The Power and Pitfalls of Big Data Research in Obstetrics and Gynecology: A Consumer's Guide

Abstract

Importance: Research in obstetrics and gynecology (OB/GYN) increasingly relies on "big data" and observational study designs. There is a gap in practitioner-relevant guides to interpret and critique such research.

Objective: This guide is an introduction to interpreting research using observational data and provides explanations and context for related terminology. In addition, it serves as a guide for critiquing OB/GYN studies that use observational data by outlining how to assess common pitfalls of experimental and observational study designs. Lastly, the piece provides a compendium of observational data resources commonly used within OB/GYN research.

Evidence acquisition: Review of literature was conducted for the collection of definitions and examples of terminology related to observational data research. Data resources were collected via Web search and researcher recommendations. Next, each data resource was reviewed and analyzed for content and accessibility. Contents of data resources were organized into summary tables and matched to relevant literature examples.

Results: We identified 26 observational data resources frequently used in secondary analysis for OB/GYN research. Cost, accessibility considerations for software/hardware capabilities, and contents of each data resource varied substantially.

Conclusions and relevance: Observational data sources can provide researchers with a variety of options in tackling their research questions related to OB/GYN practice, patient health outcomes, trends in utilization of medications/procedures, or prevalence estimates of disease states. Insurance claims data resources are useful for population-level prevalence estimates and utilization trends, whereas electronic health record-derived data and patient survey data may be more useful for exploring patient behaviors and trends in practice.

FIG. 1

Classifying study design. *If the investigator assigns the exposure, as in the case of a nonrandomized treatment versus control group study, then the study design would be classified as having an investigator-assigned quasi-experimental design. If the investigator does not assign the exposure in a treatment versus control group study, then the study design would be classified as an observational quasi-experimental design. An example of this is natural experiments, where some change like a new clinical guideline is enacted and the investigator compares the outcome on groups affected differently after the change.

FIG. 2

A 4-step approach for identifying and interpreting observational studies.