An imputation method for calculating and comparing autoimmune disease incidence using partial case review. Issue 48 (4th December 2017)

• Record Type: Journal Article
• Authors: Slezak, Jeff; Meyer, Kristin; Sy, Lina S.; Chao, Chun; Takhar, Harpreet; Ackerson, Brad; Cheetham, T. Craig; Jacobsen, Steven
• Is Part Of: Vaccine. Volume 35:Issue 48(2017)Part B
• Page Start: 6672
• Page End: 6675
• Publication Date: 2017-12-04
• Subjects: AD autoimmune disease -- CI confidence interval -- EHR electronic health record -- IR incidence rate -- IRR incidence rate ratio; Incidence -- Review of reported cases -- Data accuracy -- Epidemiologic methods; Vaccines -- Periodicals; 615.372
• DOI: 10.1016/j.vaccine.2017.10.022
• Languages: English
• ISSNs: 0264-410X

Abstract

Highlights: Accurate assessment of disease incidence can be difficult and costly. Partial case review combined with missing data methods may be a solution. Multiple imputation methods can provide estimates of true incidence rates. Abstract: Purpose: Estimate incidence of autoimmune conditions in a population who received HPV4 vaccine and a comparison unvaccinated population. Electronic health record (EHR) data may contain inaccurate or incomplete coding, while manual chart review of all cases may not be feasible. We propose a method to estimate incidence using EHR data and case review for a sample. Methods: Suspected incident cases were identified using ICD-9 codes, laboratory results and medications related to the condition. A random sample of charts was reviewed to confirm the diagnosis and determine disease onset date. Multiple imputation, using a Monte Carlo model including age and disease indicators was used to impute case status of non-reviewed cases. Incidence rate was calculated in each imputed dataset, with median and percentiles giving a distribution for the estimated incidence rate. Sensitivity analyses compared modeled results to results without imputation and results where imputation was applied to the subset of cases identified using specific ICD-9 codes. Results: The model accounted for differential case confirmation rates by age and method of case identification, identifying a potential safety signal that was missed relying on EHR data alone. Conclusions: This method may be useful for computing incidence when full case review is not feasible.