Healthcare data reporting ranges from mandatory regulated quality reports to reports that facilitate treatment planning over active patients. The main objective is to transform consolidated medical information into answers to business questions, such as:

• Determine which patients are outside treatment bandwidth. This is especially important for chronic diseases, where the patient might participate and have limited real-life interaction with care professionals.
• Quickly determine a cohort of patients with a particular set of observations relevant to a new analytical study.
• Benchmark treatment centers A & B for a particular treatment of chronic patients.
• Determine treatment efficacy over total patient population.

Conceptually progressing from medical data to a report requires the following steps:

• Data ingress and consolidation
• Report definition: identify the required information.
• Data preparation: transform repository facts to potential aggregated report facts.
• Visualisation: deliver a list or graph (csv, pdf).
• Validation: given test data, does the transformation process report what we expect?

Data can be prepared in a number of ways:

• Using Dataset Builder (DSB) to distill the clinical repository to the required facts.
• For commonly used ingest formats such as CDA R2, CDA R3 and CCDA, MGRID offers a process that automatically returns a view of all observations per patient.

BI Explorer facilitates report definition by drilling down into the prepared base reporting tables. Here queries can be constructed visually, and shared between users. Note that queries which underlie government mandated reports can be identified and shared between all Explorer users.

When all queries needed for a particular report have been constructed, they can be embedded into a report on the MGRID report server, which allows clinicians or administrators direct access to a PDF rendition for a specified period. Drill-down is supported by diving down into a specific stored query in BI Explorer. Our policy is to develop mandatory reports for a particular locality once, and then distribute them to all of our customers.

Healthcare data analytics ranges from basic statistical summaries of data and inferences to advanced predictive models. These models help to gain insight into disease progression and they support of the process of selecting the right treatment for individual patients. Statistical methods used include:

• Prediction
• Quantitative analysis
• Pattern matching
• Structured data analysis
• Identification of abnormal patterns
• Processing large time-series data
• Regression analysis
• Unsupervised / supervised learning
• Descriptive / inferential statistics
• Risk stratification
• Statistical analysis of large historical

One factor that these methods have in common is that they reveal patterns in data, and from these patterns, inferences can be made to support decisions.

Data analysts spend the bulk of their time on getting, cleaning and preparing data, which is the process of preparing and selecting variables for data sets that will be used by analytical tools. Each data analysis project has its own requirements for variables (features), and during the analysis, more knowledge is gained about variables which prove to be useful, and, about the additional variables that are required. Given a repository of healthcare data, the ability to quickly prepare data sets with selected and constructed features will enable data analysts to test hypotheses quickly, thus increasing productivity.

To aid in dataset creation for healthcare research, MGRID has developed Dataset Builder. DSB removes the need to manually craft SQL or scripts to select, merge and manipulate data data, which speeds up the time of data set creation from weeks to hours. Please refer to the DSB product page for more information.

Dataset Builder speeds up creation of datasets from structured information. It can be used to extract features from a wide variety of sources into tables.

DSB includes the following features:

• Exploration of source features like lab results, diagnosis, risk factor,allergy, medication, care plan, organisation, practitioner, device and other variables
• In-database feature construction
• Feature selection
• Cohort selection
• Data set creation and extraction using CSV and SQL
• Scheduling of tabular data generation
• Accountable transformation process for reproducible research

DSB is complementary to existing tools for data analysis and analytic workspace management, acting as a pre-processing step to produce tabular data for consumption by statistics and analytics tools. With DSB the complete transformation process from raw data to tabular data is documented and versioned, which enables re-use of constructed features, and also helps reproducible research and quality assurance of the transformation process.

Built on the MGRID Healthcare Data Model, DSB can perform in-database operations to transform physical quantity continuous variables from one unit to another, for instance from mg/dL to mmol/L. In addition, for categorical coded values, DSB has support for selections and calculations using knowledge from clinical and drug ontologies, to perform data selection and integration based on concept subtree search for hierarchical code systems such as SNOMED-CT.

Query Builder is a web-based tool for visual query building on existing databases

The Query Builder is used by researchers to select research data from existing healthcare databases, such as clinical datawarehouses, and export the data to a research location for analysis.

Actual data delivery can be controlled by an approval process that allows data stewards to either approve or deny each query before processing. If the source database contains personally identifiable data, the data can be masked or de-identified before extraction. This de-identification can be either pre-configured for the source database, or be configured per query (subject to approval by a data steward).

MGRID Message Transformation XFM provides messaging based infrastructure for the exchange of healthcare data between medical data sources and systems for secondary usage reporting and analytics. Built upon RabbitMQ from the Pivotal Big Data suite, with XFM thousands of healthcare messages per second can be received, transformed and loaded into a Healthcare Data Model (MGRID HDM or your own model).

XFM includes the following features:

• Routing of HL7v3 and FHIR messages
• Message validation and aggregation
• Message transformations (e.g. XML to SQL, XML to JSON)
• For large installtions: parallel data pre-processing
• Database loading
• Automatic deployment and scaling of components on virtualization and private clouds.
• Management dashboard

A good case study of XFM was during the Portavita Benchmark, the first Big Data Healthcare Benchmark. In this benchmark many thousands of synthetic CDA and FHIR messages are generated per second, that need to be validated, transformed and loaded into a lake datawarehouse. XFM handles this pipeline, whilst allowing scaleout if more processing hardware is available.

The Messaging SDK (MSG) is delivered as a part of XFM. With this SDK, parsers can be generated for any existing or custom HL7v3 message type. The resulting parsers can convert incoming HL7 XML into SQL or JSON. Parsers for common message types such as CDA and Consolidated-CDA are included by default.

The MGRID Healthcare Data Model (HDM) provides database support for the HL7v3 and HL7 FHIR standards. After loading data, the HDM is the source of information for reporting and analytics use cases which require data from multiple sources or longer periods of time.

HDM includes the following features:

• No loss of information
• Complete, auditable information is stored
• Support for all HL7 message standards
• Full SQL support for query of data: JSON data types combined with SQL JOIN, UNION, CTE queries and other set operators
• Database support for reasoning with physical quantities, coded values and time intervals

The MGRID Healthcare Data Model is the only model that provides seamless access to all HL7 standards from the same unified interface

The MGRID HDM is available on PostgreSQL and Pivotal Big Data Suite databases. PostgreSQL is a highly versatile and server suitable for OLTP deployments, and has been reported to reach over 300.000 TPS on a single node server. For Extract, Load, Transform deployments where the transform step from source data to query format data is performed on-the-fly, Pivotal Greenplum distributed parallel database server ensures query performance for large databases.

Product brief

Information about the MGRID healthcare data platform.

Adding HL7 v3 data types to PostgreSQL

Technical background about adding support for specialized healthcare datatypes to the PostgreSQL relational database server.