Predictive Analytics 101: An Introduction to the Future of Healthcare

Transcription

1 MGMA 2017 ANNUAL CONFERENCE OCT ANAHEIM, CA Predictive Analytics 101: An Introduction to the Future of Healthcare Frank Cohen, MBB, MPA Director, Analytics, Doctors Management LLC Clearwater, Fla. Frank Cohen does not have any financial conflicts to report at this time. 1

2 Learning Objectives 1. Explain the basic concepts of predictive analytics 2. Illustrate how predictive algorithms are built 3. Incorporate predictive analytics or nearpredictive analytics into audit plans Defining Analytics in General often involves studying past historical data to research potential trends, to analyze the effects of certain decisions or events, or to evaluate the performance of a given tool or scenario. The goal of is to improve the business by gaining knowledge which can be used to make improvements or changes. [ 2

3 Analytical Categories Descriptive Exploratory Inferential Predictive Causal Mechanistic Prescriptive Descriptive Descriptive is a preliminary stage of data processing that creates a summary of historical data to yield useful information and possibly prepare the data for further analysis. For example: Overpayment rate for a specific sample Median work RVUs reported by a physician population Percent of patients that are late or no shows Accounts receivable over time Claims summary information Descriptive describe a current/past state or condition 3

4 Exploratory In statistics, exploratory data analysis (EDA) is an approach to analyzing data sets to summarize their main characteristics, often with visual methods. For example: Profiles of provider transactions Provider similarity according to profiles Visual summaries of large amounts of data Eligibility data link to provider billing In EDA, the researcher takes a bit of a bird s eye view of the data in order to make sense of what is available to be reviewed Inferential Inferential statistics draws valid inferences about a population based on an analysis of a representative sample of that population. For example: Estimating overpayment from a sample to a population Gain a better understanding on how work RVUs will impact compensation Improving scheduling by getting a better estimate of uncertainty Using time series analysis to understand A/R over time Absolutely dependent upon a statistically valid random sample (SVRS) Inferential models include estimation, prediction and assessments 4

5 Predictive Predictive analytics is the branch of statistics which is used to make predictions about unknown future events. For example: Predicting the likelihood that a given physician will be audited in the future Predicting which procedure codes/modifiers are most likely to be targets Predicting how likely a provider is to be sued for malpractice Predicting the likelihood that a patient will return to the hospital within 30 days Predicting the time it will take for a new physician to break even Predictive analytics uses many techniques from data mining, statistics, modeling, machine learning, and artificial intelligence to analyze current data to make predictions about future. Prescriptive Prescriptive analytics is the area of business analytics (BA) dedicated to finding the best course of action for a given situation. Prescriptive analytics is related to both descriptive and predictive analytics. 5

6 An Analytical Approach Prediction vs. Estimation Estimation uses data to estimate (or guess) at a parameter of for some already known variable Prediction uses the data to estimate (or guess) at some random value that is not a part of the known variables or data set Estimation or prediction? Extrapolation Audit risk Impact of new drugs Healthcare insurance premiums Driving a car Piano tuners in Chicago Predictors usually have larger uncertainties than estimators 6

7 We cannot solve our problems with the same thinking we used when we created them. Albert Einstein LET S TALK PREDICTIVE Why Predictive Analytics? The big picture creates the need for: Strategic and financial planning Improving access to care Improving RC and profitability Improving outcomes and access Limit and mitigate risk Helping to achieve focus on priorities Leaders want to be able to look into the future What should we expect next year? What can we do to meet our objectives? 7

8 What can PA provide to the business? Improving efficiencies (more with less) Improve financial forecasting, ensuring long term survival Resource allocation and categorization Understand the market better More competitive New products and services Understand patient care better Predicting areas of patient dissatisfaction Statistical analysis Text Mining Machine Learning Artificial Intelligence Forecasting Optimization Exploratory analysis The Many Faces of PA 8

9 Supervised vs. Unsupervised Learning Supervised learning Supervised learning is an approach to machine learning where both input and desired output data are provided in the form of expected answers. Input and output data are labelled for classification to provide a learning basis for future data processing Unsupervised learning Unsupervised learning is an approach to machine learning whereby software learns from data without being given correct answers. It is an important type of artificial intelligence as it allows an AI to self improve based on large, diverse data sets such as real world experience. Basic difference in layman terms : In supervised learning, the output datasets are provided which are used to train the machine and get the desired outputs whereas in unsupervised learning no datasets are provided, instead the data are clustered into different classes Application of learning types Supervised Facial and visual recognition Sorting good claims from bad claims Cloning of EHR records Identification of patients at risk Unsupervised Facial and visual recognition Self driving cars Human behavior patterns Robotic vacuums Machine learning 9

10 Types of Models Classifiers place data points into unique buckets Recommenders recommends products and service a consumer is most likely to purchase based on prior behavior Numerical models include regression, time series and other areas of canonical statistics Text Mining enables high dimensional analysis of unstructured data, like text found in EHR and documentation in patient charts Neural Networks used complex weighting to find the best predictor Naïve Bayes Kth Nearest Neighbor (KNN) CART (Classification and Regression Trees) Random Forests Regression (linear, multiple, logistic, etc.) Time Series analysis MARS Boosted Trees Majority classifier Support Vector Machine (SVM) Neural Networks Natural language processing Social Network analysis (SNA) And new ones just about every day PA Algorithms 10

11 KNN, Clusters, CART, Decision Trees CLASSIFIERS Classifcation Classifies each data point based some set of attributes Specialty Time in practice Prior lawsuits Prior audits Internal coding reviews Patient satisfaction surveys Resignations and terminations The goal is to assign an unknown variable or record to a class based on location 11

12 Classification types KNN (Kth Nearest Neighbor) Classification trees and forests Neural networks Support Vector Machines (SVM) KNN (Kth Nearest Neighbor) In KNN, we place each data point into a class that is most appropriate We measure the distance between the items in a class using specific metrics Centroid measures the distance from the center of a class Medoid measures the distance from some representative point A new data point is placed into the given class based on it s distance from the closest class 12

13 KNN Example 1 Classifying whether a given claim may be a high risk target for audit or review Using any number of variables, such as number of ICD codes, procedure code, presence of a modifier, patient demographics, cohort comparison, etc., a claim is classified as either risk or no risk based on its distance to the cluster centroid Risk type (high, medium, low) is a function of distance thresholds Closer to centroid higher risk Classification and Regression Trees (CART) Classification and regression trees are machine learning methods for constructing prediction models from data. The models are obtained by recursively partitioning the data space and fitting a simple prediction model within each partition. As a result, the partitioning can be represented graphically as a decision tree. 13

14 Building a classification model Training Data Testing Data Tree ID Attrib 1 Attrib 2 Attrib 3 Class 1 Yes High 3125 Yes 2 Yes Medium 5280 Yes 3 No High 5800 No 4 Yes Low 2100 No 5 No Medium 17 No 6 No High 7050 No 7 No High 1104 No 8 Yes Low 2304 Yes 9 No Low 187 No 10 Yes Low 1874 No 11 Yes Medium 6102 Yes Tree ID Attrib 1 Attrib 2 Attrib 3 Class 12 Yes Medium 3104? 13 Yes High 2877? 14 No Low 788? 15 Yes Low 3671? 16 No Medium 2252? CART Example 2 Bivariate Tree ID Attrib 1 Attrib 2 Attrib 3 Class 12 Yes Medium 3104? 13 Yes High 2877? 14 No Low 788? 15 Yes Low 3671? 16 No Medium 2252? 14

15 CART Example 2 Multivariate Tree ID Attrib 1 Attrib 2 Attrib 3 Class 12 Yes Medium 3104? 13 Yes High 2877? 14 No Low 788? 15 Yes Low 3671? 16 No Medium 2252? CART Example 1 Predicting hospital stays of greater than 30 days This is a binary tree that bifurcates each step based on specific scores (or values) associated to encounter times with different healthcare teams 15

16 Support Vector Machines A Support Vector Machine (SVM) is a discriminative classifier formally defined by a separating hyperplane. In other words, given labeled training data (supervised learning), the algorithm outputs an optimal hyperplane which categorizes new examples. Regression and Time Series NUMERICAL MODELS 16

17 Regression Analysis Regression analysis is a form of predictive modelling technique which investigates the relationship between a dependent (target) and independen t variable (s) (predictor). This technique is used for forecasting, time series modelling and finding the causal effect relationship between the variables. For example: Predicting how long it will take a new physician to break even Predicting how many (and which) patients will be no shows Predicting body fat using BMI Regression analysis example 1 Predicting revenue for a newly hired physician Take revenue amounts for some number of physicians over some time period and create a slope formula Use formula top calculate outside of data range, for example, at 18 months, it is (638.5 * 36) ( ) = $12,711.5 Value Fitted Line Plot Value = Value Value 2^2 S R-Sq 98.9% R-Sq(adj) 98.8% Value

18 Regression analysis example 2 Predicting body fat percentages using BMI data Plot BMI against percent of body fat Using slope formula to predict outside of the regression Where BMI = 33.2, body fat = (3.286 * 33.2) ( * ) = Regression analysis example 3 Predicting long term recovery after discharge from a hospital Chart prognosis scale against number of days since discharge Use slope formula to predict some number outside of the data set At 42 days, prog = * =

19 Time Series In descriptive modeling, or time series analysis, a time series is modeled to determine its components in terms of seasonal patterns, trends, relation to external factors, and the like. In contrast, time series forecasting uses the information in a time series (perhaps with additional information) to forecast future values of that series {Page 18 19, Practical Time Series Forecasting with R: A Hands On Guide} Time Series Example 1 Collection Moving Average Plot for Collection Month Variable Actual Fits Forecasts 95.0% PI Moving Average Length 3 Accuracy Measures MAPE 20 MAD 3493 MSD Charge Collection Run Bonus Salary Cost P/L 1, , , (20,887.61) 3, , , (20,426.12) 3, , , (20,334.69) 5, , , , (19,915.64) 6, , , , (19,112.33) 8, , , , (18,210.51) 10, , , , (16,696.81) 14, , , , (15,155.12) 18, , , , (12,634.05) 25, , , , (9,516.35) 33, , , , (4,498.93) 45, , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , ,

20 Neural Networks In between the input units and output units are one or more layers of hidden units, which, together, form the majority of the artificial brain. Most neural networks are fully connected, which means each hidden unit and each output unit is connected to every unit in the layers either side. An often stated advantage of neural networks over conventional programs lies in their ability to solve problems that either do not have an algorithmic solution or a solution is too complex to find. Neural networks are well suited to tackle problems that people are good at solving, like prediction and pattern recognition NEURAL NETWORKS 20

21 Neural Network Example 1 Predicting Mortality Neural Network Example 2 Predicting Heart Disease 21

22 TEXT ANALYTICS Text Analytics Text analytics converts unstructured text data, which account for over 70% of healthcare records, into meaningful data that can be used for analysis, feedback, search engines and other purposes. For example Cloning detection Patient sentiment analysis Outcomes assessments 22

23 Text Analytics Example The Truth about PA Predictive modeling is all about probabilities and uncertainty Some likelihood an event occurs, not certainty PA does not necessarily work for every issue or problem Highly complex systems where co dependencies cannot be modeled How accurate do my predictive mode 23

24 In Conclusion... Continuing Education ACMPE credit for medical practice executives. 1 AAPC Core B, CPPM credit 1 ACHE credit for medical practice executives 1 CME AMA PRA Category 1 Credits.. 1 CNE credit for licensed nurses 1 CPE credit for certified public accountants (CPAs) 1.2 CEU credit for generic continuing education 1 Let the speakers know what you thought! Evaluations are available on the MGMA mobile app 24

25 Frank Cohen Doctors Management LLC 2075 San Marinio Way North Clearwater, FL