点击显示 收起
1 From the Vanderbilt Center for Human Nutrition, Vanderbilt University Medical Center, Nashville, TN, and the Nutrition Department, The Pennsylvania State University, University Park.
2 Supported in part by the Robert Wood Johnson Foundation and the Geisinger Foundation.
3 Reprints not available. Address correspondence to GL Jensen, The Vanderbilt Center for Human Nutrition, 514 Medical Arts Building, Vanderbilt University Medical Center, Nashville, TN 37212. E-mail: gordon.jensen{at}mcmail.vanderbilt.edu.
See corresponding editorial on page155.
ABSTRACT
Background: The potential for the use of nutritional screening to identify older persons at risk of hospitalization has not been contrasted with the use of tools developed for predicting hospital admissions.
Objective: Our goal was to compare the associations of items from the Level II Nutrition Screen (LII) and the Probability of Repeated Admissions (Pra) questionnaire with the outcome of hospitalization.
Design: This was a cohort study of participants in a Medicare managed-risk health plan who completed both the LII and Pra (n = 386). All hospitalizations within 1 y of screening were recorded. Hierarchical multivariate logistic regression was used to model associations with hospitalization.
Results: Pra items that retained significant associations with hospitalization were self-reported health, hospitalization in the past year, and >6 doctor visits in the past year (positive predictive value: 20%; sensitivity: 53.1; specificity: 69.7). LII items that retained significant associations with hospitalization were eating problems and polypharmacy (positive predictive value: 17.9%; sensitivity: 58.0; specificity: 56.3). Those persons designated by the Pra score as being at high risk of hospitalization (Pra 0.30, 75th percentile) were also more likely to report weight loss, polypharmacy, consumption of a special diet, and functional limitation on the LII.
Conclusions: Retained items from the Pra and the LII were comparable in identifying participants at risk of hospitalization. These observations suggest that nutritional risk factors such as eating problems, weight loss, and consumption of special diets should be considered in the management of older persons at risk of hospitalization, irrespective of the screening approach selected.
INTRODUCTION
The Nutrition Screening Initiative was established in 1990 as a multifaceted effort to promote better nutritional care of older Americans (1–3). It is a collaborative effort of the American Academy of Family Physicians, the American Dietetic Association, and the National Council on Aging. The Nutrition Screening Initiative promotes a 3-tiered approach to screening for nutritional risk: the Determine Checklist, a 10-item questionnaire designed to heighten the awareness of older adults and health care providers to the warning signs of poor nutritional health, and the Level I (LI) and Level II (LII) Nutrition Screens, more comprehensive assessments used to evaluate nutritional risk by professionals in health and social services settings or medical settings, respectively.
These screening tools have been widely disseminated and are being used in capacities that far exceed those for which they were originally intended. Limited validation testing of the use of these screens to predict specific health or nutrition status outcomes has been performed (4, 5). Jensen et al (6) reported that selected LI and LII screening items were associated with self-reported functional limitation and health care charges in a population of rural older Pennsylvanians participating in a Medicare managed-risk health plan. They concluded that further testing of the screening instruments is needed with use of objective resource measures, such as hospital admissions. A focus on hospital admission is warranted given that many older persons are malnourished at the time of hospital admission (7). Hospital admission is also a proxy indicator of poor health status.
The Probability of Repeated Admission (Pra) is a screening instrument specifically developed to identify older persons most likely to be hospitalized (8, 9). This instrument focuses on age, sex, self-rated health, prior hospital admissions, physician contacts, chronic disease processes, and caregiver status, variables that are strongly related to risk of hospitalization. The reliability and validity of this instrument was previously reported (10–12).
The ability of nutritional risk screening to identify persons at risk of hospital admission has not been rigorously shown. The objective of this investigation was 2-fold: 1) to compare the associations of items from 2 screening instruments, the LII and Pra, with hospitalization outcomes in screened older persons, and 2) to then compare the nutritional risk status of those designated by the Pra as being at high risk of hospital admission with those who were not.
SUBJECTS AND METHODS
Subjects
Subjects for this study were patients of the Penn State Geisinger Health System (PSGHS), a nonprofit group practice serving a predominantly rural population in central and northeastern Pennsylvania. This service area has one of the largest concentrations of rural older persons in the United States (13). This project was approved by the Office for Regulatory Compliance at the Pennsylvania State University and by the PSGHS Institutional Research Review Board.
Data collection
Nutritional screening data were collected at >100 clinic sites as part of a regional program initiated by the PSGHS in September 1994 that targets all participants in a Medicare managed-risk health plan. A modified LII questionnaire was included in the health plan enrollment packet and was completed by health plan members at the time of a required enrollment history and physical examination. The questionnaire has been extensively tested, as described previously (6). The overall response rate has averaged 52%, with responders well representing nonresponders (6). A validated scoring algorithm is not available. The questionnaire includes check boxes for 34 items and also requests self-reported height and weight. It requires 5 min to complete and includes queries about recent weight change (volitional or not), living and eating habits, alcohol and medication use, depression, dentition, and limitations in functional status. Arm circumferences and triceps skinfold thicknesses were deleted from the LII because it was not possible to obtain reliable measures at multiple clinic sites. Formal testing of cognition and depression were also deleted because of the limitations in training multiple clinic staff and the time constraints in administering the screening tool at busy clinics. The single question, "Do you feel depressed?" was substituted. Members' responses were reviewed with the clinic nurse, who also recorded measured height and weight. Height and weight were measured by using metric wall rules and counterweight balance-beam scales. Subjects were clothed but did not wear overgarments or shoes for these measurements. The subjects were assisted as needed in completing the screening questionnaire, and proxies were used at the discretion of clinic personnel to provide all or part of the information.
Serum albumin and cholesterol data were obtained from routine surveillance measurements taken at the discretion of the primary physicians within 12 mo before or 2 wk after completion of the screening questionnaire. Serum albumin and cholesterol results were available for 114 and 144 individuals, respectively; 70% of the measurements were made within 90 d of the screening. All laboratory analyses were completed at the accredited laboratory facilities of the PSGHS by using a Hitachi 717 analyzer (Hitachi, Indianapolis) and Boehringer Mannheim (Indianapolis) reagents and manufacturer-provided guidelines. The Boehringer Mannheim reagent system uses an enzymatic cholesterol method (cholesterol/HP; catalog no. 1039033) and a bromcresol green dye-binding albumin method (albumin BCG; catalog no. 1039025).
The Pra health survey was conducted semiannually by the PSGHS. Computer-scannable surveys and preaddressed, postage-paid return envelopes were mailed to all current enrollees of the Medicare managed-risk health plan who did not have a completed Pra form on file. Nonresponders were targeted with 2 subsequent mailings. The overall response rate has approached 70%. The Pra screen, originally developed by Boult et al (8–12), consists of 8 questions. The 8 factors used to identify those at high risk of hospitalization are older age, male sex, self-rated health, availability of an informal caregiver, having ever had coronary artery disease, and having had a hospital admission, >6 doctor visits, or diabetes in the past year. A Pra risk score between 0 and 1 is assigned by using a regression equation, with higher values indicating higher risk. A cutoff point at the 75th percentile (0.30) was considered to represent high risk in the present analysis.
Six hundred twenty-four individuals aged >65 y who were new enrollees in the PSGHS Medicare managed-risk health plan completed the LII between 1 December 1996 and 28 February 1997. A total of 15337 individuals responded to a mailing of the Pra health survey in February 1997. The sample population consisted of the 386 individuals who completed both the LII and Pra surveys during this time frame. The sample was 99% non-Hispanic white. Hospital admissions data for this sample were available through 31 January 1998. Hospitalization outside the PSGHS is rare and is captured by the admissions database used in this study. A cutoff of 1 y from the time of screening to hospital admission was used for analysis.
Statistical analysis
Statistical analyses were conducted by using SAS for PC (version 8; SAS Institute Inc, Cary, NC). Data analysis consisted of 4 phases. First, descriptive statistics were calculated for all of the individual items in the LII and Pra screens. Continuous variables were summarized by using means and SDs and categorical variables were summarized as frequencies and percentages. Second, potential differences in all of the individual data items between hospitalized and nonhospitalized subjects and between subjects with high- and low-risk Pra scores were examined by using t tests for continuous variables and analysis of variance for categorical variables, with P < 0.05 considered significant. Third, summary models were developed to predict hospitalization from 1) the Pra survey items, 2) the LII items, and 3) a hybrid model of retained items from both the Pra and LII. In each case, a hierarchical modeling scheme was used with retained items having adjusted P values < 0.10. Fourth, receiver operating characteristics curves were used to select the best cutoff for maximizing sensitivity and specificity for the complete Pra score, the retained items from the Pra, the retained items from the LII, and the hybrid model of retained items from both the Pra and the LII. Positive predictive values were calculated as (true positives)/(all predicted to be hospitalized).
RESULTS
Of the 386 subjects, 13.0% (n = 50) were hospitalized at least once in the year after screening. The total sample was nearly evenly composed of men and women (47% and 53%, respectively) and about one-third (32%) were aged 75 y (data not shown). Eighty-three percent of the sample rated their health as good or better (data not shown). Fifteen percent reported being hospitalized within the previous year and 14% reported seeing a physician >6 times in the previous year (data not shown). The frequency of positive responses to the 8 Pra items for those hospitalized compared with those not is shown in Table 1. Those who were hospitalized within the 12 mo after the screening were more likely to have reported hospitalization and physician contact in the previous year and to have reported having heart disease. Analysis of variance also indicated significant trends favoring older age and poorer reported health status among those who were hospitalized. At a high-risk Pra score of 0.30, 23.5% of subjects were hospitalized compared with 10.0% of subjects who did not meet this Pra cutoff (P < 0.007 by t test).
View this table:
TABLE 1.. Frequency of positive responses for scored Probability of Repeated Admission (Pra) items by admission outcome
The most commonly reported items on the LII were polypharmacy (ie, the use of >3 drugs, over-the-counter medications, or supplements daily; reported by 41% of subjects) and intakes of vegetables, milk, juices, and grains below the indicated frequencies (>50% of subjects for each). Those who were hospitalized in the 12 mo after screening were more likely to report polypharmacy and having a body mass index (BMI; in kg/m2) >27 (Table 2). Those designated by the Pra as being at high risk of hospitalization were more likely to report risk factors such as weight loss, polypharmacy, special diets, eating problems, and functional limitation on the LII than were those not at high risk of hospitalization (Table 3).
View this table:
TABLE 2.. Patient characteristics and frequency of positive responses for selected Level II (LII) Nutrition Screen items by admission outcome
View this table:
TABLE 3.. Frequency of positive responses for selected Level II (LII) Nutrition Screen items by Probability of Repeated Admission (Pra) risk level1
The complete Pra score had a positive predictive value for hospitalization of 21.3% (sensitivity: 52.0; specificity: 71.3). Those items that retained significant associations with hospitalization by multivariate logistic regression were self-reported health, hospitalization in the past year, and >6 doctor visits in the past year (Table 4). These Pra items gave a positive predictive value of 20% (sensitivity: 53.1; specificity: 69.7). The LII items that retained associations with hospitalization by multivariate logistic regression were eating problems and polypharmacy. These LII items had a positive predictive value of 17.9% (sensitivity: 58.0; specificity: 56.3). When those items that maintained significant associations with hospitalization from both the Pra and the LII were combined in a hybrid multivariate logistic regression model, there was little improvement in positive predictive value (21%; sensitivity: 59.2; specificity: 68.2) and the polypharmacy item was not retained.
View this table:
TABLE 4.. Multivariate logistic regression models predicting hospitalization within 1 y of screening1
DISCUSSION
The LII was developed by the Nutrition Screening Initiative to be used in clinical settings to identify older adults at risk of poor nutritional status (1–3). Despite widespread dissemination of the LII, data are lacking on its application in relation to specific health or nutritional status outcomes (4, 5). The Pra was specifically designed to predict hospitalization in older adults and its validity and reliability were previously shown (8–12).
The overall frequencies of the LII items reported in Table 2 are similar to those previously reported for a much larger sample drawn from the same population (6). In that report, the following items retained a significant association by multivariate regression with average monthly health care charges: age 75 y, male sex, albumin < 35.0 g/L, polypharmacy, loss of 4.5 kg (10 lb) over past 6 mo, BMI > 27, cholesterol < 4.14 mmol/L, and any limitation in activities of daily living or instrumental activities of daily living. Findings in the present study may differ because of the smaller sample size and a different outcome measure (hospitalization).
Although the Pra was developed to predict hospitalization risk within 24 mo (8–12), the Pra risk items in the present investigation were also more prevalent among those who were hospitalized within 12 mo. A Pra score of 0.30 corresponding to the 75th percentile for hospitalization risk is also consistent with previous reports (8–12).
Those items from the Pra and the LII that retained significant associations with hospitalization were comparable in the identification of at-risk participants. There was general agreement in the subjects identified by these items and there was no apparent added value of a hybrid model. Observed values for positive predictive value, sensitivity, and specificity suggest that the selected Pra and LII items would have acceptable utility for screening applications. An important new observation is that those subjects designated by the Pra as being at high risk of hospitalization were also more likely to report risk factors such as weight loss, polypharmacy, special diets, and functional limitation on the LII than were those not at high risk of hospitalization. These observations suggest that nutritional risk factors such as weight loss and special diets should be considered in the management of older persons at risk of hospitalization, irrespective of the screening approach selected. Of particular value may be the risk item "eating problems," comprising chewing or swallowing difficulties or pain in the mouth, teeth, or gums, which was retained in the hybrid model. Indeed, others reported the number of general oral problems to be a strong predictor of involuntary weight loss (14).
The Institute of Medicine (15) recently highlighted the limitations of nutritional risk screening at the time of hospital admission in a report titled The Role of Nutrition in Maintaining Health in the Nation's Elderly. Many older persons are already malnourished at the time of hospitalization (7). Because of this, an approach to identifying older persons at risk of hospitalization that includes nutritional risk items may be the best method for facilitating appropriate interventions before the need for hospitalization. Case management interventions for high-risk individuals might diminish the need for hospitalization and prevent deterioration in nutritional status. Consistent with this approach is the development of the Pra Plus by the HMO Workgroup on Care Management (16). This 4-page computer-scannable survey contains the original 8 Pra questions for identification of hospitalization risk and additional items to help guide intervention that include medical history, current medical conditions, living environment, functional impairment (activities of daily living and instrumental activities of daily living), medication use, involuntary weight loss (4.5 kg), and depression. Because weight loss is the only additional nutritional item, consideration of some of the nutritional risk factors identified above might also be warranted. Boult et al (17) also suggested that the Determine Checklist might be used as a secondary screen of nutritional risk for those older persons already identified as being at risk of adverse outcomes such as hospitalization.
It is not clear whether our results can be generalized to other populations of older adults beyond the larger sample of enrollees in the Medicare managed-risk health plan in the rural setting studied. Small cell sizes preclude meaningful analysis of the association of some items with the outcome of hospitalization. Larger samples and longer follow-up will be required to clarify some possible associations. Further assessment of these screening instruments in other populations of older adults will also be necessary to confirm these findings. Potential hybrid screening instruments will need to be tested a priori to evaluate predictive value.
It is difficult to discern whether the predictive value of the screening items for hospitalization risk lies solely in their utility as indicators of general health status or whether nutritional status indicators add value. Individuals in poor health will often have resulting nutritional compromise. Malnutrition can hasten decline in health status by compromising host defense and blunting response to medical therapies. It will be of interest to ultimately learn whether interventions, nutritional or otherwise, can alter hospitalization outcomes among candidates designated as high risk by screening.
ACKNOWLEDGMENTS
We gratefully acknowledge the staff of the Clinical Nutrition Research Center at the Geisinger Medical Center in Danville, PA, for their assistance in data collection and management.
REFERENCES