Abstract
Purpose :
Applying singleton codes from electronic health records (EHR) to distinguish cases from controls for specific disorders leads to inaccurate taxonomy. To accurately detect keratoconus (KC) cases we created an EHR-based algorithm using a combination of variables to reduce error. The relationship between KC cases and controls and demographic, lifestyle and comorbid factors was examined in the Million Veteran Program (MVP).
Methods :
The KC algorithm was derived from International Classification Disease Clinical Modification (ICD-CM) and Current Procedural Terminology (CPT) codes in the Veteran’s Administration (VA) Computerized Patient Health Records. The algorithm was refined at the Cleveland VA and then exported to Buffalo and Providence VAs. Final positive and negative predictive values (PPV, NPV) were calculated across sites. We extracted cases and controls, and evaluated the relationship between KC with sex, genetic ancestry, smoking history, and comorbidities in the MVP.
Results :
Chart reviews at Cleveland, Buffalo, and Providence yielded an average PPV and NPV of 91% and 98%, respectively. We identified 1,080 KC cases and 184,007 controls. Multivariable adjusted KC models included age, sex, ancestry, and smoking or Charlson Comorbidity Index (CCI). We observed that women are at higher risk than men (odds ratio (OR)= 2.6) with a 3.1% diagnosis in women compared to 0.5% in men (p-value (P)< 9.0 x 10-7). By ancestral group, African Americans (AA) were at greater risk for KC vs European Americans (EA) (OR= 1.7; P<1.0x 10-5). Hispanic Americans (HA) (OR= 1.5) and Asian Americans (ASN) (OR= 0.76) showed no association, but the direction of ORs was consistent with recent reports. A history of smoking reduced risk of KC (OR= 0.51; P< 3.4x10-9). KC cases have greater comorbidities vs controls, OR= 2.7 (P<3.6x10-170) per one step increase in CCI.
Conclusions :
The greatest risk for KC was in women and AAs. KC is rare and only 10% of the VA population is female, thus the disparity in male vs female risk compared to other studies needs to be investigated more thoroughly. Smoking was protective against KC, and prior reports speculate that cigarette smoke protects by crosslinking corneal proteins. In MVP, KC cases have more comorbidities than controls. Our validated EHR-based KC algorithm is portable and will be valuable for analyses of datasets from other health systems.
This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.