Oral nutritional interventions in frail older people who are malnourished or at risk of malnutrition: a systematic review

Article history

The research reported in this issue of the journal was funded by the HTA programme as project number NIHR128729. The contractual start date was in December 2019. The draft report began editorial review in August 2021 and was accepted for publication in May 2022. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HTA editors and publisher have tried to ensure the accuracy of the authors’ report and would like to thank the reviewers for their constructive comments on the draft document. However, they do not accept liability for damages or losses arising from material published in this report.

Copyright statement

Copyright © 2022 Thomson et al. This work was produced by Thomson et al. under the terms of a commissioning contract issued by the Secretary of State for Health and Social Care. This is an Open Access publication distributed under the terms of the Creative Commons Attribution CC BY 4.0 licence, which permits unrestricted use, distribution, reproduction and adaption in any medium and for any purpose provided that it is properly attributed. See: https://creativecommons.org/licenses/by/4.0/. For attribution the title, original author(s), the publication source – NIHR Journals Library, and the DOI of the publication must be cited.

2022 Thomson et al.

Malnutrition and frailty in older people in the UK

The UK population is ageing rapidly. The proportion of population aged ≥ 65 years is set to increase from 12% in 2016 to 18% by 2041 and further, to 26%, by 2066, with the fastest growth expected in the ≥ 85 years age group. 9 Ageing is associated with increased risk of multimorbidity10 and disability,11 which represents a major challenge for future health and social care service provision and funding. 12 There is a critical need to identify effective interventions to mitigate age-related morbidity in populations who are likely to benefit most. Chronic undernutrition or malnutrition is an important contributor to morbidity and mortality in older adults and is amenable to treatment, thereby providing a potential target for intervention.

Malnutrition is the deficiency of energy, protein, vitamins and minerals that causes weight loss, muscle loss and functional limitations,7 and it is common among older adults aged ≥ 65 years. Although malnutrition affects < 10% of independent community-dwelling older adults,13 prevalence is much greater in settings where there are increased care needs. 14 National surveys have detected malnutrition in 28% of hospital admissions, 27% of residential care home residents and 41% of nursing home residents. 15 Malnutrition has serious adverse consequences, including physical decline, and poorer outcomes of diseases and increased complications, such as infections, delayed recovery, hospital readmissions, increased length of hospital stays, more general practitioner visits, and poor quality of life (QoL) and well-being. 2,16

Frail older people are at a particularly high-risk of malnutrition and are three to four times more likely to be malnourished. 3–6 Frailty is conceptualised as an abnormal health state relating to loss of biological reserves causing increased vulnerability to small environmental or health changes, which can lead to disability, falls, long-term care, hospital admissions and mortality. 17,18 Different tools have been used to measure or operationalise frailty, such as the Fried frailty phenotype and the cumulative deficit model. 19,20 Around 1 in 10 people aged > 65 years and around one-quarter to half of people aged > 85 years are living with frailty. 17 Malnutrition and frailty are closely interlinked. Poor nutritional status and weight loss increase the risk of frailty,18,21 and the presence of malnutrition further worsens the health status of frail older people. 4 Nutrition supplementation is recommended as one of the mainstays of intervention in treating frailty (European Society for Clinical Nutrition and Metabolism); however, much of the evidence is based on short-term protein synthesis studies or micronutrient interventions (e.g. amino acids, omega 3, vitamin D) that have not shown consistent benefits on muscle mass and function. 21 Furthermore, a micronutrient treatment approach is unlikely to benefit malnutrition and broader clinical and functional outcomes that are important in frailty.

Description of current service provision

The National Institute for Health and Care Excellence (NICE) CG32 guidelines recommend that health-care professionals consider oral nutrition support to improve nutritional intake for people who can swallow safely and are malnourished or at risk of malnutrition. 7 The guidelines states that oral nutrition support includes any of the following to improve nutritional intake: food fortified with protein, carbohydrate and/or fat, plus minerals and vitamins; snacks; ONS; altered meal patterns; and dietary advice. Dietary advice is recommended (e.g. meal planning, adding nutrients to meals) for older adults at risk of malnutrition, while powdered or liquid supplements (ONS) can be prescribed to those with existing malnutrition or at high risk of developing malnutrition. 7 The cost-effectiveness of these interventions is also unknown. Evidence from reviews so far suggest that prescribed ONS is effective in reducing malnutrition and its consequences, such as delayed wound healing and infections. 22 ONS is often viewed as a mode of managing malnutrition when it is difficult for individuals to consume energy and/or nutrients from food, for example in the case of acute illness or lack of availability of food. 22 Systematic reviews have also reported the cost-effectiveness of ONS in the management of malnutrition. 23–25 Cost-effectiveness evidence suggests that the use of ONS in community settings can reduce hospital stays and admissions (estimated savings of ≥ £119,200 per 100,000 people). 2 However, a key research gap, highlighted in current guidelines, is evidence specifically among frail older people on oral nutritional interventions that are effective in reducing malnutrition.

Individual study findings are not, however, entirely consistent for clinical outcomes, probably because of differences in the type of ONS evaluated and study methodology. 26 Evidence is mainly derived from small trials conducted in heterogeneous populations and across health-care settings. Some reviews have included only hospital patients post surgery,27,28 whereas others have focused on community-dwelling adults29 and mixed populations;25,30,31 this makes it difficult to draw conclusions about the effectiveness of ONS for high-risk populations such as frail older people.

A further gap in knowledge is whether or not prescribed ONS offer additional benefits above other oral nutrition support strategies such as fortified food or expert dietary advice. Dietary counselling is often the first means of nutritional interventions in practice. 32 This includes supporting older people with planning their diet and making meal plans and is delivered by dietitians in the community or in hospitals. Food fortification, including adding specific nutrients (e.g. vitamins, proteins) to the diet, is another form of oral nutritional support. 33 However, although ONS have also been shown to be cost-effective, the costs of other forms of nutritional support, including dietary advice, food snacks and food fortification, to manage malnutrition remain unclear and need to be elucidated. 22 In addition, reviews so far have mostly compared ONS with routine care (i.e. no nutritional support), not necessarily with dietary advice. 22,24,27

In summary, much of the focus of previous reviews on oral nutritional interventions includes disease-related malnutrition and adult populations aged ≥ 18 years, and not frail older people specifically. 31,34–37 Many of these reviews and studies have mostly looked at interventions to treat malnutrition related to diseases [e.g. cancer or human immunodeficiency virus (HIV)] and after surgery, which will have different underlying mechanisms from malnutrition in frail older people. The evidence in current guidelines is also mostly from studies on disease-related malnutrition. 38,39 As noted by topic experts in the NICE CG32 guidelines,7 there is a lack of emphasis on effective interventions to reduce malnutrition among frail older people.

Determinants of malnutrition in frail older people: understanding factors affecting adherence to and acceptability of interventions

The effective treatment and management of malnutrition should be tailored to meet the needs of frail older people. Malnutrition is multifactorial. In addition to comorbidities, several other factors may affect the nutrition of older people. These include physiological changes with ageing (loss of appetite, poor taste and smell, disability), psychosocial aspects (social support, resilience, lack of knowledge about food) and personal resources (poverty, inability to shop for food). 40–43 These factors then lead to slower eating and lack of diet variety, which in turn lead to poor dietary intake (low energy, protein, and key nutrients such as B-vitamins, vitamin A, vitamin C, iron, calcium, zinc), potential weight loss and, ultimately, a state of malnutrition. 32

Issues of compliance and acceptability also play a crucial role in inadequate nutritional support. 22,32,44 Although ONS have been found to be effective, the uptake of and compliance with them can be poor. The taste, texture, temperature and mode of ONS (liquid, powder) all influence the extent to which ONS are consumed, particularly over prolonged periods of time. For example, change in energy density can improve compliance and uptake of ONS. 22 Similar issues of compliance are also relevant for dietary advice and counselling to ensure that diet plans are acceptable and sustainable over time. The delivery and implementation of nutritional support by clinicians and healthcare professionals can also be very variable. 44 This could be due to lack of consistency in guidelines on whether ONS with or without dietary advice is effective in older people. 44 Clinical practice has been reported by dietitians to be influenced by lack of knowledge, ease of implementation, published research and local departmental protocols. 44 Understanding ways to improve the adoption and implementations of evidence-based nutritional support interventions into routine practice is a particular gap in the existing evidence.

The initial logic model developed prior to the review drew on current evidence and feedback from the preparatory patient and public involvement/engagement (PPIE) work that was undertaken (Figure 1). During the project, the logic model was iteratively refined with emerging findings along with input from stakeholders to produce a final logic model.

FIGURE 1.

Initial logic model.

Aims and objectives

The aim of the study is to evaluate the effectiveness and cost-effectiveness of oral nutritional interventions in frail older people who are malnourished or at risk of malnutrition.

The research objectives are to:

• systematically review the effectiveness and cost-effectiveness of oral nutritional interventions which include ONS in frail older people who are malnourished or at risk of malnutrition

• identify components of interventions that are associated with increased effectiveness or adherence

• systematically review qualitative studies to assess issues related to acceptability of ONS among frail older people who are malnourished or at risk of malnutrition

• undertake economic modelling to identify the cost-effectiveness of different models of oral nutritional interventions in frail older people who are malnourished or at risk of malnutrition

• refine and develop a logic model for oral nutritional interventions (including determinants, components and outcomes) to reduce malnutrition in frail older people

• collate findings and consult with stakeholders to identify (1) recommendations for interventions with potential for testing in future research and (2) implications for practice and policy.

Search strategy

The search strategy was initially developed in MEDLINE combining the concepts frail older people and nutritional support. Search terms, both text words and subject headings, were identified by an information specialist in conjunction with the project team. Articles previously identified by scoping were also used to identify relevant terms. Population terms included those relating to age, frailty, or care/nursing home settings. Nutritional support included ONS, food fortification, dietary support and malnutrition prevention. Results were restricted to human studies and those published in English. No geographic filters were applied. Publication filters were also not used as a range of publication types were relevant, which allowed the same set of papers to be screened for the cost-effectiveness review.

The searches were run on 26 and 27 February 2020, with updates conducted on 13 September 2021 (see Appendix 1). In total, 11,753 articles were retrieved; these were exported to EndNote (Clarivate Analytics, Philadelphia, PA, USA) reference management software and duplicate records were removed. Following this, 8428 records remained and were exported to Covidence (Melbourne, VIC, Australia) for screening. The databases searched were Ovid MEDLINE^® and Epub Ahead of Print, In-Process & Other Non-Indexed Citations, daily and versions^®, Ovid EMBASE 1974 to 2020 September 13, EBSCOhost CINAHL, Scopus and Cochrane Library (CDSR and CENTRAL); all databases were searched from inception.

Grey literature searching encompassed a range of sources, including OpenGrey, NHS EED (NHS Economic Evaluation Database), DARE (Database of Abstracts of Reviews of Effects), HTA, IDEAS/REPeC (https://ideas.repec.org), HMIC (Healthcare Management and Information Consortium), ASPEN (American Society for Parenteral and Enteral Nutrition), BAPEN (British Association for Parenteral and Enteral Nutrition), ESPEN (European Society for Clinical Nutrition and Metabolism), European Natural Health Alliance, Canadian Malnutrition Task Force, United Kingdom Malnutrition Task Force, as well as trial registries, conference abstracts, theses and charities (659 unique resources were identified for screening). Finally, reference lists of all included studies and citations including relevant systematic reviews were screened for inclusion.

Inclusion and exclusion criteria

Deviations from the protocol

In the protocol, ‘change in nutritional intake’ was a measure of malnutrition. However, we changed this to just energy and protein during the extraction process. Similarly, ‘serious adverse events’ were not defined in the protocol but were later defined during the extraction process as kidney injury, hyperglycaemia, constipation, diarrhoea, nausea, vomiting, refeeding syndrome and micronutrient deficiency. In addition, we altered our definition of frailty to encompass more than standard measures, using the following proxy measures: hospitalised for a fall, any fracture or an emergency orthopaedic admission at the time of recruitment to the study; or permanently residing in a care or nursing home.

Selection of studies

Three reviewers (OA, EJ, CM) screened all title and abstracts identified by the search using Covidence. Full texts of potentially eligible studies were sought and then screened. Any disagreements were resolved by a third reviewer (CM, KT, SER). Where multiple reports of the same study were identified, we combined these into a single study to extract and analyse these at study level (see Appendix 2).

For the cost-effectiveness review, one reviewer (WM) screened the title and abstracts of the studies retrieved by the search in Covidence. For studies deemed eligible or for which it was impossible to decide eligibility from the abstract, the full text was retrieved, and two reviewers (WM, SR) independently assessed the full text for inclusion. This was conducted alongside the study selection of effectiveness studies. Two reviewers (SR and WM) made the final selection decisions about the included studies.

Data extraction

A data extraction form was created and piloted on 10% of included studies. Based on this piloting, the form was modified appropriately (e.g. introduction of the TIDieR framework47 for reporting interventions). One reviewer extracted 50% of included studies, with a second extracting the other 50% (OA and EJ). The reviewers then checked each other’s data extraction. Any disagreements between the two reviewers were resolved by arbitration to a third reviewer (LT). For the cost-effectiveness review, one reviewer (WM) extracted 100% of included studies, with a second (SR) checking the data extracted. Any changes suggested by SR were discussed and agreement was reached.

Quality assessment of included studies

The Cochrane RoB 1.0 tool was used to assess parallel-arm, crossover and cluster-RCTs. 48 The following domains were assessed: random sequence generation; allocation concealment; blinding of participants and personnel; blinding of outcome assessment; incomplete outcome data; selective reporting; and other bias. One review author (of OA or EJ) assessed the risk of bias for each included paper. A second reviewer (either OA or EJ) checked the assessment. Any discrepancies between the two reviewers were adjudicated by a third reviewer (LT). A tool for non-randomised studies was not needed because no studies of this type met the eligibility criteria. The quality of the included cost-effectiveness studies was assessed using the BMJ checklist. 49

The Grading of Recommendations, Assessment, Development and Evaluation (GRADE) methodology was used to address the quality of the evidence. 50 Quality of evidence for each outcome was assessed based on study design, risk of bias, imprecision of estimates, inconsistency of results from different studies, indirectness of study results (i.e. lack of applicability) and publication bias. 51 The GRADE approach was used to assess the certainty of evidence for all outcomes (where possible) using the principles outlined in the Cochrane Handbook. 52 Two review authors (EJ and OA) independently assessed the certainty of evidence for each of these outcomes, resolving any disagreements by discussion and, if necessary, through arbitration with a third review author.

Meta-analysis and narrative synthesis

Meta-analysis was undertaken for RCTs with outcomes where at least two studies compared rates of an outcome (for binary variables) or mean values (for continuous variables) between persons receiving ONS (intervention recipients) and those who received SC (the control group). All studies included in the systematic review that had outcome measures that could be combined were included in meta-analyses if they met the following criteria:

• The required data were reported or calculable (mean and SD for continuous variables, number of events and sample size for binary variables).

• Trial arms included ONS versus SC as defined by triallists (e.g. the study by Parsons et al. 53 was excluded as it lacked an appropriate ‘SC’ arm).

• Outcome measures from different studies could be combined (it was deemed that data for quadriceps strength were inappropriate for inclusion in a meta-analysis in which all other studies reported handgrip strength).

Subgroup analysis was implemented on those studies deemed adequately randomised. Adequacy of randomisation was assessed using domain 1 (random sequence generation – selection bias) from the RoB 1.0 tool. The reasons for this approach included the following: the studies were expected to be at varying risk of bias across categories (there was a low expectation of finding studies at low risk of bias across all categories); random sequence allocation and allocation concealment were considered the most important items, especially as most of the outcome measures were not considered to be subjective (ADL is an exception to this); and we sought to minimise the number of sensitivity analyses. The studies included in the adequately randomised sensitivity analysis are not necessarily at low risk of bias.

Studies were deemed to be adequately randomised where a random component (e.g. using a computer random number generator) was used in the sequence generation process. Five out of 11 studies that were included in the meta-analyses were deemed inadequately randomised. 54–57 No meta-regression analyses were conducted to investigate the effect of variation in population characteristics and intervention components across studies due to the small number of studies included in the review for each outcome.

There were a mix of studies reporting final values and change from baseline (CFB) values. CFB outcomes were preferred as they remove a component of between-person variability from the analysis. 58 Sensitivity analyses were also conducted using final values. CFB, final values and standard deviations were calculated where they were not reported. The methods used to determine the standard deviation are described in Appendix 3.

For the binary outcomes (mortality and hospitalisation), we performed an analysis comprising all studies that reported relevant data. A Mantel–Haenszel random-effects meta-analysis was conducted. For continuous outcomes with a uniform measure across studies, an inverse variance random-effects meta-analysis was conducted for the mean difference in outcomes. For continuous outcomes with different measures across studies, standardised mean differences (SMDs) were calculated using the Hedges’ g (adjusted) method. 59 Generic inverse variance random-effect meta-analyses were conducted. In reporting the results, statistical significance was defined at a 95% level of confidence.

As multiple measures of the same outcome were often included in a study (e.g. calf circumference as a measure of fat-free muscle mass), an evidence hierarchy was employed to decide which outcome was preferentially included in the analysis. This is displayed in Table 1.

Data from the longest follow-up time point available in each were included in the meta-analysis so that the longer-term impacts on outcomes could be assessed. In addition, too few studies reported multiple time points, meaning that it would not have been possible to run a meta-analysis for multiple follow-up periods. The degree of heterogeneity was estimated using the I² statistic, and the p-value of the chi-squared statistic was used to measure the strength of evidence for heterogeneity. I² values of 0–40% (heterogeneity might not be important), 30–60% (may represent moderate heterogeneity), 50–90% (may represent substantial heterogeneity) and 75–100% (considerable heterogeneity) were used to guide interpretation. 60 Publication bias and other small-study effects were evaluated using Egger’s test and funnel plots if 10 or more studies were included in an analysis. If there were fewer than 10 studies, the power of the test would usually be too low to distinguish real asymmetry from chance. 61 All analysis was conducted in RevMan.

Narrative synthesis methods were used either to analyse outcomes with insufficient data or for those studies that did not meet the criteria for meta-analysis (e.g. cohort studies). Patterns in the data, including statistical significance and direction of effect, are summarised narratively. The results reported are included alongside the meta-analysis outcomes. A narrative synthesis was undertaken for the cost-effectiveness review to describe the similarities and differences in the study questions, methods and results.

Network meta-analysis

There were multiple comparators investigated in the studies included in the systematic review. The effectiveness of ONS compared with these was evaluated using network meta-analysis (NMA). NMA enables direct and indirect evidence of a treatment effect to be combined in the estimation of the effect. For example, if one study (AB) compares A with B, one study (AC) compares A with C and one study (BC) compares B with C, then study AC provides direct evidence for A compared with C, and studies AB and BC provide indirect evidence for A compared with C. NMA also enables an effect to be estimated for A compared with C when only indirect evidence is available.

Only RCTs were included in the NMA. All interventions included in the studies that met the inclusion criteria were included in the NMA, for example different dietary interventions and dietary interventions with exercise. The purpose was to estimate the effectiveness of ONS compared with all of the different comparators found in the review studies. The network diagrams are presented in Appendix 4 (see Figures 19 and 20) and show that there are no cases of both direct and indirect evidence for any one comparison. The purpose of conducting NMAs here is to estimate treatment effects using indirect evidence. The effectiveness of every treatment compared with every other treatment can be estimated. The effect estimates for ONS compared with every other treatment are produced here. The mean and 95% credible interval of the effect estimates are calculated.

A NMA was conducted for an outcome for which there were at least three studies reporting one comparison, generally ONS compared with SC, and there was a connected network of three or more interventions. These conditions were met for two continuous outcomes. One outcome was analysed on the SMD scale and one outcome was analysed on the mean difference scale. 58,62 Analyses were conducted in WinBUGS 1.4.3 (MRC Biostatistics Unit, Cambridge, UK). 63 For the mean difference analysis, the WinBUGS program 5a code for a random-effects analysis with multiarm trials from the NICE Technical Support Document 2 was used. 64 For the SMD analysis, the WinBUGS program 7a code for random-effects analysis with multiarm trials from the NICE Technical Support Document 2 was used. The code requires that the data set include the variance of the baseline treatment in each trial with more than two trial arms. For the SMD analysis, the variance of baseline treatment was approximated as shown in the following equations. The equation presented here for the variance of baseline treatment was not specifically reported in Introduction to Meta-analysis,62 Chapter 4, but it makes use of formulae 4.18, 4.19, 4.20, 4.22 and 4.24 presented there:

µ_C, SD_C, and N_C are the mean, standard deviation and sample size of the control group; µ_T, SD_T and N_T are the mean, standard deviation and sample size of the intervention group.

A common between-study variance was assumed across treatment comparisons. Multiple studies were reported for only one treatment comparison, ONS compared with SC, so the common between-study variance estimate is determined by those studies. For continuous outcomes, the between-study variance and standard deviation are on the outcome scale. The choice of prior distribution for the between-study standard deviation should be based on the specific scale. Where there are many trials with which to estimate the between-study standard deviation, the upper limit of the uniform prior distribution should be sufficiently high that the upper end of the posterior distribution of the between-study standard deviation is barely, if at all, truncated.

Where there are few studies with which to estimate the between-study standard deviation, the uniform prior distribution can have a significant effect on the posterior distribution. The mean of a uniform prior distribution is (maximum – minimum)/2, and it is not entirely ‘uninformative’. One approach is to identify an informative prior from a published meta-analysis that does not include the same trials as the current study, or to elicit a prior distribution from experts. For the analyses planned, there were four different outcome scales across the analyses and few studies in each analysis. Therefore, a pragmatic decision taken here was to set the upper limit of the uniform distribution for the between-study standard deviation to be the difference between the greatest and smallest effect size for any one comparison in the network (only two analyses were eventually included in the review). For example, comparison A versus B has estimates (–0.4, –0.8, 0.3) and comparison B versus C has estimates (0.6, 0.1). The greatest difference in effect sizes is 0.3 minus –0.8 = 1.1. This is straightforward when there are no comparisons with direct and indirect evidence, as in this study. For networks with direct and indirect evidence, the difference in these estimates would need to be taken into account. The mean (0.55) of the uniform distribution (0 to 1.1) is the maximum possible between-study standard deviation described by the mean effect estimates. But these priors are not as vague as would normally be recommended. Recommended vague priors allow for a huge range of true effect estimates,64 far greater than seen in practice. A review65 of between-study variance estimators reported that a Bayesian approach may overestimate the between-study variance when it is close to zero and when there are few studies.

Convergence was assessed using the Brooks–Gelman–Rubin diagnostic along with a visual inspection of the trace and density plots. 66 The initial 20,000 simulations were discarded, and the results were based on a further sample of 50,000 simulations. As there were no closed loops in the network (no cases of direct and indirect evidence for any one comparison), there was no possibility of inconsistency in the network. The probability that an intervention was most effective was then estimated.

Public and patient involvement/engagement

Public and patient involvement/engagement was undertaken throughout the project, initially helping to develop the proposal and inform the initial logic model, and then scope of the review, discussing results of the review and the implications of findings. In addition, AR (PPIE lead) helped shape the research as part of the project team. The PPIE groups comprised six to eight older people (all of whom were female).

The participants in the PPIE group were members of the Newcastle branch of the Elders Council, a local organisation of older people interested in sharing their views about making the city ‘a great place in which to grow old’. Recruitment to the focus groups was organised by the chairperson of the Elders Council in Newcastle and PPIE lead (AR), and the sessions were facilitated by researchers at Newcastle University. The format of the sessions was a short presentation about review progress to date, followed by open questions to discuss as a group. The online sessions were recorded, and detailed notes were taken by researchers. These notes were subsequently written up and shared with the research team. Key concepts and broad themes were identified and used to complement the data collated in the review.

Following the completion of the review, the findings were presented to a panel of practice or policy partners to allow understanding of how different stakeholders conceptualised the results and their experiences more generally concerning the use of oral nutritional interventions in this population. The main online event comprising eight stakeholders was supplemented with three one-on-one sessions with additional partners to ensure that we collated views from a range of individuals. Geriatricians, dietitians and nurse practitioners were involved in the discussions.

Characteristics of included studies

Eleven studies were included in the effectiveness review, all of which were RCTs. One study was a crossover RCT,56 four studies were multiarm RCTs57,67–69 and the remaining six studies used a parallel-group design with two groups. 53–55,70–73 In total, 882 people were recruited across the 11 studies. The smallest study recruited 39 participants56 and the largest recruited 104. 53 Table 2 describes the characteristics of the included studies in the review.

Two studies took place in Australia,68,71 two in France,57,67 one in Germany,72 two in Sweden,56,69 and one each in the UK,73 Russia,70 Canada55 and Taiwan (Province of China). 54 Five studies were set in nursing/residential homes,54,56,57,67,73 of which four took place in multiple nursing homes. 56,57,67,73 We acknowledge that definitions of nursing/residential homes vary internationally; however, our groupings were purely for descriptive purposes. Two further studies were set in the community55,69 and three were set in hospital. 68,70,71 One study stated that it had been conducted with patients after they had been discharged from hospital. 72 Reporting on intervention duration and follow-up was often inadequate and lacked detail.

The type of ONS used and the comparisons varied across the studies. Six studies54,55,57,68,71,74 compared ONS with usual care or SC in one of their arms, and two53,72 compared ONS with dietary counselling or advice. The remaining studies contained a number of comparisons; these are detailed in Table 2. One study68 either combined or compared ONS with exercise programmes. The duration of the ONS intervention ranged from 28 days74 to 6 months,69 with a maximum follow-up of 12 months. 69 The timing of follow-up, particularly in relation to the intervention period, was poorly reported and difficult to ascertain from the studies. Of the studies that were included in the effectiveness review, six were either fully funded or part-funded by industry. Of these, four were fully funded (including one with an unrestricted grant) and two were part-funded. A further three were not funded by industry and two studies did not include details of funding/conflict of interests.

Across the studies, participants varied in age, BMI and body weight. Most studies included both men and women, although often more women participated in the studies than men. Two studies included women only. 69,71 Participants’ level of malnutrition at baseline between groups was measured using a variety of tools. Three studies used the MNA54,57,67 and one used the MUST score. 53 One study each used mid-upper arm circumference and albumin levels,71 and a further two studies reported excess weight loss. 55,70 One study stated that most participants were at risk of malnutrition, but it was unclear whether this was assessed using the MNA-SF. 56 Four studies did not report specific levels of malnutrition between groups at baseline. 68–70,72 Further details of the characteristics of participants in the included studies can be found in Appendix 5.

Some studies reported on comorbidities that may contribute to malnutrition (see Appendix 6). Two studies53,67 included participants with dementia, three53,56,68 included participants who had cognitive impairment, and one53 included participants who had cardiovascular disease. No studies reported on participants who had diabetes, stroke or cancer. Only one study57 reported on specific oral health issues that may affect malnutrition and related outcomes. One study57 reported on participants who had complete or partial denture or participants who had no dentures. Two studies54,67 reported on participants’ need for assistance with feeding.

In general, studies did not report clearly on potential social determinants of malnutrition, with the exception of their living arrangements or whether the participants were receiving household or other help (see Appendix 7). No studies reported on the ethnicity of the participants. Of the five studies that did not take place in either nursing or residential care homes, four55,68,71,75 reported the participants’ living situation to some extent (e.g. living along, married). Luo et al. 74 did not describe living arrangements. Tidermark et al. 69 noted that their participants lived at home and were non-institutionalised but did not provide any further details. Otten et al. 72 reported the number of participants living alone, but not by study arm. The interventions, comparisons and modes of delivery across the studies varied and are described in Appendix 8.

Quality assessment of included studies

Quality assessment was conducted using the Cochrane risk-of-bias tool (n = 11) for parallel-arm RCTs, multiarm RCTs, crossover RCTs or cluster-RCTs. A summary of risk-of-bias assessments across all included studies can be seen in Appendix 9.

Randomised controlled trials

We assessed the 11 included studies using the Cochrane risk-of-bias tool. Fewer than half of the RCTs were judged to be at low risk of bias for random sequence generation (45%),55,57,67,69,73 allocation concealment (45%),56,68–70,73 blinding the outcome assessor (45%)54,55,57,68,73 and selective reporting (45%). 55,57,67,69,73 Forty-five per cent of studies were judged to be at high risk of performance bias. 57,67,69–71 Thirty-six per cent of RCTs were judged to be at high risk of attrition bias,56,57,67,70 with 27% of RCTs also judged at unclear risk for this domain. 55,68,72 Most of the included RCTs were at unclear risk of other bias (64%). 56,57,67–69,72,73 Figure 3 shows the assessments across studies for each domain.

FIGURE 3.

Risk-of-bias assessments across studies for each domain.

Random sequence generation

One included study67 was assessed as being at high risk of bias for random sequence generation. Five studies54–57,72 did not provide enough detail about their method of randomisation and so were assessed as unclear risk of bias. Five studies53,68–71 were assessed as being at low risk of bias for this domain.

Allocation concealment

One included study71 was rated as being at high risk of bias for allocation concealment. Five studies54,55,57,67,72 were rated as unclear. Five studies53,56,68,69,74 were assessed as being at low risk of bias for this domain.

Blinding of participants and personnel

Five included studies57,67,69–71 were judged to be at high risk of bias for this. Four53,56,68,72 were assessed as being at unclear risk of bias, mainly because the methods of blinding were not clearly reported. Two studies54,55 were assessed as being at low risk of bias for this domain.

Blinding of outcome assessment

One study70 was deemed to be at high risk of bias for this domain. Five included studies56,67,69,71,72 were judged to be at an unclear risk of bias. Five studies53–55,57,68 were assessed as being at low risk of bias for this domain.

Incomplete outcome data

Four included studies56,57,67,70 were deemed to be at high risk of attrition bias. Three studies55,68,72 were judged to be at unclear risk of bias for this domain. Four studies53,54,69,71 were assessed as being at low risk of bias for this domain.

Selective reporting

Two studies70,72 were judged to be at high risk of selective reporting bias. Four studies54,56,68,71 were judged to be at unclear risk of bias. Five studies53,55,57,67,69 were assessed as being at low risk of bias for this domain,

Other bias

Three included studies67,70,71 were judged to be at high risk of other bias. Six studies53,56,57,68,69,72 were assessed as unclear risk of bias. Two studies54,55 were assessed as being at low risk of bias for this domain.

Summary of effectiveness results

Pairwise meta-analyses were undertaken to assess the effects of ONS compared with SC on the outcomes of interest in this review. Ten studies53–57,67–69,71,74 were included in the pairwise meta-analyses (see Appendix 10 for reasons why studies/outcomes were excluded from the meta-analysis). The meta-analysis results are presented alongside a narrative synthesis of the outcomes that were unable to be pooled. As fewer than 10 studies were incorporated into the meta-analyses for any outcome, it was not possible to use funnel plots and other tests for publication bias. Analysis was run using both final values and change of baseline. CFB analysis will be presented here (where possible); final value results are in Appendices 11 and 12.

The outcomes reported below are broadly split into three key categories, which correspond to the period over which the outcomes might be expected to induce a noticeable change. Nutritional intake outcomes and those that relate to visceral protein level (albumin) are presented first; these include total energy, protein and albumin. Following this, body composition outcomes are discussed (body weight, BMI, fat-free muscle mass, lean body mass). Then longer-term outcomes are reported (ADL, grip strength, hospitalisation, MNA, morbidity, mortality, QoL). Finally, other outcomes are narratively synthesised, including adverse events, reduction in falls and compliance. Owing to uncertainty in the duration of follow-up (and the small number of studies identified), meta-analysis was undertaken aggregating all follow-up time points together.

Nutritional intake outcomes

Six studies53,55–57,67,70 reported data on the effect of ONS on nutritional intake outcomes. A meta-analysis was possible for energy and protein intake.

Energy (kcal) intake

Four studies reported data on the effect of ONS compared with SC on kilocalories (kcal) consumed for a CFB analysis; one55 was undertaken in the community and three56,57,67 were undertaken in care homes. All four studies55–57,67 measured energy intake in kcal, which refers to the energy from food consumption. One study reported data on energy intake at final visit but no data were reported at CFB. 53 The mean and standard deviation could not be calculated as there were insufficient data. This study was not included in the CFB analysis. 53 The follow-up time point, where reported, varied between 6 weeks,56 90 days/3 months,57 16 weeks55 and 60 days. 67 The pooled results of the meta-analysis (Figure 4a) show a positive effect of ONS versus SC on energy intake (SMD 1.02, p = 0.002, 95% CI 0.15 to 1.88). There was evidence of statistical heterogeneity (p < 0.0001, I²=87%). A sensitivity analysis could not be run as there were no adequately randomised studies. GRADE scores showed very low-quality evidence for energy intake (see Appendix 13).

FIGURE 4.

Forest plots of nutritional intake outcomes (CFB). (a) Energy intake; and (b) protein intake.

Protein intake

Four studies55–57,70 reported data on the effect of ONS compared with SC on protein for a CFB analysis. One study53 reported data on protein intake at final visit, but no data were reported on CFB. Insufficient data were reported in this study, and means and standard deviations at CFB could not be calculated. Two studies55,56 measured protein intake in grams (g). Three studies measured total protein intake in grams per day (g/day),55,56,67 while one study measured protein in g/kg. 57 The follow-up period varied: 6 weeks,56 60 days,70 90 days/3 months57 or 16 weeks. 55 The pooled result (see Figure 4b) of the meta-analysis of CFB scores comprising all four studies shows a slightly positive effect of ONS versus SC on protein (SMD 1.67, p = 0.05, 95% CI –0.03 to 3.37). The data show a substantial degree of statistical heterogeneity (p < 0.00001, I² = 97%). GRADE scores showed very low-quality evidence for protein intake (see Appendix 13).

Visceral protein level

Albumin

Five studies54,56,57,70,71 reported data on the effect of ONS compared with SC on serum albumin, measured using the analysis of serum derived from fasting blood samples in grams per litre (g/l) for a CFB analysis. Two studies70,71 were included in the sensitivity analysis as they had been adequately randomised. The meta-analysis results of the main analysis (Figure 5) show no evidence of effect of ONS versus SC on albumin (MD 1.48, p = 0.13, 95% CI –0.44 to 3.41). There was evidence of statistical heterogeneity (p < 0.00001, I² = 95%). The pooled results of the sensitivity analysis show a slightly positive effect of ONS versus SC on serum albumin (MD 2.86, p = 0.010, 95% CI 0.69 to 5.03). There was moderate evidence of statistical heterogeneity (p = 0.08, I² = 68%). GRADE scores showed very low-quality of evidence for albumin (see Appendix 13).

FIGURE 5.

Forest plot of albumin levels (CFB).

Body composition outcomes

Eight studies53–57,69–71 reported data on the effect of ONS on change in body composition outcomes. A meta-analysis was possible for body weight, BMI and fat-free muscle mass.

Body weight

Five studies54,56,69–71 reported appropriate data for inclusion in the meta-analysis of CFB scores between participants receiving ONS versus SC. Three studies69–71 were included in the sensitivity analysis as they had been adequately randomised. The pooled results (Figure 6a) of the main meta-analysis comprising all five studies showed no evidence of effect of ONS versus SC on body weight CFB (MD 1.31, p = 0.06, 95% CI –0.05 to 2.66). Substantial statistical heterogeneity was found (p = 0.004, I² = 74%) indicating a variation between sample estimates beyond what would be expected by chance when samples are derived from the same population. Three studies69–71 were included in the sensitivity analysis consisting of only adequately randomised studies. The pooled results (see Figure 6a) of the sensitivity analysis also indicate that there was no evidence of effect of ONS compared with SC on body weight (MD 1.28, p = 0.26, 95% CI –0.95 to 3.52). Similar to the main analysis, there was substantial heterogeneity (p = 0.01, I² = 78%) indicating the presence of a variable confounding factor across the studies. GRADE scores showed very low-quality evidence for body weight (see Appendix 13).

FIGURE 6.

Forest plots of body weight and BMI body outcomes (CFB). (a) Body weight; and (b) BMI.

Body mass index and proxy measures

Five studies54,56,57,67,71 reported appropriate data for inclusion in the meta-analysis of CFB scores between participants receiving ONS and those receiving SC. Two studies67,71 were included in the sensitivity analysis as they were adequately randomised. The pooled results (see Figure 6b) of the main meta-analysis comprising all five studies presented no evidence of effect of ONS compared with SC on BMI at CFB (MD 0.54, p = 0.06, 95% CI –0.03 to 1.11). There was evidence of statistical heterogeneity (p = 0.03, I² = 62%). The pooled results of the sensitivity analysis indicate a mixed effect of ONS compared with SC on BMI (MD 0.44, p = 0.54, 95% CI –0.82 to 1.71). There was significant evidence of heterogeneity (p = 0.01, I² = 84%), indicating that there may be a variable confounding factor across the studies. GRADE scores showed very low-quality evidence for BMI (see Appendix 13).

One study, by Lee et al. ,54 rated as being at unclear risk of bias for random sequence generation and allocation concealment, assessed the impact of ONS compared with SC on arm circumference, providing data at baseline and post intervention. 54 The authors reported a mean change in mid-arm circumference among people who were malnourished or at risk of malnutrition, at 24-week follow-up, of 0.3 cm in the intervention group and –0.8 cm in the control group. GRADE was unable to be assessed for arm circumference, as meta-analysis was not undertaken.

Fat free muscle mass

Three studies54,55,69 reported data for the effect of ONS versus SC on fat-free muscle mass for a CFB analysis. Calf circumference and lean body mass were the outcomes used to measure fat-free muscle mass. A hierarchy of outcomes was applied, and calf circumference was chosen as the preferred outcome. Two studies54,55 used calf circumference, measured in centimetres (cm), and one study69 measured lean body mass in kilograms using dual energy X-ray absorptiometry.

As the studies used different outcome measures, a SMD was calculated using Hedges’ g (adjusted) statistics (a measure of effect size) to standardise the different data across the three studies. Follow-up data were available for 12 weeks54 to 16 weeks55 and 12 months. 69 The pooled result (Figure 7) of the main meta-analysis of CFB scores comprising all three studies showed that the individual study estimates are inconsistent in the direction of effect (SMD 0.23, p = 0.34, 95% CI –0.24 to 0.69). There was evidence of heterogeneity in this analysis (p = 0.09, I² = 58%). The evidence of this analysis shows that there is a variable confounding factor across studies. A sensitivity analysis could not be conducted as there were no adequately randomised studies. GRADE scores showed low-quality evidence for fat-free muscle mass (see Appendix 13).

FIGURE 7.

Forest plots of fat-free muscle mass (CFB).

Longer-term outcomes

Eight studies53,56,57,67–71 reported data on the effect of ONS on longer-term outcomes related to malnutrition. A meta-analysis was possible for ADL, grip strength, MNA, mobility, hospitalisation and mortality. It was not possible to undertake a meta-analysis for outcomes on QoL; therefore, a narrative synthesis of the results for this outcome was undertaken.

Activities of daily living

Three studies57,71,74 reported data on the effect of ONS compared with SC on ADL, which refers to the ability to perform everyday tasks (or ‘activities of daily living’) as a measure of disability or level of physical functioning. One study71 measured ADL using the Barthel Index, which comprises 10 items in relation to which participants are assigned points, with a higher score indicating an increased ability to perform a task. Luo et al. 74 used a modified version of the Barthel Index. Van Wymelbeke et al. 57 used the Katz score for ADL,57 and compared ADL in participants who received supplements, those who received an alternative dietary intervention (brioche) and those receiving SC. Data from the supplement and brioche groups were combined and compared with the SC group in the analyses presented here. Data from the longest follow-up time available from each study were used in the analyses presented here. This varied across the studies, from 90 days/3 months in Van Wymelbeke et al. 57 to 4 months in the study by Cameron et al. 71 and 24 days in the study by Luo et al. 74

Post-intervention data were used for the meta-analysis, as CFB data could not be calculated for the study by Luo et al. 74 The pooled result of the main meta-analysis comprising all studies (Figure 8a) demonstrated no evidence of an effect of ONS compared with SC on ADL (SMD 0.30, p = 0.55; 95% CI –0.69 to 1.29).

FIGURE 8.

Forest plots of longer-term outcomes including (a) ADL (final values); (b) grip strength; and (c) hospitalisation (CFB). M–H, Mantel–Haenszel.

A sensitivity analysis in which the study by Van Wymelbeke et al. ,57 which was not adequately randomised, was omitted also showed no evidence of an effect of ONS compared with SC on ADL (SMD 0.68, p = 0.27; 95% CI –0.54 to 1.90). Substantial heterogeneity was present in the main analysis (I² = 89%) and in the sensitivity analysis (I² = 88%). GRADE scores showed very low-quality evidence for ADL (see Appendix 13).

Grip strength

Seven studies55–57,67–69,71 reported data on the effect of ONS compared with SC on grip strength. Five56,57,67,69,71 of these reported data for a CFB meta-analysis. Each of these five studies reported data assessing handgrip strength, our primary outcome measure. Several instruments were used to measure grip strength, including the Jamar Hydraulic Hand Dynamometer57,71 and the Harpenden(R) dynamometer,69 both of which use kilograms as measurement units; and the Martin vigorimeter,56 which measures handgrip strength using kilopascal. Data from two trial arms (one in which participants were provided with ONS and the other in which participants were given brioche) were combined and compared with the SC arm for the study by Van Wymelbeke et al. 57 In the study by Lauque et al. ,67 data were compared between participants who were at risk of malnutrition and received either ONS or SC. The longest follow-up time points across the five studies ranged from 3 to 12 months. The results of the pooled meta-analysis (see Figure 8b) comprising studies using CFB data indicated no evidence of an effect of ONS compared with SC on grip strength (SMD 0.17, p = 0.40; 95% CI –0.23 to 0.58). There was also no evidence of a difference for studies with adequate randomisation (SMD 0.27, p = 0.43; 95% CI –0.40 to 0.94). Substantial statistical heterogeneity was found in both analyses (I² > 50%), possibly reflecting variation in the follow-up times between studies.

Two studies were not included in the pairwise meta-analyses for this outcome (see Appendix 10). One of these studies from which data could be extracted68 reported an improvement in quadriceps strength (measured in kg using the Nicholas Manual Muscle Tester) among ONS recipients compared with people receiving SC when this was assessed on a non-injured limb (mean CFB scores were 6.5 in the ONS group and 4.8 in the SC group) but not when injured limbs were assessed (mean CFB scores were 2.3 in the ONS group and 2.7 in the SC group). GRADE scores showed very low-quality evidence for grip strength (see Appendix 13).

Hospitalisation

Five studies53,56,57,68,71 considered the impact of ONS on hospitalisation. All five reported data that were suitable for inclusion in the meta-analysis comparing ONS with SC on the number of hospital readmissions,68,71 our preferred measure, or admissions. 56,57 Van Wymelbeke et al. 57 compared hospital admissions between participants who received supplements, brioche or SC. The brioche and SC groups were combined into one intervention arm for the analyses presented here. Follow-up time points, where reported, varied from 6 weeks56 to 90 days/3 months. 57 Only two studies68,71 had been adequately randomised. The pooled result of the main value meta-analysis (see Figure 8c) comprising all five studies showed no evidence of an effect of ONS on hospitalisation (RR 0.97, p = 0.94, 95% CI 0.46 to 2.04). The pooled result of the sensitivity analysis of adequately randomised studies also showed no evidence of an effect of ONS on hospitalisation (RR 0.80, p = 0.59; 95% CI 0.35 to 1.82). Heterogeneity was not detected in either the main or the sensitivity analysis (I² = 0%). GRADE scores showed very low-quality evidence for hospitalisation (see Appendix 13).

Change in malnutrition

Two studies57,67 reported data for the effect of ONS compared with SC on MNA score, a validated screening tool for the assessment of malnutrition risk. A higher MNA score indicates that a person has a better nutritional status. Both studies reported appropriate data for inclusion in the meta-analysis of post-intervention scores between participants receiving ONS and those receiving SC (Figure 9a). A CFB analysis could not be performed, as one of the studies67 did not report baseline data that are required to calculate CFB scores. The 18-item MNA score was the outcome measure used in both studies. 57,67 Van Wymelbeke et al. 57 assessed MNA at 90 days/12 weeks, whereas Lauque et al. 67 assessed MNA at 60 days/8 weeks. The pooled results of the meta-analysis of post-intervention data provided no evidence of an effect of ONS compared with SC on MNA (SMD –0.36, p = 0.11, 95% CI –0.81 to 0.09). Low heterogeneity was detected between the studies (I² = 6%). Neither Van Wymelbeke et al. 57 nor Lauque et al. 67 was adequately randomised; therefore, a sensitivity analysis was not undertaken. GRADE scores showed very low-quality evidence for change in malnutrition (see Appendix 13).

FIGURE 9.

Forest plots of longer-term outcomes including (a) MNA (CFB); (b) mobility; and (c) mortality (final values). M–H, Mantel–Haenszel.

Mobility

Three studies68,71,74 reported data for the effect of ONS compared with SC on mobility, assessed using gait speed (in m/second) in two studies,68,71 albeit over different distances, and pace (in seconds/m in one study). 74 The data from the study that measured mobility using pace were converted to speed by dividing the number of metres walked by the average time taken. This ensured that all studies used the same outcome measure, with a larger number indicating a positive outcome, and that MD could be used for the analyses. All three studies were included in the meta-analysis, which analysed post-intervention (final value) scores as CFB data were unavailable. The longest follow-up time points were 4 months in Cameron et al. ,71 12 weeks in Miller et al. 68 and 24 days in Luo et al. 74 The pooled results of the main meta-analysis indicated a positive effect of ONS compared with SC (MD 0.03), which was statistically significant (p < 0.00001, 95% CI 0.02 to 0.04) (see Figure 9b). The results of a sensitivity analysis, including two adequately randomised studies,68,71 demonstrated no evidence of an effect of ONS versus SC on mobility (MD 0.02, p = 0.65, 95% CI –0.06 to 0.09). Statistical heterogeneity was not detected in the main or sensitivity analyses (I² = 0%). GRADE scores showed very low-quality evidence for mobility (see Appendix 13).

Mortality

Four studies53,68,69,71 assessed the effects of ONS on mortality and reported data that were suitable for inclusion in the pairwise meta-analysis (see Figure 9c) using final value analysis. Follow-up time points, where reported, varied from 12 weeks53,68 to 4 months71 (follow-up time was not reported in one of the studies69). All four studies had been adequately randomised and, therefore, a sensitivity analysis was not undertaken. The pooled result of the meta-analysis showed no evidence of an effect of ONS on mortality (RR 0.93, p = 0.90, 95% CI 0.28 to 3.06). There was no evidence of statistical heterogeneity (I² = 0%). GRADE scores showed very low-quality evidence for mortality (see Appendix 13).

Quality of life

Four studies55,68,72,73 reported on the effect of ONS on QoL; two72,73 reported overall QoL scores, and two55,68 reported data from psychological and physical subdomains of quality-of-life tools. Tidermark et al. 69 measured QoL using the EuroQol-5 Dimensions (EQ-5D) but reported only baseline data and so this study is not discussed here further. The results across the four studies reporting on the impact of ONS on overall quality-of-life and physical function domains were mixed,53,55,68,72 although one68 out of two studies55,68 reported a positive effect of ONS on psychological aspects of QoL compared with SC. It was not possible to undertake pairwise meta-analysis using the data from studies reporting overall QoL scores, as none of the studies reported suitable data (see Appendix 10). The reasons varied between studies and included lack of SC group53,68,72,73 and lack of data on comparable QoL tools and domains between studies. 55,68

Parsons et al. 53 compared EQ-5D scores between participants who received ONS and those who received dietary advice (not SC). 53 Their analysis demonstrated higher post-intervention QoL scores, assessed using the EQ-5D time trade-off (TTO) valuation technique and the EQ-5D VAS rescaled tool, among recipients of ONS than among those who received dietary advice. Intention-to-treat (ITT) analysis at week 12 for the ONS and dietary advice groups were 0.496 and 0.364, respectively, on the EQ-5D TTO measure, and 0.535 and 0.457 on the VAS rescaled tool. Mean post-intervention scores from the ITT analysis were higher (indicating increased QoL) among participants who received ONS (mean post-intervention EQ-5D score of 67.4) than among those who received dietary advice (mean post-intervention EQ-5D score of 57.3). Parsons et al. 53 also compared overall QoL between participants who received ONS and those who received dietary advice using the EQ-5D TTO valuation technique. 53 Based on ITT analysis, QoL was significantly higher in the ONS than in the dietary advice group at the 12-week follow-up [EQ-5D TTO scores (mean ± SE) were 0.50 ± 0.04 vs. 0.36 ± 0.05 for the ONS and dietary advice groups, respectively (p = 0.005)]. Otten et al. 72 compared QoL before and after ONS using the EuroQol visual analogue scale (EQ-VAS). CFB data indicated a mean increase in QoL of 10.8 points among ONS recipients after 3 months.

For the studies that reported data on the subdomains of QoL tools, it was not possible to carry out meta-analysis owing to a lack of at least two studies reporting comparable data (pertaining to psychological or physical aspects of QoL) that compared ONS with SC and reported mean and SD values at CFB or post intervention. With regard to psychological aspects of QoL, Payette et al. 55 reported data for the emotional role functioning domain of the 36-item Short Form Survey (SF-36) that showed that the ONS group had a higher post-intervention mean score (better QoL) (84.1, SD 31.4) than the control group (75.4, SD 35.8). Using the mental component score of the 12-item Short Form Survey (SF-12), Miller et al. 68 reported data indicating that participants who received ONS alone had higher (better) scores (post intervention mean 51.4) than those who received exercise (post-intervention mean 51.3), nutrition plus exercise (post-intervention mean 49.8) or SC (post-intervention mean 49.5).

In relation to physical aspects of QoL, Payette et al. 55 reported that, for the physical role functioning domain of the SF-36, mean post-intervention scores were lower among ONS recipients (63.1, SD 35) than among the SC group (69.5, SD 37.7). Miller et al. 68 reported a higher post-intervention mean score for the physical domain of the SF-12 among ONS recipients (post-intervention mean 31.6) than among those who received an exercise intervention (post-intervention mean 31.5), SC (post-intervention mean 30.1) and ONS plus exercise (post-intervention mean 26.9). GRADE was not assessed for QoL as no meta-analysis was undertaken.

Adherence and acceptability

In addition to the effectiveness outcomes detailed above, we looked for specific evidence regarding factors that may affect adherence to and acceptability of ONS. Specifically, we looked for research detailing barriers and facilitators, determinants and active components that may aid understanding about why some interventions may be more (or less) effective for certain groups of people. Although an inclusive search strategy was used, little relevant information was found in the included studies. The limited information that was collated was derived from the studies included in the effectiveness review. Typically, the data presented, particularly those detailing the acceptability of the ONS, were not assessed through qualitative research methods with patients/health-care professionals but were instead derived from informal observations by the research team. The results are summarised below.

Compliance with ONS was reported in seven studies. 55,57,67,68,70,71 In two studies,67,69 data regarding compliance were reported narratively in brief with no supporting data, making it difficult to draw any firm conclusions about how well participants adhered to ONS. Lauque et al. 67 reported ‘good’ compliance among those at risk of malnourishment who received ONS and those who were malnourished who received ONS but did not give further details. Tidermark et al. 69 measured only compliance with nandrolone injections and not with ONS.

Across the remaining studies,55,57,67,68,70,71 the methods of measuring and reporting compliance were heterogeneous. The lowest level of compliance with ONS was reported in the study by Payette et al. ,55 who assessed adherence in accordance with the number of remaining 250 ml cans of supplement and a mean increase in total energy intake of ≥ 250 kcal per day over the study period. In this study, 23 out of 42 participants (54.8%) were noted to be compliant at 16 weeks. The highest compliance was reported by Miller et al. 68 In this study, 76% of 25 participants in the nutrition -only group adhered to the prescribed volume of nutritional supplement, compared with 66% of 24 participants in the ONS plus exercise group. It may be possible that the differences in compliance between these two studies are related to their setting: Payette et al. 55 was community based, whereas Miller et al. 68 was set in the orthopaedic ward of a hospital. Luo et al. 70 reported that 91–100% of 22 participants in the intervention group consumed their recommended intake of ONS. Van Wymelbeke et al. 57 was the only included study in which it was possible to directly compare compliance with ONS and compliance with another intervention. The study reported that, at 90 days, 74% of 17 participants were consuming all ONS, compared with 83% of 29 participants consuming all the brioche provided.

The precise nature of the interventions included in the effectiveness review was examined closely to understand how the delivery of the intervention may affect its effectiveness. The consistency of reporting on these active components varied across studies. Figure 10 visually displays this information as a rose plot in which the numbers of studies reporting compliance and the energy intake, flavour and frequency of the ONS are shown in blue. Seven studies53–57,68,70 reported frequency of ONS consumption. Five studies53,55–57,67 reported the flavour of ONS. The flavours reported were both sweet and savoury, with strawberry the most common flavour available across all five of the studies. 53,55–57,67 The impact of ONS flavour was not linked with adherence in these five studies. Eight studies53–57,67,70,71 reported the energy intake of the ONS they provided.

FIGURE 10.

Chart showing the level of reporting of various comparators in studies.

Characteristics of included studies

One study23 was included in the systematic review of full economic evaluations. This section reviews that study. Tables 5 and 6 present the key study characteristics and results of the study. Appendix 14 details the reasons for excluding studies from the cost-effectiveness review.

Quality appraisal of included studies

The quality of the included studies was assessed using the BMJ checklist. 49 The completed checklist is reported in Appendix 15. Many of the study design features that would be expected in a well-conducted study were reported in Elia et al. 23 Overall, 30 out of 36 items were assessed as ‘yes’ or ‘not applicable’. Items assessed as ‘no’ included lack of viewpoint and justification for alternatives. Unit costs were not reported, major outcomes were not presented in a disaggregated or aggregated form, and generalisability was not discussed.

Summary of cost-effectiveness results

Summary results from Elia et al. are reported in Table 5. The study found that ONS was associated with greater benefit than the control. ONS was associated with greater QALYs and with higher cost. 23 ONS was cost-effective at a cost-effectiveness threshold of £20,000 per QALY (one of the cost-effectiveness thresholds used by the NICE) with a 0.83 probability. 23

Model design

The effectiveness of ONS can be measured using a variety of outcomes, and the systematic review of effectiveness investigated many of these. Outcomes researched in the systematic review are presented in Figure 11 in the sequence in which outcomes may occur. The cost-effectiveness of ONS depends on health-related QoL and health-care resource outcomes.

FIGURE 11.

Outcomes related to ONS.

The evidence on outcomes identified in the systematic review was considered for use in two ways to evaluate the cost-effectiveness of ONS. First, the association between BMI and mortality, hospitalisation and EQ-5D was modelled, which was used to estimate the effect of ONS on these outcomes using the effectiveness estimate for ONS on BMI compared with SC and brioche from the systematic review. This approach enabled the cost–utility of ONS to be evaluated for patient cohorts with different BMI values at baseline. BMI was selected as the short-term outcome, as it is commonly used to study the relationship between health status and outcomes, and several included studies in the review reported on BMI using the same scale, meaning that the NMA could be conducted on the original scale. Second, a cost–utility analysis was planned for ONS compared with SC on mortality, hospitalisation and EQ-5D, using the effectiveness evidence from the systematic review. The analysis was to be conducted if there was evidence that ONS might be more effective than SC.

Approach 1 has its limitations. BMI is an imperfect measure of benefit from ONS consumption, as it masks gain in muscle mass versus fat mass, which is a limitation of using this outcome. More accurate measures of malnutrition, such as MNA or MUST scores, muscle mass or functional measures would have been more appropriate; however, these could not be used because of limited evidence from the effectiveness review and limited evidence for the association between MNA or MUST and outcomes such as mortality, hospitalisation and EQ-5D index utility. In addition, estimates of improved longer-term outcomes based on changes in BMI may underestimate the benefit from improved nutrition.

There is a non-linear relationship between BMI and mortality hazard, odds of hospitalisation and EQ-5D. The BMI of the model population is, therefore, very important in determining the cost-effectiveness of ONS. For the base-case analysis, a distribution of BMI values was generated from the data reported in the studies from the effectiveness review (Figure 12). As the cost-effectiveness of ONS can be expected to differ according to baseline BMI values, a number of analyses were conducted with a cohort population with a different baseline BMI value in each cohort.

FIGURE 12.

The base-case BMI distribution at baseline.

Association between body mass index and longer-term outcomes

A focused search of the literature was conducted to identify evidence of the association between BMI and mortality, hospitalisation and QoL, such as EQ-5D and ADL measures in the elderly. The focused search is described in Appendix 16. Only one study with appropriate evidence was found for each of the following outcomes: mortality, hospitalisation and EQ-5D. None of the economic evaluations identified in the review of economic evaluations reported statistics of the association between BMI and these outcomes. Details of these studies are presented in Table 7. The identification of only one study for each outcome meant that it was not possible to estimate the variation in BMI–outcome association that may be found across studies with different characteristics. The uncertainty associated with this could not be captured in the model. No studies estimating the association between BMI and ADL in the population of interest were identified.

EuroQol-5 Dimensions

One study80 was found to provide appropriate evidence for the association between BMI and unplanned hospitalisation. The study was cross-sectional. The difference in the EQ-5D index from the mean was reported for many different levels of BMI. The 95% CIs were also reported. The data from the published figure were extracted using WebPlotDigitizer version 4.6 (Pacifica, CA, USA) and from the published figure were extracted using WebPlotDigitizer and are reproduced in Figure 13. Using linear interpolation, each individual was assigned an EQ-5D index (difference from mean) normal distribution with mean and standard error.

FIGURE 13.

Association between BMI and EQ-5D from mean index. Derived from Hunger. 80

An increase in BMI is associated with different changes in BMI depending on the baseline BMI. These data are cross-sectional, and there may be some bias in estimating the change in EQ-5D associated with change in BMI from these data. This is particularly the case if the increase in BMI is associated with an increase in protein intake and the increase in weight is fat-free mass. The baseline BMI values assigned to the population in the analyses range from 17 to 23 kg/m², so change occurs in the upward part of the curve. Increased EQ-5D index utility due to increased BMI was assumed to persist for 1 year.

Effectiveness

No NMA was conducted for the BMI outcome. The effectiveness estimates used in the model were those derived from the BMI meta-analysis. In the base case, the results from all studies were included. Uncertainty in the estimates were modelled using a normal distribution.

Four out of the five studies with evidence for ONS were set in nursing homes, and the remaining study was set in a hospital. Sensitivity analysis using the effectiveness estimate from the meta-analysis included the adequately randomised trials only.

Intervention cost

The incremental cost of ONS compared with SC was calculated for each study included in the NMA of the BMI outcome. For all studies apart from Cameron et al. ,71 the cost of SC was assumed to be zero. For Cameron et al. ,71 SC was high-protein milk, which was costed as a sachet of MCTprocal^® per day. Most studies stated the brand of ONS used in the study. The price of the specific product, or as close as possible, was found on online retail sites and in the British National Formulary. 84

The unit costs of each product identified are reported in Table 10 along with the sources. The exchange rates used to convert to GBP, where necessary, were 1 AUD to 0.55 GBP and 1 euro to 0.86 GBP. 85,86

The daily resource use, number of days, per-day cost and total cost of each resource and intervention are reported in Table 11. For most interventions, 1 minute per day of staff time was assumed for delivering the product at normal meal or snack times. In the study by Lauque et al. 67 an average of 2 minutes per day for each patient was costed for a dietitian who visited once per week.

The total cost per intervention in each study is reported in Table 12. The average cost per type of intervention was included in the analysis: £369.28 for ONS. The assumption was made in the model that ONS is given to the older person for the specific period stated in the studies and no longer. It was also assumed that all of the ONS was used (opened and either consumed or discarded). No information was provided in the studies about unopened products. The settings were nursing homes and hospitals, so it is assumed that the older person continues to be given the ONS during the intervention delivery period.

Incremental cost-effectiveness analysis

The cost-effectiveness of ONS and other comparators was evaluated by estimating the incremental cost-effectiveness ratio (ICER) derived from an incremental cost-effectiveness analysis. The ICER was the incremental cost per QALY gained. This is calculated as the difference in the total discounted cost between the intervention (e.g. ONS) and the comparator (e.g. SC) divided by the difference in the total discounted utility between the intervention and the comparator:

When there are more than two technologies, the ICER for each is compared with the next most cost-effective. A technology is strictly dominated if it costs more and is less effective than a comparator. A technology is dominated by extension if there is a more effective technology with a lower ICER than the next most effective technology. If the ICER of a health technology is less than the accepted cost-effectiveness threshold, then the health technology is considered cost-effective and the decision-maker is willing to adopt the technology. The cost-effectiveness thresholds of £20,000 per QALY and £30,000 per QALY recommended by NICE77 are used as reference cost-effectiveness thresholds in this report.

Analysis of uncertainty

The investigation into how much uncertainty in the evidence influences decision uncertainty, and the uncertainty regarding whether a health-care technology should be adopted, is a key part of an economic evaluation. When evidence is available, we specify probability distributions to represent the uncertainty in the effectiveness estimates. Uncertainty in mortality, hospitalisation and utility outcomes was described in Chapter 5, and uncertainty in the effectiveness estimates was described in Chapter 3. Probabilistic sensitivity analysis was conducted using Monte Carlo simulation, which samples from every distribution n times to produce a joint distribution of the costs and effects of each intervention. The number of iterations was 8000.

The net benefit of adopting a health technology is calculated for different cost-effectiveness thresholds using the following equation:

The proportion of simulation estimates for which the intervention has the highest net benefit represents the probability that the intervention is cost-effective. The probability that an intervention is cost-effective at different cost-effectiveness thresholds is presented in a cost-effectiveness acceptability curve. 95

Base-case and sensitivity analyses

The base-case analysis used the average cost for ONS across the studies included in the BMI NMA. This was £369 per person. The average BMI of 23 kg/m² of the populations in the included studies evaluating the effect of ONS on BMI was assumed for the model population. The effect estimates from the meta-analysis were used. Sensitivity analysis was conducted based on the meta-analysis result using adequately randomised trials only.

Base-case results

The incremental cost-effectiveness results for ONS compared with SC in the base-case analysis are presented in Table 13. The cost of ONS intervention was £369 per person and the baseline BMI was 23 kg/m². The setting for the analysis was assumed to be a care home. Four out of the five studies with BMI effectiveness evidence for ONS were set in nursing homes, and the remaining study was set in a hospital. QALYs and costs were calculated in the model as incremental QALYs and costs compared with SC.

Oral nutritional supplements are associated with a greater expected benefit (0.0145 QALYs) and a higher cost (£359) than SC. The ICER for ONS was £24,390. Although this is below the cost-effectiveness threshold of £30,000 per QALY, SC is more likely to be cost-effective than ONS: the probability that ONS is cost-effective is 0.28 at the £20,000-per-QALY threshold and 0.36 at the £30,000-per-QALY threshold. The probability that ONS is cost-effective compared with SC alone is presented as a cost-effectiveness acceptability curve across the threshold range £5000–50,000 per QALY in Figure 14. This is because the incremental net benefit distribution is skewed. The incremental net benefit distribution is presented in Figure 15 and shows that more than 3000 of the 8000 incremental net benefit estimates are slightly negative. The average is £81. The variation in incremental cost estimates is very low, from £320 to £384. The shape of the incremental net benefit distribution is, therefore, driven by the distribution in incremental QALY estimates. The incremental QALY distribution is presented in Figure 16. The incremental QALY value at which the incremental net benefit is zero, the break-even value, is 0.0118. More than 2500 of the 8000 incremental QALY estimates are positive but less than 0.0118.

FIGURE 14.

Cost-acceptability curve for ONS vs. SC.

FIGURE 15.

Base-case incremental net benefit distribution.

FIGURE 16.

Base-case incremental QALY distribution.

Sensitivity analyses

Summary of cost-effectiveness results

Two approaches to estimating the cost-effectiveness of ONS based on the systematic review were considered. There was limited evidence of the effectiveness of ONS on longer-term outcomes. Based on the available evidence, there was no evidence that ONS were cost-effective using mortality, hospitalisation and EQ-5D outcome evidence. The cost-effectiveness of ONS and SC was also estimated using the meta-analysis effectiveness evidence for the BMI outcome. The benefit of increased BMI was modelled using evidence from the literature on the association between BMI and mortality, hospitalisation and EQ-5D index utility. This evidence showed that there was a greater benefit of an increase in BMI at lower baseline BMI levels.

Oral nutritional supplements were unlikely to be cost-effective at a threshold of £30,000 per QALY for a population cohort with a baseline BMI of 23 kg/m². ONS were even less likely to be cost-effective when using only the adequately randomised controlled trial evidence. Using the adequately randomised trial evidence, there was no strong evidence that ONS were cost-effective at any baseline BMI level.

Using the all randomised trial evidence, ONS were cost-effective at a baseline BMI of 19–21 kg/m² with a high level of certainty when ONS cost no more than £2 per person. It was also cost-effective at a baseline BMI of 17 kg/m² with a high level of certainty when ONS cost no more than £400 per person.

Summary

Eleven primary studies were identified in the effectiveness review. A summary of characteristics of the included studies and participants, evidence quality and findings can be found in Figure 17. Many of the studies had industry funding. Of the 11 studies identified, six (55%) were either fully (n = 4) or partially funded by industry (n = 2). Three were funded from alternative sources and two studies did not include details of funding/conflict of interests. Given the insufficient information on role of funders and the lack of clarity about independent research, the potential limitation of conflict of interest in reporting findings cannot be ruled out in these studies.

FIGURE 17.

The GOfER (Graphical Overview for Evidence Reviews) diagram of included studies in the effectiveness review.

Meta-analyses suggested positive effects of ONS versus SC for energy intake (kcal) (SMD 1.02, 95% CI 0.15 to 1.88; very low-quality evidence) or poor mobility (MD 0.03, p < 0.00001, 95% CI 0.02 to 0.04; very low-quality evidence), and no evidence of an effect for body weight (MD 1.31, 95% CI –0.05 to 2.66; very low-quality evidence) or BMI (MD 0.54, 95% CI –0.03 to 1.11; very low-quality evidence). Pooled results for other outcomes related to malnutrition and its adverse consequences were statistically non-significant. There was mixed narrative evidence regarding the effect of ONS on QoL. All evidence was graded as low or very low quality. NMAs were conducted only for the body weight and grip strength outcomes. The results of the NMA indicated there was evidence of an effect for ONS compared with SC for the body weight outcome only. Study quality was mixed; the method of randomisation was typically poorly reported, and, therefore, all evidence was assessed as low or very low quality using GRADE.

Although the studies included looked at the effectiveness of ONS on all their participants, there was heterogeneity in length and definition of follow-up between the studies. Follow-up spanned from 40 days to 1 year. 54,71 It was difficult to define the length of follow-up in studies, as reports often did not clearly define whether outcome assessments had been undertaken immediately post intervention or after time had elapsed. Additionally, no study reported the impact of the intervention on specific groups (e.g. ethnicity, education or marital status, or by comorbidities). Although demographic information was often reported in the methods section, it was not possible to evaluate the differential impact of ONS in these groups.

Furthermore, there was no systematic reporting in the identified studies of the types or characteristics of ONS that can influence compliance /uptake, such as flavour, and specifically sweetness. Our PPIE group reported that taste was an important factor in whether or not supplements would be consumed. Previous research has reported that age-related changes in taste lead overt sweetness to be one of the major factors contributing to the dislike of ONS. 101 Furthermore, research has demonstrated that the viscosity of the ONS also plays an important role in oral-sensory stimulation and satiety. 102 Den Boer et al. 102 showed that lower thickness of ONS increased intake by one-third without affecting satiation or satiety. Whereas some studies gave a choice of which ONS products could be consumed,53,55,57,67 others either did not report this or used a single brand of ONS.

The studies in the review also lacked any detailed qualitative findings. A single study reported barriers to the use of ONS generally,71 but none examined patient viewpoints in any detail. The discussions with our PPIE group of older people and other stakeholders (e.g. dietitians and clinicians) showed that there was a range of reasons why supplements may work for some people and not others. A lack of reporting on patient experiences in the review is especially surprising, as similar qualitative explorations of the views of both dietitians103 and general practitioners104 views on malnutrition management have already been undertaken.

Compliance data were reported in 7 of the 11 studies. However, there was considerable between-study heterogeneity in how compliance was determined and reported (e.g. aggregated across study arms), which made it difficult to draw firm conclusions. A previous systematic review completed by Hubbard et al. 105 suggested that mean compliance with ONS was 78%; interestingly, compliance was found to be lower in a hospital setting than in the community (67% vs. 81%).

One economic evaluation was identified in the systematic review. This study was conducted in a care home setting. It was well conducted and showed that ONS could be cost-effective in a care home setting when compared with dietary advice.

Two approaches to estimating the cost–utility of ONS using the evidence identified in the effectiveness systematic review were taken. The first approach was to use evidence of the effectiveness of ONS on longer-term outcomes from the systematic review, but there was little or no evidence that ONS was effective using the direct evidence of the effect of ONS on the longer-term outcomes, and so this cost–utility analysis was not conducted. The second approach was to estimate cost-effectiveness using results from the meta-analysis or NMA of effectiveness evidence for BMI in the systematic review. The longer-term consequences of changes in BMI were then modelled using evidence from the literature on the association between BMI and mortality, hospitalisation and EQ-5D index utility. The evidence base for this model design was limited; consequently, the results should be interpreted with caution. This linked evidence showed that there was a greater benefit of an increase in BMI at lower baseline BMI levels. ONS was not likely to be cost-effective at a £30,000 per QALY threshold for a population cohort with a baseline BMI of 23 kg/m² using all randomised trial evidence. ONS was even less cost-effective when only the adequately randomised control trial evidence was used; there was no strong evidence that ONS was cost-effective at any baseline BMI level. Using the all randomised trial evidence, ONS was cost-effective at a baseline BMI level of 19–21 kg/m² with a high level of certainty when ONS cost no more than £200 per person. It was also cost-effective at a baseline BMI level of 17 kg/m² with a high level of certainty when ONS cost no more than £400 per person.

The incremental QALYs in the economic model for ONS versus SC was 0.0145 was slightly lower than the incremental QALYs (0.0174) in the single-study-based economic evaluation by Elia et al. 23 The incremental cost compared with SC (£359) was higher than the incremental cost compared with dietary advice (£191). It is not known if dietary advice is cost-effective. If dietary advice were not cost-effective, then the most appropriate comparator would be SC without dietary advice.

Strengths

This review has many strengths. Our search strategy was broad and wide-ranging and included multiple databases supplemented by searching citations, reference lists of included studies and relevant systematic reviews and comprehensive grey literature sources. We included primary studies that focused on malnourished, frail older people (aged ≥ 65 years). However, we took the pragmatic decision to include studies in which the mean age of participants was ≥ 65 years to ensure that we maximised the evidence base. Two studies53,70 included some participants aged < 65 years. Furthermore, all screening, data extraction and quality assessments were carried out in duplicate to minimise human error. We also included a wide range of outcomes to ensure that the effects of ONS could be investigated across a range of health outcomes, including those hypothesised to respond relatively quickly to ONS (e.g. kcal and protein) and those that may change over a longer period (e.g. hospitalisations, morbidity and mortality). Study authors were contacted when critical missing information was required. Meta-analyses and NMA were undertaken where possible, and the results were compared. For the meta-analyses, both CFB and final values were imputed where one or the other was missing and meta-analyses were conducted for each. A sensitivity analysis was also carried out based on study quality.

A systematic review of economic evaluations of ONS in a frail older population was conducted. Only one study was included, indicating the paucity of cost-effectiveness evidence for ONS in this population. The economic analysis conducted made use of the effectiveness evidence of ONS on BMI from the systematic review. The use of an economic model enabled cost-effectiveness to be explored for different cost assumptions and for different population cohorts defined by BMI. The key uncertainties in the model associated with the effectiveness evidence and the evidence for the association between BMI and outcomes was investigated by conducting sensitivity analyses.

Limitations of the research and deviations from protocol

There are some limitations to this review. The review identified only a small number of included studies focusing on a population of frail older adults who were either malnourished or at risk of malnutrition. This important and growing population, nonetheless, is at high risk of adverse outcomes from malnutrition, and there is limited evidence on the effectiveness of ONS to mitigate malnutrition risk in frail older adults. Therefore, we defined the scope of the review along these lines. A single search was undertaken for both the effectiveness and the cost-effectiveness reviews, which included only studies published in English, which may have excluded some potentially eligible studies. Furthermore, we also deviated from our original protocol in two ways. First, we refined the eligibility criteria with regard to frailty. In the original protocol, we specified that frailty needed to be defined according to a standardised measure such as Fried’s frailty phenotype. However, on screening of the search we found that very few studies described their population as frail in these terms. We therefore decided to expand the eligibility criteria for frailty by using proxy criteria (see Methods). Although these were added after the screening process had begun, to avoid bias we made attempts to prevent being data-driven by asking clinical members of the wider study team for their input into and suggestions about these criteria without indicating the studies and types of data we were encountering. We also updated the searches at a later stage before finalising the results, which will have identified any missing studies related to the extended criteria of frailty. In addition, serious adverse events were defined in discussion among the research team, to only include kidney injury, hyperglycaemia, constipation, diarrhoea, nausea, vomiting, refeeding syndrome and micronutrient deficiency. These were considered the most important serious adverse events, although it is possible that others were not extracted as a result. Finally, we used GRADE to understand the strength of the evidence base, but this was not included in the protocol.

We identified a small evidence base that looked at the impact of ONS on frail, older people specifically. Most studies were based on small samples and the duration of interventions reported was typically ≤ 3 months. The effectiveness of ONS on malnutrition-related outcomes (e.g. grip strength, ADL, hospitalisation) is difficult to establish over a relatively short term. Furthermore, the dose of ONS typically varied across the studies, which could add to the inconsistency observed. Although 11 studies were described as RCTs, the risk-of-bias assessment suggested that there were issues with the allocation sequence generation used and a lack of reporting on how the randomisation was generated, and so forest plots of both all studies and those adequately randomised were reported. 68–71 Additionally, only a few studies reported on certain key outcomes in relation to measures of malnutrition and its consequences (e.g. wound healing, reduction of infections, and falls), and therefore a meta-analysis was not possible. There was also heterogeneity in the measures or scales used for certain outcomes (e.g. ADL, MNA and protein intake). To enable meta-analysis to be undertaken, SMDs were used to aggregate different measures of specific outcomes. As a result, the meta-analysis results are less intuitive to interpret. Another major limitation was the lack of outcome data related to QoL and physical function outcomes, which our PPIE group highlighted as particularly important. Although QoL data were reported in four studies,55,68,72,73 they could not be pooled because of differences in reporting, and, therefore, it was not possible to carry out a meta-analysis. The paucity of robust and consistently reported QoL data highlighted in this review is perhaps not surprising, as measures are often disparate and difficult to combine. The narrative results suggested that the effect of ONS in relation to physical function and overall QoL assessment scores were mixed; however, both studies that reported data for the psychological aspects of QoL reported a positive effect of ONS. The PPIE group questioned the impact of a slight (albeit statistically significant) improvement in energy (for example) and wanted to know whether or not ONS would improve their ability to engage with and be active in their daily lives.

One of the major findings was a lack of data on the effectiveness of interventions by key determinants. Although population characteristics (e.g. socioeconomic status, living arrangement) were reported to some extent at baseline, no studies looked at effectiveness according to these characteristics. As a result, we were unable to examine to whom ONS might be most suited. Similarly, we also set out to investigate what components of the intervention (frequency, length, type, flavour, etc.) led to greater effectiveness. However, owing to both a lack of studies and poor reporting of the nature of the intervention, it was not possible to undertake this analysis. Similarly, there were no qualitative intervention studies identified in the review that examined how patients experienced taking the supplements. Descriptive qualitative studies were identified in this population, but these were not related to a specific intervention and therefore were not eligible for inclusion. 106 Although a wide range of ONS products are available, only five studies gave a choice of flavour or type. 53,55–57,67 Older adults, who are the focus of this review, represent a diverse group, with differences in age and other factors, such as ethnicity and comorbidities. Our PPIE evidence suggested that this was likely to have a strong influence on compliance (and, therefore, potentially on effectiveness), and more research in this area is needed.

Only 10 studies met our inclusion criteria for our meta-analysis. As fewer than 10 studies were included for any outcome, funnel plots could not be used to assess publication bias. This is especially important in this review, as 6 of the 11 studies have a direct link to a company that produces the ONS (through a grant, employees as authors, or free product supply for the trial; see Table 2). Previous research has found evidence between pharmaceutical company sponsorship and results that strongly favour the sponsors’ interest. 107 As a result, there may be trials not published that show negative results. The few studies included in the meta-analyses and the often high level of statistical heterogeneity meant that it was difficult to draw firm conclusions from the results of the meta-analyses.

Sensitivity analysis was conducted on a subset of studies that were assessed as adequately randomised. This was considered to be the most useful sensitivity analysis given the low expectation that studies would be at low risk of bias and that most of the outcome measures were considered to be objective. The exception is the ADL outcome. The classification of the subset of studies in the sensitivity analysis does not imply that the studies included are at low risk of bias and, as expected, no studies were assessed as being at low risk of bias across all categories. Overall, the quality of the reporting and the methods across the included studies was low, and this further limits the conclusions that can be drawn from the results of the meta-analyses.

We included data from the longest follow-up time points available in each study when selecting those for inclusion in the meta-analyses. Too few studies were included in the review to undertake analyses for multiple time points. The reason for choosing the longest follow-up time point was to allow for assessing longer-term impacts on outcomes. Furthermore, as in most meta-analyses, SC was very diverse and varied among the included studies. We made the pragmatic decision to include comparators that we felt were relatively ‘light touch’ or routine and were similar to SC in similar settings (e.g. dietary advice information sheet).

There were also several limitations associated with evaluating cost-effectiveness. There was very limited evidence identified in the systematic review on health-care resource outcomes and outcomes that affect QoL, and this limited the cost-effectiveness conclusions that could be drawn from this evidence. The cost-effectiveness of ONS was also modelled utilising the effectiveness evidence identified in the systematic review focused on linking a short-term measure of malnutrition (i.e. BMI) with longer-term outcomes (i.e. QoL, hospitalisation, mortality). The cost-effectiveness modelling method used, linking BMI effectiveness evidence with outcomes, may underestimate the cost-effectiveness of ONS. This is because change in BMI may be an imperfect proxy for improved nutrition, and the mortality, hospitalisation and EQ-5D index utility outcomes may not capture all of the health-care resource use and health outcomes that may be affected by ONS. The evidence base for the association between change in BMI and QoL and health-care resource outcomes was limited. The use of the cross-sectional study estimating the association of BMI and EQ-5D to model the association between change in BMI and EQ-5D is particularly prone to bias. 80 We also found no information on how the association between BMI and the outcomes might vary by characteristics such as age and setting. Although ONS may not be cost-effective its the average cost, it is possible that it could be cost-effective if a cheaper ONS intervention could be found.

Compared with cost-effectiveness analysis based on effectiveness evidence from a systematic review, the advantage of a trial-based economic analysis is that a comprehensive evaluation of the health-care resource and health outcomes associated with ONS can be undertaken in which all QoL outcomes can be measured using the same preference-based health-related QoL instrument. We conducted a systematic review of the trial-based economic evaluations. There was only one reasonably well-conducted economic evaluation that showed that ONS may be cost-effective in a care home setting. 23 Given the possible variations in settings and older people’s characteristics, additional well-conducted studies may be needed to provide strong evidence of the cost-effectiveness of ONS in a care home. There were no economic evaluations identified conducted in other settings.

Implications for practice/decision-makers

Insufficient evidence was available to make any firm conclusions regarding the effectiveness and cost-effectiveness of ONS in frail older adults. Overall, there was limited evidence on the effectiveness of ONS with or without other dietary interventions in reducing the risk of malnutrition among frail older adults. There seemed to be positive effects of ONS on energy and mobility, but these were typically small, and the extent to which these are of clinical significance needs to be understood in more detail. Furthermore, it was not possible to make any recommendations about for whom ONS may be best suited or which intervention components (e.g. ONS type, with or without other dietary interventions) lead to more successful outcomes. As the effects of ONS varied greatly between studies, considerable uncertainty remains. ONS are one method of oral nutritional support and are part of a wider toolkit of dietary interventions available to health-care professionals. There remains a need to better understand the role of ONS and the extent to which this works alongside other dietary interventions. Given the limited body of evidence in the review, we were unable to make recommendations for practitioners. Our stakeholder discussions with practitioners (e.g. dietitians, care home staff, clinicians in hospitals) showed a need for further research on dietary interventions (or ‘food-first’ approaches), whereby dietary changes to meals and food are encouraged; and to understand the evidence for ONS in the context of these approaches, along with addressing issues underlying malnutrition in frail older adults (e.g. living conditions, comorbidities).

Implications for research

Considerable uncertainty remains about the effectiveness of ONS in reducing the risk of malnutrition in frail older adults. There is a need for high-quality, adequately powered primary studies that report on short- and long-term health outcomes as well as assess the determinants and participant characteristics that could help understand specific groups for whom ONS might be more (or less) effective.

There was also only one ONS cost-effectiveness trial-based economic evaluation in a care home setting. Further research is need in both this population and others (e.g. in hospital and at home). The cost of ONS varied greatly across studies in terms of both the ONS and staff time preparing and administering ONS. There is a lack of evidence on the relationship between the duration of ONS provision and long-term outcomes such as hospitalisation, mortality and QoL. A primary study that includes both short-term malnutrition outcomes and longer-term outcomes would provide further information about those relationships.

High-quality research with a sufficient sample size should investigate the effect of characteristics such as baseline BMI, MNA, mobility and other comorbidities on the effect of ONS. Primary studies often assume that dietary advice (including the use of ONS) can be followed and will be effective. However, in frail older people, malnutrition can be exacerbated by other factors (e.g. dementia, inability to prepare food, residential status), so there is a need for primary studies that take place in a variety of settings and accurately record (and report by) baseline characteristics. More transparent reporting of the nature of the intervention would also allow effectiveness to be measured against intervention components.

Further research on nutritional support for malnourished, frail older people should aim to close the evidence gaps identified in our review:

• Outcomes in different subgroups of frail older adults. Older people represent a diverse group. More targeted research is needed from high-quality primary studies on the differential impacts of ONS in older adults representing different groups, for example age (i.e. < 85 or > 85 years), stage of frailty, deprivation, social isolation, place of residence (e.g. at home, care homes), marital status, ethnicity and comorbidities. Future research should also routinely collect and report outcomes in relation to mediators. This would allow the further development of potential tools that could assess the effectiveness of prescribable ONS in particular groups of older adults.

• Comparison of ONS with other dietary interventions. Our review did not find studies with a wide range of dietary interventions as comparators; for example, no studies had trial arms that promoted energy-dense meals through food enrichment (e.g. choosing full-fat and full-sugar products, nourishing drinks and food enrichment). Future studies should incorporate these dietary approaches compared with, and in combination with, the use of ONS.

• More evidence on comparisons with multicomponent interventions. While some trial arms incorporated ONS alongside exercise, steroids or dietary advice, further research should examine the impact of combined interventions beyond prescribable ONS, including protein/protein-energy supplementation and exercise.

• Detailed qualitative research to explore the acceptability and perspectives of patients. There was a lack of detailed qualitative work to discuss the lived experiences of patients prescribed ONS (vs. other treatments). A mixed-methods approach incorporating detailed interviews/focus groups should be added to subsequent trials.

• Collection of outcome data most relevant to patients. Although a variety of outcomes were reported in the primary studies, our PPIE group members were more concerned with functional or QoL-related outcomes than more clinically driven outcomes. Functional status assessed with either self-report or performance-based measures should be routinely collected alongside clinical outcomes so that patients can realistically assess the likely impact of ONS and other nutritional support interventions on their daily life.

• Duration of follow-up. The length of follow-up varied greatly across the studies, and few studies looked at the impact of outcomes in the longer term. Specifically, few examined the impact on hospitalisation, morbidity and mortality.

• Detailed reporting of trial arms. Although the ONS intervention was often described in some depth, too often there was little information on SC (which varied in studies and settings) or the precise nature of dietary counselling/advice. More comprehensive reporting (e.g. using a standard checklist) of standard/usual care should be included in published trial outputs.

• Cost-effectiveness of ONS interventions across different settings. Only one study evaluated the cost-effectiveness of ONS in a nursing home, and no cost-effectiveness studies were conducted in the home or hospital setting. More primary studies with detailed cost-effectiveness reporting are needed. In addition, there was considerable variation in the cost of ONS across studies. Further research should clarify if there is any difference in effectiveness across types of ONS.

Contributions of authors

Katie Thomson (https://orcid.org/0000-0002-9614-728X) (Research Associate) co-led the writing of the final report and co-supervised the conduct and write-up of the systematic review. She also oversaw the public engagement and led the stakeholder dissemination workshop.

Stephen Rice (https://orcid.org/0000-0002-6767-0813) (Senior Research Associate, Health Economics) co-led the writing of the protocol and the final report. He also supervised the meta-analysis, conducted the network meta-analysis and led the development of the cost-effectiveness model.

Oluwatomi Arisa (https://orcid.org/0000-0001-8581-6073) (Research Assistant, Evidence Synthesis Group) contributed to the screening of studies for the systematic review, data extraction of the systematic review, meta-analysis and the bias elicitation assessment. She also contributed to the writing of the final report.

Eugenie Johnson (https://orcid.org/0000-0003-3324-7141) (Training Fellow, Evidence Synthesis Group) contributed to the screening of studies for the systematic review, data extraction of the systematic review and the bias elicitation assessment. She also contributed to the writing of the final report.

Louise Tanner (https://orcid.org/0000-0003-2340-8677) (Research Associate, Evidence Synthesis Group) contributed to the data extraction of the systematic review and meta-analysis. She also contributed to the writing of the final report.

Christopher Marshall (https://orcid.org/0000-0002-7970-681X) (Research Associate, Evidence Synthesis Group) led the development of the protocol, was involved in screening and contributed to the writing of the final report.

Tumi Sotire (https://orcid.org/0000-0002-5330-9568) (Research Assistant, Health Economics) contributed to the cost-effectiveness model. He also contributed to the writing of the final report.

Catherine Richmond (https://orcid.org/0000-0002-2940-5197) (Research Assistant, Evidence Synthesis Group) designed and ran the searches for the systematic review, helped manage the references and referencing, and contributed to the writing of the final report.

Hannah O’Keefe (https://orcid.org/0000-0002-0107-711X) (Research Assistant, Evidence Synthesis Group) managed the references and referencing and contributed to the writing of the final report.

Wael Mohammed (https://orcid.org/0000-0003-0370-4903) (Research Assistant, Health Economics) contributed to the screening of studies and to the final report.

Margot Gosney (https://orcid.org/0000-0001-9004-7571) (Elderly Care Consultant and Professor of Elderly Care Medicine) contributed to the design and the writing of the protocol, provided advice on the review and contributed to the final report.

Anne Raffle (https://orcid.org/0000-0003-4684-2230) (Chairperson, Elders Council of Newcastle) contributed to the development of the protocol, facilitated the organisation of the PPIE sessions, provided input into the project and contributed to the final report.

Barbara Hanratty (https://orcid.org/0000-0002-3122-7190) (Professor of Primary Care and Public Health) contributed to the design and the writing of the protocol, provided advice on the effectiveness and cost-effectiveness review and contributed to the final report.

Claire T McEvoy (https://orcid.org/0000-0001-8512-3293) (Senior Lecturer in Nutrition and Ageing Research) contributed to the design and writing of the protocol, provided advice on the effectiveness and cost-effectiveness review and contributed to the final report.

Dawn Craig (https://orcid.org/0000-0002-5808-0096) (Professor of Practice in Evidence Synthesis) contributed to the design and writing of the protocol, provided advice on the effectiveness and cost-effectiveness review and contributed to the final report.

Sheena E Ramsay (https://orcid.org/0000-0002-0391-9704) (Clinical Senior Lecturer) was the principal investigator, and she oversaw the running of the project and provided clinical guidance throughout. She contributed to the design of the protocol, screening and data extraction for the systematic review, the bias elicitation assessment and the writing of the final report.

Publication

Thomson KH, Rice S, Arisa O, Johnson E, Tanner L, Marshall C, et al. Effectiveness and cost-effectiveness of oral nutritional supplements in frail older people who are malnourished or at risk of malnutrition: a systematic review and meta-analysis. Lancet Healthy Longev 2022;3:e654–66.

Data-sharing statement

All data requests should be submitted to the corresponding author for consideration. Access to anonymised data may be granted following review.

Disclaimers

This report presents independent research funded by the National Institute for Health and Care Research (NIHR). The views and opinions expressed by authors in this publication are those of the authors and do not necessarily reflect those of the NHS, the NIHR, the HTA programme or the Department of Health and Social Care. If there are verbatim quotations included in this publication the views and opinions expressed by the interviewees are those of the interviewees and do not necessarily reflect those of the authors, those of the NHS, the NIHR, the HTA programme or the Department of Health and Social Care.

MEDLINE (via Ovid)

1. exp Frail Elderly/

2. exp Frailty/

3. frail*.ti,ab,kw,kf.

4. ((older or aged) adj (person* or people or patient* or population*)).ti,ab,kw,kf.

5. ((geriatric or elder*) adj2 (people or person* or patient* or population*)).ti,ab,kw,kf.

6. exp Nursing Homes/ or exp Homes for the Aged/

7. ((residential or nursing or care) adj home*).ti,ab,kw,kf.

8. exp Respite Care/

9. exp Long-Term Care/

10. “home* for the aged”.ti,ab,kw,kf.

11. “old age home*”.ti,ab,kw,kf.

12. “skilled nursing facilit*”.ti,ab,kw,kf.

13. “intermediate care facilit*”.ti,ab,kw,kf.

14. “respite care”.ti,ab,kw,kf.

15. “long term care facilit*”.ti,ab,kw,kf.

16. or/1-15

17. Dietary Supplements/

18. Malnutrition/dh, dt, pc, th [Diet Therapy, Drug Therapy, Prevention & Control, Therapy]

19. Nutritional Support/

20. Food, Fortified/

21. Food, Formulated/

22. “oral nutrition*”.ti,ab,kw,kf.

23. “dietary counselling”.ti,ab,kw,kf.

24. “dietary supplement*”.ti,ab,kw,kf.

25. (food adj2 (fortif* or formulat*)).ti,ab,kw,kf.

26. “nutritional intervention*”.ti,ab,kw,kf.

27. “liquid supplement*”.ti,ab,kw,kf.

28. “sip feed*”.ti,ab.

29. “nutrition* management”.ti,ab,kw,kf.

30. (nutri* adj2 (supplement* or therapy)).ti,ab,kw,kf.

31. (maln* adj2 (prevent* or management or risk factor*)).ti,ab,kw,kf.

32. or/17-31

33. 16 and 32

34. exp animal/ not human/

35. 33 not 34

36. limit 35 to english language.

EMBASE (via Ovid)

1. exp frail elderly/

2. exp frailty/

3. frail*.ti,ab,kw.

4. ((older or aged) adj (person* or people or patient* or population*)).ti,ab,kw.

5. ((geriatric or elder*) adj2 (frail* or people or person* or patient* or population*)).ti,ab,kw.

6. exp nursing home/

7. exp home for the aged/

8. ((residential or nursing or care) adj home*).ti,ab,kw.

9. exp respite care/

10. long term care/

11. “home* for the aged”.ti,ab,kw.

12. “old age home*”.ti,ab,kw. 

13. “skilled nursing facilit*”.ti,ab,kw.

14. “intermediate care facilit*”.ti,ab,kw.

15. “respite care”.ti,ab,kw.

16. “long term care facilit*”.ti,ab,kw.

17. or/1-16

18. dietary supplement/

19. malnutrition/dt, pc, th [Drug Therapy, Prevention, Therapy]

20. nutritional support/

21. fortified food/

22. elemental diet/

23. “oral nutrition*”.ti,ab,kw.

24. “dietary counselling”.ti,ab,kw.

25. “dietary supplement*”.ti,ab,kw.

26. (food adj2 (fortif* or formulat*)).ti,ab,kw.

27. “nutritional intervention*”.ti,ab,kw.

28. “liquid supplement*”.ti,ab,kw.

29. “sip feed*”.ti,ab.

30. “nutrition* management”.ti,ab,kw.

31. (nutri* adj2 (supplement* or therapy)).ti,ab,kw.

32. (maln* adj2 (prevent* or management or risk factor*)).ti,ab,kw.

33. or/18-32

34. 17 and 33

35. exp animal/

36. exp human/

37. 35 not 36

38. 34 not 37

39. limit 38 to english language.

Cochrane

1. #1 MeSH descriptor: [Frail Elderly] explode all trees

2. #2 MeSH descriptor: [Frailty] explode all trees

3. #3 (frail*):ti,ab,kw

4. #4 (((older or aged) NEAR/1 (person* or people or patient* or population*))):ti,ab,kw

5. #5 (((geriatric or elder*) NEAR/2 (people or person* or patient* or population*))):ti,ab,kw

6. #6 MeSH descriptor: [Nursing Homes] explode all trees

7. #7 (((residential or nursing or care) NEAR/1 home*)):ti,ab,kw

8. #8 MeSH descriptor: [Respite Care] explode all trees

9. #9 MeSH descriptor: [Long-Term Care] explode all trees

10. #10 (home* NEXT “for the aged”):ti,ab,kw

11. #11 (“old age” NEXT home*):ti,ab,kw

12. #12 (“skilled nursing” NEXT facilit*):ti,ab,kw

13. #13 (“intermediate care” NEXT facilit*):ti,ab,kw

14. #14 (“respite care”):ti,ab,kw

15. #15 (“long term care” NEXT facilit*):ti,ab,kw

16. #16 (OR #1-#15)

17. #17 MeSH descriptor: [Dietary Supplements] this term only

18. #18 MeSH descriptor: [Malnutrition] this term only and with qualifier(s): [therapy - TH, diet therapy - DH, drug therapy - DT, prevention & control - PC]

19. #19 MeSH descriptor: [Nutritional Support] this term only

20. #20 MeSH descriptor: [Food, Fortified] this term only

21. #21 MeSH descriptor: [Food, Formulated] this term only

22. #22 (oral NEXT nutrition*):ti,ab,kw

23. #23 (“dietary counselling”):ti,ab,kw

24. #24 (dietary NEXT supplement*):ti,ab,kw

25. #25 ((food NEAR/2 (fortif* or formulat*))):ti,ab,kw

26. #26 (nutritional NEXT intervention*):ti,ab,kw

27. #27 (liquid NEXT supplement*):ti,ab,kw

28. #28 (sip NEXT feed*):ti,ab,kw

29. #29 (nutrition* NEXT management):ti,ab,kw

30. #30 ((nutri* NEAR/2 (supplement* or therapy))):ti,ab,kw

31. #31 ((maln* NEAR/2 (prevent* or management or risk factor*))):ti,ab,kw

32. #32 (OR #17-#31)

33. #33 #16 AND #32.

CINAHL (via EBSCOhost)

1. S38 S37 AND LA English

2. S37 S33 NOT S36

3. S36 S34 NOT S35

4. S35 (MH “Human”)

5. S34 (MH “Animals+”)

6. S33 S16 AND S32

7. S32 S17 OR S18 OR S19 OR S20 OR S21 OR S22 OR S23 OR S24 OR S25 OR S26 OR S27 OR S28 OR S29 OR S30 OR S31

8. S31 TI ( (maln* N2 (prevent* or management or risk factor*)) ) OR AB ( (maln* N2 (prevent* or management or risk factor*)) )

9. S30 TI ( (nutri* N2 (supplement* or therapy)) ) OR AB ( (nutri* N2 (supplement* or therapy)) )

10. S29 TI “nutrition* management” OR AB “nutrition* management”

11. S28 TI “sip feed*” OR AB “sip feed*”

12. S27 TI “liquid supplement*” OR AB “liquid supplement*”

13. S26 TI “nutritional intervention*” OR AB “nutritional intervention*”

14. S25 TI ( (food N2 (fortif* or formulat*)) ) OR AB ( (food N2 (fortif* or formulat*)) )

15. S24 TI “dietary supplement*” OR AB “dietary supplement*”

16. S23 TI “dietary counselling” OR AB “dietary counselling”

17. S22 TI “oral nutrition*” OR AB “oral nutrition*”

18. S21 (MH “Food, Formulated”)

19. S20 (MH “Food, Fortified”)

20. S19 (MH “Nutritional Support”)

21. S18 (MH “Malnutrition/DH/DT/PC/TH”)

22. S17 (MH “Dietary Supplements”)

23. S16 S1 OR S2 OR S3 OR S4 OR S5 OR S6 OR S7 OR S8 OR S9 OR S10 OR S11 OR S12 OR S13 OR S14 OR S15

24. S15 TI “long term care facilit*” OR AB “long term care facilit*”

25. S14 TI “respite care” OR AB “respite care”

26. S13 TI “intermediate care facilit*” OR AB “intermediate care facilit*”

27. S12 TI “skilled nursing facilit*” OR AB “skilled nursing facilit*”

28. S11 TI “old age home*” OR AB “old age home*”

29. S10 (MH “Long Term Care”)

30. S9 (MH “Respite Care”)

31. S8 TI ( ((residential or nursing or care) N1 home*) ) OR AB ( ((residential or nursing or care) N1 home*) )

32. S7 TI home* for the aged OR AB home* for the aged

33. S6 (MH “Nursing Homes+”)

34. S5 TI ( ((geriatric or elder*) N2 (people or person* or patient* or population*)) ) OR AB ( ((geriatric or elder*) N2 (people or person* or patient* or population*)) )

35. S4 TI ( ((older or aged) N1 (person* or people or patient* or population*)) ) OR AB ( ((older or aged) N1 (person* or people or patient* or population*)) )

36. S3 TI frail* OR AB frail*

37. S2 (MH “Frailty Syndrome”)

38. S1 (MH “Frail Elderly”).

Scopus

( ( TITLE-ABS-KEY ( frail* ) OR TITLE-ABS-KEY ( ( ( older OR aged ) W/1 ( person* OR people OR patient* OR population* ) ) ) OR TITLE-ABS-KEY ( ( ( geriatric OR elder* ) W/2 ( people OR person* OR patient* OR population* ) ) ) OR TITLE-ABS-KEY ( ( ( residential OR nursing OR care ) W/1 home* ) ) OR TITLE-ABS-KEY ( “home* for the aged” ) OR TITLE-ABS-KEY ( “old age home*” ) OR TITLE-ABS-KEY ( “skilled nursing facilit*” ) OR TITLE-ABS-KEY ( “intermediate care facilit*” ) OR TITLE-ABS-KEY ( “respite care” ) OR TITLE-ABS-KEY ( “long term care facilit*” ) ) ) AND ( ( TITLE-ABS-KEY ( “oral nutrition*” ) ) OR ( TITLE-ABS-KEY ( “dietary counselling” ) ) OR ( TITLE-ABS-KEY ( “dietary supplement*” ) ) OR ( TITLE-ABS-KEY ( ( food W/2 ( fortif* OR formulat* ) ) ) ) OR ( TITLE-ABS-KEY ( “nutritional intervention*” ) ) OR ( TITLE-ABS-KEY ( “liquid supplement*” ) ) OR ( TITLE-ABS-KEY ( “sip feed*” ) ) OR ( TITLE-ABS-KEY ( “nutrition* management” ) ) OR ( TITLE-ABS-KEY ( ( nutri* W/2 ( supplement* OR therapy ) ) ) ) OR ( TITLE-ABS-KEY ( ( maln* W/2 ( prevent* OR management OR “risk factor*” OR therapy ) ) ) ) ) AND NOT INDEX ( medline ) AND ( LIMIT-TO ( LANGUAGE , “English” ) )

Change from baseline

For CFB values, the decision rule was guided by the standard deviation values for studies reporting the same outcome measure. There were two options.

Final values

For final values, the lowest correlation value was set that enabled the average standard deviation to get as close to the average standard deviation at baseline as possible. The average standard deviation was always higher due to the conservative solution used for the standard deviation for the final value.

The ratios of the calculated standard deviations to the baseline standard deviations, the ratios of the reported standard deviations to the baseline standard deviations, and the set correlations are reported in Table 20 for each outcome. These include results for a few studies from which data were extracted but were not finally included in the review. The outcomes and context of the excluded studies are still relevant for assessing the plausibility of the imputed values.

Body weight

FIGURE 21.

Body weight: final values.

FIGURE 22.

Body weight: CFB.

Body mass index

FIGURE 23.

Body mass index: final values.

FIGURE 24.

Body mass index: CFB.

Search strategy for body mass index outcome: MEDLINE

1. Body Mass Index/

2. BMI.ti,ab,kw,kf.

3. Body Mass Index.ti,ab,kw,kf. 

4. MNA.ti,ab,kw,kf.

5. EQ-5D.tib,kw,kf.

6. Barthel Index.ti,ab,kw,kf.

7. Kartz Index.ti,ab,kw,kf.

8. Readmission.ti,ab,kw,kf.

9. Hospital admission.ti,ab,kw,kf.

10. Falls.ti,ab,kw,kf.

11. associat*.ti,ab,kw,kf. 

12. or/5-10

13. Statistics as Topic/

14. relationship.ti,ab,kw,kf.

15. statistical.ti,ab,kw,kf.

16. regression.ti,ab,kw,kf.

17. exp Frail Elderly/

18. exp Frailty/ 

19. frail*.ti,ab,kw,kf.

20. ((older or aged) adj (person* or people or patient* or population*)).ti,ab,kw,kf.

21. ((geriatric or elder*) adj2 (people or person* or patient* or population*)).ti,ab,kw,kf.

22. exp Nursing Homes/ or exp Homes for the Aged/

23. ((residential or nursing or care) adj home*).ti,ab,kw,kf.

24. exp Respite Care/

25. exp Long-Term Care/

26. “home* for the aged”.ti,ab,kw,kf.

27. “old age home*”.ti,ab,kw,kf. 

28. skilled nursing facilit*.ti,ab,kw,kf

29. .intermediate care facilit*.ti,ab,kw,kf.

30. respite care.ti,ab,kw,kf.

31. long term care facilit*.ti,ab,kw,kf.

32. or/17-31

33. 11 or 13 or 14 or 15 or 16

34. 1 or 2 or 3 or 4

35. 12 and 32 and 33 and 34

36. limit 35 to (English language and full text and humans and yr=“1980-Current”).