Validity of claims-based diagnoses for infectious diseases common among immunocompromised patients in Japan

Area of Science:

• Health Informatics
• Epidemiology
• Infectious Diseases

Background:

• Claims data are increasingly used for disease surveillance and research.
• Validating algorithms that identify specific diseases from claims data is crucial for accuracy.
• This study focused on herpes zoster (HZ), Mycobacterium tuberculosis (MTB), nontuberculous mycobacteria infections (NTM), and Pneumocystis jirovecii pneumonia (PJP) in Japan.

Purpose of the Study:

• To validate Japanese claims-based disease-identifying algorithms for HZ, MTB, NTM, and PJP.
• To assess the positive predictive values (PPVs) of these algorithms for both prevalent and incident cases.

Main Methods:

• A multicenter, cross-sectional, retrospective study (VALIDATE-J) was conducted using claims data and medical records from two Japanese hospitals.
• Algorithms were developed to identify HZ, MTB, NTM, and PJP cases treated between 2012-2016.
• Diagnosis confirmation used three gold standard definitions, and PPVs were calculated for prevalent and incident cases.

Main Results:

• PPVs for herpes zoster (HZ) ranged from 67.4-70.9% and for Mycobacterium tuberculosis (MTB) from 67.0-90.0% across definitions and case types.
• Positive predictive values for nontuberculous mycobacteria infections (NTM) were 18.3-63.4% and for Pneumocystis jirovecii pneumonia (PJP) were 20.0-45.0%.
• Adding treatment information to algorithms improved PPVs for HZ and, to a lesser extent, for prevalent NTM cases.

Conclusions:

• Claims-based algorithms demonstrated moderate to high positive predictive values for identifying herpes zoster (HZ) and Mycobacterium tuberculosis (MTB) in Japan.
• The algorithms showed lower performance for identifying nontuberculous mycobacteria infections (NTM) and Pneumocystis jirovecii pneumonia (PJP).
• Algorithm refinement, such as incorporating treatment data, can enhance accuracy for specific conditions.