Background: Accuracy of outcome ascertainment is crucial to ensure validity when mining electronic healthcare records (EHR) databases for drug safety signal detection.[1-3] Objective: To evaluate and compare the accuracy of various coding algorithms used to identify cases of acute myocardial infarction (AMI) from three European EHR databases. Methods: We conducted a validation study in three databases of the EU-ADR network:[4] (1) IPCI (GP database, Netherlands); (2) HSD (GP database, Italy): and 3) Aarhus (claims, Denmark). We identified cases of AMI from GP medical records, primary hospital discharge diagnoses, and death registries using coding algorithms which employed different disease terminology schemes: (1) ICPC; (2) ICD9-CM; (3) ICD-10th revision. We also used free text using key words consistent with AMI. A random sample of 200 cases per database was obtained from all potential cases identified. Additional 200 cases identified by free text search were obtained in IPCI. Manual review of medical records and hospitalisation charts was performed using standardised questionnaire implemented as computerised data entry via custom-built software Chameleon,ª locally installed in each database. Positive predictive values (PPV) were calculated overall and for each code and free text query. Results: The study population comprised healthcare data from 4 034 232 individuals with 22 428 883 person-years (PYs) of follow-up during the period 1995–2011. Within this population, a total of 42 774 potential cases of AMI were identified. From the random sample of 800 potential cases of AMI selected for validation, 748 records were retrieved (93.5%) and reviewed. All ICD-10 codes used (I21.0, I21.1, I21.2, I21.3, I21.4, and I21.9) had 100% PPV. Overall the ICD9-CM codes had very good PPV, with 410.9/410.90, the most frequently occurring code having a PPV of 96.5% (95%CI 93.5-100.4). The ICPC code K75 had a PPV of 75% (67.4-82.6). Use of free text had a lower PPV: 60% (95% CI 17.1-102.9) in HSD and 19.7% (95%CI 12.9-26.5) in IPCI. Conclusion: The results obtained in this study are consistent with the PPV estimates for ICD-9CM and ICD-10 cited in the literature. Strategies are necessary to further optimise the value of free text search in the identification of AMI in EHR databases.
Coloma, P., Valkhoff, V., Mazzaglia, G., Nielsson, M., Pedersen, L., Molokhia, M., et al. (2012). Accuracy of Coding-Based Algorithms in Identification of Acute Myocardial Infarction from Multi-Country Electronic Healthcare Record Databases. Intervento presentato a: Annual Meeting of the International-Society-of-Pharmacovigilance (ISoP), Cancun, Mexico.
Accuracy of Coding-Based Algorithms in Identification of Acute Myocardial Infarction from Multi-Country Electronic Healthcare Record Databases
Mazzaglia G;
2012
Abstract
Background: Accuracy of outcome ascertainment is crucial to ensure validity when mining electronic healthcare records (EHR) databases for drug safety signal detection.[1-3] Objective: To evaluate and compare the accuracy of various coding algorithms used to identify cases of acute myocardial infarction (AMI) from three European EHR databases. Methods: We conducted a validation study in three databases of the EU-ADR network:[4] (1) IPCI (GP database, Netherlands); (2) HSD (GP database, Italy): and 3) Aarhus (claims, Denmark). We identified cases of AMI from GP medical records, primary hospital discharge diagnoses, and death registries using coding algorithms which employed different disease terminology schemes: (1) ICPC; (2) ICD9-CM; (3) ICD-10th revision. We also used free text using key words consistent with AMI. A random sample of 200 cases per database was obtained from all potential cases identified. Additional 200 cases identified by free text search were obtained in IPCI. Manual review of medical records and hospitalisation charts was performed using standardised questionnaire implemented as computerised data entry via custom-built software Chameleon,ª locally installed in each database. Positive predictive values (PPV) were calculated overall and for each code and free text query. Results: The study population comprised healthcare data from 4 034 232 individuals with 22 428 883 person-years (PYs) of follow-up during the period 1995–2011. Within this population, a total of 42 774 potential cases of AMI were identified. From the random sample of 800 potential cases of AMI selected for validation, 748 records were retrieved (93.5%) and reviewed. All ICD-10 codes used (I21.0, I21.1, I21.2, I21.3, I21.4, and I21.9) had 100% PPV. Overall the ICD9-CM codes had very good PPV, with 410.9/410.90, the most frequently occurring code having a PPV of 96.5% (95%CI 93.5-100.4). The ICPC code K75 had a PPV of 75% (67.4-82.6). Use of free text had a lower PPV: 60% (95% CI 17.1-102.9) in HSD and 19.7% (95%CI 12.9-26.5) in IPCI. Conclusion: The results obtained in this study are consistent with the PPV estimates for ICD-9CM and ICD-10 cited in the literature. Strategies are necessary to further optimise the value of free text search in the identification of AMI in EHR databases.File | Dimensione | Formato | |
---|---|---|---|
ABSTRACTS__12th_ISoP_Annual_Meeting__New.10.pdf
Solo gestori archivio
Dimensione
1.15 MB
Formato
Adobe PDF
|
1.15 MB | Adobe PDF | Visualizza/Apri Richiedi una copia |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.