The VA and the military continue to have problems managing large IT projects and particularly those relating to EHRs (see, for example: Why the Military and the VA Healthcare Systems Are Not Amenable to Change; Epic Partners with IBM for Military EHR Proposal; This May Be a Problem; VA-Cerner EHR Deal Paused Over Interoperability Concerns). The U.S. Coast Guard turns out to be no exception to this rule based on a recent story about the travails of this branch with an attempted Epic EHR install (see: U.S. Coast Guard’s $67 Million EHR Fiasco). Below is an excerpt from the article:
In late January, the U.S. House of Representatives’ Subcommittee on Coast Guard and Maritime Transportation held a hearing to review the United States Coast Guard’s $14 million, five-year electronic health record (EHR) system project. The project, which began in September 2010, ballooned into a $67 million fiasco that the USCG [U.S. Coast Guard] finally ended in September 2015. But the Coast Guard didn’t officially confirm its termination until April 2016. At the time, the USCG public affairs office vaguely explained that there were concerns about whether the project could be completed in a reasonable time and at a reasonable cost....Finally, at the House subcommittee meeting, the reason for the agency’s uncommunicativeness became crystal clear: sheer embarrassment....[T]he project was originally envisioned and sold as an EHR modernization effort based on Epic’s EHR software that would begin deployment to the Coast Guard’s 41 clinics and 125 sick bays sometime in late 2011. It soon morphed into a “whole new Coast Guard Integrated Health Information System.”....Soon, 25 different vendors were working on the IHiS, at a total cost of $56 million, without active USCG management oversight. The GAO’s report...revealed a cringe-worthy litany of poor or non-existent system development, management, and governance practices over the duration of the IHiS project.... A memo from last May by Epic also details myriad operational blunders that explain why the project tanked....According to the GAO, the $67 million reported cost is probably a huge understatement of the actual cost. The Coast Guard, in computing its cost of failure, “did not include labor costs for the agency’s personnel (civilian or military) who spent approximately 5 years managing, overseeing, and providing subject matter expertise on the project.
There is a large literature about why large IT projects fail. A 2016 Forbes article summarized the major reasons for such failures (see: Are These The 7 Real Reasons Why Tech Projects Fail?). Here are the main reasons listed in this article:
- Poorly defined (or no defined) outcome.
- Lack of leadership.
- Lack of accountability.
- Insufficient communication.
- No plan or timeline.
- A lack of real-world user testing before launch.
- Solving the wrong problem.
It's useful to compare this list with the Coast Guard project problems enumerated in the excerpt above. The Coast Guard project seems to have failed primarily because of (1) scope creep; (2) incompetent project management with multiple vendors; and (3) leadership failure. Although the perhaps understated cost of the aborted EHR project was $67M, its important to keep in mind that it only encompassed clinics and "sick bays" without any hospital involvement.
For me, one of the major future changes in diagnostic medicine will be the use of predictive analytics based on deep learning and big data (see: Integrated Clinical Research Databases: A New Way to "Monetize" Clinical Data; What Are the Consequences of Big Tech Entering the Healthcare Market?). This new science will enable the prediction of future "outcomes" for patients. This trend was emphasized by former Google CEO Eric Schmidt in a lecture at HIMSS 2018 (see: HIMSS 2018: ‘Run to the Cloud,’ says Former Google CEO Eric Schmidt). He referred to the trend as leveraging the power of predictive analytics. Below is an excerpt from the article:
Chief among innovations — what Schmidt called “the really powerful stuff at the edge” of his work — are predictive algorithms, he said. While it’s one thing to be able to classify, it’s another thing entirely to be able to predict next steps. “We have physicians within our company who believe that if these algorithms for prediction work, we can predict outcomes in the ER, for example, 18 to 24 hours earlier than any other observation system,” Schmidt said. “We can’t predict our own fates, but machines can. That’s what I want as I age: I want the computer and all this work I’ve done over my whole career to make sure that I have a healthy life.”
It's interesting that Eric Schmidt chose to emphasize the value of predicting outcomes in the ER during his recent lecture at HIMSS. ER visits are a major source of revenue for hospitals in that they account for a large percentage of inpatient admissions (see: The Hospital Emergency Department Is Now the Admissions Department). ER visits can also result in malpractice lawsuits for a hospital (see: AMA Study: EPs Rank 5th in Liability Claims Frequency). One common challenge are ER patents presenting with a history of chronic headaches. ER physicians may order expensive brain scans with computed tomography (CT) or magnetic resonance imaging (MRI) for such patients, a practice often viewed as defensive medicine (see: Signs and symptoms of patients with brain tumors presenting to the emergency department). Emergency physicians are thereby caught between a rock and a hard place, obliged to rule out serious disease but also subjecting themselves to criticism for over-utilization of services.
The ER is an example of a hospital unit where predictive analytics could potentially provide guidance for the workup of ER patients presenting with headaches. The analysis of many thousands of records of patients presenting with a history of headaches and perhaps no other obvious symptoms could yield a useful set of recommendations that would avoid over-utilization of services. The best way to diagnose early brain tumors may depend on some subtle differences in routine blood tests or physical changes that are not recognized at the present time. I like the quote used by Schmidt in the excerpt above relating to predictive analytics: We can’t predict our own fates, but machines can.