Abstract Details

Title Effect of Neurological Manifestations on SARS-CoV-2 Infection Prognosis using Machine Learning Models

Topic General Neurology

Presentation(s) General Neurology Posters (7:00 AM-5:00 PM)

Poster/Presentation
Number 019

Objective

Investigate the effect of neurological manifestations on the prognosis of SARS-CoV-2 infection.

Background

Mounting evidence suggests that neurological manifestations of SARS-CoV-2 infection are associated with poorer prognosis.

Design/Methods Using the Cerner Real-World Data dataset, we investigated outcomes in COVID-19 patients with and without neurological complications. We implemented a preliminary risk stratification machine learning (ML)-based logistic regression model to predict the likelihood of one of the following events: death, discharge to hospice, and discharge to skilled nursing facilities (SNF) using neurological complications and other clinical data.

Results Our logistic regression model resulted in at least 87% accuracy and 84% area under the receiver operator characteristic curve. Among the 25,261 COVID-19 and 92,235 non-COVID-19 patients, 11% vs. 14% developed neurological complications (COVID-19N), respectively. The average length of stay in days (LoS) and mortality rate were significantly higher in COVID-19N compared to non-COVID-19N and COVID-19 without neurological complications (COVID-19WN), respectively 7.5, 12% vs. 2.95, 5.5% vs. 4, 7% (p<0.05). The rate of home discharge was significantly lower in the COVID-19N compared to non-COVID-19N and COVID-19WN, respectively 52% vs. 88% vs. 68% (p<0.05). The rate of discharge to a hospice was significantly higher in the COVID-19N compared to non-COVID-19N and COVID-19WN, respectively 5.5% vs. 2.6% vs. 3.0% (p<0.05). The rate of discharge to skilled nursing facilities (SNF) was significantly higher in the COVID-19N than non-COVID-19N and COVID-19WN, respectively 11.5% vs. 10.6% vs. 6% (p<0.05).

Conclusions

Our study demonstrates that neurological manifestations of COVID-19 infection are associated with significantly increased mortality, length of hospitalization, and discharges to hospice and SNF. ML models are highly accurate in predicting the likelihood of death and discharge destination. Ongoing work will investigate specific factors for poor prognosis for individual neurological complications and refine ML-based models by tuning their hyperparameters, adding performance metrics, and validating models.

Authors/Disclosures
Parisorn Thepmankorn (Rutgers New Jersey Medical School) PRESENTER	Ms. Thepmankorn has received personal compensation for serving as an employee of Johnson and Johnson.
Keyvan Heshmati, MD	Dr. Heshmati has nothing to disclose.
	No disclosure on file
	No disclosure on file
	No disclosure on file
	No disclosure on file
Nizar Souayah, MD, FAAN (NJMS)	Dr. Souayah has received personal compensation in the range of $500-$4,999 for serving on a Speakers Bureau for Takeda. Dr. Souayah has received publishing royalties from a publication relating to health care.

��ɫ��

��ɫ��