Plateletpheresis efficiency and mathematical correction of software-derived platelet yield prediction: A linear regression and ROC modeling approach

Predicting donor-related factors for high platelet yield donations by classification and regression tree analysis

INTRODUCTION

Collecting high-dose (HD) or double-dose (DD) apheresis platelets units from a single collection offers significant benefit by improving inventory logistics and minimizing the cost per unit produced. Platelet collection yield by apheresis is primarily influenced by donor factors, but the cell separator used also affects the collection yield.

OBJECTIVES

To predict the cutoff in donor factors resulting in HD and DD platelet collections between Trima/Spectra Optia and MCS+ apheresis equipment using Classification and Regression Trees (CART) analysis.

METHODS

High platelet yield collections (target ≥ 4.5 × 10¹¹ platelets) using MCS+, Trima Accel and Spectra Optia were included. Endpoints were ≥ 6 × 10¹¹ platelets for DD and ≥ 4.5 to < 6 × 10¹¹ for HD collections. The CART, a tree building technique, was used to predict the donor factors resulting in high-yield platelet collections in Trima/Spectra Optia and MCS+ equipment by R programming.

RESULTS

Out of 1,102 donations, the DDs represented 60% and the HDs, 31%. The Trima/Spectra Optia predicted higher success rates when the donor platelet count was set at ≥ 205 × 103/µl and ≥ 237 × 103/µl for HD and DD collections. The MCS+ predicted better success when the donor platelet count was ≥ 286 × 103/µl for HD and ≥ 384 × 103/µl for DD collections. Increased donor weight helped counter the effects of lower donor platelet counts only for HD collections in both the equipment.

CONCLUSIONS

The donor platelet count and weight formed the strongest criteria for predicting high platelet yield donations. Success rates for collecting DD and HD products were higher in the Trima/Spectra Optia, as they require lower donor platelet count and body weight than the MCS+.

Assessing the performance of a plateletpheresis unit at a tertiary-care academic medical center in Mexico: A 6-year experience

BACKGROUND

Optimization of platelet (PLT) apheresis collection is a priority to satisfy the increasing demand of hemato-oncology patients. We assessed the performance of a plateletpheresis unit supporting hematology patients.

STUDY DESIGN AND METHODS

This descriptive retrospective study included 561 plateletpheresis collections from 2013 to 2018. For data analysis, descriptive statistics and receiver operating characteristic (ROC) curve were used. A 5-item satisfaction questionnaire was analyzed.

RESULTS

Ninety percent of the donors were males. The median plateletpheresis time was 89 minutes; its success rate was 92.5%; median donor PLT count was 232 × 10⁹ /L, women median PLT count was 247 × 10⁹ /L vs 231x10⁹ /L in men (P = .017). Seventy-seven percent donors were candidates for a double product and 24.5% were processed; 20.8% of these donors had a weight ≤75 and 79.2% >75 kg, P = .003, and 6.6% were women and 93.4% men, P = .161. Thirty-six of donors had ≥250 × 10⁹ /L and 16.8% was processed as a triple product. ROC analysis showed that with donor PLT counts ≥200 × 10⁹ /L the sensitivity for obtaining double products was 0.981 and specificity 0.714, with an area under the curve (AUC) = 0.877. The adverse effect rate was 4.3%. Of the potential donors, 6.3% were rejected. The cost of processing single or double products was 430 USD. Comfort and time spent during plateletpheresis were areas for improvement.

CONCLUSION

Platelet count and donor weight predicted PLT yield and obtaining double products. Women had higher PLT counts, but no significant difference was found between donor gender and processed products. Assessment of the apheresis unit can help to improve its performance.

Donor platelet and leukocyte count as predictive factors of the quality of platelet concentrates obtained from whole blood by semiautomated fractionation

Platelet concentrates (PCs) obtained from whole blood are produced by fractionation of the buffy coat (BC) or the platelet-rich plasma. Despite the improvements in the technologies used for the hemocomponent fractionation, the proportion of PCs that do not accomplish the quality requirements is high. This study aimed to determine whether the basal platelet and leukocyte counts are predictive factors of the quality of the PCs obtained from BC by semiautomated fractionation. Quality control registers of 196 PCs were analyzed. Gender- and age-dependence of the blood cell count and the characteristics of PCs were evaluated. Platelet yield and residual leukocytes in the PCs were correlated with the platelet and leukocyte counts and the age of the donors. Predictive efficacy was assessed, and an optimal cut-off was established. The proportions of PCs accepted and rejected by using or not the optimal cut-off were compared. 50.0% of the PCs accomplished all the quality control requirements. Female donors had a higher basal platelet count than males. A correlation was observed between basal platelets and platelet yield, but not between basal leukocytes and residual leukocytes. The basal platelet count predicted the quality of the PCs. A cut-off of 231,000 platelets/mm³ was established, but it did not improve the proportion of accepted PCs. In conclusion, we found that the basal platelet count is correlated with the platelet yield. The basal leukocyte count is not correlated with the residual leukocytes. The established cut-off for the basal platelet count did not improve the proportion of accepted PCs.

Assessment of software-derived predictive algorithms for platelet yield and blood cell count after apheresis

INTRODUCTION

Algorithms have been developed to predict the platelet yield after apheresis from the donor's data, as well as the effect on the blood cell count, to extract an acceptable platelet number without affecting the donor. However, the evaluation of these algorithms has not been widely reported. This study aimed to assess the accuracy of the predictive algorithms of the Trima Accel v. 6 blood collection system.

METHODS

Platelet concentrates (PCs) obtained by apheresis were analyzed. Platelet count and hematocrit were compared pre- and post-apheresis. Calculated post-apheresis platelet count (CPAPC), hematocrit (CPAH), and platelet yield (CPY), and their actual values were correlated. The bias of the algorithms was assessed with Bland-Altman plots, and the prediction of the extraction of single or double platelet products was evaluated.

RESULTS

Two hundred and seventy-nine PCs were analyzed. Post-apheresis platelet count (PAPC) and hematocrit were decreased. A moderate correlation was observed between CPY and the actual yield, with a negative bias, and a trend to increase alongside the magnitude of the measurements. CPAPC and CPAH were strongly correlated with their actual values without bias. Prediction of single or double platelet product extraction showed a significant agreement with the actual outcomes.

CONCLUSIONS

The predictive algorithm for the platelet yield showed bias, and a trend to underestimate the actual platelet yields when they are higher. The algorithms for the prediction of the PAPC and hematocrit did not show bias, proving their accuracy. Prediction of a single or double platelet product extraction has a strong agreement with the APY.