Table 1

Data validation issues and solutions

ClassVariation in data entryEffect on measuresSolutions to minimise distortion of measures
Identification of ICU patientsCCU patients (PTS) were classified as cardiology treating specialty not as ICU PTSCCU patients were not included in ICU cohort, Distorted image of ICU careAdd CCU to ICU treating specialty fields, scan each ICU quarterly for variation in proportion of major diagnoses benchmarked to type of ICU
SICU PTS after discharge to medical step down unit electronically remained in SICU treating specialty (to track workload)Treating specialty dates determined length of stay (LOS) in ICU, this prolonged SICU length of stayAdd surgical step down to treating specialty fields. Monitor with # patient days/maximal # days (>100% is impossible)
PTS in a step-down ward were assigned an ICU treating specialtyArtificial reduction in severity of illness in ICU cohort that included step-down patients1. Communicate with local ICU managers to validate each individual ward assigned to an ICU treating specialty. 2. Track quarterly variation by ICU in severity of illness
PTS were electronically moved out of the ICU location when ready for ward but physically remained in ICU when no acute care bed was available.Artificial shortening or length of stay1. Use two sources to determine LOS, one with treating specialty, the other with ward location. Feedback differences to local leadership and managers
Duplicate deathsPTS readmitted and died within 24 h of index hospital discharge may be counted dead twice.Inflates risk adjusted and unadjusted mortality ratesAdd screening for duplicate deaths
Capture of medication ordersMedications orders are captured by dispensed status. Those meds dispensed from ward stock were invisible (subq heparin, insulin)Underestimates adherence to deep venous prophylaxis, treatment for hyperglycemiaRevise extraction program
Satellite pharmacies used different procedures for completing ordersSameUse data from bar coded medication administration
Laboratory data locationLocation of lab values varied across the countryIn the risk model, a normal value is inserted for ‘unmeasured’ labs–could underestimate severity of illness, or adherence (proportion of patients with therapeutic INR on Coumadin)Map initial laboratory value locations, track proportion of missing labs for each lab value for each quarterly report
Change in lab value location occurs with addition of new reagents and new testing machines commonlyDeveloped system to allow identification of changes in the laboratory maps each quarter
Laboratory normal valuesNormal range of newer laboratory tests varies (ex. Troponin 10% coefficient of variation ranges from 0.03 to >5)Inability to use laboratory test in risk modelCategorise data (troponin normal, high, abnormal)
Determine lab test manufacturer quarterly
Classify type of instrument (point of care)
  • CCU, cardiac care unit; SICU, surgical care unit; ICU, intensive care unit; PTS, patients.