The Study Design Typology

Introduction

Goal of the Typology

The goal of the Study Design Typology is to classify any quantitative human study into one of 8 clearly defined study design types, each of which has a distinct set of biases and interpretive pitfalls that affect how evidence from these studies should be used. This typology is meant to support the critical appraisal of evidence and the classification of new and ongoing research for scientific portfolio management and analysis.

How the Classification Works

We identified the most parsimonious set of unambiguous questions about the design of any study on individual humans, that when answered, will correctly classify that study into 8 main design types (darker grey bubbles below). These design types represent distinct approaches to human investigations. Four of the design types are for Interventional studies; four are for Observational studies. The questions ask about factual features of a study's design and require little to no subjective judgement. Using our typology, a study can be classified into a main interpretive context with only 3 to 4 questions. For each main design type, '''Additional Descriptors''' (see tables below) elaborate on secondary design and analytic features that introduce or mitigate additional biases and interpretive pitfalls. As the tables show, some Additional Descriptors apply only to some design types.

How this Typology Differs from Others

Some other design typologies include those of Hartling, et al and the Cochrane Collaboration. Here are some distinguishing features of our approach and our typology:

• Our typology is targeted to quantitative human studies only. We have a separate classification system for distinguishing quantitative human studies from population, non-human, "meta-" studies, and qualitative studies.
• Our objective is to classify studies into design types that have distinct interpretive concerns. Administrative or other distinctions are not included.
• This typology is part of the Ontology of Clinical Research. It follows best practices for building ontologies. For example, we use a tree structure for classification, in which each branch of the tree has exhaustive and mutually exclusive options. We have formulated the hierarchical structure to minimize the number of questions needed to be answered to fully classify a study (i.e., we optimized for the shortest path to a leaf node).
• Moreover, all the options under a branch are "children" of the "parent" to facilitate automated reasoning. For example, we place all interventional studies under one parent, and all multiple-group interventional studies under one parent under interventional studies. This contrasts with the Hartling classification, which does not hierarchically classify interventional and observational study types.
• The questions have been designed to require as little subjective judgement as possible to answer. The permissible answers are listed and are exhaustive and mutually exclusive. The results of a preliminary evaluation of this typology are presented in this paper. We welcome all feedback and look forward to improving this typology through community use and revision. Please email us with comments.

Please note that the tables above are images, so the hyperlinks currently are not functional. Update in progress.