Relationship between markers of malnutrition and clinical outcomes in older adults with cancer: systematic review, narrative synthesis and meta-analysis

Malnutrition predicts poorer clinical outcomes for people with cancer. Older adults with cancer are a complex, growing population at high risk of weight-losing conditions. A number of malnutrition screening tools exist, however the best screening tool for this group is unknown. The aim was to systematically review the published evidence regarding markers and measures of nutritional status in older adults with cancer (age ≥ 70). A systematic search was performed in Ovid Medline, EMBASE, Web of Science, CINAHL, British Nursing Database and Cochrane CENTRAL; search terms related to malnutrition, cancer, older adults. Titles, abstracts and papers were screened and quality-appraised. Data evaluating ability of markers of nutritional status to predict patient outcomes were subjected to meta-analysis or narrative synthesis. Forty-two studies, describing 15 markers were included. Meta-analysis found decreased food intake was associated with mortality (OR 2.15 [2.03–4.20] p = < 0.00001) in univariate analysis. Prognostic Nutritional Index (PNI) was associated with overall survival (HR 1.89 [1.03–3.48] p = 0.04). PNI markers (albumin, total lymphocyte count) could be seen as markers of inflammation rather than nutrition. There a suggested relationship between very low body mass index (BMI) (<18 kg/m2) and clinical outcomes. No tool was identified as appropriate to screen for malnutrition, as distinct from inflammatory causes of weight-loss. Risk of cancer-cachexia and sarcopenia in older adults with cancer limits the tools analysed. Measures of food intake predicted mortality and should be included in clinical enquiry. A screening tool that distinguishes between malnutrition, cachexia and sarcopenia in older adults with cancer is needed.


Introduction
Older adults with cancer are a growing population who require complex, multi-layered care to achieve the best possible clinical outcomes from anticancer treatment [1]. One important, but often overlooked, aspect of this is nutritional care, which has been consistently shown to be one of the most predictive and treatable components of comprehensive oncogeriatric assessment [2].
Malnutrition is caused by a lack of intake or uptake of nutrition [3,4], and risk screening is recommended [3] for all inpatients on admission and outpatients at their first appointment [5]. A number of malnutrition screening tools exist [6,7], although the most appropriate tool for identifying malnutrition in older adults with cancer is unknown. The varying diagnostic criteria for malnutrition between screening tools is reflected in the varying prevalence estimates; for example, the prevalence of malnutrition in older adults with gastrointestinal cancer varies between 20 and 52%, depending on the screening tool [8].
Malnutrition screening tools have often been validated against the subjective global assessment (SGA) [9]. The SGA was initially validated for use in end-stage renal disease [10], but has recently been shown to be less reliable than other nutritional screening tools to predict clinical outcomes in certain populations [11], such as the NRS-2002 screening tool which possesses higher specificity and positive predictive value for post-operative complications [12], and mortality [13] in hospitalised patients.
As well as varying markers, the marker thresholds used to determine nutritional risk differ between tools. For example, with regard to weight loss, the British Association for Parenteral and Enteral Nutrition screening tool uses any unintentional weight loss [14]; the Short Nutritional Assessment Questionnaire uses >3 kg in 1 month or >6 kg in 6 months [15]; the 3 Minute Nutrition Screening uses >7 kg in an unspecified time frame [16]; and the European Society for Clinical Nutrition and Metabolism (ESPEN) screening tool uses >10% in an unspecified time frame [17]. Older adults with cancer exhibit further complexity given their higher risk of other weight-losing conditions, including sarcopenia and cachexia due to cancer or other co-morbidities. Cachexia, sarcopenia and malnutrition have similar clinical presentations and diagnostic criteria [18,19]. However, malnutrition has a specific focus on the 'intake and utilisation' of nutrition, therefore a screening tool that can also identify problems with oral intake is required.
To establish which screening tool is most appropriate to identify malnutrition in older adults with cancer, markers of malnutrition and their thresholds must be investigated in relation to their ability to predict poorer clinical outcomes. The objective of this systematic review is to identify and synthesise the published evidence about markers of nutritional status in the older cancer patient. The findings will inform the most appropriate nutritional screening tool to use in this population.

Methods
The study protocol was registered with PROSPERO [20], and is reported in accordance with the Preferred Reporting Items for Systematic Review and Meta-Analyses (PRISMA) guidelines [21].

Literature search
Searches were performed by AB and SG between the 6th and 8th December 2018, from data-based inception to search date in; Ovid ® MEDLINE (Ovid MEDLINE ® ) and An initial search combining keywords related to malnutrition, cancer and older adults, using MeSH and text terms was conducted. On review of the findings, an additional supplementary search was conducted to include text terms for individual screening tools that were previously identified. See online Supplementary information 1 for the final MEDLINE search strategy. Forward and backward citation searching of all included studies, and relevant systematic reviews [22][23][24], was completed: we examined the reference lists of included studies and identified articles citing included studies in Web of Science.

Inclusion and exclusion criteria
Eligible studies had participants aged 70 years or older with any cancer diagnosis. Studies investigating markers of nutritional status, used in nutritional screening tools or objective nutritional indexes [6,7], against any patientrelated outcome were included. All observational studies were included, and randomised control trials (RCTs) were included if study interventions were not nutrition related (e.g. nutritional interventions). Editorials, case studies, case reports and conference abstracts without subsequent full text publication were excluded along with review articles. Nutritional markers used in screening tools such as disease state and functional performance were excluded as all participants had cancer diagnoses. The relationship between functional performance and patient outcomes is an established individual risk factor for poor patient outcomes [25].

Study selection
All titles and abstracts retrieved by electronic searching were downloaded to an Endnote X8 library and duplicates were removed according to a published protocol [26]. The remaining records were uploaded to the online citationscreening tool Abstrackr [27]. Studies were initially dual screened independently (by AB and SG) on the basis of title and abstract against the eligibility criteria. Where one or more of the investigators were uncertain whether the article met the inclusion criteria, the abstract was included and the full-text article was included for review. All potentially relevant studies were retrieved and full-texts were reviewed by AB and SG, with any unresolved disagreements resolved by consensus or adjudication by a third reviewer (MJ).
Data were extracted by AB, using a custom data extraction form [20]. Data extraction was piloted, reviewed and modified before a final extraction from the main papers of the included studies, with use of supplementary materials as necessary.

Risk of bias; quality appraisal
Each study was evaluated using the Critical Appraisal Skills Program checklist [28] items 1-10. The cohort study checklist was used for all study designs. All included papers were evaluated by AB with a random 25% independently reviewed by GM. See online Supplementary information 2 for quality assessment of studies.

Analysis
A narrative summary with descriptions and comparisons was completed. Meta-analyses were conducted with sufficient study data (n ≥ 3 studies) with homogeneity of proxy marker thresholds and patient outcomes. Review Manager 5.3 [29] was used to conduct meta-analyses. The I 2 statistic was used to assess heterogeneity, with a random-effects model chosen if significant heterogeneity was indicated [30]. Results were considered significant if confidence intervals did not include the null value, with corresponding significance values of p < 0.05.

Dietary intake
Two studies [33,35] investigated five markers of food intake: declining [33] or decreasing food intake, number of daily full meals, protein-rich food intake, fruit and vegetable intake and mode of feeding [35]. Only one study [33] performed multivariate analysis, observing 'declining food intake' to be associated with overall mortality. All other markers of food intake reported associations between patient mortality and declining food intake, regardless of the threshold or marker used for food intake. Two studies [33,35] investigated three comparable scales of declining food intake at univariate level, allowing meta-analysis of results.

Meta-analysis
A random-effects model was used to combine odds ratios (ORs) for mortality, with meta-analysis suggesting that declining food intake is associated with worse increase risk of mortality in univariate analysis (OR 2.15 [95% CIs 1.61-2.86, p = < 0.0001]), Fig. 1.
Three studies [33,35,42] investigated the relationship between fluid intake and patient outcomes; finding an association in two studies between fluid intake <3 cups/day with chemotherapy toxicity in univariate analysis [42], and fluid intake <5 cups/day with overall mortality in univariate analysis [33]. However, one study observed no relationship between fluid intake and mortality [35].

Meta-analysis
Two studies investigated PNI to predict risk of postoperative complications, although this only met statistical significant in univariate analysis [58,61].

Controlling nutritional status score (CONUT)
One study [56] reported an association between CONUT and OS in multivariate analysis, but no relationship with post-operative complications. A second smaller (n = 68) study [57] found no association between CONUT and OS or cancer-specific survival.

Nutritional risk index (NRI)
Two studies investigating NRI found low NRI was associated with worse patient outcomes [36,68]. One [68] investigated NRI as a predictor of outcomes after anticancer therapies in oesophageal cancer and found that NRI was associated with poorer 2-year OS and distant metastasis-free survival in multivariate analysis. The second [36] undertook a smaller study (n = 71) and found low NRI to be associated with post-operative complications in univariate analysis, but not with either major or infectious complications.

Body mass index (BMI)
Due to variable BMI thresholds and patient outcomes, metaanalysis of results was not possible. Four studies [44,45,47,50] conducted multivariate analysis of BMI on patient outcomes; with one [45] finding an association between BMI < 18 kg/m 2 and death within 3 months of surgery. Another found BMI < 18 kg/m 2 associated with shorter survival [47]. Multivariate analysis also identified associations with BMI and OS [50] and the clinical decision of active versus palliative treatment [44].
Participants in the three studies [45,47,56] investigating BMI < 18 kg/m 2 on patient outcomes were all diagnosed with NSCLC. These studies observed associations between low BMI and poorer patient outcomes.

Weight loss
Only one study [45] conducted multivariate analysis of weight loss on patient outcomes. A 5% weight loss in 3 months was associated with post-operative early death within three months [45].
Three studies investigated the effect of weight loss on mortality. Two studies [33,35] found an association between weight loss and mortality, where weight loss of between 5 and 10%, >10%, >3 kg or unknown weight loss were associated with 1-year mortality [35]. Weight loss in the past 6 months was also associated with mortality [33]. The largest study, of 12,979 patients with colon cancer reported no association between 'weight loss' and 90-day or 1-year mortality rates [52]. Three studies [36,61,67] investigating weight loss and treatment complications found no association.

Mid arm circumference (MAC) and calf circumference (CC)
Only one study investigated MAC and CC in relation to patient outcomes [73], finding CC < 31 cm and MAC < 21 cm to be associated with mortality in patients receiving chemotherapy in univariate analysis.

Muscle strength
Two measures of muscle strength were identified in the reviewed articles; hand-grip strength [39,70] and lean skeletal muscle-mass by CT [51]. A pilot study with 24 participants found no association between grip-strength and chemotherapy toxicity [39]. Two studies reported associations between lean skeletal muscle mass with POD in multivariate analysis [51], and grip-strength with caregiver burden in univariate analysis [70].

Albumin
Four studies [31,32,34,45] conducted multivariate analysis of albumin to predict patient outcomes; with only one study [34] finding an association with OS, and one study with major post-operative complications [45]. No association with mortality [31,32], completion of chemotherapy [31,32] or death within 3 months of surgery were found [45]. Univariate associations between Alb and postoperative and chemotherapy-related complications were seen in four studies [40,58,61,67], and OS in two [41,72]. There were no observed associations between Alb and OS or disease-free survival [46], functional decline [38], or chemotherapy toxicity [37] in three other studies. Thresholds of Alb varied between 35 [31] and 40 g/l [40].

C-reactive protein
An association between increasing CRP and OS was seen in one study [34] through multivariate analysis. There were no observed relationships between CRP and OS [46] or functional decline [38].

Discussion
Forty-two papers, representing 21,032 participants, investigating the associations of 15 makers of nutritional status with patient outcomes, were identified for review. Our meta-analysis of three questions regarding declining food intake shows an association between reduced food intake and mortality, but does not assess utilisation. Our metaanalysis of four studies shows an association between poorer PNI scores and clinical outcomes, but this score measures inflammatory markers (which may indicate increased energy requirement) but does not assess poor oral intake. PNI alone, therefore cannot distinguish between cachexia and malnutrition).
Measures of dietary intake and utilisation are essential in diagnosing malnutrition, as these changes in consumption or assimilation can lead to net calorific deficit and consequent weight loss. Assessments of eating and drinking, despite being a direct measure of intake, are inadequately, assessed in commonly used malnutrition screening tools (e.g. ESPEN criteria, MUST). Several screening tools included an assessment of appetite. Appetite may correlate with dietary intake in patients with cancer, although it is only a proxy marker of malnutrition; for example a patient with dysphagia due to localised oesophageal cancer may be hungry but unable to eat. Food and fluid intake arguably have the greatest face and content validity for determining nutritional risk. From the available evidence, there appears to be some evidence that reduced food and fluid intake were associated with adverse patient outcomes in older adults with cancer, with meta-analyses suggesting an association between declining food intake with mortality, However, there is an urgent need for more evidence, and in particular studies which appropriately control for potential confounding variables via multivariable analyses.
Whilst proxy markers of malnutrition can be easily used and are commonly available, their value against direct anthropometric markers or measures of food and fluid intake is limited, see Table 3 for comparison of malnutrition screening tool and objective indexes content, compared with malnutrition markers identified in this review.
PNI was devised in 1984 as a risk score relating postoperative complications with baseline nutrition, using albumin and lymphocyte counts [7]. Our finding of an association between low PNI and worse OS is consistent with other recent meta-analyses of all adults with cancer undergoing surgery [74][75][76]. Albumin and common laboratory tests for inflammation (e.g. CRP and white cell counts) are useful as predictors of prognosis in people with cancer e.g. Glasgow Prognostic Score [77]. However, they are not specific to malnutrition and are not recognised as a diagnostic markers for malnutrition [78].
The single biomarkers identified in this review suggest no clear association with patient outcomes. Although reduced haemoglobin can be caused by dietary deficiency, it may also be a feature of inflammation, chronic disease, bone marrow suppression from anticancer treatments and other wasting diseases (e.g. cachexia and sarcopenia [79,80]). Although the clinical presentation of malnutrition, cachexia and sarcopenia overlap, Table 4, the management of each differs [4,19,79,80]. Therefore, the use of nonspecific biochemical and clinical markers, or objective indices, which identify inflammation-albeit giving information about increased metabolic and therefore nutritional requirements-tell us nothing about dietary intake. Therefore, in the absence of information about dietary intake, they may reduce the specificity for malnutrition in an older population at high risk of all three conditions. Four anthropometric markers were examined in this review: BMI, weight loss, MAC and CC. We found weight loss was associated with worse clinical outcomes in older adults with cancer. The varying thresholds in required percentage weight loss and the timeframes for weight loss used in the analysed literature, precluded meta-analysis or identification of an appropriate threshold for weight loss to indicate malnutrition in older adults with cancer. However, weight loss does have face validity as a marker of malnutrition. Weight loss is used in most malnutrition screening tools [6].
As with weight loss, varying thresholds prohibited metaanalysis of BMI. We found a low BMI (<18 kg/m 2 ) predicts poorer outcomes, particularly in lung cancer patients [45,47,56]. MAC is known to correlate with BMI in hospital inpatients [81]. BMI is a simple measure, easy to implement in clinical practice but does not differentiate between fat and muscle and repeat measures are needed to be clinically useful. Adiposity mass increases with age and muscle decreases without significant changes to BMI [82,83], and the presence of sarcopenic obesity should be considered.

Strengths and limitations
A strength of this study was the broad inclusion criteria of patients with any cancer diagnosis, markers of nutritional status and patient outcomes. This allowed a comprehensive analysis of potential markers of nutritional status, and appraisal of the evidence surrounding the validity of outcomes in older adults with cancer. We chose to focus on adults aged 70 years and over with cancer as this population is both growing and complex; we address an important clinical issue and identify a gap in clinical practice. This patient group may present with multimorbidity and coexistent cachexia and sarcopenia. Cancer patients are frequently neglected from clinical trials and surgical and pharmacological interventions require correction of nutritional deficits before treatment commences.
There are a number of limitations. Firstly, due to the heterogeneity in markers, marker thresholds, cancer diagnoses, treatment types and study quality, meta-analysis of most extracted data was not possible. Secondly, our aim was to study malnutrition, therefore the search strategy was not designed to capture all studies of general prognostic markers in older adults with cancer. Few studies included biomarkers. We acknowledge that some studies investigating Hb, Alb and CRP outside of a focus on malnutrition may have been missed for this population. However, we are unlikely to have missed any critical markers of malnutrition. Finally, although lower weighting was given to lower quality studies within results synthesis, due to the number of lower quality studies, results may be treated with caution.

Implications for clinical practice and research
Measures of dietary intake should be sought as part of routine nutritional assessment. The appropriateness of using 'proxy' markers of malnutrition should be reconsidered, especially those overlapping with inflammation in older adult patient groups with co-morbid conditions or acute illness. Further research is required into the appropriate thresholds for markers of nutritional status in this complex population. A screening tool that can identify and   differentiate between malnutrition, cachexia and sarcopenia in older adults with cancer, and which is usable in clinical practice, may allow targeted and appropriate treatment of these conditions. Currently, there is none which can assess all three conditions.

Conclusion
We could not identify a single tool suitable to screen for malnutrition risk in older adults with cancer. Markers of inflammation and measures or oral intake are used and are associated with clinical outcomes. However, alone, they cannot distinguish between risk of malnutrition, sarcopenia and cachexia (which may co-exist in older adults with cancer). Dietary intake measures in conjunction with others, which measure nutritional utilisation, would be helpful. The value, and best way, of differentiating between malnutrition, cachexia and sarcopenia for older adults with cancer remains unanswered.
Funding This work was funded by Yorkshire Cancer Research (Award reference number HEND405PhD). Publication costs will be provided by Yorkshire Cancer Research TRANSFORM programme.
Author contributions AB and MJ designed the project. AB, SG and MJ designed the protocol. AB and SG conducted the review. AB and GM performed the extraction of data. AB performed the analysis. AB wrote the manuscript. All authors revised the manuscript critically. AB and MJ has overall responsibility for the final content.

Compliance with ethical standards
Conflict of interest The authors declare no conflict of interest.
Publisher's note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons. org/licenses/by/4.0/.