The validity of International Classification of Diseases, Ninth Revision, Clinical Modification diagnosis codes for identifying patients hospitalized for COPD exacerbations

Abstract

Background: Acute exacerbations of COPD (AE-COPD) are a leading cause of hospitalizations in the United States. To estimate the burden of disease (eg, prevalence and cost), identify opportunities to improve care quality (eg, performance measures), and conduct observational comparative effectiveness research studies, various algorithms based on the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes have been used to identify patients with COPD. However, the validity of these algorithms remains unclear.

Methods: We compared the test characteristics (sensitivity, specificity, positive predictive value, and negative predictive value) of four different coding algorithms for identifying patients hospitalized for an exacerbation of COPD with chart review (reference standard) using a stratified probability sample of 200 hospitalizations at two urban academic medical centers. Sampling weights were used when calculating prevalence and test characteristics.

Results: The prevalence of COPD exacerbations (based on the reference standard) was 7.9% of all hospitalizations. The sensitivity of all ICD-9-CM algorithms was very low and varied by algorithm (12%-25%), but the negative predictive value was similarly high across algorithms (93%-94%). The specificity was > 99% for all algorithms, but the positive predictive value varied by algorithm (81%-97%).

Conclusions: Algorithms based on ICD-9-CM codes will undercount hospitalizations for AE-COPD, and as many as one in five patients identified by these algorithms may be misidentified as having a COPD exacerbation. These findings suggest that relying on ICD-9-CM codes alone to identify patients hospitalized for AE-COPD may be problematic.

Figure 1.

Patient selection. Algorithms 1 (expanded COPD algorithm), 2 (primary COPD or respiratory failure codes), 3 (multiple primary COPD codes), and 4 (single primary AE-COPD code) are shown. **Excluded if 491.20 (obstructive chronic bronchitis without exacerbation) in the primary or secondary diagnosis positions. AE-COPD = acute exacerbation of COPD; ICD-9-CM = International Classification of Disease, Ninth Edition, Clinical Modification.

Figure 2.

Test characteristics of ICD-9-CM algorithms for identifying AE-COPD. Algorithms 1 (expanded COPD algorithm), 2 (primary COPD or respiratory failure codes), 3 (multiple primary COPD codes), and 4 (single primary AE-COPD code) are shown. Records were weighted by 1/probability of being sampled. The circle or square represents the estimated test characteristic, and whiskers represent the 95% CI. The 95% CI for specificity ranged between 99% and 100% for all algorithms. All pairwise comparisons of sensitivity between algorithms were statistically significant (P < .05), except for algorithms 1 vs 2 (P = .40). All differences in the NPV between algorithms were significant (P < .05) except for the difference between algorithms 1 and 2 (P = .51). Differences in specificity were significant for algorithms 1 vs 3, algorithm 1 vs 4, and algorithm 2 vs 4 (all P < .05); other pairwise comparisons were not significant. The PPV of algorithm 4 was significantly better than that of algorithm 1 (P = .01) and algorithm 2 (P = .04); other pairwise comparisons were not significant. NPV = negative predictive value; PPV = positive predictive value. See Figure 1 legend for expansion of other abbreviation.

Figure 3.

Test characteristics of ICD-9-CM algorithms for identifying AE-COPD (restrictive reference standard). Algorithms 1 (expanded COPD algorithm), 2 (primary COPD or respiratory failure codes), 3 (multiple primary COPD codes), and 4 (single primary AE-COPD code) are shown. Records were weighted by 1/probability of being sampled. The circle or square represents the estimated test characteristic, and whiskers represent the 95% CI. The 95% CI ranged from 98% to 99% for specificity and 99% to 100% for NPV for all algorithms. See Figure 1 and 2 legends for expansion of abbreviations.