Table 1.
Challenge category | Specific challenge | Example |
Selecting the correct data elements | Selecting the right action state and timestamp | Selecting the right action state in medication order data |
Selecting the right timestamp | ||
Other timestamp-related considerations | Delayed documentation—verbal orders | —a |
Timestamp conversion and formatting | — | |
Visualizing time series data to understand temporal patterns | Raw time data table versus data visualization | |
Metadata attributes | Misleading (meta)data labels | Shown to user data column label |
Workflow imprints | Issues that affect performance and capabilities of algorithms | Patient deterioration in the Neonatal Intensive Care Unit necessitates verbal orders |
Delayed action on active orders | — | |
Priming pumps: Speeding up infusion pump rates to prime may look like an error, but has no clinical consequence | — | |
Unstructured data entry | Complexity of human language | Free text dosing of Total Parenteral Nutrition |
Heterogeneity of human language | — | |
Fusing datasets and the role of device integration | Merging datasets from multiple sources requires valid linking identifiers | The nonintegration of smart pumps with electronic health records |
Clinical decision support blind spots | Use of smart infusion pump drug libraries | |
Retrospective versus prospective data or detection | Retrospective data and real-time data are processed and accessed differently | The order audit modification issue |
Technical versus clinical validity | — | — |
aNot applicable.