Table 1

Performance of the algorithm for predicting daily dosage in the validation sample

ModelsPrescription typeNAccuracy (95% CI)G-means (95% CI)
Non-informativeness prediction
(NLP1 model)
Regular prescriptions9820.992 (0.986 to 0.998)0.969 (0.958 to 0.980)
Titrated prescriptions181.000 (1.000 to 1.000)1.000 (1.000 to 1.000)
All prescriptions10000.992 (0.968 to 0.998)0.973 (0.963 to 0.983)
Dosage prediction
(NLP2 model)
Regular prescriptions8520.964 (0.951 to 0.976)0.969 (0.957 to 0.981)
Titrated prescriptions180.944 (0.839 to 1.000)0.972 (0.896 to 1.000)
All prescriptions8700.963 (0.951 to 0.976)0.968 (0.956 to 0.980)
Overall predictionRegular prescriptions9820.968 (0.957 to 0.979)N/A
Titrated prescriptions180.944 (0.839 to 1.000)N/A
All prescriptions10000.968 (0.957 to 0.979)N/A
  • N/A, not available; NLP1, first natural language processing model; NLP2, second natural language processing model.