| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
| ||||||||||||||||||||||||||||||||
1 Geriatric Research Education and Clinical Center, Central Arkansas Veterans Healthcare System, Little Rock.
2 Donald W. Reynolds Department of Geriatrics, University of Arkansas for Medical Sciences, Little Rock.
3 Geriatric and Extended Care Service, Central Arkansas Veterans Healthcare System, Little Rock.
4 Department of Family and Community Medicine, University of Arkansas for Medical Sciences, Little Rock.
5 Division of Biometry, University of Arkansas for Medical Sciences, Little Rock.
Address correspondence to Dennis H. Sullivan, MD, Geriatric Research Education Clinical Center (182/LR), Central Arkansas Veterans Healthcare System, 4300 W. 7th Street, Little Rock, AR 72205. E-mail: sullivandennish{at}uams.edu
 |
Abstract |
|---|
 Top Abstract MethodsResultsDiscussionReferences |
|---|
Methods. For this study, 900 nursing home residents with a recently identified nutritional problem from 96 long-term care facilities in 8 states were randomly selected. At study entry, nutritional, health status, and demographic data were extracted from the nursing home chart or the Minimum Data Set. Weights obtained by the nursing home staff were recorded at baseline and during each of the subsequent 6 months. Cox proportional hazards regression analysis was used to assess relationships between weight parameters and mortality risk during the 7 months of observation.
Results. During the study, 435 (48%) participants had at least 1 monthly weight that differed from the previous month's weight by 
5%; 163 (18%) participants had this magnitude of a 1-month weight change more than once. By controlling for the linear trend (i.e., slope) of each residents' weight change between the first to the last weight, the average month-to-month residual variability in the resident's weights was calculated. This residual variability or "average random fluctuation" in weight from 1 month to the next was 2% in 229 (25%) participants and 3% in 82 (9%) participants. Despite the large random fluctuation in the residents' weights, weight loss was an important prognostic indicator. Those who lost 5% in any month had a 10-fold increased risk for death compared with those who gained weight (adjusted relative risk, 10.6; 95% confidence interval, 3.2 to 35.5). The average random fluctuation in weight was associated with an increased risk for death only at the upper 10th percentile for the population.
Conclusions. Many nutritionally compromised elderly nursing home residents experience significant bidirectional fluctuations in their weight from month to month. How much of this variability is due to measurement error is not known. However, a weight loss of 5% in any month has important prognostic implications.
Retrospective analyses of large government data sets indicate 10% to 15% of nursing home residents meet the MDS criteria for significant weight loss at least once every year (2,12). Consequently, the potential impact of the MDS guidelines on overall mortality risk is large. However, it is not known whether targeting residents for more in-depth assessments based on these criteria is a valid strategy. Although weight change in older persons has prognostic significance, it is not known whether monthly weight measurements obtained by the clinical staff in long-term care institutions are reliable enough to be useful in identifying at-risk residents in a prospective manner. The few studies that identified a link between weight loss and death within the nursing home setting were retrospective and generally focused on very select resident populations (3–5,11). Several studies indicate that nursing home residents are just as likely to gain significant amounts of weight as they are to lose weight (3,12). This suggests that at least some of the weight fluctuation may be due to errors in measurement techniques or alterations in fluid balance rather than a true change in lean or fat mass. If measurement error is the major contributor to weight fluctuations in many nursing homes, short-term weight change may not be a useful prognostic indicator. If fluid shifts are the more common cause of the fluctuations, the amount of month-to-month fluctuation in weight may indicate metabolic instability and have greater prognostic significance than simply the amount of weight lost during any measurement interval.
The purpose of this study was to determine whether there is an association between mortality risk and either (a) the 6-month average intraresident monthly variation in recorded weight measurements, or (b) the relative change in weight during any single time interval among a national sample of nutritionally at-risk elderly nursing home residents. A secondary objective was to determine what baseline resident characteristics were associated with greater 6-month average intraresident monthly variation in weight.
|
METHODS |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
Inclusion Criteria: Definition of Nutritional Risk
NCS Healthcare, a leading long-term care pharmacy provider, recruited 96 facilities for the study. The characteristics of these facilities have been described previously (13). During a 3-month period from August to November 1998, consultant pharmacists reviewed the nursing home medical records and most recent MDS of each resident within these facilities. Residents were considered eligible for inclusion in the registry if they were capable of oral feeding at the time of the initial review and they had a nutritional problem identified within the previous 3 months that placed them at nutritional risk. For the purpose of the study, nutritional risk was defined as weight loss, poor appetite, or both according to one of the following: (a) MDS documentation of poor appetite (the resident leaves 25% of food uneaten at most meals [MDS 2.0, Section K 4c; MDS+, Section L 4e]), (b) chart documentation of poor appetite prompting dietary consultation, (c) MDS documentation of a weight loss of 5% or more within a 1-month period (MDS Section K 3a), or (d) chart documentation of weight loss prompting dietary consultation.
To limit costs, GAIN registry entry was limited to a stratified random sample of residents. From the list of residents who met the entry criteria, a maximum of 20 were chosen at random for registry entry from each facility. Subsequent to enrollment, each resident's nursing home records and any updated MDS data were reviewed monthly for 6 months by the long-term care consultant pharmacist in charge of data collection. Survival data were collected for 7 months.
To be eligible for entry into the current study, residents had to remain in an intermediate or skilled nursing home bed for at least 2 months. Of the 1000 residents from 96 long-term care facilities initially enrolled in the registry, 100 failed to meet the study entry criteria. This included 15 who were residing in an assisted living bed and 85 who either died or were discharged before the third month of observation. This left a final study sample of 900 residents.
Baseline Variables
Variables collected at baseline included demographic data, diagnoses, medications, appetite assessment, overall activity of daily living (ADL) functional capacity, level of feeding dependence, and number of pressure sores. Appetite was rated using a scale of 0 to 3: poor = 0 (resident consumes 0%–49% of meals); fair = 1 (resident consumes 50% to 74% of meals); good = 2 (resident eats 75%–89% of meals); and excellent = 3 (resident eats 90%–100% of meals). Functional capacity was rated as 0 (completely independent), 1 (needs assistance with some ADLs), or 2 (needs assistance with all ADLs). Level of feeding dependence was rated as 0 (independent, feeds self), 1 (requires set-up only), 2 (requires assistance with feeding beyond set-up), or 3 (totally dependent on staff for feeding). The presence of pressure sores was based on chart and MDS documentation. Residents were not examined and the severity of pressure sores was not determined. When available, baseline variables were taken from categories reported on the MDS. Otherwise, they were obtained from the nursing home chart.
Weight and Mortality Data
During the first month of observation, all participants were weighed 2 or more times by the nursing home staff. The first 2 weights obtained were recorded. In the subsequent 6 months (i.e., the second through the seventh month of observation), the first weight obtained by the nursing home staff in that month was recorded. Any death, discharge, or loss to follow-up during the 7 months of observation was recorded.
Statistical Methods
All analyses were conducted using SAS statistical software (Version 8.0, SAS Institute, Cary, NC). A probability value of.05 or less was considered significant.
Intraresident weight variability.-- In the first set of analyses, several methods were used to evaluate intraresident weight variability from month to month. The average of the first 2 weights obtained for each resident was considered the resident's baseline weight. To determine how many times a resident's weight changed by >5%–10%, the percentage change in weight during each monthly interval was calculated. To determine the average monthly intraresident variability in weight, other analytic approaches were used. One method (method 1) involved taking the absolute value of the difference between each weight and the weight from the previous month, expressing the difference as a percentage of the baseline weight, and then calculating the average. However, this method did not consider the fact that part of the intraresident month-to-month variability in the weight measurements was a result of the trend (i.e., slope) in weight over time. For residents who gained or lost weight in a steady (linear) manner, their actual monthly weight would equal their predicted weight calculated from the slope. Any variability about the slope could be considered to represent random fluctuations in weight each month. This "random fluctuation" is equivalent to the residuals in least-squares regression. To obtain a better estimate of possible random fluctuations in the participants' weights with time, 2 additional methods were used to evaluate intraresident month-to-month weight variability.
For the first of the alternate methods (method 2), the slope (i.e., predicted change in weight per month) was set equal to the difference between the last weight and the first weight and then divided by the number of months of observation. From the slope, each participant's "predicted" weight for each month of observation was calculated. The participant's average month-to-month weight variability was set equal to the average of the absolute value of the difference between the actual weight and the predicted weight each month and expressed as a percentage of baseline weight. In recognition that this method would be strongly influenced by outlier weight measurements at the beginning or end of the observation period, intraresident weight variability was also calculated using least-squares regression (method 3). This was accomplished using SAS software and the "by" statement within the procedure regressing weight (expressed as a percentage of baseline) by month. In this manner, the regression line for weight by time and the residuals were calculated for each participant. The participant's average month-to-month weight variability was then set equal to the average of the absolute value of the residuals for each month.
We expected that the latter 2 methods of calculating average month-to-month weight variability would produce much larger values than the first method for participants who had a large weight change during 1 month but otherwise very little weight change in the other months. For this reason, we assumed that the method that produced the smallest estimate of average month-to-month weight deviation for a participant also gave the best estimate of the average amount of possible random fluctuation in that person's weight with time. Consequently, we set the variable "average random fluctuation" equal to the smallest of the three estimates of variability for the participant. We evaluated associations between the average random fluctuation and baseline participant characteristics using the Wilcoxon rank sum test.
Relationship between weight change and mortality risk.--
In the second set of analyses, we evaluated the relationship between weight change and mortality risk using Cox proportional hazards regression analysis. We tested the following hypotheses: (a) compared with the remaining residents, those with the greatest (i.e., upper 10th percentile) average random fluctuation in weight each month would have a greater mortality risk during the 7 months of observation; (b) compared with the remaining residents, those who lost 10% of their weight within 6 months would have a greater mortality risk during the 7 months of observation.
To test each hypothesis, we created a time-to-event and a status variable for each participant. The status variable indicated whether the event was death or the last follow-up. Because death was recorded at each monthly follow-up visit, the time to event was entered as the month of death or last follow-up (i.e., months 2 to 7). All participants who survived the 7-month review period were assigned an event time of 7. Four dummy weight loss variables were also created. If a resident's last weight represented a loss of 10%, the variable "lost 10%" was set to 1 and the other 3 dummy variables set to 0. If a participant's last weight represented a loss of 0 to <10%, the variable "lost <10%" was set to 1 and the other 3 dummy variables were set to 0. The other 2 dummy variables, "gain <10%" and "gain 10%" were set in a similar manner.
To test the first hypothesis, a dummy weight fluctuation variable was created and set to 1 if the resident's average random fluctuation in weight each month was in the upper 10th percentile (i.e., 3%). Otherwise it was set to 0. To determine whether 3% fluctuation in weight each month was associated with greater mortality risk after adjusting for the direction of weight change, all 5 dummy variables were entered into the analysis. One of these 5 variables served as the reference category as determined automatically by SAS software.
To test the second hypothesis, the 4 dummy weight loss variables were entered into the analysis. To control for potential confounders, a second analysis was run with 10 other correlates of death (i.e., indicators of health and nutritional status at study entry) entered into the analysis using a stepwise procedure after the 4 dummy weight loss variables were forced into the model. The 12 control variables used were age, appetite score, body mass index, length of stay, number of diagnoses, number of pressure sores, ADL functional score, feeding dependence score, sex, number of medications, and diagnosis of congestive heart failure or chronic obstructive pulmonary disease. These variables were chosen because they had previously been found to be associated with death in this cohort of nursing home residents (13).
Time-dependent covariate analyses.--
To test the hypothesis that a weight loss of 5% in any month was associated with an increased risk for death, Cox proportional hazards regression analyses were conducted using time-dependent variables. As before, a time-to-event and a status variable were created for each participant. Three time-dependent dummy variables were also created (lost 5%, no change, and gained 5%) and forced into the model. At each event time (i.e., each month), the variables were reset to 0 or 1 depending on how much the participant's weight had changed from the previous month. To control for potential confounders, a second analysis was conducted in the same manner, except the covariates were entered using a stepwise procedure.
Another set of similar analyses was conducted to determine whether a weight loss of 10% in any 3-month interval was associated with an increased risk for death. In these analyses, the time-to-event variable was entered as the month of death or last follow-up beginning with month 3. During each event time, the time-dependent variables were reset depending on how much the resident's weight had changed from 3 months before.
|
RESULTS |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
|
|
5%. One hundred sixty-three (18%) participants had this magnitude of a 1-month weight fluctuation more than once. The maximum 1-month loss of weight for each participant ranged from 0% to 38%. Sixteen (1.8%) residents had at least 1 weight that differed from the previous month's weight by 15%. A similar number (14 residents) had an apparent 1-month weight gain of 15% on at least 1 occasion.
|
2% per month in 229 (25%) residents and 3% per month in 82 (9%) residents. When only the first method of calculating monthly intraresident weight fluctuation was used (i.e., method 1, which did not account for weight trend), 523 (58%) residents had an average weight change of 2% per month, 249 (28%) residents had an average weight change of 3% per month, and 117 (13%) residents had an average weight change of 4% per month. Table 4 lists the baseline resident characteristics that were significantly (p <.01) associated with monthly average random fluctuation in weight. In all cases, the direction of the association was the same, and the more medically complex residents (e.g., recent admission, active infection, partially or completely dependent on staff for assistance with ADLs) had the greatest variability. However, the clinical relevance of these findings is questionable because the actual differences in the variability between the groups were rather small for all of the variables listed. Medical diagnoses (e.g., previous cardiovascular accident, congestive heart failure, hip fracture, chronic venous insufficiency) were not associated with average random fluctuation in weight.
|
As shown in Table 5, a weight loss of 10% within 6 months was associated with an increased risk for death during the 7 months of observation. After controlling for age and health status at baseline, the results remained highly significant (ARR, 8.3; 95% CI, 3.1 to 21.8).
|
5% of their weight in any month had a 10-fold higher adjusted risk for death compared with those who had a similar magnitude of weight gain. Similarly, a weight loss of 10% in any 3-month interval was associated with an 8-fold increased risk for death after controlling for age and other health status variables.
|
|
DISCUSSION |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
As defined in this study, the average random fluctuation in weight was associated with death only when it was relatively large (i.e., the upper 10th percentile for the cohort). This association may be confounded by illness severity. As we found in this study, those with the greatest month-to-month variability in weights tended to be the frailer, more functionally dependent residents. These residents are often the most difficult to weigh. They are probably also more likely to experience significant fluid shifts. However, we could not investigate this possibility in this study. Although having heart failure, venous insufficiency, dependent edema, or other similar diagnoses listed on the problem list was not associated with greater weight fluctuations, a more careful assessment of monthly medication adjustments and changes noted on physical examination would be needed to determine how much of the weight fluctuation was a result of fluid shifts. A listing of diagnoses is not usually a good indicator of disease activity or mortality risk (17,18). Given the weakness of the associations between average random fluctuation in weight and both baseline characteristics and outcomes, further study is needed to determine whether this type of weight variability is an important health status indicator.
Confirming the results of previous investigations (3–5,11), our study clearly shows that the relative amount of weight change during any single time interval among a national sample of nutritionally at-risk elderly nursing home residents is strongly associated with death. Although the difference in mortality risk between weight gain and weight stability was not statistically significant, we observed a trend toward lower mortality risk with weight gain. In contrast, weight loss was strongly associated with death. After we adjusted for baseline health status, those who lost 10% of their weight within 6 months were more than 8 times as likely to die during the observation period compared with those who gained this amount of weight. Despite the amount of what appeared to be random fluctuations in the residents' weights, we also found that a 1-month weight loss of 5% or more was a powerful predictor of death. The ARR of death for those with this amount of weight loss was more than 10 times that of those who gained weight during the month. A loss of >10% of baseline weight in 1 month was associated with a 20-fold increased risk for death (ARR of death is >20, data not shown).
We could not determine how aggressively the participants were being treated for their medical or nutritional problems, or whether the nursing homes were responding appropriately to their weight changes. Some residents may not have been considered candidates for aggressive nutritional or medical care. However, loss of weight is an important quality indicator in nursing homes. In a future study, it would be important to determine how frequently the MDS-associated resident assessment protocols were appropriately triggered with each episode of weight change. Particularly for those residents who lose weight, facilities should be conducting careful assessments to exclude reversible causes. They should also evaluate the quality of their feeding assistance programs, because this is often neglected in many nursing homes (19,20).
This study has several potential limitations. Among these was the short period of observation. We do not know whether the weight trends or the relationship between weight change and death remain significant over longer time intervals. Another limitation was the targeting criteria used. Because we included only residents identified to be at nutritional risk in this study, the results may not provide an accurate assessment of the amount of month-to-month weight variability within the total population of nursing home residents. Another study comparing weight variability in residents identified to be at nutritional risk and the remaining residents might be revealing. In the current study, the month-to-month variability in the weights was large even after the trend (or slope) of weight change in each participant was considered. Poor measurement technique, medical instability of the residents, or other factors may have contributed to this variability. Assuming that nutritionally stable residents would be more medically stable, a comparison study of nutritionally at-risk and nutritionally stable residents would provide further evidence of the quality of the weight measurements. A careful assessment of each resident who has a weight change of >3% to 5% in any month would also be important but may not lead to the identification of the source of the change in all cases (21).
This study relied solely on the data recorded by the clinical staff within each facility. It was beyond the scope of the study to verify the accuracy of these data by repeating any of the weight measurements. It is logical to assume that the best way to ensure the accuracy of weights obtained from nursing home residents is to have appropriately trained staff with adequate time and equipment to obtain these measurements. Among other things, the training would emphasize the importance of using a consistent approach to performing weight measurements, such as using the same scale, taking the weights at the same time each day, and ensuring that the resident is wearing the same amount of clothing. Having a team trained in this manner to recheck the nursing home weights would provide more insight into the possible causes of the weight variability.
Although this type of data verification was not possible in this study, the study results are still important. Clinicians in practice must rely on the weight measurements obtained by the nursing homes. This study indicates that these measurements have important prognostic implications, even if they are not as accurate as they could be. Many nutritionally compromised elderly nursing home residents experience significant bidirectional fluctuations in their weight from month to month. How much of this variability is a result of measurement error is not known. However, a weight loss of 5% in any month has important prognostic implications.
|
Acknowledgments |
|---|
The authors thank Albert Barber, PharmD, from NCS Healthcare, Akron, Ohio; Jeffery S. Olson, PharmD, and Michael R. Stevens, PharmD, from Bristol-Myers Squibb, Plainsboro, New Jersey; and Beverly D. Yamashita, BSN, Stephen P. Reinhart, MBA, Jeffrey P. Trotter, MM, and Xavier E. Olave, BA, from Ovation Research Group, Highland Park, Illinois, for their contributions to this study, including data collection protocol development, facility recruitment, and data collection, entry, and checking.
Received March 17, 2003
Accepted March 18, 2003
|
References |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
This article has been cited by other articles:
 |
|
 |
|
  A. C. Milne, A. Avenell, and J. Potter Meta-Analysis: Protein and Energy Supplementation in Older People Ann Intern Med, January 3, 2006; 144(1): 37 - 48. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. L. Locher, C. O. Robinson, D. L. Roth, C. S. Ritchie, and K. L. Burgio The Effect of the Presence of Others on Caloric Intake in Homebound Older Adults J. Gerontol. A Biol. Sci. Med. Sci., November 1, 2005; 60(11): 1475 - 1478. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. S. Lee, S. B Kritchevsky, T. B Harris, F. Tylavsky, S. M Rubin, and A. B Newman Short-term weight changes in community-dwelling older adults: the Health, Aging, and Body Composition Weight Change Substudy Am. J. Clinical Nutrition, September 1, 2005; 82(3): 644 - 650. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||
| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
