Data mining is a relatively new technique used in clinical research in anesthesiology. It is, effectively, a retrospective chart review process conducted on a large computer database of patient information. As with all retrospective analyses of medical record data, it suffers from a lack of adequate controls and, therefore, it is generally assumed to provide interesting information regarding the association of events but not to prove causation. The results, however, can provide specific direction for the design of prospective, randomized, controlled studies, which can provide more definitive conclusions on a specific question. Despite these limitations, when one is asking questions about rare events, mining of large databases may be the only feasible way of conducting the research. Often these studies can report important, and even provocative, findings (1).
Computerized anesthesia record keepers have been available for more than 20 yr, although their adoption into our specialty has been slow (2). Over the past decade these systems, which were originally focused on reproducing the intraoperative anesthetic record, have gradually evolved to become perioperative information systems. Perioperative systems not only include electronic preoperative, intraoperative, and postoperative records but are linked with the hospital’s many databases and are therefore linked to the patient’s comorbidities, laboratory values, and outcomes data, i.e., length of stay, postoperative morbidity, patient satisfaction, and financial information (charges, cost, and revenue) (2–6). It is now possible for a perioperative system to electronically link what we do to patients with the patients’ outcome. In this issue of the journal, Reich et al. (7) have done exactly that. They looked at a common physiologic occurrence in anesthesia (postinduction hypotension) and tried to determine what, in the patient’s preoperative history or in the drug management of the induction, was associated with the most frequent incidence of postinduction hypotension. Again, because this is a retrospective, electronic review, it does not definitely prove a cause-and-effect relationship, but the authors were able to review 5244 records, all of which were clearly documented (another advantage of an external record), to reach a conclusion. Propofol is associated with an increased incidence of clinically significant hypotension in ASA ≥III or greater patients who were older than 50 yr of age. This hypotension was, in turn, associated with increased hospital stay or mortality (7). That hypotension is associated with induction doses of propofol is not new information, but recent work has elucidated the mechanism as it applies to the aging cardiovascular system (8,9). In addition, Monk et al. (10) have reported an association of intraoperative hypotension with an increased mortality during the first year after surgery. The implication is that there may be a causative association when, in fact, it may be more of an indicator of patients at risk. That is, our anesthetic care is also providing the patient with a physiologic stress test that may predict 1 yr of survival.
Although everyone may not agree with the conclusion of Reich et al. that propofol should be avoided as an induction drug in ASA ≥III patients who are older than 50 yr of age, one must be impressed with the potential of the process used in this analysis. As of January 2003, 14 of the 143 academic training programs in the United States used perioperative information systems (email survey conducted by Kevin K. Tremper, PhD, MD and Amy Shanks, MS, Department of Anesthesiology, University of Michigan Health System). At that time, 30% of the department chairs anticipated they would be implementing a system within 2 yr. Even if this has not occurred, the planning for implementing systems in our specialty has significantly increased over the past few years. This parallels the desire to computerize other aspects of the medical record. It is surprising that anesthesiology, which was one of the first specialties to develop an electronic record, may be one of the lagging specialties in implementing an electronic patient care system as a routine. Nevertheless, it is clear that we will be soon documenting all of our care electronically. Given the limited number of systems available, many institutions will be using the same system with the same database structure. This should enable multiple institutions to pool data such that the study conducted here by Reich et al. could have been conducted on 52,000 or 520,000 anesthetics. These linked databases containing terabytes of information on physiologic outcomes will provide tremendous new opportunities in clinical research. This sudden availability of data is analogous to the explosion of information in molecular biology with the advent of the gene chip. Given the granularity of physiologic data that anesthesiologists collect, we may be considered the phenotype doctors for acute care, physiology and pathophysiology, and clinical outcomes, just as the molecular biologists are documenters of genotype. A new generation of phenotype database clinical researchers will need advanced skills in database management and statistics—the same skills required by the genotype data mining researchers.
As the vendors of anesthesia information systems accelerate their implementation in hospitals throughout the country, it begs the question: is it time for our specialty to develop a standardized preoperative assessment, intraoperative record, and postoperative visit? If we hope to pool our data on a large scale, it is important that we are collecting the same data elements. This standardization of data collection is a direct logical offshoot of the current efforts of a group known as the Data Dictionary Task Force (11). In 2002, a first step to standardization of terminology was taken by a task force of the Anesthesia Patient Safety Foundation: Data Dictionary Task Force (11). This project will add anesthesiology terminology to SNOMED (Systemized NOmenclature of MEDicine), which is the most commonly accepted list of medical terminology (12). The next step would be to add definitions to these terms and, finally, to develop standardized data fields for documenting our perioperative care.
The great potential for clinical research with database mining will probably never eliminate the prospective controlled study, but this new opportunity will be crucial in hypothesis generation and for answering questions regarding rare events that would be impossible to study in any other fashion. We may ultimately be able to have the denominator for the events that had been brought to our attention through the closed-claim studies (13). This ability to manage perioperative data will allow us to close the loop on how we care for patients and their outcomes; providing an effective way of continually improving the quality of our anesthetic care.
1. Silber JH, Kennedy SK, Even-Sloshan, et al. Anesthesiologist direction and patient outcome. Anesthesiology 2000;93:152–63.
2. Tremper KK, O’Reilly M, Kazanjian P, et al. A perioperative information system: design and implementation. Semin Anesth Perioperative Med Pain 2004;23:72–85.
3. O’Reilly M, Tremper KK, Baumgart P. Anesthesia information management systems: workflow engineering and return on investment. Semin Anesth Perioperative Med Pain 2004;23:151–7.
4. Kheterpal S, Tremper KK, O’Reilly M, Kazanjian P. A perioperative information system: technical considerations and tradeoffs. Semin Anesth Perioperative Med Pain 2004;23:125–32.
5. Reich DL, Krol M. Using AIMS data for quality improvement and research. Semin Anesth Perioperative Med Pain 2004;23:99–103.
6. Iaconetti DJ, Lussos SA, Koch J, Falcone C. Anesthesia information management systems and operating room management: the Inova Fairfax Hospital experience. Semin Anesth Perioperative Med Pain 2004;23:104–14.
7. Reich DL, Hossain S, Krol M, et al. Predictors of hypotension after induction of general anesthesia. Anesth Analg 2005;101:622–8.
8. Schnider TW, Minto CF, Shafer SL et al. The influence of age on propofol pharmacodynamics. Anesthesiology 1999;90:1502–16.
9. Memtsoudis SG, Andrew HS, Heerdt PM. Autonomic mechanisms in the age-related hypotensive effect of propofol. Anesth Analg 2005;100:111–5.
10. Monk TG, Saini V, Weldon C, Sigl JC. Anesthetic management and 1-year mortality after noncardiac surgery. Anesth Analg 2005;100:4–10.
11. Monk TG, Sanderson I. The development of an anesthesia lexicon. Semin Anesth Perioperative Med Pain 2004;23:93–8.
12. SNOMED® International, a division of the College of American Pathologists. Available at: www.snomed.org
. Accessed June 23, 2005.
13. Cheney FW. The American Society of Anesthesiologists Closed Claims Project: what have we learned, how has it affected practice, and how will it affect practice in the future? Anesthesiology. 1999;91:552–6.