TY - JOUR
T1 - Deriving Dose Limits for Warnings in Electronic Prescribing Systems Statistical Analysis of Prescription Data at University Hospital Birmingham, UK
AU - Coleman, Jamie
AU - Hodson, James
AU - Ferner, Robin
PY - 2012/1/1
Y1 - 2012/1/1
N2 - Introduction: Electronic decision support can reduce medication errors, and dose-range checking is one element of that support.
Objective: The aim of this study was to design an approach to setting upper dose warning limits in electronic prescribing systems where there are historical data on dosing.
Method: We used historical data on 56 drug-form combinations for which over 100 prescriptions had been issued between I June 2009 and 31 May 2010 in a bespoke electronic prescribing system at University Hospital Birmingham, UK. First, two experts derived dose limits for each drug-form combination, then the drugs were randomly divided into a training set and a test set. A variation of the 'Nearest Rank' approach to estimate statistical limits was used to derive the percentile with the optimal sensitivity and specificity.
Results: For the 28 drug-form combinations in the test set, the 86th percentile of dose gave a mean sensitivity of 95.3% and a mean specificity of 97.9% for warning limits, representing the highest reasonable dose; the 96th percentile gave a mean sensitivity of 90.2% and mean specificity of 99.5% for disallow limits, beyond which no dose should be prescribed.
Conclusions: Dosing decision support within electronic prescribing systems can be derived by statistical analysis of historical prescription data. We advocate a combined theoretical and statistical derivation of dose checking rules in order to ensure that prescribers are alerted appropriately to potentially toxic doses.
AB - Introduction: Electronic decision support can reduce medication errors, and dose-range checking is one element of that support.
Objective: The aim of this study was to design an approach to setting upper dose warning limits in electronic prescribing systems where there are historical data on dosing.
Method: We used historical data on 56 drug-form combinations for which over 100 prescriptions had been issued between I June 2009 and 31 May 2010 in a bespoke electronic prescribing system at University Hospital Birmingham, UK. First, two experts derived dose limits for each drug-form combination, then the drugs were randomly divided into a training set and a test set. A variation of the 'Nearest Rank' approach to estimate statistical limits was used to derive the percentile with the optimal sensitivity and specificity.
Results: For the 28 drug-form combinations in the test set, the 86th percentile of dose gave a mean sensitivity of 95.3% and a mean specificity of 97.9% for warning limits, representing the highest reasonable dose; the 96th percentile gave a mean sensitivity of 90.2% and mean specificity of 99.5% for disallow limits, beyond which no dose should be prescribed.
Conclusions: Dosing decision support within electronic prescribing systems can be derived by statistical analysis of historical prescription data. We advocate a combined theoretical and statistical derivation of dose checking rules in order to ensure that prescribers are alerted appropriately to potentially toxic doses.
U2 - 10.2165/11594810-000000000-00000
DO - 10.2165/11594810-000000000-00000
M3 - Article
C2 - 22263779
VL - 35
SP - 291
EP - 298
JO - Drug Safety
JF - Drug Safety
IS - 4
ER -