Validation of administrative data used for the diagnosis of upper gastrointestinal events following nonsteroidal anti-inflammatory drug prescription

Authors

  • N. S. ABRAHAM,

    1. Houston Center for Quality of Care and Utilization Studies, Section of Health Services Research
    2. Section of Digestive Diseases, Michael E. DeBakey VA Medical Center
    3. Baylor College of Medicine, Houston, TX, USA
    Search for more papers by this author
  • D. C. COHEN,

    1. Houston Center for Quality of Care and Utilization Studies, Section of Health Services Research
    2. Baylor College of Medicine, Houston, TX, USA
    Search for more papers by this author
  • B. RIVERS,

    1. Houston Center for Quality of Care and Utilization Studies, Section of Health Services Research
    2. Baylor College of Medicine, Houston, TX, USA
    Search for more papers by this author
  • P. RICHARDSON

    1. Houston Center for Quality of Care and Utilization Studies, Section of Health Services Research
    2. Baylor College of Medicine, Houston, TX, USA
    Search for more papers by this author

Dr N. S. Abraham, Michael E. DeBakey VA Medical Center, 2002 Holcombe Blvd. (152), Houston, TX, USA.
E-mail: nabraham@bcm.tmc.edu

Abstract

Summary

Aims

To validate veterans affairs (VA) administrative data for the diagnosis of nonsteroidal anti-inflammatory drug (NSAID)-related upper gastrointestinal events (UGIE) and to develop a diagnostic algorithm.

Methods

A retrospective study of veterans prescribed an NSAID as identified from the national pharmacy database merged with in-patient and out-patient data, followed by primary chart abstraction. Contingency tables were constructed to allow comparison with a random sample of patients prescribed an NSAID, but without UGIE. Multivariable logistic regression analysis was used to derive a predictive algorithm. Once derived, the algorithm was validated in a separate cohort of veterans.

Results

Of 906 patients, 606 had a diagnostic code for UGIE; 300 were a random subsample of 11 744 patients (control). Only 161 had a confirmed UGIE. The positive predictive value (PPV) of diagnostic codes was poor, but improved from 27% to 51% with the addition of endoscopic procedural codes. The strongest predictors of UGIE were an in-patient ICD-9 code for gastric ulcer, duodenal ulcer and haemorrhage combined with upper endoscopy. This algorithm had a PPV of 73% when limited to patients ≥65 years (c-statistic 0.79). Validation of the algorithm revealed a PPV of 80% among patients with an overlapping NSAID prescription.

Conclusions

NSAID-related UGIE can be assessed using VA administrative data. The optimal algorithm includes an in-patient ICD-9 code for gastric or duodenal ulcer and gastrointestinal bleeding combined with a procedural code for upper endoscopy.

Ancillary