Abstract Details

Title Identification of Patients with Incident Amyotrophic Lateral Sclerosis Using ICD-10-CM Codes in Claims Data

Topic Global Health and Neuroepidemiology

Presentation(s) P7 - Poster Session 7 (8:00 AM-9:00 AM)

Poster/Presentation
Number 20-005

Objective To develop a predictive model to identify incident amyotrophic lateral sclerosis (ALS) cases using Medicare data.

Background ALS is a rapidly progressive and nearly universally fatal neurodegenerative disease. Delayed diagnosis results in unnecessary morbidity and early mortality in many patients. We sought to develop a claims-based predictive model using a large, population-based Medicare dataset.

Design/Methods We used age-eligible Medicare beneficiaries to identify all incident ALS patients in 2017 and 2018 (N=8,050) and randomly selected controls (N=32,200) using a 1:4 ratio frequency matched on year and month of diagnosis. We used Medicare data from 2015-2018 to create predictor variables for demographics (age, sex, race/ethnicity), a measure for use of care, and all ICD-10-CM diagnosis codes assigned by beneficiaries from October 2015 up to the beneficiary diagnosis or control reference date. We randomly divided the data into 80% training and 20% validation and then used the training data to run a penalized regression model (elastic net) to identify demographic and diagnosis predictors of ALS-case status. We selected the two hyper-parameters (alpha and lambda) using five-fold cross validation. The predictive accuracy of the model was determined using area under the receiver operating characteristic curve (AUC) and 95% confidence intervals (CIs).

Results There were 4,462 ICD-10-CM diagnosis codes observed among our cases and controls. The AUC of a model with only demographic characteristics and use of care was 0.705 (95% CI: 0.694-0.717). Our elastic net model with demographics and diagnosis codes resulted in an AUC in the validation set of 0.903 (95% CI: 0.893-0.913) using the best alpha of 0.3 and 735 predictors. For the optimal cut point for the model, the sensitivity was 79% and specificity was 88%.

Conclusions A claims-based, ALS predictive model using only demographic characteristics and diagnosis codes had excellent discriminability and may be useful to facilitate earlier diagnosis of ALS patients.

Authors/Disclosures
Timothy R. Fullam, MD, FAAN PRESENTER	Dr. Fullam has received personal compensation in the range of $500-$4,999 for serving on a Speakers Bureau for Amylyx. Dr. Fullam has received publishing royalties from a publication relating to health care. Dr. Fullam has received personal compensation in the range of $500-$4,999 for serving as a ALS CDRMP Programmatic Panel with Department of War.
Jordan A. Killion, PhD (Barrow Neurological Institute)	Dr. Killion has received personal compensation for serving as an employee of CommonSpirit Health. The institution of Dr. Killion has received research support from The Michael J. Fox Foundation for Parkinson's Research (MJFF-000939). The institution of Dr. Killion has received research support from Department of Defense Grant (PD190057). The institution of Dr. Killion has received research support from Barrow Neurological Foundation. The institution of Dr. Killion has received research support from Kemper and Ethel Marley Foundation. The institution of Dr. Killion has received research support from Moreno Family.
Brad A. Racette, MD, FAAN (Barrow Neurological Institute)	Dr. Racette has received personal compensation in the range of $5,000-$9,999 for serving as a Consultant for American Regent. Dr. Racette has received personal compensation in the range of $500-$4,999 for serving as a advisory council with NIEHS.

��ɫ��

��ɫ��