HADM 250 Data Anytc in Health SVR Ad MN Notes (April 20-27 2020) PDF

Preview

Summary

This document contains all of my lecture notes, screenshots and discussion post writings for the last two lectures given in April of 2020....

Description

4.20.2020 Lecture Notes ●

Term Paper will be due before May 4th so that Professor can provide feedback. Professor would like it submitted preferably on May 1st.

●

Final Exam will be made available on May 15th, 2020, will be very similar in format to the midterm. Multiple choice on Blackboard.

●

We have the knowledge and expertise to intervene on pandemics, unfortunately it takes time to intervene.

Overview of Modeling Techniques ●

●

Machine Learning: A Branch of AI; Concerns the construction and study of systems that can learn from data. (AI was originally conceived in the 1970's with mainframes and has continued to evolve significantly with the data we can actually store ) ○ Decision tree, Neural Networks, Inductive Logic, Association Rules, Clustering, Bayesian Networks ○ Natural Language Processing Statistical Analysis: Study of the collection, organization, analysis, interpretation and presentation of data. ○ Descriptive (Inferential), Predictive & Prescriptive ○ Visualization & Confidence

●

Operations Research: Simulation Models

●

Descriptive is really the historic analysis

●

Prescriptive is more judgmental to what could happen if we change parameters in the environment. We look to stimulation modeling with different variables that are always changing

Step 1: Data Preparation ●

Can bring amendments to the data forward important process that requires domain knowledge

●

ALL DATA IS WRONG! ○ There are duplicate records all the time literally and in the sense of billing that have to be removed first ○ There is a monthly billing record that is a duplicate ○ There is always a change in admission and billing records

Multilevel 3-D Framework For Forecast Decision Support ●

Defining geographic location where patients are coming from is largely based on market share

●

Development of services is very important

Time Granularity For a Forecasting Data Framework ●

Data can be seen and go down to weekly or hourly data

●

If your looking at the ER the arrival times for patients can be significant indicators (covid19 for example)

●

Data has to be as complete as possible. NYSDOH (New York State Department of Health) are 3-5 months behind on the data collection

●

Data we are working with often has some incentive to maximize the reimbursement that providers want to achieve Source of the data is one thing, but the reliability is another and have to consider the cost.

●

●

To acquire data costs at least $1,500 a year to download data from the DOH

●

A single spreadsheet of data can cost up to $30,000

●

Ease of use and accessing and managing and mining are steps that are critical

Making Improvements In Data Quality ● ● ● ● ● ●

Consistency-data that retains the same significance for demand prediction over time Accuracy-the extend that data are reliable and free from significant error Timeliness-the committed availability timeframe meets the forecaster's schedule Reliable-can be depended on when and as promised Affordable- cost of data acquisition is within budget Ease of use-how readily users can access data

Non-seasonal and Seasonal Forecasts with Prediction Limits ●

Physician on the left (left graph) relatively new physician

●

Upper limit and lower limit this group is a maternity group and its not unusual to see this activity in seasonality (right graph)

Making Adjustments (Overrides) To Multi-Level Forecasts ●

●

At lowest level, overrides ○ May be too detailed and not realistic ○ Can be aggregated ("rolled up" to higher levels) At summary levels, overrides ○ Require business rules (algorithm) to drill down ○ May result inappropriate allocations ○ Can lead to inconsistent trends and seasonality at lower levels Solution: Document all judgmental adjustments and periodically evaluate performance

●

Ex. NYU missing their data, its possible to manually go in and falsify or put adjustments in because there were no numbers in there

●

Have to document the adjustments and justification for the adjustments

The Intelligent Dashboard Environment for Decision Support ●

Not just useful for demand forecasting, but also applicable to visualize:

● ● ● ●

Managing strategic opportunities Monitoring competitive environment Enhancing demand intelligence Providing a reporting capability

Confidence, Bias and Precision Biased consistently off the mark in one region, its biased to upper right side Unbiased means we are in the target almost all the time Precision hit the target exactly every time Need to recognize the difference between the terms in the context of diagnostic codes for example that its biased because its being economically driven Forecasts: three months into the future can make a drastic difference Finance department would want a forecast for 12-18 months into the future, have to understand how much statistical noise is in the data Goodness of Fit is Not Same As Forecast Accuracy! ● ●

Goodness of fit statistic- Fit period Accuracy measure-Forecast period ○ Model fit is designed to model historical patterns-May not necessarily translate into accurate forecasts ○ Model fit may add complexity -May not be beneficial to forecast periods

Creating a Waterfall Chart

S1-Create Hold-Out samples (or use stored forecasts) S2- Make forecasts with hold-out sample in a table (This will look like an upside down waterfall) S3-Measure performance with multiple measures

Remedies For Handling Data Exceptions Replace of outliers if warranted Use a model to predict unusual value Replace value with prediction Run model over extended period and refine replacement predicted value Include "dummy" variable in models Utilize alternative (robust/resistant)methods and compare results with conventional method Be on guard for "Black Swans" www.youtube.com/watch?v=BDbuJtAiABA Have to understand there will be black swan events that we can’t foresee (natural disasters, pandemics, epidemics, etc)

Discussion Post-ONC Delays Final Interoperability Rule Deadline Due to Coronavirus I thought this article was very interesting and an appropriate move to ensure that those affected by COVID-19, particularly those who have gotten sick from the virus are receiving the care that they really need following discharge from the hospital, nursing home or any other healthcare facility where they are receiving care. When I first read that CMS had extended the deadline to implement 21st Century Cures Acts due to COVID-19, my initial reaction was, if there was ever a time to not extend a deadline when it relates to patient care in some way, now is the absolute worst time. But after reading the article further, I realized if they did not extend the deadline, the focus on curing and successfully discharging those who have tested positive for the virus would not be the same and right now, that is what takes priority as it is affecting millions of people AND their families. However, implementing the final interoperability rule is also very important so that patients can gain access to their own electronic health information as well as providers to help them understand the necessary steps that need to be taken for their health as well as make information sharing secure, this is still a process that needs to be done mainly for events like this (pandemic) because time is very vital, and how healthcare is delivered in the United States still has a negative effect on the entire country if it is not run in the most efficient way possible.

It is a tough situation but I feel that this extension is necessary given the extent of the pandemic and trying to get it under control first. Classmates Discussion Post

Response

4.27.2020 Lecture Notes ●

We have faced epidemics before, but we have not faced anything as dramatic as this pandemic

●

The amount of data we have is growing exponentially

●

We are very accustomed to reading highly structures data (billing records, listing of procedures performed, classifications of what has happened to a patient, can link multiple episodes that has happened to a patient overtime).

●

Over the last decade we have been evolving from paper records to the electronic health record

●

We are developing the machine learning technology to be able to interpret the data

●

Where are the high points in the data?

●

We have been evolving now into data hubs, we have been trying to grapple with for the past 10 years, trying to share data over multiple hospitals but it has been oriented to highly structured data formats.

●

We are looking at going national as far as the electronic health records and have interoperability

●

GIS: Geographic Information System

Governance Readings ●

Is a very important of this whole process (governance), we have to be concerned with privacy and regulatory

●

There are conflicts that can arise and we as leaders have to prepared for these things

● ● ●

Information Governance: Principles for Healthcare Governance Practices & Benchmarking: Survey & Infographics ○ Quality of care we are providing in the industry ○ Multiple measures we need to be concerned with Data Management: SAS overview ○ How to manage large data bases, SAS is one of the bigger databases

●

Information Governance ●

How good is the data? Where is it coming from?

●

When high tech was created in 2000, we had very few hospitals using ehrs, the VA was far ahead of many organizations at the time

●

Over 90% of hospitals, physicians and physician offices were working with ehrs in 2018.

●

New technologies and work flows are going to become a part of the new generations

●

What can we do to allow the machines to help us store information?

Governance: ●

● ● ● ● ●

● ●

Accountability ○ Who is responsible for what? What are they allowed to do with it? ○ Goes all the way up to the board of trustees to ensure things are being done correctly and so that the system can not be hacked Transparency Integrity Protection Compliance Availability ○ Can the physician be able to acquire his office records to recall what the history of a patient might be? This is why we need interoperability to be there? ○ How long should the records be there? Retention Disposition

Overview of Modeling Techniques ●

Geographic Mapping & Visualization

●

○ Specialized Area of Software ○ Preliminary Descriptive Analysis & Demographics ○ Visualization & Confidence Operations Research: ○ Simulation Models: Monte Carlo ○ Probabilistic Models: Decision Rules

Primary Geographic Market Newburgh and Cornwall Areas ●

Politics is alive and well in the healthcare industry!

Primary Geographic Market Graph ● What are the financial characteristics? ● The decrease is an Indicator that there was a significant change in the population, it was based on the closure of Steward Airforce base?** ● Which segments of the populations, which zip codes were declining, ● ●

The green areas of the two hospitals What were the market shares? We see trending the tertiary care hospital had declined 15%

Primary Market Areas Competitors Graph A portion of the population would always be going to other tertiary hospitals such as Mt Sinai, NYU, Memorial Sloan Kettering

How Is This Accomplished ●

Database Query’s ○ Demonstration ○ Visual Studio Query: we identify the various fields of data we are concerned with, on the right side we are trying to identify what was going on (-6947) ○ What the machine does is write the code for us when we point and click into these fields. No need to learn programming language to understand how to use this query ○ Within the zip codes are geographic census tracks which are considerably small in size. Represents 6,000 lives approximately ○ Zip code in an urban area represents approx. 50,000 people

●

Spreadsheet Analysis ○ Demonstration

Environmental Change ●

Berger Commission Modeling of Closures ○ Demonstration

● ● ●

Predictive Modeling: Rules & Ratios Monte Carlos Modeling: Multiple Distribution Models & Random Cycles

Berger Commission Model WE can close an entire hospital or an individual service at the highest level and identify any # of hospitals that will close these services

Intermediate Decisions ● ● ● ●

Initial Normal Competition in Market Place Mercy & Franklin: Close Trauma, ○ Pediatrics & Maternity Services Long Beach Hospital Closure Mercy Hospital Closure

● ●

We picked and choose what the Berger Commission might be addressing The hospital today continues to operate at 22,000 patients per year

Utilization Detail ● ● ●

Can understand multiple things when the right questions are asked Very sophisticated spreadsheet opportunity but using machine learning in order to generate these kinds of tools These are a team sport and you as a manager have to prepared to ask the right questions so these tables can be made for you...