Newsletter

Development and Implementation of Local Cardiovascular Databases

Collection of valid, reliable, and relevant clinical data is important in an institution of any size-be it academic or private practice-for the following reasons:¹

• High-quality data are essential to document the efficacy, quality, and cost-effectiveness of the procedures that we perform.
• Data can be used to monitor and document improvement in quality of health care (or decline in quality caused by decreases in funding).
• A database can yield scientific data, suitable for publication.

Designing the Database

In designing or improving a local cardiovascular database, an essential element is the epidemiological/biostatistical consultant. Some key questions and issues that may be considered include the following:²

1. What is the function of the database? The developer should determine the eventual use of the data: quality improvement, cost effectiveness, publications, or another purpose.
2. What is the focus of the database, and what aspects of patient care are to be captured? This is usually defined by the intended analyses, e.g., outcome assessments, including clinical, economic, or functional (quality of life) endpoints.
3. What form will the database take? Is it designed to achieve its defined goals without becoming too labor-intensive and cumbersome? Methods of data entry, storage, retrieval, and analysis must be considered. In the best situations, data collection is unobtrusive, systematic, and computer-based for later analysis.
4. How flexible is the database? Is it possible to assess new techniques in cardiovascular medicine? Can new variables be added and unimportant or outdated data deleted?
5. How easy is data extraction and analysis?
6. Are data definitions standardized? Does the database allow comparisons within your own institution as well as with other institutions collecting similar data? Standardization is vital for comparing outcomes between centers.
7. Have you considered front-end quality improvement (error checking)? Early notification of data entry personnel if there is a problem is important. If possible, regular review of the variables and computerized fields and updating of software should be considered. Training and retraining data entry staff is essential to maintain data quality. In other words, "Garbage in is garbage out".
8. Can you get input from colleagues who may use the database later (ideally, broad-based input from potential end-users during the design phase)?
9. Is the selection of software appropriate to support the functions listed above?

Queries

A specific research plan is necessary when designing a formal query for a database. If appropriate, a brief literature review should be completed, and a list of references attached to the plan. The broad objective and specific aims of the analysis should be defined, along with the primary and any secondary hypotheses. Inclusion and exclusion criteria should also be listed. Endpoints and predictor variables should be defined and identified by the appropriate and exact fields in the computer.

Analyses

Methods used for the statistical analysis will depend upon whether individual variables are discrete or continuous and on other characteristics of each variable (e.g., normal versus skewed distribution). Univariate statistical analyses are done first, after which significant associations can be optimally assessed with multivariate modeling (e.g., logistic regression, Cox proportional hazards regression).

Observational databases have limitations, as they are typically used for retrospectively identifying associations between defined risk factors and outcomes. This often necessitates much larger sample sizes than are needed for well-designed prospective evaluations. Also, important risk factors, e.g., preoperative inability to ambulate or intraoperative identification of severe proximal aortic atherosclerosis are often omitted from even the most sophisticated databases. Finally, in the case of local databases, site-specific factors may prevent generalization of results and conclusions. Still, with adequate sample size and proper statistical modeling, retrospective studies are appropriate for discovering important occurrences and associations, which can then be studied prospectively. Other types of studies emphasize "process measurement"; these have become popular for analysis of practice patterns, costs, and outcomes of medical care.²

Summary

By analyzing data regarding risk factors, medical events, complications, and expense of cardiovascular procedures, we may be able to substantially impact the quality, accessibility, and cost of health care.

References:

1. Grover FL: The Society of Thoracic Surgeons National Database: Current status and future directions. Ann Thorac Surg 1999;68:367-73.
2. Tardiff BE, Jollis JG, Lubarsky DA: The use of information systems and large databases in cardiovascular medicine. In: Tuman KJ, editor: Outcome Measurements in Cardiovascular Medicine. A Society of Cardiovascular Anesthesiologists Monograph. Lippincott Williams & Wilkins, 1999.

 

Nancy A. Nussmeier, M.D. and
William K. Vaughn, Ph.D.
Texas Heart Institute
Houston, Texas

© Society of Cardiovascular Anesthesiologists
Questions or comments? Please send email to webmaster@scahq.org