henry domenico vanderbilt university medical center
TRANSCRIPT
Biostatistics in Quality Improvement
Henry DomenicoVanderbilt University Medical Center
Guidelines for a Data Driven Project◦ What can Biostatistics do for a Project?◦ Defining a Question◦ Data Collection◦ Analysis◦ Conclusion
Readmissions at Vanderbilt◦ Background◦ Identifying Important Factors◦ Predicting the Probability of Readmission◦ Future Work
Outline
Statistics is a powerful tool.
Statistics should be part of improving the care we give, not just for research.
My focus is on using statistics to…◦ Answer questions about quality issues◦ Understand where we are and where we’re going◦ Identify the driving force behind the issues we face◦ Develop new strategies for improving care◦ Understand the effectiveness of an intervention
Statistics in Clinical Improvement
Understand your population Decide if previously held notions are correct Discover what factors are important to your
problem Make Predictions Test an Intervention Present your findings in a convincing way
Wealth of data collected at Vanderbilt
By Utilizing Data You Can…
Biostatistics can be built into each stage of a project.◦ Question
◦ Data Collection
◦ Analysis
◦ Conclusions
Data Driven Improvement
You can make your life easier by starting with a carefully defined question.◦ What factors lead to increased patient satisfaction?
Clearly identify the population.
What are the possible factors?
How will the response be measured?
What is a clinically significant result?
Defining a Question
Data should be collected from the population of interest.
Ideally would like to have any variable that effects the response.
Observational or Experimental?
At what level should the data be gathered?
A lot of effort can go into gathering data that won’t answer your question.
Data Collection
If the question is carefully defined and data is gathered correctly, analysis becomes the easy part.
Identify important factors.
Determine statistical/clinical significance.
Account for confounding factors.
Analysis
We want to present our results in a way that is convincing to others.
Use the results of the analysis to present a clear picture of what is occurring.
Directly answer the original question.
Conclusions
Hospital Readmission rate is being viewed as a quality metric.
In the near future, Vanderbilt will see financial penalties for patients discharged and readmitted within 30 days.
Readmissions at Vanderbilt
Readmission at Vanderbilt
Can we identify which patients are at risk of being readmitted?
Can we model a patient’s probability of being readmitted within 30 days?
Can we present this information to providers at the point of decision?
Is there an intervention that will reduce the probability that a patient is readmitted?
Readmissions at Vanderbilt
Working with a data set of all 2009 Inpatients.
Created a readmission flag.
Demographic and diagnostic variables are included.
Missing lab and vital sign information.
Data Collection
Identifying Important Factors
Identifying Important Factors
Can examine each DRG’s readmission rate.
Several have only one observation.
May be more meaningful to examine which had a higher number of readmissions than average.
Which DRGs Have High Readmissions?
DRG Description Residual
280 AMI Disc. Alive 6
847 Chemo 5.5
293 Heart Failure 5.1
291 Heart Failure 4.5
216 Cardiac Valve 3.6
More than Expected DRG
Used same method to determine:
Patients from Davidson County had a higher readmission rate.
ICD-9’s 780( Malaise and Fatigue) and 789.09(Symptoms involving abdomen and Pelvis) had higher readmission rates.
Patients on Blue Cross/Blue Shield were at higher risk.
Other Factors
Logistic regression models the probability of readmission based on a patient’s explanatory variables.
Used logistic regression to model the odds ratio for different factors.
The odds ratio tells us the increased probability of readmission associated with these factors.
Logistic Regression
Again using logistic regression, we can develop a model that will provide each patient’s readmission probability.
Specify which variables we want to use to make predictions.
Use a statistical software package to build a model.
Modeling a Patient’s Readmission Probability
Model a patient’s probability of readmission based on:◦ Age◦ Length of Stay◦ Number of Medications◦ DRG
Model should only be applied to patients from same population used to build the model.
Demonstration
Alternative non-parametric approach to logistic regression.
Breaks population into subsets and identifies factors important to each subset.
Provides predicted probabilities like logistic regression.
Tree Based Model
Tree Based Model
Our goal is to be able to inform providers of a patient’s probability of readmission.
Let them know what factors need to be addressed before discharge.
Goal is to reduce preventable readmissions.
Providing Info at POD
Obtain a more comprehensive data set.◦ All admitted patients◦ Gender◦ Lab Values◦ Vital Signs◦ Readmission Status
Develop specific models for individual DRGs.
Future Work
Eventual goal is to work with subject matter experts to develop an intervention that reduces readmissions.
Show effectiveness using a Randomized Controlled Trial.
Testing an Intervention
This is just one example of how statistics can improve a project.
Hopefully demonstrates the value of being data driven and using statistics.
Biostatistics Free Clinic
Conclusions
Thank You
Questions?