Symposium

Progress ¡nthe Development of a Nutritional Risk Index123 FREDRIC D. WOLINSKY,* RODNEY M. COE,t WM. ALEX MclNTOSHJ KAREN S. KUBENAS JOHN M. PRENDERO AST,\\ M. NOEL CHAVEZ,^ DOUGLAS K. MILLER** JAMES C. ROMEIS,tf AND WENDALL A. LANDMANN,§ *Department of Medicine, Indiana University (Indianapolis); ^Departments of Community Medicine and "Internal Medicine, and ttCenter for Health Seruices Education and Research, St. Louis Uniuersity School of Medicine; ^Departments of Sociology and ^Animal Science/Human Nutrition, Texas A & M Uniuersity; \\Program on Aging, Mercy Hospital (Pittsburgh); and ^Department of Community Health Sciences, Uniuersity of Illinois School of Public Health.

The NRI was initially developed, in part, from the National Health and Nutrition Examination Survey (NHANES I) (6). Building on and modifying the items contained in the NHANES I protocol, 16 questions were ultimately selected that tap five important di mensions of nutritional risk, including the mechanics of food intake, prescribed dietary restrictions, morbid conditions affecting food intake, discomfort associated with the outcomes of food intake, and significant changes in dietary habits. The questions themselves are shown in Table 1 along with their frequency dis tributions obtained from the three-wave panel study described below. From the outset, it was decided that

ABSTRACT The development of a 16-item nutritional risk index (NRI) is chronicled from its inception through its application in three studies designed to assess its reliability and validity. Study I involved a survey of 401 community-dwelling elderly in St. Louis, Missouri who were interviewed at baseline, 4-5 mo later, and 1 yr later. Study II involved a cross-sectional survey of 377 male outpatients attending two clinics at the St. Louis Veterans Administration Medical Center. Study HIin volved a cross-sectional survey of 424 communitydwelling elderly in Houston, Texas. Internal consistency reliability coefficients ranged between 0.47 and 0.60, and test-retest reliability coefficients ranged between 0.65 and 0.71. Validity was established by using the NRI to predict the use of health services, as well as by correlating it with a variety of anthropométrie,labo ratory, and clinical markers of nutritional status. The utility of the NRI for future applications is discussed. J. Nutr. 120:1549-1553, 1990.

' Presented as part of a conference, "Nutrition Monitoring and Nutrition Status Assessment", at the first fall meeting of the Amer ican Institute of Nutrition, Charleston, South Carolina, December 8-10, 1989. The conference was supported in part by cooperative agreement HPU880004-02-1 with the DHHS Office of Disease Pre vention and Health Promotion, the USDA Human Nutrition In formation Service, the DHHS National Center for Health Statistics, and the International Life Sciences Institute-Nutrition Foundation. 1 The Planning Committee for the meeting consisted of Drs. He

INDEXING KEY WORDS:

•nutritional studies

risk •screening

measures

field

len A. Guthrie, Roy J. Martin, Linda D. Meyers, James A. Olson, Catherine E. Woteki, and Richard G. Allison (ex officio). The sym posium papers were edited by a committee consisting of Dr. James Allen Olson (coordinator), Dept. of Biochemistry & Biophysics, Iowa State University, Ames, IA¡Dr. Cathy C. Campbell, Division of Nutritional Sciences, Cornell University, Ithaca, NY; Dr. Roy J. Martin, Dept. of Foods &. Nutrition, University of Georgia, Athens, GA; and Dr. Catherine E. Woteki, Food & Nutrition Board, National Academy of Sciences, Washington, DC. 3 Supported in part by grants to Dr. Coe from the Alexian Brothers Hospital in St. Louis and the National Institute on Aging (K07-AG00302), to Dr. Wolinsky by the National Institute on Aging (K.04AG-00328 and RO1-AG-06618), to Dr. Prendergast by the Veterans Administration (HSR&.D 84-017), and to Dr. Mclntosh by the Na tional Institute on Aging (R01-AG-04043|. Each of the studies was approved by the appropriate Institutional Review Boards. Direct all correspondence to Dr. Wolinsky, Department of Medicine, Indiana University, 1001 West Tenth Street, Indianapolis, Indiana 462022859(317)630-7269.

The purpose of this article is to provide an overview of the work (1-5) that has been done over the past 6 yr on the development and validation of a nutritional risk index (NRI). Designed specifically for use with older persons, the goal of the NRI is to screen for those at risk of developing nutritionally related disabilities who could benefit from interventions to maintain their health status while reducing their health services uti lization. Those interventions might involve nutritional counseling, supplementation, and other modifications of the treatment regimen. As such, this work has been and continues to be motivated by tandem concerns for health care quality and cost containment. 0022-3166/90

S3.00 ©1990 American Institute of Nutrition.

Received 10 December 1989. Accepted 11 July 1990.

1549 Downloaded from https://academic.oup.com/jn/article-abstract/120/suppl_11/1549/4738656 by guest on 15 March 2018

1550

WOLINSKY

the NRI must be relatively brief, suitable for use in telephone surveys, easily scored, and readily inter pretable. That is, the NRI was to be capable of being reliably used in a variety of settings by relatively un trained and unskilled personnel. As a result, the ques tions shown in Table 1 are both straightforward and simple. The remainder of this article is arbitrarily divided into four parts. First is an overview of the psycho metric properties of the NRI as established in a threewave panel study of community-living elderly in St. Louis. Second is an overview of the clinical validation of the NRI obtained from a study of older outpatients at the St. Louis Veterans Administration Medical Center (VAMC). Third is a review of the preliminary findings emerging from a subsequent validation study of community-living elderly in Houston. The article concludes with a discussion of what the NRI actually represents, and the work that remains to be done.

STUDY

I

Community-living elderly in St. Louis. This study involved a two-stage random sampling design

ET AL.

resulting in 401 respondents 65 yr of age or older who resided in the 18 Census tracts within a two-mile ra dius of the sponsoring agency in south-central met ropolitan St. Louis (1-4). At baseline, face-to-face in terviews were conducted in the respondents' homes, with reinterviews by telephone at ~3-4-mo and 12mo intervals. The response rate at baseline was 59%. The resulting sample is representative of the elderly in these Census tracts, although it is slightly healthier than the national population of the elderly. Table 1 contains the frequency distributions on the 16 items comprising the NRI at each wave of the study. As shown, these questions tap risks that range from the relatively common (i.e., items 5, 14, and 11) to the relatively rare (i.e., items 7, 8, and 10). For the most part, the frequency distributions are rather stable across the three waves. The exceptions (excluding those relatively rare items 8 and 10, whose means are too unstable for such comparisons) involve items 15, 9,1,6, and 2, for which the occurrence rates all decline. Initially, this was thought to reflect either of two things. First, panel attrition may have resulted in a significantly healthier group of survivors by the third wave (7). Comparative analyses of survivors and de cedents, however, failed to support that hypothesis (8). Second, the differential stability in the frequency

TABLE 1 Frequency distributions

and exact wording of the NRI responses at three time points Percent responding "yes" (indicating risk)

Item itemDo number51411159431131612627810Actual dentures?In you wear adoctor?Have the past month, have you taken any medicines prescribed by abdomen?In you ever had an operation on your notprescribed the past month, have you taken any medicines that were doctor?Do by your youconstipated you have any troubles with your bowels that make diarrhea?Are or gives you any theydisagree there any kinds of foods that you don't eat because you?Do with food?Do you have trouble biting or chewing any kind of youreating?Do you now have an illness or condition that interferes with

now?Are you smoke cigarettes regularly diet?Have you on any kind of a special (hadiron you ever been told by a doctor that you were "anemic" blood)?Have poor inyour you had any spells of pain or discomfort for 3 d or more month?Do abdomen or stomach in the past appetite?Did you have an illness that has cut down on your month?Did you have any trouble swallowing at least 3 d in the last month?Have you have any vomiting at least 3 d in the last d?(Note: you gained or lost any weight in the last 30 pounds.)Sample net gain/loss must have exceeded 10 meansSample sizesT-l757359453635252423221914125554.77401T-272695731293921182017141092224.11334T-3767351322234221422171676

Downloaded from https://academic.oup.com/jn/article-abstract/120/suppl_11/1549/4738656 by guest on 15 March 2018

NUTRITIONAL

distributions may reflect two dimensions of the NRI, one being a stable set of items (i.e., traits), and the other being a variable set of items (i.e., states). Sub sequent comparative analyses of the trait vs. state subscales of the NRI, however, did not suggest the disaggregation of the index (8). The reliability of the NRI was primarily assessed using Cronbach's coefficient alpha, a measure of in ternal consistency (9). At baseline, the reliability coef ficient obtained was 0.603. Although not as robust as one would like, it was more than adequate for the ini tial stages of development of a new scale. Examination of the item-total correlations did not suggest dropping any items from the scale. Accordingly, all 16 items were retained for further analyses and administration in the subsequent waves of data collection. Coefficient alphas of 0.544 and 0.515 were obtained, respectively, at waves two and three. Because of the longitudinal design of this study, it was also possible to assess the test-retest reliability of the 16-item NRI. This was done by correlating the NRI scores across the three waves. The resulting coef ficients, which range between 0.65 and 0.71, indicate considerable test-retest reliability. Moreover, these coefficients were invariant regardless of whether pairwise or listwise deletion algorithms were used to con trol for compositional change in the sample over time. Thus, panel attrition did not affect the reliability of the NRI. Exploratory factor analyses were used to assess the pattern structure of the NRI. The results confirmed the theoretically posited five-factor structure consist ing of the mechanics of food intake (items 3 and 5), prescribed dietary restrictions (items 4 and 16), morbid conditions affecting food intake (items 1, 2, and 1115), discomfort with the outcomes of food intake (items 6-9), and significant changes in dietary intake (item 10). These five factors accounted for 47.9% of the variance in the 16 items. Further factor analyses of various age and sex groupings revealed generally consistent pattern structures, providing further sup port for the hypothesized dimensionality of the NRI (8). Concurrent predictive validity was assessed by add ing the NRI into standard regression models (10) used to forecast the elderly's use of health services. The results indicated that after allowing 18 other measures of the predisposing, enabling, and need characteristics of the elderly individuals to predict their physician, emergency room, and hospital utilization, the en hancement obtained by adding the NRI was substan tial. By itself, the net increment of the NRI to the explained variance was 14% for physician utilization (or ~61% of the variance that could be explained), 4% for emergency room utilization (or ~33% of what could be explained), and 5% for hospital utilization (or ~38% of what could be explained). Thus, the NRI was predictive of the use of health services, and that Downloaded from https://academic.oup.com/jn/article-abstract/120/suppl_11/1549/4738656 by guest on 15 March 2018

1551

RISK INDEX

predictive ability was not jointly shared with the stan dard measures of health status or other background characteristics. Finally, to provide further evidence of the validity of the NRI, the respondents were divided into two groups based on their risk levels. High risk was defined as a score of eight or more on the NRI (12% of the sample were in this category), with low risk involving a score of seven or less. This determination was ini tially made on theoretical grounds, inasmuch as a score of eight or more represented an "at risk" response on at least half of the dimensions that the NRI taps. Sub sequent cut-point or threshold analyses indicated that this division point was also the most powerful for dis criminating between the two groups with regard to health status and health services utilization (8). Indeed, the high risk group was significantly more likely to be rated by the interviewers as emaciated, had approxi mately twice as many physician visits and hospital ep isodes, and had nearly three times as many visits to the emergency room than the low risk group. The high risk group also had significantly poorer scores on stan dard measures of perceived health, functional health status, and morale (4).

STUDY II

Older outpatients

at the St. Louis VAMC.

Although the results described above were sufficient to establish the basic psychometric properties of the NRI (i.e., reliability, factor pattern structure, and pre dictive and known groups validity), they did not dem onstrate its clinical validity. That is, the unresolved question was whether the NRI was in fact tapping a specifically nutritional aspect of health status. It was felt that if clinical validation could be obtained, then the NRI could serve as an initial screen, or perhaps even as a substitute, for the more difficult and expen sive traditional approaches to nutritional assessment like anthropometry, laboratory assay, and clinical as sessment. Accordingly, a second study was initiated to provide that clinical validation (5). Data were collected on a sample of 377 male out patients aged 55 or greater who were randomly se lected from consecutively presenting individuals at the general medicine or geriatric clinics of the St. Louis VAMC. The response rate was 64%. Included in the protocol were a face-to-face interview conducted at the clinic, anthropométrieprocedures, laboratory as say of blood, a 3-d food intake record, and a medical chart review. Although the reliability of the NRI was slightly attenuated in this sample (coefficient alpha = 0.472), the same factor pattern structure emerged from the analysis. Comparable results on the other psychometric properties of the NRI were also ob served (5).

1552

WOLINSKY

Two strategies were employed to assess the clinical validity of the NRI. First, bivariate correlations be tween the NRI and a variety of clinical markers were calculated. Statistically significant relationships (at the 0.05 level or beyond, and in the anticipated direction) were found between the NRI and the body mass index (BMI) and total energy intake (KCal). In addition, re lationships approaching statistical significance (be tween the 0.10 and 0.05 levels) were found between the NRI and dietary intake and hemoglobin ratings. Moreover, although the correlations between the NRI and hematocrit, serum albumin, and serum total pro tein levels were not statistically significant, they were of the expected sign. Because the failure of several of the bivariate cor relations to reach statistical significance could have been due to the nonlinear nature of the relationship between the NRI and these clinical markers, a second strategy was employed. Cut-point or threshold anal yses were conducted on three composite measures of nutritional status. The first was the BMI, rescored as "at risk" if the value was 30 (obese). The second composite measure was of labo ratory risk, and was the sum of abnormally low results by VAMC standards for hemoglobin (