In this retrospective observational study, which is to our knowledge the largest cohort studied using indirect calorimetry as opposed to predictive equations to determine energy requirements in mechanically ventilated, critically ill patients, we have demonstrated a non-linear, significant association between the percent AdCal/REE and mortality by 60 days. The results suggest that increasing the AdCal/REE to 70 % was associated with decreased mortality, while increases above that point, particularly as the curve increased >100 %, were associated with increasing mortality. In addition, protein ingestion was independently and significantly associated with decreased mortality. However, achieving calorie targets was associated with a longer ICU stay and length of ventilation.
Uncertainty about the optimal goals for nutritional support continues, fueled by conflicting results in the recent literature. Thus, an apparent lack of benefit with higher caloric intake was demonstrated by Arabi et al.  who compared permissive underfeeding to standard feeding (835 ± 297 vs 1299 ± 467 kcal/day, respectively) in patients who received the same amount of protein, and by Rice et al. in two studies comparing trophic to standard therapy (300 ± 149 vs 1418 ± 686 and 425 ± 141 vs 1385 ± 46 kcal/day, respectively) [5, 19]. On the other hand, a positive association between achieving caloric goals and outcome was demonstrated in observational studies [2, 3] and in RCTs by our group in the TICACOS study , and others that used indirect calorimetry [20, 21]. As suggested by Heyland et al., this discrepancy may be explained at least in part by the statistical methods used in the studies showing an apparent lack of effect, in particular the method accounting for duration of exposure to nutrition or length of stay. In this regard we have shown in our study an association between achieving 70 % of the AdCal/REE and improved survival, which remained consistent even when only measurements after the second day and adding evaluable nutrition days as a predictor were considered.
In contrast to many previous studies that used an arbitrary, predefined cutoff of caloric intake to define optimal nutrition, mainly based on predictive equations, we examined the AdCal/REE as a continuous variable which allowed us to assess the relationship between administered calories and mortality at various levels of intake. The resultant U-shaped curve revealed a decrease in mortality as the caloric intake was increased to 70 % of target calories, but this was followed by an increase in mortality, particularly as the curve increased >100 %.
Our findings are in accordance with those of Heyland et al. who used similar methodology to show that providing more than two-thirds of prescribed calories was associated with reduced mortality and suggested that providing >85 % of the caloric goal was associated with the best outcome . In addition, they did not find an additional benefit of feeding >100 % of the target. However, the Heyland study used predictive equations to assess calorie goals, IC being used in only 0.5 % of their cases. By contrast, all our patients were assessed with IC, many of them having repeated measurements during the course of their stay so that their true and possibly changing metabolic needs may have been more accurately assessed. This difference in methods of assessing REE may have accounted for the difference in optimal targets noted in the two studies, namely 70 % vs. 85 %. In this regard, a recent position paper from the Multicentric Study Group for Indirect Calorimetry stated that IC is a tool of paramount importance necessary to optimize the nutrition therapy of patients with various pathological conditions, including critically ill patients . It is important to note that due to a number of factors, there is typically a negative gap between administered and targeted calories; thus, calorie targets should be 100 % of REE.
In our study, increasing the delivered caloric intake >70 % of REE was associated with both a longer ICU stay and days of ventilation. These results confirm those found in the TICACOS study . This might be explained by the fact that higher amounts of calories add to the work of breathing, thus, resulting in a longer time to wean from ventilation and thus, a longer ICU LOS. These complications were not evident in the Arabi and Rice studies in the groups receiving standard nutritional support [4, 5]. However, in both these studies, the standard therapy groups received lower calorie loads compared to our study, namely 1299 and 1300 kcal/day, respectively, compared to 1651 kcal/day. Complications were not mentioned in the Heyland study. This finding of more prolonged ventilation and ICU LOS could account for the U-shaped curve we noted where an optimal amount of calories may exist, while deviations from it, either above or below, may be associated with harm. The fact that caloric intake is not a binary variable, that the caloric goal is as yet unknown and that surrogate measures for IC that are often used are inaccurate, complicate the quest to find this optimal goal and might explain the contradicting results of past studies. If the effect of caloric intake on mortality is indeed U-shaped, it supports the more widespread use of accurately measured metabolic requirements, either by IC or by improved predictive equations.
Our study also supports the increasingly appreciated importance of protein in improving survival, as protein intake was linearly associated with decreased mortality in the multivariable model (HR 0.99, 95 % CI 0.98–0.99, p = 0.018) suggesting a 1 % reduction in mortality for every gram of daily protein ingested. These results echo other observational studies [22, 23] highlighting the importance of protein intake. In addition, it is worth noting that even with the use of supplemental parenteral nutrition, the realities of ICU make it difficult to achieve 100 % of caloric and protein targets.
Finally, it is interesting to note the reduced REE in patients who died, which has been suggested to be the result of multi-organ dysfunction in sepsis leading to metabolic shutdown . This issue requires further elucidation. Again, this highlights the importance of IC- based REE measurements as metabolic needs may shift through the course of a critical illness.
Our study has several limitations. By the very nature of the observational design, inclusion of indirect calorimetry patients and the non-randomized administration of calories might introduce selection bias. As in these designs, especially in nutrition assessment trials, results need to be interpreted with caution as there is a risk that non-random allocation itself might influence the results, i.e. well-fed patients have a better prognosis irrespective of caloric requirements or that higher levels of caloric intake pre-select longer-surviving patients. In addition, while we have tried to account for confounders, we cannot rule out additional, as yet unknown ones.
Ours is a single-center study describing a unique critically-ill population, and the center-specific practices, which might limit the external validity. In addition the differences might not manifest in a short duration of stay. We have attempted to address these limitations in the design of our study as described in “Methods”. Thus, regarding selection and time bias, we restricted the sample size, excluded short-stay patients and used several sensitivity analyses that showed the stability of the results. The fact that the REE and the AdCal/REE percent were relatively stable in time >48 hours of hospitalization further supports a lack of time bias. In addition, our results show that “slightly underfed” patients, i.e. those receiving 70 % of target calories, fared better than optimally fed patients (those receiving 100 % of target calories), which argues against the notion that well-fed patients might have a better prognosis irrespective of nutrition demands. In addition we have tried to account for confounders, including disease severity.