Physical functional ability and nutritional status are two major indicators for predicting mortality risk in older adults. A decline in functional status is a profound predictor of mortality(Reference Thomas, Kamel and Azharrudin1). The death rate increases from 8 % in individuals with no disability, to 15 % with one or more Instrumental Activities of Daily Living (IADL) disabilities, to 21 % in persons with one or two ADL dependencies, and up to 37 % in those with five or six ADL dependencies during a 2-year period(Reference Manton2). On the other hand, nutritional status is also a strong predictor of mortality. Numerous studies have shown that death rate is much higher for the elderly who are malnourished than those who are well nourished(Reference Beck, Ovesen and Osler3–Reference Chan, Lim and Ernest7). In a recent study, we have shown that the relative mortality risk is about seven times for individuals who are malnourished and 2·5 times for those who are at risk of malnutrition compared with those who are normal when rated with the Mini-Nutritional Assessment (MNA)(Reference Tsai, Yang and Wang8). The two indicators have been shown to be highly correlated, but the complementary effect between the two has not been well studied. Hence, the study was aimed to determine the extent to which the two tools (functional status and nutritional assessment) together can improve the ability to identify those who otherwise would not be identified by either one alone.
Data and subjects
The present study analysed datasets of the ‘Survey of Health and Living Status of the Elderly in Taiwan’ (SHLSET), a population-based longitudinal cohort study conducted by the Bureau of Health Promotion of Taiwan. The goal of the survey was to gain an understanding of the role of demographic, socio-economic, environmental, lifestyle and healthcare factors on health, well-being and quality of life of older Taiwanese(9). The SHLSET employed a three-stage national probability sampling method. Stage 1 stratified the Taiwanese population into 361 primary sampling units and (after excluding thirty lightly populated mountainous primary sampling units) randomly selected fifty-six primary sampling units for further sampling. Stage 2 was a proportional-to-size random selection of blocks from the selected primary sampling units and the final stage was a random selection of two eligible persons from each of the selected blocks. A total of 4412 men and women aged 60 years or older were drawn with this process and, among them, 4049 completed the first interview in 1989. A second sample of 2462 subjects, 50–66 years old, was drawn from the population with the same procedure in 1996 to maintain/extend the age of the cohort. Subjects in the combined cohort were interviewed with a structured questionnaire for eliciting demographic, socio-economic, lifestyle and health or healthcare-related data every 3 or 4 years (in 1989, 1993, 1996, 1999, 2003 and 2007). The completion rates ranged from 79·1 to 91·8 %. The 1999 survey had a special emphasis on diet and nutrition and the questionnaire included most items in the MNA, meeting the need of the present study and thus was chosen as the baseline of this analysis. The method and procedure were reviewed and approved by government-appointed representatives. The detailed sampling process of the survey has been described previously(Reference Tsai and Chang10). The present study was conducted according to the guidelines laid down in the Declaration of Helsinki and all procedures involving human subjects were approved by the ethics committee. Written informed consent was obtained from all subjects.
The Mini-Nutritional Assessment
We graded the nutritional status of each subject with both the long form (LF) and short form (SF) of the MNA which is well recognised as a simple, easy-to-use, reliable and non-invasive tool for screening/assessing the risk of malnutrition in older adults. The MNA is a two-part and two-stage scale. The LF includes eighteen items and evaluates four aspects of nutritional status (dietary, anthropometric, global and self-view)(Reference Guigoz, Langue and Vellas11). The elderly who are suspected of malnutrition are first screened with the SF consisting of six key questions. Those who are rated as at risk of malnutrition are further evaluated with the rest of the LF (additional twelve questions) to complete the diagnosis(Reference Rubenstein, Harker and Salva12). Under most conditions, the SF predicts the LF well and can function independently as a screening tool.
Because of population-related differences, we rated the nutritional status of subjects in the present study with a slightly modified version (Taiwan version-2) of the MNA. The MNA-Taiwan version-2 adopted the Taiwanese-specific anthropometric cut-off points(Reference Tsai, Ho and Chang13) and also replaced item R (calf-circumference) for item F (BMI)(Reference Tsai, Ku and Tsai14). The MNA-Taiwan version-2 has been shown to perform better or at least equal to the MNA-Taiwan version-1 (a normalised version by adopting population-specific anthropometric cut-off points). Data for all items in the MNA-LF, except items I (pressure sore/skin ulcers) and M (fluid intake), were available in the survey database. So, the MNA was based on fifteen items with a maximum score of 28 points, rather than seventeen items for 30 points. However, the total score was proportionately adjusted on the full-score basis. A final score of ≤ 16·5 suggests malnourishment; 17–23·5, at risk of malnutrition; and 24 or more, normal. All MNA-SF items were available in the questionnaire. The MNA-SF has a maximum score of 14. A score of 0–7 indicates malnourished; 8–11, at risk of malnutrition; and 12–14, normal(Reference Rubenstein, Harker and Salva12).
Functional status was assessed using a self-report questionnaire containing the ADL adapted from the 1984 National Health Interview Survey Supplement on Aging(Reference Fitti and Kovar15). The following two types of scores could be derived from the scale: one measured difficulty in carrying out the items (the method used in the present study) and the other measured the degree of difficulty in performing each of the six items (bathing, dressing, transferring, eating, walking or toileting). An item that one ‘cannot do independently’ is considered ‘dependent’ for that item. The number of items that one ‘cannot do’ was then totalled(Reference Johnson, Lui and Yaffe16). Mortality data were obtained from records maintained by the SHLSET project and confirmed with records of the Universal Health Insurance Program and the National Household Registration.
Subjects’ baseline characteristics, nutritional status, number of ADL dependencies and follow-up 4-year mortality were computed with simple statistics. We used Cox regression analysis (with the Akaike information criterion) and the net reclassification improvement (NRI)(Reference Pencina, D'Agostino and D'Agostino17) to quantify the improvement offered by the two markers, the MNA and the ADL, in predicting follow-up 4-year all-cause mortality. In Cox regression analysis, we calculated the lengths of survival time for subjects who died during the follow-up 4 years according to records maintained by the SHLSET project and took 48 months as the censoring time for subjects who survived, and we used the changes in Akaike information criterion values to indicate the improvement in predictive abilities. In an adaptation(Reference Jacobs and Brumback18) of NRI analysis(Reference Pencina, D'Agostino and D'Agostino17), we computed the differences in proportions of upward and downward reclassifications in the fatality groups minus the differences in proportions of upward and downward reclassifications in the survivor groups after the inclusion of the MNA ratings to the ADL ratings(Reference Pencina, D'Agostino and D'Agostino17). We considered the number of ADL dependencies (0, 1–2 or ≥ 3) to be the base risk, whereas the augmented risk function was the combination of the number of dependencies with the MNA ratings. For each ADL stratum, we defined upward reclassification as nutritional status worse than normal (at higher risk) and downward as normal nutritional status (at lower risk). All statistical analyses were performed with the SAS statistical software package (SAS Institute, Inc.). Statistical significance for all analyses was accepted at α = 0·05.
Table 1 shows the characteristics of the subjects. The sample included slightly more men, which reflected the composition of the age group. More than half (64·9 %) of the subjects were 65–74 years of age and the rest were over 75 years; only 21·4 % of the subjects had more than 6 years of formal education; 11 % lived alone; 25 % were current smokers; 14 % drank once or more per week and 55 % exercised more than three times per week.
* Average age 73·04 (sd 6·11) years.
† Or with non-family members.
‡ Classified according to answers to questions that asked about the frequency and length of doing routine physical exercise. The cut-off was >30 min/time.
Table 2 shows the classification of ADL status and the follow-up 4-year mortality according to the nutritional status rated at baseline. The MNA-LF rated 2·7 % as malnourished, 12·7 % as at risk and 84·6 % as normal, while the MNA-SF rated 3·5, 19·0, and 77·5 %, respectively. Greater proportions of subjects who were malnourished or at risk of malnutrition had higher numbers of ADL dependency. Applying the MNA and ADL ratings simultaneously identified additional elderly with increased follow-up 4-year mortality risk who would otherwise be missed by either scale alone. Compared with the MNA-LF, the MNA-SF showed relatively good ability in predicting follow-up 4-year mortality risk.
MNA, Mini-Nutritional Assessment; LF, long form; SF, short form.
* Baseline nutritional status was rated with the LF and SF of MNA Taiwan-version-2. Values are weighting-adjusted.
† n and (%) of the total sample.
‡ n and (%) of subjects within the nutrition category.
§ n and (%) of death within the ADL dependency–nutrition category.
Table 3 shows that both MNA (both forms) and ADL presented significant independent mortality-predictive abilities and complemented each other in predicting follow-up 4-year mortality risk (all P< 0·001). It also exhibits the improvement in predicting the follow-up mortality according to the changes in Akaike information criterion values by adding nutritional status and/or ADL status to the models. The MNA-LF showed greater improvement than the MNA-SF.
LF, long form; SF, short form; AIC, Akaike Information Criterion; CV, control variables.
* All models are weighting-adjusted.
† All models were controlled for demographic (age, sex and years of formal education), lifestyle (smoking status, alcohol drinking and routine physical activity) and health-related (hypertension, diabetes, heart disease, stroke and cancer) variables.
‡ MNA score was rated with the MNA-LF and MNA-SF, respectively, and ADL score was rated with the ADL scale. Both variables were tested as continuous parameters (MNA score or number of ADL dependencies).
Table 4 shows the cross-classification according to the ADL and MNA ratings for subjects who died and for those who survived during the 4-year follow-up. Among those who died, 205 subjects were reclassified in a higher risk and 374 in a lower risk by the MNA-LF; among those who survived, 238 subjects were reclassified in a higher risk and 2055 in a lower risk. The unstratified NRI was estimated to be 0·51 (P< 0·001), which suggested a significant improvement after the inclusion of the MNA-LF. The MNA-SF showed comparable improvement (unstratified NRI = 0·53, P< 0·001). The stratified NRI after the inclusion of the MNA-LF were 0·33 (se 0·042, P< 0·001), 0·24 (se 0·254, P= 0·355) and 0·05 (se 0·193, P= 0·790) for 0, 1–2 and ≥ 3 ADL dependencies, respectively. The MNA-LF improved the mortality prediction of the ADL alone for those with 0 dependencies, but not for those with 1–2 or ≥ 3 dependencies. The stratified NRI after the inclusion of the MNA-SF were 0·38 (se 0·047, P< 0·001), 0·28 (se 0·255, P= 0·260) and 0·08 (se 0·191, P= 0·688), similar to that observed with the MNA-LF. The stratified NRI shows that most of the improvement in classification over the ADL alone arises within the people with 0 dependencies, and little or no improvement in classification occurs in the people with ≥ 3 dependencies.
* The analysis of adding ADL classifications to MNA classifications is shown in Appendix A.
† Subjects were identified as upward reclassification if their nutrition status was worse than normal and downward if they were normal. The proportion of reclassified upward (downward) was the number of upward (downward) divided by the total in each ADL category.
‡ The standard error of the NRI was computed as the square root of (2 × (p up,eventq up,event/n event+p up,noneventq up,nonevent/n non-event)), where q= 1 − p; p up is the proportion reclassified upward; and n is the total number of subjects in each ADL category.
Table 5 shows the cross-classification of the results rated with the MNA-SF v. that rated with the MNA-LF. Among the 101 elderly rated as malnourished by the SF, thirty-nine were rated as at risk by the LF; among 545 rated as at risk by the SF, sixteen were rated as malnourished and 241 as normal by the LF; and among 2226 rated as normal by the SF, thirty-eight were rated as at risk by the MNA-LF.
Fig. 1 shows the follow-up 4-year Cox regression curves stratified by the number of ADL dependencies and nutritional status (rated with the MNA-LF and MNA-SF, respectively) adjusted for sociodemographic (age, sex and education), lifestyle (smoking, drinking and exercise) and health (hypertension, heart disease, diabetes, stroke and cancer) variables. The impact on follow-up 4-year survival by malnutrition and functional decline reached statistical significance (P< 0·05) in most cases.
Healthcare providers often rely heavily on measures of functional status such as the ADL to get an idea of the risk of mortality in the frail elderly. However, that approach may miss a significant proportion of the vulnerable individuals. Assessing nutritional status (with the MNA) in addition to measuring physical functional status (with the ADL) may add useful information. The present results suggest that measuring both ADL and MNA provides a more accurate indication of mortality risk than assessing either one alone. When the general older adults were graded into three levels of both nutritional and ADL status and the relative mortality risk was calculated, people who were malnourished had a markedly elevated risk of mortality in individuals seemingly functionally independent (those without any dependency). Subjects who had a greater number of ADL dependencies and poorer nutritional status were associated with a greater follow-up mortality risk. These results suggest that the MNA can improve the ability of the ADL to predict follow-up mortality risk in frail elderly.
The MNA has long been shown to predict the risk of mortality of the elderly living in various settings. van Nes et al. (Reference Van Nes, Herrmann and Gold19) observed that elderly hospital in-patients with poor nutritional status as rated with the MNA were associated with a longer length of stay and increased in-hospital mortality. Similarly, Donini et al. (Reference Donini, Savina and Rosano5) observed that a low MNA score was predictive of a greater incidence of adverse clinical events during hospitalisation and of higher follow-up mortality in elderly patients. Beck et al. (Reference Beck, Ovesen and Osler3) found that an MNA score ≤ 23·5 points predicted mortality risk in a group of 70–75-year-old Danes. In a prospective follow-up study, Persson et al. (Reference Persson, Brismar and Katzarski4) observed that the MNA was able to predict 3-year follow-up mortality risk in geriatric patients.
The MNA is a multidimensional and multifunctional scale. Both the LF and SF scales evaluate multi-aspects of nutritional health. The MNA has been shown to have the ability to predict functional status, hospital cost and mortality risk, in addition to nutritional status(Reference Cereda, Valzolgher and Pedrolli8,20–23). Thus, it is not surprising that the MNA can enhance the ADL in predicting the risk of mortality in the elderly.
Our regression models suggest that nutritional status and ADL dependency are two major independent predictors of follow-up mortality risk in elderly persons. The models also show that the MNA-SF is comparable with the MNA-LF in improving the ability of ADL in predicting follow-up mortality risk. Because the SF is much simpler and time-saving to administer than the LF, it is often much more desired in routine practice. Although the MNA-LF is stronger than the MNA-SF in predicting nutritional risk, the SF appears as effective as the LF in predicting follow-up mortality risk.
The results also show that among the 101 elderly rated as ‘malnourished’ by the MNA-SF, the MNA-LF ‘clarifies’ that thirty-nine of these (39 %) are actually at lesser risk (at risk). Among the 545 rated as ‘at risk’ by the MNA-SF, almost half (241) are relabelled as ‘normal’. Thus, among those in whom any action should be taken, the LF shows that a lesser action would suffice in 280 (43 %) out of 646 individuals. Changes to a diagnosed greater risk state are less frequent, only fifty-four people. Thus, the MNA-SF has a great deal of similar signal to the MNA-LF, but certain classifications are different in many people. Whether this matters depends on the cost of further diagnosis or intervention action. The 241 individuals who are rated as at risk by the MNA-SF but as normal by the MNA-LF are probably those who are having emerging nutritional problems.
The complementarity can also be demonstrated by adding the ADL classifications to the MNA classifications (Appendix A). With any dependency taken as upward reclassification and no dependency taken as downward reclassification, the unstratified NRI is 0·34 (se 0·033, P< 0·001), while the stratified NRI are 0·12 (se 0·027, P< 0·001) for those who are rated as normal by the MNA-LF, 0·35 (se 0·087, P< 0·001) for those who are rated as at risk, and 1·10 (se 0·211, P< 0·001) for those who are rated as malnourished. Similar results are obtained by adding the ADL classifications to the MNA-SF classifications. Thus, addition of the ADL classifications to the MNA classifications leads to significant reclassification improvement, particularly in the malnourished subgroup.
Strengths and limitations
A major strength of the study is that the data were from a longitudinal cohort study involving a relatively large population-representative sample. Baseline physical functional (ADL) and nutritional (MNA) status can infer a causal relationship with the follow-up mortality risk. However, there are also limitations. (1) Despite a relatively large sample size, when classified by the ADL and MNA simultaneously, the number of observation is relatively small in some classifications. (2) Most data were obtained through interviews. In individuals who were cognitively impaired, data were mostly obtained through proxies. The quality of such data might be compromised. (3) No biochemical data were available in the datasets to support the present findings. Thus, confirmation of the nutritional status graded with the MNA with biomarkers is not possible.
Using Cox regression analysis and the NRI, we have conclusively shown that nutritional (MNA) and functional (ADL) ratings complement each other in predicting the follow-up 4-year mortality risk in the elderly. Rating the elderly with both tools simultaneously can provide a more accurate prediction of mortality risk than with either one alone. The MNA and ADL are two simple and non-invasive scales that can be easily administered to the elderly living in almost any setting. Simultaneous application of the MNA and ADL appears to be an effective and economical way of identifying those elderly who are in need of intervention.
The authors thank David Jacobs, PhD, for contributions concerning the use of the NRI in this study. The present study is based on the data from the ‘SHLSET’, provided by the Bureau of Health Promotion, Department of Health, ROC (Taiwan). Descriptions or conclusions herein do not represent the viewpoint of the Bureau. The authors declare that they have no competing interests. A. C. T. conceived the idea and directed the study, and wrote the manuscript. L.-C. L. and J.-Y. W. performed the statistical analysis and reviewed the manuscript.