Towards a standard query model for sharing decision-support applications.

Walter Sujansky, Russ B. AltmanPubMed 1994

TL;DR

This paper tackles the challenge of sharing clinical decision-support (CDS) applications across different healthcare sites by proposing a "Standard Query Model Framework." This framework automates the translation of queries from a universal clinical data model to site-specific database formats, aiming to reduce costs, errors, and delays associated with manual customization.

What Problem Does This Paper Address?

Clinical decision-support (CDS) applications hold immense potential for improving healthcare, but their adoption is hampered by a significant technical hurdle: integrating them with existing clinical databases. The core problem is database heterogeneity. Different hospitals and clinics use diverse database systems with varying schemas, query languages, and terminology. When a CDS application, developed with a specific data model in mind, needs to query these disparate databases, its queries must be painstakingly translated to match each local database's specifications. This manual translation process is not only expensive and time-consuming but also highly prone to errors, making it impractical to share CDS applications widely. The authors argue that a standardized approach to representing clinical data and translating queries is essential for unlocking the full potential of CDS.

Methods

The paper introduces the Standard Query Model Framework, which consists of three key components:

  1. A Standard Reference Schema: This is a site-independent model of clinical data, allowing CDS developers to formulate queries in a universal language.
  2. A Formal Mapping Language: Healthcare sites use this language to define how their local database schemas correspond to the standard reference schema.
  3. A Translating Compiler: This compiler automatically translates the site-independent queries into semantically equivalent queries that can be executed by the local database, using the provided mappings.

The authors have developed a prototype implementation called TransFER, which embodies this framework for relational databases. TransFER utilizes:

  • FER (Functional Entity-Relationship model): A novel semantic data model designed to be abstract and expressive enough to represent diverse clinical database implementations.
  • ReFER query language: A high-level query language that corresponds to the FER data model, allowing users to express data retrieval requests abstractly.
  • ERA (Extended Relational Algebra) mapping language: A declarative language based on relational algebra, used to define mappings between FER schemas and site-specific relational database schemas.
  • TransFER query-translation module: This module applies the ERA mappings to translate ReFER queries into site-specific queries.

Datasets

Not discussed in the paper. The paper focuses on a conceptual framework and a prototype for relational databases, rather than empirical evaluation on specific datasets.

Key Insights and Ideas

  • The "Query Model" is Key: The paper highlights that a "query model" encompasses the data model, database schema, query language, and domain terminology. Heterogeneity in any of these aspects creates integration challenges.
  • Query Translation is Superior to Data Translation: The authors argue that automating query translation is preferable to data translation because it avoids data duplication and the associated costs and integrity issues.
  • Declarative Mappings are Crucial: The use of a declarative mapping language is emphasized for its benefits in inspection, validation, and maintenance, which are critical for managing complex translation rules.
  • Abstraction is the Solution: By providing a standard, abstract reference schema (FER) and a corresponding query language (ReFER), the framework allows CDS developers to work independently of local database specifics.
  • Relational Theory as a Foundation: The authors leverage relational algebra to provide a formal basis for their mapping language and translation compiler, enabling sound optimization and verification.

Key Results

The paper presents a conceptual framework and a prototype (TransFER) for automating query translation. While specific quantitative results are not detailed, the key outcome is the demonstration of a methodology that:

  • Enables CDS applications to be formulated against a site-independent clinical data model.
  • Automates the translation of these queries to site-specific relational database implementations.
  • Aims to significantly reduce the cost, complexity, and error rate associated with sharing CDS applications.

The prototype is designed to handle representational heterogeneity in relational databases, including differences in entity naming, data storage (derived vs. stored), and type hierarchy representation.

Limitations and Future Work

The authors acknowledge several limitations and outline future research directions:

  • Relational Database Constraint: The current TransFER prototype is limited to relational databases. The authors note that non-relational sites will still benefit from the standard query model framework but won't fully leverage the automated translation.
  • Incorporation of Standard Medical Terminology: A significant challenge is integrating a standard medical terminology and developing methods to automatically resolve differences with legacy terminologies.
  • Definition of Sound Temporal Semantics: The temporal aspects of clinical data and queries are crucial and require well-defined temporal constructs and mapping methods.
  • Specification of a Useful but Finite Reference Schema: Defining a reference schema that is rich enough for many CDS applications yet general enough to subsume most clinical databases is an ongoing challenge.

The authors suggest that knowledge-based methods and further investigation into temporal semantics are key areas for future work to make the standard query model framework practically useful.

Summary updated

Abstract

Many clinical decision-support applications are created in a centralized manner, but distributed widely for local use. When such applications include queries to electronic patient databases, the queries must be translated to conform to local database specifications. Because no well-defined standard model of clinical data exists, the translation process is ad hoc, costly, and error-prone. In this paper, we propose an abstract formalism, called the Standard Query Model Framework, for specifying a standard clinical data model and for supporting the automated and reliable translation of queries that appear in shared decision-support applications. We present the components of this framework, discuss their desirable features, and describe a prototype that we have developed for relational patient databases. We also highlight the outstanding research issues relevant to our approach.
BibTeX
@inproceedings{sujansky1994towards,
  author = {Walter Sujansky and Russ B. Altman},
  title = {{Towards a standard query model for sharing decision-support applications.}},
  booktitle = {PubMed},
  year = {1994},
}
Figure