Accelerated Weight Loss May Precede Diagnosis in Alzheimer Disease FREE

Some reports suggest that mild weight loss in old age may be a downstream effect of normal aging processes associated with reduced metabolic demands^{1 - 2} and appetite^{3 - 5} and diminishing physical stature and height.^{6 - 7} However, unintentional weight loss at any age may be a marker for disease.^{8 - 9} Even mild weight loss in old age is associated with specific diseases and increased mortality.^{10 - 19} More recently, Alzheimer disease (AD) has emerged as a contributor to age-related weight loss. In late-stage dementia of the Alzheimer type (DAT), there is a loss of up to 2 pounds of body weight per year.²⁰ Weight loss is associated with faster progression of dementia and with nursing home placement.^{20 - 24} In studies that compared weight changes in people who eventually develop DAT vs those who remain without dementia, decreasing body weight increased the risk for dementia,^{18 ,25} and participants who eventually developed DAT lost weight faster than their counterparts without dementia.^{26 - 29}

The exact course of weight loss and its temporal relationship to the onset of DAT remain unknown. Our goals were to determine the interval between the occurrence of weight loss and the onset of dementia and to characterize the rate of weight change over time in older (aged 65-95 years) adults with and without dementia.

We examined archival data from all available research participants without dementia and with weight measurements who were enrolled in a longitudinal study of memory and aging between January 1, 1991, and March 1, 2005 (demographics reported in the Table). Of the 449 individuals who were initially without dementia, 60 men and 65 women developed DAT; 132 men and 192 women remained without dementia through their last assessment. All participants spoke English and lived in the greater St Louis metropolitan area. Twenty-three were African American and the remainder were white.

At enrollment and annual follow-up, experienced clinicians assessed each participant for the presence and severity of dementia using the Clinical Dementia Rating (CDR).³⁰ The CDR evaluates cognitive function in each of 6 categories (memory, orientation, judgment and problem solving, performance in community affairs, home and hobbies, and personal care). Ratings are based on information gathered during a 90-minute semistructured clinical assessment with the research participant and a knowledgeable collateral source (usually a spouse or an adult child) but without reference to psychometric performance or results of previous evaluations.^{31 - 33} A CDR of 0 indicates no dementia; and a rating of 0.5, 1, 2, or 3 corresponds to very mild, mild, moderate, or severe dementia. The CDR has high interrater reliability, with weighted κ values ranging from 0.75 to 0.94.^{34 - 35} Embedded in the clinical evaluation are standard cognitive screening tools, such as the Short Blessed Test for cognitive impairment³⁶ and the Mini-Mental State Examination.³⁷ The CDR has been validated in studies of all forms of dementia, is sensitive to clinical progression, and is highly correlated (93%) with autopsy-confirmed AD.^{31 - 32 ,38}

At each time of assessment, the participant was weighed using a professional medical scale with light clothes on and shoes off. Experienced clinicians took the participant's medical history from the collateral source and the participant, which included questions pertinent to weight loss. Clinicians summarized the interviews to indicate whether specific disorders were present (stroke and cardiovascular incidents, diabetes mellitus, hypertension, depression, head trauma, anosmia, dysphagia, and caregiving) or rated symptoms along a continuum (physical health³⁹ and socioeconomic status⁴⁰ ). The presence of APOE (apolipoprotein) ε4 alleles was determined for 395 participants (278 without dementia and 117 with DAT) using restriction enzyme isotyping.⁴¹ Participants who did not undergo genotyping were assessed before the initiation of genotyping at our center.

Although multiple medical diagnoses were allowed for the participants with dementia, DAT was the primary diagnosis. People with other potentially dementing disorders (eg, Parkinson disease) were excluded. Participants reporting depressive symptoms but not meeting the criteria for major depression⁴² were included. Longitudinal data for the participants without dementia included all times of assessment from study enrollment through last available assessment. Longitudinal data for the participants with DAT included all available assessments at which they received a CDR of 0 or 0.5, except for 8 who progressed directly from a CDR of 0 to a 1. We did not examine any data from participants rated as having a CDR of 1 unless it was their first assessment with a diagnosis of DAT. In all, there were 2224 times of observation.

Group comparisons of quantitative measures at enrollment and at last assessment were conducted using t tests. The χ² test of independence was used for nominal variables. Because the longitudinal course of weight data did not meet standard linear assumptions, several competing models were fit to the data to determine an appropriate analytic strategy, including quadratic and cubic transformations and piecewise linear regression. This model-fitting procedure is described in detail by Fitzmaurice et al.⁴³

Piecewise linear regression fit the data best and showed that the longitudinal course of body weight data are best represented by 2 linear segments with different slopes joined together by a single inflection point (Figure). Using the linear element-effects model (PROC ELEMENT in SAS statistical software, version 9.1; SAS Institute Inc, Cary, NC), we applied simultaneous between-group tests to these segmented data. A element-effects model (random intercept and slopes) has an advantage over standard repeated-measures designs in that intraindividual and interindividual variability is estimated more efficiently and is less susceptible to differences in premorbid ability. For example, the weight of the individual at study enrollment is accounted for by the random intercept specification of the model. These models are also less affected by missing data or unequally spaced follow-ups, and can account for multiple slopes over time.

Element models can also be used to account for potential mediating effects (eg, age and diabetes mellitus).^{43 - 45} Thus, in a second set of analyses, we controlled for the mediating effects of variables that may affect declining body weight as a function of dementia status. The longitudinal contributions of a variety of health risks were tested to determine potential covariates, thus yielding a purer index of dementia-related weight loss because age, sex, and other confounding effects were statistically controlled. Longitudinal tests of covariates differ from cross-sectional tests in that they determine not only group differences but also whether differences are related to weight loss over time (weight × group × time interaction).

Descriptive characteristics for the 2 groups are reported in a time-sequential fashion in the Table. The 2 groups were similar with respect to the proportion of men and women, educational level, and socioeconomic status. Although the DAT group had a slightly higher proportion of the APOE ε4 allele (as measured by at least 1 allele) compared with the group without DAT, this difference was not significant. All genotypic comparisons were conducted, but no group differences were found. The group that developed DAT was older (4.6 years) and weighed less (8.2 pounds) at study enrollment compared with those who remained without DAT. Those who developed DAT performed slightly worse at enrollment on the 2 cognitive screening measures (Mini-Mental State Examination and Short Blessed Test). The 2 groups were not different at enrollment in marital status, health rating, or various medical conditions or in taking potentially anorectic medications.

Follow-up for the participants with and without DAT was between 5 and 6 years. The 2 groups differed on the 2 cognitive screening measures at the last assessment (P<.001 for both). The participants with DAT (n = 449) were more likely to be widowed (χ²₁ = 4.87, P<.05) and were in poorer physical health (as reported on the Physical Health Rating Scale, P<.001) than the group without dementia. They reported more depressive symptoms (n = 447) (χ²₁ = 6.88, P<.001) and were more likely to take cholinesterase inhibitor and N-methyl-D-aspartate agonist drugs (n = 441) (χ²₁=18.76, P<.001). This sample is relatively naïve for dementia medications approved for the symptomatic treatment of AD (186 observations of 2224 total). Several factors explain why so few participants were treated with approved drugs. First, all the participants were research volunteers; clinical management was provided by their primary care physicians, not by the research clinicians. Second, most participants (117 of 125) in the group with dementia were staged CDR 0.5 and may have been too minimally impaired to trigger treatment decisions by their primary physicians. Third, some participants were enrolled and assessed before market introduction of approved symptomatic therapies for AD. All analyses were performed with and without observations during which participants reported taking dementia medications. The pattern and magnitude of results were equivalent. In general, dementia medications slightly accentuated weight loss, but a separate study with more people taking these medications is needed to gain reliable estimates of the contribution these medications make to weight loss. The groups with and without dementia did not differ significantly at last assessment in terms of any of the other medical conditions or medications (P>.05 for all).

The longitudinal course of body weight data did not meet standard linear assumptions. There was a significant group × time squared interaction (t_53.3 = 2.53, P<.01), indicating that the overall course of weight loss differed between groups. There was a significant linear trend in the group without dementia (t₁₈₉ = 5.17, P<.001), but not quadratic (t_39.9 = 0.13, P>.05), whereas the linear and quadratic terms (t_55.9 = 5.08, P<.001; and t₁₃₀ = 2.48, P<.01, respectively) were significant for the DAT group. That is, both groups lost weight, but the rate of weight loss changed over time for the DAT group. A piecewise linear regression was applied to account for differential acceleration in weight loss between the groups (Figure).

Seven time points were tested, including the time of assessment when dementia was first diagnosed (T0), 3 times of assessment before the dementia diagnosis (T − 1 through T − 3), and 3 times of assessment after the dementia diagnosis (T + 1 through T + 3). Likelihood ratio tests confirmed that body weight among the older adults with DAT was best fit by a piecewise linear function with 2 segments joined 1 year before the diagnosis of dementia (G² ₂ = 101.6, P<.001). Thus, weight loss in the DAT group is best described by 2 linear functions that bend sharply at the assessment 1 year before diagnosis (Figure). In contrast, the data for the group without dementia were best fit by a constant linear function that did not change over time.

In the group without dementia, weight loss was 0.65 lb (SE, 0.13) per year for the entire study period. In the group with dementia, weight loss was similar (0.68 [SE, 0.27] lb per year; t₅₃₇ = 1.05, P>.05), until 1 year before the dementia diagnosis, when there was a sharp acceleration such that the slope roughly doubled in magnitude (−1.34 [SE, 0.30] lb per year; t₅₂₈ = 2.36, P<.05). Although the groups with and without DAT did not differ in their rate of weight loss before the T − 1 assessment, there was a significant main effect of weight (t₅₃₇ = 3.19, P<.01), indicating that the group with DAT was already lighter at study enrollment (8.2 lb).

A second set of analyses examined the role of potential mediating influences to weight loss. The older a participant was at study enrollment, the more weight that individual lost over time (t₄₀₅ = −2.19, P<.05), especially as a function of cognitive status. Sex and health status were also significant (t₃₆₀ = 4.23, P<.05; and F_4,1288 = 2.53, P<.05, respectively). The same was true for hypertension (t₁₄₆₉ = −2.12, P<.05) and stroke history (t₁₂₄₃ = −2.63, P<.05). Marital status (t₁₃₁₇ = −0.52, P>.05) and depressive mood (t₁₄₁₆ = −0.20, P>.05), however, were associated with dementia status but were not related to weight loss over time. All other health risks investigated were not related to either group or time (educational level, APOE ε4 status, diabetes mellitus, and reported changes in appetite). Thus, the second set of analyses of weight loss and dementia status included age at study enrollment, sex, health status, hypertension, and stroke history as covariates.

The covariate model showed a similar pattern of significance to the first model. Compared with healthy aging participants, individuals who progress to DAT begin to lose about twice as much weight 1 year before the onset of symptoms. Older adults without dementia lost 0.48 (SE, 0.13) lb per year, while the older adults with dementia lost 1.17 (SE, 0.30) lb per year. Controlling for health risks attenuated weight loss equally in both groups.

This study examined weight loss and its temporal correspondence to dementia onset. Although other studies^{18 ,25 - 29} have shown that weight loss is an antecedent of DAT, to our knowledge, this is the first to characterize the time course of weight loss and its relationship to dementia onset. In a sample of clinically well-characterized older adults who were cognitively healthy at enrollment, an acceleration in the rate of weight loss was a harbinger of the change from nondemented status to DAT. Participants lost about 0.6 lb per year while without dementia, but 1 year before the first symptomatic detection of DAT, the rate of weight loss in individuals doubled to 1.2 lb per year. Also, participants with DAT already weighed less (about 8 lb) than participants without dementia at study enrollment, and this difference persisted throughout the study.

Small multiplicative effects of age, sex, health, hypertension, and stroke with weight were found. Similar to other studies^{18 ,20 - 24 ,26 - 27 ,29} of weight loss and aging, older people lost more weight than younger people, men lost more weight than women, people in poorer health lost more weight than people in good health, and individuals with cardiovascular symptoms lost more weight than those without such symptoms. Although weight loss in those with dementia was greatest when these health risks were present, none of these risks mediated the doubling relationship between dementia status and weight loss. Dementia of the Alzheimer type was the driving cause of weight loss in this longitudinal sample. Other health risks investigated (depressive symptoms, marital status, use of potentially anorectic medications, APOE ε4 expression, diabetes mellitus, and appetite change) were not related to weight loss and dementia progression. Cardiovascular factors, such as myocardial infarction, transient ischemic attacks, stroke history, smoking, and hypertension, also seem to be unrelated in previous studies.^{25 - 26 ,29}

Consistent with studies^{1 - 7 ,11 - 12 ,15 ,17 - 18} that suggest very modest weight loss in later life may be because of normal age-related physiologic changes, we found that aging without dementia was marked by gradual declines in body weight. These effects are small and do not account for the weight loss in those with DAT. Individuals who eventually develop dementia (preclinical DAT)⁴⁶ have already lost a significant amount of weight 4 to 6 years before diagnosis. The onset of dementia exaggerates mild weight loss, doubling its rate in the year before the clinical detection of the mildest behavioral symptoms.

Taken together, findings from the present study and from previous epidemiologic studies^{26 - 29} suggest that the preclinical course of DAT is marked by 2 phases of increasing weight loss. At midlife, participants who will eventually have dementia weigh as much as their peers without dementia. At late midlife or early late life, these individuals begin to lose weight at faster rates, and by 6 years before dementia detection, participants who will eventually develop DAT are 6 to 8 pounds lighter on average. The present study shows that at least 1 year before dementia detection, the rate of weight loss again increases. Thus, weight loss associated with DAT probably begins very early in the course of the disease and then accelerates in the 1 to 2 years before the onset of cognitive symptoms.

The mechanisms responsible for these findings are unknown. A psychosocial hypothesis is that individuals with dementia forget to eat; however, this seems unlikely given that weight loss starts before the onset of symptoms and that marital status (the presence of a caregiver for most participants) is unrelated to weight loss in this and other studies.²⁶ Caregiver burden outstripping the ability to prepare food for the individual with dementia seems unlikely for the same reasons. The individuals with DAT were more depressed, but depression did not vary with body weight. This is similar to other findings from this center, in which depression did not affect the rate of cognitive decline in the same sample.⁴⁷ Furthermore, 3 other studies^{25 - 26 ,29} controlled for depressive symptoms and no associations were found. These observations contradict the hypothesis that weight loss in late life is caused by depression.^{48 - 50}

Medically, DAT could affect appetite; however, our study supports findings from Barrett-Connor and colleagues²⁶ indicating that appetite remains unchanged (as reported by the individual with dementia and a knowledgeable collateral source). An unchanged appetite does not, however, indicate a stable diet. There are reports of mild to moderate changes in taste and smell in healthy aging populations^{51 - 53} and in populations with dementia,^{54 - 57} and these factors need to be measured rigorously in future studies. Subtle gustatory changes could result in cumulative decreases in caloric intake or decreases in the quality of food consumed by individuals with DAT. Given the present findings, these changes would have to be present for at least 6 years while without dementia and then significantly worsen 1 to 2 years before dementia detection.

Overall, we believe that these findings suggest that ongoing pathophysiologic changes in preclinical AD are related to weight loss. Future investigation of weight loss during the preclinical stage may shed light on the pathogenesis of DAT. Examining the overlapping biological processes associated with cachexia, anorexia, and other wasting syndromes may provide new insights. If true, weight loss is not a risk factor for DAT; rather, it is an early manifestation of the disease. Given the time course of weight loss as healthy aging individuals convert to DAT, future studies of frailty and weight loss should closely monitor and control for participants' cognitive status over time.

The present study is limited in making conclusions about the preexisting 8-lb group difference because we did not observe individuals from midlife. It is possible that heavier individuals lose body weight faster than lighter individuals. Conversely, heavier individuals who would otherwise have dementia may have disease in midlife, preventing their inclusion in this late-life longitudinal research study. Optimally, more sophisticated weight indexes (eg, body mass index or percentage of fat) are needed to address these issues. Proportional estimates of weight loss may be associated with cognitive status; thresholds of weight loss deserve further investigation. Furthermore, some of the group without dementia may actually contain individuals who have preclinical DAT. If weight loss accelerates before dementia diagnosis, the inclusion of individuals with DAT in the sample without dementia would attenuate group differences in observed weight loss and weaken results. Thus, the reported results may slightly overestimate weight loss in aging persons without dementia (compare with estimates from White and colleagues²⁸ ). Also, neuropathologic follow-up of these individuals is needed to rule out confounding non–AD-associated dementias and may yield insight into weight loss and atypical dementias. It would be useful to characterize atypical dementias to determine if weight loss is DAT specific or more general to other types of late-life brain disease. Finally, too few participants were treated with cholinesterase inhibitor drugs, precluding the analysis of the effect that these drugs may have on weight loss in dementia. These data are preliminary, and a larger more detailed study is required.

Correspondence: John C. Morris, MD, Alzheimer's Disease Research Center, Washington University School of Medicine, 4488 Forest Park, Suite 130, St Louis, MO 63108 (morrisj@abraxas.wustl.edu).

Accepted for Publication: May 10, 2006.

Author Contributions:Study concept and design: Johnson, Wilkins, and Morris. Acquisition of data: Johnson and Morris. Analysis and interpretation of data: Johnson. Drafting of the manuscript: Johnson. Critical revision of the manuscript for important intellectual content: Johnson, Wilkins, and Morris. Statistical analysis: Johnson. Obtained funding: Morris. Administrative, technical, and material support: Morris. Study supervision: Morris.

Funding/Support: This study was supported by grants P01 AG03991 and P50 AG05681 from the National Institute on Aging (Dr Morris); and by the Alan A. and Edith L. Wolff Charitable Trust.

Acknowledgment: We thank Martha Storandt, PhD, Chengjie Xiong, PhD, and Elizabeth Grant, PhD, Washington University, for their assistance in preparing the manuscript; and the Clinical (Dr Morris, leader) and Genetics (A. M. Goate, DPhil, leader) Cores of the Alzheimer's Disease Research Center for providing diagnostic and genotyping data.