Demand forecasting for platelet usage: From univariate time series to multivariable models

Platelet products are both expensive and have very short shelf lives. As usage rates for platelets are highly variable, the effective management of platelet demand and supply is very important yet challenging. The primary goal of this paper is to present an efficient forecasting model for platelet demand at Canadian Blood Services (CBS). To accomplish this goal, five different demand forecasting methods, ARIMA (Auto Regressive Integrated Moving Average), Prophet, lasso regression (least absolute shrinkage and selection operator), random forest, and LSTM (Long Short-Term Memory) networks are utilized and evaluated via a rolling window method. We use a large clinical dataset for a centralized blood distribution centre for four hospitals in Hamilton, Ontario, spanning from 2010 to 2018 and consisting of daily platelet transfusions along with information such as the product specifications, the recipients' characteristics, and the recipients' laboratory test results. This study is the first to utilize different methods from statistical time series models to data-driven regression and machine learning techniques for platelet transfusion using clinical predictors and with different amounts of data. We find that the multivariable approaches have the highest accuracy in general, however, if sufficient data are available, a simpler time series approach appears to be sufficient. We also comment on the approach to choose predictors for the multivariable models.

An unsupervised learning approach to identify immunoglobulin utilization patterns using electronic health records

BACKGROUND

Managing Canada's immunoglobulin (Ig) product resource allocation is challenging due to increasing demand, high expenditure, and global shortages. Detection of groups with high utilization rates can help with resource planning for Ig products. This study aims to uncover utilization subgroups among the Ig recipients using electronic health records (EHRs).

METHODS

The study included all Ig recipients (intravenous or subcutaneous) in Calgary from 2014 to 2020, and their EHR data, including blood inventory, recipient demographics, and laboratory test results, were analyzed. Patient clusters were derived based on patient characteristics and laboratory test data using K-means clustering. Clusters were interpreted using descriptive analyses and visualization techniques.

RESULTS

Among 4112 recipients, six clusters were identified. Clusters 1 and 2 comprised 408 (9.9%) and 1272 (30.9%) patients, respectively, contributing to 62.2% and 27.1% of total Ig utilization. Cluster 3 included 1253 (30.5%) patients, with 86.4% of infusions administered in an inpatient setting. Cluster 4, comprising 1034 (25.1%) patients, had a median age of 4 years, while clusters 2-6 were adults with median ages of 46-60. Cluster 5 had 62 (1.5%) patients, with 77.3% infusions occurring in emergency departments. Cluster 6 contained 83 (2.0%) patients receiving subcutaneous Ig treatments.

CONCLUSION

The results identified data-driven segmentations of patients with high Ig utilization rates and patients with high risk for short-term inpatient use. Our report is the first on EHR data-driven clustering of Ig utilization patterns. The findings hold the potential to inform demand forecasting and resource allocation decisions during shortages of Ig products.

A data-informed system to manage scarce blood product allocation in a randomized controlled trial of convalescent plasma

BACKGROUND

Equitable allocation of scarce blood products needed for a randomized controlled trial (RCT) is a complex decision-making process within the blood supply chain. Strategies to improve resource allocation in this setting are lacking.

METHODS

We designed a custom-made, computerized system to manage the inventory and allocation of COVID-19 convalescent plasma (CCP) in a multi-site RCT, CONCOR-1. A hub-and-spoke distribution model enabled real-time inventory monitoring and assignment for randomization. A live CCP inventory system using REDCap was programmed for spoke sites to reserve, assign, and order CCP from hospital hubs. A data-driven mixed-integer programming model with supply and demand forecasting was developed to guide the equitable allocation of CCP at hubs across Canada (excluding Québec).

RESULTS

18/38 hospital study sites were hubs with a median of 2 spoke sites per hub. A total of 394.5 500-ml doses of CCP were distributed; 349.5 (88.6%) doses were transfused; 9.5 (2.4%) were wasted due to mechanical damage sustained to the blood bags; 35.5 (9.0%) were unused at the end of the trial. Due to supply shortages, 53/394.5 (13.4%) doses were imported from Héma-Québec to Canadian Blood Services (CBS), and 125 (31.7%) were transferred between CBS regional distribution centers to meet demand. 137/349.5 (39.2%) and 212.5 (60.8%) doses were transfused at hubs and spoke sites, respectively. The mean percentages of total unmet demand were similar across the hubs, indicating equitable allocation, using our model.

CONCLUSION

Computerized tools can provide efficient and immediate solutions for equitable allocation decisions of scarce blood products in RCTs.

From demand forecasting to inventory ordering decisions for red blood cells through integrating machine learning, statistical modeling, and inventory optimization

BACKGROUND

The demand and supply of blood are highly variable over time. Blood inventory management that relies heavily on experience-based decisions may not be adaptive to real demand, leading to high operational costs, wastage, and shortages.

METHODS

We combined statistical modeling, machine learning, and optimization methods to develop a data-driven demand forecasting and inventory management strategy for red blood cells (RBCs). We then used the strategy to inform daily blood orders. A secondary semi-weekly (twice per week) ordering strategy was developed to handle the last-mile split delivery problem for blood suppliers, characterized by multi-deliveries to the same location multiple times during a short period of time. Both strategies were evaluated using the TRUST database including all patient data across four hospitals in Hamilton, Ontario.

RESULTS

We identified 227,944 RBC transfusions for 40,787 patients in Hamilton, Ontario from 2012 to 2018. The predicted daily demand from the hybrid demand forecasting model was not significantly different from the actual daily demand (paired t-test p-value = 0.163); however, the proposed daily ordering quantity from the model was significantly lower than the actual ordering quantity (p-value <0.001). The proposed daily ordering strategy reduced inventory levels by 38.4% without risk of shortages, leading to an overall cost reduction of 43.0% (95% confidence interval [CI]: 42.3%, 43.7%) compared with the actual cost. The semi-weekly ordering strategy reduced ordering frequency by 62.6% (95% CI: 61.5%, 63.7%).

CONCLUSION

The proposed data-driven ordering strategy combining demand forecasting and inventory optimization can achieve significant cost savings for healthcare systems and blood suppliers.