Identification of smoking using Medicare data--a validation study of claims-based algorithms

Area of Science:

• Health Informatics
• Epidemiology
• Medical Record Analysis

Background:

• Accurate identification of smoking status is crucial for epidemiological research and clinical studies.
• Self-reported smoking data is the gold standard but can be subject to recall bias.
• Claims data offers a potentially scalable alternative for identifying smoking behavior in large patient populations.

Purpose of the Study:

• To evaluate the accuracy of Medicare claims-based algorithms in identifying current smokers.
• To compare algorithms using diagnosis/procedure codes versus those including anti-smoking prescriptions.
• To assess algorithm performance using different look-back periods for claims data.

Main Methods:

• Two claims-based algorithms were developed to identify smoking status in Medicare beneficiaries.
• Algorithm 1 utilized only diagnosis and procedure codes; Algorithm 2 included these plus anti-smoking prescriptions.
• Performance was assessed against self-reported smoking data (gold standard) using sensitivity, specificity, NPV, PPV, and AUC.

Main Results:

• Claims-based algorithms demonstrated high specificity (100%) but low sensitivity (ranging from 9.8% to 27.9%).
• Incorporating pharmacy claims and using all available pre-index data improved sensitivity, NPV, and AUC compared to limited data.
• The algorithm using diagnosis/procedure codes and all pre-index claims had an AUC of 0.64.

Conclusions:

• Claims-based algorithms can identify smokers with high specificity, though sensitivity remains a limitation.
• These algorithms may be cautiously considered for identifying smoking in observational studies when self-report data is unavailable.
• Further refinement of claims-based algorithms could enhance their utility in epidemiological research.