The Use of Funnel Plots & Multi- Year Cumulative Data to Track Hospital Performance Herbert MA, Hamman BH, Roper KL, Ring WS, Edgerton JR, Texas Quality.

Slides:



Advertisements
Similar presentations
How would you explain the smoking paradox. Smokers fair better after an infarction in hospital than non-smokers. This apparently disagrees with the view.
Advertisements

Experiments and Variables
Sta220 - Statistics Mr. Smith Room 310 Class #14.
Gall C, Katch A, Rice T, Jeffries HE, Kukuyeva I, and Wetzel RC
Process Control Charts An Overview. What is Statistical Process Control? Statistical Process Control (SPC) uses statistical tools to observe the performance.
APPLICATIONS OF DIFFERENTIATION
Line Plots and Histograms Similar but Different. Line Plots Easy and visual way to organize data. Easy and visual way to organize data. consists of a.
Full time and part time employment Coventry population in employment by gender Source: Annual Population Survey, Office for National Statistics
 There are times when an experiment cannot be carried out, but researchers would like to understand possible relationships in the data. Data is collected.
Slide 1 SOLVING THE HOMEWORK PROBLEMS Simple linear regression is an appropriate model of the relationship between two quantitative variables provided.
Diane Stockton Trend analysis. Introduction Why do we want to look at trends over time? –To see how things have changed What is the information used for?
Quality Assessment 2 Quality Control.
Sharing and explaining the standardized infection ratio (SIR): Does your audience prefer words, colors, and/or δymβφĨs? Dana Burshell, MPH, CPH, CIC HAI.
3 CHAPTER Cost Behavior 3-1.
The Standardized Infection Ratio Steven P Hudson, MBA, MA Statistician Health Care Excel, Inc.
Overview Public Reporting Cardiovascular Data Recommendations.
How do scientists show the results of investigations?
Association between 2 variables We've described the distribution of 1 variable in Chapter 1 - but what if 2 variables are measured on the same individual?
Employment, unemployment and economic activity Coventry working age population by disability status Source: Annual Population Survey, Office for National.
The Argument for Using Statistics Weighing the Evidence Statistical Inference: An Overview Applying Statistical Inference: An Example Going Beyond Testing.
Equations of Lines Chapter 8 Sections
CHAPTER 18: Inference about a Population Mean
 Frequency Distribution is a statistical technique to explore the underlying patterns of raw data.  Preparing frequency distribution tables, we can.
Source: Annual Population Survey, Office for National Statistics. Full time and part time employment Coventry population.
Scientific Method.
Social Science Research Design and Statistics, 2/e Alfred P. Rovai, Jason D. Baker, and Michael K. Ponton Within Subjects Analysis of Variance PowerPoint.
Employment, unemployment and economic activity Coventry working age population by ethnicity Source: Annual Population Survey, Office for National Statistics.
Issues concerning the interpretation of statistical significance tests.
Copyright © 2010 Pearson Education, Inc Chapter Twenty-Three Report Preparation and Presentation.
Association between 2 variables We've described the distribution of 1 variable - but what if 2 variables are measured on the same individual? Examples?
Chapter 8: Simple Linear Regression Yang Zhenlin.
How to Construct a Seasonal Index. Methods of Constructing a Seasonal Index  There are several ways to construct a seasonal index. The simplest is to.
Insert name of presentation on Master Slide HCAI Charts HCAI Information for Action, November 2010 Presenter: Mari Morgan, Wendy Harrison.
Statistical Inference for the Mean Objectives: (Chapter 8&9, DeCoursey) -To understand the terms variance and standard error of a sample mean, Null Hypothesis,
The normal approximation for probability histograms.
Additional Regression techniques Scott Harris October 2009.
Describing Data Week 1 The W’s (Where do the Numbers come from?) Who: Who was measured? By Whom: Who did the measuring What: What was measured? Where:
Scientific Method The 7-step process to scientific investigations.
How to investigate hospital mortality statistics
The Second Patient Report of the National Emergency Laparotomy Audit
Modeling Distributions of Data
Two-Sample Hypothesis Testing
Tennessee Adult Education 2011 Curriculum Math Level 3
Data Analysis of EnchantedLearning.com vs. Invent.org
SIMPLE LINEAR REGRESSION MODEL
Public Reporting of Cardiovascular Data
Regression model Y represents a value of the response variable.
Analyzing Reliability and Validity in Outcomes Assessment Part 1
Gerald Dyer, Jr., MPH October 20, 2016
Cumulative sum techniques for assessing surgical results
CHAPTER 18: Inference about a Population Mean
Frederick L Grover, MD  The Annals of Thoracic Surgery 
Visual Search and Attention
How to Start This PowerPoint® Tutorial
Topic 7: Visualization Lesson 1 – Creating Charts in Excel
Scatter Plot 3 Notes 2/6/19.
Honors Statistics Review Chapters 4 - 5
Quality Control Lecture 3
Volume 46, Issue 5, Pages (May 2007)
James Brevig, MD, Julie McDonald, BSN, Edy S
What Are the Odds? The Annals of Thoracic Surgery
CHAPTER 18: Inference about a Population Mean
Cumulative sum failure analysis for eight surgeons performing minimally invasive direct coronary artery bypass  David M. Holzhey, MD, Stephan Jacobs,
(-4)*(-7)= Agenda Bell Ringer Bell Ringer
GRAPHING MOTION Distance vs. Time.
CHAPTER 18: Inference about a Population Mean
Ruyun Jin, MD, Anthony P. Furnary, MD, Stephanie C. Fine, MA, Eugene H
GRAPHING MOTION Distance vs. Time.
Cardiac surgery report cards: making the grade
GRAPHING MOTION Distance vs. Time.
Presentation transcript:

The Use of Funnel Plots & Multi- Year Cumulative Data to Track Hospital Performance Herbert MA, Hamman BH, Roper KL, Ring WS, Edgerton JR, Texas Quality Initiative The American Association of Thoracic Surgeons April 26, 2015 Seattle, Washington

Nothing to Disclose

The Texas Quality Initiative 27 Hospitals in North Texas agreed to share clinical and administrative data 27 Hospitals in North Texas agreed to share clinical and administrative data All participated in the STS Database All participated in the STS Database 26,634 cardiac procedures from 1/ / ,634 cardiac procedures from 1/ / ,379 isolated CABG were analyzed for observed to expected (O/E) operative mortality 13,379 isolated CABG were analyzed for observed to expected (O/E) operative mortality There was a need to graphically represent the data There was a need to graphically represent the data – Simple – Easy to understand

The Funnel Plot

Methods A funnel plot is centered on a benchmark with 95% confidence intervals drawn on the graph. A funnel plot is centered on a benchmark with 95% confidence intervals drawn on the graph. To assess operative mortality and allow for risk correction, the observed to expected (O/E) ratio is used. To assess operative mortality and allow for risk correction, the observed to expected (O/E) ratio is used. The case volume is plotted on the horizontal axis The case volume is plotted on the horizontal axis O/E ratio on the vertical axis; O/E ratio on the vertical axis; either annual data or multi-year data can be shown. either annual data or multi-year data can be shown.

The Funnel Plot X Axis: Volume of Cases (CABG) Y Axis: O/E Ratio for isolated CABG An O/E of 1 is expected 95% Confidence intervals surround “1”

The Funnel Plot Worse than 1, but not statistically different Better than 1, but not statistically different Outlier for poor performance Outlier for good performance

The Funnel Plot At low volume it is very hard to become an outlier At low volume it is very hard to become an outlier In fact, at less than 200 cases, you cannot become an outlier for good performance In fact, at less than 200 cases, you cannot become an outlier for good performance The Problem: The Problem: Most hospitals analyze their data on an annual basis Most hospitals analyze their data on an annual basis Most Hospitals do less than 200 cases per year Most Hospitals do less than 200 cases per year They cannot reveal They cannot reveal themselves as outlier themselves as outlier due to wide confidence due to wide confidence interval at low volume interval at low volume Year after year they find Year after year they find that their results are “OK” that their results are “OK”

Five Year Data of all Hospitals allows Comparison of Results Five Year Data of all Hospitals allows Comparison of Results Five hospitals are above the upper 95% confidence interval, statistically worse than the target value of 1.0. Five hospitals are above the upper 95% confidence interval, statistically worse than the target value of 1.0. Seven Hospitals O/E is <1, but still within the confidence interval Seven Hospitals O/E is <1, but still within the confidence interval

The Funnel Plot PROBLEM No Problem

How can we account for this problem? Plotting running totals moves the result to the right, where the funnel is narrower Plotting running totals moves the result to the right, where the funnel is narrower The results can begin to show statistically significant differences from “1” The results can begin to show statistically significant differences from “1” More important….. More important….. Trends become visually apparent Trends become visually apparent Poorly performing hospitals can be identified Poorly performing hospitals can be identified – Even before the results reach statistical significance – Urgent interventions can be put in place

Plotting Running 5 Year Totals Plotting Running 5 Year Totals Year 1 Year Year 1+ 2 Year Year

Plotting Running 5 Year Totals  Year 6 2 Year 2 Year Year 2+ 3 Year Year

How Does This Help Let’s see some examples from the TQI Data Let’s see some examples from the TQI Data These are real data from real hospitals These are real data from real hospitals Some of the examples are from different time intervals, because… Some of the examples are from different time intervals, because… I Selected graphs to illustrate different scenarios I Selected graphs to illustrate different scenarios Colored dots represent annual data Colored dots represent annual data Green line represents the running 5 year total Green line represents the running 5 year total

Even with excellent outcomes, a hospital with case volumes under 200 cannot become an outlier for good performance However, Cumulative data will reveal excellence (in 1 more year)

Annual Data tightly clustered: O/E doesn’t change much With Cumulative Data … The Curve is flat, but at higher volume becomes an outlier

Annual Data is all within the funnel Hospital perceives “No Problem” Cumulative Data Unmasks Outlier for Poor Performance

Annual Data is Inconclusive: 3 out of 5 years are within the funnel Cumulative Data  Slope of the Curve is Predictive of Poor Performance At Year 3 Intervention is Needed, This is even before hospital becomes an outlier in year 5

Annual Data: 4 of 5 years O/E is > 1 Cumulative Data Shows a downward slope We have no concerns about this hospital

Conclusions The use of funnel plots allows easy comparison of individual programs The use of funnel plots allows easy comparison of individual programs Analyzing only annual data can lead to a false sense of satisfaction Analyzing only annual data can lead to a false sense of satisfaction The plotting of a five year running total will provide sufficient volume to reveal an accurate assessment The plotting of a five year running total will provide sufficient volume to reveal an accurate assessment the trend (slope) may give an indication of effectiveness of quality improvement programs in place. the trend (slope) may give an indication of effectiveness of quality improvement programs in place.

Background Behind the effort for more transparency and better outcomes is a need to measure and analyze data to present an accurate, clear picture. Since annual physician and hospital case volumes are often low, confidence intervals for many measures are wide and it is difficult to separate performance improvements from noise. The use of funnel plots with annual and especially multi-year data provide a more reliable estimate of performance compared to national benchmarks.

Results Plotting single year data indicates performance with reference to the STS benchmark (O/E=1) and shows whether the outcome is outside the 95% confidence interval. Usually volumes are small enough that confidence intervals are wide. Multi-year plots of annual results show year over year changes but suffer from similar annual volumes, which still leave the data in the wide part of the funnel. The running total adds annual volumes moving the result to a larger volume position where the funnel is narrower. At this point it is often possible to determine that the results are showing statistically significant differences from 1. Trends also become more evident.