Supplementary material to the paper

Predicting Blood Donor Arrival

by Vidar Bosnes1, Magne Aldrin2 and Hans Erik Heier1

1) Department of Immunology and Transfusion Medicine, Ullevål University Hospital, Oslo.

2) The Norwegian Computing Center, Oslo

To be published in [ Publication information will be inserted here ]


Regression coefficients of the ordinary logistic regression (OLR) model, with standard errors and p-values (adjusted for overdispersion).

  Variablea Estimate  Standard error P-valueb
  Intercept -0.968 0.221 <0.001
  male -0.047 0.015 0.002
  personal.contact 1.418 0.018 <0.001
  platelet.donation 0.322 0.097 0.001
  firsttime.donor 0.335 0.041 <0.001
  donation.site 0.131 0.016 <0.001
  min(donor.age, 33) -0.001 0.003 0.561
  max(donor.age, 33) 0.015 0.001 <0.001
  n.donations 0.004 <0.001 <0.001
  min(n.arrivals.2y, 8) 0.124 0.005 <0.001
  max(n.arrivals.2y, 8) 0.018 0.013 0.175
  n.deferrals.2y -0.089 0.013 <0.001
  n.cancellations.2y -0.005 0.002 0.038
  min(n.noshows.2y, 10) -0.148 0.005 <0.001
  max(n.noshows.2y, 10) -0.055 0.017 0.001
  min(arrival.ratio.2y, 0.5) 0.176 0.096 0.067
  max(arrival.ratio.2y, 0.5) 1.195 0.069 <0.001
  max(time.since.prev.visit, 90) -0.001 <0.001 <0.001
  min(time.since.appt.made, 15) -0.075 0.002 <0.001
  max(time.since.appt.made, 15) -0.009 0.001 <0.001
  (day.of.year-183)/365 -0.103 0.067 0.124a
  ((day.of.year-183)/365)2 0.032 0.107 0.769a
  ((day.of.year-183)/365)3 -0.145 0.421 0.731a
  Tuesday (=1 if day.of.week=2) 0.206 0.021 <0.001
  Wednesday (=1 if day.of.week=3) 0.281 0.023 <0.001
  Thursday (=1 if day.of.week=4) 0.311 0.021 <0.001
  Friday (=1 if day.of.week =5) 0.353 0.023 <0.001
  (time.of.day - 12)/12 0.021 0.073 0.775
  ((time.of.day - 12)/12)2 1.620 0.182 <0.001
  ((time.of.day - 12)/12)3 -3.002 0.648 <0.001
  (day.number - 500)/1000 0.028 0.067 0.675
  ((day.number - 500)/1000)2 0.231 0.104 0.026
  ((day.number - 500)/1000)3 -0.886 0.447 0.047

 a   All explanatory variables had significant effects, although some of the transformed components were non-significant. In particular, all three coefficients related to the variable day.of.year were non-significant alone, but a Wald test for the simultaneous effect of these gave p<0.001.
 b   Non-significant coefficients are often deleted, but this is of negligible practical importance in the present study, with nearly 180 000 observations. The purpose of performing the transformations was to approximate the additive logistic regresion (ALR) model. Since it was observed that keeping the non-significant coefficents gave a better approximation than deleting them, we decided to keep non-significant coefficients.
The non-significant coefficients were either close to 0 (e.g. min(donor.age,33)), related to parts of the curves with little data (e.g. max(n.arrivals.2y,8)), or significant together with other coefficients (day.of.year).