//penelope/nlg/publications/all-hands2005/ah2005queries.dvi

Intuitive Querying of e-Health Data Repositories
Catalina Hallett, Richard Power, Donia Scott
Abstract
At the centre of the Clinical e-Science Framework (CLEF) project is a repository of well organised,detailed clinical histories, encoded as data that will be available for use in clinical care and in-silicomedical experiments. An integral part of the CLEF workbench is a tool to allow biomedical re-searchers and clinicians to query – in an intuitive way – the repository of patient data. This paperdescribes the CLEF query editing interface, which makes use of natural language generation tech-niques in order to alleviate some of the problems generally faced by natural language and graphicalquery interfaces. The query interface also incorporates an answer renderer that dynamically gener-ates responses in both natural language text and graphics.
Background
databases involves expressing queries in a The Clinical e-Science Framework (CLEF) aims language that is understood by the database at providing a data repository of well organised management system (typically SQL). Direct SQL clinical histories, which can be queried and querying requires specialist knowledge of the summarised both for biomedical research and both the query language and the structure of the underlying database, and – in the case of medical the query interface is to provide efficient databases – usually also knowledge of precise access to aggregated data for performing a variety of tasks, e.g., assisting in diagnosis be counter-productive to require this additional or treatment, identifying patterns in treatment, level of technical expertise of the clinicians and selecting subjects for clinical trials, monitoring biomedical researchers who want to access the the participants in clinical trials. The intended users of this service are clinicians, biomedicalresearchers, and hospital administrators.
Attempts to overcome this problem in user current domain is cancer; however, the framework interfaces to medical databases have traditionally in principle supports a wide range of clinical made use of graphical devices such as forms, diagrams, menus, or pointers to communicate to An analysis of free text queries written by the user the information content of a database medical professionals show that they are mostly (e.g., KNAVE (Shahar and Cheng, 1999) and very complex and often ambiguous. This makes TrialDB (Deshpande et al., 2001)), and research the design of the query interface to the CLEF shows that they are much preferred over textual repository particularly difficult, since our users query languages such as SQL, especially by will need to construct complex queries containing conditional and temporal structures.
empirical studies have reported high error rates by The CLEF repository of clinical histories domain experts using graphical modelling tools currently contains some 20000 records of cancer (Kim, 1990) and a clear advantage of text over graphics for understanding nested conditional or ICD, and is implemented as a relational database that stores patient records modeled However, it is also well-known that queries on the archetype for cancer developed at UCL expressed in free natural language are sensitive two general types of queries, as exemplified ungrammaticalities) or processing (at the lexical, syntactic or semantic level). A further drawback of natural language interfaces to databases is that such systems normally understand only a subset of natural language, and it is not always clear to casual users which are the valid constructions and whether the lack of response from the system is due to the unavailability of an answer or to an unaccepted input construction. On the positive side, natural language is far more expressive than SQL, so it is generally easier to ask complex questions and manipulate temporal constructions using natural language than using a database In the first example, the expected answer is a comparison between a certain statistical The CLEF query interface
measure (in this case, percentage) applied on twogroups of patients differentiated by the treatment The CLEF query system is designed to answer questions relating to patterns in medical histories a statistical measure (average) computed for a over sets of patients in the data repository.
certain parameter (number of investigations of The current interface is designed for casual type ”body scan”) of a group of patients with some and moderate users who are familiar with the semantic domain of the repository but not withits technical implementation (e.g., clinicians, For either of these queries, the attributes medical researchers and hospital administrators).
involved in constructing the query can vary within For the reasons we described above, the guiding a certain range: any statistical measure can be principle in the design of our interface is that its used, the differentiating parameter could be the use requires no prior knowledge of the structure diagnosis instead of the treatment, etc.
Additionally, there are a number of variations to access languages such as SQL, no familiarity these two main types of queries. For both types, with medical codes, and only minimal prior the user may ask for simple assessment queries repository is not through SQL, or graphics or free by interacting with an automatically-generated Natural Language feedback text (currently only There are also cases where several similar the WYSIWYM technology developed by Power queries are combined into one more complex et al (Power et al., 1998), allows users of the profile described above to construct in an intuitive way, unambiguous, syntactically correct, complex For all these queries, there is practically no limit to the complexity that can be achieved description can in fact be a conjunction or disjunction of diagnoses, and the same applies for every concept included in a query. Therefore, theuser can construct queries such as: Query analysis
Types of queries
An analysis of real queries from clinical trials and invented queries supplied by clinicians identified interface for editing the conceptual meaning of a supported by the query editor, and they are The WYSIWYM interface presents the contents not considered separate types of queries, nor of a knowledge base to the user in the form of the content of the knowledge base is a yet to Modeling queries
be completed formal representation of the user’s For presentation reasons, queries have to be decomposed into constituents that can be easily a natural language text that corresponds with edited by the user. By way of exemplification, let the incomplete query and guides them towards us consider the query type (1). There are three editing a semantically consistent and complete elements to the query: the set of relevant patients, defined by a problem; the partition of this set control the interpretation that the system gives according to treatment; and the further partition according to outcome, from which the percentages a basic query frame, where concepts to be instantiated (anchors) are clickable spans of text sentences, we consider a format in which these with associated pop-up menus containing options for expanding the query. For example, one canstart constructing a query that asks for a group of Relevant
subjects:
patients fulfilling some conditions by editing the Treatment profiles:
Relevant subjects:
Outcome measure:
Treatment
profile:
received [some treatment]
Outcome: [measure] of [patients with
This breakdown allows the following basic Relevant subjects:
Once the user selects an anchor and a new value for the concept represented by the anchor, the Treatment profiles:
semantic representation of the query is updated and a new text is generated on the basis of combination of features or events of the same Outcome measure:
type, thus allowing for complex queries, with nested conditional structures to be built. Some concept instances can also be typed in manually, Each of the bracketed elements are complex which is useful for numerical values or other fields descriptions that model the concept definition in with unpredictable content, such as names. This the CLEF archetype. For example, the concept is also a way of enriching the ontology with new diagnosis consists of the following obligatory and optional components: tumour name, locus, editor with a partially constructed query.
type (metastatic, primary, secondary) and TNM staging code. Each of the subcomponents can be extended through boolean operations (negation, selection over the feedback text is treated as an intermediate query, which is sent to the DBMS.
In return, the DBMS will transmit to the interface Query editing interface
a feedback answer. At this point, the feedbackanswer is a set of paired values representing the General features
number of patient records that match the query Conceptual authoring through WYSIWYM editing and the percentage from the total number of (Power et al., 1998) alleviates the need for expensive syntactic and semantic processing of patient records by sex, which was considered a the queries by providing the users with an good discriminatory feature. For example, for an intermediate query such as Number of patients of some real queries that could be given multiple over the age of 60., the feedback answer could be 100 records (20% of 500), 55 men (55%), 45 ambiguities are presented below, along with the solution provided by the CLEF query interface.
As a further consistency checking mechanism, When the phrase describing a relevance set the interface provides an additional rendering of includes a conjunction or disjunction, there may the query in running text, which is performed be ambiguity over whether the intended query is once the editing of the feedback query has single or multiple. Compare these three patterns: been completed, the user is presented with an alternative natural language query corresponding to the structure that has been edited (output schematic to allow for more intuitive editing, the output query resembles in every respect a free text query, thus being more natural and easier to read.
The natural language interface is database- Example 6a is likely to be interpreted as two separate queries, while the others are ambiguous.
knowledge of the database structure.
Disjunctions like 6c occur often in real life structure of the database is not only completely transparent to the user, but also to the interface developer: changes at the database level require no changes in the query editor. Queries can be saved for later re-use, which is particularly useful for frequent users who formulate queries with Dealing with ambiguities
Since the processing of an edited query is deterministic and transparent to the user, the main challenge is not to construct valid database queries from edited queries but to ensure that the query myelodysplastic syndrome only and for acute the user is editing corresponds to the intended myelogenous leukaemia caused by bad prognosis meaning. Therefore we want to ensure that the myelodysplastic syndrome, or if it make sense to layout of the query conveys one meaning only to give a single answer lumping these two groups The process of defining a specific unambiguous layout for the queries was based on the analysis feedback texts by using different realisations for conjunctions/disjunctions that imply multiple Specifying constraints and temporal
relevance sets, and conjunctions/disjunctions that relations
do not. For example, we use bulleted lists for the Guiding users towards editing correct and former, and conjunction words (and, or) for the complete queries is essential and is one of the main points where our approach improves on classical natural language query interfaces. This is achieved by defining and implementing a system of semantic static and dynamic constraints.
Static (or ontological) constraints relate to
the structure of the queries as defined in the query prognosis myelodysplastic syndromefor at least 6 months model. This includes specifying the super-classof an instance (for example, the anchor cancer can only be instantiated with names of cancers), its type (for example, age is numeric and editable, while cancer is a static string) and its status Dynamic constraints are triggered at runtime
by the user selection of certain instances. Most constraints simply serve the role of restricting In 9a we have two relevance sets; in 9b we have the user selection so that the resulting query Similar ambiguities can be found when several however, allowing the user to construct queries treatment profiles are mentioned, or several outcome measures. In each case, the ambiguity can be avoided in the WYSIWYM feedback texts Dynamic contraints can be either conceptual, the same way as before, by using bullets to mark which are compiled from a medical knowledge base and represent depedencies between medicalconcepts (for example, nephroblastoma is a type of kidney cancer, so users shouldn’t be allowed properties. A description can be elaborate either to query for nephroblastoma in the left breast), or because it contains many boolean operators, numerical (for example, patients between 60 and 30 years of age is a disallowed construction).
boolean combinations in running prose means As medical records mirror the evolution in time that the scope of the operators can become of a patient, it is important to be able to access ambiguous to the user. For this reason, layout the patient’s status at a certain point in time. The is used to present boolean combinations more easy specification of time in natural language is an important advantage of natural language queryinterfaces over graphical interfaces. All temporal concepts in the medical record are stamped with a valid time stamp, i.e. the (precise1) moment in time when the event took place. Typically, a time interval is represented as a pair of start and end dates, where start and end are discrete time values of a certain predefined granularity.
interface associates specific linguistic expressions to time intervals. For example, between [date 1] and [date 2] is interpreted as a closed interval [date 1, date 2], in [this year] is interpreted as [01/01/this year, 31/12/this year]. Such time expressions cover most temporal queries, such as: patients diagnosed with cancer before 1999, 1to a certain level of granularity imposed by the representation of time instances in the database Gender Age adenocarcinoma small cell carcinoma squamous cell carcinoma death
patients who received chemotherapy within 5 Conclusions and further work
We have presented in this paper a query interfaceto a repository of patient records which makes Answer generation
use of natural language generation techniques.
A typical result set received from the DBMS The query interface allows the editing of complex consists of lists of patients that fulfilled the queries and is a viable alternative to natural requirements of the query, for each patient having language interfaces and visual query interfaces specified the age, gender, and the values for provided in textual format using natural language query such as Select all patients between the generation techniques and also as tables and ages of 30 and 60 with a clinical diagnosis charts. The main features that set our approach of malignant neoplasm of bronchus or lungs apart from other querying interfaces to medical and histopathology diagnosis of adenocarcinoma, small cell carcinoma or squamous cell carcinoma, • users require little training for using the who were alive after 10 years of the diagnosis, The result set is processed in such a way as to • a set of semantic constraints are used to guide allow the rendering of various groups of patients according to the age/gender breakdown and each only, therefore incorrect queries are not individual query term. For each individual search parameter, the data is split into a dynamicallydetermined number of age groups, and for each • the constructed queries are unambiguous, age group the number of patients is further split since ambiguity is dealt with in the editing processed is presented to the user in three types of format: tables, charts and text. Each individual chart also contains an automatically generated • the query interface has wider applicability caption that explains the content of the chart.
The captions are generated using template- based techniques, where fillers are provided by the same result set that was used for generatingthe chart. For the bar chart in Fig. 3, a fragment Whilst the query editing interface is fully of the explanation provided in the caption reads: implemented, extending the range of queries This chart displays the distribution of patients supported is an ongoing effort. This is performed in 4 age groups according to their gender and in parallel with an evaluation of the usability histopathology diagnosis. 42 patients have been and user-friendliness of the interface.
returned as a result to your query: expected that the evaluation will help formulate -in the 29-38 years age group there were 1 an extended range of queries and improve the patients (0 men and 1 woman): all patients were editing interface. The improved query interface diagnosed with adenocarcinoma. [.] will provide means of interactively defining -in the 49-58 age group, there were 27 patients default values for instances that support them (14 men and 13 women): 11 were diagnosed (for example, one may want to default all index with adenocarcinoma, 5 were diagnosed with events to the date of the first diagnosis). We also squamous cell carcinoma, 11 were diagnosed with plan to extend the range of temporal operators to include, for example, trend operators for clinical Figure 3: Generated bar chart: histopathology diagnosis/age/gender breakdown time-oriented clinical data. In Proceedings blood pressure, stationary haemoglobin count) and define independent variables for reportingstatistical results (such as age groups, sex,education level).
References
A. Deshpande, C. Brandt, and P. Nadkarni.
Meeting the needs of clinical studies.
Journal Informatics Association, 9(4):369–382.
Dipak Kalra, Anthony Austin, A. O’Connor, D. Patterson, David Lloyd, and DavidIngram, 2001. Design and Implementationof a Federated Health Record Server,pages 1–13. Medical Records Institute forthe Centre for Advancement of ElectronicRecords Ltd.
Y. Kim. 1990. Effects of conceptual data modelling fomalsms on user validationand analyst modelling of informationrequirements. Ph.D. thesis, University ofMinnesota.
M. Petre. 1995. Why looking isn’t always Multilingual authoring using feedbacktexts. In Proceedings of 17th InternationalConference on Computational Linguisticsand Association for Computational Linguistics(COLING-ACL 98), Intelligent visualization and exploration of

Source: http://www.cis.upenn.edu/~rjprasad/papers/ah2005queries.pdf