Impact of Predictive Scoring Model and Email Messaging on African American Blood Donors

  • Lohith Bachegowda (Weil-Cornell University/New York Blood Center)
  • Debra Kessler (New York Blood Center, New York, NY)
  • Christopher France (Ohio University)
  • Pinaki Dasgupta (Hindsait Inc)
  • Brad Timm (Hindsait Inc)
  • Beth Shaz (New York Blood Center, New York, NY)

Abstract

African Americans (AA) are under-represented in the national blood donor pool, but are increasingly recruited to manage the transfusion needs of sickle cell disease (SCD) patients. To address this shortage, we hypothesized that computational analytics utilizing donor-related predictor variables and email communication could identify repeat AA blood donors who would return to donate. Within our database (BOSS Solutions, France), we identified all AA donors and calculated several donor-level metrics including: age, gender, blood type, days since last donation (minimum, maximum and median during each of eight years), total number of units donated and the proportion of visits they were eligible to donate between 2005-2012. We employed these metrics as predictor variables in models calculating the likelihood of repeat blood donations for the year 2013 from those donors who donated at least once during 2011-2012, using a suite of proprietary machine learning algorithms (Hindsait's Inc, NJ). We then applied these models to all 2011-2012 donors in a cross-validation framework to generate a Predicted Repeat Donor Score (PRDS) for each donor. PRDS values ranged from 0-1 with higher scores indicating a greater likelihood of donation. Additionally, as a pilot study to assess the effect of email communication on blood donation rates, an email was sent out in April 2015 to all eligible AA donors, seeking donations within two months. Two different email templates were created and sent to approximately an equal number of donors, one focusing on the benefits to SCD patients (majority of whom are AA) receiving blood from other healthy AA donors, and another, a generic request for blood donation. From 2004 through 2014, 95,228 unique AA donors contributed 347,016 donations. From this cohort, 30,786 donors donated at least once during 2011-2012 and had a PRDS calculated for 2013. Only 8,547 (27.8%) returned to donate at least once during 2013. Blood donors in the year 2013 had a significantly higher mean PRDS compared to non-donors (0.649 vs 0.268) (p<0.001). High PRDS (≥0.6) compared to low PRDS (<0.6) was associated with 20% more email opening rate (p <0.001), 89% higher rate of presentation in two months (p <0.001), and specifically amongst those who opened the email, 159% higher two months’ donation presentation rate (p<0.001), respectively. In the email non-opener group, high PRDS donors were 45% more likely to present for donation, compared to the low PRDS group (p=0.03). The generic or culturally tailored message did not significantly alter blood donation rate (1.4% vs 1.3%) (p=0.79). Our data suggests computation algorithms utilizing donor metrics can identify highly committed AA blood donors. Targeted email communication with AA blood donors has the potential to increase future blood donations in a cost-effective manner.