Per-patient nutrition delivery
Figure
2 shows a large variation in nutrition rates achieved per day, per patient, narrowing and rising as the patient-specific metabolic state stabilises [
21], similar to that seen in Heyland et al. [
67]. However, the median variation per patient was only 12.9 [4.6–20.4] % (Table
4), suggesting patients do not deviate significantly from their mean nutrition rate. This result and the large IQR of the mean feed rates achieved (74.3–98.2%, Table
4) suggest the lower nutritional delivery to the 5th and 25th percentile are a result of a few patients who had a lower ability to tolerate glucose intake.
It is very uncommon for patients not on GC in Christchurch ICU to have their feed rates changed due to the strong clinical culture of patients achieving their caloric goal. In addition, prior to STAR’s predecessor, SPRINT, being implemented (2005) [
14], feed rates were fixed at 100% caloric goal during GC for all patients. Hence, if they are not on GC, they are likely to have a fixed 100% caloric goal nutrition rate and have a BG within 4.4–8.0 mmol/L, having a relatively constant glucose tolerance and reduced insulin sensitivity variability [
21,
68]. Therefore, if all patient data were considered, the intra-patient variability seen in Table
4 would likely go down.
Considering STAR feeds the maximum possible nutrition, while safely maintaining normo-glycaemia, the nutrition rates achieved give a good indication of the patient-specific ability to tolerate glucose and thus of their ‘STAR ideal’ nutrition rate. In essence, every patient is fed the maximum they can achieve with added insulin, within the bounds of the future predicted variability. Therefore, the spread of nutrition rates per patient in the results infer this ‘STAR ideal’ nutrition rate is very patient specific and evolves with time.
The ‘STAR ideal’ nutrition rate achieved by STAR was less than the 100% caloric goal for more than 50% of patients, over all days. However, the best unit surveyed in Cahill et al. [
59] was still considerably lower than this predetermined caloric goal suggesting these generalised approximations do not represent all ICU patients well, as seen in the results for STAR in Christchurch. In addition, over 56% of patients exceeded the lower 85% ‘Heyland ideal’ of [
51] by day 3, as shown in Table
3.
Limitations
Cahill et al. [
59] provides the percentage caloric goal nutrition achieved by each ICU. However, caloric goals may vary across ICUs. Additionally, the estimation of patient body weight in Christchurch ICU [
69], as shown in Table
2, may also bias the caloric goal feed estimate which outlines the need for patients to be weighed on the day of ICU admission in Christchurch Hospital. As a result, some ICUs may thus achieve caloric goal nutrition targets ‘more easily’ than others, making comparison difficult. However, the 25 kcal/kg/day ACCP guideline [
62] used in the Christchurch ICU, or a similar value guideline (25–30 kcal/kg/day SCCM/ASPEN [
70], and 20–25 kcal/kgBW/day initial phase and recovery phase 25–30 kcal/kgBW/day ESPEN [
71]), is commonly used and these cover the range used with STAR patients.
In addition, Cahill et al. review nutrition achieved during the first day of ICU stay, which is not necessarily when GC starts for all patients. Although GC commonly starts at the beginning of ICU stay, it may not always be the case. However, as an ICU patient is under the most amount of stress immediately post-surgery or insult [
72], they are most likely to require GC at or near the beginning of their ICU stay [
1‐
3]. In this study, 59.3% of patients started GC within 24 h of being admitted to the ICU (Median 15.5 h, Table
1).
Moreover, Cahill et al. survey the nutrition given to all ICU patients. However, this study only considered patients who required GC. The 25–35% of patients who require GC in the ICU [
28] are the most metabolically stressed and, as a result, have a reduced glucose uptake capacity. They are thus often harder to deliver the target nutrition rates [
20,
52,
57,
58]. In addition, given that 158 ICUs over 20 countries were surveyed by Cahill et al. [
59], and this ICU was 1 of the 22 surveyed in Australia, and New Zealand the mixed medical surgical ICU in Christchurch Hospital would likely have patients with similar parameters. Therefore, achieving nutrition rates with high performance GC similar to that achieved for all ICU patients, normo-glycaemic and hyperglycaemic, in the best ICU reviewed by Cahill et al. [
59] is a significant outcome. More importantly, this outcome and the inter-patient variability in the results indicate high nutrition delivery and safe, effective GC are not exclusive, an equally, that nutrition restriction to obtain GC does not necessarily reduce total nutrition in an international context.
The insulin–glucose model used by STAR has been shown to be effective in predicting a patient’s response [
35,
36,
73]. However, as STAR doses based on the 5th and 95th percentile future metabolic variability [
38], ensuring only a 5% risk of hypoglycaemia [
74], the majority of patient’s future BG will fall within the targeted band. Hence, many patients could possibly remain within the targeted BG range (4.4–8.0 mmol/L) if given a higher than recommended nutrition rate. As a result, some patients may be able to receive higher nutrition rates than reported here and still be able to be provided effective GC. However, this choice would also increase the likelihood of hyperglycaemia, reducing the safety of GC provided by STAR.
The 85% caloric goal presented in Heyland et al. is calculated by a model fit to retrospective data from 158 ICUs and clinical practices, and while it represents a significant body of multi-centre data, it may not be causative. Many prospective trials have found improved outcomes for even lower hypo-caloric feeding [
50,
75‐
77]. Therefore, this ‘Heyland ideal’ value may overestimate the caloric goal required for improved outcomes and may be reflective of ‘less sick’ patients tolerating higher nutrition. This study is designed to show that STAR can provide high nutrition rates while still providing safe and effective GC. In addition, STAR is designed to be flexible to different nutrition goals while still providing effective GC.
Other factors, such as mechanical ventilation, neurologic injury, gastric emptying and paresis patients, are well known to influence the nutritional requirements of ICU patients. This is another strong limitation of this retrospective analysis, as this detailed information was not available. However, the cohort was typical of medical ICU in Christchurch.
The STAR GC protocol uses model-based patient-specific control in conjunction with a stochastic model to predict the best treatment for a patient. As shown in Table
1 and [
32], STAR is able to achieve very good GC with a compliance of over 96.8% in all interventions and near identical results across multiple ICUs [
32]. However, in many clinical practices, the idea of protocol-driven changes in the nutrition given to a patient for GC is foreign and thus clinically unacceptable. Thus, the main focus of this study is to show that protocol-driven changes in nutrition rate do not preclude in achieving better nutrition delivery rates than those of 158 ICUs from 20 countries. In addition, the concept of nutritional tolerances in relation to glucose tolerances provides a potentially new method of calculating patient-specific feed rates and should be investigated further in future studies.