Deriving ICD-10 Codes for Patient Safety Indicators for Large-scale Surveillance Using Administrative Hospital Data

Existing administrative data patient safety indicators (PSIs) have been limited by uncertainty around the timing of onset of included diagnoses.

We undertook de novo PSI development through a data-driven approach that drew upon "diagnosis timing" information available in some countries' administrative hospital data.

Administrative database analysis and modified Delphi rating process.

All hospitalized adults in Canada in 2009.

We queried all hospitalizations for ICD-10-CA diagnosis codes arising during hospital stay. We then undertook a modified Delphi panel process to rate the extent to which each of the identified diagnoses has a potential link to suboptimal quality of care. We grouped the identified quality/safety-related diagnoses into relevant clinical categories. Lastly, we queried Alberta hospital discharge data to assess the frequency of the newly defined PSI events.

Among 2,416,413 national hospitalizations, we found 2590 unique ICD-10-CA codes flagged as having arisen after admission. Seven panelists evaluated these in a 2-round review process, and identified a listing of 640 ICD-10-CA diagnosis codes judged to be linked to suboptimal quality of care and thus appropriate for inclusion in PSIs. These were then grouped by patient safety experts into 18 clinically relevant PSI categories. We then analyzed data on 2,381,652 Alberta hospital discharges from 2005 through 2012, and found that 134,299 (5.2%) hospitalizations had at least 1 PSI diagnosis.

The resulting work creates a foundation for a new set of PSIs for routine large-scale surveillance of hospital and health system performance.