Data Analytics Chapter 1: Master Data Management to create longitudinal patient data.

Master Data Management for Longitudinal Patient Data

What’s the most challenging piece of business problem in the US healthcare? There are many we can name including the ineffective preventive care, existence of chronic diseases, old age population, unhealthy eating habits, Lack of effective data points to take decisions, lack of effective technology,  lack of quality care delivery by the providers and ineffective payment mechanism etc. 

All these fuels millions of dollars to the per capita total healthcare cost of care. To reduce the cost of care we need to get the right information to the right person at the right time to make right intervention to the patient care. The information provided to the care taker should be insights and meaningful. The bigger piece of information needs to be sliced and diced while we maintain the longitudinal view.

How do we create the longitudinal patient data with relationships to other entities like physician and family? The answer is data integration through Master Data Management.

Below picture depicts the patient journey.

The healthcare data is being collected and transformed between many entities which are using certified and non-certified Electronic medical record systems. HIPAA mandates certain standards to be followed to minimize the disparities. The variety, volume and velocity of healthcare data completely different from other industry. Below picture depicts the data types and standard where the patient data is stored and exchanged.

• Standard Data Types
• ICD
• HCPCS
• SNOMED-CT
• UBO4CODES
• NDC
• Rx NORM
• ZIP
• Custom Data Types
• Charts
• SOAP Notes
• Observations
• Lab Results
• Standard Data Formats
• EDI
• HL7
• DICOM
• QRDA
• Custom Formats
• XML
• JSON
• FLAT FILES
• APCD
• Formats
• Text
• Numeric
• Paper
• Digital
• Pictures
• Videos
• Multimedia

How do we integrate the data from desperate sources? We are still not started utilizing the block chain to save and communicate the patient records. A long way to go to these new technologies. As of now best option to tackle this is to adopt the Master Data management (MDM) principle.

Master Data Management is a discipline to harmonize the critical data of a business for example Customer data, product data, Patient data etc. It’s not a cake walk to adopt the master data management as it requires organizational characteristics to know the real value of data and its benefit when harmonized with strong willingness to change leadership.

MDM Styles

The important part of the change is identifying the right method that fits the business need as MDM has many variations in its architecture style. 

MDM COTS products

Based on the need enterprises need to adopt the right architecture and MDM solution from the market? There are many proven COTS products in the mark. Following are the some of the market leaders.

• Informatica
• Reltio
• SAP
• IBM Infophere
• Talend
• Visionware Multivue
• SAS
• Etc..

Matching and Merging Engine

All of these products has matching and merging engine at the heart to create the golden record. The mostly used algorithms for the matching are the following, which basically utilizes the deterministic and fuzzy matching techniques.

• Exact
• Synonyms
• Phonetic
• Edit Distance
• Date  Interchange

Below is an example of how data from different sources are matched for a patient and creates the unique golden records by using multiple algorithm or match types. While creating golden records we could prioritize each field values based on trustiness of the source or latest of the data or most commonly occurred data etc.

Good Match

These are the records which are coming from same or different source but they are the same entity by all the means. The confidence level of these records are very and they could be merged, create a golden record and assign a unique identifier to them.

Potential Match

This is the grey area after matching, they could be either a match or not a match. These are the records which requires manual intervention by the data stewards with business knowledge to take a business decision to merge them or unmatched them.

No Match

This the bucket where the matching engine couldn’t find any match with other records i.e. these are the singleton records.

Processes Involved

Since master data management is a discipline which improve by practices and process, it’s important to follow each process and make necessary actions at each step. At very high level we would have below processes

• Data Profiling
• Data Cleansing and Standardization
• Data De-duplication
• Data Enrichment
• Matching
• Merging / Unmerging
• Data DE identification

Healthcare industry need to adopt right master data management process to improve the quality of data to derive the meaningful insights. In addition to adoption it’s also important that the educational process and data governance activities.

Ref: https://www.gartner.com/doc/3824466/comparison-master-data-management-implementation .