Multicenter Study of the Validity and Reliability of Subjective Global Assessment in the Hemodialysis Population

Objective: Subjective Global Assessment (SGA) is a nutrition assessment tool recommended by the 2000 NKF K/DOQI Nutrition Guidelines. However, the validity and reliability of this tool have not been established in hemodialysis (HD) patients. The purpose of this observational study was to determine the reliability and validity of SGA in the HD population. Renal dietitians (RD) were recruited to perform SGA (7-point scale version) and collect data on demographics, clinical status, biochemistries, dietary intake, and quality of life (Medical Outcomes Short Form-36) on 3 HD patients at baseline and 6 months later. Design: The 54 participating RDs were trained to perform SGA and collect data via a website created for this study. Interrater reliability for SGA was tested in a subset of 76 patients, via an SGA performed by a second RD at baseline, while intrarater reliability was assessed by the original RD repeating the SGA at 1 month. Data collection occurred at HD facilities in the United States (109 patients), Canada (35 patients), and New Zealand (9 patients). Results: Of the 153 patients, 46% were female, 64% were Caucasian, 6% were Hispanic, 21% were African American, and 6% were Asian. The primary etiologies were hypertension (33%), type 2 diabetes mellitus (DM) (27%), type 1 DM (10%), and glomerular nephritis (10%); 59% had cardiovascular disease. The mean age, body mass index (BMI), serum albumin, and duration on HD were 64 (cid:1) 14 years (mean (cid:1) SD), 28 (cid:1) 7 kg/m 2 , 3.7 (cid:1) 0.4 mg/dL, and 41 (cid:1) 34 months, respectively. SGA scores were well nourished (7)–30%; mildly malnourished (MN 6)–41%; moderately MN 5–21%, 4–7%, and 3–2%; and severely MN (2 and 1)–0%. SGA training via the Internet achieved fair interrater reliability (weighted Kappa (cid:2) 0.5, Spearman’s Rho (cid:2) 0.7) and substantial intrarater reliability (weighted Kappa (cid:2) 0.7, Spearman’s Rho (cid:2) 0.8) ( P (cid:3) .001). Validity was demonstrated through statistically signiﬁcant differences in mean BMI and serum albumin across the 5 categories of SGA (7–28 (cid:1) 7, 6–29 (cid:1) 7, 5–28 (cid:1) 8, 4–21 (cid:1) 4, 3–24 (cid:1) 2, P (cid:3) .05; and 7–3.8 (cid:1) 0.3, 6–3.8 (cid:1) 0.4, 5–37 (cid:1) 0.05, 4–3.4 (cid:1) 0.07, 3–2.9 (cid:1) 1.2, P (cid:3) .001, respectively). Nutritional status varied by age ( P (cid:3) .05), but not ethnicity or nationality.

tional biochemical indices are altered by fluid and inflammation status (e.g., serum albumin), or need to be interpreted using different thresholds (e.g., serum prealbumin) for the CKD population.Anthropometric measures (e.g., bioelectric impedance, or skinfold measurements) for the CKD population must be carefully administered posttreatment to prevent confounding influences of fluid status.This is often challenging because of a lack of training by healthcare providers.Dietary interviews are often problematic, because they rely on patient memory of recent intake, and in this patient population may be negatively influenced by age-related and uremia-related influences on patients' memories.These collective issues compound the already challenging task of determining CKD patients' nutritional status.For example, clinicians must determine whether malnutrition exists when serum albumin is low although 24-h dietary recall and body mass index (BMI) indicate sufficient nutrient intake, or whether unintentional weight loss in a patient with normal serum prealbumin and albumin concentrations indicates insufficient nutritional intake vs. disease-related protein catabolism.
Subjective global assessment (SGA) is a nutritional tool commonly used by both clinicians and researchers. 1,23][4][5] Research studies compared and contrasted different nutrition status markers with SGA in CKD patients.Early studies analyzed SGA in two groups: those with or without some degree of nutritional loss.Statistically significant differences were observed between the two groups using established nutrition indices such as BMI, mid-arm muscle circumference (MAMC), lean body mass, serum albumin, creatinine, total iron-binding capacity, ferritin, and C-reactive protein. 2,6 -9More recent studies examining reliability and specificity found that SGA does not always correspond with other nutritional markers. 2,3Therefore, questions regarding the ability of SGA to detect nutritional status and to distinguish differing levels of nutritional loss have been raised.The purpose of the present National Research Question was to examine the intrarater and interrater reliability, and concurrent and predictive validity, of SGA in stage 5 CKD patients receiving hemodialysis by utilizing clinical renal dietitians for patient recruitment and data collection.

Subjects
Renal dietitians were recruited for participation in this study through the National Kidney Foundation's Council on Renal Nutrition and the Canadian Association of Nephrology Dietitians.Two batches of letters were sent to dietitians, explaining the study and asking them to participate.Participation required that they complete an online training course, including a pretest and posttest.They were also asked to provide contact information for their local institutional review board (IRB

SUBJECTIVE GLOBAL ASSESSMENT IN HEMODIALYSIS
lowed a process to randomly recruit three patients from their center.Patients indicated their willingness to participate by signing a written consent form.Dialysis centers with more than one renal dietitian were encouraged to have a second dietitian participate for assessment of interrater reliability.

Data Collection
At baseline and 6 months from baseline, participating dietitians used standard forms to collect the data outlined in Table 1.The SGA examination was repeated at baseline by a second dietitian (if available) to test interrater reliability, and was performed again after 1 month by the initial dietitian, to test intrarater reliability.The frequency and duration of hospitalizations within the 6-month collection period were documented.All data were sent to Case Western Reserve University for compilation and analysis.
Dietary intake data were collected the multiple-pass method. 11This method reviews the food intake of patients over a 24-h period five times to elucidate information on food items' brand names, serving sizes, and preparation methods.Dietary data were analyzed using Nutritionist Pro, version 2.5.1 (Axxya Systems, Stafford, TX).
The SGA was performed using the same core components as those of Detsky et al. 4 (weight status, dietary intake, gastrointestinal symptoms, functional status, metabolic stress, and a physical examination) and the 7-point scale developed during the Canada-USA Peritoneal Dialysis (CANUSA) Study. 5A rating of 7 indicates no nutritional loss, and a rating of 1 indicates severe nutritional loss.Information on weight and metabolic status for SGA was collected from the patient's medical record.Information on the remaining components of SGA was obtained from the patient interview.Dietitians were asked to evaluate all areas on the physical-examination portion, as described on the study website.Current medical records and laboratory results were used to collect other specific clinical and biochemical data.

Data Analysis
Statistical analysis was performed using SPSS, version 14.0 (SPSS, Inc., Chicago, IL).Distribution of variables was examined, and analyses were performed, according to the individual variables examined.Descriptive analyses were used to describe the sample of patients studied and the frequency of distribution of variables.Correlation analyses were used to test bivariate associations.The SGA was treated as a categorical variable; therefore, both one-way analysis of variance (ANOVA) and Kruskal-Wallis analyses were used to compare continuous variables across the SGA categories for criterion (concurrent and predictive) validity.When the variable was significantly different between SGA categories, a Tukey post hoc analysis was used to determine which SGA categories were significant compared with the others.Weighted and unweighted kappa scores were used to measure intrarater and interrater reliability.Statistical significance was defined as P Ͻ .05.

Results
Baseline characteristics of patients (n ϭ 154) were compared to the United States Renal Data System, 13 and by country (Tables 2 and 3).Seventy percent of patients had some degree of malnutrition according to SGA, ranging from very mild to moderate (Fig. 1).Among patients with malnutrition, 29% had moderate nutritional loss (scores of 3-5 on the 7-point scale), and no patient was categorized as severely malnourished (a score of 1 or 2).After 1 and 6 months, the number of patients with malnutrition increased to 75% and 76%, respectively.
Intrarater reliability for baseline and 1-month assessments performed by the same dietitian was measured in 111 patients.The agreement was 68% (unweighted kappa ϭ 0.56; 95% CI, 0.44-0.68;weighted kappa ϭ 0.68; 95% CI, 0.58-0.78),indicative of substantial intrarater reliability.Figure 2 depicts rater discordance, i.e., the difference between scores for the same patient.When ratings differed, the vast majority differed by a single point on the 7-point scale, and never  differed by Ͼ2 points.This demonstrates parallelism, even when scores were not identical.
To establish concurrent validity, baseline BMI and serum albumin were examined across baseline SGA scores of 1-7.Body mass index differed between SGA scores (P Ͻ .004,Kruskal-Wallis nonparametric test for skewed distribution).Specifically, both SGA scores 3 and 4 were different from both scores 6 and 7 (P Ͻ .05,Tukey's honestly significant difference [HSD]).Serum albumin was significantly different between separate SGA scores (P Ͻ .001,one-way ANOVA).Specifically, a score of 3 was significantly different from scores 5, 6, and 7 (P Ͻ .05,Tukey's HSD) (Table 4).To tease out the different influences of malnutrition and inflammation, an analysis was performed, examining the incidence of patients with low (Ͻ3.8 mg/dL) and normal (Ն3.8 mg/dL) serum albumin within each SGA score.It could be hypothesized that those patients with a low SGA score, but normal albumin concentrations, have pure malnutrition.Likewise, patients with low serum albumin and a normal SGA score could have inflammation alone, with no loss of nutritional status.Table 4 depicts these scenarios.
To establish predictive validity, the frequency and duration of hospitalization during the 6-month study period were examined across baseline SGA scores of 1-7.Frequency of admission was significantly different between SGA scores (P Ͻ .05,Kruskal-Wallis test for skewed distribution), and duration of hospitalization approached statistical difference (P ϭ .089)(Fig. 3).When patients were analyzed by serum albumin concentration (Ͻ3.8 or Ն3.8 mg/dL), patients with a concentration Ն3.8 mg/dL were not significantly different according to either number of admissions or days of hospital stay between SGA scores.However, patients with serum albumin Ͻ3.8 mg/dL were significantly different in terms of number of times hospitalized (P Ͻ .05),but not in terms of number of days hospitalized by SGA score.

Discussion
The purpose of this National Research Question was to examine the reliability and validity of SGA in hemodialysis patients.The methods used to achieve these aims were novel in that all the renal dietitians who participated were trained by way of an interactive website, and the training was monitored through pretests and posttests.The dietitians were active practitioners from Canada, New Zealand, and 30 different states throughout the United States (US).These dietitians had markedly different levels of educational backgrounds, years of work experience, and familiarity and training in terms of SGA.The patients randomly recruited for this study had characteristics similar to those reported for the US dialysis population in general. 13Therefore, extrapolation of this study's results to the US dialysis patient population is reasonable.all measured interrater reliability in stage 5 CKD patients, using two clinicians to conduct all SGA assessments.These studies reported significant reliability using intercorrelation coefficients (ICCs), percent agreement, or kappa scores.Our study found the same weighted kappa scores as did Cooper et al., 3 but lower percent agreement than in Kalantar-Zadeh et al.  8 That study had a much higher ICC for intrarater reliability than this study, possibly because of the shorter time (2 weeks) between assessments and the limited number of assessors (n ϭ 2).During a 2-week period, there would be a decreased chance for actual clinical changes to occur.However, both studies found statistically significant results.
Many different studies examined the concurrent validity of SGA. 2,3,8,9,12,14 -17A landmark study of SGA was the international CANUSA Study. 5The prospective, cohort CANUSA Study followed a large number of peritoneal dialysis patients from 14 dialysis centers across Canada and the United States.It found SGA to be independently predictive of death (P Ͻ .05) and days hospitalized (P Ͻ .05).
Jones et al. 2 studied SGA in a sample of 72 hemodialysis patients.They used both the original 3-point (A, B, and C) and 7-point scale methods of rating SGA, and compared SGA to biochemical and anthropometric nutritional markers as well as the composite nutritional score (CNS).They demonstrated concurrent validity through significantly different mean values for mid-arm circumference, MAMC, serum creatinine, C-reactive protein, and dietary sodium intake between SGA categories A and B (all P Ͻ .05).Using the 7-point scale, significant differences were found between SGA categories for the same variables, except for dietary sodium intake.Discrepancies were seen between CNS and SGA, causing concern about the reliability of SGA compared with CNS.
In this study, we teased out the influence of inflammation by examining serum albumin (Ͻ3.8 mg/dL and Ն3.8 mg/dL) to determine whether patients had malnutrition alone, malnutrition and inflammation, the presence of inflammation alone, or neither inflammation nor malnutrition.Our data indicate that malnutrition alone (SGA score of 3-5, with albumin Ն3.8 mg/dL) existed in 11% of our sample, and 18% had a combination of malnutrition and inflammation.Interestingly, when those patients with a low SGA but normal serum albumin were analyzed for differences in rate of hospitalization, no statistical significance was found.Significance was found only in the number of times hospitalized between baseline and 6 months in those patients with both low serum albumin and a low SGA score.
A limitation of our study was the 1-month period between the two assessments for intrarater reliability, which may have influenced agreement between the two assessments, because actual changes in nutritional status could have occurred, in which case it would have been accurate for the assessments not to agree.
In conclusion, this study found SGA to have fair interrater reliability, substantial intrarater reliability, and both concurrent and predictive validity when performed in a diverse hemodialysis population by a large and varied group of dietitians who were prepared for study participation using a web-based training program.

Figure 2 .
Figure 2. Rates of disagreement between raters according to degree of disagreement.Zero represents complete agreement; 1 represents a 1-point difference between raters on the 7-point scale; 2 represents a 2-point difference.There were no occasions when raters disagreed by Ͼ2 points on the 7-point scale.SGA, subjective global assessment.
Visser et al. (n ϭ 22; 13 hemodialysis and 9 peritoneal dialysis patients), 8 Kalantar-Zadeh et al. (n ϭ 12 hemodialysis patients), 14 and Cooper et al. (n ϭ 76; 52 hemodialysis and 24 peritoneal dialysis patients) 14  and Visser et al.8 Differences in study design included use of the Detsky 3-point (A, B, and C) version of SGA vs. the 7-point scale, using two assessors to conduct SGA vs. 54 different dietitians, and in-person vs. web-based training.Only one other published study of renal patients measured intrarater reliability.

Table 2 .
Patient Demographics of Total Sample SGAVP, Subjective Global Assessment Validation Project; USRDS, United States Renal Data System.*Data are presented as mean (Ϯ SD where available).

Table 3 .
Patient Demographics by Country *Data are presented as mean Ϯ SD.