healthcare data analytics implementation
TRANSCRIPT
INNOVATE INTEGRATE TRANSFORM
March 31, 2016
PREDICITIVE ANALYTICS IN HEALTHCARE
©ALTEN Calsoft Labs
HIMSS describes healthcare analytics as the “systematic use of data and related clinical and business (C&B) insights developed through applied analytical disciplines such as statistical, contextual, quantitative, predictive, and cognitive spectrums to drive fact-based decision making for planning, management, measurement and learning
Objectiveso Healthcare providers are improving the clinical outcomes of patients
via treatments and protocolso Promotion of wellness and disease management
Overview
The key objective of healthcare analytics is to gain insight for making informed healthcare decisions.
©ALTEN Calsoft Labs
Patient Re-admission within 30 days – A case study
Business ProblemA healthcare facility wants to identify a patient's chance of getting re-admitted upon discharge within 30 days.
BenefitsClinicians can be prepared to provide better post-discharge
care for patients who are likely to get re-admitted and hospitals can avail benefits from Government.
©ALTEN Calsoft Labs
Patient admission If a patient is hospitalized for more than 24 hrs, it is considered as Patient admission.
Re-Admission Patient gets admitted for more than 24 hrs within 30 days of the last discharge date. If a patient comes back to the hospital after 30 days, it is not considered as Re-Admission.
Definitions
©ALTEN Calsoft Labs
Data Analysis Process
The below figure shows the typical processes of Data analysis of a Dataset.
Receive the Datasets (.csv)
Process the Datasets for
Analysis
Analyse the Datasets
Build the Model
Visualize the Analysed data
©ALTEN Calsoft Labs
In order to predict the re-admission, following data fields/predictors were considered.
Demographics – Age, Sex Lab data – Includes lab tests Vitals – Includes BP, Sugar, Weight, etc. Visit types – Emergency, In-patient, Outpatient Diagnosis – Diseases/ailments – Heart,Pnuemonia Previous hospital visit Length of stay
The Predictors – Predictive Analytics
©ALTEN Calsoft Labs
The data was received as a set of .csv files which gave the complete
details of Demographics, Admission, vitals, lab tests of selected sample
of patients over a period of time.
The processing of the data included the following activities:
o Removing commas, uploading .csv files to HDFS (Horton works)
o The required DDL scripts were written in Hive
o The necessary joins were written
o The result was refined datasets
The refined datasets are passed on to Data Analysis team for analysis
Data Processing …
©ALTEN Calsoft Labs
Predictive Analysis Process
Use the Module – Analyse the dataset
Identify the Suitable Algorithm
Build the model
Evaluate/Deploy the model
Monitor/Refactor the model
©ALTEN Calsoft Labs
Datasets
The refined datasets are divided into train and test datasets in order to build the Model
30%
70%
Train Test
©ALTEN Calsoft Labs
The best model is arrived at by testing the data under different classifiers
and precision, recall and F1 score metrics calculated for each classifier.
Gradient Boosting
Random Forest
Support Vector Machines
Logistic Regression
K-Nearest Neighbor
Ridge
Evaluate Models
©ALTEN Calsoft Labs
Model - Process
ModelTrainingDataset
TestDataset
Model Final Analyzed Dataset
©ALTEN Calsoft Labs
Model – Fine tuning K- Fold Means
For tuning parameters and model selection, k-Fold cross validation was used where data was split into K equal partitions. 1 fold was used for testing and the remaining for training. This was repeated K( K=4) times and using the average testing accuracy.
Dataset
©ALTEN Calsoft Labs
Accuracy
Accuracy is measured by area under the ROC curve as shown below0.77 accuracy is achieved by Random Forest as shown in the below curve
©ALTEN Calsoft Labs
Process – Data AnalysisThe below figure shows the typical processes of Data analysis of a dataset.
1.Offline Processed data is dumped into Staging Data Mart
Rest Client
CAF Analytics
EngineRest API
1. Builds Model running Python scripts
2. Scores model
©ALTEN Calsoft Labs
Data Visualization – Report on the Model
©ALTEN Calsoft Labs
Actual Report on a Dataset
©ALTEN Calsoft Labs
Screen shots
Write to US @ [email protected]
©ALTEN Calsoft Labs
18INNOVATE INTEGRATE TRANSFORMCopyright 2016 © ALTEN Calsoft Labs. All such documents and related graphics are provided "as is" without warranty of any kind and are subject to change without prior
notice. ALTEN Calsoft Labs reserves the right, in its sole discretion, to correct any errors or omissions in any portion of this document
Visit: www.Altencalsoftlabs.com