Validation and Comparison of Psoriasis Case-Finding Algorithms Using International Classification of Diseases Nosology

Abstract

Background

The methodology of utilizing ICD nosology to identify psoriasis case cohorts within an EMR database has not been validated extensively.

Objective

To appraise the validity of the psoriasis case definition utilizing ICD-9 nosology and to compare performances of six distinct algorithms.

Design

Retrospective chart review of a randomly selected subset (n=966) of 2,737 cases with ≥1 ICD-9 diagnosis code for psoriasis in 2004–2013.

Results

Positive Predictive Values for the six algorithms ranged from 84.7% {Any Provider (AP) ≥ 1} to 97.7% {Dermatologist (Derm) ≥ 3}. Negative Predictive Values for the six algorithms evaluated ranged from 24.1% (‘Derm ≥ 3’) to 80.9% (‘Derm≥1’). NPV for ‘Derm≥1’ was 80.9%, while NPV for all other algorithms was ≤ 35.1%. Percent misclassification for the six algorithms ranged from 11.0% (‘Derm≥ 1’) to 46.3% (‘Derm≥ 3’). Increased frequency of code applied also resulted in corresponding significant increases in percent misclassification. Within class, percent misclassification for ‘Derm ≥ 1’ (11.0%) was significantly lower than for ‘AP≥ 1’ (15.3%, P<0.01).

Conclusions

At least one ICD-9 code applied by a dermatologist represents the strongest performing algorithm to identify psoriasis cases.