Predictive Analytics – Reducing Risk of Heart Attack Mortality During Hospital Stays

 OVERVIEW: Performance analytics provide the empirical basis for the strategic and tactical management of healthcare outcomes.  As discussed in the referenced JAMA study, for every percentage increase in the guideline adherence rate in hospitals there was an equivalent decrease in the likelihood the patients treated would die before discharge.  Ironically, in the 350 hospitals studied, up to 25% of the opportunities to provide guideline-recommended care were missed in current treatment practice.

This is an introduction to “performance analytics” a relatively new and sophisticated way of helping manage complexity, in this instance focusing on healthcare outcomes.  A healthcare analytic performance model taken from research published in the Journal of the American Medical Association (JAMA) is provided for context and a reference to “the real world.”   The goal of the JAMA study was to determine the impact of following guide-line recommended treatments on positive health care outcomes, i.e. not dying in the hospital of a myocardial infarction (MI) acute coronary syndrome (ACS).

 Performance Analytics

Performance analytics  have been successfully used to judge both outcomes (how well did we do in the past) and to predict performance (how well are we likely to do in the future).  An outcome orientation implies that we can learn from past behavior and apply lessons learned to the current operation.  By studying the past we can improve the present.  It helps if the organization being studied is operating in a relatively stable environment where the emphasis is on efficiency (doing the right thing correctly) and not on effectiveness (figuring out what is the right thing to do) – it’s the difference between exploiting knowledge and experience, and exploring for new alternatives.  In risk management this is often called “failure mode and effect” analysis, where we assess the underlying reasons why something went wrong/or could go wrong and what were/could be the outcomes.

The companion to outcome analysis is predictive analysis.  Here we are trying to increase the odds of gaining a certain goal.  If we are interested in reducing process time we may try to determine the critical path of the process, perhaps making use of new network based modeling techniques or even agent based modeling and behavioral modeling to assess the likely impact of changing the rules-based activity.

Performance analytics can pay dividends in risk management when we are trying to prevent something bad from happening rather than correcting an existing problem.  They can, for example, help homeland security officials better assess where grant money would do the most good in reducing vulnerabilities in our transportation, communication, energy and financial infrastructures.   They can also help state and local governments dexplore the best way to invest for disasters, natural or manmade.  Modeling the distribution of vaccine in response to a possible pandemic would be one application. As you will see in the JAMA study, performance analytics can help hospitals to focus operational and financial resources where they get the most return for their money.

The focus can be either financial or operational. Senior management may be primarily interested in financial outcomes, such as return on assets.  Line operations and employees may be more interested in operational outcomes, i.e. lives saved.  But regardless of whether you are interested in predicting or assessing, finance or operations the basic concerns are still the same: “does the model hold water,” is it valid and reliable?  Assessing validity and reliability can prove to be knotty problems.

Validity means are we measuring or predicting what we think we are predicting? For example, if I measure hospital process performance will it tell me anything about outcomes among coronary patients?  Did we pick the correct metrics?  Is the data good?  And alternately is the model reliable? Can I generalize to other hospitals?  If others replicate my study will they get the same results?  So when we talk about reliability we are really interested in getting consistent readings and results.

 Case Study – Journal of the American Medical Association (JAMA) “Association Between Hospital Process Performance and Outcomes Among Patients With Acute Coronary Syndrome”[1]

What role does quality play in healthcare delivery?  A profoundly significant question, encompassing government agencies, accrediting agencies, insurance companies, professional associations and last, but not least patients who have suffered acute coronary syndromes, i.e. heart attacks.  Specifically do we have any evidence these processes make a difference?

Hospitals have bought into the idea that process-based performance systems will result in better patient outcomes, but prior to this JAMA study there was not much empirical evidence.  Non-ST-segment elevation (NSTE) myocardial infarction (MI) account for 1.6 million annual admissions in US hospitals and represent over 75% of all MI cases in US hospitals.  Here we have a study that is primarily focusing on operations metrics and assessing outcomes.  Although even for the casual reader, it is not hard to see the possibilities for further inference about financial impacts and potential predictive value of this baseline study.

The Study

The study evaluated 9 individual ACC/AHA class I (useful and effective) American College of Cardiology and American Heart Association guide-line-recommended therapies including four acute process-of-care measures (aspirin, β-blocker, heparin, and intravenous glycoprotein inhibitors) used in the first 24 hours as well as five discharge regimens.  Patient eligibility was consistent and determined by ACC/AHA guidelines.  Patients who died during the first 24 hours were excluded from the assessment of the acute process-of-care; and those who died at anytime during the hospital stay were excluded from the discharge regimen assessment.  Overall the assessment was just about as rigorous and professionally controlled as you could possibly hope for. The software used in the analysis was SAS version 8.2. Data was collected from CRUSADE an ongoing voluntary, observational data collection and quality improvement initiative, and data was abstracted by professionally trained data collector at each hospital.  Once collected, sanitized data (identity striped) was entered via a Web-based data collection tool and aggregated into an analytical database.  The audited resulting error from missing data was a remarkable 5% across all collected data elements.  It really doesn’t get any better than this.

The Results

Quality of care is defined as the degree to which health service for individuals and populations increases the likelihood of desired health outcomes and are consistent with current professional knowledge.  This was the first study to attempt to link variability in hospital process performance with patient outcomes.

–          Of the 350 hospitals studied, up to 25% of the opportunities to provide guideline-recommended care were missed in current treatment practice.

–          After adjustment, every percentage increase in the guideline adherence rate of hospitals was associated with an equivalent decrease in the likelihood the patients treated would die before discharge.   Let me be more specific: a 10% increase in composite adherence gave a corresponding 10% reduction in the odds of dying in the hospital…and this at the 90-95% confidence interval.

–          Processes were more important than structural features, i.e. that new cardiology emergency room was not as significant as whether or not ACA/AHA guidelines were followed.

The Significance

We are entering a new era of risk management and performance improvement in healthcare.  Performance analytics can save lives.  The evidence is valid and reliable, and significant.   Your hospital can build on this study, improving outcomes and reducing risk.  Performance analytics are signaling a new cooperative era between healthcare providers, insurers, government and practitioners.

[1] Peterson, et al, “Association Between Hospital Process Performance and outcomes Among Patients With Acute Coronary Syndromes, JAMA April 26m 2006, Vol. 295, No. 16.


4 thoughts on “Predictive Analytics – Reducing Risk of Heart Attack Mortality During Hospital Stays

  1. John,
    I really enjoyed your blog. I haven’t seen much in the past on linking process measures to outcomes which will be important if they are to stimulate change.

  2. Pingback: Viva The Revolution in Analytics! Viva Performance Analytics! A Shameless Parody « Prologue

  3. I truly have a tendency to agree with the whole thing that is written throughout Window
    Treatments Ideas “Performance Analytics – Reducing Risk of Heart Attack Mortality During Hospital Stays |
    Prologue”. I am grateful for all of the actual
    details.I appreciate it,Jacquelyn

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s