Enteral versus parenteral nutrition and enteral versus a combination of enteral and parenteral nutrition for adults in the intensive care unit

Summary of findings for the main comparison. Enteral versus parenteral nutrition for adults in the intensive care unit

Enteral versus parenteral nutrition for adults in the intensive care unit
Patient or population: critically ill adults admitted to the ICU for trauma, emergency, or surgical care; population excluded people with acute pancreatitis Setting: intensive care units in: Brazil, China, Germany, Iran, Italy, Turkey, UK, and USA Intervention: EN Comparison: PN
Outcomes	*Anticipated absolute effects (95% CI)**		Relative effect (95% CI)	Number of participants (studies)	Certainty of the evidence (GRADE)
	Risk with EN	Risk with PN
Mortality	In‐hospital mortality		RR 1.19 (0.80 to 1.77)	361 (6 studies)	⊕⊕⊝⊝ Low^a
	Study population
	229 per 1000 (154 to 340)	192 per 1000
	Mortality within 30 days		RR 1.02 (0.92 to 1.13)	3148 (11 studies)	⊕⊕⊝⊝ Low^b
	Study population
	304 per 1000 (274 to 336)	298 per 1000
	Mortality within 90 days		RR 1.06 (0.95 to 1.17)	2461 (3 studies)	⊕⊝⊝⊝ Very low^c
	Study population
	393 per 1000 (352 to 434)	371 per 1000
	Mortality within 180 days		RR 0.33 (0.04 to 2.97)	46 (1 study)	⊕⊝⊝⊝ Very low^d
	Study population
	130 per 1000	43 per 1000 (5 in 387)
Number of ICU‐free days up to day 28	–	–	–	–	Not measured
Number of ventilator‐free days up to day 28	Mean number of ventilator‐free days: 14.2 (SD ± 12.2)	Mean difference 0 days (0.97 fewer to 0.97 more)	N/A	2388 (1 study)	⊕⊝⊝⊝ Very low^d
Adverse events: aspiration (as reported by study authors at end of study follow‐up period)	Study population		RR 1.53 (0.46 to 5.03)	2437 (2 studies)	⊕⊝⊝⊝ Very low^e
	5 per 1000 (2 to 17)	3 per 1000
Adverse events: sepsis (as reported by study authors at end of study follow‐up period)	Study population		RR 0.59 (0.37 to 0.95)	361 (7 studies)	⊕⊕⊝⊝ Low^f
	123 per 1000 (77 to 199)	209 per 1000
Adverse events: pneumonia (as reported by study authors at end of study follow‐up period)	Study population		RR 1.10 (0.82 to 1.48)	415 (7 studies)	⊕⊕⊝⊝ Low^f
	314 per 1000 (234 to 423)	268 per 1000
Adverse events: vomiting (as reported by study authors at end of study follow‐up period)	Study population		RR 3.42 (1.15 to 10.16)	2525 (3 studies)	⊕⊝⊝⊝ Very low^g
	11 per 1000 (4 to 32)	3 per 1000
*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; EN: enteral nutrition; ICU: intensive care unit; N/A: not applicable; PN: parenteral nutrition; RR: risk ratio; SD: standard deviation.
GRADE Working Group grades of evidence High: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate: we are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low: our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect. Very low: we have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.
^aAll studies had a high risk of performance bias; downgraded one level for study limitations. Studies included a variety of primary diagnoses and evidence was less direct; downgraded one level for indirectness. ^bAll studies had a high risk of performance bias; downgraded one level for study limitations. Studies included a variety of primary diagnoses and study designs and evidence were less direct; downgraded one level for indirectness. ^cAll studies had a high risk of performance bias; downgraded one level for study limitations. Studies included a variety of primary diagnoses and study designs and evidence were less direct; downgraded one level for indirectness. Few studies and one included study had a large number of participants relative to other included studies; downgraded one level for imprecision. ^dData from only one study that had a high risk of performance bias; downgraded one level for study limitations and two levels for imprecision. ^eAll studies had a high risk of performance bias; downgraded one level for study limitations. Studies included a variety of primary diagnoses and evidence was less direct; downgraded one level for indirectness. Few studies and one included study had a large number of participants relative to other included studies; downgraded one level for imprecision. ^fAll studies had a high risk of performance bias; downgraded one level for study limitations. Studies included a variety of primary diagnoses and evidence was less direct; downgraded one level for indirectness. ^gAll studies had a high risk of performance bias; downgraded one level for study limitations. Studies included a variety of primary diagnoses and study designs and evidence were less direct; downgraded one level for indirectness. Few studies, with very few events, and one included study had a large number of participants relative to other included studies; downgraded one level for imprecision.

See: Summary of findings for the main comparison Enteral versus parenteral nutrition for adults in the intensive care unit; Summary of findings 2 Enteral versus enteral and parenteral nutrition for adults in the intensive care unit

Summary of findings 2. Enteral versus enteral and parenteral nutrition for adults in the intensive care unit

Enteral versus enteral and parenteral nutrition for adults in the intensive care unit
Patient or population: critically ill adults admitted to the ICU for trauma, emergency, or post‐surgical care; population excludes participants with acute pancreatitis Setting: intensive care units in: France, Italy, Switzerland, Turkey, and USA Intervention: EN Comparison: EN + PN
Outcomes	*Anticipated absolute effects (95% CI)**		Relative effect (95% CI)	Number of participants (studies)	Certainty of the evidence (GRADE)
	Risk with EN	Risk with EN + PN
Mortality	In‐hospital mortality		RR 0.99 (0.84 to 1.16)	5111 (5 studies)	⊕⊕⊝⊝ Low^a
	Study population
	106 per 1000 (90 to 124)	107 per 1000
	Mortality within 30 days		RR 1.64 (1.06 to 2.54)	409 (3 studies)	⊕⊝⊝⊝ Very low^b
	Study population
	216 per 1000 (140 to 335)	132 per 1000
	Mortality within 90 days		RR 1.00 (0.86 to 1.18)	4760 (2 studies)	⊕⊕⊝⊝ Low^c
	Study population
	115 per 1000 (99 to 135)	115 per 1000
	Mortality within 180 days		RR 1.00 (0.65 to 1.55)	120 (1 RCT)	⊕⊝⊝⊝ Very low^d
	Study population
	400 per 1000 (260 to 620)	400 per 1000
Number of ICU‐free days up to day 28	–	–	–	–	Not measured
Number of ventilator‐free days up to day 28	–	–	–	–	Not measured
Adverse events: aspiration (as reported by study authors at end of study follow‐up period)	–	–	–	–	Not measured
Adverse events: sepsis (as reported by study authors at end of study follow‐up period)	–	–	–	–	Not measured
Adverse events: pneumonia (as reported by study authors at end of study follow‐up period)	350 per 1000 (228 to 538)	250 per 1000	RR 1.40 (0.91 to 2.15)	205 (2 studies)	⊕⊝⊝⊝ Very low^d
Adverse events: vomiting (as reported by study authors at end of study follow‐up period)	–	–	–	–	Not measured
*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; EN: enteral nutrition; ICU: intensive care unit; PN: parenteral nutrition; RCT: randomized controlled trial; RR: risk ratio.
GRADE Working Group grades of evidence High: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate: we are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low: our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect. Very low: we have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.
^aAll studies had high risk of performance bias; downgraded one level for study limitations. Studies included a variety of primary diagnoses and evidence was less direct; downgraded one level for indirectness. ^bAll studies had high risk of performance bias; downgraded one level for study limitations. Studies included a variety of primary diagnoses and evidence was less direct; downgraded one level for indirectness. Few studies with increased risk of imprecision; downgraded one level. ^cBoth studies had high risk of performance bias; downgraded one level for study limitations. Studies included a variety of primary diagnoses and evidence was less direct; downgraded one level for indirectness. ^dData from only one study that had a high risk of performance bias; downgraded one level for study limitations and two levels for imprecision.

Antecedentes

available in

Descripción de la afección

La desnutrición se asocia con una mayor mortalidad y morbilidad que incluye sensibilidad a las complicaciones infecciosas, como las infecciones pulmonares, las infecciones urinarias, las infecciones de la herida y la sepsis, así como sensibilidad a las complicaciones no infecciosas, como la insuficiencia respiratoria y las arritmias cardíacas (Correia 2003; Mogensen 2015).

La enfermedad aguda y crónica, el traumatismo y la inflamación inducen el catabolismo relacionado con el estrés, que aumenta la tasa metabólica a la cual el cuerpo descompone los alimentos. Además de lo anterior, los efectos secundarios relacionados con los fármacos pueden afectar la ingestión y dar lugar a la pérdida del apetito o a náuseas y vómitos, o ambos, y se ha indicado que las rutinas del hospital y la falta de concientización entre el personal de enfermería puede afectar la atención nutricional (Norman 2008). Los pacientes con enfermedades graves, que pueden estar inconscientes, ser incapaces de alimentarse solos o ser incapaces de recibir apoyo nutricional oral, o ambos, presentan una mayor sensibilidad a la desnutrición.

El apoyo nutricional es un aspecto complejo de la atención de los pacientes con enfermedades graves. Esta revisión sistemática procuró específicamente considerar la vía de administración que optimiza la asimilación de la nutrición. No consideró la administración de suplementos nutricionales específicos debido a que algunos ya se examinaron (Allingstrup 2016; Dushianthan 2016; Tao 2014).

Descripción de la intervención

La nutrición enteral (NE) se refiere a la administración de alimentos nutricionalmente completos a través de una sonda en el estómago, el duodeno o el yeyuno (NICE 2006). Este método es apropiado para los pacientes que presentan una ingesta oral inadecuada, pero un sistema digestivo funcional y alguna evidencia indica que es un método efectivo para proporcionar nutrición a grupos particulares de pacientes (p.ej., pacientes con sepsis (Elke 2013); pacientes con pancreatitis aguda (Al‐Omran 2010)). La NE puede ayudar a mantener la función y la integridad de la barrera intestinal (Altintas 2011; King 1999; Kyle 2006), y se asocia con un aumento en la producción de inmunoglobulina A, que a su vez puede proporcionar una mayor protección contra las infecciones de las vías respiratorias. Sin embargo, los pacientes con enfermedades graves pueden no tolerar bien la alimentación enteral y pueden ocurrir efectos secundarios como náuseas y vómitos (Harvey 2015) y necrosis intestinal no oclusiva (Marvin 2000). Además, los volúmenes altos de residuos gástricos pueden permitir la colonización con bacterias y aumentar el riesgo de aspiración y complicaciones, como neumonía asociada al respirador (Altintas 2011), aunque un estudio que evaluó la monitorización del volumen residual gástrico no mostró diferencias en la neumonía asociada al respirador con la ausencia de monitorización (Reignier 2013). Además, la NE puede ser perturbada por la atención del paciente y las intervenciones diagnósticas, en particular en los pacientes que reciben ventilación (Corley 2017), lo que puede afectar la capacidad de la NE de mantener las metas nutricionales (Kyle 2006; Seres 2013). Se ha encontrado algún beneficio a partir de la colocación de la sonda en el duodeno o el yeyuno en lugar del estómago (Alkhawaja 2015).

La nutrición parenteral (NP) no es fisiológica y elude el sistema digestivo y el sistema venoso portal. Administra alimentos nutricionalmente completos por vía intravenosa a través un catéter venoso central o periférico y se puede utilizar como una alternativa para los pacientes que necesitan apoyo nutricional. Tiene la ventaja de la facilidad de administración al paciente (Seres 2013), a menudo sin una intervención adicional necesaria para proporcionar apoyo nutricional cuando todos los componentes se administran a través de un sistema de "todo en uno". Aunque la interrupción de la alimentación durante la asistencia al enfermo no es necesaria, la NP puede aumentar el riesgo de sobrealimentación (Singer 2009). La NP se asocia con una tasa mayor de hiperglucemia; posteriormente, los pacientes pueden requerir control glucémico junto con la NP. Los estudios anteriores han informado una mayor sensibilidad a las complicaciones infecciosas, como las infecciones de la sangre relacionadas con el catéter (Peter 2005).

La NP se puede utilizar para complementar la NE y para lograr los requerimientos energéticos pretendidos cuando la NE sola es inadecuada (Singer 2011).

Los resultados de investigación no están claros con respecto a la ingesta calórica suficiente necesaria para satisfacer los requerimientos energéticos de los pacientes con enfermedades graves, y en la actualidad no hay evidencia que apoye la presuposición de que estos pacientes se benefician con una ingesta normocalórica (80% a 100% de las necesidades energéticas) en lugar de la subalimentación permisiva (menos del 70% de las necesidades energéticas) (Marik 2016). De igual manera, los investigadores han debatido sobre el momento adecuado para el inicio de la nutrición, con ensayos controlados aleatorios (ECA) grandes (p.ej., el estudio EDEN [Early versus delayed enteral feeding to treat people with acute lung injury or acute respiratory distress syndrome] (ARDS Clinical Trials Network 2012); el estudio EPaNIC [Early Parenteral Nutrition Completing Enteral Nutrition in Adult Critically Ill Patients) (Casaer 2011]) que aportaron evidencia que está en contradicción con las guías nutricionales europeas (ESPEN; Singer 2009), que recomiendan la alimentación temprana durante las enfermedades graves (Casaer 2014). Aunque esta revisión procuró específicamente considerar la vía de nutrición, la ingesta calórica y el momento de inicio también son consideraciones importantes.

Por qué es importante realizar esta revisión

La mayoría de las guías de la Society for Parenteral and Enteral Nutrition (ASPEN) recomiendan el uso de NE sobre la NP (Taylor 2016) e indican una reducción de la morbilidad infecciosa y la duración de la estancia hospitalaria en la unidad de cuidados intensivos (UCI) en los pacientes que reciben NE. Estos datos son equivalentes a las guías de la European Society for Parenteral and Enteral Nutrition (ESPEN) (Kreymann 2006; Singer 2009) y del National Institute for Health and Care Excellence del Reino Unido (NICE) (NICE 2006). Sin embargo, estas guías reflejan solamente los resultados de investigación de ECA pequeños publicados antes de estas guías y la evidencia actual se contradice con algunos resultados, por ejemplo, el riesgo de complicaciones infecciosas con la NP.

Existe un gran debate en cuanto a si, cómo y cuándo el apoyo nutricional puede contribuir con una mejoría en los resultados de los pacientes (Casaer 2014; Preiser 2015; Schetz 2013). La nutrición de los pacientes con enfermedades graves tiene una relevancia global, logra beneficios para el paciente y reduce la repercusión sobre los recursos de asistencia sanitaria. Esta revisión procuró específicamente considerar si la vía de administración de la nutrición es un factor significativo en el tratamiento de los pacientes adultos con enfermedades graves e incorpora los hallazgos recientes para evaluar la evidencia de efectos beneficiosos y el riesgo de eventos adversos.

Objetivos

available in

Comparar los efectos de los métodos de nutrición enteral versus parenteral y los efectos de la nutrición enteral versus la combinación de los métodos enterales y parenterales, en pacientes adultos con enfermedades graves, en cuanto a la mortalidad, el número de días sin UCI hasta el día 28 y los eventos adversos.

Métodos

available in

Criterios de inclusión de estudios para esta revisión

Tipos de estudios

Se incluyeron todos los ensayos controlados aleatorios (ECA), incluidos los estudios cuasialeatorios (p.ej., estudios en los que el método de asignación se basó en la alternancia, la fecha de nacimiento o el número de historia clínica) y los estudios aleatorios con asignación al azar grupal.

Tipos de participantes

Se incluyeron todos los pacientes adultos, con más de 16 años de edad, que habían estado en una UCI por al menos 24 horas.

Se incluyeron participantes que habían ingresado por cualquier tipo de trastorno, excepto la pancreatitis aguda, debido a que este grupo de pacientes se examina en otra revisión (Al‐Omran 2010). Se procuró incluir estudios que tuvieran una población combinada que incluyera la pancreatitis aguda, si menos del 50% de los participantes presentaba pancreatitis aguda; se procuró establecer contacto con los autores de los estudios para solicitar información adicional de ser necesario.

Se incluyeron participantes con traumatismos, trastornos urgentes, médicos o posquirúrgicos electivos. Se incluyeron participantes con ventilación mecánica y sin ventilación mecánica.

Cuando no todos los participantes del estudio estaban en la UCI, el estudio se incluyó si los autores informaron que más del 75% de los participantes estaban en la UCI.

Tipos de intervenciones

Se incluyeron los estudios que compararon NE versus NP y los estudios que compararon NE versus NE y NP. Los mismos representan dos grupos de comparación; se analizaron por separado los datos de los estudios que compararon NE versus NP y de los estudios que compararon NE versus NE y NP.

Se incluyó la NE administrada a través de una sonda en el estómago, el duodeno o el yeyuno y la NP que se administró a través de un catéter venoso central o un catéter venoso periférico. Se previó que el protocolo utilizado para administrar la nutrición diferiría entre los estudios. Se incluyó NE y NP iniciada de forma temprana o tardía, y administrada para lograr una meta normocalórica o hipocalórica.

Tipos de medida de resultado

Se procuró establecer si un tipo de método de alimentación reducía la tasa de mortalidad entre los participantes del estudio y se consideraron los datos obtenidos en diferentes puntos temporales, hasta los 180 días. La duración de la estancia hospitalaria en la UCI fue un resultado importante para este tema de la revisión. Debido a que las tasas de mortalidad pueden ser altas en la población incluida y para evitar el efecto de la muerte sobre este resultado, se planificó informar los datos presentados como días sin UCI. De igual manera, se planificó evaluar la duración de la ventilación mecánica como el número de días sin respirador. Estos resultados representan el número de días que un paciente está vivo o que ya no utiliza la ventilación mecánica; por lo tanto, un participante que ha muerto se contaría como con cero días sin UCI o sin respirador. Los eventos adversos representan un resultado importante para esta revisión, y la NE y la NP pueden dar lugar a diferentes eventos adversos. Para cada evento adverso informado por los autores de los estudios, se recopilaron datos durante el período de seguimiento del estudio.

Resultados primarios

Mortalidad (medida: en el hospital, en el transcurso de 30 días, en el transcurso de 90 días y en el transcurso de 180 días).

Resultados secundarios

Número de días sin UCI hasta el día 28.
Número de días sin respirador hasta el día 28.
Eventos adversos según lo informado por los autores de los estudios (incluida la hiperglucemia, la neumonía por aspiración, las infecciones de la sangre relacionadas con el catéter y los eventos gastrointestinales).

Métodos de búsqueda para la identificación de los estudios

Búsquedas electrónicas

We identified RCTs through literature searching with systematic and sensitive search strategies as outlined in Chapter 6.4 of the Cochrane Handbook of Systematic Reviews of Interventions (Higgins 2011). We applied no restrictions to language or publication status.

We searched the following databases for relevant trials:

Cochrane Central Register of Controlled Trials (CENTRAL; 2017, Issue 9);
MEDLINE (OvidSP, 1946 to 3 October 2017);
Embase (OvidSP, 1974 to 3 October 2017).

We developed a subject‐specific search strategy in MEDLINE and used that as the basis for the search strategies in the other listed databases. The search strategy was developed in consultation with the Information Specialist. Search strategies can be found in Appendix 1; Appendix 2; and Appendix 3.

We scanned the following trial registries for ongoing and unpublished trials (8 January 2018):

World Health Organization International Clinical Trials Registry Platform (www.who.int/ictrp/en/);
ClinicalTrials.gov (clinicaltrials.gov).

Búsqueda de otros recursos

We carried out citation searching of identified included studies in Web of Science (apps.webofknowledge.com), on 24 March 2017 and conducted a search of grey literature through Opengrey (www.opengrey.eu./), on 27 April 2017. We scanned reference lists of relevant systematic reviews to search for additional trials. We did not contact study authors or organizations to ask if they were aware of other completed or ongoing studies.

Obtención y análisis de los datos

Two review authors (SL and OSR) independently completed all data collection and analyses before comparing results and reaching consensus. We consulted with a third review author (AS) to resolve conflicts when necessary.

Selección de los estudios

We used reference management software to collate the results of searches and to remove duplicates (Endnote). We used Covidence software to screen results of the search of titles and abstracts and identify potentially relevant studies (Covidence). We sourced the full texts of all potentially relevant studies and considered whether they meet the inclusion criteria (see Criteria for considering studies for this review). We reviewed abstracts at this stage and included these in the review only if they provided sufficient information and relevant results that included denominator figures for each intervention/comparison group. We recorded the number of papers retrieved at each stage and reported this information using a PRISMA flow chart. We reported in the review brief details of closely related but excluded papers.

Extracción y manejo de los datos

We used Covidence software to extract data from individual studies (Covidence). A basic template for data extraction forms is available at www.covidence.org. We adapted this template to include the following information.

Methods: type of study design; setting; dates of study; funding sources.
Participants: number of participants randomized to each group; baseline characteristics (to include "Acute Physiology and Chronic Health Evaluation II" (APACHE II) scores, whether mechanically ventilated and length of time in the ICU before study commencement).
Interventions: details of intervention and comparison nutrition (kilocalories per kilogram received, time of initiation, duration of delivery, use of glycaemic controls).
Outcomes: all relevant review outcomes as measured and reported by study authors.
Outcome data: results of outcome measures.

We considered the applicability of information from individual studies and the generalizability of data to our intended study population (i.e. the potential for indirectness in our review). If we found associated publications from the same study, we created a composite dataset based on all eligible publications.

Evaluación del riesgo de sesgo de los estudios incluidos

Two review authors (SL and OSR) independently assessed study quality, study limitations, and the extent of potential bias by using the Cochrane ’Risk of bias’ tool (Higgins 2011). We considered the following domains.

Sequence generation (selection bias).
Allocation concealment (selection bias).
Blinding of participants, personnel, and outcome assessors (performance and detection bias).
Incomplete outcome data (attrition bias).
Selective outcome reporting (reporting bias).
Other: use of concomitant drugs.

We anticipated that it would not be feasible for studies to blind participants and personnel and, in the absence of any description of personnel blinding, we assumed that no blinding occurred. However, we anticipated that it was feasible to blind outcome assessors and we considered risk of detection bias (outcome assessor blinding) by each outcome. We considered whether investigators used standard criteria for diagnosis of outcomes, for example, aspiration pneumonia or ventilator‐acquired pneumonia, which may be subject to clinician bias.

For each domain, we judged whether study authors had made sufficient attempts to minimize bias in their study design. We made judgements using three measures; high, low, and unclear risk of bias. We recorded this judgement in ’Risk of bias’ tables and presented a summary ’Risk of bias’ figure.

Medidas del efecto del tratamiento

We collected dichotomous data for mortality and adverse events, and continuous data for number of ICU‐free days and number of ventilator‐free days. We reported dichotomous data as risk ratios (RR) to compare groups, and continuous data as a mean difference (MD). We reported 95% confidence intervals (CI).

Cuestiones relativas a la unidad de análisis

(See Differences between protocol and review.) We conducted separate analysis for the comparison arms PN, and EN and PN; this method avoided double‐counting in multi‐arm studies.

In the event of cluster trials, we would have defined the unit of allocation as the ICU or the hospital rather than the individual participant and analysed data accordingly, calculating effect estimates using the generic inverse variance method (Higgins 2011).

Manejo de los datos faltantes

In the event that study authors did not account for missing data, we would have contacted them for information. We considered data to be complete if losses were reported and explained by study authors and we combined no incomplete data in the meta‐analysis.

Evaluación de la heterogeneidad

We assessed whether evidence of inconsistency was apparent in our results by considering heterogeneity. We assessed clinical heterogeneity by comparing similarities in our included studies between study designs, participants, interventions, and outcomes, and used the data collected as stated under Data extraction and management. We assessed statistical heterogeneity by calculating the Chi² test or I² statistic and judged any heterogeneity above an I² value of 60% and a Chi² P value less than or equal to 0.05 to indicate moderate to substantial statistical heterogeneity (Higgins 2011).

In addition to looking at statistical results, we considered point estimates and overlap of CIs. If CIs overlap, then results are more consistent. Combined studies may show a large consistent effect but with significant heterogeneity. Therefore, we planned to interpret heterogeneity with caution (Guyatt 2011a).

Evaluación de los sesgos de notificación

We attempted to source published protocols for each of our included studies by using clinical trials registers. We compared published protocols with published study results to assess the risk of selective reporting bias. If we identified sufficient studies reporting on an outcome (i.e. more than 10 studies (Higgins 2011)), we planned to generate a funnel plot to assess risk of publication bias in the review; an asymmetrical funnel plot may suggest publication of only positive results (Egger 1997).

Síntesis de los datos

We completed meta‐analyses of outcomes for which we had comparable effect measures from more than one study, and when measures of heterogeneity indicated that pooling of results was appropriate. We did not pool studies that had a high level of clinical heterogeneity and moderate to high statistical heterogeneity indicated by I² statistics and Chi² P values. We used the statistical calculator provided in Review Manager to perform meta‐analysis (Review Manager 2014).

For dichotomous outcomes, for example, mortality rate, we calculated the RR using summary data presented in each trial. We used the Mantel‐Haenszel effects model. If events had been extremely rare (one per 1000), we would have used the Peto odds ratio (Higgins 2011). For continuous outcomes, we aimed to use the MD. We used a fixed‐effect statistical model. In the event of finding evidence of moderate statistical or clinical heterogeneity, we would have investigated this by performing subgroup analyses, as below, and analysed data using a random‐effects model to incorporate unexplained heterogeneity.

We calculated CIs at 95% and used a P value less than or equal to 0.05 to judge whether a result was statistically significant. We considered imprecision in the results of analyses by assessing the CI around an effects measure; a wide CI would suggest a higher level of imprecision in our results. A small number of identified studies may also reduce precision (Guyatt 2011b).

Análisis de subgrupos e investigación de la heterogeneidad

Study designs may differ in relation to time of initiation of feeding and target energy requirements given to participants. Therefore, we considered these subgroups for each of our outcomes. We used cut‐offs for time of feeding from the most recent ASPEN guidelines (Taylor 2016), and cut‐offs for target energy requirements from Marik 2016. Critically ill people who are elderly may have different nutritional requirements and metabolism (ASPEN 2002), leading to different responses to EN and PN methods as compared with younger participants. We did not supply a cut‐off age for this subgroup but aimed to separate participants described as 'frail elderly' by study authors from remaining participants. Heterogeneity may be introduced by the types of procedures that participants have undergone or by their reason for admission; people who have had abdominal or bowel surgery and people admitted with gastrointestinal complications may have greater difficulty with ingestion and digestion. In summary, we aimed to perform subgroup analysis as follows.

Early initiation of feeding (less than 48 hours) versus late initiation of feeding (48 hours or greater).
Normocaloric intake (to match 80% to 100% of energy expenditure) versus hypocaloric intake (less than 70% of energy expenditure).
'Frail elderly' versus other participants.
Gastrointestinal medical or surgical participants versus non‐gastrointestinal medical or surgical participants.

We performed subgroup analysis only when study authors reported outcome data for identified subgroups. In the absence of numerical data, we planned to present qualitative analysis of these factors as a possible source of heterogeneity.

We aimed to perform subgroup analyses on the following outcomes: mortality, number of ICU‐free days up to day 28, and number of ventilator‐free days up to day 28.

Análisis de sensibilidad

We explored the potential effects of decisions made as part of the review process as follows.

We excluded all studies that we judged at high or unclear risk of selection bias.
We assessed decisions made regarding missing data, excluding studies that provided incomplete data.
We conducted meta‐analysis using the alternate meta‐analytical effects model (fixed‐effect or random‐effects).

We compared effect estimates from the above results with effect estimates from the main analysis. We aimed to report differences that altered interpretation of effects.

We aimed to perform sensitivity analyses on the following outcomes: mortality, number of ICU‐free days up to day 28, and number of ventilator‐free days up to day 28.

'Summary of findings' table and GRADE

Two review authors (SL and OSR) independently used the GRADE system to assess the certainty of the body of evidence associated with the following outcomes (Guyatt 2008):

mortality (at time points: in‐hospital, 30 days, 90 days, 180 days);
number of ICU‐free days up to day 28;
number of ventilator‐free days up to day 28;
adverse events as reported by study authors (aspiration, sepsis, pneumonia, and vomiting).

The GRADE approach appraises the certainty of a body of evidence based on the extent to which one can be confident that an estimate of effect or association reflects the item being assessed. Evaluation of the certainty of a body of evidence considers within‐study risk of bias, directness of the evidence, heterogeneity of the data, precision of effect estimates, and risk of publication bias.

We constructed two 'Summary of findings' tables using the GRADE profiler software for the following comparisons in this review (www.guidelinedevelopment.org/):

EN versus PN for adults in the ICU;
EN versus EN and PN for adults in the ICU.

We reached consensus without consulting a third review author.

Results

Description of studies

Results of the search

We screened 4173 titles and abstracts, of which we identified 1369 through forward and backward citation searches. We screened titles from clinical trials registers and grey literature searches. We assessed 147 full texts for eligibility. See Figure 1.

Figure 1

Flow diagram of search strategy.

Included studies

See Characteristics of included studies table.

We included 25 studies (37 publications), with 8816 participants (Abdulmeguid 2007; Abrishami 2010; Adams 1986; Altintas 2011; Bauer 2000; Bertolini 2003; Borzotta 1994; Casaer 2011; Cerra 1988; Chiarelli 1996; Dunham 1994; Engel 1997; Fan 2016; Gencer 2010; Hadfield 1995; Harvey 2014; Heidegger 2013; Justo Meirelles 2011; Kudsk 1992; Peterson 1988; Radrizzani 2006; Rapp 1983; Xi 2014; Young 1987). Two studies were quasi‐randomized (Altintas 2011; Fan 2016), and the remaining studies were RCTs. We found no cluster‐randomized studies. Reports for Bertolini 2003 and Radrizzani 2006 were for the same study, but participants were divided according to criteria for sepsis (severe sepsis in Bertolini 2003; and non‐severe sepsis in Radrizzani 2006), and we included them as separate studies for the purpose of this review. We included one study for which we could only source the abstract (Abdulmeguid 2007); we sourced the full text of all remaining studies.

Study population

Participants had a wide variety of primary diagnoses but all were critically ill. Sixteen studies reported that participants were mechanically ventilated (Abdulmeguid 2007; Adams 1986; Altintas 2011; Bertolini 2003; Borzotta 1994; Cerra 1988; Chiarelli 1996; Dunham 1994; Harvey 2014; Heidegger 2013; Justo Meirelles 2011; Kudsk 1992; Radrizzani 2006; Rapp 1983; Wischmeyer 2017; Xi 2014); nine studies did not describe mechanical ventilation status as part of the inclusion or exclusion criteria (Abrishami 2010; Bauer 2000; Casaer 2011; Engel 1997; Fan 2016; Gencer 2010; Hadfield 1995; Peterson 1988; Young 1987).

Study setting

All studies were conducted in the ICU (Abdulmeguid 2007; Abrishami 2010; Adams 1986; Altintas 2011; Bauer 2000; Bertolini 2003; Casaer 2011; Cerra 1988; Chiarelli 1996; Dunham 1994; Fan 2016; Gencer 2010; Hadfield 1995; Harvey 2014; Heidegger 2013; Justo Meirelles 2011; Kudsk 1992; Peterson 1988; Radrizzani 2006; Rapp 1983; Wischmeyer 2017; Xi 2014), or assumed to be in the ICU (Borzotta 1994; Engel 1997; Young 1987). Eight studies were undertaken in the USA (Adams 1986; Borzotta 1994; Cerra 1988; Dunham 1994; Kudsk 1992; Peterson 1988; Rapp 1983; Young 1987); three were in Italy (Bertolini 2003; Chiarelli 1996; Radrizzani 2006); two were in the UK (Hadfield 1995; Harvey 2014); two were in Turkey (Altintas 2011; Gencer 2010); two were in China (Fan 2016; Xi 2014); and one each in Iran (Abrishami 2010); France (Bauer 2000); Belgium (Casaer 2011); Germany (Engel 1997); Switzerland (Heidegger 2013); and Brazil (Justo Meirelles 2011). One international study was undertaken in Belgium, Canada, France, and the USA (Wischmeyer 2017). One study did not report the country in which it was conducted (Abdulmeguid 2007).

Intervention and comparisons

Seventeen studies compared an EN feeding protocol to a PN feeding protocol (Abdulmeguid 2007; Adams 1986; Altintas 2011; Bertolini 2003; Borzotta 1994; Cerra 1988; Engel 1997; Gencer 2010; Hadfield 1995; Harvey 2014; Justo Meirelles 2011; Kudsk 1992; Peterson 1988; Radrizzani 2006; Rapp 1983; Xi 2014; Young 1987); Engel 1997 was a multi‐arm study with two EN groups (one standard EN formula and one formula supplemented with arginine, omega‐3 fatty acid, nucleotide, and selenium). Six studies compared an EN feeding protocol to a protocol in which EN was supplemented with PN (Abrishami 2010; Bauer 2000; Casaer 2011; Chiarelli 1996; Heidegger 2013; Wischmeyer 2017). Two studies compared EN versus PN and versus combined EN and PN (Dunham 1994; Fan 2016).

Study authors reported initiation of both EN and PN within 48 hours of ICU admission in 13 studies (Adams 1986; Altintas 2011; Bauer 2000; Bertolini 2003; Dunham 1994; Engel 1997; Fan 2016; Harvey 2014; Justo Meirelles 2011; Kudsk 1992; Peterson 1988; Radrizzani 2006; Wischmeyer 2017). One study reported that all participants were given PN within 24 to 36 hours and that EN was initiated in one group at four days (Chiarelli 1996). One study reported that initiation of EN in one group was after at least 14 days of fasting, and study authors did not state at which time point PN was initiated (Xi 2014). Two studies initiated supplemental PN after all participants had been given EN for three days (Casaer 2011; Heidegger 2013). The remaining five study authors did not report time of initiation of feeding (Abdulmeguid 2007; Abrishami 2010; Gencer 2010; Hadfield 1995; Young 1987).

Outcomes

All studies, except Engel 1997, reported participant deaths; some studies did not clearly report time points and some studies reported deaths as participant losses with mortality as a reason for withdrawal from the study. No studies reported data for number of ICU‐free days up to day 28, and one study reported data for number of ventilator‐free days up to day 28 (Harvey 2014). Study authors reported adverse events which were: mechanical (Borzotta 1994; Casaer 2011; Harvey 2014); metabolic (Borzotta 1994; Fan 2016; Harvey 2014); gastrointestinal (Adams 1986; Altintas 2011; Bauer 2000; Borzotta 1994; Casaer 2011; Cerra 1988; Chiarelli 1996; Fan 2016; Harvey 2014); and infective (Abdulmeguid 2007; Adams 1986; Altintas 2011; Borzotta 1994; Casaer 2011; Engel 1997; Fan 2016; Gencer 2010; Justo Meirelles 2011; Heidegger 2013; Kudsk 1992; Rapp 1983; Wischmeyer 2017; Xi 2014; Young 1987).

Funding sources

Study authors reported funding sources in 11 studies (Abrishami 2010; Adams 1986; Bertolini 2003; Borzotta 1994; Casaer 2011; Hadfield 1995; Harvey 2014; Heidegger 2013; Radrizzani 2006; Rapp 1983; Wischmeyer 2017). Three studies noted no involvement in trial management from funders (Casaer 2011; Heidegger 2013; Radrizzani 2006); remaining studies reported no details of funders' involvement.

Excluded studies

We excluded 99 articles after reading the full text. We reported details of 32 of these studies in Characteristics of excluded studies. Of these 32 studies, we excluded studies in which the setting was not reported and we could not assume it was the ICU (Arefian 2007; Baigrie 1996; Braga 1996; Braga 1998; Braga 2001; Chen 2004; DiCarlo 1999; Dong 2010; Hermann 2004; Kim 2012; Klek 2008; Klek 2011; Malhotra 2004; McArdle 1981; Moore 1989; Reynolds 1997; Ryu 2009; Sand 1997; Suchner 1996; Van Barneveld 2016; Xiao‐Bo 2014; Yu 2009; Zhang 2016; Zhu 2012), or too few participants were in the ICU (Woodcock 2001). We excluded two studies of participants with acute pancreatitis (Abou‐Assi 2002; Pupelis 2001). One study compared EN versus PN in people in the ICU (Zhang 2005), but reported that participants in the EN group were also given PN as required for the first three to four days and it was therefore ineligible for our review, and one study was described as an RCT but not all participants randomized to the control group received EN (Doig 2013). We excluded one study that compared early‐goal directed nutrition (EGDN) versus EN in people in the ICU; the EGDN group were given EN supplemented with PN; however, the supplemented PN was only given if required to meet feeding goals and study authors did not report which participants had and did not have PN (Allingstrup 2017). One study compared EN versus PN but feeding took place for only one day in the ICU and feeding was continued for an additional six days on the ward (Fujita 2012). We excluded one abstract that had not been published as a full report (Zanello 1992); the abstract included no outcomes of interest and was published in 1992.

Studies awaiting classification

We were unable to assess eligibility in 11 studies (Braga 1995; Cao 2014; Chen 2011; NCT00522730; NCT01802099; Ridley 2015; Soliani 2001; Theodorakopoulou 2016; Xiang 2006; Xiu 2015; Yi 2015). We identified one study in clinical trials registers that was completed without published results (NCT00522730). From database searches we identified one protocol for a terminated study (NCT01802099), and one study that had been completed but not published (Ridley 2015). We were unable to source full texts for three trials and could not assess eligibility from abstracts (Braga 1995; Chen 2011; Soliani 2001). Four trials were published only as abstracts with insufficient information to assess eligibility (Cao 2014; Theodorakopoulou 2016; Xiu 2015; Yi 2015), and one study report requires translation before we assess eligibility (Xiang 2006). See Characteristics of studies awaiting classification table.

Ongoing studies

We identified two ongoing studies from clinical trials registers (NCT00512122; NCT02022813). See Characteristics of ongoing studies table.

Risk of bias in included studies

We have included a summary of risk of bias assessments in Figure 2 and Figure 3. Blank spaces in the risk bias summary figures indicate that study authors did not report the review outcome.

Figure 2

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies. Blank spaces in tables indicated that study authors did not report the review outcome.

Figure 3

Risk of bias summary: review authors' judgements about each risk of bias item for each included study. Blank spaces in tables indicate that study authors did not report the review outcome.

Allocation

All studies were described as randomized and nine studies provided sufficient information on the method of randomization (Bertolini 2003; Borzotta 1994; Casaer 2011; Harvey 2014; Heidegger 2013; Kudsk 1992; Peterson 1988; Radrizzani 2006; Wischmeyer 2017). Two studies randomized participants according to hospital record number (Altintas 2011; Fan 2016), and we judged these to have high risk of bias. The remaining 14 studies reported insufficient information on method of randomization and we recorded these as having unclear risk of bias (Abdulmeguid 2007; Abrishami 2010; Adams 1986; Bauer 2000; Cerra 1988; Chiarelli 1996; Dunham 1994; Engel 1997; Gencer 2010; Hadfield 1995; Justo Meirelles 2011; Rapp 1983; Xi 2014; Young 1987).

Ten studies described adequate allocation concealment methods, and we judged these to have low risk of selection bias (Altintas 2011; Bertolini 2003; Borzotta 1994; Casaer 2011; Harvey 2014; Heidegger 2013; Kudsk 1992; Peterson 1988; Radrizzani 2006; Wischmeyer 2017) The remaining 15 studies reported no description of methods to conceal allocation and we recorded these as having unclear risk of bias (Abdulmeguid 2007; Abrishami 2010; Adams 1986; Bauer 2000; Cerra 1988; Chiarelli 1996; Dunham 1994; Engel 1997; Fan 2016; Gencer 2010; Hadfield 1995; Justo Meirelles 2011; Rapp 1983; Xi 2014; Young 1987).

Blinding

One study reported that participants in the enteral group were given a placebo parenteral solution and we judged this study to have low risk of performance bias (Bauer 2000). One study reported that personnel were not blinded to the intervention group (Heidegger 2013). The remaining studies did not report whether attempts had been made to blind personnel; adequate blinding would involve intrusive procedures (e.g. central line placement or nasogastric tube placement) and if these procedures were not described we assumed that blinding had not occurred (Abdulmeguid 2007; Abrishami 2010; Adams 1986; Altintas 2011; Bertolini 2003; Borzotta 1994; Casaer 2011; Cerra 1988; Chiarelli 1996; Dunham 1994; Engel 1997; Fan 2016; Gencer 2010; Hadfield 1995; Harvey 2014; Justo Meirelles 2011; Kudsk 1992; Peterson 1988; Radrizzani 2006; Rapp 1983; Wischmeyer 2017; Xi 2014; Young 1987). We judged these studies, and Heidegger 2013, to have high risk of performance bias.

We did not believe that lack of blinding of outcome assessors would influence data for the mortality outcome and we judged all studies to have low risk of detection bias for mortality, regardless of whether study authors reported blinding of outcome assessors. Three studies adequately reported blinding of outcome assessors for the remaining review outcomes and we judged these to have low risk of detection bias (Casaer 2011; Dunham 1994; Heidegger 2013). Nineteen studies reported insufficient details of outcome assessor blinding and we judged these to have unclear risk of detection bias (Abdulmeguid 2007; Adams 1986; Altintas 2011; Bauer 2000; Bertolini 2003; Borzotta 1994; Cerra 1988; Chiarelli 1996; Engel 1997; Fan 2016; Gencer 2010; Harvey 2014; Justo Meirelles 2011; Kudsk 1992; Peterson 1988; Rapp 1983; Wischmeyer 2017; Xi 2014; Young 1987). Three studies included data only for mortality and our assessment of detection bias was limited to this outcome (Abrishami 2010; Hadfield 1995; Radrizzani 2006). We noted whether studies had included criteria for diagnoses of outcomes and we were not concerned that lack of information or type of measurement tools had introduced risk of clinician bias.

Incomplete outcome data

We judged 21 studies to have low risk of attrition bias, as there appeared to be no reported losses (Abdulmeguid 2007; Adams 1986; Altintas 2011; Bertolini 2003; Chiarelli 1996; Engel 1997; Fan 2016; Gencer 2010; Hadfield 1995; Justo Meirelles 2011; Rapp 1983; Wischmeyer 2017; Xi 2014), or losses were few and adequately explained by study authors (Abrishami 2010; Bauer 2000; Casaer 2011; Cerra 1988; Dunham 1994; Harvey 2014; Kudsk 1992; Radrizzani 2006). Four studies had a large number of losses or losses were unevenly distributed between groups and we were unclear whether these losses could influence outcome data (Borzotta 1994; Heidegger 2013; Peterson 1988; Young 1987).

Selective reporting

We were able to source prospective clinical trials registration reports for four studies, of which we judged three studies to have low risk of reporting bias (Casaer 2011; Harvey 2014; Wischmeyer 2017). We noted changes to the clinical trials registration documents after completion of the study with regard to data collection time points and we could not be certain whether these changes may have introduced bias to the results; we judged this study to have an unclear risk of selective reporting bias (Heidegger 2013). We were unable to judge reporting bias for the remaining 21 studies because study authors did not report clinical trials registration reports or published protocols (Abdulmeguid 2007; Abrishami 2010; Adams 1986; Altintas 2011; Bauer 2000; Bertolini 2003; Borzotta 1994; Cerra 1988; Chiarelli 1996; Dunham 1994; Engel 1997; Fan 2016; Gencer 2010; Hadfield 1995; Justo Meirelles 2011; Kudsk 1992; Peterson 1988; Radrizzani 2006; Rapp 1983; Xi 2014; Young 1987).

Baseline characteristics

Three studies reported some baseline imbalances between groups and we did not know if these differences could influence outcome data (Altintas 2011; Bertolini 2003; Radrizzani 2006). One study was an abstract and did not provide sufficient detail on baseline characteristics (Abdulmeguid 2007). We judged 21 studies to have low risk of bias for baseline characteristics because data for characteristics were comparable between groups (Abrishami 2010; Adams 1986; Bauer 2000; Borzotta 1994; Casaer 2011; Cerra 1988; Chiarelli 1996; Dunham 1994; Engel 1997; Fan 2016; Gencer 2010; Hadfield 1995; Harvey 2014; Heidegger 2013; Justo Meirelles 2011; Kudsk 1992; Peterson 1988; Rapp 1983; Wischmeyer 2017; Xi 2014; Young 1987).

Other potential sources of bias

We considered whether differences between intervention and comparison groups could have introduced bias; in particular we considered nutritional protocols, patient management and use of concomitant medication, and glycaemic controls. We noted some differences in 12 studies (Abdulmeguid 2007; Adams 1986; Bertolini 2003; Borzotta 1994; Chiarelli 1996; Dunham 1994; Gencer 2010; Hadfield 1995; Harvey 2014; Kudsk 1992; Radrizzani 2006; Young 1987). We were not able to judge if these differences influenced study outcome data and we reported these as having unclear risk of bias. We identified no other sources of bias in the remaining studies.

Effects of interventions

Enteral nutrition versus parenteral nutrition

Primary outcome

1. Mortality

We noted that some studies reported loss of randomized participants from analysis due to death (Borzotta 1994; Dunham 1994; Kudsk 1992; Peterson 1988; Young 1987). We included participants from three of these studies (Borzotta 1994; Dunham 1994; Kudsk 1992), as data for our primary analysis; we did not include mortality data from Young 1987 or Peterson 1988 because study authors did not report to which intervention group these participants belonged. Data were grouped according to time point.

In hospital

Deaths were reported in Abrishami 2010 during the seven‐day study period, at various time points up to day 18 in Borzotta 1994, and at four days in Kudsk 1992 and we assumed that these occurred in hospital. Three studies reported ICU mortality and did not report hospital mortality (Bertolini 2003; Cerra 1988; Heidegger 2013); in this instance, we included data for ICU mortality in this analysis.

One feeding route rather than the other may make little or no difference to in‐hospital mortality (RR 1.19, 95% CI 0.80 to 1.77; 361 participants; 6 studies; I² = 3%; low‐certainty evidence; Analysis 1.1). We used GRADE to downgrade by two levels; we were concerned by study limitations and indirectness. See summary of findings Table for the main comparison.

Within 30 days

Six studies did not specify time points and we reported outcome data for these studies with mortality at 30 days (Abdulmeguid 2007; Chiarelli 1996; Fan 2016; Gencer 2010; Hadfield 1995; Justo Meirelles 2011).

One feeding route rather than the other may make little or no difference to mortality within 30 days (RR 1.02, 95% CI 0.92 to 1.13; 3148 participants; I² = 32%; low‐certainty evidence; Analysis 1.2). We used GRADE to downgrade by two levels; we were concerned by study limitations and indirectness. See summary of findings Table for the main comparison.

Within 90 days

Three studies reported mortality within 90 days (Harvey 2014; Rapp 1983; Young 1987). It is uncertain whether one feeding route rather than another reduces mortality within 90 days because the certainty of the evidence is very low (RR 1.06, 95% CI 0.95 to 1.17; 2461 participants; I² = 55%; Analysis 1.3). We used GRADE to downgrade by three levels; we were concerned by study limitations, indirectness, and imprecision. See summary of findings Table for the main comparison.

Within 180 days

One study comparing EN versus PN reported mortality from one to four months; we assumed that there were no earlier deaths and we included it as mortality data within 180 days (Adams 1986). Study authors reported one death in the EN group (23 participants) and three deaths in the PN group (23 participants). We used the Review Manager 5 calculator to calculate an effect estimate (RR 0.33, 95% CI 0.04 to 2.97) (Review Manager 2014). It is uncertain whether either feeding route affects number of people who die within 180 days because the certainty of this evidence is very low; we were concerned by study limitations and imprecision. See summary of findings Table for the main comparison.

Secondary outcomes

1. Number of intensive care unit‐free days up to day 28

No studies reported data for number of ICU‐free days.

2. Number of ventilator‐free days up to day 28

One study (2388 participants included in the analysis) reported data for number of ventilator‐free days up to day 28 (Harvey 2014). Study authors reported little or no difference in number of days free of respiratory support (EN group: mean 14.2 (standard deviation (SD) ± 12.2) days versus PN group: mean 14.2 (SD ± 12.1) days; P = 0.94). It is uncertain whether either feeding route affected the number of ventilator‐free days up to day 28 because the certainty of the evidence is very low; we were concerned by study limitations and imprecision. See summary of findings Table for the main comparison.

3. Adverse events as reported by study authors

Study authors did not always describe outcomes as 'adverse events.' We collected outcomes as described by study authors, which we categorized as mechanical events, metabolic events, gastrointestinal events, and infective events. We combined data when more than one study reported an event, and when data were reported as 'number of participants' with an event rather than number of events. We reported single study data of adverse events in Table 1.

Table 1. Adverse events for single studies: enteral nutrition versus parenteral nutrition

Study ID	Description of event	EN group (n/N)	PN group (n/N)
Mechanical events
Adams 1986	Clogged jejunostomy tube	9/23	N/A
	Disconnected line	N/A	1/23
	Line eroded into right upper lobe bronchus	N/A	1/23
	Malfunctioned line	N/A	7/23
Dunham 1994	Transpyloric tube occlusion	2/12	0/15
	Failure to intubate	0/12	0/15
	Withdrawal of tube by participant	1/12	N/A
Metabolic events
Adams 1986	Hepatic failure	1/23	1/23
	Acute renal failure	1/23	1/23
	Pancreatitis	2/23	1/23
Fan 2016	Hypoproteinaemia	22/40	32/40
Harvey 2014	Electrolyte disturbance	5/1197	8/1191
Gastrointestinal events
Adams 1986	Nausea, cramps, bloating	19/23	16/23
Adams 1986	Gastrointestinal bleeding	0/23	0/23
Dunham 1994	Gastric reflux	0/12	0/15
	Ileus	1/12	0/15
	Small bowel ileus	0/12	1/15
Fan 2016	Stress ulcer	7/40	19/40
Harvey 2014	Elevated liver enzymes	7/1197	3/1191
	Jaundice	1/1197	1/1191
	Ischaemic bowel	0/1197	1/1191
Xi 2014	Anastomotic leak	2/22	6/23
Infective events
Adams 1986	Persistent fever without obvious cause	1/23	5/23
Altintas 2011	Catheter infection	2/30	4/41
Borzotta 1994	Meningitis	2/28	0/21
	Sinusitis	3/28	6/21
	Bronchitis	6/28	6/28
	Clostridium difficile	2/28	4/21
	Peritonitis	0/28	1/21
Fan 2016	Intracranial infection	7/40	13/40
Fan 2016	Pyaemia	3/40	19/40
Gencer 2010	Pulmonary infection	2/30	2/30
Kudsk 1992	Empyema	1/51	4/45
Young 1987	Aspiration pneumonia	9/28	3/23
Young 1987	Infection (type of infection not described)	5/28	4/23

EN: enteral nutrition; n: number of participants with an event; N: total number randomized to group; N/A: not applicable; PN: parenteral nutrition.

Mechanical events

Two studies comparing EN versus PN reported data for aspiration (Borzotta 1994; Harvey 2014). It is uncertain whether one feeding route rather than another reduces aspiration because the certainty of this evidence is very low (RR 1.53, 95% CI 0.46 to 5.03; 2437 participants; I² = 0%; Analysis 1.4). We used GRADE to downgrade by three levels; we were concerned by study limitations, indirectness, and imprecision. See summary of findings Table for the main comparison.

Two studies comparing EN versus PN reported data for pneumothorax, with little or no difference in incidences of pneumothorax according to feeding group (RR 1.46, 95% CI 0.19 to 11.22; 2437 participants; I² = 0%; Analysis 1.5) (Borzotta 1994; Harvey 2014).

Single studies reported data for malfunctioned line, clogged jejunostomy tube, accidental disconnected line, and eroded line (Adams 1986), and one study reported data for transpyloric tube occlusion, failure to intubate, and withdrawal of tube by participant (Dunham 1994). See Table 1.

Metabolic events

Two studies comparing EN versus PN reported data for hyperglycaemia (Borzotta 1994; Harvey 2014). We found fewer people had hyperglycaemia who were given EN (RR 0.57, 95% CI 0.35 to 0.93; 2437 participants; I² = 0%; Analysis 1.6). We noted that very few people in either group had hyperglycaemia in one large study (Harvey 2014).

Single studies reported data for hepatic failure, acute renal failure, and pancreatitis (Adams 1986), electrolyte disturbance (Harvey 2014), and hypoproteinaemia (Fan 2016). See Table 1.

Gastrointestinal events

Three studies comparing EN versus PN reported data for vomiting (Altintas 2011; Cerra 1988; Harvey 2014). It is uncertain whether PN leads to a reduction in vomiting because the certainty of this evidence is very low (RR 3.42, 95% CI 1.15 to 10.16; 2525 participants; I² = 0%; Analysis 1.7). We used GRADE to downgrade by three levels; we were concerned by study limitations, indirectness, and imprecision. See summary of findings Table for the main comparison.

Six studies comparing EN versus PN reported data for diarrhoea (Adams 1986; Altintas 2011; Borzotta 1994; Cerra 1988; Fan 2016; Young 1987). We found that fewer people had diarrhoea when given PN (RR 2.17, 95% CI 1.72 to 2.75; 363 participants; I² = 57%; Analysis 1.8).

Three studies comparing EN versus PN reported data for abdominal distension, with little or no difference in incidence of abdominal distension according to feeding regimen (RR 1.53, 95% CI 0.34 to 6.96; 2505 participants; I² = 0%; Analysis 1.9) (Altintas 2011; Harvey 2014; Peterson 1988).

Single studies also reported data for nausea, bloating or cramps, and gastrointestinal bleeding (Adams 1986); gastric reflux, ileus, and small bowel ileus (Dunham 1994); stress ulcer (Fan 2016); jaundice, ischaemic bowel, and elevated liver enzymes (Harvey 2014); and anastomotic leak (Xi 2014). See Table 1.

Infective events

Seven studies comparing EN versus PN reported data for sepsis (Altintas 2011; Engel 1997; Justo Meirelles 2011; Kudsk 1992; Peterson 1988; Xi 2014; Young 1987). EN may reduce incidences of sepsis (RR 0.59, 95% CI 0.37 to 0.95; 361 participants; I² = 27%; low‐certainty evidence; Analysis 1.10). We used GRADE to downgrade by two levels; we were concerned by study limitations and indirectness. See summary of findings Table for the main comparison.

Seven studies comparing EN versus PN reported data for pneumonia, or aspiration pneumonia, or ventilator‐acquired pneumonia (Adams 1986; Altintas 2011; Borzotta 1994; Fan 2016; Justo Meirelles 2011; Kudsk 1992; Young 1987). One study reported data for pneumonia and aspiration pneumonia and we included only data for pneumonia in the analysis (Young 1987). One feeding regimen rather than another may make little or no difference to pneumonia (RR 1.10, 95% CI 0.82 to 1.48; 415 participants; I² = 55%; low‐certainty evidence; Analysis 1.11). We used GRADE and downgraded by two levels; we were concerned by study limitations and indirectness. See summary of findings Table for the main comparison.

Three studies comparing EN versus PN reported data for intra‐abdominal infection or intra‐abdominal abscess (Adams 1986; Gencer 2010; Kudsk 1992). We found fewer intra‐abdominal infections when EN was given (RR 0.26, 95% CI 0.07 to 0.89; 202 participants; I² = 0%; Analysis 1.12).

Three studies reported data for wound infection (Adams 1986; Borzotta 1994; Gencer 2010). We found little or no difference between groups in number of participants with a wound infection (RR 1.45, 95% CI 0.55 to 3.82; 155 participants; I² = 55%; Analysis 1.13).

Three studies reported data for urinary tract infection (Borzotta 1994; Gencer 2010; Young 1987). We found little or no difference between groups in number of participants with a urinary tract infection (RR 1.48, 95% CI 0.65 to 3.40; 160 participants; I² = 49%; Analysis 1.14).

Single studies also reported data for: persistent fever without obvious cause (Adams 1986); catheter infections (Altintas 2011); meningitis, sinusitis, bronchitis, Clostridium difficile, and peritonitis (Borzotta 1994); intracranial infection and pyaemia (Fan 2016); pulmonary infection (Gencer 2010); empyema (Kudsk 1992); and aspiration pneumonia and infection (type of infection not described) (Young 1987). See Table 1.

Bertolini 2003 reported that there were no severe adverse events related to the intervention or comparison and Radrizzani 2006 reported no severe adverse events related to the intervention. Hadfield 1995 did not report adverse events. Abdulmeguid 2007 collected data for nosocomial bloodstream infections and septic morbidity but these were not clearly reported in the abstract. Rapp 1983 reported data for some participants who had sepsis but this was not clearly reported.

Subgroup analysis

1. Early initiation of feeding (less than 48 hours) versus late initiation of feeding (48 hours or greater)

Eleven studies comparing EN versus PN initiated feeding within 48 hours (Adams 1986; Altintas 2011; Bertolini 2003; Dunham 1994; Engel 1997; Fan 2016; Harvey 2014; Justo Meirelles 2011; Kudsk 1992; Peterson 1988; Radrizzani 2006). No studies reported late initiation of EN and late initiation of PN, therefore we could not conduct subgroup analysis for this comparison.

2. Normocaloric intake (to match 80% to 100% of energy expenditure) versus hypocaloric intake (less than 70% of energy expenditure)

We considered possible subgroup analysis based on terms used by study authors to describe whether intake was formulated to be normocaloric or hypocaloric; we did not make judgements based on other information such as target rates (measured as kilocalories/kilogram). No studies described intake as normocaloric or hypocaloric, therefore, we did not conduct subgroup analysis.

3. 'Frail elderly' versus other participants

We identified no studies that specified inclusion of frail elderly participants, or subdivided participant characteristics by this description.

4. Gastrointestinal medical or surgical participants versus non‐gastrointestinal medical or surgical participants

Three studies comparing EN versus PN included participants with only abdominal injury or who had gastrointestinal surgery (Gencer 2010; Kudsk 1992; Peterson 1988). For the relevant outcomes, we compared these with studies in which participants did not have a primary diagnosis of gastrointestinal medical or surgical conditions. We could not be certain of primary diagnoses in Abdulmeguid 2007, and did not include this study in subgroup analysis.

Subgroup analysis showed little or no difference in rates of in‐hospital mortality (Chi² = 0.05, degrees of freedom (df) = 1 (P = 0.83), I² = 0%) based on whether participants were gastrointestinal surgical or medical participants (RR 0.88, 95% CI 0.06 to 13.74; 98 participants; 1 study), or participants who did not have a primary diagnosis of a gastrointestinal surgical or medical condition (RR 1.19, 95% CI 0.80 to 1.77; 361 participants; 6 studies; I² = 22%; Analysis 1.15).

Subgroup analysis showed no difference in rates of mortality at 30 days (Chi² = 0.25, df = 1 (P = 0.62), I² = 0%) based on whether participants were gastrointestinal surgical or medical participants (RR 0.67, 95% CI 0.12 to 3.71; 60 participants; 1 study), or participants who did not have a primary diagnosis of a gastrointestinal surgical or medical condition (RR 1.03, 95% CI 0.93 to 1.15; 3008 participants; 9 studies; I² = 41%; Analysis 1.16).

Sensitivity analysis

1. Selection bias

We assessed six studies as having low risk of selection bias for both sequence generation and allocation concealment (Bertolini 2003; Borzotta 1994; Harvey 2014; Kudsk 1992; Peterson 1988; Radrizzani 2006). In sensitivity analysis, we excluded studies that had high or unclear risk of both sequence generation and allocation concealment. For in‐hospital mortality, this altered the effect estimate with fewer deaths for participants given PN (RR 2.66, 95% CI 1.04 to 6.85; 196 participants; 3 studies; I² = 0%). There was no difference in effect estimates for mortality within 30 days, and within 90 days.

2. Attrition bias

We judged two studies to have unclear risk of attrition bias and performed sensitivity analysis by excluding them from appropriate analyses (Borzotta 1994; Young 1987). For the comparison EN versus PN, we noted no change in effect for in‐hospital mortality, mortality at 30 days, and mortality at 90 days.

3. Effects model

We reanalysed our mortality data using a random‐effects model; this did not change the effect.

Enteral nutrition versus enteral nutrition and parenteral nutrition

Primary outcome

1. Mortality

One study reported loss of randomized participants from analysis due to death (Dunham 1994), and we included these participants for our primary analysis.

In hospital

Five studies comparing EN versus EN and PN reported data for in‐hospital mortality (Abrishami 2010; Casaer 2011; Dunham 1994; Heidegger 2013; Wischmeyer 2017). One feeding regimen rather than the other may make little or no difference to in‐hospital mortality (RR 0.99, 95% CI 0.84 to 1.16; 5111 participants; I² = 0%; low‐certainty evidence; Analysis 2.1). We used GRADE to downgrade by two levels; we were concerned by study limitations and indirectness. See summary of findings Table 2.

Within 30 days

One study did not report the time point for mortality and we included this study in the analysis as mortality within 30 days (Fan 2016). We included three studies in the analysis comparing EN versus EN and PN (Chiarelli 1996; Fan 2016; Heidegger 2013). It is uncertain whether combined EN and PN reduced mortality at 30 days because the certainty of the evidence is very low (RR 1.64, 95% CI 1.06 to 2.54; 409 participants; I² = 0%; Analysis 2.2). We used GRADE to downgrade by three levels; we were concerned by study limitations, indirectness, and imprecision. See summary of findings Table 2.

Within 90 days

Two studies comparing EN versus EN and PN reported data for mortality within 90 days (Bauer 2000; Casaer 2011). One feeding regimen rather than the other may make little or no difference to mortality at 90 days (RR 1.00, 95% CI 0.86 to 1.18; 4760 participants; I² = 0%; low‐certainty evidence; Analysis 2.3). We used GRADE to downgrade the evidence by two levels; we were concerned by study limitations and indirectness. See summary of findings Table 2.

Within 180 days

One study (120 participants) comparing EN versus EN and PN reported mortality at two years; interpretation of a figure of cumulative survival over time reported by study authors showed that all deaths were within 180 days (Bauer 2000). Study authors reported 24 deaths in each group (60 participants per group). We used the Review Manager 5 calculator to obtain the effect estimate (RR 1.00, 95% CI 0.65 to 1.55) (Review Manager 2014). It is uncertain whether one feeding regimen rather than another reduces mortality within 180 days because the certainty of this evidence is very low. We used GRADE to downgrade by three levels; we were concerned by study limitations (one level) and imprecision (two levels). See summary of findings Table 2.

Secondary outcomes

1. Number of intensive care unit‐free days up to day 28

No studies reported data for number of ICU‐free days.

2. Number of ventilator‐free days up to day 28

No studies reported data for number of ventilator‐free days.

3. Adverse events as reported by study authors

Study authors did not always describe outcomes as 'adverse events.' We collected outcomes as described by study authors, which we categorized as mechanical events, metabolic events, gastrointestinal events, and infective events. We combined data when more than one study reported an event. We reported single study data of adverse events in Table 2. Abrishami 2010 did not report adverse event outcomes.

Table 2. Adverse events for single studies: enteral nutrition versus enteral nutrition and parenteral nutrition

Study ID	Description of event	EN group (n/N)	EN + PN group (n/N)
Mechanical events
Casaer 2011	CVC obstruction	9/2328	15/2312
	Nasal bleeding	18/2328	14/2312
	Pneumohaemothorax after CVC placement	0/2328	2/2312
	Subclavian artery puncture	0/2328	2/2312
Dunham 1994	Withdrawal of tube	1/12	0/10
Dunham 1994	Failure to intubate	0/12	2/10
Metabolic events
Fan 2016	Hypoproteinaemia	22/40	7/40
Gastrointestinal events
Casaer 2011	Vomiting or aspiration	284/2328	295/2312
Dunham 1994	Gastric reflux	0/12	2/10
Fan 2016	Stress ulcer	7/40	9/40
infective events
Fan 2016	Pyemia	3/40	10/40
Fan 2016	Intracranial infection	7/40	5/40
Wischmeyer 2017	Catheter bloodstream infection	0/73	7/52
	Intra‐abdominal infection	0/73	4/52
	Upper urinary tract infection	0/73	1/52
	Surgical deep infection	0/73	1/52

CVC: central venous catheter; EN: enteral nutrition; EN + PN: combined enteral and parenteral nutrition; n: number of participants with an event; N: total number randomized to group.

Mechanical events

Two studies comparing EN versus EN and PN reported data for feeding tube obstruction (Casaer 2011; Dunham 1994). There was little or no difference in events between groups (RR 0.96, 95% CI 0.70 to 1.32; 4662 participants; I² = 0%; Analysis 2.4).

One study reported data for failure to intubate, and withdrawal of tube by participant (Dunham 1994), and one study reported data for nasal bleeding, central venous catheter obstruction, pneumohaemothorax, and subclavian artery puncture (Casaer 2011). See Table 2.

Metabolic events

One study comparing EN versus EN and PN reported data for hypoproteinaemia (Fan 2016). See Table 2.

Gastrointestinal events

Four studies comparing EN versus EN and PN reported data for diarrhoea (Bauer 2000; Casaer 2011; Chiarelli 1996; Fan 2016). We noted substantial statistical heterogeneity between studies (I² = 88%) and did not pool data (Analysis 2.5).

Single studies reported data for vomiting or aspiration (Casaer 2011), gastric reflux (Dunham 1994), and stress ulcer (Fan 2016). See Table 2.

Infective events

Two studies reported pneumonia (aspirated pneumonia in Fan 2016; pneumonia in the ICU in Wischmeyer 2017). It is uncertain whether one feeding regimen rather than another reduced pneumonia because the certainty of this evidence is very low (RR 1.40, 95% CI 0.91 to 2.15; 205 participants; I² = 31%; Analysis 2.6). We used GRADE to downgrade the evidence by three levels; we were concerned by study limitations (one level) and imprecision (two levels). See summary of findings Table 2.

Two studies reported wound infections (Casaer 2011; Wischmeyer 2017); we used data for skin/soft tissue wounds in Wischmeyer 2017. We found little or no difference in events between groups (RR 0.67, 95% CI 0.50 to 0.92; 4765 participants; I² = 46%; Analysis 2.7).

Two studies reported bloodstream infections (Casaer 2011; Wischmeyer 2017); we used data for primary bloodstream infections in Wischmeyer 2017. We found little or no difference in events between groups (RR 0.81, 95% CI 0.66 to 1.01; 4765 participants; I² = 0%; Analysis 2.8).

Three studies reported urinary tract infections (Bauer 2000; Casaer 2011; Wischmeyer 2017); we used data for 'lower urinary tract infections' in Wischmeyer 2017. We found little or no difference in events between groups (RR 0.87, 95% CI 0.65 to 1.17; 4885 participants; I² = 52%; Analysis 2.9).

Three studies reported airway infections (Bauer 2000; Casaer 2011; Wischmeyer 2017); we used data for 'lower respiratory tract infection' in Wischmeyer 2017, and 'respiratory infection' in Bauer 2000. We noted substantial statistical heterogeneity between studies (I² = 78%) and did not pool data (Analysis 2.10).

Single studies reported data for pyaemia and intracranial infection (Fan 2016); and surgical deep infections, catheter bloodstream infections, upper urinary tract infections, and intra‐abdominal infections (Wischmeyer 2017). See Table 2. One study reported number of infections after day nine and reported this as number of events rather than by participant; we did not include these data because we could not be certain whether participants had more than one infection (Heidegger 2013).

Subgroup analysis

1. Early initiation of feeding (less than 48 hours) versus late initiation of feeding (48 hours or greater)

Four studies comparing EN versus EN and PN initiated feeding with 48 hours (Bauer 2000; Dunham 1994; Fan 2016; Wischmeyer 2017). One study comparing EN versus EN and PN initiated a late feeding protocol for the EN group after four days of all participants being given PN (Chiarelli 1996); PN was initiated early and weaning to EN was initiated late. Two studies comparing EN versus EN and PN initiated a late feeding protocol for the PN group after three days of all participants being given EN (Casaer 2011; Heidegger 2013); EN was initiated early and supplemental PN was initiated late. We did not conduct subgroup analysis for this comparison because there were few studies.

2. Normocaloric intake (to match 80% to 100% of energy expenditure) versus hypocaloric intake (less than 70% of energy expenditure)

3. 'Frail elderly' versus other participants

We identified no studies that specified inclusion of frail elderly participants, or subdivided participant characteristics by this description.

4. Gastrointestinal medical or surgical participants versus non‐gastrointestinal medical or surgical participants

Two studies comparing EN versus EN and PN included participants who were only non‐gastrointestinal surgical or medical participants (Abrishami 2010; Heidegger 2013). Two studies included participants with a mix of primary diagnoses which included gastrointestinal medical or surgical conditions (Casaer 2011; Wischmeyer 2017). One study did not report whether participants had gastrointestinal medical or surgical conditions (Fan 2016). We did not conduct a subgroup analysis because there were few studies.

Sensitivity analysis

1. Selection bias

We assessed five studies as having high or unclear risk of sequence generation (Abrishami 2010; Bauer 2000; Chiarelli 1996; Dunham 1994; Fan 2016). We excluded these studies from the analysis and found no difference in interpretation of effect estimates for in‐hospital mortality. It was not feasible to conduct sensitivity analysis for mortality at 30 days and mortality at 90 days because only one study remained.

2. Attrition bias

We judged one study to have unclear risk of attrition bias and performed sensitivity analyses by excluding it from appropriate analyses (Heidegger 2013). There was no difference in effect for mortality in hospital and at 30 days.

3. Effects model

We reanalysed our mortality data using a random‐effects model; this did not change the effect.

Discusión

available in

Resumen de los resultados principales

Se incluyeron 25 estudios que compararon NE versus NP o versus NE y NP administradas a pacientes adultos con enfermedades graves en la UCI. Además, se identificaron nueve estudios en espera de clasificación (tres estudios completados o terminados sin publicación del informe completo, dos estudios publicados solo como resúmenes con información insuficiente, tres estudios en los que no fue posible tener acceso a los informes completos, un estudio que requiere traducción) y dos estudios en curso.

Se encontró evidencia de certeza baja y muy baja que no mostró diferencias entre la NE versus NP en la mortalidad en el hospital, en el transcurso de 30 días, en el transcurso de 90 días ni en el transcurso de 180 días. Ningún estudio informó el número de días sin UCI hasta el día 28. Un estudio informó el número de días sin respirador hasta el día 28 y no hay seguridad con respecto a si una vía de alimentación en lugar de otra alteró el número de días sin respirador debido a que la certeza de la evidencia es muy baja. Se encontró evidencia de certeza baja y muy baja que no mostró diferencias entre la NE versus la combinación de NE y NP en la mortalidad en el hospital, en el transcurso de 90 días y en el transcurso de 180 días. No hay seguridad con respecto a si la combinación de NE y NP reduce la mortalidad a los 30 días debido a que la certeza de la evidencia es muy baja.

Los eventos adversos informados por los estudios fueron: mecánicos (aspiración, neumotórax, hemorragia nasal, punción de la arteria subclavia, obstrucción de la sonda o la línea, malfuncionamiento de la línea, fracaso en la intubación); metabólicos (hiperglucemia, hipoproteinemia y trastorno de electrólitos); gastrointestinales (diarrea, vómitos, distensión abdominal, náuseas, timpanismo abdominal o retortijones, ictericia, úlcera de estrés, enzimas hepáticas elevadas y reflujo gástrico); e infecciosos (sepsis, neumonía, infecciones del catéter, infección pulmonar, infección intracraneal, infecciones primarias de la sangre, infecciones de la herida, infección intraabdominal, infecciones de las vías urinarias, infecciones quirúrgicas, infección de las vías respiratorias, piemia, empiema y sepsis de la línea).

Se encontró evidencia de certeza baja y muy baja que no mostró diferencias entre la NE versus NP en los participantes con aspiración o neumonía. Se encontró que la NE puede reducir la sepsis (evidencia de certeza baja) y no hay seguridad con respecto a si la NP reduce los vómitos debido a que la certeza de la evidencia es muy baja. Además, no se encontró evidencia de una diferencia entre NE versus NP en: la incidencia de neumotórax, la distensión abdominal, las infecciones de la herida y las infecciones de las vías urinarias. Se encontró que menos pacientes que recibieron NE presentaron hiperglucemia y presentaron infecciones intraabdominales y que menos pacientes que recibieron NP presentaron diarrea. No se utilizó GRADE para evaluar la certeza de la evidencia en estos eventos adversos adicionales y se observó que la evidencia provenía de pocos estudios.

No hay seguridad con respecto a si la combinación de NE y NP en comparación con NP reduce la neumonía debido a que la certeza de la evidencia es muy baja. Además, se encontró poca o ninguna diferencia entre la NE versus la combinación de NE y NP en los participantes con infecciones de la herida, infecciones de la sangre, infecciones de las vías urinarias u oclusión de la sonda de alimentación.

Compleción y aplicabilidad general de las pruebas

Se identificaron 25 estudios que incluyeron a 8816 participantes que habían ingresado a la UCI con una variedad amplia de diagnósticos. Aunque se observó heterogeneidad estadística limitada en la mayoría de los análisis de resultados de la revisión, es posible que el rango de diagnósticos primarios pueda haber introducido heterogeneidad y reducido la aplicabilidad de estos hallazgos y se utilizó la evaluación GRADE para reducir la certeza en las estimaciones del efecto. A pesar del número de estudios incluidos, no fue posible realizar análisis de subgrupos en algunos de los subgrupos propuestos, lo que limitó la exploración de las diferencias entre los estudios incluidos. También se señaló que los estudios variaron en la fecha de publicación desde 1983 a 2017 y, aunque no se evaluó la posible influencia de la fecha en los resultados, es posible que los cambios en el tratamiento de los pacientes en la UCI puedan significar que algunos datos de los estudios puedan no ser generalizables al ámbito de la UCI actual.

Calidad de la evidencia

Nutrición enteral versus nutrición parenteral

Se señaló que todo el personal conocía el tipo de régimen de alimentación de cada grupo de participantes y todos los resultados; se consideró que lo anterior introdujo un alto riesgo de sesgo de realización. Mediante el enfoque GRADE, la certeza de la evidencia para la mortalidad en cada punto temporal, para el número de días sin respirador hasta el día 28 y cada evento adverso (aspiración, sepsis, neumonía y vómitos) se disminuyó en un nivel debido a las limitaciones de los estudios.

Los estudios incluyeron a participantes con diagnósticos primarios variados (p.ej., pacientes con trastornos médicos o quirúrgicos gastrointestinales o no gastrointestinales, y si los participantes recibían ventilación mecánica). Se cree que lo anterior redujo la posibilidad de generalizar la evidencia para algunos resultados; fue posible que los participantes con algunos diagnósticos pudieran haber respondido de forma diferente a cada alimentación. Mediante el enfoque GRADE, la certeza de la evidencia para la mortalidad hospitalaria, la mortalidad en el transcurso de 30 días y 90 días y los eventos adversos se disminuyó en un nivel debido a la imposibilidad de generalizar la evidencia.

Se observó que la mayoría de los estudios incluyó un escaso número de participantes y dos estudios incluyeron tamaños de la muestra grandes (Casaer 2011; Harvey 2014); estos estudios introdujeron una ponderación más grande en las estimaciones del efecto a través de algunos análisis, que fue particularmente notorio en los análisis de la aspiración y los vómitos en los que solo dos estudios informaron datos para varios resultados de los eventos adversos. Se consideró el efecto de estos estudios grandes sobre los resultados y, mediante el enfoque GRADE, la certeza de la evidencia de la aspiración y los vómitos se disminuyó en un nivel debido a la imprecisión. Para la mortalidad a los 180 días, se encontró solamente un estudio pequeño y se consideró que este resultado solo proporcionó un efecto impreciso, por lo que la certeza de la evidencia para este resultado se disminuyó en un nivel debido a la imprecisión.

Nutrición enteral versus nutrición enteral y nutrición parenteral

Se observó que todo el personal conocía el tipo de régimen de alimentación de cada grupo de participantes y todos los resultados, por lo que se consideró que hubo un alto riesgo de sesgo de realización. Mediante el enfoque GRADE, la certeza de la evidencia para la mortalidad (en cada punto temporal) y para la neumonía se disminuyó en un nivel debido a las limitaciones de los estudios.

Los estudios incluyeron a participantes con diagnósticos primarios variados (p.ej., pacientes con trastornos médicos o quirúrgicos gastrointestinales o no gastrointestinales, y si los participantes recibían ventilación mecánica). Se cree que lo anterior redujo la posibilidad de generalizar la evidencia para algunos resultados; es posible que los participantes con algunos diagnósticos puedan haber respondido de forma diferente a cada alimentación. Mediante el enfoque GRADE, la certeza de la evidencia para la mortalidad hospitalaria y la mortalidad en el transcurso de 30 días y 90 días, se disminuyó en un nivel debido a la imposibilidad de generalizar la evidencia.

Para la mortalidad a los 180 días se encontró solamente un estudio pequeño y para la neumonía se encontraron dos estudios pequeños. Se consideró que estos resultados solo proporcionaron un efecto impreciso y la certeza de la evidencia para estos resultados se disminuyó en un nivel debido a la imprecisión.

Sesgos potenciales en el proceso de revisión

Se realizó una búsqueda minuciosa y dos autores de la revisión evaluaron la elegibilidad de los estudios, extrajeron los datos y evaluaron el riesgo de sesgo de los estudios incluidos y, por lo tanto, se redujo el sesgo potencial en el proceso de revisión. Sin embargo, algunas decisiones con respecto a la elegibilidad se tomaron sobre la base de la información presentada solamente en los informes de los estudios y no se estableció contacto con los autores de los estudios para obtener aclaraciones; Se excluyeron algunos estudios que no establecieron con claridad que los participantes estaban en la UCI. Se respetó la decisión del protocolo de incluir los datos de resultado informados como número de días sin UCI y número de días sin respirador, hasta el día 28 (Lewis 2016). Muchos de los estudios incluidos habían informado la duración de la estancia hospitalaria en la UCI o la duración de la ventilación mecánica y estos datos de resultado no se incluyeron en la revisión.

Acuerdos y desacuerdos con otros estudios o revisiones

Se observó que revisiones realizadas por otros autores de la revisión utilizaron criterios diferentes para decidir si los participantes presentaban enfermedades graves y, por lo tanto, estas revisiones no incluyeron los mismos estudios (Elke 2016; Simpson 2005). Simpson 2005 informó una reducción en la mortalidad con la NP; sin embargo, estos datos contradijeron los de la revisión más reciente de Elke 2016 cuyos resultados fueron consistentes con los resultados de la presente revisión de que no hay efectos sobre la mortalidad cuando se comparan la NE versus la NP.

No se encontró evidencia de una diferencia en el número de participantes con eventos adversos. Las revisiones de Elke 2016 y Simpson 2005 informaron una reducción en las complicaciones infecciosas al administrar NP. Aunque estas revisiones incluyeron algunos estudios diferentes a los de la presente revisión, los autores de la revisión presentaron datos compuestos del número de infecciones como los informaron los autores de los estudios. En la presente revisión los datos de las infecciones se informaron mediante el número de participantes con tipos particulares de infección, en lugar de una cifra compuesta y, por lo tanto, la revisión difirió en el tipo de datos informados para las infecciones.

Figure 1

Flow diagram of search strategy.

Figure 2

Figure 3

Risk of bias summary: review authors' judgements about each risk of bias item for each included study. Blank spaces in tables indicate that study authors did not report the review outcome.

Analysis 1.1

Comparison 1 Enteral (EN) versus parenteral nutrition (PN), Outcome 1 In‐hospital mortality.

Analysis 1.2

Comparison 1 Enteral (EN) versus parenteral nutrition (PN), Outcome 2 Mortality at 30 days.

Analysis 1.3

Comparison 1 Enteral (EN) versus parenteral nutrition (PN), Outcome 3 Mortality at 90 days.

Analysis 1.4

Comparison 1 Enteral (EN) versus parenteral nutrition (PN), Outcome 4 Aspiration.

Analysis 1.5

Comparison 1 Enteral (EN) versus parenteral nutrition (PN), Outcome 5 Pneumothorax.

Analysis 1.6

Comparison 1 Enteral (EN) versus parenteral nutrition (PN), Outcome 6 Hyperglycaemia.

Analysis 1.7

Comparison 1 Enteral (EN) versus parenteral nutrition (PN), Outcome 7 Vomiting.

Analysis 1.8

Comparison 1 Enteral (EN) versus parenteral nutrition (PN), Outcome 8 Diarrhoea.

Analysis 1.9

Comparison 1 Enteral (EN) versus parenteral nutrition (PN), Outcome 9 Abdominal distension.

Analysis 1.10

Comparison 1 Enteral (EN) versus parenteral nutrition (PN), Outcome 10 Sepsis.

Analysis 1.11

Comparison 1 Enteral (EN) versus parenteral nutrition (PN), Outcome 11 Pneumonia.

Analysis 1.12

Comparison 1 Enteral (EN) versus parenteral nutrition (PN), Outcome 12 Intra‐abdominal infection.

Analysis 1.13

Comparison 1 Enteral (EN) versus parenteral nutrition (PN), Outcome 13 Wound infection.

Analysis 1.14

Comparison 1 Enteral (EN) versus parenteral nutrition (PN), Outcome 14 Urinary tract infection.

Analysis 1.15

Comparison 1 Enteral (EN) versus parenteral nutrition (PN), Outcome 15 In‐hospital mortality: gastrointestinal (GI) medical/surgical vs non‐GI medical/surgical.

Analysis 1.16

Comparison 1 Enteral (EN) versus parenteral nutrition (PN), Outcome 16 Mortality at 30 days: GI medical/surgical vs non‐GI medical/surgical.

Analysis 2.1

Comparison 2 Enteral (EN) versus combined EN and parenteral nutrition (PN), Outcome 1 In‐hospital mortality.

Analysis 2.2

Comparison 2 Enteral (EN) versus combined EN and parenteral nutrition (PN), Outcome 2 Mortality at 30 days.

Analysis 2.3

Comparison 2 Enteral (EN) versus combined EN and parenteral nutrition (PN), Outcome 3 Mortality at 90 days.

Analysis 2.4

Comparison 2 Enteral (EN) versus combined EN and parenteral nutrition (PN), Outcome 4 Feeding tube obstruction.

Analysis 2.5

Comparison 2 Enteral (EN) versus combined EN and parenteral nutrition (PN), Outcome 5 Diarrhoea.

Analysis 2.6

Comparison 2 Enteral (EN) versus combined EN and parenteral nutrition (PN), Outcome 6 Pneumonia.

Analysis 2.7

Comparison 2 Enteral (EN) versus combined EN and parenteral nutrition (PN), Outcome 7 Wound infection.

Analysis 2.8

Comparison 2 Enteral (EN) versus combined EN and parenteral nutrition (PN), Outcome 8 Bloodstream infection.

Analysis 2.9

Comparison 2 Enteral (EN) versus combined EN and parenteral nutrition (PN), Outcome 9 Urinary tract infection.

Analysis 2.10

Comparison 2 Enteral (EN) versus combined EN and parenteral nutrition (PN), Outcome 10 Airway infection.

Summary of findings for the main comparison. Enteral versus parenteral nutrition for adults in the intensive care unit

Enteral versus parenteral nutrition for adults in the intensive care unit
Patient or population: critically ill adults admitted to the ICU for trauma, emergency, or surgical care; population excluded people with acute pancreatitis Setting: intensive care units in: Brazil, China, Germany, Iran, Italy, Turkey, UK, and USA Intervention: EN Comparison: PN
Outcomes	*Anticipated absolute effects (95% CI)**		Relative effect (95% CI)	Number of participants (studies)	Certainty of the evidence (GRADE)
	Risk with EN	Risk with PN
Mortality	In‐hospital mortality		RR 1.19 (0.80 to 1.77)	361 (6 studies)	⊕⊕⊝⊝ Low^a
	Study population
	229 per 1000 (154 to 340)	192 per 1000
	Mortality within 30 days		RR 1.02 (0.92 to 1.13)	3148 (11 studies)	⊕⊕⊝⊝ Low^b
	Study population
	304 per 1000 (274 to 336)	298 per 1000
	Mortality within 90 days		RR 1.06 (0.95 to 1.17)	2461 (3 studies)	⊕⊝⊝⊝ Very low^c
	Study population
	393 per 1000 (352 to 434)	371 per 1000
	Mortality within 180 days		RR 0.33 (0.04 to 2.97)	46 (1 study)	⊕⊝⊝⊝ Very low^d
	Study population
	130 per 1000	43 per 1000 (5 in 387)
Number of ICU‐free days up to day 28	–	–	–	–	Not measured
Number of ventilator‐free days up to day 28	Mean number of ventilator‐free days: 14.2 (SD ± 12.2)	Mean difference 0 days (0.97 fewer to 0.97 more)	N/A	2388 (1 study)	⊕⊝⊝⊝ Very low^d
Adverse events: aspiration (as reported by study authors at end of study follow‐up period)	Study population		RR 1.53 (0.46 to 5.03)	2437 (2 studies)	⊕⊝⊝⊝ Very low^e
	5 per 1000 (2 to 17)	3 per 1000
Adverse events: sepsis (as reported by study authors at end of study follow‐up period)	Study population		RR 0.59 (0.37 to 0.95)	361 (7 studies)	⊕⊕⊝⊝ Low^f
	123 per 1000 (77 to 199)	209 per 1000
Adverse events: pneumonia (as reported by study authors at end of study follow‐up period)	Study population		RR 1.10 (0.82 to 1.48)	415 (7 studies)	⊕⊕⊝⊝ Low^f
	314 per 1000 (234 to 423)	268 per 1000
Adverse events: vomiting (as reported by study authors at end of study follow‐up period)	Study population		RR 3.42 (1.15 to 10.16)	2525 (3 studies)	⊕⊝⊝⊝ Very low^g
	11 per 1000 (4 to 32)	3 per 1000
*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; EN: enteral nutrition; ICU: intensive care unit; N/A: not applicable; PN: parenteral nutrition; RR: risk ratio; SD: standard deviation.
GRADE Working Group grades of evidence High: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate: we are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low: our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect. Very low: we have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.
^aAll studies had a high risk of performance bias; downgraded one level for study limitations. Studies included a variety of primary diagnoses and evidence was less direct; downgraded one level for indirectness. ^bAll studies had a high risk of performance bias; downgraded one level for study limitations. Studies included a variety of primary diagnoses and study designs and evidence were less direct; downgraded one level for indirectness. ^cAll studies had a high risk of performance bias; downgraded one level for study limitations. Studies included a variety of primary diagnoses and study designs and evidence were less direct; downgraded one level for indirectness. Few studies and one included study had a large number of participants relative to other included studies; downgraded one level for imprecision. ^dData from only one study that had a high risk of performance bias; downgraded one level for study limitations and two levels for imprecision. ^eAll studies had a high risk of performance bias; downgraded one level for study limitations. Studies included a variety of primary diagnoses and evidence was less direct; downgraded one level for indirectness. Few studies and one included study had a large number of participants relative to other included studies; downgraded one level for imprecision. ^fAll studies had a high risk of performance bias; downgraded one level for study limitations. Studies included a variety of primary diagnoses and evidence was less direct; downgraded one level for indirectness. ^gAll studies had a high risk of performance bias; downgraded one level for study limitations. Studies included a variety of primary diagnoses and study designs and evidence were less direct; downgraded one level for indirectness. Few studies, with very few events, and one included study had a large number of participants relative to other included studies; downgraded one level for imprecision.

Summary of findings for the main comparison. Enteral versus parenteral nutrition for adults in the intensive care unit

Summary of findings 2. Enteral versus enteral and parenteral nutrition for adults in the intensive care unit

Enteral versus enteral and parenteral nutrition for adults in the intensive care unit
Patient or population: critically ill adults admitted to the ICU for trauma, emergency, or post‐surgical care; population excludes participants with acute pancreatitis Setting: intensive care units in: France, Italy, Switzerland, Turkey, and USA Intervention: EN Comparison: EN + PN
Outcomes	*Anticipated absolute effects (95% CI)**		Relative effect (95% CI)	Number of participants (studies)	Certainty of the evidence (GRADE)
	Risk with EN	Risk with EN + PN
Mortality	In‐hospital mortality		RR 0.99 (0.84 to 1.16)	5111 (5 studies)	⊕⊕⊝⊝ Low^a
	Study population
	106 per 1000 (90 to 124)	107 per 1000
	Mortality within 30 days		RR 1.64 (1.06 to 2.54)	409 (3 studies)	⊕⊝⊝⊝ Very low^b
	Study population
	216 per 1000 (140 to 335)	132 per 1000
	Mortality within 90 days		RR 1.00 (0.86 to 1.18)	4760 (2 studies)	⊕⊕⊝⊝ Low^c
	Study population
	115 per 1000 (99 to 135)	115 per 1000
	Mortality within 180 days		RR 1.00 (0.65 to 1.55)	120 (1 RCT)	⊕⊝⊝⊝ Very low^d
	Study population
	400 per 1000 (260 to 620)	400 per 1000
Number of ICU‐free days up to day 28	–	–	–	–	Not measured
Number of ventilator‐free days up to day 28	–	–	–	–	Not measured
Adverse events: aspiration (as reported by study authors at end of study follow‐up period)	–	–	–	–	Not measured
Adverse events: sepsis (as reported by study authors at end of study follow‐up period)	–	–	–	–	Not measured
Adverse events: pneumonia (as reported by study authors at end of study follow‐up period)	350 per 1000 (228 to 538)	250 per 1000	RR 1.40 (0.91 to 2.15)	205 (2 studies)	⊕⊝⊝⊝ Very low^d
Adverse events: vomiting (as reported by study authors at end of study follow‐up period)	–	–	–	–	Not measured
*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; EN: enteral nutrition; ICU: intensive care unit; PN: parenteral nutrition; RCT: randomized controlled trial; RR: risk ratio.
GRADE Working Group grades of evidence High: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate: we are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low: our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect. Very low: we have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.
^aAll studies had high risk of performance bias; downgraded one level for study limitations. Studies included a variety of primary diagnoses and evidence was less direct; downgraded one level for indirectness. ^bAll studies had high risk of performance bias; downgraded one level for study limitations. Studies included a variety of primary diagnoses and evidence was less direct; downgraded one level for indirectness. Few studies with increased risk of imprecision; downgraded one level. ^cBoth studies had high risk of performance bias; downgraded one level for study limitations. Studies included a variety of primary diagnoses and evidence was less direct; downgraded one level for indirectness. ^dData from only one study that had a high risk of performance bias; downgraded one level for study limitations and two levels for imprecision.

Summary of findings 2. Enteral versus enteral and parenteral nutrition for adults in the intensive care unit

Table 1. Adverse events for single studies: enteral nutrition versus parenteral nutrition

Study ID	Description of event	EN group (n/N)	PN group (n/N)
Mechanical events
Adams 1986	Clogged jejunostomy tube	9/23	N/A
	Disconnected line	N/A	1/23
	Line eroded into right upper lobe bronchus	N/A	1/23
	Malfunctioned line	N/A	7/23
Dunham 1994	Transpyloric tube occlusion	2/12	0/15
	Failure to intubate	0/12	0/15
	Withdrawal of tube by participant	1/12	N/A
Metabolic events
Adams 1986	Hepatic failure	1/23	1/23
	Acute renal failure	1/23	1/23
	Pancreatitis	2/23	1/23
Fan 2016	Hypoproteinaemia	22/40	32/40
Harvey 2014	Electrolyte disturbance	5/1197	8/1191
Gastrointestinal events
Adams 1986	Nausea, cramps, bloating	19/23	16/23
Adams 1986	Gastrointestinal bleeding	0/23	0/23
Dunham 1994	Gastric reflux	0/12	0/15
	Ileus	1/12	0/15
	Small bowel ileus	0/12	1/15
Fan 2016	Stress ulcer	7/40	19/40
Harvey 2014	Elevated liver enzymes	7/1197	3/1191
	Jaundice	1/1197	1/1191
	Ischaemic bowel	0/1197	1/1191
Xi 2014	Anastomotic leak	2/22	6/23
Infective events
Adams 1986	Persistent fever without obvious cause	1/23	5/23
Altintas 2011	Catheter infection	2/30	4/41
Borzotta 1994	Meningitis	2/28	0/21
	Sinusitis	3/28	6/21
	Bronchitis	6/28	6/28
	Clostridium difficile	2/28	4/21
	Peritonitis	0/28	1/21
Fan 2016	Intracranial infection	7/40	13/40
Fan 2016	Pyaemia	3/40	19/40
Gencer 2010	Pulmonary infection	2/30	2/30
Kudsk 1992	Empyema	1/51	4/45
Young 1987	Aspiration pneumonia	9/28	3/23
Young 1987	Infection (type of infection not described)	5/28	4/23
EN: enteral nutrition; n: number of participants with an event; N: total number randomized to group; N/A: not applicable; PN: parenteral nutrition.

Table 1. Adverse events for single studies: enteral nutrition versus parenteral nutrition

Table 2. Adverse events for single studies: enteral nutrition versus enteral nutrition and parenteral nutrition

Study ID	Description of event	EN group (n/N)	EN + PN group (n/N)
Mechanical events
Casaer 2011	CVC obstruction	9/2328	15/2312
	Nasal bleeding	18/2328	14/2312
	Pneumohaemothorax after CVC placement	0/2328	2/2312
	Subclavian artery puncture	0/2328	2/2312
Dunham 1994	Withdrawal of tube	1/12	0/10
Dunham 1994	Failure to intubate	0/12	2/10
Metabolic events
Fan 2016	Hypoproteinaemia	22/40	7/40
Gastrointestinal events
Casaer 2011	Vomiting or aspiration	284/2328	295/2312
Dunham 1994	Gastric reflux	0/12	2/10
Fan 2016	Stress ulcer	7/40	9/40
infective events
Fan 2016	Pyemia	3/40	10/40
Fan 2016	Intracranial infection	7/40	5/40
Wischmeyer 2017	Catheter bloodstream infection	0/73	7/52
	Intra‐abdominal infection	0/73	4/52
	Upper urinary tract infection	0/73	1/52
	Surgical deep infection	0/73	1/52
CVC: central venous catheter; EN: enteral nutrition; EN + PN: combined enteral and parenteral nutrition; n: number of participants with an event; N: total number randomized to group.

Table 2. Adverse events for single studies: enteral nutrition versus enteral nutrition and parenteral nutrition

Comparison 1. Enteral (EN) versus parenteral nutrition (PN)

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 In‐hospital mortality Show forest plot	6	361	Risk Ratio (M‐H, Fixed, 95% CI)	1.19 [0.80, 1.77]

2 Mortality at 30 days Show forest plot	11	3148	Risk Ratio (M‐H, Fixed, 95% CI)	1.02 [0.92, 1.13]

3 Mortality at 90 days Show forest plot	3	2461	Risk Ratio (M‐H, Fixed, 95% CI)	1.06 [0.95, 1.17]

4 Aspiration Show forest plot	2	2437	Risk Ratio (M‐H, Fixed, 95% CI)	1.53 [0.46, 5.03]

5 Pneumothorax Show forest plot	2	2437	Risk Ratio (M‐H, Fixed, 95% CI)	1.46 [0.19, 11.22]

6 Hyperglycaemia Show forest plot	2	2437	Risk Ratio (M‐H, Fixed, 95% CI)	0.57 [0.35, 0.93]

7 Vomiting Show forest plot	3	2525	Risk Ratio (M‐H, Fixed, 95% CI)	3.42 [1.15, 10.16]

8 Diarrhoea Show forest plot	6	363	Risk Ratio (M‐H, Fixed, 95% CI)	2.17 [1.72, 2.75]

9 Abdominal distension Show forest plot	3	2505	Risk Ratio (M‐H, Fixed, 95% CI)	1.53 [0.34, 6.96]

10 Sepsis Show forest plot	7	361	Risk Ratio (M‐H, Fixed, 95% CI)	0.59 [0.37, 0.95]

11 Pneumonia Show forest plot	7	415	Risk Ratio (M‐H, Fixed, 95% CI)	1.10 [0.82, 1.48]

12 Intra‐abdominal infection Show forest plot	3	202	Risk Ratio (M‐H, Fixed, 95% CI)	0.26 [0.07, 0.89]

13 Wound infection Show forest plot	3	155	Risk Ratio (M‐H, Fixed, 95% CI)	1.45 [0.55, 3.82]

14 Urinary tract infection Show forest plot	3	160	Risk Ratio (M‐H, Fixed, 95% CI)	1.48 [0.65, 3.40]

15 In‐hospital mortality: gastrointestinal (GI) medical/surgical vs non‐GI medical/surgical Show forest plot	6	361	Risk Ratio (M‐H, Fixed, 95% CI)	1.19 [0.80, 1.77]

15.1 GI medical/surgical	1	98	Risk Ratio (M‐H, Fixed, 95% CI)	0.88 [0.06, 13.74]
15.2 Non‐GI medical/surgical	5	263	Risk Ratio (M‐H, Fixed, 95% CI)	1.20 [0.80, 1.79]
16 Mortality at 30 days: GI medical/surgical vs non‐GI medical/surgical Show forest plot	10	3068	Risk Ratio (M‐H, Fixed, 95% CI)	1.03 [0.93, 1.14]

16.1 GI medical/surgical	1	60	Risk Ratio (M‐H, Fixed, 95% CI)	0.67 [0.12, 3.71]
16.2 Non‐GI medical/surgical	9	3008	Risk Ratio (M‐H, Fixed, 95% CI)	1.03 [0.93, 1.15]

Comparison 1. Enteral (EN) versus parenteral nutrition (PN)

Comparison 2. Enteral (EN) versus combined EN and parenteral nutrition (PN)

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 In‐hospital mortality Show forest plot	5	5111	Risk Ratio (M‐H, Fixed, 95% CI)	0.99 [0.84, 1.16]

2 Mortality at 30 days Show forest plot	3	409	Risk Ratio (M‐H, Fixed, 95% CI)	1.64 [1.06, 2.54]

3 Mortality at 90 days Show forest plot	2	4760	Risk Ratio (M‐H, Fixed, 95% CI)	1.00 [0.86, 1.18]

4 Feeding tube obstruction Show forest plot	2	4662	Risk Ratio (M‐H, Fixed, 95% CI)	0.96 [0.70, 1.32]

5 Diarrhoea Show forest plot	4		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected

6 Pneumonia Show forest plot	2	205	Risk Ratio (M‐H, Fixed, 95% CI)	1.40 [0.91, 2.15]

7 Wound infection Show forest plot	2	4765	Risk Ratio (M‐H, Fixed, 95% CI)	0.67 [0.50, 0.92]

8 Bloodstream infection Show forest plot	2	4765	Risk Ratio (M‐H, Fixed, 95% CI)	0.81 [0.66, 1.01]

9 Urinary tract infection Show forest plot	3	4885	Risk Ratio (M‐H, Fixed, 95% CI)	0.87 [0.65, 1.17]

10 Airway infection Show forest plot	3		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected

Comparison 2. Enteral (EN) versus combined EN and parenteral nutrition (PN)