team 6 - unitypoint health powerpoint

Post on 14-Apr-2017

92 Views

Category:

Documents

1 Downloads

Preview:

Click to see full reader

TRANSCRIPT

Team #6Indra Kiran Chowdavarapu

Brittany CopponexJohn HardenAjay Mohan

The Issue at HandProblem

If UPH hospitals reach a certain readmission threshold, CMS will apply a penalty to insurance payments remitted to UPH hospitals for all Medicare/Medicaid patients ranging from 0-3%

ObjectiveProtocol must be developed that limits unplanned 30-day readmissions by providing (a) high-quality care and (b) financial viability

Deliverable 1-Scoring patient risk-Methodology

Deliverable 2-Cost of care plans-Effect of care plans

Deliverable 3-Protocol

Deliverable 4-GUI-Implementation plan

Analytics Framework

Data Preparatio

n

Input Selection

Regression Analysis

Model Selection

Deliverable 1 Deliverable 2 Deliverable 3 Deliverable 4

Challenges with the Data• Skewed data, outliers and extreme values

Transformed variables to bring skewness closer to zero• Categorical data analysis

Categorized data into levels and created dummy variables • Dimensionality of data and collinearity among variables

Cluster analysis, variable selection • Over-fitting and over-sampling

Data partition, stratified sampling

Deliverable 1 Deliverable 2 Deliverable 3 Deliverable 4

Why Logistic Regression?Parameter readmit30 Estimate

Chi-Square Pr > ChiSq Estimate

Intercept 1 -0.5748 0.95 0.0049LOG_ed_visits 1 1.7203 9.6 0.0019 0.1104LOG_los 1 1.2282 5.18 0.0228 0.0851SQRT_meds_cardio_agents 1 1.1313 4.13 0.0421 0.0715LOG_meds_neuromuscular 1 1.175 4.35 0.0369 0.0742SQRT_meds_endocrine 1 0.6852 5.78 0.0162 0.0924diabetes 0 1 -0.746 8.73 0.0031hemoglobin 1 -0.0823 5.4 0.0201 -0.0936icu_yn 0 1 -0.5936 16.08 <.0001

Prob(Readmit30=1) = -0.5748 + 1.7203* LOG_ed_visits + 1.2282 *LOG_los + 1.1313* SQRT_meds_cardio_agents + 1.175*LOG_meds_neuromuscular + 0.6852*SQRT_meds_endocrine - 0.746*diabetes -0.0823*hemoglobin-0.5936*icu_yn

Deliverable 1 Deliverable 2 Deliverable 3 Deliverable 4

Model Selection Criteria• Misclassification rates – training and validation• Average squared errors – training and validation• Sensitivity• Specificity• AIC & BIC values

Alternative Methodology• Linear Regression for estimating readmission costs

Deliverable 1 Deliverable 2 Deliverable 3 Deliverable 4

Moving Forward• Infrastructure integration• Phased launch/test• Workflow implementation• Evaluation and expansion

• Alternative implementation

Deliverable 1 Deliverable 2 Deliverable 3 Deliverable 4

Thank You!

top related