Design and implementation of GRIP: a computerized glucose control system at a surgical intensive care unit

Abstract

Background: Tight glucose control by intensive insulin therapy has become a key part of critical care and is an important field of study in acute coronary care. A balance has to be found between frequency of measurements and the risk of hypoglycemia. Current nurse-driven protocols are paper-based and, therefore, rely on simple rules. For safety and efficiency a computer decision support system that employs complex logic may be superior to paper protocols.

Methods: We designed and implemented GRIP, a stand-alone Java computer program. Our implementation of GRIP will be released as free software. Blood glucose values measured by a point-of-care analyzer were automatically retrieved from the central laboratory database. Additional clinical information was asked from the nurse and the program subsequently advised a new insulin pump rate and glucose sampling interval.

Results: Implementation of the computer program was uneventful and successful. GRIP treated 179 patients for a total of 957 patient-days. Severe hypoglycemia (< 2.2 mmol/L) only occurred once due to human error. With a median (IQR) of 4.9 (4.2-6.2) glucose measurements per day the median percentage of time in which glucose fell in the target range was 78%. Nurses rated the program as easy to work with and as an improvement over the preceding paper protocol. They reported no increase in time spent on glucose control.

Conclusion: A computer driven protocol is a safe and effective means of glucose control at a surgical ICU. Future improvements in the recommendation algorithm may further improve safety and efficiency.

Figures

Figure 1

Glucose control cycle. Nurse-driven glucose control consists of the repetitive execution of the cycle depicted in this figure. First, the nurse acquires blood from the patients and gets the glucose level. A protocol or doctor then decides what action should be taken (in most cases a change of the rate of the insulin pump), and subsequently this action is performed at the bedside by the nurse.

Figure 2

High-level overview of GRIP. Grip contains four major components: a component to interface with the hospital information system, a component to interface with the nurse, a component that calculates the advice that GRIP gives, and a component that logs errors and stores all data GRIP generates. HL7 : Health Level 7. SQL : Standard Query Language.

Figure 3

Main screen of GRIP. The main screen of GRIP. An overview of the ICU is shown replicating the arrangements of beds on the floor. Beds have colors according to pending action: green – no action has to be taken, orange – action has to be taken and a small icon indicates what action, in this case a new glucose value was detected in the hospital data system, which needs a validation from the nurse, and red – urgent action required, for example occurrence of hypoglycemia, or an advised measurement that is more than 30 minutes late. Each bed shows the current insulin pump rate, the most recent glucose value and the time it was taken, and the time the next glucose value needs to be taken. Empty beds are shown in grey. Each bed is clickable to yield a more detailed information panel shown in figure 4.

Figure 4

Patient overview in GRIP. The patient overview panel shows more detailed information for a single patient. On the left general characteristics such as patient ID number, birth-date and sex are shown. The middle three panels show the current status, the recommendation of GRIP, and the tasks GRIP thinks the user should perform.

Figure 5

Glucose control. Median and interquartile range of glucose levels are shown for the 109 patients with a length of stay longer than 1 day. The dashed line equals the glucose target of GRIP.

Figure 6

Results of the nurse questionnaire. Results of the nurse questionnaires 1 month before (N = 32) and 6 months after (N = 22) implementing GRIP. Questions pertaining to GRIP were only asked in the questionnaire held after 6 months. The median and interquartile range of responses are shown for each question asked.