Challenges associated with missing data in electronic health records: A case study of a risk prediction model for diabetes using data from Slovenian primary care

Abstract

The increasing availability of data stored in electronic health records brings substantial opportunities for advancing patient care and population health. This is, however, fundamentally dependant on the completeness and quality of data in these electronic health records. We sought to use electronic health record data to populate a risk prediction model for identifying patients with undiagnosed type 2 diabetes mellitus. We, however, found substantial (up to 90%) amounts of missing data in some healthcare centres. Attempts at imputing for these missing data or using reduced dataset by removing incomplete records resulted in a major deterioration in the performance of the prediction model. This case study illustrates the substantial wasted opportunities resulting from incomplete records by simulation of missing and incomplete records in predictive modelling process. Government and professional bodies need to prioritise efforts to address these data shortcomings in order to ensure that electronic health record data are maximally exploited for patient and population benefit.

Development of a screening tool using electronic health records for undiagnosed Type 2 diabetes mellitus and impaired fasting glucose detection in the Slovenian population

“…As demonstrated in Stiglic et al . where an UT2DM predictive model was built on increasing fraction of complete samples, the predictive performance stabilizes at ~600 complete records used to build the model.…”

“…As reported in the study by Stiglic et al . on a smaller subset ( n =952) of data from three healthcare centres in Slovenia, the imputation of missing data has a significant impact on the performance of a predictive model, but only for smaller samples. As demonstrated in Stiglic et al .…”

“…The only significant difference could be found in the age variable, with a difference of up to 2.5 years; however, as the distribution of the difference was equal over all three groups [UT2DM, IFG and normal fasting glucose (NFG)], this difference should not have affected the model significantly. The high number of missing values can be attributed to the fact that entry of detailed FINDRISC questionnaire data is not mandatory at the primary healthcare level in Slovenia .…”

Off-label use of pentazocine and the associated adverse events among pediatric surgical patients in a tertiary hospital in Northern Nigeria: a retrospective chart review

“…Another limitation is the retrospective nature of the study, which, like previous AEs studies, is characterized by missing and inaccurate data [51,52]. Similar to previous retrospective studies, we may not have complete ascertainment of all the AEs experienced by our cohort during the study period [53]. The indications for pentazocine exposure and the dosing regimen of the drug were not documented for three children, while the case files for two children were irretrievable.…”

Peer Review #2 of "Building interpretable models for polypharmacy prediction in older chronic patients based on drug prescription records (v0.1)"

“…With the rapid introduction of the electronic health records (EHR), particularly at the primary healthcare level, it will be possible to effectively monitor and identify groups of patients or individuals at high risk for drug-induced or related health problems [6]. Additionally, linking different EHR repositories together [7] and solving challenges in capturing the data in electronic form [8] will allow further improvements of data driven techniques.…”