Automated data extraction—A feasible way to construct patient registers of primary care utilization

Figures & data

Table I. The CXP arranged extracted data into nine tables that correspond to the nine modules in the original EMR. Data in the top eight tables are extracted from structural data, whereas “Terminology” is extracted from narrative data.

Name of table	Data to be customized for second step in the extraction process
Contacts	Type of contact registration (e.g. doctor, nurse, telephone, administrative), user ID, and date or time interval of contact
Diagnosis	Name, code according to ICD –9 –10, and date or time interval of diagnosis
Biometrics	Weight, height, and BMI sorted by date
Documents	Referrals and other documentation sorted by date
Drugs	Prescriptions (name, ATC code, iteration, and dosage) sorted by date
Biochemical analysis	Analyses, values, and units sorted by date
Measurement	Biometric measurements documented outside the table “Terminology”
Patients	Gender, age, alive/dead
Terminology	Narrative data in the core journal text are found under selected subheadings (e.g. current disease, heritage, physical status)

Figure 1. Flow chart for the procedure of selecting personal identification numbers (PIN) to assess the congruity of CXP extracted data to the original EMRs. In step B, the PIN already selected in step A was excluded. In step C, PINs from step A and step B were excluded. Altogether 3,045 data items were compared.

Figure 2. Venn diagrams for data extraction by CXP and by eXtractor, illustrating distribution of inclusion criteria (ICD-10 code, ATC code, laboratory value) for each patient. Patients extracted by Pygargus CXP (n = 445) and by eXtractor (n = 433).

Figure 3. Data items (n = 3,045) compared to the original EMR distributed over nine CXP modules and by inclusion (ATC code, laboratory values, or ICD code).