Yale University's Global Health & Innovation Conference is the world's leading and largest global health conference. The organizers' mission is to facilitate conversation that could improve health on a global scale. We were honored when they asked us to present our readmission reduction predictive model.
Here we are with all the brains at Yale. (See what we did there?) We were asked to present after Yale reviewed our readmission reduction abstract. Our data science team assembled this abstract to study the validity of our readmission reduction solution, Medalogix Touch. Vanderbilt University then peer reviewed our study and gave it the stamp of approval. If you're a proud data or healthcare geek, here's the aforementioned abstract for your reading pleasure.
READMISSION RISK PREDICTIVE MODEL REVIEW
Bryan Mosher, Medalogix LLC, Nashville, TN Reviewed by Christine Lai, Ph.D., Vanderbilt University
Purpose: Amidst pay for performance healthcare initiatives, many healthcare organizations are exploring various technologies to improve care and decrease costs. One specific challenge post acute organizations are turning to technology to solve is hospital readmissions.
Methodology: In our abstract example, the model is sourced by data from a standardized home health admissions questionnaire, OASIS C. The model’s time window is 1/1/2010 to 3/31/2013. There are a total of 8,842 patients with 13,281 episodes from care provided by 6 home health agencies in Ohio. We estimate a 60-day transfer risk using generalized linear models.
Results: The model identified a set of 8 significant risk factors for the 60-day transfer. The biggest risk factor is the patient’s Status and Diagnosis Risk (a combination of health condition and severity attributes) with odds ratio of 1.208. The next is the patient’s History of Transfer with odds ratio of 1.203. Patients with either the presence of Status and Diagnosis Risk or a History of Transfer are nearly 20% more likely to transfer within a 60-day window of care. Patients that have Joint Replacement Therapy Need were 30% less likely to transfer. Overall the variable effect sizes were relatively stable as suggested within the 95% CI of the odds ratios per the bootstrapping distribution. In addition, the model performance was stable having the area under the ROC curve (c statistics) centered on 0.761.
Conclusion: Home health agencies can use a predictive model to design and implement clinical interventions to reduce preventable and costly hospital readmissions. This model can be simulated in various care settings to identify at-risk patients, schedule interventions, improve care and reduce costs.
After our abstract was selected to be featured at Yale's Global Innovation Conference, our lead data scientist, Michael Faron, developed the academic poster previewed below.