Responses to the introduction of SGC
Both education and the possibility to discuss experiences have been pointed out as success-criteria when implementing new protocols . When teaching the ICU staff about the principles and practice of SGC, the main focus was on practicalities for the nurses and the benefits for the patients. During these sessions discussion was encouraged. We received few objections to the increased workload. The 1:1 nurse: patient ratio in our ICU may have contributed to the positive attitude. The staff already had experience in managing insulin infusions, albeit less strictly, when the BG target range was 4 to 8 mmol/L. The main reactions from both the nursing staff and physicians were interest, support and satisfaction. Such positive responses are described in the literature to be related to understanding the protocols rationale and benefits [13, 20]. With very few exceptions doctors prescribed BG targets according to the new algorithm. From time to time it was necessary to curb enthusiasm from wards that wanted to participate in the study, and physicians who wanted to use the algorithm outside the ICU.
Contrary to our expectations, posters with performance statistics generated little interest. On the other hand, e-mails with the same information caused discussions and proved valuable. It seems that e-mails allowed our co-workers to assess the feedback at their own pace and when it did not interfere with other obligations. We find this experience useful with respect to future quality enhancement projects.
During 2004–5, 77.5 % of blood-glucose samples were within the target-range specified at that time, i.e. 4–8 mmol/L. The number of cases of severe hypoglycaemias was low. Thus, nurses were able to reach set targets even with no algorithm to guide therapy. The statistics indicate that nurses conservatively preferred to keep the BG levels in the higher end of the target range. This finding was confirmed in discussions with the nurses. The nurses' rationale was to avoid SHG. Several experienced nurses expressed uncertainty with the use of an algorithm to control the BG level instead of the 'good old way' of intuition and experience. However, following introduction of new therapeutic targets, an algorithm to guide therapy and extensive teaching and feedback, glycaemic control improved significantly, albeit at the cost of increased expenditure and more frequent hypoglycaemia. Was the reduction in median BG from 6.6 mmol/L to 5.9 mmol/L important enough to justify these risks? It is evident from Figs 3 and 4 that a downward shift of all BG values was the result of our project. Thus, the frequency of samples with BG > 8 mmol/L was reduced by 32 %. Finney et al found a strong association between higher level of BG and risk of death and suggested an upper tolerance limit of 8 mmol/L . We are convinced that this can not be achieved unless one aims for a narrower range, i.e. normoglycaemia.
The studies by van den Berghe et al. [1, 6] indicated that primarily patients who remained in the ICU for several days profited from SGC. This has resulted in suggestions that SGC be withheld until patients have remained in the ICU for three days. In our study 210 and 206 patients had LOS > 3 days in the periods before and after implementation of SGC, i.e. 44 %. Does this mean that the remaining patients were exposed to an unnecessary risk? Our interpretation of the results of the randomised trials is that any benefit of SGC accumulates with increasing LOS. As it is extremely difficult to predict LOS for patients who are admitted to the ICU, we believe it to be a sensible strategy to initiate SGC as early as possible. Routines must be developed, however, to avoid hypoglycaemia (see below).
Glycaemic control improved for the majority of our patients. These results may be partially explained by the bedside follow-up and continuous attention to the rationale for SGC by the project-nurses. Also, the 1:1 nurse: patient ratio ensured that nurses attended to only one patient at a time. Furthermore, almost all our ICU patients were fed using enteral nutrition, possibly another contributing factor to achieving SGC . Our continued goal is that our nurses become comfortable with BG levels in the middle range of normal, thus avoiding hyperglycaemia that would otherwise occur because they are "playing it safe".
The number of patients who experienced severe hypoglycaemia increased from 12 (2.4 %) to 40 (8.9 %) following implementation of SGC. This compares well with published results [1, 6, 12] but is nevertheless a cause for concern. It is difficult to establish causality, if any, between episodes of hypoglycaemia and subsequent deaths and to avoid bias we have consulted with independent expert who is now evaluating the cause of death in patients who were hypoglycaemic. The death rate among hypoglycaemic patients did not exceed expected mortality rate as assessed by SAPS II score. Also, patients who died after hypoglycaemia were at higher baseline risk of death than survivors (Table 3, Additional file 1). In a recent review, Cryer refers to laboratory work in non-human primates that demonstrated a need for prolonged (i.e. several hours) of severe hypoglycaemia (< 1.0 mmol/L) to reliably produce brain damage . The situation may well be different in critically ill patients, in particular those whose metabolic demands are only marginally met due to their illness . On the other hand, hyperglycaemia has consistently been shown to be associated with severity of brain damage in head injured patients or those with anoxic coma [22, 23]. Kanji et al (2004) claims that to eliminate the incidence of hypoglycaemia completely is unrealistic since ICU patients have significant fluctuations in metabolic and endocrine demands .
Although mortality was not formally assessed, ICU- and 30-day mortality in 2005 and 2006 do not raise cause for alarm.
Need for algorithm revision?
The hypoglycaemias that resulted from the introduction of our algorithm led to a suggestion that unless BG was truly stable, no more than two hours should pass between blood-glucose measurements. If uncertain, or with changes in the patients condition, nurses were encouraged to measure the BG even more frequent. Also, nurses were encouraged to compare their own assessment, based on experience, with the algorithm. Thus, the status of the algorithm changed from "binding" to "guiding". Most hypoglycaemic episodes occurred early after admission to our ICU (Fig 5). This may indicate that insufficient attention was given to the issues of nutrition and glycaemic control in the early phase when patients were being stabilised in the ICU. Also, our algorithm may be criticised because it contains some subjective language and therefore requires interpretation by experienced nurses. This could potentially represent a problem for new nurses in training. However, new nurses in our ICU always work together with an experienced coach and it is therefore unlikely that hypoglycaemia occurred as a result of errors made by inexperienced nurses.
On the other hand our algorithm is not dose defining and allows for some variability in how aggressively nurses may treat blood-glucose aberrations. A number of different protocols used achieve SGC were reviewed by Meijering et al who concluded that dynamic scale protocols, similar to ours, yielded the best results . More recently published protocols are often dose defining and computer guided, and claim superiority over less rigorous protocols, such as ours, particularly with respect to avoiding hypoglycaemia. However, the numbers of patients included in many of these most recent studies are often small, typically ranging from twenty or less [14, 24–27] to 50–60 patients [28–30] (but see below). In our study only a small fraction of blood samples were hypoglycaemic (0.25 %), but this nevertheless affected 8.9 % of patients following introduction of SGC (Table 3). This may indicate that a fairly large number of patients should be included before superiority of any particular protocol is claimed.
A good example of a dose defining insulin infusion protocol was recently published by Braithwaite et al. who reported the results in 24 trauma patients . The protocol was arranged as a table with six columns containing detailed instructions on insulin infusion rates depending on BG. The patient was assigned a particular treatment column depending on his or her response to therapy. BG was controlled hourly until stable and thereafter every 2 hours. Target (blood-glucose < 6.1 mmol/L) was reached at a median time of 11 hours and BG thereafter fluctuated very little. The authors observed no cases of severe hypoglycaemia. If similar safety and efficacy can be demonstrated in a large patient sample this protocol is an attractive alternative to our algorithm, especially because it lends itself to computerization.
Thomas et al used a web-based insulin dose calculator to achieve tight glycaemic control in a 16-bed ICU . A total of 601 patients were subjected to this protocol, which was amended after 502 patients to allow for a higher insulin dose. Only 19 episodes of severe hypoglycamia were noted throughout this implementation (3.1 %), but the frequency seemed to increase after the amendment. Also, Davidson and coworkers reported experiences with a computer-directed intravenous insulin system that resulted in rapid glycaemic control and a low incidence (2.6 %) of hypoglycaemia in a large patient cohort . These authors contrast the use of a computer-directed system with that of van den Berghe , which depends on trained nurses. However, in a smaller study Shulman et al reported that computerised decision-supported intensive insulin therapy did not result in tight glycaemic control and that 10 % of patients became hypoglycaemic . To us it seems only reasonable that nurse-commitment and skill will contribute to success in obtaining glycaemic control in the ICU, whether they are assisted by computerised protocols or not.