Outcomes of gastrostomy in children with neurological impairment: a systematic review

1 Introduction

Children with severe neurological impairment carry a high burden of oropharyngeal dysphagia, and that feeding difficulty has consequences on their growth, organ development, and family life (1). Conditions such as cerebral palsy and hypoxic-ischemic encephalopathy (HIE) disrupt the coordinated swallow, so oral intake becomes slow, unsafe or insufficient. A frequent consequence is malnutrition, where roughly one-third of neurologically impaired children referred for enteral support are already malnourished at presentation (2). Disruption of coordinated swallow can also lead to aspiration, a leading cause of death in children with co-existing gastro-esophageal reflux disease (GERD) (3). The nutritional deficits extend beyond physiological development, as childhood malnutrition impairs neurodevelopment, cognition and academic achievement as shown in cohorts followed into adulthood (4, 5). For a child whose brain is already injured, undernutrition therefore threatens whatever developmental reserve remains.

Gastrostomy feeding answers the nutritional problem directly, and the evidence for that narrow endpoint is consistent. Weight-for-age and BMI z-scores rise in the months after tube placement across single-centre and multicentre cohorts, regardless of the underlying diagnosis (2, 6, 7). Timing of gastrostomy placement impacts this growth response. Children who receive a gastrostomy earlier, before prolonged undernutrition, show faster catch-up and are more likely to normalise their nutritional status (6, 8). Additional benefits reported by caregivers describe easier and shorter feeds, better general health, and improved family life after placement, with satisfaction near 90% in population-based follow-up (9, 10).

Respiratory outcomes remain less settled. One prospective cohort found that chest infections requiring antibiotics and hospital admissions for chest infection both fell after gastrostomy, arguing against the fear that tube feeding worsens aspiration (11). Conversely, a large propensity-matched analysis linked gastrostomy placement to higher subsequent pneumonia hospitalisation, a signal the authors attributed to confounding by indication rather than to harm (12). Although serious complications are uncommon, placement itself carries a risk of granulation tissue, stomal infection, tube displacement, and occasional revision (10, 13). Health-related quality of life for the child is harder to interpret, because neurological impairment predicts lower scores whatever the feeding route (14).

Despite this body of observational work, the evidence base lacks methodological quality and breadth. Sleigh and Brocklehurst's systematic review found no randomized controlled trials and no controlled cohorts beyond a single comparison, relying instead on case series with risk of bias, and concluded that the true magnitude of both benefit and surgical risk could not be quantified (15). The subsequent Cochrane update identified no studies meeting inclusion criteria at all, concluding that considerable uncertainty about the effects of gastrostomy in children with cerebral palsy remains and that no reliable evidence base exists to guide families through this decision (16). Both reviews closed more than a decade ago, and both were restricted to cerebral palsy. Controlled comparisons have appeared since, but they contrast feeding routes over short follow-up windows, and the most recent Cochrane review examined how reflux should be managed alongside gastrostomy rather than what the operation does to a child's development. Nutritional and caregiver-reported outcomes have been examined repeatedly, but neurodevelopmental outcomes such as cognition, gross motor function, communication, and school participation have not been synthesised (4, 5, 17). This systematic review explores whether gastrostomy tube placement, compared with continued oral or nasogastric feeding, improves neurodevelopmental outcomes, nutritional status, respiratory health, and caregiver quality of life in children with neurological impairment and feeding difficulties.

2 Methods

This systematic review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines. The review protocol was registered with PROSPERO (registration number: CRD420261483110) (18, 19).

2.1 Study design

This is a systematic review of quantitative and mixed-methods primary studies, without meta-analysis. It is reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 statement (18). The review question was framed using the PICOS structure set out in sections 2.2 to 2.5. No meta-analysis was undertaken, for the reasons given in section 2.10.

2.2 Participants

Eligible participants were children and adolescents aged 0 to 18 years with neurological impairment, defined as a static or progressive disorder of the central or peripheral nervous system associated with feeding difficulty. Qualifying diagnoses included all conditions of neurological impairment.

2.3 Interventions

The intervention of interest was placement of a gastrostomy for enteral feeding, by any technique and using any device. This included percutaneous endoscopic gastrostomy, laparoscopic and open surgical gastrostomy, radiologically inserted gastrostomy, low-profile balloon and non-balloon button devices, and gastrojejunal tubes placed through an existing or new gastrostomy tract. Studies in which a fundoplication or other antireflux procedure was performed concurrently with, or subsequent to, gastrostomy were eligible, and the presence, timing and type of any antireflux procedure was extracted for every study so that its contribution to the outcome could be examined separately.

2.4 Comparators

Three comparator structures were eligible: parallel comparison against children continuing oral or nasogastric feeding; within-subject comparison against the child's own pre-gastrostomy status, whether analysed as a simple before-and-after contrast or through a self-controlled or exposure-crossover design; and comparison between gastrostomy techniques, devices, or the presence or absence of a concurrent antireflux procedure. Single-arm descriptive studies without any comparator were eligible only for the complication, mortality and cross-sectional status outcomes, where they contribute rate data, and their contribution to before-and-after outcome domains was classified as not assessable.

2.5 Outcomes

Outcomes were specified across seven domains before data extraction began. The primary outcome domain was neurodevelopmental function, comprising cognition, gross and fine motor function, communication, adaptive behaviour, school participation and seizure control, measured by any standardised or non-standardised instrument. Secondary outcome domains were: nutritional status, including weight-, height-, BMI- and weight-for-height z-scores, anthropometry, catch-up growth, micronutrient status and excess adiposity; respiratory outcomes, including chest infection frequency before and after gastrostomy, respiratory and aspiration-related hospitalisation, and discrete aspiration events; minor complications, comprising peristomal infection, granulation tissue, leakage, dislodgement, obstruction and site pain; major complications, comprising buried bumper syndrome, colonic or other visceral injury, gastrocolic fistula, peritonitis, major haemorrhage, unplanned reoperation and procedure-related death; gastro-oesophageal reflux outcomes, including new or worsened reflux and subsequent fundoplication; and quality of life and caregiver-reported outcomes, including validated child and caregiver instruments, feeding time, medication delivery reliability and caregiver burden. All-cause and cause-specific mortality was extracted for every study.

2.6 Search strategy

The following search strategy was applied across all databases, combining four concept blocks with the Boolean operator AND and terms within each block with OR:

(“cerebral palsy” OR “neurological impairment” OR “neurologic impairment” OR “neurologically impaired” OR “neurodevelopmental disorder” OR “developmental disability” OR “intellectual disability” OR “neurological disability” OR “neurodegenerative” OR “Batten disease” OR “epilepsy” OR “epileptic encephalopathy” OR “developmental encephalopathy” OR “seizure disorder” OR “hypoxic ischemic encephalopathy” OR “HIE” OR “traumatic brain injury” OR “TBI” OR “spinal muscular atrophy” OR “SMA”) AND (“gastrostomy” OR “gastrostomy tube” OR “G-tube” OR “PEG” OR “percutaneous endoscopic gastrostomy” OR “gastrojejunostomy” OR “G-J tube” OR “enteral feeding” OR “tube feeding” OR “gastrostomy placement” OR “gastrostomy insertion” OR “button gastrostomy” OR “low-profile gastrostomy” OR “radiologically inserted gastrostomy” OR “RIG” OR “laparoscopic gastrostomy” OR “surgical gastrostomy” OR “open gastrostomy”) AND (“child” OR “infant” OR “adolescent” OR “pediatric” OR “paediatric” OR “children”) AND (“developmental outcome” OR “cognitive development” OR “motor function” OR “GMFCS” OR “gross motor function” OR “nutritional status” OR “weight gain” OR “growth” OR “malnutrition” OR “anthropometric” OR “BMI” OR “body mass index” OR “height velocity” OR “weight velocity” OR “respiratory” OR “pneumonia” OR “aspiration” OR “pulmonary” OR “hospitalization” OR “quality of life” OR “QOL” OR “HRQoL” OR “health-related quality of life” OR “caregiver burden” OR “family experience” OR “parent satisfaction” OR “gastroesophageal reflux” OR “GERD” OR “vomiting” OR “fundoplication” OR “complication” OR “adverse event” OR “infection” OR “granulation tissue” OR “survival” OR “mortality” OR “nasogastric” OR “NG tube” OR “oral feeding” OR “feeding time” OR “feeding duration” OR “sleep quality” OR “school attendance”).

No language restriction was applied at the search stage; non-English publications were excluded at screening.

2.7 Data sources

The following sources were searched from inception to 15 July 2026: PubMed, Embase, Scopus, CENTRAL and ClinicalTrials.gov. The reference lists of all included studies and of the two previous systematic reviews in this field were hand-searched for additional eligible records. Conference abstracts were eligible where sufficient methodological and outcome detail could be extracted.

2.8 Study selection & data extraction

Records retrieved from all sources were imported into Covidence (Veritas Health Innovation, Melbourne, Australia), which removed duplicates automatically. Screening was carried out by two independent reviewer teams working in parallel, each team comprising two reviewers (MD and HJ; MS and MA). Each team screened every record independently at title and abstract, and the same two-team structure was then repeated in full for full-text screening, so that every record and every full text was assessed twice. Conflicts arising at either stage, whether within a team or between the two teams, were resolved by the lead reviewer (MKI). No automation tool was used to make inclusion decisions.

Data were extracted onto a standardised extraction form developed for this review and piloted on a sample of studies before full extraction. The form captured bibliographic and design characteristics (author, year, country, design, single- or multi-centre, study period, follow-up duration, comparator); population characteristics (total enrolled, number with neurological impairment, primary neurological diagnosis and its distribution, proportion with cerebral palsy, age at gastrostomy, GMFCS level, baseline feeding method, aspiration history); intervention characteristics (gastrostomy type, presence and timing of any concurrent fundoplication); and every outcome listed under Outcomes, together with a free-text field for extractor notes and ambiguities and a field flagging suspected cohort overlap. Extraction was distributed across five reviewers, and the completed forms were checked centrally by the lead reviewer for internal consistency, unit errors and misassigned outcomes before synthesis.

Where the same cohort appeared in more than one publication, the reports were linked and counted once for any given outcome, with the most complete or most recent report used as the primary source and the companion publications identified in the footnotes of the relevant outcome table. Cohort overlap was identified from institution, recruitment period, sample size and author overlap.

Studies of mixed populations were eligible where children with neurological impairment formed the majority of the cohort, or where outcomes for the neurologically impaired subgroup were separable from those of other participants. Where a study reported only whole-cohort results for a mixed population, this is stated explicitly alongside the finding and the result was not treated as neurology-specific.

2.9 Risk of bias assessment

Risk of bias was assessed with the tool matched to each study design. The Cochrane Risk of Bias 2 (RoB 2) tool was applied to the randomised trial (20). The Risk Of Bias In Non-randomised Studies of Interventions (ROBINS-I) tool was applied to non-randomised comparative designs, including retrospective and prospective cohorts and chart reviews (21). The Joanna Briggs Institute (JBI) critical appraisal checklists were applied to cross-sectional studies, surveys and questionnaire-based research (22). The Critical Appraisal Skills Programme (CASP) qualitative checklist was applied to the qualitative interview studies (23). Assessments were performed independently in duplicate, with disagreements resolved by the external reviewer. Reporting bias across studies was not formally assessed with funnel plots or statistical tests, since these require pooled effect estimates; it is addressed narratively.

2.10 Data analysis

Clinical and methodological heterogeneity across the included studies was substantial. The studies differed in neurological diagnosis, impairment severity, age at placement, device and surgical technique, presence of a concurrent antireflux procedure, follow-up duration, and, most consequentially, in outcome definition: chest infection, aspiration, and minor complication were each operationalised differently across studies, and the same nutritional outcome was reported variously as an absolute weight, a percentage change, a z-score at a single time point, or a change in z-score. Meta-analysis was therefore not appropriate, and no pooled effect estimate is presented. This decision was taken after data extraction, once the extent of outcome-definition heterogeneity became apparent.

Findings were synthesised narratively within each outcome domain, using structured direction-of-effect vote counting. Each study contributing to a domain was assigned to one of five categories on the basis of the direction of the reported effect, irrespective of its magnitude or statistical significance: favourable, where the outcome improved after gastrostomy or was better than in the comparator group; unfavourable, where it deteriorated or was worse than in the comparator; no difference, where the study reported no significant change or between-group difference; mixed, where the study reported effects in opposing directions across measures or subgroups; and not assessable, where the outcome was mentioned but no baseline, denominator or numerator sufficient to judge direction was reported. Vote counts are presented as frequencies and are not interpreted as effect estimates, since a count of studies does not weight for sample size, study quality or precision. This approach follows Cochrane guidance on vote counting by direction of effect and is reported in line with the Synthesis Without Meta-analysis (SWiM) guideline. Studies were additionally classified according to whether their authors were, on the evidence and conclusions they presented, in favour of gastrostomy, against it, or mixed or neutral.

Within each domain, findings were then interpreted with explicit attention to whether the direction of effect tracked study design, on the reasoning that agreement between small uncontrolled cohorts and large comparator-controlled datasets is informative and disagreement between them is more informative still. Results are reported for each domain in the same order: the study-level table, the vote count, the highest-quality and largest contributing studies, the discordant studies, and a one-sentence synthesis. Subgroup observations were made where the included studies themselves reported them, specifically for age at placement, presence of a concurrent antireflux procedure, and device type; these were not pre-specified as formal subgroup analyses. No sensitivity analysis was possible in the absence of pooling.

Because of the extreme clinical and methodological heterogeneity of the observational data, including inconsistent outcome definitions, reliance on retrospective caregiver recall, and pervasive confounding by indication, a formal GRADE assessment of the certainty of evidence was deemed infeasible. Consequently, all clinical conclusions are presented as observational associations rather than established causal effects.

3 Results

3.1 PRISMA flow diagram

3.1.1 Study selection & characteristics

The characteristics of all included studies are presented in Table 1. Data were extracted from 120 records.

Table 1. Characteristics of included studies.

Several included records derive from the same or an overlapping source population but are retained as separate rows because each reports different outcome domains for this review; they are not treated as independent samples in any synthesis. Studies drawn from the same cohort: (1) Kawahara 2010 and Kawahara 2014; (2) Sullivan 2004, Sullivan 2005, and Sullivan 2006 (Oxford Feeding Study); (3) the Clayton 2024 medical-record cohort (n = 124) and the Clayton 2024 caregiver-survey cohort (n = 65), which may include overlapping caregivers and are therefore not combined; and (4) Downs 2019, Wong 2019, Wong 2020, and Jacoby 2020, which draw on the same Western Australian population-linked data collections.

CP, cerebral palsy; GERD, gastro-oesophageal reflux disease; NI, neurological impairment; PEG, percutaneous endoscopic gastrostomy.

The evidence base is overwhelmingly observational. One randomised controlled trial was identified, and this trial is labelled by its own journal as an observational study despite describing random-number-table allocation, so its design classification is uncertain. The remaining studies comprised retrospective cohorts and chart reviews, prospective cohorts, retrospective case series, cross-sectional and registry-based studies, caregiver surveys, and three qualitative interview studies. Most cohorts were single-centre. The most common analytical structure by a wide margin was the single-arm before-and-after comparison, in which children served as their own controls; genuinely parallel comparisons against continued oral or nasogastric feeding were reported in only a small minority of studies, and the largest of these are administrative or population-linked datasets rather than clinical cohorts.

Studies originated from 30 countries, most frequently the United Kingdom (n = 20), the United States (n = 17), Australia (n = 9), Canada (n = 9) and Italy (n = 7), with smaller contributions from across Europe, the Middle East, East and South Asia and Latin America. Publication dates spanned more than four decades, from 1981 to 2026, and the older surgical case series report techniques, devices and perioperative care that differ materially from contemporary practice.

Cohort size ranged from 10 children to more than 24,000, the largest datasets being administrative or registry-linked. Age at gastrostomy varied widely both within and between studies, from the neonatal period to late adolescence, with most cohorts reporting a median age between one and eight years. Follow-up ranged from 28 days to a maximum of 19 years and was under one year in a substantial proportion of studies. Reporting of impairment severity was poor: GMFCS level was recorded in only a minority of cohorts, and many studies grouped participants under an undifferentiated heading of neurological impairment without specifying the underlying diagnosis or its distribution. Cerebral palsy was the single most common diagnosis where diagnosis was reported, but a substantial number of studies did not separate cerebral palsy from other causes of neurological impairment, which limits the precision with which findings can be attributed to any specific population.

3.2 Synthesized findings

Findings are presented by outcome domain. The direction-of-effect vote count for each domain is summarised in Table 2, and study-level data are given in Tables 3–6.

Table 2. Direction-of-effect vote count by outcome domain.

Counts are frequencies of studies and are not effect estimates; they are not weighted for sample size, precision or study quality, and should be read alongside the design-stratified interpretation in Sections 3.3, 4.1. Studies contributing to more than one domain are counted once in each.

Table 3. Nutritional outcomes after gastrostomy.

WAZ, weight-for-age z-score; HAZ, height-for-age z-score; WHZ, weight-for-height z-score; WLZ, weight-for-length z-score; W/L-Z, weight/length z-score; StatZ, stature(-for-age) z-score; BMIZ, BMI z-score; MUAC, mid-upper arm circumference; NS, not significant; GT, gastrostomy(-fed); NG, nasogastric; NFD, nasal-feeding device.

Sullivan 2005 is shown as the representative publication for this 57-child multicentre cohort; companion papers from the same cohort (Sullivan 2004, caregiver QoL only—no nutritional outcome; Sullivan 2006b, interim report deferring growth data to this paper) are not shown separately.

Salomone 2022 is the larger, later-cutoff analysis (n = 24, data to March 2022) of the same Naples cohort reported in Salomone 2023 (n = 17, data to December 2020); the earlier report is not shown separately.

Gębora-Kowalska 2012 shares four of its five authors with Toporowska-Kowalska 2011; data were extracted from the abstract only, as the full text was unavailable.

Table 4. Respiratory outcomes after gastrostomy.

Chest-infection frequency before versus after gastrostomy, respiratory and aspiration-related hospitalisations, and aspiration events. GT, gastrostomy(-fed); NG, nasogastric; GJ, gastrojejunal; LRTI, lower respiratory tract infection; IRR, incidence rate ratio; HR, hazard ratio; NS, not significant.

Companion publications reporting the same cohort.

Data-quality caution, as noted in the cell.

Table 5. Minor complications after gastrostomy.

RTI, respiratory tract infection; GOR, gastro-oesophageal reflux; GT, gastrostomy(-fed); NG, nasogastric; NS, not significant.

Franken 2020a is the larger, later cross-sectional analysis (n = 150) of an overlapping Netherlands cohort; the earlier interim report from the same group (Franken 2019, n = 50) is not shown separately for this outcome.

Clayton 2024 reports two potentially overlapping caregiver/patient subgroups (a case-note record review and an anonymised UK-wide survey); the authors state some individuals may appear in both and the overlap cannot be quantified; the two counts should therefore not be treated as independent.

Table 6. Major complications after gastrostomy.

Buried bumper syndrome, colonic or other visceral injury, gastrocolic fistula, peritonitis, major haemorrhage, unplanned reoperation and procedure-related death. GT, gastrostomy; PEG, percutaneous endoscopic gastrostomy; PEG-J, percutaneous endoscopic gastrostomy with jejunal extension; GJ, gastrojejunal; FG, fundoplication with gastrostomy; BBS, buried bumper syndrome; NI, neurological impairment.

Companion publications reporting the same cohort.

Data-quality caution, as noted in the cell.

3.2.1 Nutritional outcomes

Weight and weight-based indices improved after gastrostomy in most cohorts reporting them, but the studies with a parallel comparator group did not agree. Fifty-six studies assessed weight, height, BMI, or micronutrient status before and after gastrostomy placement; six used a parallel oral-, nasogastric-, or alternative-feeding comparator instead of a pure pre/post design (24–28), though in one of these the surgical arm received gastrostomy together with Nissen fundoplication. We classified the direction of effect as favourable in 39 studies, mixed in eight, no significant change in five, and not assessable in three lacking a baseline measurement (29–31). No study reported nutritional deterioration. These counts give equal weight to 12-child small case series and to a 24,920-child cohort, and do not incorporate methodological quality: two studies reached low risk of bias, fourteen moderate, nine moderate-to-high, twenty-four serious and two high, with four not appraised. We therefore ordered the synthesis below by study design and risk of bias rather than by how often a favourable classification appeared.

The six comparator-controlled studies split in direction. Day et al. (25), an administrative cohort of 24,920 children and one of two studies at low risk of bias, found that gastrostomy-fed children gained more weight than orally-fed children across every age band studied, and that gastrostomy-fed children with severe cerebral palsy grew taller than orally-fed peers with the same diagnosis. Caselli et al. (24) found greater arm circumference, triceps skinfold and adipose arm area than in orally-fed peers, and no difference on other measures. Vergara et al. (26), a conference abstract, reported weight-percentile improvement in 29 children treated surgically and none in 26 treated non-surgically; gastrostomy and Nissen fundoplication were performed together, so the contribution of gastrostomy alone cannot be separated. Against these, Sullivan (32) found no difference in WAZ from an oral-fed comparator after adjustment for baseline status and GMFCS level (mean difference 0.002, p = 1.0), and Wu et al. (28) found a numerically larger weight gain with gastrostomy than with continued nasogastric feeding during inpatient rehabilitation for acquired brain injury that was not significant (5.1 kg vs. 2.9 kg, p = 0.2). Both were small, both carried serious risk of bias, and in Wu the groups differed at baseline in nutritional status.

Among the uncontrolled cohorts, the largest datasets pointed the same way. Sullivan et al. (33), a prospective multicentre cohort of 57 children, reported the largest single median WAZ change in this domain: −3.03 at baseline to −1.60 at 12 months (p < 0.0001), with catch-up growth in 91% of children. Franken et al. (34), the largest single dataset for this outcome (n = 300), reported a similar direction of effect (WLZ improved from −0.98 to −0.26, p < 0.0005); a quarter of that cohort had a non-NI diagnosis, including cystic fibrosis, and growth data were not stratified by diagnosis. Both were rated at serious and moderate risk of bias respectively. Catto-Smith (35) observed a mean increase in weight-for-age z-score of 1.05 after a median of 6.1 months in a cohort of 98 neurologically impaired children.

Motil et al. (36) found that gastrostomy feeding improved linear growth and lean body mass alongside a shift in body composition toward proportionally higher adiposity, a pattern also seen in Seguy et al. (37) and Franken et al. (34). Arrowsmith (38) similarly noted a shift in body composition, finding that while weight SDS improved significantly (−4.8 to −3.0, p < 0.05), percent body fat by skinfold also increased significantly from 10.7% to 16.3%, with no significant catch-up in height (38). Grayeb et al. (39), the longest follow-up in this domain (median 64 months) and the second study at low risk of bias, found persistent iron deficiency in 64% of tested children and iron-deficiency anaemia in 31% despite significant WAZ improvement. Sanders et al. (40) found a graded relationship between timing of gastrostomy and nutritional recovery: children treated earliest after their neurological insult had the fastest weight-gain velocity (3.3 g/kg/day) and the greatest reversal of malnutrition, against 1.5 g/kg/day in those treated latest. Gębora-Kowalska (41) reported a similarly steep trajectory: BMIZ rose from −4.83 at baseline to −2.39 at 12 months (p < 0.001), with cachectic children falling from 52% to 19%. Pinheiro et al. (42), the only study in a population defined by congenital Zika syndrome, found significant WAZ improvement but incomplete catch-up in 39% of children. Finbråtten et al. (30), the largest single-diagnosis registry in this review (661 children with cerebral palsy), reports no pre/post comparison but found growth retardation associated with both later gastrostomy insertion and greater GMFCS severity. Neither Sanders nor Finbråtten randomised the timing of placement, and children who received gastrostomy earlier may differ from those treated later in disease severity and clinical indication.

Several further cohorts reported the same direction of effect. O'Loughlin et al. (43), the second-largest single cohort in this domain (n = 122), found WAZ improved from −2.57 to −1.65 (p < 0.00005). Demirbaş et al. (44) compared gastrostomy alone with gastrostomy plus concurrent Nissen fundoplication; both arms showed significant WAZ and BMIZ improvement. Chae-ri Suh et al. (7) found significant improvement in BMIZ, HAZ, and WAZ within 6 months under a structured paediatric nutrition-team protocol. Samuel (n = 32) (45) found WAZ improved from −1.86 to −1.47 at 3 months. Chao et al. (46) found no significant overall change in weight percentile (15th to 17th, p = 0.64), though 20% of children caught up by more than 10 percentiles.

Zhang et al. is the only randomised evidence in this domain: 84 children with traumatic brain injury and dysphagia were randomised to PEG or continued nasogastric feeding, and weight, serum protein, albumin, and haemoglobin all rose more in the PEG group by 28 days (p < 0.05 for each), although within the control group the weight change itself was not significant (p = 0.089). Follow-up was 28 days; nutritional values were presented graphically only, the study was not prospectively registered, and it was rated at high risk of bias. The authors note their findings require confirmation in larger-scale studies.

Six studies found no significant nutritional change; two of these, Sullivan (32) and Wu (28), are the comparator-controlled studies described above. Franken (47) followed patients for four months or less. Lefton-Greif (48) assessed a paired subsample of ten patients. Jennuvat (49), followed to three years, attributed the absence of catch-up growth to low household income. None of these studies had the power to exclude a modest effect, and a null result here does not establish that gastrostomy has none.

Gastrostomy feeding is associated with improvement in weight and weight-based nutritional indices across the majority of included studies. Height gain and micronutrient correction lag behind weight gain in a minority of cohorts. The one comparator-controlled study that adjusted for baseline nutritional status and motor severity found no difference from oral feeding at 12 months, and both comparisons against nasogastric feeding ran for 28 to 43 days only.

3.2.2 Respiratory outcomes

Fifty-six of the included studies reported a respiratory outcome; chest-infection frequency before vs. after gastrostomy, respiratory or aspiration-related hospitalisations, or discrete aspiration events. These records represent approximately fifty-four independent cohorts: Sullivan 2005 and Sullivan (11, 33) report complementary respiratory data from the same 57-child Oxford Feeding Study, and Jacoby and Downs (50, 51) both analyse the same linked Western Australian and New South Wales birth cohort (n = 673), so neither pair should be counted twice in any overall count; two further duplicate Covidence records (52, 53) and a duplicated Kehar (54) record were resolved to a single entry each. The overall direction is favourable but considerably less consistent than for nutritional outcomes: twenty-four studies reported a reduction in chest infections, respiratory admissions, or aspiration events after gastrostomy, four found no significant change, eight described a mixed picture, and four reported an unfavourable signal, while sixteen mentioned respiratory outcomes without a baseline comparison, a denominator, or a numerator sufficient to judge direction. Where improvement was quantified it was often substantial: Sullivan et al. (11, 33), which were assessed as having a moderate-to-high risk of bias and serious risk of bias, found that the mean number of chest infections requiring antibiotics fell from 1.8 to 0.9 at twelve months, that admissions for chest infection fell from a mean of 0.5 to 0.09 (p = 0.04), and that the proportion of children admitted at least once with a chest infection fell from 26% to 7% (p = 0.021), with no child admitted for a chest infection by twelve months. Comparable magnitudes were reported by Chao (46) (moderate risk of bias), AlNamshan (55) (serious risk of bias), Ramelli (56) (serious risk of bias), and Jennuvat (49) (moderate/high risk of bias; pneumonia-related admissions 1.6 to 0.6 per patient, p < 0.001; ICU admissions 0.45 to 0.13, p = 0.006; reflux- and aspiration-related admissions 0.67 to 0.09 per person-month, p < 0.001; 26 of 29 children with a pre-operative history of chest infection had none at review; aspiration-pneumonia admissions median 1 to 0.5 per year, p < 0.001).

The countervailing evidence comes from the largest and most methodologically rigorous datasets, all of which are population-linked and comparator-controlled. Lin et al. (12), a lower-risk study with a moderate risk of bias, a 1:2 propensity-score-matched cohort of 2,490 children with neurological impairment and a prior pneumonia admission, found subsequent severe pneumonia hospitalisation in 32% and an increased risk of the composite outcome of death or severe pneumonia after new gastrostomy placement (HR: 1.21), even though mortality alone fell (HR: 0.47); Jacoby et al. (50), assessed at low risk of bias, and Downs et al. (51), assessed at serious risk of bias due to reporting from an abstract, analysing 673 Australian children against their own pre-gastrostomy year, found that lower-respiratory-tract-infection admissions rose in the first post-operative year (IRR: 1.26) and remained elevated at six years or more (IRR: 1.42) despite an overall decline in all-cause admissions; and Nelson et al. (57), also among the higher-quality evidence with a low/moderate risk of bias, an exposure-crossover analysis of 948 children, found reflux-related hospital days and acute health-care use unchanged before and after tube placement (adjusted RR: 0.92, 95% CI: 0.57–1.48). This split is systematic rather than random: the favourable studies are predominantly small, single-centre, at serious or moderate/high risk of bias, and dependent on caregiver recall or clinician-recorded infection counts, whereas the null and unfavourable studies are large, administratively linked, comparator-controlled, and generally carry a lower risk of bias (low to moderate), which suggests that part of the apparent respiratory benefit reflects recall and selection effects rather than a true treatment effect. Mechanistic and high-risk-population data point the same way: Thomson et al. (53), judged to have moderate/high risk of bias, showed that the proportion of reflux events reaching the pharyngeal space rose from 56% to 82% after PEG (p = 0.022) with no clinical change in chest-infection frequency, and Vajsar et al. (58) recorded aspiration pneumonia in 5 of 12 and respiratory failure requiring PICU admission in 4 of 12 children with SMA type 1 within thirty days of insertion. In summary, gastrostomy is associated with fewer chest infections and fewer respiratory admissions in most small before-and-after cohorts, though confidence in these findings is limited by widespread moderate to serious risk of bias; conversely, the methodologically stronger studies with lower risk of bias show no reduction—and in some analyses an increase—in pneumonia-related hospitalisation, indicating that gastrostomy should not be expected to abolish aspiration risk, particularly where salivary aspiration or uncontrolled reflux is the dominant mechanism.

3.2.3 Minor complications

Minor complications after gastrostomy were common but rarely severe. Forty-two studies reported infection, granulation tissue, leakage, or dislodgement in sufficient detail for this table. All were observational and none was rated at low risk of bias: 16 carried moderate risk of bias, two moderate/high, 21 serious and one high; two conference abstracts were not formally appraised. We classified the direction of effect as mixed in 21 studies, unfavourable in 13, and favourable in eight. These counts give equal weight to a 239-child prospective multicentre series and to a twenty-child retrospective case series; the synthesis below is therefore ordered by study design and risk of bias rather than by the frequency of any single classification. Reported rates for any minor complication ranged from 20% (complications tracked only to 30 days) (59) to 95% (caregiver interview) (9), a range corresponding to differences in ascertainment method more closely than to differences between populations. AlNamshan et al. (55) reported one of the highest rates in this table (n = 178): 71.3% site infection alone, with leakage (36.5%) and dislodgement (30.3%) also frequent, although the cohort-level total reported in that study (61%) is lower than the sum of its individual event rates. Franken et al. (34), the largest cohort in this domain (n = 300), reported a comparable overall rate of 73.7% alongside a low major-complication rate of 2.0%. Catto-Smith (35) reported fever in 17%, cellulitis in 7%, and peritonism in 4% of patients. Across studies reporting both categories, minor events exceeded major events by approximately an order of magnitude.

Wong et al. (60) offers a distinct signal drawn from emergency-department data rather than surgical follow-up: tube dislodgement accounted for 70% of gastrostomy-related presentations, and gastrojejunostomy tubes required hospital admission far more often than gastrostomy tubes (40% vs. 7%, p = 0.001). Viswanath et al. (61) adds a further nuance to the fundoplication pattern discussed elsewhere in this review: wound infection was more frequent with gastrostomy alone (11.0% vs. 3.3%), while gas-bloat syndrome occurred more often with concurrent fundoplication. Langer et al. reported that bowel obstruction, wrap herniation and gas-bloat occurred only where an additional anti-reflux procedure had been performed, and Cameron et al. (62) found skin erosion or infection in four children in each of the percutaneous endoscopic and surgical groups. Rates drawn from cohorts undergoing concurrent fundoplication or gastrojejunal extension are therefore not directly comparable with those from gastrostomy alone. Minor complications after paediatric gastrostomy are frequent but low acuity overall, and tube type, concurrent anti-reflux surgery, complication definition, ascertainment method, and follow-up duration account for most of the variability in rates reported across studies.

Study-level data for this domain are presented in Table 5.

3.2.4 Major complications

Fifty-nine of the included studies reported major gastrostomy- or fundoplication-related complications; peritonitis, visceral or colonic injury, gastrocolic fistula, buried bumper syndrome, major haemorrhage, reoperation, or procedure-related death. Eight further studies were excluded from this table after a data-quality check: Grayeb et al. (39) (low risk of bias) and Aumar et al. (63) (low/moderate risk of bias), whose authors explicitly did not report complication data; Benoit (64) (serious risk of bias), in which no complications of any kind were recorded; and Petersen (65), Kehar (54) (serious risk of bias) (54), Nakao (66) and Arnbjörnsson (67) (serious risk of bias) (65, 66), whose major-complication fields contain only minor events, such as granulation tissue, peristomal erythema, leakage and tube dislodgement, that belong in the minor-complication synthesis.

Across the 59 usable studies, the dominant pattern is that serious events are uncommon but not negligible: twenty-nine studies reported a low or nil major-complication rate, sixteen reported a burden high enough to be classified as unfavourable, ten described a mixed picture, and four reported major complications only qualitatively, in a mis-assigned field, or in a form that could not be separated from a mixed population. Among studies confined to gastrostomy alone, rates cluster tightly between 0% and 3.5%: Franken (34) (n = 300) 2.0% (moderate risk of bias), Lertudomphonwanit (59) (n = 201) 1.5% at 30 days (serious risk of bias), Işık and Akbulut (68) (n = 91) 3.3% (serious risk of bias), Sakamoto (69) 2.8% (moderate risk of bias), Sathesh-Kumar (70) 1.3% (moderate/high risk of bias), Jadi (71) 1.6% (serious risk of bias), and 0% in Borowitz (72) (serious risk of bias), Seguy (37) (serious risk of bias), Saleem (73) (serious risk of bias), Demirbaş (44) (serious risk of bias) and Sullivan (32) (serious risk of bias).

The specific events named in the review question are individually rare across observational literature. Catto-Smith (35) recorded isolated cases of esophageal tear (1%) and late gastrocolic fistula (1%). Buried bumper syndrome was reported by Koca (74) (1/34, 2.9%, managed without surgery, serious risk of bias), Jennuvat (49) (1/39, 2.6%, moderate/high risk of bias), Işık and Akbulut (68) (1/91, serious risk of bias) and Noviello (75)(5/148, 3.4%, moderate risk of bias); colonic injury or gastrocolic fistula by Suh (7) (transverse colonic perforation 1/18, 5.6%, requiring surgical re-insertion, moderate risk of bias), Fascetti-Leon (76) (6/239 cumulative, 2.5%, moderate risk of bias), Puntis (77) (colonic interposition preventing PEG placement in 2/29 and asymptomatic gastrocolic fistula in 1/29, serious risk of bias), Park (78) (1/32, 3.1%, requiring laparotomy, serious risk of bias), Noviello (75) (1/148, moderate risk of bias) and Khattak (79) (4/120, moderate risk of bias); peritonitis by Lertudomphonwanit (59) (2/201, 1%, serious risk of bias), Sakamoto (69) (1.4%, moderate risk of bias), Işık and Akbulut (68) (1/91, serious risk of bias), Sullivan (33) (1/57, 2%, moderate risk of bias), Viswanath (61) (1/82, 1.2%, serious risk of bias) and Khattak (79) (8/120, moderate risk of bias); and procedure-related death by Viswanath (61) (one gastric perforation, serious risk of bias), Cameron (62) (two 30-day deaths, both in the Stamm gastrostomy arm, serious risk of bias) and Khattak (79) (one sepsis-related death, moderate risk of bias). Meanwhile, several large cohorts—Grayeb (39) (low risk of bias, none of 94 childhood deaths attributed to the procedure), Dreznik (80) (serious risk of bias, 93 deaths, none procedure-related), Åvitsland (81) (moderate risk of bias) and Smith (9) (moderate/high risk of bias) which explicitly documented no procedure-attributable mortality.

The clearest and most consistent signal in this domain is that excess serious morbidity is driven by concurrent antireflux surgery and by jejunal extension rather than by gastrostomy itself. Viswanath et al. (61) (serious risk of bias) found post-operative complication rates significantly higher after gastrostomy with fundoplication than after gastrostomy alone (18/60, 30% vs. 12/82, 14.6%; p = 0.036), including the series' only procedure-related death; Guillén Redondo et al. (82), a comparative study at serious risk of bias, reported gastric perforation, gastro-oesophageal junction stenosis or dumping in 5 of 53 primary fundoplications (9%); Langer (83) (serious risk of bias) observed that bowel obstruction, wrap herniation and gas bloat occurred exclusively after an antireflux procedure; and Wales et al. (84), a comparatively detailed dataset despite serious risk of bias, recorded any major complication in 65.1% of fundoplication-plus-gastrostomy vs. 47.9% of gastrojejunal patients (p = 0.08), with fundoplication-specific events (wrap failure 14.3%; other major events 9.5% vs. 0%, p = 0.04) accounting for the difference while gastrojejunal tubes carried their own excess of intussusception and adhesive obstruction (20.8% vs. 7.9%, p = 0.06).

The same device effect appears in Stewart et al. (85) (moderate risk of bias), the largest dedicated buried-bumper series (29/535, 5.4%; 86% requiring laparotomy; mean 1,025 days to presentation), in which the syndrome affected approximately 27% of PEG-J vs. 2.7% of PEG placements, and in Wong et al. (86, 87) (low/moderate risk of bias), where 40% of gastrojejunal-tube emergency presentations required admission vs. 7% for gastrostomy tubes (p = 0.001). The highest reported rates belong to specific high-risk populations and broad pre-specified definitions rather than to technique: Vajsar et al. (58) recorded aspiration pneumonia in 5 of 12, respiratory failure requiring PICU admission in 4 of 12 and death in 2 of 12 children with SMA type 1 within thirty days, and Lefton-Greif et al. (48) (moderate risk of bias) reported 30-day mortality of 11% (3/28) from pneumonia, pneumothorax and respiratory failure, which are figures that reflect the fragility of the children selected for the procedure more than the procedure itself.

In summary, major complications after paediatric gastrostomy for neurological impairment are uncommon, affecting roughly 0%–3% of children when the procedure is performed alone, with buried bumper syndrome, colonic injury and peritonitis each reported in fewer than 1 in 20. Although confidence in precise rates is limited by the predominance of studies at serious or moderate risk of bias, the finding that serious-morbidity burden rises several-fold with concurrent antireflux procedures or jejunal extensions is consistent across study types, and the highest mortality figures come from cohorts of profoundly fragile children in whom deaths were consistently attributed to underlying disease rather than to the gastrostomy.

Study-level data for this domain are presented in Table 6.

3.2.5 Neurodevelopmental outcomes

The primary outcome domain of this review is lacking, where only eleven studies reported anything at all bearing on neurodevelopmental function, and of these only one applied a standardised instrument. We classified the direction of effect as favourable in three studies, all of which describe development qualitatively and none of which measured it; as no difference in three; and as not assessable in five, in which the domain was either not measured or reported in a form from which no numerator could be extracted. No study reported an unfavourable neurodevelopmental effect, and none reported cognition, gross or fine motor function, communication, adaptive behaviour or school participation using a validated developmental instrument before and after gastrostomy. For every other included study, the neurodevelopmental field of the extraction form was completed as not reported or left blank. Importantly, none of the studies contributing to this domain was at low risk of bias. Among the studies with the lowest risk-of-bias ratings, Sanders et al. (40)was rated moderate, while Clayton et al. (88), Tawfik et al. (89), and Wesley et al. (90) were rated moderate/high. Even within this comparatively stronger subset, neurodevelopmental outcomes were not assessed using validated instruments.

The single exception is instructive. Wu et al. applied the WeeFIM Cognition scale at admission and discharge to 74 children undergoing inpatient rehabilitation after acquired brain injury, comparing those who received a new gastrostomy with those who continued nasogastric feeding. Cognitive function improved substantially across the whole cohort, from 12.2 ± 6.5 to 19.4 ± 7.9 (p < 0.001), but the improvement was unrelated to the feeding route: tube type had no effect on the subtotal at either admission or discharge (p > 0.8), both arms improved within the nil-by-mouth subgroup, and every between-group comparison was non-significant. Eating function, measured on the IDDSI Functional Diet Scale, improved from a median of 0 to 8 (p < 0.001) with no difference between arms at discharge. The study is small, non-randomised, confined to a single inpatient rehabilitation admission of mean 43 days, and reports its cognition test statistic inconsistently between the Results and the Abstract, and was rated at serious risk of bias. Accordingly, although cognitive recovery occurred during follow-up, the study did not demonstrate a difference in recovery attributable to gastrostomy compared with continued nasogastric feeding.

The remaining ten studies measured proxies. Four reported seizure control, which is the most frequently recorded neurological outcome in this literature and the least consistently measured. Clayton et al. (88) documented improved seizure control in 11 of 21 gastrostomy recipients (52%) from case-note review of a Dravet syndrome cohort, without a seizure diary, and separately asked caregivers to report perceived change in convulsive seizures, cognition and walking; the survey proportions appear only as pie charts and are not numerically extractable. The study was rated at moderate/high risk of bias. Samuel (45) found seizure frequency unchanged in 76% of 21 children with epilepsy, improved in three and worsened in two and was rated at high risk of bias. Tawfik et al. (89) found epilepsy control improved in three, worse in three and unchanged in 14, not statistically significant; this study was rated at moderate/high risk of bias. Stringel (91) reported the most favourable seizure result in the review, with 73% of 35 children under neurology follow-up showing a significant reduction in seizure activity and 90% a reduced anticonvulsant dose, but the procedure in that series was a Nissen fundoplication and no standardised measure was applied. It was also rated at serious risk of bias. Sanders et al. (40) and Wesley et al. (90) describe developmental progress qualitatively, in the language of good outcomes and improved neurologic growth, without measurement. Sanders et al. was rated at moderate risk of bias, the lowest rating among studies in this domain, while Wesley et al. was rated moderate/high; thus, even the comparatively stronger evidence for favourable developmental change relied on qualitative, non-standardised assessment. Zhang et al. (27), Tarek et al. (92) and Kaur et al. (93) measured swallowing function or recovery of oral feeding, which are outcomes of the feeding route rather than of development. Benoit et al. (64) recorded developmental delay in 93% of a subgroup as a baseline characteristic, with no outcome applied.

Overall, studies relying on narrative or caregiver-reported assessments tended to describe neurodevelopmental improvement, whereas the only study using a standardised cognitive measure found no improvement attributable specifically to gastrostomy compared with continued nasogastric feeding. Given the small number of studies, absence of low-risk evidence, and lack of standardised neurodevelopmental assessment, the available evidence is insufficient to determine whether gastrostomy has a favourable, unfavourable, or neutral effect on neurodevelopment.

The question this review set out to answer, and the question families most often ask when a gastrostomy is proposed, therefore cannot be answered from the existing literature. This is the clearest finding of the review.

Study-level data for this domain are presented in Table 7.

Table 7. Neurodevelopmental outcomes after gastrostomy.

Cognition, gross and fine motor function, communication, adaptive behaviour, school participation and seizure control, by any instrument. Only Wu 2026 applied a standardised instrument. WeeFIM, Functional Independence Measure for Children; IDDSI, International Dysphagia Diet Standardisation Initiative; NI, neurological impairment. Swallowing function and recovery of oral feeding are outcomes of the feeding route rather than of neurodevelopment and are classified as not assessable for this domain.

3.2.6 Gastrointestinal outcomes

Gastro-oesophageal reflux after gastrostomy is the one outcome domain in this review where the direction of effect depends less on study design than on the child's reflux status before the operation. Only 12 of the 56 studies (21%) explicitly stratified patients according to the presence or absence of reflux before gastrostomy, allowing postoperative outcomes to be interpreted in relation to baseline reflux status (26, 34, 56–58, 75, 79–81). A further four studies provided partial subgroup or symptom-based tracking, whereas the remaining 40 studies (71%) did not stratify outcomes by pre-gastrostomy reflux status. Consequently, direct assessment of de novo reflux among patients without pre-existing disease and changes in reflux among those with established disease was possible in only a minority of studies.

Fifty-nine studies reported new or worsened reflux, physiological reflux measurement before and after placement, or the rate of subsequent (non-concurrent) antireflux surgery. We classified the direction of effect as favourable in 19 studies, unfavourable in 14, mixed in 15, no difference in three, and not assessable in eight, which either reported baseline reflux prevalence only, did not separate the neurologically impaired subgroup, or did not delineate reflux from a composite gastrointestinal outcome.

The physiological studies are the most informative and they agree with one another. Thomson et al. (53) measured pH and impedance before and after PEG in ten children and found total reflux events rose from 183 to 355 (p = 0.047), acid events from 27 to 173 (p = 0.028), and the median distal reflux index from 0.25% to 2.95% (p = 0.032); no child had an abnormal reflux index before PEG and three of ten did afterwards. Machado et al. (94) found a significant increase in weakly acid reflux episodes after PEG (mean difference +17.4, 95% CI: 1.6–33.5, p = 0.04), and Demirbaş et al. (44) found the total bolus exposure index rose in the gastrostomy-alone arm from 4.01 to 5.67 (p = 0.041). Toporowska-Kowalska et al. (95) add a qualification rather than a contradiction: the number of children with pathological reflux fell from 6/15 to 2/15, but the refluxate became significantly more acidic (p < 0.001). This study was judged to have moderate risk of bias.

Kawahara et al. (96, 97), both at moderate risk of bias, provide the mechanism that reconciles the apparently conflicting clinical literature. In both of their cohorts the effect of gastrostomy on reflux ran in opposite directions depending on baseline status. Among children without pre-operative pathological reflux the median reflux index rose significantly (2.1% to 4.5%, p = 0.004) and roughly a third crossed into the abnormal range, while among children who already had pathological reflux the index fell significantly (11.2% to 9.8%, p = 0.04) and a third normalised. Park et al. (78) found the same pattern clinically, with aggravated reflux in 6.3% overall but less severe disease in 13 of 15 children who had histological GERD before the operation, although this study was at serious risk of bias. Taken together, the moderate-risk studies suggest that gastrostomy may worsen reflux in children who did not have it and improve it in some children who did, although this pattern should be interpreted cautiously given the observational evidence base. Franken et al. (34), the largest cohort in this domain (n = 300) and judged to have moderate risk of bias, illustrates the consequence: overall GERD symptom prevalence was essentially unchanged (57.8% to 54.2%, p = 0.824), a null result that conceals de novo reflux in 34% of children who had none beforehand and resolution in 28% of those who did.

De novo reflux after gastrostomy alone is common. Langer (1988) (58) reported the highest rate in this table, with 22 of 50 surviving children (44%) who had no demonstrable reflux pre-operatively developing symptomatic reflux after gastrostomy alone, at a mean of nine months, however, this study was at serious risk of bias. Puntis et al. (77) found symptomatic reflux in 15 of 27 children (55.5%) after PEG, also in a study at serious risk of bias, Saadah et al. (98) in 16 of 25 (64%), and Heine et al. (1995) (83), at moderate risk of bias, found the clinical severity of reflux increased in 8 of 30 (p < 0.005), exacerbating disease in roughly a quarter of children with previously no or mild reflux. Aumar et al. (63), the largest prospective dataset in this domain and one of the lower-risk studies (low/moderate risk of bias), found de novo GERD in 12% and worsening in 25% of those with pre-existing disease. Against these, several cohorts report much lower de novo rates, most of them retrospective and symptom-based rather than physiological: Khattak et al. (79) 4% of neurologically impaired children, Viswanath et al. (61) 3.8%, and Sakamoto et al. (69) 2.4%.

Subsequent antireflux surgery is the outcome with the most direct bearing on practice, and its rate is consistently modest across studies, although confidence in individual estimates varied with risk of bias. Across the studies reporting it, the proportion of children who went on to require a fundoplication after gastrostomy alone clustered between 0% and 17%, most commonly between 4% and 8%: Franken (34) 0.7%, Wadman (99) 4.3%, Åvitsland (81) 4%, Zambaiti (100) 6.3%, Guillén Redondo (82) 7.2%, Ulman (101) 7.5%, Puntis (77) 7.4%, Dreznik (80) 10.1%, and Williams (102) 3.3% at five years alongside a further 10.2% converted to a gastrojejunal tube. These estimates were drawn predominantly from observational studies with varying risk of bias, although the broadly similar rates across studies provide some consistency to the finding. Of note, Nelson et al. (57), a study with low/moderate risk of bias, reported subsequent antireflux surgery in 5.3% at five years, mirroring the findings of the higher-risk studies. The outliers are informative rather than anomalous. Viswanath et al. (61) reported 17.1% in the era when their unit performed gastrostomy alone and reserved fundoplication for medical failure, which is the expected consequence of that policy rather than evidence of harm, although the observational design limits causal interpretation. Harrington et al. (103) reported 23% within two years and identified the reason: a seizure disorder raised the risk of requiring subsequent antireflux surgery more than fourfold (p = 0.03), with a negative predictive value of 87%, however, this association arose from a study at serious risk of bias and should therefore be interpreted cautiously. Dreznik et al. (80) found the converse in the same population, with 88% of neurologically impaired children with refractory seizures having recurrent reflux symptoms or aspiration pneumonia even after fundoplication. As for Catto-Smith (35), they reported that 14% of their cohort developed clinically significant reflux with aspiration post-gastrostomy, all of whom required subsequent fundoplication, placing their reoperation rate at the higher end of the reported 0%–17% cluster (35).

Two controlled comparisons address whether a concurrent antireflux procedure prevents this. Redondo et al. (82) found that among children with pre-existing reflux, symptoms worsened in 36% after gastrostomy alone against 15% after gastrostomy with a concurrent antireflux technique. Barnhart et al. (104), a propensity-matched cohort of 4,163 infants, found no difference between matched pairs in reflux-related hospitalisation (difference −0.05 per year, 95% CI: −0.20 to 0.15), so the available controlled evidence does not demonstrate that adding a fundoplication translates into fewer reflux-related admissions. Although these comparative data are methodologically stronger than the uncontrolled series, residual confounding remains inherent to the non-randomised designs. Set against the major-complication findings reported in section 3.3.4, where concurrent antireflux surgery accounted for most of the excess serious morbidity, these findings illustrate the potential trade-off between reflux control and procedural morbidity, although the evidence is insufficient to establish the optimal approach.

Overall, despite the predominantly observational evidence and variable risk of bias, reflux after gastrostomy is common, is frequently new in children who did not have it before, and improves in a substantial minority of those who did. Across studies, most cases did not progress to antireflux surgery, with subsequent fundoplication generally reported in fewer than 10% of children, and studies with low/moderate risk of bias observed broadly similar estimates. Seizure disorder was identified as a potential predictor of subsequent antireflux surgery in one study, but this finding requires cautious interpretation given its serious risk of bias.

Study-level data for this domain are presented in Table 8.

Table 8. Gastro-oesophageal reflux and subsequent antireflux surgery after gastrostomy.

New or worsened reflux, physiological reflux measurement before and after placement, and rate of subsequent (non-concurrent) antireflux surgery. GER, gastro-oesophageal reflux; GERD, gastro-oesophageal reflux disease; MII-pH, multichannel intraluminal impedance with pH monitoring; PPI, proton pump inhibitor; GJ, gastrojejunal; NS, not significant. Concurrent antireflux procedures performed at the same operation are not counted as subsequent surgery.

Companion publications reporting the same cohort.

3.2.7 Mortality

The evidence was mostly observational and methodologically limited, with most studies assessed at moderate to serious risk of bias and only a small number contributing lower-risk evidence. Accordingly, mortality rates from uncontrolled cohorts were interpreted primarily as descriptive estimates, while greater weight was given to comparator-controlled and population-linked analyses.

Mortality in this population is high, is dominated by the underlying neurological disease, and is only rarely attributable to the gastrostomy itself. Sixty-seven studies reported mortality. We classified the direction of effect as favourable in 40 studies, mixed in 14, unfavourable in six, no difference in one, and not assessable in six. The favourable classification here denotes the absence of procedure-attributable death rather than a survival benefit, since only three studies in the review compared survival against a non-gastrostomy comparator, and those three do not agree.

All-cause mortality during follow-up ranged from 0% in the short-follow-up prospective cohorts to 43% in Lefton-Greif et al. over a mean of 4.11 years, and clustered between 8% and 30% in cohorts followed for two years or longer: Åvitsland (81) 8.6%, Vergara (26) 10.1%, Stringel (91) 10%, Kawahara (97) 13.0%, Ramelli (56) 15.6%, O'Loughlin (43) 16%, Tawfik (89) 17%, Franken (47) 17.3%, Puntis (77) 17.2%, Guillén Redondo (82) 19.3%, Sathesh-Kumar (70) 22.4%, Heine (105) 23%, Grayeb (39) 26%, Smith (9) 26.2%, Koca (74) 26.4%, Dreznik (80) 26.9%, Aumar (63) 29%, Zambaiti (100) 33% and AlNamshan (55) 35%. These figures describe the fragility of the population selected for the procedure and should not be read as an outcome of it. Among the more informative longitudinal datasets, Nelson et al. (57), assessed at low/moderate risk of bias, found two-year survival of 87.4% (95% CI: 85.2–89.4) and five-year survival of 75.8% (95% CI: 72.8–78.4) in 948 children, and Smith et al. (9), although at higher risk of bias, reported the longest survival curve in this review, falling from 91% at one year to 55% at ten. Catto-Smith (35) reported progressive long-term mortality of 11% at 1 year, 21% at 2 years, 27% at 3 years, and 39% at 13 years, with only one early death (cardiac failure) at 7 days.

Procedure-attributable death is rare and, where it occurs, is concentrated in the same places as major complications. Wesley et al. (90) recorded one death directly attributable to the operation, on the eleventh post-operative day following sepsis and wound dehiscence. Viswanath et al. (61) recorded one, on the fourth post-operative day after re-exploration for a gastric perforation sustained during laparoscopic fundoplication, with no deaths in the gastrostomy-alone group. Aumar et al. (63) attributed two of 94 deaths to early procedural complications. Wales et al. (84) recorded two device-attributable deaths in the fundoplication arm, from small-bowel obstruction after fundoplication and from peritonitis following a gastrostomy tube change, and Langer (83) one reflux-attributable death after gastrostomy alone. Against these, a large number of cohorts examined the question explicitly and found nothing: Grayeb (39) (none of 94 deaths), Dreznik (80) (none of 93), Guillén Redondo (82) (none of 40), Franken (34) and Franken (14) (none of 26 in either), Sathesh-Kumar (70), Ramelli (56), Koca (74), Kaur (93), Smith (9), Stringel (91), Park (78) and Åvitsland (81) all state that no death was related to the procedure.

Thirty-day mortality, where reported separately, was 0% to 3% in most cohorts: AlNamshan et al. (55) 3% (5/178), Nelson et al. (57) 1.8% (17/948), and zero in Sathesh-Kumar 2009 (70), Chao (46), Khattak (79), Cameron (62) (PEG arm) and O'Loughlin (43). The two exceptions are the two smallest and most fragile cohorts in the review, and both are informative about patient selection rather than technique: Vajsar et al. (58) recorded 2 deaths among 12 infants with spinal muscular atrophy type 1 within 30 days (16.7%), and Lefton-Greif et al. (48) 3 among 28 (11%), from pneumonia, pneumothorax and respiratory failure.

Three studies compared mortality against children who did not receive a gastrostomy, and their findings conflict significantly. Lin et al. (12), a 1:2 propensity-matched cohort, found that new gastrostomy was associated with a reduced risk of death (HR: 0.47). Downs et al. (60) found survival rates similar to children without a gastrostomy. Crucially, Wong et al. (86), a propensity-score-matched cohort of 212 children with severe intellectual disability (assessed at low/moderate risk of bias), found a significantly increased risk of death in the gastrostomy group (HR: 2.87, 95% CI: 1.13–7.26) over a median of 6.25 years compared to matched controls fed orally. Given that Wong et al. and Lin et al. are the highest-quality comparative studies in this review and point in opposite directions, the evidence does not definitively support a survival benefit and suggests long-term mortality risk may be elevated in some gastrostomy-fed populations (12).

Two further observations bear on timing, although both derive from small studies at serious risk of bias. Sanders et al. (40) found mortality rising steeply with delay to intervention, from 7% in the earliest-treated group to 40% in the latest, and Heine et al. (105) found that a falling weight-for-age z-score after PEG identified children at risk of death, with four of the eight children whose z-score fell subsequently dying. Both were uncontrolled and susceptible to confounding by the severity that prompted delayed intervention; these findings should therefore be considered hypothesis-generating rather than evidence that earlier gastrostomy reduces mortality.

Overall, children receiving gastrostomy for neurological impairment had substantial all-cause mortality, but deaths were predominantly attributed to underlying disease rather than the procedure itself. Procedure-attributable mortality was rare, and 30-day mortality was generally 0% to 3% outside the most fragile diagnostic groups. Although confidence in precise mortality estimates is limited by the predominance of observational studies at moderate to serious risk of bias, the lower-risk evidence similarly suggests that gastrostomy itself is rarely a cause of death. Evidence that gastrostomy improves survival, however, remains limited, with only one moderate-risk propensity-matched study demonstrating a reduction in short-term mortality.

Study-level data for this domain are presented in Table 9.

Table 9. Mortality after gastrostomy.

All-cause and procedure-attributable mortality. A favourable direction denotes the absence of procedure-attributable death rather than a demonstrated survival benefit; only Lin 2021, Downs 2019, and Wong 2020 compared survival against a non-gastrostomy comparator. Attribution of cause is as stated by the original authors and was not independently adjudicated. GT, gastrostomy; PEG, percutaneous endoscopic gastrostomy; SG, Stamm gastrostomy; GJ, gastrojejunal; HR, hazard ratio; ID, intellectual disability; NI, neurological impairment.

Companion publications reporting the same cohort.

Data-quality flag, as noted in the cell.

3.2.8 Quality of life & caregiver-reported outcomes

Quality of life is the outcome domain in which the volume of favourable reporting is least matched by the strength of the measurement behind it, and the synthesis therefore required an additional classification step before the direction of effect could be judged.

Twenty-seven studies contained a statement about quality of life. Three were set aside before classification: Sullivan et al. (33), which is not itself a quality-of-life report and refers readers to its companion paper for that outcome; Stringel (91), in which every child also had a fundoplication, so any reported change cannot be attributed to gastrostomy rather than to the combined procedure; and Barnhart et al. (104), which never measured quality of life. The remaining 24 were classified by the source of the quality-of-life claim. Seventeen (70.8%) drew on a questionnaire put to the patient or caregiver, which we designate Category A. Four (16.7%) reported a narrative account from the patient or caregiver with no questionnaire behind it, designated Category B (40, 106–108). Three (12.5%) rested on the study authors' own inference, with the claim never passing through the patient or family, designated Category C. Category C was excluded from further analysis, because treating an author's interpretation of an unrelated outcome such as weight gain as evidence of a quality-of-life benefit would inflate the apparent consistency of the reported association. The analysis set for this domain is therefore the 21 studies in Categories A and B, which between them contribute 16 child-focused results (Table 10) and 17 caregiver-focused results (Table 11).

Table 10. Quality of life after gastrostomy—child outcomes.

Claim category: A, the claim rests on a questionnaire put to the patient or caregiver; B, the claim is a narrative account from the patient or caregiver with no questionnaire behind it. Category C studies, in which the claim rests on the authors’ own inference and never passes through the patient or family, were excluded from the analysis set. PedsQL, Pediatric Quality of Life Inventory; CP-QOL-Child, Cerebral Palsy Quality of Life Questionnaire for Children; SAGA-8, Satisfaction Questionnaire with Gastrostomy Feeding; QI-Disability, Quality of Life Inventory-Disability; HRQoL, health-related quality of life; NI, neurological impairment.

Table 11. Quality of life after gastrostomy—caregiver outcomes.

Categories as for Table 10. SF-36, 36-Item Short Form Health Survey; VAS, visual analogue scale. Sullivan 2004 and Sullivan 2005 report the same Oxford Feeding Study cohort, and the record and survey subgroups of Clayton 2024 may overlap by the authors’ own statement, so the 21 studies in the analysis set are not 21 independent cohorts. Three studies appear in both Tables 10, 11 with different findings, since they report a child and a caregiver outcome separately.

3.2.9 Child quality of life

Among the 16 studies reporting a child outcome we classified the reported direction of change as favourable in 13, mixed in one (109), and not assessable in two for want of a baseline (14, 110). No study reported an unfavourable direction of change in either group. Because no included study was randomised or controlled, these are reported directions of change and not estimates of an effect of gastrostomy, and the count cannot be read at face value. Only one record is robust, scoring 11 of 12 because it alone combined a validated instrument, a prospective pre-operative baseline and a directly measured quality-of-life endpoint (109), and its graded direction is mixed rather than favourable. Of the three records of intermediate strength, meaning they satisfy most of the rubric but fail at least one of those three requirements, the highest-scoring is cross-sectional and therefore not estimable for want of a baseline (11 of 12) (14). The remaining twelve rows are weaker still: six of limited strength, typically with no validated instrument and a pre-operative state reconstructed from memory; three anecdotal, with no instrument, no numbers and no statistical testing; two qualitative and therefore ungraded; and one available only as a conference abstract, whose methods cannot be appraised at all. It is these weaker records that carry almost all of the favourable signal, in most cases with the pre-operative state reconstructed by the same respondent who supplied the post-operative rating.

The instruments behind these results are weak. Within the 17 Category-A studies only four (23.5%) used a validated instrument, three of them for the child outcome: the parent-proxy PedsQL 4.0 (14, 109) and CP-QOL-Child [Frediani (111), available only as an abstract]. For the child, only two studies applied a validated instrument, only one of which had a baseline against which change could be measured, and that study found no significant change in total health-related quality of life (p = 0.30), with improvement confined to the psychosocial summary (+5.4 points, p = 0.03) and driven by the social scale (+10.1 points, p = 0.04); the improvement was not modified by neurological impairment status (p = 0.66), but the change scores are reported for the whole mixed cohort rather than the impaired subgroup. Frediani (111) found significantly higher CP-QOL-Child scores in gastrostomy-fed than orally fed children (p < 0.002), but this is a between-group comparison in an abstract-only report rather than a within-child change. No study obtained child self-report.

Weighting therefore reverses the impression given by counting: the only record capable of supporting a moderate-certainty statement is mixed, so the child outcome is best described as no demonstrated change in measured health-related quality of life, alongside consistent caregiver-reported improvement in the child's well-being at low to very low certainty. At outcome level, certainty that gastrostomy is associated with a measured change in the child's own health-related quality of life is very low: the body of evidence is inconsistent between its single interpretable record and the remainder, imprecise, indirect in that the endpoint is proxy-reported throughout, and incompletely appraised for risk of bias.

3.2.10 Caregiver quality of life

Among the 17 reporting a caregiver outcome we classified 14 as favourable and three as mixed (9, 107, 112). Here the weighted picture is more favourable, because the single robust record falls on the favourable side: Sullivan (113) scores 12 of 12, the only full-mark record in either table, on a validated SF-36 version 2 administered prospectively with a pre-operative baseline. Of the three records of intermediate strength, each prospectively measured but relying on a purpose-built instrument or a proxy endpoint rather than a validated quality-of-life measure, one is favourable [Avitsland (81), 9 of 12] and two are graded mixed [Heine (105), 8 of 12; Thorne (112), 7 of 12]. The remaining thirteen rows are weaker: seven of limited strength, generally lacking a validated instrument and reconstructing the pre-operative state by recall; four anecdotal, with no instrument and no statistical testing; and two qualitative and therefore ungraded.

The fourth validated administration is the caregiver's own SF-36 version 2 (113). The remaining 13 (76.5%) of the Category-A instruments were purpose-built for the study in question, most commonly a researcher-developed multi-item questionnaire (49, 72, 89, 105), and otherwise a Likert-agreement survey (Clayton 2024), numeric rating scales (81), a visual analogue scale (112), or a named satisfaction instrument, SAGA-8 (110), which measures gastrostomy-specific satisfaction rather than general quality of life. Two further studies used validated frameworks (QI-Disability and the Beach Center Family Quality of Life Scale) to code qualitative interview transcripts rather than administering them as scored questionnaires (108), which strengthens the analysis but produces no scores, baselines or statistics. The domains tested are correspondingly narrow. Caregiver satisfaction was the commonest domain, appearing in 12 of the 21 studies, against only three studies that tested the validated physical, social, emotional and pain constructs that make up a generic health-related quality-of-life instrument.

For the caregiver, the strongest quantitative result in the whole review is available: Sullivan et al. (113) recorded significant and sustained SF-36 gains at 12 months in social functioning (+13.51, 95% CI: 1.55 to 25.48, p = 0.028), mental health (+9.88, 95% CI: 2.03 to 17.72, p = 0.015), energy and vitality (+9.94, 95% CI: 2.29 to 17.58, p = 0.012), general health perception (+6.35, 95% CI: 0.15 to 12.56, p = 0.045) and the mental component summary (+7.06, 95% CI: 0.90 to 13.22, p = 0.026), with most of the gain occurring in the first six months and sustained thereafter. By 12 months most domains were no longer significantly below UK population norms, and caregiver worry about the child's nutritional state fell from 78% before gastrostomy to 15% at 12 months (p < 0.0001). Åvitsland et al. (81) provides the next-best-quantified dataset, with prospective measurement at baseline, 6 and 18 months showing improvement in parent satisfaction (5.40 to 3.50, p < 0.001), parental stress (5.10 to 3.77, p = 0.015), parent-child communication during meals (4.57 to 2.57, p = 0.001) and proxy-rated child well-being (4.97 to 2.81, p < 0.001), on scales where a lower score is better.

Two qualifications run through the caregiver evidence. The first is that satisfaction measures acceptability rather than quality of life. When Borowitz et al. (72) report that 95% of parents would elect gastrostomy again and 100% would recommend it, or Koca et al. (74) that all 20 responding families would choose PEG again, that is a statement about whether families were glad they did it, not a measurement of what their quality of life became. The second is that where burden was measured rather than satisfaction, the picture is less uniform. Thorne et al. (112) found device satisfaction consistently above 80% on a visual analogue scale and unrelated to complication frequency, yet caregivers described long-term management as a continuing challenge requiring adaptive coping. Smith et al. (9) found that 86% of caregivers rated tube feeding as having a positive impact on the child's and family's life but that 28% reported tube-attributable family stress including difficulty arranging respite, restricted mobility and a changed relationship with the child. Mahant et al. (107) describes intense caregiving needs, sleep deprivation, social isolation and financial strain alongside the benefits. Avitsland et al. (81) found that meal duration did not fall significantly after gastrostomy, which directly contradicts Heine et al. (105), and that oral intake increased in a notable minority (34% at 6 months, 49% at 18 months) and decreased in none. A third qualification follows from the appraisal: where a published instrument replaced a bespoke one the favourable signal attenuated sharply, Russell (110) recording high satisfaction in only 50% of caregivers on SAGA-8 against 90 to 100% on the older bespoke surveys, and the most credible estimate of the unfavourable minority comes from the strongest sampling frame in the review, Smith's (9) population-based 28%.

Because one robust favourable record outweighs a larger number of limited and anecdotal records, the caregiver outcome is graded favourable at moderate certainty, with an unfavourable minority that is consistently present and consistently small. At outcome level, certainty that gastrostomy is associated with improved caregiver-reported quality of life is moderate, and is held below high by the absence of any controlled comparison, by the graded estimate resting on a single record, and by incomplete risk-of-bias appraisal across the remainder; the available evidence supports an association and not a demonstrated effect.

Limitations of the weighted result. Three features of the weighting limit what these results can be taken to show. First, the weight is concentrated rather than distributed: one record per outcome carries the only moderate-certainty statement available, so both conclusions rest on a single study each rather than on the 33 rows counted, and revision of either record's strength category would change the reported direction. Second, the weighting axis is the interpretability of each study's quality-of-life measurement rather than a complete appraisal of internal validity, because risk of bias is documented for only two of the records in these tables; a record can therefore be weighted highly here and remain at unappraised risk of bias. Third, the strength categories and certainty grades involved reviewer judgement at the margins, were performed by a single reviewer against extracted evidence, and await second-reviewer verification. The direction reported for each outcome is accordingly a weighted and partly subjective judgement about a body of uncontrolled evidence: it can suggest an association between gastrostomy and reported quality of life, but it cannot establish that gastrostomy caused the change.

Study-level data for this domain are presented in Table 10 (child outcomes) and Table 11 (caregiver outcomes).

3.3 Assessment of risk of bias

One hundred and two of the included studies were formally appraised. The remaining eighteen were conference abstracts, excluded from formal critical appraisal as they do not provide sufficient methodological detail to be reliably scored. Study-level judgements for every domain of every tool are given in Tables 12a–d, split by the instrument applied: RoB 2 for the single randomised trial (Table 12a), ROBINS-I for the non-randomised comparative and cohort studies (Table 12b), the JBI critical appraisal checklists for cross-sectional, survey and questionnaire-based studies (Table 12c), and the CASP qualitative checklist for the interview studies (Table 12d). Conference abstracts (n = 16) were excluded from formal critical appraisal, as they do not provide sufficient methodological detail to be reliably scored.

Table 12a. Risk of bias assessment of the included studies: RoB 2—randomised trial.

Table 12b. Risk of bias assessment of the included studies: ROBINS-I—non-randomised comparative and cohort studies.

Table 12c. Risk of bias assessment of the included studies: JBI critical appraisal—cross-sectional, survey and questionnaire-based studies.

Table 12d. Risk of bias assessment of the included studies: CASP qualitative checklist—interview studies.

D1 to D7 are the domains of the tool named in each sub-table heading. RoB 2 domains: D1 randomisation process, D2 deviations from intended intervention, D3 missing outcome data, D4 measurement of the outcome, D5 selection of the reported result. ROBINS-I domains: D1 confounding, D2 selection of participants, D3 classification of interventions, D4 deviations from intended interventions, D5 missing data, D6 measurement of outcomes, D7 selection of the reported result. Where a study contributed to more than one outcome domain and was appraised separately for each, the outcome is given in parentheses after the study name.

Overall methodological quality was low, and the weakness was systematic rather than idiosyncratic. Of the 49 studies appraised with ROBINS-I, 38 (86.4%) were rated at serious overall risk of bias, four at moderate and two at low to moderate. Not one reached an overall low rating. The domain responsible is almost always the same: confounding was rated serious in the great majority of these studies, and this is intrinsic to the field rather than a failure of individual investigators. Children are selected for gastrostomy because their feeding is already failing and their impairment is most severe, so any comparison against children who did not receive one is confounded by the reason the decision was made. Selection bias was rated moderate in nearly every ROBINS-I study for the related reason that cohorts were assembled from those who had already undergone the procedure. By contrast, the domains under the investigators' direct control performed well: classification of the intervention and deviations from intended intervention were rated low in almost every study, since gastrostomy placement is an unambiguous, well-documented exposure.

The two studies with the lowest ROBINS-I risk are the two population-linked administrative analyses, Nelson et al. and Wong et al. (57, 86) both rated low to moderate, and the next best are Lin et al. (12), Barnhart et al. (104), Ulman et al. (101) and Fascetti-Leon et al. (76) at moderate. This matters for interpretation rather than as a bookkeeping observation, because these are the same studies that most often disagree with the aggregate direction of effect in the respiratory and mortality domains. Where the highest-quality evidence and the majority vote diverge, the divergence tracks study quality.

Among the 50 studies appraised with the JBI checklists, four were rated low risk overall (25, 39, 50, 114), two low to moderate, 27 moderate and 17 moderate to high. The recurrent weaknesses here were confounding control and outcome measurement rather than sampling. Studies rated moderate to high were characteristically those relying on retrospective caregiver recall of the pre-gastrostomy state collected at the same sitting as the post-gastrostomy opinion (9, 43, 49, 89, 110), those reporting only cross-sectional status with no baseline (14, 24, 115), and the abstract-only reports in which the methods could not be checked against the results (111).

The single randomised trial, Zhang et al. (27), was rated at high overall risk of bias on RoB 2, with high risk in both the randomisation and deviations domains and some concerns for outcome measurement and selective reporting. This is consequential out of proportion to the study's size, because it is the only randomised evidence in the entire review and it is cited in the nutritional, respiratory, reflux and major-complication domains. Its own journal labels it an observational study despite a described random-number-table allocation, which is consistent with the randomisation-domain judgement. Its results should be treated as supportive rather than confirmatory.

The three qualitative studies performed best of any group against their own instrument. Nelson et al. (57) and Glasson et al. (108) were rated low risk on the CASP checklist and Russell et al. (110) low to moderate, with the only concerns being researcher reflexivity and recruitment. This does not make their findings quantitatively poolable, and they are not used that way in this review, but it does mean the caregiver-burden narrative they contribute is methodologically sound on its own terms.

Several cross-cutting reporting weaknesses affect the appraisal irrespective of tool. Functional severity was rarely classified with a standardised instrument such as the GMFCS, so residual confounding by impairment severity cannot be excluded even where confounding was otherwise addressed. Many studies grouped heterogeneous neurological diagnoses into a single composite category without disaggregating by aetiology, and several reported complication, growth, reflux and mortality outcomes for whole mixed cohorts rather than stratified by neurological-impairment status, which dilutes the estimates specific to this population. The studies drawing on administrative or claims data (50, 102) trade clinical granularity for scale and carry the misclassification risk inherent in diagnostic and procedural coding. Attrition was a specific concern in three cohorts: Martínez-Costa et al. (6) excluded 33 of 98 recruited children, mainly for missing anthropometric data; Pinheiro et al. (42) excluded 5 of 23 operated children including one death; and Kaur et al. (93) lost 16 children to follow-up without a stated follow-up duration. Where attrition is concentrated among sicker children, as is plausible in this population, the reported outcomes describe a healthier surviving subset and are likely to overstate benefit. Martínez-Costa's exclusions are documented and reasoned, which makes Kaur's unexplained loss the more confidence-reducing of the two.

Cohort size compounds these issues in a subset of studies. Shapiro (n = 19) (106), Pinheiro et al. (n = 18) (42), Brant (n = 20) (13), Thomson et al. (n = 10) (53), Toporowska-Kowalska et al. (n = 15) (95) and Valletta et al. (n = 11) (116) each enrolled fewer than 25 children. In cohorts this small a single atypical case moves a reported percentage substantially, so their point estimates are less stable than those from the larger cohorts, and the synthesis weights them accordingly.

The overall picture is an evidence base in which the exposure is well defined and well documented, the outcomes are variably defined and often retrospectively ascertained, and confounding by indication is pervasive and largely unaddressed. Eighty-six per cent of the comparative studies carry a serious risk of bias, no comparative study reaches low risk, and the only randomised trial is at high risk. Every directional conclusion in this review should be read against that background.

4 Discussion

4.1 Summary of main findings

This systematic review of 120 independent studies reveals a stark contrast between the well-documented nutritional and caregiver benefits of gastrostomy, and its unexpectedly unproven effects on respiratory health and child neurodevelopment. By synthesizing data spanning over four decades, it becomes clear that while gastrostomy is consistently associated with resolving the narrow endpoint of malnutrition, its broader systemic impacts require a highly nuanced clinical approach.

The most definitive benefit of gastrostomy placement is rapid nutritional recovery, supported by moderate certainty evidence. The majority of included cohorts demonstrated highly significant improvements in weight-for-age and body mass index (BMI) z-scores following tube placement. For example, prospective data highlighted a median weight-for-age z-score improvement from −3.03 to −1.60 within 12 months, resulting in robust catch-up growth. Crucially, the literature demonstrates a distinct dose-response relationship between the timing of the intervention and the speed of recovery; children treated earliest after neurological insult experience weight-gain velocities more than double those treated later (3.3 g/kg/day vs. 1.5 g/kg/day), alongside higher rates of full nutritional normalization. This aligns closely with broader international consensus, including the 2019 ESPGHAN guidelines and the Cochrane review of gastrostomy in cerebral palsy (16), which similarly identified weight gain as the most consistent benefit across literature. While these findings suggest that earlier placement, before prolonged undernutrition occurs, is associated with more robust catch-up growth, confounding by indication and severity cannot be excluded. The evidence supports offering gastrostomy without undue delay, though it does not establish a causal effect of timing alone.

Although gastrostomy is often used in practice to help prevent aspiration, the highest-quality data provides moderate-certainty evidence that it does not actually offer respiratory protection. While smaller, uncontrolled caregiver-reported cohorts frequently cite reductions in chest infections, large population-linked and propensity-matched datasets repeatedly contradict this. Population analyses have demonstrated that lower-respiratory-tract infection admissions often rise or remain unchanged post-gastrostomy (with incidence rate ratios remaining elevated up to six years post-placement), and propensity-matched data even indicated an increased hazard ratio (HR: 1.21) for severe pneumonia hospitalizations. This difference between study types is likely due to confounding by indication. In practice, the sickest children with the highest risk of aspiration are the ones most likely to receive a gastrostomy, which biases observational studies to make the procedure look harmful for respiratory health. When higher-quality studies adjust for this baseline severity using propensity matching, the apparent respiratory benefits disappear. This is consistent with the Cochrane review by Gantasala et al. (16), which also found conflicting evidence on respiratory illness. Ultimately, while gastrostomy bypasses oral feeding issues, it cannot prevent respiratory problems caused by salivary aspiration or poor airway clearance.

While the review includes various insertion techniques (e.g., percutaneous endoscopic, laparoscopic, and open surgical gastrostomy), comparative data isolating the specific effect of insertion technique on long-term outcomes remain sparse. Surgically, gastrostomy placement alone appears to be relatively safe (supported by low-to-moderate-certainty evidence), with major complications (such as peritonitis or colonic injury) and procedure-attributable mortality occurring in roughly 0% to 3% of cases. However, serious morbidity rises several-fold when a concurrent antireflux procedure (fundoplication) or a jejunal extension is added. Interestingly, broader recent literature supports a conservative approach to concurrent surgeries; a 2021 retrospective cohort study published in MDPI found that neurologically impaired children receiving gastrostomy alone actually experienced significantly fewer long-term reflux-related hospitalizations per year than those receiving a concurrent fundoplication, although the observational design limits causal interpretation. Within this review, physiological studies offer a possible explanation for the heterogeneous reflux outcomes observed across studies: gastrostomy was associated with de novo reflux in children who previously had none, while improvement in reflux indices was observed in a proportion of those with pre-existing disease. Although these findings were broadly consistent across several studies, confidence is limited by the predominantly observational evidence and moderate-to-serious risk of bias. Given that subsequent antireflux surgery was generally required in fewer than 10% of children, the available evidence does not support routine concurrent fundoplication in the absence of a clear clinical indication. The American Academy of Pediatrics (AAP) notes that adding a G-J tube or fundoplication increases the risk of surgery-related issues like bleeding and infection, as well as long-term risks like overfeeding and worsening reflux (117). Because of these risks, concurrent procedures should be reserved for children with severe vomiting, persistent aspiration, or highly symptomatic GERD. This cautious approach is also supported by recent Cochrane reviews on managing GERD in children with neurological impairments (17).

For families, gastrostomy significantly reduces caregiver burden (moderate-certainty evidence). Utilizing validated tools like the SF-36, studies have documented sustained improvements in caregiver mental health, vitality, and social functioning, alongside a massive drop in nutrition-related anxiety (falling from 78% pre-operatively to 15% at 12 months). However, this clear benefit for caregivers contrasts sharply with the lack of data on the child's neurodevelopment, where we found very low-certainty to no evidence. Across 120 studies, only one applied a standardized cognitive instrument (the WeeFIM scale), finding that developmental improvements were tied to natural rehabilitation rather than the feeding route itself. No robust evidence exists proving gastrostomy directly improves cognition, gross motor function, or communication.

Ultimately, synthesizing these discrete clinical trade-offs will be highly useful to help physicians and parents make a shared, evidence-based decision about whether gastrostomy will be beneficial for a child. By replacing generalized assumptions of respiratory protection with concrete data on nutritional timing and surgical safety, clinicians can guide families through a much more transparent and accurate consent process.

4.2 Limitations

The limitations of this review fall into three groups: limitations of the evidence, limitations of the review process, and limitations of the synthesis method.

The evidence itself is the principal constraint, and the formal appraisal quantifies it: 86% of the comparative studies assessed with ROBINS-I carry a serious overall risk of bias, no comparative study reaches low risk, and the only randomised trial is at high risk. Only one randomised comparison exists, its design classification is disputed by its own journal, and its follow-up is 28 days. The remainder is observational, and most of it is uncontrolled. Confounding by indication is unavoidable and largely unaddressed: gastrostomy is offered to the children who are least well, so any comparison with children who did not receive one is confounded by the severity that prompted the decision, and few studies adjusted for impairment severity or recorded GMFCS level at all. Outcome definitions are inconsistent to a degree that limits comparison across studies, particularly for chest infection, aspiration and minor complication. Follow-up is short in a large proportion of studies, which matters most for the outcomes that develop slowly, including linear growth, adiposity and buried bumper syndrome. Several cohorts report mixed populations in which the neurologically impaired subgroup is not separable, and where this is the case the finding is flagged in the relevant table and should not be read as neurology-specific. Two domain-specific constraints deserve separate mention. In the reflux domain, most studies report a single cohort-wide incidence without stratifying by pre-operative reflux status, which, given that the effect runs in opposite directions in the two groups, makes those incidences difficult to interpret and impossible to compare across studies with different baseline reflux prevalence. In the quality-of-life domain, almost every result is proxy-reported by a caregiver or clinician rather than by the child, most non-validated instruments ask families to recall the pre-gastrostomy state at the same sitting as they give their post-gastrostomy opinion, and the counts of studies are not counts of independent cohorts, since two reports describe the same Oxford Feeding Study cohort and one study's record and survey subgroups may overlap by its own authors' statement. In the mortality domain, all-cause mortality is reported far more often than procedure-attributable mortality, attribution is made by the original authors rather than by any independent adjudication, and only three studies had a non-gastrostomy comparator, of which one found reduced short-term mortality and one found increased mortality, so the absence of procedure-attributable death across this literature is reassuring but not equivalent to a demonstrated absence of excess risk. Older series describe surgical and perioperative practice that has since changed, and cohorts recruited over periods of up to two decades are additionally vulnerable to secular change in surgical technique and perioperative care within a single study. Functional severity was rarely classified with a standardised instrument such as the GMFCS, and several studies grouped heterogeneous neurological diagnoses into one composite category without disaggregating by aetiology, so residual confounding by severity and by diagnosis cannot be excluded. Reporting bias is likely and could not be assessed formally: complication data in particular are more likely to be reported when notable, and several studies recorded no complications in a manner that does not distinguish absence of events from absence of reporting.

The review process has its own limitations. The search was restricted to five sources and grey literature was not searched systematically beyond conference abstracts retrieved by those searches, so eligible unpublished or non-indexed studies may have been missed. Non-English publications were excluded at full-text screening, and 15 otherwise potentially eligible reports were lost on that basis, which risks language bias in a field with substantial European, Latin American and East Asian output. Screening was duplicated throughout by two independent reviewer teams, but extraction was distributed across five reviewers rather than duplicated, so single-extractor error is possible. Two instances found during central checking illustrate the risk: one included study reports individual complication rates that sum to more than its own stated total, and one Covidence record was extracted twice with directly contradictory mortality data that could not be reconciled, so that study is not counted in either direction in the mortality synthesis. Cohort overlap was identified from institution, period, sample size and authorship rather than by contacting authors, so residual double-counting cannot be excluded, and one included study explicitly reports two subgroups whose overlap the original authors state cannot be quantified. Study authors were not contacted for missing or ambiguous data. Risk of bias was assessed for 102 of the 120 included studies, with 18 being conference abstracts and therefore not eligible for appraisal.

The synthesis method also constrains interpretation. Direction-of-effect vote counting was chosen because outcome-definition heterogeneity made pooling inappropriate, but a count of studies does not weight for sample size, precision, or quality. Therefore, a domain in which forty small, favourable series outvote three large, null datasets should not be read as forty-to-three evidence. To address this limitation, future research could use quality-weighted or Bayesian evidence synthesis methods. Future systematic reviews could also focus their main conclusions on high-quality propensity-matched or administrative cohorts, especially for outcomes like respiratory infections. This would mirror the Cochrane approach of separating higher- and lower-quality non-randomised studies. Until then, readers should treat every directional conclusion in this review as qualitative rather than quantitative.

4.3 Conclusions

Gastrostomy placement in children with neurological impairment and feeding difficulty is associated with a consistent and clinically meaningful improvement in weight-based nutritional status, and the improvement is greater when the procedure is performed earlier. Serious surgical harm is uncommon when gastrostomy is performed alone. Because of the added surgical risks, concurrent antireflux surgery or jejunal extension should be used cautiously. Following AAP guidance, these additions should be reserved for cases of severe, symptomatic GERD or persistent aspiration.

Gastrostomy should not, however, be presented to families as a means of preventing aspiration or respiratory infection. The largest and most rigorously controlled datasets show no reduction in pneumonia-related hospitalisation, and in some analyses an increase, and this discordance with the smaller before-and-after cohorts is systematic enough to warrant caution in how the procedure is discussed at the point of consent.

The evidence supports a caregiver benefit. The best-evidenced quality-of-life finding in this review is that caregiver health-related quality of life improves and caregiver worry about the child's nutrition falls substantially, measured with a validated instrument against population norms. The corresponding claim about the child cannot be made: it has been tested once with a validated instrument and a baseline, and the total score did not change.

Two further points should be communicated to families: de novo reflux is hypothesized to occur in a subset of previously reflux-free children, while reflux indices may improve in some children with pre-existing disease. As such, the clinical conversation should differ according to the child's pre-operative reflux status; most new reflux is managed medically, and fewer than one child in ten goes on to need antireflux surgery, with a seizure disorder the strongest predictor of those who will. And while mortality in this population is high, it reflects the underlying disease rather than the operation: deaths attributable to the procedure are rare enough to be counted individually across the whole literature, thirty-day mortality is 0% to 3% outside the most fragile diagnostic groups, though the two propensity-matched comparisons available disagree: one found short-term mortality halved, the other found it more than doubled, so this review cannot determine whether gastrostomy raises, lowers, or has no effect on survival.

The most important conclusion of this review is negative: there is no rigorously captured data on how gastrostomy impacts a child's cognition, motor function, or communication. To address this evidence gap, future studies need a different approach. Prospective cohorts should use standardized, validated pediatric developmental tools (like the WeeFIM or CP-QOL) and report outcomes stratified by GMFCS levels to adjust for baseline severity. These studies also need clearly defined developmental endpoints and follow-up periods long enough to capture real changes in cognitive and physical development.

Ultimately, these findings highlight the importance of shared decision-making. As emphasized by AAP guidelines on nonoral feeding, clinicians should work with families to weigh the clear benefits of gastrostomy (like nutritional recovery and improved caregiver quality of life) against the remaining clinical uncertainties (like the lack of respiratory protection and unknown neurodevelopmental impacts). By using clear data on surgical safety and the importance of early placement, clinicians can help families make more informed and realistic decisions.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fped.2026.1963093/full#supplementary-material

Supplementary material Supplementary material at the source

Footnotes

Edited by: Yakun Liu, Soochow University, China

Reviewed by: Marco Di Mitri, University of Florence, Italy

Fabiola Cassaro, University of Messina, Italy