Bias due to re-used databases: Coding in hospital for extremely vulnerable patients

Abstract

Objective

This paper interrogates bias caused by heterogeneity in coding processes through an analysis of electronic medical records EMR databases in France. In general, researchers and professionals often apply data not only for its primary function but also for multiple alternative purposes. However, how they code information might be inconsistent with alternative purposes that exploit existing databases.

Methods

Using the EMR acute care and the EMR rehabilitation care databases, we select more than 800,000 patients coded as socially vulnerable during their rehabilitation stay. Statistical analysis was conducted to describe the types of heterogeneity and to compare the distribution of vulnerability coding processes across different hospital statuses and individual social vulnerability roles. Coding process rates were also analyzed.

Results

This paper shows the heterogeneity in this process of social vulnerability coding, exploiting acute care database and rehabilitation care database. For groups of patients with ICD-10 coded as socially vulnerable during their rehabilitation stays, the probability of being previously coded as such during their acute care stay is 11.4 % higher in the public sector than in the private one.

Conclusion

Implementing the EMR system leads to heterogeneity in the coding process. The paper concludes by arguing that heterogeneity in coding is not random but rather calculated. Applying this database in epidemiologic studies or health economics projects that factor in patients’ vulnerability information may lead to unintended biased results. These findings might also be useful for policymakers using EMR to plan for implementing new reforms in many healthcare settings.