Acquisition of ICU data: concepts and demands

Applying Machine Learning to Pediatric Critical Care Data

OBJECTIVES

To explore whether machine learning applied to pediatric critical care data could discover medically pertinent information, we analyzed clinically collected electronic medical record data, after data extraction and preparation, using k-means clustering.

DESIGN

Retrospective analysis of electronic medical record ICU data.

SETTING

Tertiary Children's Hospital PICU.

PATIENTS

Anonymized electronic medical record data from PICU admissions over 10 years.

INTERVENTIONS

None.

MEASUREMENTS AND MAIN RESULTS

Data from 11,384 PICU episodes were cleaned, and specific features were generated. A k-means clustering algorithm was applied, and the stability and medical validity of the resulting 10 clusters were determined. The distribution of mortality, length of stay, use of ventilation and pressors, and diagnostic categories among resulting clusters was analyzed. Clusters had significant prognostic information (p < 0.0001). Cluster membership predicted mortality (area under the curve of the receiver operating characteristic = 0.77). Length of stay, the use of inotropes and intubation, and diagnostic categories were nonrandomly distributed among the clusters (p < 0.0001).

CONCLUSIONS

A standard machine learning methodology was able to determine significant medically relevant information from PICU electronic medical record data which included prognosis, diagnosis, and therapy in an unsupervised approach. Further development and application of machine learning to critical care data may provide insights into how critical illness happens to children.

Design and implementation of a portable physiologic data acquisition system

OBJECTIVE

To describe and report the reliability of a portable, laptop-based, real-time, continuous physiologic data acquisition system (PDAS) that allows for synchronous recording of physiologic data, clinical events, and event markers at the bedside for physiologic research studies in the intensive care unit.

DESIGN

Descriptive report of new research technology.

SETTING

Adult and pediatric intensive care units in three tertiary care academic hospitals.

PATIENTS

Sixty-four critically ill and injured patients were studied, including 34 adult (22 males and 12 females) and 30 pediatric (19 males and 11 females).

INTERVENTIONS

None.

MEASUREMENTS AND MAIN RESULTS

Data transmission errors during bench and field testing were measured. The PDAS was used in three separate research studies, by multiple users, and for repeated recordings of the same set of signals at various intervals for different lengths of time. Both parametric (1 Hz) and waveform (125-500 Hz) signals were recorded and analyzed. Details of the PDAS components are explained and examples are given from the three experimental physiology-based protocols. Waveform data include electrocardiogram, respiration, systemic arterial pressure (invasive and noninvasive), oxygen saturation, central venous pressure, pulmonary arterial pressure, left and right atrial pressures, intracranial pressure, and regional cerebral blood flow. Bench and field testing of the PDAS demonstrated excellent reliability with 100% accuracy and no data transmission errors. The key feature of simultaneously capturing physiologic signal data and clinical events (e.g., changes in mechanical ventilation, drug administration, clinical condition) is emphasized.

CONCLUSIONS

The PDAS provides a reliable tool to record physiologic signals and associated clinical events on a second-to-second basis and may serve as an important adjunctive research tool in designing and performing clinical physiologic studies in critical illness and injury.

Improving the ICU: part 2

ICUs are a vital but troubled component of modern health-care systems. Improving ICU performance requires that we shift from a paradigm that concentrates on individual performance, to a systems-oriented approach that emphasizes the need to assess and improve the ICU systems and processes that hinder the ability of individuals to perform their jobs well. This second part of a two-part treatise establishes a practical framework for performance improvement and examines specific strategies to improve ICU performance, including the use of information systems.

An evaluation of the use of hand-held computers for bedside nursing care

Both keyboard-based and pen-based devices for data input have advantages and disadvantages. The suitability of the two input devices for entering different types of clinical data into computers remains unclear. This study aimed to determine the usefulness of different types of technology for nurses' data entry by comparing the utility and efficiency of keyboard and pen-based data input for clinical tasks. The study was conducted in the six largest specialties of an acute care hospital. In each specialty, several wards were randomly selected, and all nurses working in these wards were invited to participate. The input prototype was designed according to the type of text that was to be entered into the system. Task 1 mainly consisted of structured data, Task 2 contained equal amount of structured and textual data, and Task 3 was mainly in textual form. Each nurse was asked to complete the three simulations of nursing records (Tasks 1, 2 and 3). Preliminary findings showed that nurses found the pen-based interface easier to use than the keyboard for completing Tasks 1 and 2, but not Task 3. In terms of accuracy, the nurses preferred the keyboard to the pen when the data were more structured. The pen-based device is not a panacea for all kinds of user interface, and more importantly, the choice of input device should depend on the amount of structured and textual data.

Electronic medical record in the intensive care unit

The EMR in the ICU has the utility of providing the necessary information to make sound clinical decisions for critically ill patients. For it to be optimized, the EMR must be more than just what is being replicated in the written record or merely a documentation tool; it must add value that supports and enhances clinical decision support. The EMR is too expensive a tool just to be a computer designed to ease documentation and retrieve data faster. Gardner and Huff have suggested that the EMR must answer three questions: Why, What, and So What. The "Why" is relatively easy to answer, but the "What" data to use so that the information is meaningful to a provider and the "So What" are more difficult to answer. Provided one can qualitatively assess "What" information is important for a health care provider, then "So What" becomes an important objective in the empirical quantification of the benefits that the EMR provides. It is clear that to analyze some of the outcomes that health care delivery provides, one needs some mechanism to automate the information at the point of care, particularly now that the regulatory agencies are requiring it. Given the fact that there is no single integrated computerized patient record, this becomes the daunting task for the next century. Making it easier for health care providers to interact with the system and providing them with instantaneous feedback that changes their medical decision so they can deliver better care (clinical pathways, clinical practice guidelines) will be the task required of the next generation of CISs.

Benefits of an electronic clinical information system: an intensive care nursing perspective

Due to advancements in surgery, medicine and equipment, the modern intensive care unit (ICU) patient necessitates the recording of vast amounts of data. The management of these data is increasingly impinging on nursing time. A review of the literature describes the potential benefits of using a clinical information system (CIS) to record and save data electronically rather than transcribing it onto conventional charts. Commercial systems that can perform these tasks are available but have severe financial implications in terms of initial costs, maintenance and upgrading. Work is continuing at Killingbeck Hospital on a project developing and introducing a CIS into intensive care using standard hardware, software and programming tools thus minimizing these costs.

Design of a summary screen for an ICU patient data management system

The paper describes a used-centred design for the summary screen of a computerised ICU patient data management system (PDMS). The screen also forms the resting state display, or default screen, and provides the principal navigation tool to other functionality within the system. The design process identified the most frequent potential users of this screen to be the nurses. Their tasks and the information resources required to perform them were analysed. The analysis identified that the nurses' main task of planning and implementing patient care required an awareness of a set of physiological parameters which provided an overview of the patient's general condition. Novel formats are proposed for displaying the trends in physiological parameters and these have been incorporated into a proposed screen design. These display formats have been evaluated by ICU nurses; they were adjudged to be clear, relevant, easy to learn and simple to use. Nurses considered the content of the screen, and the display formats used, to be suitable for maintaining an awareness of a patient's state during routine patient management.

Aberdeen polygons: computer displays of physiological profiles for intensive care

The clinician in an intensive therapy unit is presented regularly with a range of information about the current physiological state of the patients under care. This information typically comes from a variety of sources and in a variety of formats. A more integrated form of display incorporating several physiological parameters may be helpful therefore. Three experiments are reported that explored the potential use of analogue, polygon diagrams to display physiological data from patients undergoing intensive therapy. Experiment 1 demonstrated that information can be extracted readily from such diagrams comprising 8- or 10-sided polygons, but with an advantage for simpler polygons and for information displayed at the top of the diagram. Experiment 2 showed that colour coding removed these biases for simpler polygons and the top of the diagram, together with speeding the processing time. Experiment 3 used polygons displaying patterns of physiological data that were consistent with typical conditions observed in the intensive care unit. It was found that physicians can readily learn to recognize these patterns and to diagnose both the nature and severity of the patient's physiological state. These polygon diagrams appear to have some considerable potential for use in providing on-line summary information of a patient's physiological state.

Patient data management systems in intensive care--the situation in Europe

OBJECTIVE

Computerized Patient Data Management Systems (PDMS) have been developed for handling the enormous increase in data collection in ICUs. This study tries to evaluate the functionality of such systems installed in Europe.

DESIGN

Criteria reflecting usefulness and practicality formed the basis of a questionnaire to be answered accurately by the vendors. We then examined functions provided and their implementation in European ICUs. Next, an "Information Delivery Test" evaluated variations in performance, taking questions arising from daily routine work and measured time of information delivery.

SETTING

ICUs located in Vienna (Austria), Antwerp (Belgium), Dortmund (Germany), Kuopio (Finland).

PARTICIPANTS

5 PDMS were selected on the basis of our inclusion criteria: commercial availability with at least one installation in Europe, bedside-based design, realization of international standards and a prescribed minimum of functionality.

RESULTS

The "Table of Functions" shows an overview of functions and their implementation. "System analyses" indicates predominant differences in properties and functions found between the systems. Results of the "Information Delivery Tests" are shown in the graphic charts.

CONCLUSIONS

Systems with graphical data presentation have advantages over systems presenting data mainly in numerical format. Time has come to form a medical establishment powerful enough to set standards and thus communicate with industrial partners as well as with hospital management responsible for planning, purchasing and implementing PDMS. Overall, communication between clinicians, nurses, computer scientists and PDMS vendors must be enhanced to achieve the common goal: useful and practical data management systems at ICUs.

Data quality of bedside monitoring in an intensive care unit

Computerized record keeping promises complete, accurate and legible documentation. Reliable measurements are a prerequisite to fulfill these expectations. We analyzed the physiological variables provided by bedside monitoring devices in 657 bedside visits performed by an experienced Intensive Care nurse during 75 Intensive Care rounds. We registered which variables were displayed. If a variable was displayed, we assessed whether it could be used for documentation or should be rejected. If a value was rejected the reason was registered as: the measurement was not intended (superfluous display), the current clinical situation did not allow proper measurement, or other reasons. Basic variables (vital signs and respiration related variables) were displayed in more then 90%, specific variables (e.g. intracranial pressure) were displayed in less than 50% of the situations. Displayed variables were superfluous on an average of 11% because measurement was not intended. Variables like heart rate, temperature, airway pressure, minute volume of ventilation, arrhythmia, pulmonary arterial pressure, non-invasive blood pressure, and intracranial pressure provide high quality measured values (acceptance of more than 90%). Invasive arterial pressure, central venous pressure, respiration rate and oxygen saturation (via pulse oximetry) provided lower quality values with a rejection rate higher than 10%. Inappropriate sensor technology to match the clinical environment seems to be the root cause. In future the request for automatic documentation will increase. In order to avoid additional staff workload and to ensure reliable documentation, sensor technology especially related to respiration rate, blood pressure measurements, and pulse oximetry should be improved.

Lessons learned while building an integrated ICU workstation

The project LUCY (Linked Ulm Care sYstem) is described. The goal of this project was to build a research workstation in an Intensive Care Unit which enables evaluation of data/information processing and presentation concepts. Also evaluation of new devices and functions considering not only one device but the workplace as an entirety was an aim of the project. We describe the complete process of building from the stage of design until its testing in clinical routine. LUCY includes a patient monitor, a ventilator, 4 infusion pumps and 8 syringe pumps. All devices are connected to a preprocessing computer via serial interfaces. A high performance graphic workstation is used for central display of physiological and therapeutic variables. A versatile user interface provides touch screen, keyboard and mouse interaction. For fluid administration a bar code based control and documentation facility was included. While our scheduled development efforts were below 4 man-years, the overall man-power needed until the first routine test amounts to 8 man-years. Costs of devices and software sum up to 160,000 US$. First experiences in clinical routine show good general acceptance of the workplace concept. Analysing the recorded data we found 90% of the items to be redundant: individual filtering algorithms are necessary for each of nowaday's devices. The flexibility of the system concerning the implementation of new features is far from our expectations. Technical maintenance of the system during clinical operation requires continuous effort which we cannot afford in the current situation.

Information transfer in high dependency environments: an ergonomic analysis

We have studied the information flow in HDE (with special focus on the information transfer process) using data provided by a group of experienced health care professionals. A model of the information flow in HDE was built up. It postulates the existence of quanta of information (due to the artificial fragmentation of the information flow produced by the clinical working processes: organization in shifts, demand of simultaneous activities from different staff members, etc.). This fragmentation is described by using the so-called Clinical Information Process Units (CIPUs), which correspond to patient care activities going on in parallely and serially linked blocks, performed by the staff in the specific environments. Due to a transfer in responsibility over the patient the CIPUs are linked by information transfer events which are described using transfer modules (TraMs). We exemplified 32 CIPUs related to the clinical environments (PreOp, Surgery, Recovery Intensive Care, Ward, Diagnostics, Outpatient) and the health care professional groups (Anesthesiologist/Intensivist, Surgeon, Nurse, Physician, Diagnostic Physician, Physical Therapist). A matrix was established providing the transfer situations among the CIPUs enabling a systematic classification of the TraMs. The contents of the TraMs are built up of information link elements, which are assembled according to the specific settings of the transfer situation given by the emitter, receiver and purpose. In summary we modelled the process of information transfer in HDE through CIPUs, TraMs and information links in a way, which may be useful to design information technology applications or to reorganize the information management in HDE.