The Effectiveness of Oral Nutritional Supplements for the Management of Malnutrition

Please note - all figures, graphs, appendices, and identifying information have been removed. Please contact me for a copy of the original submission.

IV. Abstract

Malnutrition is prevalent in many disease states and all care settings and causes measurable adverse outcomes for both individuals and the healthcare economy. The National Institute of Health and Care Excellence (NICE) Clinical Guideline 32 recommends the provision of nutrition support for anyone with malnutrition. Oral nutritional supplements (ONS) can be prescribed when food alone is insufficient to meet nutritional requirements. Research on ONS has been limited to specific patient groups and clinical settings. There is insufficient evidence that can be generalised to all patients that meet the indications for malnutrition. The aim was to evaluate the effectiveness of ONS for the entire malnourished population by meeting set objectives: completing a systematic search and meta-analysis of recent randomised controlled trials (RCT) and evaluating which factors, and to what extent, they influenced nutritional outcomes. Eleven academic databases were searched, using combinations of 8 search terms and 12 studies were selected for the meta-analysis.

Weight, BMI and MUAC increased in intervention groups in all but 1 RCT but not to a significant degree. Improvements were slightly higher in care and community settings than hospitals. Nutritional improvements were significant for malnourished and fracture patients but not cancer patients and compliance to the intervention had a significant effect of weight change. To conclude, when adherence to the intervention is high, ONS are somewhat effective at improving nutritional outcomes, specifically in malnourished and fracture patients.

V. Key words

Oral nutrition supplement, ONS, malnutrition, enteral, systematic review, meta-analysis

Table of Contents

I. Title page 3

II. Statement of originality 4

III. Acknowledgements 5

IV. Abstract 6

1.0 List of figures and tables 9

1.1 List of abbreviations 9

1.3 Aims and objectives 10

2.0 Methodology 10

2.1 Criteria for studies 10

2.1.1 Types of studies 10

2.1.2 Types of participants 10

2.1.3 Types of setting 11

2.1.4 Types of intervention 11

2.1.5 Outcome measures 11

2.2 Search methods 11

2.2.1 Databases 11

2.2.2 Search terms 12

2.3 Study selection 12

2.3 Quality assessment 14

2.4 Ethics and risk assessment 14

2.5 Statistical analysis 14

3.0 Results 14

3.1 Study Selection 14

3.2 Quality assessment 16

3.3 Characteristics of included studies 19

3.4 Statistical analysis 22

3.4.1 Primary outcome measures 22

3.4.2 Healthcare setting and patient group 25

3.4.3 ONS prescription and compliance 26

4.0 Discussion 27

4.1 Systematic search and study selection 27

4.2 Meta-Analysis 28

4.2.1 Changes in nutritional outcomes 28

4.2.2 Nutritional outcomes by healthcare setting and disease state 29

4.2.3 Nutritional intervention and patient compliance 31

4.3 Limitations 31

4.4 Conclusions 32

4.5 Recommendations 33

5.0 Appendices 36

5.1 Appendix 1 - research proposal 36

5.2 Appendix 2 - project approval form 54

5.3 Appendix 3 - risk assessment 55

5.4 Appendix 4 - ethical evaluation 58

5.5 Appendix 5 - detailed search strategy 66

5.6 Appendix 6 - detailed study review 68

1.0 List of figures and tables

Table 1: Information to be extracted from studies being considered for review

Table 2: Inclusion and exclusion criteria for studies being considered for review

Table 3: Quality assessment of studies being considered for review

Table 4: Characteristics of studies selected for meta-analysis

Figure 1: Process of study selection for meta-analysis, adapted from PRISMA (Moher et al., 2009)

Figure 2: Percentage weight change between control and intervention groups

Figure 3: Percentage weight change in intervention and control groups in each study.

Figure 4: Percentage weight change in intervention and control groups in each study spanning 0.

Figure 5: Percentage BMI change between control and intervention groups

Figure 6: Percentage MUAC change between control and intervention groups

Figure 7: Intervention group percentage weight change in care, community & hospital settings

Figure 8: Percentage weight change correlated with energy prescribed in the nutritional intervention

Figure 9: Percentage weight change correlated with energy prescribed in the nutritional intervention

Figure 10: Percentage weight change correlated with compliance to the nutritional intervention

1.1 List of abbreviations

ONS - Oral Nutritional Supplement

NICE – National Institute for Health and Care Excellence

RCT- Randomised Controlled Trial

ACBS – Advisory Committee for Borderline Substances

BMI – Body Mass Index

MUAC – Middle Upper Arm Circumference

CG32 – Clinical Guideline 32

ASP – Academic Search Premier

BSU – Bath Spa University

DOAJ – Directory of Open Access Journals

NHLBI – National Heart, Lung and Blood Institute

UC – Usual Care

DA – Dietary Advice

DC – Dietary Counselling

NHS – National Health Service

1.3 Aims and objectives

The aim was to evaluate the effectiveness of appropriately prescribed ONS for the management of malnutrition, for patients in any care setting and disease state.

The objectives were to:

  • Carry out a systematic search into the use of ONS in randomised controlled trials from 2010-2020 and use R to analyse the changes in patient’s nutritional outcomes.

  • Identify results for each reported nutritional outcome and evaluate whether any changes in nutritional outcomes were significant.

  • Evaluate whether different patient groups or healthcare settings are more likely to have different nutritional outcomes.

  • Evaluate how nutritional outcomes changed based on the intended energy intake for the interventions and how compliant participants were.

2.0 Methodology

A literature review on ONS was conducted. This informed the aims and objectives of the systematic review and meta-analysis. The methodology was planned in accordance with guidelines published by the Cochrane Collaboration, following protocol for high-quality systematic reviews (Higgins et al., 2019). The 2009 PRISMA flow diagram was used to plan and document the systematic search technique (Moher et al., 2009). A research proposal was drafted and approved by the project supervisor (Appendix 1 – Research Proposal; Appendix 2 – Project Approval Form).

2.1 Criteria for studies

2.1.1 Types of studies

Randomised Controlled Trials (RCT) of nutritional interventions using ONS were considered for review. A 10-year time frame was applied; 2010-2020.

2.1.2 Types of participants

Participants were to consist of adults, capable of giving consent that met the Advisory Committee for Borderline Substances (ACBS) indication for ONS (BAPEN, 2016). Patients in all disease states were included because although the primary research interest was malnutrition, additional health conditions are often present in people with malnutrition.

2.1.3 Types of setting

Oral nutritional supplements are prescribed across all healthcare settings, so all were considered. Study setting was categorised as acute (hospital), care (assisted living / nursing / care homes) or community (free living). The only exclusions were laboratory studies, due to their snapshot nature and lack of generalisability to the malnourished population.

2.1.4 Types of intervention

Studies were considered if the intervention was comparable to the NICE prescribing guidelines (12 weeks at >300kcal/day) (NICE, 2017). Intervention groups usually receive a fairly uniform treatment, with only minor differences in dosage and length of intervention. However, due to ethical considerations, treatment in control groups often varies significantly, some providing dietary counselling, food fortification or additional snacks (Parsons et al., 2017). All types of control group were considered provided they passed quality assessment review.

2.1.5 Outcome measures

The impact of ONS on nutritional status was to be measured and evaluated. The primary outcome measure was percentage weight change. Percentage change of BMI and MUAC were also measured to ensure all body composition measurements were concurrent, to indicate the reliability of results. The secondary outcome measures were the prescribed energy intake in the nutritional intervention and the participant’s adherence to the intervention.

2.2 Search methods

The search was planned and conducted in accordance with the Cochrane Handbook and using the PRISMA guidelines (Higgins et al., 2019; Moher et al., 2009).

2.2.1 Databases

All database searching took place between November 2019 and February 2020. The initial searches took place in November and were repeated in February to capture any newly released publications. The databases used to identify the relevant research were, on instruction from the Cochrane Collaboration, PubMed (including MEDLINE), Cochrane CENTRAL and To ensure all relevant research was evaluated the search was extended to include: Academic Search Premier (ASP), Bath Spa University (BSU) online library, Directory of Open Access Journals (DOAJ), Google Scholar, JSTOR and Science Direct. The bibliographies of NICE Clinical Guideline 32 (CG32), the Malnutrition Pathway and all studies chosen for the analysis were also hand-searched to help identify any further research (NICE, 2017; The Malnutrition Pathway, 2017).

2.2.2 Search terms

Several search terms were used in isolation and in combinations: “ONS”, “Oral Nutritional Supplement”, “Oral Nutrition Supplement”, “enteral”, “nutrition”, “sip”, “sip feed”, “malnutrition”. Search terms were input to all databases using the same combinations. Filters including: ‘2010-2020’, ‘peer-reviewed’ and ‘research articles’ were applied.

2.3 Study selection

All results were imported into a desktop reference manager (Mendeley, 2020) and the ‘remove duplicates’ tool was used to refine the results. Some databases did not support citation export to Mendeley, for these papers, citations were added manually if they did not already appear in the imported list. Studies identified by searching reference lists were added manually. Once all studies had been collated, titles were examined for relevance and excluded if irrelevant. Table 1 shows the basic data extraction table that was created to help evaluate the abstracts of the remaining studies.

Table 1: Information to be extracted from studies being considered for review

Study characteristics

If the study was deemed relevant from the extracted data, the full paper was obtained and evaluated against the pre-set inclusion/exclusion criteria shown in table 2. If the full text could not be obtained, the study was excluded.

Table 2: inclusion and exclusion criteria for studies being considered for review

2.3 Quality assessment

The National Heart, Lung and Blood Institute (NHLBI) framework was used to determine the overall quality of each piece of research and to determine any ‘fatal flaws’ associated with each study (NHLBI, 2020). The Cochrane collaboration indicates that this tool should be used and then verified by a second researcher, however, as this was an independent project, a second researcher would not be appropriate (Higgins et al., 2019).

2.4 Ethics and risk assessment

The risks associated with the project were minimal as it was a desktop review. There was some occupational risk for the researcher which was evaluated in a risk assessment (Appendix 3 – Risk Assessment). There were no ethical issues as the project did not use human participants (Appendix 4 – Ethical Evaluation). Additionally, the study quality assessment tool ensured that all participant data was anonymised, and participants had given consent.

2.5 Statistical analysis

Due to the nature of heterogenous data in nutritional interventions, primary outcome measures were converted into percentage change. R Commander and Excel were used to produce box plots and graphs that best represented the collected data (R-Commander, 2020). R Commander was used to conduct statistical analysis, including tests for correlations, ANOVAs and linear models.

3.0 Results

3.1 Study Selection

899,170 studies were identified for the review. Applying the relevant filters (Appendix 5 – Detailed Search Strategy), refined the search to 6692 studies, these were imported to a reference management system and duplicates were excluded (Mendeley, 2020). Titles and abstracts were reviewed in 3 stages for relevance to the research question. The final 30 studies were reviewed against the inclusion/exclusion criteria and quality assessment, 12 trials were selected to be included in the analysis (Appendix 6 – Detailed Study Review).

This process is shown below in figure 1.

Figure 1: Process of study selection for meta-analysis, adapted from PRISMA (Moher et al., 2009).

3.2 Quality assessment

Three studies were eliminated from the review using the NHLBI quality assessment tool (NHLBI, 2020). The reasons for this were: lack of randomisation (Allen et al, 2013), high attrition (Seong-Hyeon et al., 2018), low intervention adherence (Seong-Hyeon et al., 2018) and a general lack of detail provided meaning quality score could not be determined (Allen et al., 2013; Grode et al., 2014). The remaining studies scored between 10-14 and had no ‘fatal flaws’. There were 3 main quality issues with the remaining studies, but none were considered grounds for exclusion.

  • First, in all studies except Lee et al. (2013), the treatment allocation concealment and blinding were not ideal. Lee effectively concealed and blinded by preparing the supplement as a warm drink and the serving the control group warm soup, the dietitians instructed care home staff to prepare and serve the residents and did not have contact with them during this time. The remaining studies did not sufficiently blind or conceal allocations, however, as stated in each review, the nature of the intervention (high energy drinks provided in bottles), made it unfeasible to provide a placebo alternative. This is a common issue in this field of research, but the risk of bias was minimal as the outcome measures (weight, BMI, MUAC) were objective measurements, unlikely to be influenced by the lack of concealment.

  • The second issue was high attrition, seen in Huynh et al. (2014) and Parsons et al. (2017). This high attrition was acknowledged by Huynh et al. and attributed to their decision to exclude all participants who were non-compliant to the intervention in order to gain a true result of the effect of ONS on body weight. The high attrition in Parsons was primarily due to death or declining health. Both studies completed an intention to treat analysis.

  • The final issue was compliance to the intervention, this was either not reported (Lee et al., 2013), was not measured (Cameron et al., 2011) or was low (Baldwin et al., 2011; Hatao et al., 2016; Zhu et al., 2019). Compliance is a common issue in nutritional interventions and as such, there is no gold standard for compliance measurement. Therefore, as long as compliance and the possible effects was discussed in the study, the papers were not excluded based on compliance.

Y = Yes, N = No, NR = Not reported, CD = Cannot determine

3.3 Characteristics of included studies

The selected studies were conducted across a range of healthcare settings. Four studies were carried out in hospitals (Cameron et al., 2011; Jiang, 2018; Myint, 2018; Wyers 2018), 4 in care (Jobse, 2018; Lee et al., 2013; Parsons et al., 2016; Stange, 2013) and 2 in the community (Baldwin et al., 2011; Hatao et al., 2016). An additional 2 studies (Huynh et al., 2014; Zhu et al., 2019) were completed across both hospital and community settings but the majority of the intervention was implemented for the 12-weeks post discharge, so each were considered community studies.

Participants fell into 3 main patient groups: cancer (Baldwin et al., 2011; Hatao et al., 2016; Jiang et al, 2018; Zhu et al., 2019), fracture (Cameron et al., 2011, Myint et al., 2018; Wyers et al., 2018) and malnutrition (Huynh et al., 2014; Jobse et al., 2018; Lee et al., 2013; Parsons et al., 2016; Stange et al., 2013). The number of participants in each study ranged from 44 to 358, aged 39-88 years. All studies except Baldwin et al. and Cameron et al. were completed over a 12-week intervention period (this complies with clinical guidance for ONS prescriptions) (NICE, 2017). The endpoint for Baldwin et al. and Cameron et al. was 6 weeks.

Due to the rigorous selection process, all trials had a similar intervention protocol, a prescription of 250-600kcal of ONS daily. However, differences were seen in the control groups. Due to ethical restrictions, controls groups either received no intervention or continued receiving ‘usual care’ if in care or hospital. Many studies evaluated beyond nutritional outcomes, including clinical and functional parameters. However, all trials used anthropometric measures to determine nutritional status. All except Myint (2018), reported weight change in either kilograms or percentage changes, 7 reported BMI and 6 reported MUAC. The results from each anthropometric measure were collated.

Using the data extraction form (table 1), the key characteristics of the studies that had been selected for meta-analysis were collated and input to table 4.

3.4 Statistical analysis

3.4.1 Primary outcome measures

The primary outcome measure was weight and all studies except Myint (2018) reported weight change. Figure 2 shows there is variance in weight change between groups, mean weight change was calculated as -0.40% (SD 4.09) for the intervention group, and -2.77% (SD 4.10) for the control group. However, although more weight was retained in the intervention groups, these differences were not statistically significant (ANOVA P=0.2).

Figure 2. Percentage weight change between control and intervention groups

Figure 3 shows the differences in weight change between both groups for each study. In all but one study (Cameron et al., 2011), positive weight change favoured the intervention group

Figure 3. Percentage weight change in intervention and control groups in each study.

Figure 4 is a variant of the data shown in figure 3, however it shows a clear graphical view of which studies reported overall weight gain which found overall weight loss. Cameron (2011), Hatao (2016) and Jiang (2018) reported weight loss in both groups. Huynh (2014) and Parsons (2017) reported weight gain in both groups. All remaining studies reported weight loss in the control groups but weight gain in the intervention groups.

Figure 4: Percentage weight change in intervention and control groups in each study spanning 0.

There was less available data for the other outcome measures, only 7 studies reported BMI. The same trend as for weight change was found (figures 2, 3 & 4). Figure 5 shows the differences in BMI between control and intervention groups. The calculated mean BMI change was non-significant, at -0.62% (SD 3.29) in the intervention groups, compared to -2.87% (SD 4.17) in the control groups (P = 0.28). Only 6 studies reported MUAC and this was the only outcome measure that reported overall weight gain in the intervention group. Figure 6 shows very small variance in the intervention group but large variance in the control. The mean difference was 0.12% (SD 0.26) in the intervention groups and -0.08% (SD 2.01) in the control. These differences were not significant (P=0.81).

Figure 5. Percentage BMI change between control and intervention groups

Figure 6. Percentage MUAC change between control and intervention groups

3.4.2 Healthcare setting and patient group

A sub-group analysis showed weight change by healthcare setting. Figure 7 revealed a trend of slight weight gain in care (0.38% (SD 1.94)), slight weight loss in the community (-1.58% (SD 5.01)) and the most weight loss seen in hospital trials (-4.22% (SD 5.11)). The most variance was found in community and hospital and the least in care. However, the differences between these groups were not statistically insignificant (P=0.152).

Figure 8 shows that there were significant differences when weight gain was stratified by patient group (P=0.0199), malnourished patients gained an average of 0.91% body weight (SD 2.10), fracture patients lost -1.78% body weight (SD 3.16) and cancer patients lost -4.61% of body weight (SD 5.30). A linear model showed that the key driver for weight change was patient group. Specifically, fracture and malnourished patients contributed the most significantly to weight change (P=0.0199; P=0.00342).

3.4.3 ONS prescription and compliance

The prescribed ONS in each intervention ranged from 250-600kcal daily. A slight positive correlation was seen between energy (kcal) prescribed and participant’s increased percentage weight change. However, a linear regression showed that this trend was not statistically significant (P=0.13).

Nine studies expanded on the intervention protocol, reporting on the participants compliance to nutritional guidance. Compliance varied from 35-90% and a statistically significant correlation was calculated when comparing the effects of compliance (%) to the specified intervention on percentage weight change (P=0.008). There were no indications as to why compliance differed between studies. A Pearson’s correlation showed that neither amount of energy prescribed, healthcare setting or patient group affected compliance (P=0.63; P=0.21; P=0.15).

4.0 Discussion

4.1 Systematic search and study selection

The data collected and presented in the results enabled each of the objectives to be met. The first objective was to carry out a systematic search and use R to analyse the nutritional outcomes data. Twelve studies were chosen for analysis, refined from over 800,000, and contained a total of 1579 participants. Four studies had less than 100 participants which lowers the reliability of the analysis. However, this is a common issue in nutritional interventions, particularly where research is conducted in care settings with a small number of residents (Baldwin et al., 2015). All studies with low participant numbers addressed this in their evaluation of results and if the sample size was not sufficient to detect 80% power, this was highlighted in the quality assessment.

Cawood et al. (2012), indicate that developing world studies should be excluded in reviews of nutritional interventions, however they provide no justification. Perhaps it is due to different healthcare systems and prevalence of malnutrition. In this research, Jiang (2018) and Zhu (2019) were conducted in China (classified a developing country) (United Nations, 2014). However, both studies passed quality assessment, were conducted in research and healthcare environments and the interventions were uniform to the developed world studies, therefore it was deemed appropriate to include these in the analysis. The outcomes from these studies were mixed, however Jiang saw extreme weight loss in comparison to the other selected studies; had this been exluded, the analysis may have found weight gain to be significant. The Cawood et al. (2012) study was funded by and completed by industry professionals and was highly supportive of ONS. It would have been useful for Cawood to publish the results of the studies excluded due to location; this could show insight into how this influenced the conclusions of the study.

Only 3 studies were excluded on the grounds of not meeting quality standards. However, it is important to note that QA was the final round of study refinement and if significant ‘fatal flaws’ were noticed early on in the selection process, they were excluded before the official QA. In retrospect, it would have been useful to retain these studies until QA so that an analysis of the conclusions drawn by low quality studies could be compared to those of high-quality research. Each of the 3 low quality studies that were excluded concluded that ONS were very effective at improving nutritional outcomes; therefore, including these studies may have influenced the results of this analysis to become preferential to ONS.

Twelve studies were selected for analysis, this is less than usual industry standard for malnutrition (comparatively Hubbard et al.’s 2012 review of compliance to ONS contained 46 trials and Cawood et al.’s 2012 review on high protein ONS contained 36 trials). To draw conclusions with higher statistical power, it would have been useful to include more studies. However, this would have invalidated the research as it would not have conformed to the pre-determined search method and inclusion/exclusion criteria.

4.2 Meta-Analysis

4.2.1 Changes in nutritional outcomes

The second objective was to identify and evaluate the significance of changes in nutritional outcomes in the selected studies. The primary finding was that ONS improve nutritional outcomes compared to control groups, however these results are not always significant. Calculated means, box plots (Figures 2, 5 & 6) and the corresponding P values for weight, BMI and MUAC showed a general trend of improvement in the intervention group. Both median and mean values were higher in the intervention group for each outcome measure. However, no significant differences were found - weight (P=0.2), BMI (P=0.28), MUAC (P=0.81).

Figure 3 showed that the intervention group had more weight gain/retention than control groups in all but one study. The most significant change in weight between groups was reported by Jiang, the intervention group lost 5.26% body weight, compared to the control group who lost 12.92% body weight (P=0.036) Cameron et al. (2011) was the anomaly, reporting more weight loss in the intervention group than the control. However, there are two methodological factors that may have caused the disparity between this and the other studies. Firstly, the study was conducted for only 6 weeks, using acutely ill, fracture patients. Guidance provided by NICE stipulates that ONS should be prescribed for no fewer than 12 weeks to see improved nutritional outcomes (NICE, 2017). Furthermore, pain is a known factor for reduced appetite, thus contributing to increased weight loss in fracture patients (Carlsson et al., 2005). Secondly, the control group were subject to standard hospital care which included daily supplementation with high protein milk, although it was not stated how energy this provided in addition to usual diet. The ideal study would have included a ‘no intervention’ group, however this is a common difficulty when practicing within ethical guidelines (Medical Research Council, 2019). In the absence of a ‘no intervention’ group, more data surrounding the energy provided to the control group and patient compliance would have been useful to evaluate the true effect of the intervention.

Six studies reported weight gain, the most significant being Huynh’s study of malnourished, free living patients. At 4%, this was significantly higher than the 2.1% measured in the control group (P=0.0007). However, the average age in this study was 39 years. This is lower than the general population on ONS and the average age across the studies in this analysis, 72 years (Nutricia, 2020). Younger patients may be more able to return to a usual state of health without additional complications, therefore influencing the overall significance of the analysis (Boyd et al., 2008). Although contradictorily, Parson’s et al., (2017) included participants with the highest average age of 88.5 years and reported the second highest weight gain in the intervention group, with a significance between groups of P<0.05.

Figures 5 and 6 showed that results for BMI and MUAC were similar to the results for weight change. However, the limited data available for these measures lowered the statistical power of the results. For the data that was available, there was a non-significant positive trend in the intervention groups (P=0.28; P=0.81). Ultimately, BMI and MUAC were most useful to compare to trends in weight changes; the results for each outcome measure are proportionate to each other, this consistency across studies suggests that the validity of these results is high.

This analysis highlighted that ONS do not always enable patients to gain weight, but that they will usually help minimise weight loss. This finding is somewhat contradictory to the majority of research, which considers ONS very effective for treating malnutrition (Stratton and Elia, 2007). This may be due to differences in the evaluated outcome measures. Studies that show general trends of nutritional improvements often reveal significantly improved clinical and functional outcomes (Cawood et al., 2012;). Therefore, it can be assumed that although a key indicator of change, body composition cannot be the sole determinant of effectiveness. It would be more appropriate to acknowledge that body composition changes are a causal pathway, resulting in clinical and functional benefits (Cawood et al., 2012).

4.2.2 Nutritional outcomes by healthcare setting and disease state

The third objective was to assess the effect of different healthcare settings and disease states on nutritional outcomes. Published research that shows significant improvements in patient outcomes is usually conducted in specific settings and patient groups (Stratton and Elia, 2007). As results were pooled from a wide range of studies, this sub-analysis sought to see whether this may be a cause of disparities in results between this research and other studies.

This analysis found that community studies saw considerably less weight loss than the acute setting and care studies found slight weight gain. Weight change in the community varied significantly, with this analysis showing a range of 13.24%. This is attributable to the individual responsibility of preparing and complying to nutritional guidance. In care, the flexibility with food and nutrient choices for residents is typically higher than in hospitals. Therefore, there are different standards in care settings and achieving weight gain is a realistic target with supplementation. Hospital studies reported the most weight loss (mean -4.22%). This finding is consistent with existing literature; during hospitalisation almost half (45.5%) of patients lose weight due to changes in living situation, distress and acute illness (Leandro-Meari et al., 2015). In these cases, the dietetic focus is to minimise loss rather than attempt weight gain (Leandro-Meari et al., 2015). Furthermore, patients that undergo surgery (a high proportion of patients on ONS), can be expected to lose 15-20% body weight in the year following hospitalisation (Hatao et al., 2016). Comparatively, less weight loss is seen in community and care settings where acute pain and illness are less common (NICE, 2017). The differences in weight changes between healthcare settings found in this sub-analysis is logical and concurs with other research, however the differences are not statistically significant (P=0.152).

A further sub-group analysis was conducted, to investigate any weight change trends between different patient groups. Although all patients were malnourished, some also had additional health conditions, so the patient groups also included fracture and cancer. Figure 8 showed that patients being treated for malnutrition were able to gain weight (0.91%) but patients with cancer or fractures lost weight (-4.61%; -1.78%). These differences were statistically significant at P=0.0199.

There was a wide range of weight change in cancer studies, perhaps due to the diversity and complexity of the patients. Hatao and Zhu studied gastrointestinal cancer patients post operatively, with opposite results. Zhu strongly indicated (P=0.008) that ONS helped patients gain weight, while Hatao stated that ONS did not prevent patients from losing weight (P=0.26). Baldwin evaluated gastrointestinal and lung cancer patients and Jiang included head and neck cancer patients; both studies assessed patients during chemotherapy treatments and both determined ONS to be significantly effective (P>0.05; P=0.036). Chemotherapy and surgery are both known factors for changed appetite and gastrointestinal symptoms that cause weight loss (O’Gorman et al., 1998). As the severity of symptoms differs between patients and patient types, it is to be expected that some types of patient tolerate ONS better than others. Fracture patients appeared to lose some weight compared to their control groups; Wyers was preferential to the ONS group (P<0.01) and Cameron was preferential to the control (P=0.74). Myint did not report on weight change and so was excluded from this aspect of the quantitative synthesis, however they concluded that ONS were significantly effective for improving the BMI of fracture patients (P=0.0012). The group for which ONS was reliably effective was malnourished patients. All studies of malnutrition patients reported that weight was significantly improved in the ONS groups, incidentally all were also care studies. Therefore, a linear model was applied. This revealed that patient group, rather than healthcare setting was the key driver for weight change (P=0.0199).

4.2.3 Nutritional intervention and patient compliance

The final objective was to evaluate how nutritional outcomes changed based on the style of the intervention and the patient compliance to the intervention. The NICE recommendation is a kcal intake of >300kcal/day, however the actual prescribed intake varied from 250-600kcal. Hubbard et al. (2012) state that compliance decreases at higher prescriptions, so a Pearson’s test was used to examine whether there was a correlation between high energy prescription and decreased compliance; no correlation was found (P=0.63).

There was a weak, positive correlation between increased energy prescription and weight change (P=0.13). However, this does not consider the actual amount of product ingested and the effectiveness of ONS cannot be assessed without considering the adherence to the intervention. Figure 10 shows that the more significant correlation was compliance and weight change (P=0.008). Although this plot is not causative, it can be assumed that increased energy intake would be the influencing factor for positive weight change. Recent research often states that low compliance is a historical issue, however these results indicate that it is firstly, still low in some cases, and secondly, that it is a vitally important driver to see improved nutritional status (Hubbard et al., 2012).

4.3 Limitations

This study has several limitations. Firstly, a heterogeneous group of studies were included. This was a necessity as the aim was to gather data across all patients and settings. For systematic reviews that are inclusive of many types of studies, heterogeneous data is common. Therefore, when the objectives were formulated to include sub-group analyses to determine where the differences lay. (Cawood et al., 2012). There were two main reasons for this heterogeneity.

Firstly, outcome measures were presented in a millimetres, centimetres, inches, pounds, kilograms and percentages. To enable analysis, all data was converted to ‘percentage change’ and a series of sub-group analyses and a linear model were completed to determine the effects of different variables. However, this did mean that additional information such as standard deviations were no longer comparable after conversion. Secondly, there were significant differences between protocols for control groups. The gold standard ‘no intervention’ control would not meet requirements for ethical approval in cases of patients that meet the indication for intervention (Medical Research Council (2019). Therefore, control groups were subject to ‘usual care’ which included a range of treatments including micronutrient supplementation, high protein diets, additional snacks and high protein milk. Most studies did not report the additional energy provided to control groups, making analyses unrepresentative of true additional energy provided to the intervention groups. Cameron et al.’s findings were anomalous, and had they been excluded, the conclusions from the analysis may have been more preferential to ONS. The cause for this may have been attributable to the nature of their control group. The control group received an unspecified amount of high protein milk. Hornsby hospital, where the study was conducted, do not publish their nutrition protocols for inpatients but if compared to UK prescribing guidelines, it could be assumed that the control patients received an additional 200kcal per day (NICE, 2017). Consequently, there would only be a difference of 275kcal between groups, over 6 weeks this may not have been sufficient to show significant improvements.

The inclusion and exclusion criteria were extremely stringent to control unnecessary variables often found with heterogenous data. However, this presented a further issue. There was a considerable lack of data to draw conclusions with high statistical power. Only 12 studies met the criteria and of those, 11 reported weight changes (the primary outcome measure), 7 reported BMI, 6 reported MUAC and 8 reported the compliance to the interventions. Ideally a systematic review should be registered, therefore specifying a number of studies to be included would lower the validity of research as it would not conform to pre-specified methods provided in the registration. Despite being a university project, the intention was to mimic a formal systematic review, so the criteria remained stringent. However, this project was completed for academic and personal development, not industry affiliated and not set for publication. Therefore, it may have been appropriate to make minor adjustments to the inclusion and exclusion criteria so that more studies could be analysed.

4.4 Conclusions

This research can conclude that ONS have a positive effect on nutritional outcomes when prescribed to treat malnutrition, hip fracture and cancer. Measurements of weight, BMI and MUAC change all show a positive correlation when supplementing with ONS but these changes were not sufficient to be deemed significant. Patients in care settings see the most improvement, community settings see some improvement and hospitals only see a slight reduction in weight loss. However, the differences between healthcare settings are not significant.

The overall significant factors were disease state and compliance to the intervention. Oral nutritional supplements proved to be most effective in malnourished and fracture patients compared to cancer and the true effectiveness was shown by the correlation to compliance. This analysis showed that when patients adhered to the intervention, nutritional outcomes were improved most. Overall, it can be concluded that ONS are to somewhat effective when compliance is high, particularly in malnutrition and fracture patients.

4.5 Recommendations

Further work would expand on the data presented in this review, there are several key points of interest to take forward in subsequent research. Although the aim was to identify the effectiveness of ONS for all disease states, due to the study selection process only 3 patient groups were analysed in this review. Further research should ensure