The syndrome of inappropriate overconfidence in computing: an invasion of medicine by the information technology industry?

Journal of American Physicians and Surgeons, Summer, 2009 by Scot M. Silverstein

I have written that the rigor of medical science is almost completely lacking in the very field that ironically is touted as revolutionizing medicine: health information technology (HIT). (1)

I arrived at this conclusion a decade ago, after observing harmful, ad hoc, and capricious decision-making by HIT leaders and vendors. Years of further observations have not changed my opinion.

Problems created by lack of scientific rigor in a cross-disciplinary, exploratory field such as HIT, which should be guided by the most rigorous scientific principles, are increasing in scope and severity. In fact, I believe medicine is suffering an unhelpful "cross-occupational invasion" by the IT industry, with the IT industry's best interests as the primary driver, not medicine's.

We've recently been informed that physicians and patients have been test subjects without consent in a mass vendor-driven experiment to develop future HIT software versions. In the style reminiscent of the Tuskegee Experiment, these computer experiments have been conducted with full approval of healthcare organization executives, while all of the legal burdens have been shifted to the test subjects in an environment of contractual secrecy about HIT defects? (2)

The rationale for this "hold harmless" environment and censorship regarding HIT defects has itself been unscientific and lacking in the critical thinking imperative in biomedicine. The assumption that HIT vendors should be held harmless for IT related morbidity and mortality (the rates of which are not known, thanks in part to the secrecy clauses), because clinicians are "learned intermediaries" between silicon chips and human flesh, is faulty.

This assumption arises, at best, from the assumption that HIT is an innocuous technology, capable only of good, and that any minor glitches should be recoverable through infallible cognitive skills of "learned" clinicians.

The belief that HIT errors can be caught with 100 percent reliability by physicians and other clinicians, learned as they might be, is in fact dangerous and risible. It reflects an ill-informed layperson's understanding of the complex, poorly bounded, unpredictable, often improvisation-driven environment of medicine. (3)

It is an assumption based on a view of HIT as an "inventory system" of medical data, not an overseer of all aspects of medical care. It is a view devoid of knowledge of medicine or of informatics, which inform us that technology malfunctions interact with other local and/or systemic issues. These include such factors as the hectic workday of clinicians, the role of trainees in patient care, cognitive overload caused by HIT's all too common "mission-hostile user experience," and so forth. These unpredictable interactions can cause medical error.

The possibilities for causing error are in fact endless, and the assumptions of the industry about HIT's unmitigated beneficence were certainly not arrived at scientifically. Wishful thinking, profit motive, hope, faith, anything but scientific rigor, were in play.

Beyond the risk of medical error, the phenomenon of "cybernetic mysticism," the belief that computers are almost magical devices, is creating worse problems. Have we suffered a complete breakdown in scientific thinking with regard to the electronic health record (EHR) and other features of HIT? I believe we are approaching that point in our current irrational exuberance over HIT, evidenced by a plan to invest tens of billions of dollars and impose economic "incentives" and "penalties" (a.k.a. government force) in a push for universal HIT by 2014.

The goal of 2014 is itself overly optimistic, based on projections in a study of HIT diffusion (4) as well as the common sense of those who've worked in HIT in real world settings. I read announcements like this one with trepidation:

"The goal," Sebelius said, "is to provide every American with a safe, secure electronic health record by 2014." The nominee also endorsed efforts to use data gleaned from electronic medical records to conduct "comparative effectiveness research to provide information on the relative strengths and weaknesses of alternative medical interventions to health providers and consumers." (5)

The use of EHR data to reliably detect uncommon (but strong and discrete) signals from a single drug or treatment is itself a medical informatics "grand challenge." A grand challenge is a fundamental scientific or technologic problem whose solution requires significant increases in current levels of scientific knowledge and/or technical capabilities. Solutions should improve the health of the population and be achievable within a decade. (6) An example would be finding the association of rofecoxib (Vioxx) with myocardial infarction earlier than we did, via an EHR-based automated postmarketing surveillance process.

Performing this type of postmarketing surveillance is in fact a "grand challenge" because of the uncontrolled nature of aggregated EHR data. The statistical methods needed to reliably pull warning signals out of such data for even a single drug are experimental. The problems are formidable if one is to perform such investigations in a scientifically sound manner. For example, to accomplish postmarketing surveillance of a single drug, innovative statistical models and methods for analysis of extremely large datasets (large number of observations or large number of dimensions), an active area of research, will be necessary to supplement and replace more simplistic methodologies such as adverse event frequency comparisons. Research in computational statistics, for example, involves experimental development of visualization and computationally intensive methods for mining large, nonhomo-geneous, multidimensional datasets so as to discover knowledge in the data. (7,8)