The Design of Demonstration Studies - Evaluation Methods in Biomedical Informatics - page 193

TABLE 7.1. Summary of descriptive, correlational, and comparative studies.
Parameter Descriptive study Correlational study Comparative study

Goal of study To describe a To explore the relations To assign causality to the

resource in terms between variables relation between variables

of variables

Independent None One or more as One or more, with at least one

variables selected by investigator created by the investigator

Dependent One or more One or more as selected One or more as selected

variables by the investigator by the investigator

Logic of data Descriptive statistics Analysis of patterns of Continuous variables: ANOVA

analysis about dependent relationships among Discrete variables: contingency

variables variables table, ROC analyses

ANOVA, analysis of variance; ROC, receiver operating characteristics.

completely different from selecting a random sample of participants for a
survey, for example. These concepts are often confused.
The contrasts between descriptive, correlational, and comparative studies
are summarized in Table 7.1. Readers are also referred to Figure 4.3 in
Chapter 4.

Generic Issues in Demonstration Study Design

As we move into a more technical discussion of study design, we introduce
three key issues that always arise: defining the intervention, choosing
the participants, and selecting the tasks for the participants to undertake.
When evaluating therapeutic technologies such as drugs, the participants
are usually patients. For evaluating interventions such as health educa-
tion, practice guidelines, or information resources, the participants can
be care providers, departments, or even hospitals. 10,11 In studies where
the care providers are participants, we usually consider the patients these
care providers treat to be tasks and average the performance of each
care provider over a set of tasks.
When studying an information resource, evaluators should be aware that
its performance or impact may vary greatly among participants (e.g., clini-
cians may vary greatly in the information they obtain from the clinical lab-
oratory system, perhaps because of differing prior experience) or between
different kinds of tasks (e.g., the resource may only improve the manage-
ment of certain groups of patients, such as those in intensive care, irre-
spective of which clinician is looking after them). This means that evaluators
should be aware of both the range of participants and the range of tasks to
be included in their study. In laboratory studies the investigator often can
directly control both the participants included in the study and the tasks
with which they are challenged. In field studies, this level of control may
not be possible or desirable, as the identity of care providers who work in