optimal cut‐off to obtain a DP assessed by receiver operating characteristic curve (ROC) modeling. Results Three hundred and two plateletpheresis procedures were performed; in 271 (89.7%) occasions, donors were men and in 31 (10.3%) women. Pre‐donation PLT count had the best direct correlation with actual PLT yield ( r = 0.486. P < .001). Means of software machine‐derived values differed significantly from actual PLT yield, 4.72 × 10 11 vs.6.12 × 10 11 , respectively, ( P < .001). The following equation was developed to adjust these values: actual PLT yield= 0.221 + (1.254 × theoretical platelet yield). ROC curve model showed an optimal apheresis device software prediction cut‐off of 4.65 × 10 11 to obtain a DP, with a sensitivity of 82.2%, specificity of 93.3%, and an area under the curve (AUC) of 0.909. Conclusion Trima Accel v6.0 software consistently underestimated PLT yields. Simple correction derived from linear regression analysis accurately corrected this underestimation and ROC analysis identified a precise cut‐off to reliably predict a DP.