Chapter 17: Neurophysiologic Decision Support Systems

The field of biomedical informatics can be defined as the scientific discipline that deals with biomedical information, data, and knowledge including their storage, retrieval, and optimal use for problem solving and decision making.¹ The fields of bioinformatics, imaging informatics, clinical informatics, and public health informatics are all grouped under this larger umbrella.^1,2 Most relevant to this chapter, clinical informatics is motivated by applied problems in clinical care encompassing both patient-specific information and knowledge-based information. The electronic medical record is an application based on patient-specific information, whereas MEDLINE, which is provided by the National Library of Medicine, is a key resource for health-related knowledge retrieval.² A clinical decision support system (CDSS) is a computer program designed to help clinicians make diagnoses or management decisions³ and often relies on both patient-specific and knowledge-based information.²

The goal of invasive neuromonitoring, or multimodality monitoring, is to provide neurophysiologic decision support at the bedside. Advanced monitoring techniques can provide real-time information regarding the relative health or distress of the brain. This information can be used to set physiologic end points to guide goal-directed therapy and thereby create and maintain an optimal physiologic environment for the comatose injured brain to heal.⁴ From a critical standpoint, multimodality monitoring as a clinical decision support system is still in its infancy. Patient-specific information from patient monitors and devices are not easy to obtain, manipulate, or visualize, and knowledge-based information is largely dependent on what is known by the clinician looking at the data. That said, the field is now starting to evolve rapidly and the realization of the potential benefits of multimodality monitoring, and CDSSs in general, is only going to increase. This chapter is designed to help you implement neurophysiologic decision support systems in your intensive care unit (ICU).

It is true that there is a paucity of CDSSs of any type operating in the ICU,⁵ and yet it may be the clinical setting in greatest need for such systems. Clinicians may be confronted with more than 200 variables⁶ during morning rounds, and yet people are not able to judge the degree of relatedness between more than two variables.⁷ Our ability to collect patient data far exceeds our intellectual capacity to understand it unassisted⁵ and greatly contributes to conditions of constant information overload that can lead to preventable medical errors.^7,8 Implementation of CDSSs to help us understand the clinical meaning of patient data is essential.^9,10

Generally speaking, there is a lack of clinical informatics infrastructure to support these systems. Adoption of electronic health record (EHR) systems is being promoted by the Department of Health and Human Services through the office of National Coordinator for Health Information Technology by means of forums and regulations. The American Recovery and Reinvestment Act of 2009 (ARRA) has also authorized centers for Medicare and Medicaid Services (CMS) to provide reimbursement incentives up to $44,000 for each eligible health care professional who uses a certified EHR in a meaningful manner. Starting in 2015, financial penalties will be levied until EHRs are utilized according to the meaningful use of the terms.^11-13 ICUs have been migrating from paper-based to computerized charting systems for over a decade.¹⁴ Despite all of these efforts, as of 2009, it is estimated that hospitals only have an EHR adoption rate of about 12%.¹²

At a minimum you should have two goals. First, you should have tools to evaluate the effectiveness of treatments that are initiated to modify physiologic end points or improve clinical conditions. If you drill a hole into a patient's head and place an intracranial pressure (ICP) microprobe into the patient's brain, you should be able to easily monitor the effectiveness of your treatment decisions. I have yet to meet a hospital administrator or information technologist who has disagreed with this point. For instance, clinicians should be able to easily answer these questions:

• How fast does ICP drop after mannitol administration?

• Are repeated mannitol doses just as effective as earlier doses, or are more frequent doses now required?

• Is it safe to continue to give mannitol? What is the osmolar gap?

• Does increasing blood pressure increase or decrease ICP?

• Does adding sedation or reducing temperature provide additional ICP reduction?

• Has medical management failed; is it time to consider a surgical option?

These basic questions are difficult to answer in the absence of neurophysiologic decision support tools.

Second, neurophysiologic decision support systems should elucidate the physiologic status of the patient and how such physiology impacts other metrics of brain health like brain oxygenation and/or cerebral metabolism. An application of this (discussed in Chapter 15) is the assessment of cerebral autoregulation status. Often understanding patient physiologic status provides a context for treatment interventions rather than leading to a specific intervention. For example, if a patient with a sub-arachnoid hemorrhage has elevated ICP in the context of disturbed autoregulation, increasing blood pressure to increase cerebral perfusion (eg, to counteract cerebral vasospasm) may risk further ICP elevations, thereby increasing the risk of tissue compression or cerebral herniation. In this context, knowledge of cerebral autoregulation status helps the clinician weigh the risk/reward profile of different treatment options rather than directly evaluating the effectiveness of a specific treatment.

There are other potential benefits that can be realized by CDSS if patient-specific data can be collected and provided to support these systems.^15,16 These include automatic early detection of secondary complications before clinical symptoms occur (eg, sepsis detection¹⁷), computerized implementation of clinical protocols (eg, administration of insulin¹⁸), and general support of clinical decision making (eg, clinical dashboard¹⁹). As ICU clinical information infrastructure improves, the potential benefits provided by CDSS will only increase in the future.

The FDA does regulate medical software as a device. One category of medical software is "software accessories," which is actively regulated by the FDA. Software accessories are attached to (or used with) other medical devices such as a patient physiologic monitor. For example, systems for digital analysis and graphical representation of electroencephalographic (EEG) data connected to EEG acquisition systems are FDA regulated. In contrast, it has been unclear how, or to what extent (if at all), stand-alone software should be regulated by the FDA.²⁰ With respect to neurophysiologic decision support systems, in general, all-in-one monitoring solutions connect directly with patient monitoring devices and, therefore, fall into the regulated FDA category of software accessory. In contrast, ICU-wide solutions often start by collecting data from patient monitors and storing it in a database. In a second separate step, data analysis and visualization software query data from the database, thereby being classified as stand-alone software that has not been regulated. There is the potential for causing harm when using CDSS,²¹ and issues regarding liability and negligence with CDSS use are not currently clear.²⁰ Two independent organizations have been focusing on these issues,^22,23 and in 2011 the FDA announced Class I requirements (ie, requiring general controls) for 'Medical Device Data Systems' and Class I, II, or III requirements for 'Mobile Medical Applications' depending on what the application does. An application that allows a user to input patient-specific information and - using formulae or processing algorithms - output a patient-specific result, diagnosis, or treatment recommendation to be used in clinical practice constitutes a regulated software application. Discuss these new regulations with vendors and if you plan to create applications to help analyze patient data call the FDA and find out if there are regulatory requirements that must be met.

High-resolution physiologic data from the patient monitor and tertiary patient monitoring devices are the most fundamental data that are required to support neurophysiologic decision support systems. Real-time information from infusion pumps about treatments and other intervention information is the second most critical patient-specific data that you need. These two types of data together will enable you to evaluate the physiologic effect of treatment interventions and to make determinations about a patient's physiologic status. Your understanding of the patient can be enhanced if you can integrate these data with laboratory, clinical examination, nutrition, and any other information that is captured digitally.

I have already implied that EEG monitoring systems are FDA regulated, and indeed they are. Historically, EEG monitoring and neurophysiologic decision support systems have been separated with little integration. This is beginning to change. EEG systems are starting to integrate patient monitoring data into its system or making quantitative EEG parameters available for export into neurophysiologic decision support systems. EEG vendors will be able to tell you specifics for their systems, and options in this regard increase every year. Kull and Emerson²⁴ cover in detail considerations related to EEG monitoring in the ICU.

The most likely answer is no. The hospital plan will probably focus on complying with "meaningful use" standards including electronic prescribing, health information exchange among clinicians and hospitals, and automated reporting of quality performance.¹¹ The benefit to the hospital is that electronic data collection is in principle more efficient, helps reduce medical errors and resulting lawsuits, improves medical device management, reduces costs,^9,25,26 and will provide the maximum financial incentives for EHR adoption.¹² This will almost certainly cover laboratory, clinical examination, and other tertiary data normally documented on paper charts, including intervention information. All of these data can be helpful to incorporate into a neurophysiologic decision support system. However, high-resolution patient monitoring data—the essential ingredient to any type of neurophysiologic decision support system—will probably not be considered a priority. Traditionally, EHR systems are not designed to capture and store high-resolution physiologic data, although some EHR systems specifically designed for the critical care settings do attempt to address this (eg, MetaVision for ICUs,²⁷ iMDsoft, Needham, MA). Still it is unlikely that even critical care EHR systems will come preconfigured to collect data from neuromonitoring devices that do not plug into the monitor. A customized solution, or more likely, an entirely separate data acquisition plan will be needed for this purpose (and for EEG monitoring as well).

There is no right answer to this question and typically the "right" solution will be different depending on the particular situation of your institution. There are many choices to consider that fundamentally revolve around the three critical areas including:

1. Collection and storage of high-resolution data from the patient monitor and secondary monitoring devices that do not plug into the patient monitor.

2. Integration of patient monitoring data with other types of patient information, such as medication infusion information.

3. Analyzing, visualizing, and otherwise synthesizing patient data converting it into useful clinical information that supports diagnosis and management decisions.

Specifically, the first and biggest choice you need to consider is whether you plan to monitor a small number of patients using a portable monitoring solution that can be moved from room-to-room, or if you plan on creating an ICU-wide solution. Small-scale solutions are akin to purchasing a medical device that takes data directly from the patient monitor and peripheral devices in order to display the data on its screen. This choice offers a convenient all-in-one solution that includes equipment for data acquisition and software for analysis and visualization in one package. Cart-based portable solutions often can include EEG monitoring with trending of patient vital signs from the patient monitor. These systems tend to be easy to use by staff, easy to maintain, and purchasing one is likely the quickest path to implementing a live system in the ICU (Figure 17-1). Cart-based neuromonitoring solutions are usually FDA approved and, therefore, are priced similar to an expensive medical device—a cost benefit is not necessarily realized with this option. These CDSS devices also have some of the other drawbacks of any medical device, including difficulty integrating it with other hospital clinical information systems, options to use the collected data for CDSS outside the all-in-one system is extremely limited, and archiving the data for clinical research purposes is often cumbersome and may not lend itself to group analysis. Finally, analysis and visualization options are fixed unless your system exports real-time data to another source that allows you to manipulate and visualize it how you choose.

In contrast, ICU-wide solutions are more complicated to initiate and expensive to maintain, but usually offer greater short- and long-term flexibility. Data from all the patient monitors and peripheral devices in the ICU are collected simultaneously and stored on an enterprise level central server.^25,28 A server is simply a class of computers with large computational and storage capacities that manage, store, and retrieve data for other computers or devices.²⁹ This central server will need to be operated and maintained over time, most often by hospital information technology, which is an expense that must be budgeted for. You will need to assess the extent to which hospital administrators and information technology (IT) understand and are committed to neurophysiologic decision support in the ICU. I recommend this as an essential prerequisite before seriously considering ICU-wide solutions. Ideally, patient data are stored in a nonproprietary Open Database Connectivity (ODBC) database like the Microsoft SQL Server. In this format, patient data can be queried for use by any number of CDSS programs in the future. CDSS programs that query data from a database, as opposed to directly from the monitor, fall into the category of stand-alone software and are not generally regulated by the FDA at this time. Data in an ODBC database will also better facilitate clinical research. Unlike the all-in-one solution, you may need additional software systems to analyze and/or view the data (Figure 17-2).

Most bedside devices including the patient monitor are equipped with RS-232 (digital) and analog (waveform) data output ports for automatic data output capabilities. Some newer models to output data from devices include Universal Serial Bus (USB), 802.3 (Ethernet), IEEE 11073, or even wireless (eg, 802.11b/g, Bluetooth) data communication capabilities. One strategy is to collect data by connecting directly to each patient monitor and isolated unconnected medical devices. This is typically the strategy for all-in-one and some ICU-wide solutions. Alternatively, most patient monitors operate on a closed intranet and are connected to a central server. This facilitates providing data from the patient monitor to a central nursing station or nursing pod area, or perhaps transferring hourly data to an EHR system for nurse verification. Some vendors that provide an ICU-wide solution to patient data collection do so by pulling the data off the patient monitor central server. However, data from devices not connected to the patient monitor must still be collected from each individual device. This can be accomplished by plugging individual devices into a communication hub that transmits the data from the device to the data collection server, where it is integrated with data pulled from the monitor.

Medical devices usually output data as a comma-delimited text string (eg, 16, 24.5, HIGH, 53.76, 2, 11, and so on) every few seconds. In order to collect these data, this text string must be converted into individual data points that are linked to its proper identification label. This translation is referred to as a device interface, or a small bit of software that automatically converts data outputted from the device into understandable data in a usable format. Each device has its own specifications for data communication, which usually is available in the device's technical manual or can be obtained from the device manufacturer. Many commercial entities have already written device interfaces for cardiopulmonary and neuromonitoring devices. Make a list of devices that you want and make sure your vendor of choice supports them. Also, consider purchasing data interfaces for future, but as of yet unknown, devices.

Yes, and I would strongly encourage it, especially if you plan to implement an ICU-wide monitoring solution. First, you will need capital funds to purchase equipment and long-term operating funds to support your monitoring program over time. Connecting neurophysiologic decision support to the budget for EHR implementation could be helpful. Even if these budgets remain separate, the best time to get a neurophysiologic decision system implemented is when the hospital already has an IT project happening in the ICU. Simply having hospital administration and IT attention on the ICU can help you get your system in place. The counterargument is that you could lose some degree of control over the system. It will be critical that you have buy-in from administrators on the goals of neurophysiologic decision support, or else you may end up with a system and still will not be able to do what you intended and need.

As was already stated, the EHR will likely be the data repository for important data like laboratory and infusion data. To fully realize all the benefits of neurophysiologic decision support you will need real-time data from the EHR to merge with patient monitoring data. Enterprise-level EHR systems have the capacity to supply this information, but you will need hospital and IT permission and help to get these data. Working with them from the beginning, again while there is a hospital project budget attached to EHR implementation in the ICU, is the best chance you will have to get this information.

Many decisions are made during EHR implementation that you may want influence over. For instance, common practice is for nurses to document in the EHR when an infusion was started. The precise time that an infusion was started is not as critical when evaluating hourly data but is vital when trying to understand the impact an infusion has on high-resolution monitoring data. Numerous studies have shown that automatic documentation of data directly to an EHR from infusion pumps is more accurate and saves nursing time.^30-32 Unless there is a voiced clinical need for precise timing of infusion pump information, this information will not be easily available for neurophysiologic decision support tools. Other types of data like ventilator data or other data that are difficult to collect might be left out. Ideally, a neurophysiologic decision support system can automatically send data to the EHR as well. I highly recommend being involved in these discussions.

This in part depends on how frequently you collect and store data, for which, unfortunately, there is no clear guideline. For some applications, data collected every 10 minutes can be suf-ficient, whereas for cerebral autoregulation indices,³³ data collection every 5 seconds is needed. In our neurologic ICU, we store approximately 200 megabytes (MB)/day/bed, collecting 5-second resolution digital data into an SQL database and 2-second resolution digital data plus 240-Hz waveform data into individual binary files. Continuous EEG recordings can generate approximately 1 gigabyte (GB)/day for EEG alone, and 20 GB/day when including video. If these data travel over the hospital network, IT administrators will want to ensure that Ethernet networks employ 1 GB per second or greater connections between switches and routers to avoid network performance slowdowns or data loss.³⁴ In some institutions, as a matter of policy, high-resolution patient data are erased several months after discharge unless there is a specific request to save it. In our ICU, all patient data are permanently stored in a data warehouse for quality control and clinical research purposes.

A data warehouse is a collection of decision support technologies to enable better and faster decisions,³⁵ and is composed of multiple servers and networked data storage to support clinical decision making and research. A data warehouse, then, is where laboratory, imaging, intervention, physiologic, and all other patient databases reside. Traditional data warehouses are set up to bring several different kinds of data (eg, laboratory and physiologic) together into a unified database to be utilized by clinical support software tools. According to the National Institutes of Health (NIH) road map, the goal of translational research is to translate scientific discovery that typically begins at "the bench" into practical applications that progress to the clinical level, or the patient's "bedside."³⁶ A data warehouse is a critical component for clinical informatics translational research, whereby new uses of patient information can be researched and then brought to the bedside for clinical use through the implementation of a CDSS.

Clinical research has yielded an understanding of many physiologic patterns that should be acted on when observed at the patient bedside. Continued clinical research is vital to advancing the field of neurophysiologic decision support systems. Dr. Urban Ungerstedt from the Karolinska Institute, Stockholm, Sweden, is most widely known for his work in cerebral microdialysis (eg, see references 37 and 38). Dr. Ungerstedt has recently initiated a collaborative project with support of the Karolinska Institute and the Neurocritical Care Society to provide free software support for research multimodal data acquisition.³⁹ This organization is also committed to connecting researchers together for collaboration on multimodal monitoring projects. This is a good place to start.

Data about the clinical course of patients will also need to be collected. The desire to document and time stamp the entire clinical course of a patient must be balanced with limited resources to do so. Try to capture too many variables and the quality and accuracy of data can go down, whereas not capturing enough information results in not being able to meaningfully address research questions. How to Research ⁴⁰ is a good reference on conducting research that covers the entire research process from formulating a question to writing a manuscript for publication. I strongly encourage you to take advantage of collaboration opportunities through www.multimodalmonitoring.org and connect with investigators already conducting research in your area of interest.

The field is expanding and more systems and configurations emerge each year. The systems listed below are ones that I am familiar with, but undoubtedly there are other viable options in the marketplace. Features are added to these systems frequently. Contact system vendors for current lists of capabilities, and absolutely search on your own because new systems seem to appear annually. Online search term suggestions include ICU information systems, ICU data acquisition systems, NeuroICU monitoring, critical care information system, and critical care information management.

Be very clear about what you want to be able to do with the data, both now and for the future, and what data you need to have. You can always delete or thin data; be biased toward collecting too much data too frequently. If you purchase a solution that stores data in an ODBC database like the Microsoft SQL server, know that you can use standard data mining packages like IBM SPSS Modeler Professional⁴¹ or Statistica Data Miner,⁴² and/or other standard data visualization packages like MATLAB⁴³ and R.⁴⁴ Working with hospital IT to help you evaluate all your options is highly recommended.

Cart-based all-in-one systems:

• Component Neuromonitoring System⁴⁵ (CNS Technology, Ambler, PA)

• Eclipse Neurologic Workstation⁴⁶ (Axon Systems, Hauppauge, NY)

• Neuro Workbench⁴⁷ (Nihon Kohden, Foothill Ranch, CA)

ICU-wide enterprise systems:

• Bedmaster EX⁴⁸ (Excel Medical, Jupiter, FL)

• Patient Monitoring Solutions⁴⁹ (General Electric, Waukesha, WI)

• Healthcare Informatics and Patient Monitoring⁵⁰ (Philips Healthcare, Andover, MA)

Hybrid systems (can be configured for portable use or for the entire ICU):

• ICM+⁵¹ (University of Cambridge, Cambridge, UK)

• ICU Pilot⁵² (CMA Microdialysis, Holliston, MA)

• Customized solutions⁴⁵ (CNS Technology, Ambler, PA)

Dashboard systems:

• Clinical Command and Control⁵³ (Global Care Quest, El Segundo, CA)

• CareAware Critical Care⁵⁴ (Cerner Corporation, Kansas City, MO)