Introduction

Malnutrition is prevalent among cancer patients, with a reported incidence ranging from 39% to 87%.1 Malnutrition not only reduces the efficacy of antitumour treatment and increases the length of hospital stay but also reduces the quality of life and survival of patients.2-5 Nutritional screening is a process to fast detect whether a patient at risk of malnutrition or not, whereas nutritional assessment finally determines nutritional status. Currently, there is no gold standard for assessing the nutritional status of cancer patients, although various nutritional assessment tools have been developed.6 The Patient-Generated Subjective Global Assessment (PG-SGA), a modified version of the Subjective Global Assessment (SGA), is the most widely used tool for assessing the nutritional status of cancer patients2-5 because it shows better sensitivity, specificity, positive, and negative predictive values than other tools.1 However, it has been considered overly time consuming in China and other countries.2, 7, 8 In addition, some items of the PG-SGA are perceived as hard to comprehend even by well-trained professionals.9

The abridged PG-SGA (abPG-SGA) is a modified version of the PG-SGA that uses only the patient-generated component (Boxes 1–4) to provide a simplified tool for nutritional assessment.10, 12 Moreover, the abPG-SGA is used as a nutritional screening tool rather than a nutritional assessment tool.11, 12 Therefore, there is a need to develop a simple validated version of the PG-SGA to be used as a nutritional assessment tool for cancer patients. The Chinese Society of Nutritional Oncology (CSNO) took on this initiative to develop a modified version of the PG-SGA without compromising its validity.10

In this study, we first surveyed clinical nutrition practitioners about their perception and use of the PG-SGA and requested advice regarding how to modify it. We then further selected individual items by rigorous statistical analyses. Subsequently, we generated and validated the modified PG-SGA (mPG-SGA) as a nutritional assessment tool for cancer patients.

Materials and methods Population

The Investigation on the Nutritional Status and Clinical Outcomes of Common Cancers (INSCOC) is an ongoing national survey conducted in China by the CSNO.13 The present study was part of the INSCOC and was conducted between May 2013 and April 2019 in 72 tertiary hospitals across China. Participants were older than 18 years with a pathologically confirmed diagnosis of cancer and were conscious, were able to communicate in Chinese, were willing to participate in this study and provided written informed consent. Participants included patients with first cancer diagnosis or recurrent disease with any tumour comorbidities. The exclusion criteria were those who underwent organ transplantation, pregnant women, or those who were admitted to the intensive care unit (ICU) at the beginning of recruitment. All admitted patients were interviewed by professionals to complete formatted questionnaires including the PG-SGA, Nutritional Risk Screening 2002 (NRS 2002), Eastern Cooperative Oncology Group (ECOG) performance status (PS), Karnofsky performance status (KPS), and others. Anthropometric measurements were performed by trained medical professionals. The anthropometric measurement included height, body weight, body mass index (BMI), upper midpoint arm circumference, triceps skin fold thickness (non-dominant arm), and hand grip strength (non-dominant hand). Laboratory tests were performed immediately after admission by the clinical labs of the participating hospitals, including haemoglobin, albumin, total protein, and C-reaction protein. All professionals involved in the nutritional status assessment are certified nutritional therapists (registered dietitian/doctors/nurses) with nutrition assessment qualifications and experiences. A total of 41 117 patients were initially included in the study. Patients with incomplete questionnaires (n = 4 669) or no data for haemoglobin or albumin (n = 2377) were excluded from the study. The current analyses included 34 071 patients. Among them, 1558 of the patients were followed up for 5 years after admission.

Patient-Generated Subjective Global Assessment score

The PG-SGA score is made up of the scores of seven separate domains. The domains are about weight loss (Box 1, Worksheet 1), food intake (Box 2), symptoms (Box 3), activities and functions (Box 4), disease (Worksheet 2), metabolic demand (worksheet 3), and the results of a physical examination (Worksheet 4). Based on the PG-SGA score, nutritional status is generally categorized into four groups14: well-nourished (0–1 points), mildly malnourished (2–3 points), moderately malnourished (4–8 points), and severely malnourished (≥9 points).

Item selection and generation of the Modified Patient-Generated Subjective Global Assessment

We first conducted a survey on the use of the PG-SGA among medical staff across the country. Seventy healthcare professionals (details shown in Figure S1 and Table S2) from 23 Chinese provinces were asked to complete a questionnaire consisting of 24 questions (Table S3) on the content validity, comprehensibility and difficulty of completing the PG-SGA (i.e. Boxes 1–4 and Worksheets 1–4). In addition, eight open-ended questions were posed after each box and worksheet to ask for the professionals' perceptions regarding its utility.

The medical professionals were also asked about whether their patients or their patients' family members could complete the patient-generated components (i.e. Boxes 1–4) independently during routine clinical work. After the results of the questionnaires were recorded, research meetings were set up to discuss which item(s) should be deleted or included or if new items should be added. The individual components (i.e., Boxes 1–4 and Worksheets 1–4) would be adjusted if more than 5% of professionals expressed doubts about their comprehensibility and would be deleted if more than 50% of professionals thought they were difficult to complete. After discussion, imperfect items were selected for further reliability analyses. Items were selected for inclusion if the item-total correlation was >0.1 and Kendall's tau-b rank correlation was >0.1; otherwise, the items were deleted. Finally, all selected items were used to generate the new mPG-SGA.

Validation in the Modified Patient-Generated Subjective Global Assessment

The internal validity of the newly generated mPG-SGA was first tested using Pearson's correlation analysis by comparing the total mPG-SGA score and the scores from each section (i.e. Boxes 1–4, age). The external validity was examined using Pearson's correlation by comparing scores of the mPG-SGA scores with the scored PG-SGA, global PG-SGA rating, NRS 2002, and KPS. For the test–retest validity, the mPG-SGA was conducted in consented participants on a single day, followed by a repeated assessment by the same investigator and other professionals on a different day within 1 week. The assessment days were separated by at least 1 days without chemotherapy, surgery, or other operations that might obviously affect the results of the patient's nutritional assessment. The test–retest validity can be also examined by comparing scores of Box 4 of the mPG-SGA with the KPS score and the ECOG PS score from the same patients.

The performance of the mPG-SGA was compared with that of the PG-SGA based on the area under the curve (AUC). The AUC of the abPG-SGA (Boxes 1–4 of the PG-SGA) was calculated. To compare the completion time of the questionnaire, the mPG-SGA was first conducted in consented participants on a single day and then the PG-SGA was conducted on the next day by the same professional. The ability of the mPG-SGA to categorize patients with different nutritional statuses was also evaluated by comparing the average nutritional parameters of each group, such as the albumin, triceps skinfold thickness (for estimate of subcutaneous fat), and hand grip strength (for estimate of muscle strength).

Finally, the relationship between the nutritional status evaluated by the mPG-SGA and patient overall survival (OS) was examined using Kaplan–Meier methods. OS was defined as the time from the date of diagnosis to the date of death from any cause, the date of the last follow-up or 30 April 2019, whichever came first. The ability to predict survival based on the categorization of the patient's nutritional status was also simultaneously compared among mPG-SGA, abPG-SGA and the original PG-SGA.

Statistical analysis

The item-total correlation and Cronbach's alpha coefficient were used to test the reliability and internal consistency of the PG-SGA. Kendall's tau-b was used to identify the rank correlation between each item and the variables in the PG-SGA score (categorized into four ranges: 0–1, 2–3, 4–8, and ≥9). External and internal consistencies were examined by concordance analyses. The mPG-SGA was validated using the original PG-SGA. Receiver operating characteristic curves were used to compare the sensitivity and specificity of the ability to accurately identify different nutritional statuses across the mPG-SGA box combinations. The sensitivity and specificity of the mPG-SGA in comparison with the PG-SGA were tested by AUC analyses. The cut-off scores were calculated from the points of maximal specificity and sensitivity (Youden's index). When assessing the consistency between nutritional statuses determined by the mPG-SGA, the abPG-SGA and the original PG-SGA, weighted Kappa values <0.5, 0.5–0.75, 0.75–0.9, and >0.9 were considered poor, moderate, good and excellent, respectively.15 Test–retest reliability was examined using Spearman correlation coefficients. Two related samples nonparametric test was used to compare scores and time. OS was evaluated by Kaplan–Meier methods. P values ≤0.05 (two-sided) were considered statistically significant. All analyses were conducted using SPSS 26 (IBM Corp, Armonk, NY, USA).

Role of the funding source

The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding authors had full access to all the data in the study and had final responsibility for the decision to submit for publication.

Results

The characteristics of patients recruited for this study are shown in Table 1. Overall, 82% (27 935 out of 34 071) of the participants were malnourished; 58% were moderately or severely malnourished. The average PG-SGA scores and percentages of patients with moderate or severe malnutrition in different tumour locations are shown in Table S1. The percentages of patients with moderate or severe malnutrition ranged from 36% in breast cancer to 80.7% in oesophageal cancer.

Table 1. Patient characteristics Characteristics n (%) Age, older than 65 years 8476 (24.9) Sex, male 18 794 (55.2) Primary tumour location Pancreatic cancer 468 (1.4) Biliary cancer 121 (0.4) Oesophageal cancer 2512 (7.4) Gastric cancer 4517 (13.3) GIST 45 (0.1) Colorectal cancer 6686 (19.6) Liver cancer 1315 (3.9) Brain cancer 342 (1.0) Leukaemia 871 (2.6) Lung cancer 6913 (20.3) Ovarian cancer 778 (2.3) Malignant lymphoma 1009 (3.0) Cervical cancer 1434 (4.2) Endometrial cancer 404 (1.2) Prostate cancer 298 (0.9) Bladder cancer 270 (0.8) Nasopharyngeal carcinoma 2308 (6.8) Breast cancer 3687 (10.8) Other cancer 1022 (3.0) Nutritional status (PG-SGA score) Well-nourished (0–1 point) 6136 (18.0) Mild malnutrition (2–3 points) 8084 (23.7) Moderate malnutrition (4–8 points) 11 095 (32.6) Severe malnutrition (≥9 points) 8756 (25.7) Recent treatment Surgery 8756 (25.7) Chemotherapy 17 399 (51.1) Radiotherapy 4490 (13.2) Use of nutritional support 25 017 (73.4) PG-SGA, Patient-Generated Subjective Global Assessment.

Figure 1 shows the results of the evaluation of the original PG-SGA questionnaire by healthcare professionals. In general, the assessment of Boxes 1–4 and Worksheet 1 of the PG-SGA fell into the predefined acceptance category. Almost all professionals (>95%) acknowledged the content validity, comprehensibility, and low difficulty for professional completion of these parts of the assessment. However, the professionals often doubted that the patients and family members themselves would be able to complete the patient-generated components without help from professionals (61%, 43%, 47%, and 54% for Boxes 1–4, respectively). Although 69% of professionals were concerned about it being difficult for patients or their families to complete the nutritional assessment by themselves, most professionals (99%) agreed that patients or their families should be encouraged to complete it by themselves (Figure 2); 69% to 79% of the professionals acknowledged the comprehensibility of Worksheets 2–4. However, 57% of professionals thought it was difficult for medical staff to complete Worksheet 4. Thus, the items in Worksheet 4 were deleted. More professional opinions are shown in Table S4.

The content validity, comprehensibility, and difficulty of each component of the Patient-Generated Subjective Global Assessment as perceived by healthcare professionals in China.

Results to the question ‘Should patients or their families be encouraged to complete the nutritional assessment by themselves?’.

As shown in Table 2, the internal consistency of PG-SGA items was acceptable (Cronbach's alpha = 0.648). The components and items with an item-total correlation <0.1 and Kendall's tau-b rank correlation <0.1 was deleted (‘mouth sores’ and ‘other’ in Box 3, all items in Worksheet 2 except for ‘age greater than 65 years’, and items in Worksheet 3 of the PG-SGA). Finally, the new mPG-SGA (Table 3) was constructed by combining Boxes 1–4 and the item ‘age greater than 65 years’ and excluding items ‘mouth sores’ and ‘other’ in Box 3.

Table 2. Internal consistency of the PG-SGA items and Kendall's tau-b rank correlation between each item and categorized PG-SGA level Items of PG-SGA Mean (SD) Corrected item-total correlationa Cronbach's alpha if item deleteda Kendall's tau-bb Box 1. Weight (& Worksheet 1) 1.11 (1.64) 0.407 0.627 0.597 Box 2. Food intake 0.73 (0.96) 0.593 0.574 0.656 Box 3. Symptoms No appetite 0.57 (1.18) 0.362 0.619 0.511 Nausea 0.10 (0.31) 0.345 0.634 0.312 Vomiting 0.18 (0.71) 0.289 0.629 0.294 Mouth sores 0.02 (0.22) 0.066 0.649 0.054 Constipation 0.08 (0.27) 0.216 0.642 0.212 Diarrhoea 0.11 (0.55) 0.103 0.649 0.191 Dry mouth 0.08 (0.27) 0.185 0.644 0.182 No taste 0.06 (0.23) 0.215 0.643 0.210 Smells bother me 0.02 (0.16) 0.155 0.647 0.141 Problems swallowing 0.11 (0.45) 0.157 0.644 0.223 Full quickly 0.06 (0.23) 0.195 0.644 0.218 Pain 0.29 (0.89) 0.224 0.640 0.336 Other 0.03 (0.17) 0.077 0.649 0.073 Box 4. Activities and function 0.47 (0.77) 0.471 0.603 0.473 Worksheet 2. Disease Age older than 65 years 0.25 (0.43) 0.108 0.647 0.143 Other diseases 0.04 (0.20) 0.012 0.651 0.025 Worksheet 3. Metabolic demand 0.06 (0.42) 0.090 0.649 0.069 Worksheet 4. Physical exam 0.54 (0.71) 0.374 0.618 0.307 PG-SGA, Patient-Generated Subjective Global Assessment. a By reliability statistics, Cronbach's alpha = 0.648. b Well-nourished or mildly malnourished (0–3 points), moderately malnourished (4–8 points), and severely malnourished (≥9 points). Table 3. Modified Patient-Generated Subjective Global Assessment (mPG-SGA) Box 1. Weight Wt loss past month Points Wt loss 6 months Box 1. Score 1.1 A summary of my current and recent weight: 10% or greater 4 20% or greater I currently weigh about _______ pounds 5–9.9% 3 10–19.9% I am about ____ feet ____ inches tall 3–4.9% 2 6–9.9% One month ago, I weighed about _________ pounds 2–2.9% 1 2–5.9% Six months ago, I weighed about _________ pounds 0–1.9% 0 0–1.9% 1.2 During the past two weeks my weight has: Use the 1 month weight data if available. Use the 6 month data only if there is no 1 month weight data. Add one extra point if the patient has lost weight during the past 2 weeks. Decreased (1) not changed (0) increased (0) Box 2. Food Intake: Compared to my normal intake, I would rate my food intake during the past month as follows: Box 2. Score Unchanged (0) Use the highest score checked, no matter how many are checked More than usual (0) Less than usual (1) If less than usual, I am now eating: Normal food but less than the normal amount (1) Little solid food (2) Only liquids or nutritional supplements (3) Very little of anything (4) Only tube feedings or only nutrition by vein (0) Box 3. Symptoms: I have had the following problems that have kept me from eating enough during the past 2 weeks (check all that apply): Box 3. Score No problems eating (0) Add all points for Box 3 to obtain a total score No appetite, just did not feel like eating (3) I have had: nausea (1) vomiting (3) constipation (1) diarrhoea (3) dry mouth (1) problems swallowing (2) pain; where? (3) __________________

Things taste funny or have no taste (1)

Smells bother me (1)

I feel full quickly (1) Box 4. Activities and function: Over the past month, I would generally rate my activity as: Box 4. Score Normal with no limitations (0) Not my normal self but able to be up and about with fairly normal activity (1) Not feeling up to most things but in bed or chair less than half the day (2) Able to do little activity and spend most of the day in bed or a chair (3) Pretty much bedridden, rarely out of bed (3) Box 5. Age Box 5. Score Age older than 65 years (1) 0 = Well-nourished; 1–2 = mildly malnourished; 3–6 = moderately malnourished; ≥7 = severely malnourished Total score/level: /

The internal consistency and external consistency were estimated by calculating the correlations of the mPG-SGA score with the individual components and other performance scores (Table 4). The correlation between the mPG-SGA score and the overall PG-SGA rating (A, B, and C) was 0.625 (P < 0.001), and the correlation with the total PG-SGA score was 0.984 (P < 0.001). The correlations between the mPG-SGA score and its individual components were significant (P < 0.001) and were 0.684, 0.722, 0.842, 0.589, and 0.189 for Boxes 1–4 and ‘age more than 65’, respectively. The correlations with other performance scores (NRS 2002, KPS) were weaker but still significant (P < 0.001). Box 4 of the PG-SGA (the activity and functioning of patients) asked the same questions as the KPS but in a different way. By re-categorizing the KPS as an ECOG performance score (PS), the alternate-form reliability was able to be examined. Good external consistency was shown between Box 4 of the PG-SGA and the KPS and the ECOG PS (Pearson r = 0.626 and 0.568 respectively, P < 0.001) (Table S5). Test–retest reliability was investigated in 134 patients, which included 89 patients for inter-rater and 45 for intra-rater testing. No significant difference and strong correlations were observed in the total mPG-SGA scores between two measurements by five independent raters (Spearman correlation coefficient = 0.964, Table 6A) and between the repeated measurements by the same professional (Spearman correlation coefficient = 0.995, Table S6B). The times needed to complete the questionnaire were recorded for 30 patients. The mean time to complete the mPG-SGA was significantly shorter than the PG-SGA [281 (±59) seconds vs. 411 (±77) seconds, P < 0.001] (Table S7 ).

Table 4. Correlations (Pearson, r) of the total mPG-SGA score with individual components and other indexes and performance scores Total mPG-SGA score Total PG-SGA score Global PG-SGA rating (A, B, C) Box 1. weight loss Box 2. food intake Box 3. symptoms with 2 deleted Box 4. activities and function Age older than 65 NRS 2002 score KPS Pearson r 0.984 0.625 0.684 0.722 0.842 0.589 0.189 0.501 0.441 P <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 N 34 071 34 071 34 071 34 071 34 071 34 071 34 071 34 071 34 071 Abbreviations: KPS, Karnofsky performance status; PG-SGA, Patient-Generated Subjective Global Assessment; mPG-SGA, modified PG-SGA.

The combinations of box scores that met the sensitivity and specificity criteria are described in Table 5. Different cut-off scores were used for these variables to determine the best sensitivity, specificity, and accuracy compared with the categorized PG-SGA scores. Four nutritional status classifications were formed, that is, well-nourished (0 points), mildly malnourished (1–2 points), moderately malnourished (3–6 points), and severely malnourished (≥7 points), based on the analyses of the area under curve (0.962, 0.989, and 0.985 for the three cut-off scores, respectively) and the point of maximal sensitivity (0.924, 0.918, and 0.945) and specificity (1.000, 1.000, and 0.938) of the mPG-SGA scores. Compared with the original PG-SGA, the mPG-SGA showed better consistency than the abPG-SGA (weighted kappa: 0.881 vs. 0.830). However, the accuracy was reduced when the weight loss score (Box 1) was removed (weighted kappa <0.8). We evaluated the diagnostic consistency between mPG-SGA and PG-SGA for 18 types of malignant tumours. The weighted kappas were ranged from 0.854 to 0.946, which indicated excellent consistency (Table S8). Thus, the items selected for inclusion in the mPG-SGA and the scoring were suitable in most common tumour types. The ability of the mPG-SGA to categorize patients into different nutritional status groups according to the anthropometric parameters is shown in Table S9. Again, the discriminatory ability of the majority of the nutritional parameters was highly significant based on the evaluation of these objective parameters (P < 0.001).

Table 5. Area under receiver operating characteristic curve and the sensitivity, specificity and agreement with the PG-SGA in the prediction of malnutrition (PG-SGA well-nourished or mildly, moderately, or severely malnourished) Method Well-nourished or mildly malnourished (PG-SGA score 0–1 points or more)a Mildly or moderately malnourished (PG-SGA score 0–3 points or more)a Moderately or severely malnourished (PG-SGA score 4–8 points or more)a Weighted AUC Cut-off score Sensitivity Specificity AUC Cut-off score Sensitivity Specificity AUC Cut-off score Sensitivity Specificity kappa A. abPG-SGA (Boxes 1–4)b 0.927 0.5 0.854 1.000 0.986 1.5 0.965 0.935 0.980 6.5 0.905 0.944 0.830 B. mPG-SGA 0.962 0.5 0.924 1.000 0.989 2.5 0.918 1.000 0.985 6.5 0.945 0.938 0.881 C. mPG-SGA-Box 1 0.933 0.5 0.866 1.000 0.921 1.5 0.809 0.939 0.887 4.5 0.787 0.819 0.739 D. mPG-SGA-Box 1 + Worksheet 4 0.975 0.5 0.951 1.000 0.935 1.5 0.862 0.899