Big-data predictive analytics offer the promise of better outcomes and lower costs for healthcare organizations, effectively allowing a patient to access the expertise of thousands of experts gained through treating millions of patients. But successfully deploying the technology isn’t always easy. How you plan for and introduce analytics is critical to acceptance by stakeholders and a willingness to take action based on the knowledge you generate.
The Locomotive from Nurnberg to Furth: The fear effect of technology
Disruptive technologies almost always elicit initial skepticism and even fear. When the first steam locomotive prepared to make its run from Nurnberg to Furth in 1835, people were concerned about noise and pollution and feared that human physiology might not support travel at speeds over 20 mph. Other useful-but-disruptive advances have also generated initial fear.
In research published over 30 years ago, I demonstrated how a lack of a sense of control generated fear when personal computers revolutionized computing at universities. Interacting with a black box that generates results as if by magic, without giving any control to the end-user, will always generate distrust.
Projects will fail if analytics technologies are perceived to usurp personal judgment and control over final decisions.
Presenting analytics to stakeholders as a tool they can use will empower them and pre-empt fear. Demonstrate how these new tools help healthcare professionals to quickly evaluate risks, potential outcomes, and what-if scenarios. Predictions, recommendations and prescriptions derived from analytics need to come with reasons why a particular risk is indicated and how recommended actions will affect outcomes.
Avoid alarm-fatigue with unambiguous, actionable alerts
People cease to pay attention when alarms are too frequent or information doesn’t present clear options for action.
Enhance rather than add to existing processes, screens and alarming rules and ensure that important information is unambiguous, actionable and consistent with existing work flows. Think through where analytic results will be used, identify benefits and ROI and make certain that information is actionable. For example, Dr. John Cromwell implemented a system at the University of Iowa Hospitals and Clinics which sends real-time, actionable risk information to the operating room that is helping surgeons avoid post-surgical infections.
Don’t add to the onslaught of computer work
General surgeon Jeffrey Singer recently noted in the Wall Street Journal that rigid electronic health records systems promote “tunnel vision in which physicians become so focused on complying with the EHR worksheet that they surrender a degree of critical thinking and medical investigation.”
Analytics technology should be entirely hidden, yet deliver reliable information about risks, best next action and alternatives. Don’t require medical professionals to complete yet another computer screen.
Know the end point and how to measure results
One of the most important things to consider before embarking on any IT project is to clearly establish what the completed project would look like and how to measure success. Avoid projects that are attractive because of the “cool” technologies involved without clear definitions of success and ROI.
Think about what ideal results look like; who would use them and how; and how something of value would be created. Involve key stakeholders and end-users to reflect their concerns and perceived barriers to success. Once you know how success is exactly and operationally defined, everything else follows, such as where to look for what data, how results are delivered, what level of integration, training, operational changes or new resources/personnel are required.
Decide what data you need
Data acquisition and preparation is always the most time-consuming and difficult part of any advanced predictive analytics project. EMR systems are mostly closed and data from different sources and repositories use different labels and metrics for the same measurements. For example, reports from different laboratories may use different formats, scales and nomenclatures.
Before you begin, think through whether you need immediate ROI for a specific project or a more complete analytics solution. A project-specific approach allows you to go after low-hanging fruit using data that are easiest to get and integrate; a longer term approach, to support diverse projects, requires building a robust general data-analysis warehouse with a Master Patient Index, terminologies and translation logic, and incorporates adapters to allow integration with relevant data sources.
Finally, there is governance, though ideally this would come first. Often overlooked, governance is important for two major reasons. First, many projects initially succeed, but then fold after the project champion departs, leaving nobody who knows and understands how it all works and where the data are. Second, regulatory oversight and scrutiny will become important when analytics affect real patient outcomes.
A role-based system with lifecycle management, version control, audit logs, approval processes, etc., will solve the issue of departing champions as well as the need to document how predictive models were built, validated, approved and implemented. Good examples of this can be found among pharmaceutical and medical device manufacturers, which have for years incorporated mature governance features to meet these challenges.