Does artificial nutrition improve outcome of critical illness?

Nutritional support is generally considered an essential component in the management of critically ill patients. The existing guidelines advocate early enteral nutrition, with the optimal timing for the addition of parenteral nutrition to insufficient enteral feeding being the subject of transatlantic controversy. The unphysiologic intervention of artificial nutrition in critically ill patients, however, may evoke complications and side effects. Besides the classically described complications, suppression of autophagy, potentially important for cellular repair and organ recovery, was elucidated only recently. The question whether artificial nutrition in critical illness improves or worsens outcome as compared with starvation has so far not been adequately addressed. This paper provides a critical analysis of the existing literature on ICU nutrition, highlighting important methodological shortcomings of many trials and meta-analyses and underlining the urgent need for high-quality research in this field. Recent adequately designed randomized controlled trials suggest that trophic enteral feeding during the first week of critical illness is as good as full enteral feeding and that early addition of parenteral nutrition to insufficient enteral nutrition does not provide any benefit and worsens morbidity.

Complications of artifi cial nutrition include aspiration pneumonia, feeding tube dislocation, diarrhea, abdomi nal hypertension, intestinal ischemia, catheter sepsis, liver steatosis, hyperglycemia, dyslipidemia and refeeding syndrome [20][21][22]. Nutrition also suppresses autophagy, a survival mechanism serving to recycle intracellular nutrients and maintain energy homeostasis during nutrient deprivation. Recent evidence suggests that autophagy is essential for the immune response and for housekeeping functions such as removal of toxic protein aggregates and damaged organelles, and thus it could be critical for recovery from organ failure [22][23][24]. A recent study of critically ill rabbits showed that early PN, especially the provision of proteins and lipids, suppressed the ubiquitin-proteasome pathway, thereby contributing to the preservation of muscle mass, but also evoked a phenotype of autophagy defi ciency in liver and skeletal muscle, suggesting that the maintenance of muscle mass might come at the price of accumulation of toxic protein aggregates, thus compromising function [25]. Th is same phenotype has also been found in the liver and muscle of fed critically ill patients [26] and was recently shown to contribute to vital organ failure in an animal model of critical illness [27]. Autophagy-defi cient mice show muscle loss with accumulation of toxic proteins and dysfunctional organelles [28,29], suggesting that nutrition-induced suppression of autophagy might even have a negative impact on fat-free mass.
A direct link between nutrition and organ failure was also shown in a series of animal experiments evaluating the impact of diff erent nutritional regimen on ischemiareperfusion injury in the kidney. In these experiments, nutrition -more specifi cally protein rather than glucoseappears to increase the extent of kidney injury [30][31][32].
Given these pros and cons of artifi cial nutrition in critical illness, clinical trials should have clinical (morbidity, mortality, long-term functional outcome) and not nutritional or other surrogate endpoints. Also surprising is that the majority of the existing literature concentrates on when, how and what to feed, whereas the benefi cial eff ect of feeding per se is rarely questioned.
What is the evidence that (more) nutrition improves outcome?

Observational trials
Several observational trials (OTs) suggest an association between caloric defi cit and adverse outcomes. Th is caloric defi cit mainly develops during the fi rst days in the ICU, when EN is insuffi cient due to hemodynamic instability, slow progression to target, poor tolerance (gastric retention/vomiting) and interruptions for procedures [33,34]. In a large multinational prospective OT, an inverse relationship between caloric input and risk of mortality was only signifi cant for patients with body mass index <25 or >35 kg/m 2 [35]. In another OT, proteins and energy were titrated separately for patients with an expected pro longed ICU stay. Th e authors concluded that achieving the energy/protein target is associated with lower 28-day mortality. However, ICU and hospital mortalities showed an opposite trend (surprisingly translated into a hazard ratio <1 in the unadjusted Cox proportional hazard analy sis, suggesting erroneous censoring as discussed below). ICU and hospital stays and duration of mecha nical ventilation were shortest for patients reaching neither the energy target nor the protein target [36]. In other OTs, higher caloric intake was associated with poorer outcomes, including mortality and infections [37,38].
Early initiation of (mainly enteral) nutrition was associated with a lower mortality in several OTs [39,40]. However the incidence of ventilator-associated pneumonia was lower and the ICU stay was shorter in the late initiation group [39]. Another OT found more gastrointestinal complications and a longer ICU stay with early initiation of EN in patients with high illness severity [41]. Early addition of PN to insuffi cient EN was not associated with shorter hospital stay in another large OT [42,43].
Th e evidence for the recommended amount of protein in artifi cial nutrition is limited. Four OTs using isotopes [44,45], nitrogen balances [46] or measurements of body composition [47] in an aggregate total of 90 septic, trauma or burn patients have shown that the net balance between protein synthesis and protein breakdown is improved with protein intakes up to 1 to 1.5 g/kg/day, whereas any further increase in protein intake is wasted in oxidative pathways. However, there are no randomized controlled trials (RCTs) showing any benefi t of more protein intake on clinically relevant outcomes.
A major methodological problem with OTs investigating ICU nutrition is confounding by indication. Patients who receive less or later feeding also have an explanatory reason why feeding is inadequate or delayed, which is mainly related to illness severity. In addition, association is never proof of causality. True evidence for the optimal amount, route and timing of nutritional support in critically ill patients can hence only be provided by RCTs.
Another methodological (statistical) issue that often results in incorrect interpretation of clinical trials is inappropriate censoring [48,49] for any type of time-toevent survival analysis (for example, Kaplan-Meier statistics or Cox proportional hazards models) when the duration of follow-up is not identical for all subjects. To illustrate this problem we refer to the trial evaluating the impact of reaching nutritional targets on survival [36], where patients were followed-up throughout their hospital stay. Patients reaching the nutritional target had longer ICU and hospital stays and also numerically higher ICU and hospital mortalities. However, un corrected Cox proportional hazard analysis, censoring patients at last follow-up, suggested a lower mortality risk for patients reaching target. Th ese data are internally inconsistent and can only be explained by inappropriate use of informative censoring.
Censoring is informative when the reason why a patient is censored (no longer included in the analysis) is not just by chance but in fact related to the outcome [48,49]. In the above-mentioned study, patients discharged earlier (and subsequently omitted from the analysis) do not have the same mortality risk as those who remain in the ICU or hospital [36]. Th e deletion/censoring of early discharge patients lowers the number of patients at risk (the denominator in the calculation) in the cohort not reaching nutritional target and, as a result, for a similar or lower numerical mortality, the proportion (altered by the lower denominator) of patients dying in that cohort seems higher (see illus tration in Figure 1). Th is problem is preventable by obtain ing landmark time-to-event analysis (for example, 90-day mortality) in all patients. Conversely, if the duration of stay in the ICU or hospital is the endpoint, patients dying before ICU/hospital discharge should be censored after the last patient has been discharged alive, to account for such competing risk between length of stay and mortality [50].

Meta-analyses of older randomized controlled trials
RCTs that compare any form or amount of (enteral or parenteral) nutrition with tolerating starvation (withhold ing/delaying nutritional support) in critically ill patients are non-existent [51]. Results from older (all largely underpowered) RCTs on other aspects of ICU nutrition have been summarized in meta-analyses, many of them including elective surgery patients. Two recent meta-analyses, limited to critically ill patients, showed a signifi cant reduction of infections with early EN compared with late EN, whereas mortality reached significance in only one of the trials [52,53]. Th e quality of the included trials was poor, however, with high risk of bias [54][55][56]. Meta-analyses comparing EN with PN in critically ill patients suggested a reduction of infectious complica tions with EN but the mortality results were confl icting [57,58]. In many of the included RCTs, PN patients received more calories and had higher blood glucose levels than EN patients. Finally, two metaanalyses of trials comparing EN with a combination of PN and EN did not establish clinical benefi t [58,59].
A common problem with all these meta-analyses is that they included small, old studies with high risk of bias (for example, lack of randomization concealment, no blinded outcome assessment, no intention-to-treat analysis). Other problems are the inclusion of diff erent patient populations with variable illness severity and baseline nutritional status, variable nutritional formulations, inconsistent defi nitions of early, widely varying caloric input, and so forth. Better quality research on ICU nutrition is therefore urgently needed.

Recent randomized controlled trials
Two recent RCTs have compared diff erent amounts of EN in ICU patients [60,61]. Th e fi rst single-center trial compared the provision of 90 to 100% of the caloric goal with permissive underfeeding, defi ned as 60 to 70% of the target in 240 (of 1,587 screened) patients receiving EN and expected to remain in the ICU for at least 48 hours [60]. Permissive underfeeding signifi cantly reduced the hospital mortality (risk ratio = 0.71; 95% confi dence interval = 0.50 to 0.99) and tended to reduce 180-day mortality (risk ratio = 0.74; 95% confi dence interval = 0.53 to 1.03). However, the actual energy delivery only diff ered by 10% (59% vs. 71% of target) [60]. Th e small number of patients, without a well-motivated sample size calculation to guarantee the required statistical power, makes this trial prone to observations evoked by chance.
Th e EDEN trial is a multicenter open-label RCT comparing trophic EN with full EN in 1,000 (of 7,968 screened) adult patients with acute lung injury requiring mechanical ventilation [61]. Th e sample size of this trial was determined by a well-rationalized calculation. In the full feeding group, EN was advanced to goal (25 to 30 kcal/kg ideal body weight/day of nonprotein calories) as quickly as possible. Patients in the trophic feeding group received 20 kcal/hour, and advanced to full feeding only if they were still on mechanical ventilation after 6 days. Th e actual caloric intake was 400 kcal/day (25% of target) for trophic EN versus 1,300 kcal/day (80% of target) for full EN. Th e feeding protocol had no signifi cant eff ect on ventilator-free days through 28 days (the primary endpoint), nor on infec tious complications, ICU-free days, organ failure-free days or 60-day mortality. Patients in the full-feeding group had signi ficantly more episodes of gastrointestinal intolerance, higher blood glucose levels and insulin requirement and a greater cumulative net fl uid balance [61]. An important limitation of this trial is the high exclusion rate, mainly for high disease severity or important comorbidity, thus limiting the generalizability of its fi ndings.
Taken together, these two RCTs suggest that limiting enteral caloric delivery during the fi rst ICU days is not deleterious for clinical outcome. Potential limitations are the insuffi cient power to prove equivalence between the nutritional strategies, the absence of prior malnutrition in included patients and the lack of long-term functional outcomes [62].
Two other RCTs evaluated the concept of combined PN and EN to prevent the development of a caloric defi cit [63,64]. Th e Tight Calorie Control Study (TICACOS) was conceived as a pilot trial and hence no sample size calculation was performed. Th is trial included 130 ICU (of 944 screened) patients on mechanical ventilation with an expected ICU stay ≥3 days [63]. Within 48 hours after admission, patients were randomized to the tight-calorie group, in which caloric intake was adjusted daily under supervision of a dietician guided by repeated indirect calorimetry, or a control group receiving a fi xed 25 kcal/kg/day. If the caloric target could not be reached by EN alone, PN was added. In the tight-calorie group, the measured target varied signifi cantly over time and caloric intake exceeded this target, resulting in a cumulative energy balance of 2,008 ± 2,177 kcal. Th ree patients received exclusive PN and 19 patients received combined PN and EN. In the control group, intake was consistently below the calculated target, resulting in a cumulative energy balance of -3,550 ± 4,591 kcal. Only one patient received exclusive PN and seven patients received combined EN and PN. Th e results for hospital mortality, the primary outcome, are only given for the per-protocol analysis (28.5% vs. 48.2%; P = 0.023 in favor of the tight-calorie group). Th e survival curve revealed a trend for decreased mortality in the intention-to-treat analysis and a signifi cantly decreased mortality in the per-protocol analysis. However, both ICU and hospital stays were shorter in the tight-calorie group, clearly introducing the statistical problem of informative censoring/competing risk that we discussed earlier. In addition, not only the ICU stay but also the duration of mechanical ventilation was signifi cantly longer in the tight-calorie group, which also revealed more infectious complications. Th e design of the study also does not allow distinguishing eff ects of indirect calorimetry guidance versus providing more nutrition [63]. A multicenter trial with similar design is currently ongoing (clinicaltrials.gov: NCT01479673).
Th e Early Parenteral Nutrition Completing Insuffi cient Enteral Nutrition in Adult ICU Patients (EPaNIC) trial is the largest RCT in the fi eld of ICU nutrition. Th is trial randomized 4,640 ICU (of 8,703 screened) patients with nutritional risk score ≥3 to either early or late initiation of PN to complete insuffi cient EN [64]. Patients with severe prior malnutrition (body mass index <17 kg/m 2 ; n = 106) were excluded. Patients in both groups received EN according to a standardized protocol. Calculations of  Of particular interest was the subgroup of 517 patients with contraindications for EN. In this subgroup the benefi cial eff ect of withholding early PN, and thus tolerating virtual fasting for 1 week, was even more pronounced. Whether this refl ects benefi t of withholding any nutrition, irrespective of the route, or more deleterious eff ects of more PN remains to be studied [64]. Th e EPaNIC trial is unique by its power, preplanned analysis of hard clinical endpoints and blinding of outcome assessors. Formulated criticisms include the inclusion of a large postcardiac surgery cohort, although showing a similar outcome, and the use of tight glycemic control. Tolerating more PN-induced hyperglycemia, however, is very unlikely to improve outcome with early PN.
In summary, both the TICACOS trial and the EPaNIC trial show a benefi cial eff ect of reduced energy intake on morbidity. Th e diff erence with regard to survival outcome in these trials with markedly diff erent inclusion criteria (the TICACOS trial only included a small and specifi c subset of ICU patients, whereas the EPaNIC trial included a very representative sample of all-comers in the ICU) requires further investigation.

Conclusion
Th e provision of artifi cial nutrition to critically ill patients is largely based on association between malnutrition and poor outcome. Whether and to what extent artifi cial nutrition can prevent/attenuate muscle wasting/weakness or accelerate recovery, however, remains unclear. In addition, artifi cial nutrition during critical illness has deleterious eff ects that should be weighed against any presumed benefi t. Future clinical trials should include hard clinical endpoints.
Th e available evidence suggests that, in patients without severe prior malnutrition, early trophic enteral feeding is at least as good as full EN and that early addition of PN to insuffi cient EN does not provide any benefi t and worsens morbidity. How long critically ill patients can be nutrient restricted or how to determine at which time point the benefi ts of artifi cial nutrition start to outweigh the risks remains to be determined.