BACKGROUND: To assess the efficacy of a centralised review of a voluntary low-budget head injury database with a retrospective analysis of data before and after a centralised review. METHOD: A computerised data collection (Neurolink) on traumatic brain injury cases admitted to three neuro-intensive care units in Milan (Italy): analysis of a three-year period (1999-2001). Data from 499 patients (epidemiology, type of lesion, clinical course, monitoring, treatment, complications and outcome). The audit involved a review of forms relating to patients enrolled in the three-year period, with the aim of improving the quality of data entry. Missing data in all empty fields were identified; evident errors and contradictory data were identified and corrected; missing and final data were analysed to test the efficacy of the review. FINDINGS: The total post-review missing data rate was significantly lower than the paired pre-review missing data rate (p = 0.001). The same was confirmed for each of the 3 years (p = 0.001 for each year). The missing data rate significantly improved over the three-year period (p = 0.001). Data for the pre-hospitalisation period had the highest missing rates; data regarding the ICU stay showed the greatest improvement after the review. A total of 407 items (0.44%) were identified as errors. CONCLUSIONS: Data quality is fundamental to avoid information bias in database analysis. This study indicates that it is possible to generate a serious data collection without significant resources. Audit seems to be an important tool before the final data is used for scientific projects.
BACKGROUND: To assess the efficacy of a centralised review of a voluntary low-budget head injury database with a retrospective analysis of data before and after a centralised review. METHOD: A computerised data collection (Neurolink) on traumatic brain injury cases admitted to three neuro-intensive care units in Milan (Italy): analysis of a three-year period (1999-2001). Data from 499 patients (epidemiology, type of lesion, clinical course, monitoring, treatment, complications and outcome). The audit involved a review of forms relating to patients enrolled in the three-year period, with the aim of improving the quality of data entry. Missing data in all empty fields were identified; evident errors and contradictory data were identified and corrected; missing and final data were analysed to test the efficacy of the review. FINDINGS: The total post-review missing data rate was significantly lower than the paired pre-review missing data rate (p = 0.001). The same was confirmed for each of the 3 years (p = 0.001 for each year). The missing data rate significantly improved over the three-year period (p = 0.001). Data for the pre-hospitalisation period had the highest missing rates; data regarding the ICU stay showed the greatest improvement after the review. A total of 407 items (0.44%) were identified as errors. CONCLUSIONS: Data quality is fundamental to avoid information bias in database analysis. This study indicates that it is possible to generate a serious data collection without significant resources. Audit seems to be an important tool before the final data is used for scientific projects.
Authors: Matthew K H Hong; Henry H I Yao; John S Pedersen; Justin S Peters; Anthony J Costello; Declan G Murphy; Christopher M Hovens; Niall M Corcoran Journal: BMJ Open Date: 2013-05-28 Impact factor: 2.692