Predictors of mortality in patients diagnosed with probable Alzheimer's disease

Physicians and public health planners are often asked to predict mortality in patients diagnosed with Alzheimer's disease (AD). Unfortunately, there are few studies to guide these predictions, and published findings often do not apply to general patient populations. Many study populations consist of subjects from geriatric or dementia clinics, psychiatric hospitals, or nursing homes. These populations are often biased with over-representation of advanced or early cases. ^[1] Other biases include the elimination of patients with comorbid conditions and the use of prevalent rather than incident cases. Few studies have evaluated predictors of mortality in incident cases ^[2,3] within a community-based population. ^[4,5]

Several factors seem to contribute to mortality in patients with AD. The age and gender of the patient, as well as the age at disease onset, are often cited. ^[5-15] Some investigators have linked mortality to AD markers, such as the severity of cognitive loss ^{[4,7,9,10,12,15-17]} or the degree of language impairment. ^[10,18] General debility ^{[8,15,16,19-21]} or behavioral changes ^[9,16,22] have been associated with decreased survival. Finally, comorbid conditions, particularly vascular disease, are often a major contributor to mortality in elderly patients, including those with dementia. ^[4,8,18,20]

This study examines mortality in a cohort of patients with AD. This population-based cohort enrolls incident cases in whom dementia has recently become diagnosed. Mortality in these demented patients was compared with that in the general population in the same county. We used univariate and multivariate models to determine baseline characteristics associated with early mortality.

Methods.

Subjects with previously undiagnosed cognitive impairment were identified within a geographically defined portion of a large health maintenance organization (Group Health Cooperative). ^[23,24] The Seattle area membership of Group Health Cooperative includes approximately 23,000 persons aged >or=to60 years. In this age group, attrition is approximately 1% per year, excluding deaths. The Alzheimer's Disease Patient Registry (ADPR) seeks to identify all new cases of dementia as they are observed by or brought to the attention of primary care physicians, specialty clinics, or other disciplines within the medical care setting. Two independent surveillance strategies have been developed: the marker event strategy and the primary care physician contact strategy. The marker event strategy includes routine review of areas where demented patients are likely to appear for medical care, including neurology, CT brain scans, admission and discharge records, mental health clinics, emergency rooms, and the computerized Treatment Record Form of all clinic visits. The primary care physician contact strategy consists of directed publicity about the study, especially monthly letters to physicians describing interesting research findings and reminding physicians to refer persons with newly recognized dementia symptoms. Approximately 40% of ADPR enrollees enter through the primary care physician contact strategy. The records of persons identified through screening strategies are reviewed to eliminate persons obviously ineligible for enrollment. Approximately 20% of subjects identified through the surveillance network decline to allow their names to be released to the ADPR, and an additional 14% decline to provide informed consent. A follow-up review of the medical records of these persons indicated that they were slightly older (mean age 80 years), more likely to be female (about two-thirds are women), and less likely to be currently married (40% married). Roughly one-half of the refusals had a diagnosis consistent with AD. Enrollees underwent a standardized diagnostic evaluation performed by ADPR physicians and psychologists. ^[24] Cases were followed annually and retested to verify the diagnoses. Diagnosis was accomplished by consensus of at least two physicians and one psychologist. Only those ADPR enrollees meeting National Institute of Neurological and Communicative Disorders and Stroke-Alzheimer's Disease and Related Disorders Association (NINCDS-ADRDA) criteria for probable AD ^[25] were included in the current analysis. Those cases categorized as possible AD, non-AD dementia, or not demented were excluded from the study.

Demographic and medical history variables were selected from a comprehensive set of measures gathered at the time of enrollment and clinical examination, including demographic variables, measures of dementia severity, and comorbid conditions. Demographic variables included age, gender, and education. Three education categories were used, including lower than high school, high school, and higher than high school. Measures of dementia severity included Mini-Mental State Examination (MMSE) scores, ^[26] Blessed dementia rating scale (DRS) scores, ^[27] Boston naming test scores, rate of cognitive decline, and presence of wandering or agitation. ^[16] The rate of cognitive decline was estimated using MMSE decline (30 minus MMSE score at enrollment) occurring since the time of symptom onset (estimated by the patient and proxy during the initial interview). Comorbid conditions included heart disease, diabetes, hypertension, smoking, alcohol use, reported weight loss, head trauma, and sensory impairment affecting the ability to hear or read. Smoking was analyzed both as ``smoked'' versus ``never smoked'' and as ``currently'' versus ``not currently'' smoking. Alcohol use was analyzed both as ``used'' versus ``never used'' and as ``currently using'' versus ``not using'' alcohol. Weight loss was evaluated using a decrease of 10% or greater in body weight reported by the patient or proxy. Sensory impairment affecting the ability to hear ^[28] or read ^[29] was evaluated using validated bedside clinical screening questions completed by the proxy. Vascular comorbid conditions were evaluated both separately and as a combined variable, including heart disease, hypertension, and diabetes. Most variables were categorized as present or not present when examining their main effect on survival. Age, education, rate of symptom progression, Blessed DRS, MMSE, and the Boston naming test were categorized in tertiles. Age and MMSE and Blessed DRS scores were used as continuous variables when entered in multivariate models. Survival intervals were calculated from the date of enrollment to the date of death. For those patients still alive at final follow-up, survival intervals were censored at that time. The source population for the ADPR is similar to the surrounding metropolitan population with regard to age (13% aged >or=to65 years in both groups), gender (55% vs. 51% female), and ethnicity (91% vs. 90% white). ^[30] Nine percent of the source population had <12 years of education (vs. 18%) and 34% had >15 years of education (vs. 24%). The annual income of the source population in 1994 U.S. dollars was less than $15,000 in 20% (vs. 23%) and more than $50,000 in 13% (vs. 19%). The source population was more likely to be married (67%) than the surrounding metropolitan population (56%). The mortality of this cohort was compared with that of the surrounding county (King County). Because mortality rates are available for King County residents only in those <85 years of age, it was necessary to restrict information contributed by cohort members to information prior to their 85th birthday. For King County residents, age- and gender-specific death rates for each 5-year age group (in which patients were included) were applied to the age-gender distribution of our AD case sample to obtain expected deaths.

The survival analysis was done using proportional hazards regression techniques. Initially, each baseline characteristic was screened using a univariate unstratified proportional hazards regression or a log-rank test, or both, based on stratifying the survival data by the covariate of interest. Kaplan-Meier curves were used to illustrate the univariate results. Significant prognostic factors identified in the univariate analyses were then candidates for inclusion in multivariate models. Age and gender were considered important confounders of the association between prognostic factors and survival and were included in the multivariate models, regardless of associations found in the univariate analyses. For confounding adjustment, models were stratified by gender because there was evidence that the gender-specific hazards were not proportional. Age was incorporated in the linear predictor as a continuous variable with both linear and (centered) quadratic terms.

The measure of disease severity most strongly associated with survival was the Blessed DRS. We searched for important prognostic factors in survival beyond the effects of disease severity; in those models, Blessed DRS was included as a covariate. Other predictors from the univariate analyses were then evaluated in the multivariate proportional hazards model if they were clinically significant and showed a trend toward statistical significance (p <or=to 0.15). Predictors were included in the final model only if they were statistically significant (p <or=to 0.05). A search for influential observations, assessments of a good fit, and tests for interactions were done. There was no evidence in the data that it was necessary to develop different models of survival based on baseline dementia severity. SAS and Splus software were used.

Results.

Of the 766 patients enrolled in the ADPR between 1987 and 1993, 327 were diagnosed with probable AD and included in this study. The baseline demographic and medical characteristics of these patients are shown in Table 1. The average age at entry was 79 years, and the mean MMSE score at enrollment was 20 (first and third quartiles 16 and 23, respectively). The interval from enrollment to final follow-up ranged from 0.1 to 6.5 years (median 3.3, first and third quartiles 1.4 and 4.0, respectively). There were 898 person-years of follow-up.

Of the 327 cases with probable AD, 93 died, 23 occurring during the first year and 51 during the first 2 years of follow-up. The unadjusted mortality for the cohort was 10.2 per 100 person-years. In the group <85 years of age, 60 deaths were observed during 668 person-years of follow-up (9.0 deaths per 100 person-years). The expected number of deaths in King County residents <85 years of age, given the same age and gender distribution, was 29 (4.3 per 100 person-years). The ratio of observed to expected deaths (Standardized Mortality Ratio) was 2.1.

Univariate analysis results are presented in Table 2. Among demographic variables, age and male gender were associated with shortened survival. Among the measures of disease severity, MMSE score, Blessed DRS score, rate of symptom progression, and wandering or agitation were associated with shortened survival. Comorbid conditions associated with shortened survival included sensory impairment affecting hearing and reading. Individual markers of vascular disease, including heart disease, diabetes, and hypertension, were not associated with shortened survival. A variable combining vascular disease, diabetes, and hypertension was modestly associated with shortened survival. Baseline variables not associated with shortened survival from the univariate comparisons included education level, Boston naming test scores, smoking, alcohol use, weight loss, head trauma, and depression.

After adjustment for age and gender stratification, functional limitation, as measured by the Blessed DRS, had the strongest association with shortened survival in the multivariate model. After inclusion of the Blessed DRS, sensory impairment affecting reading was also significantly associated with shortened survival (see Table 2). This model was improved when the interaction between Blessed DRS score and age was entered. However, the data presented in Table 2 do not include this interaction. A combined variable for heart disease, hypertension, and diabetes was associated with shortened survival after adjusting for age and gender (p = 0.04), but did not remain significant after inclusion of Blessed DRS and sensory impairment affecting reading. Additional variables were not independently associated with shortened survival. The relative risks for mortality based on age and Blessed DRS score are shown in Table 3. The interaction indicates that risk is lower for the oldest patients with high Blessed DRS scores relative to the risk for younger patients with high Blessed DRS scores. Figure 1 shows the effect on shortened survival of sensory impairment affecting the ability to read in female cohort members.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1. Kaplan-Meier plot of sensory impairment affecting the ability to read in female cohort members based on a fixed Blessed dementia rating scale score and age.

Discussion.

Our patients with newly diagnosed AD had increased mortality compared with the age- and sex-adjusted general population from the same county. On univariate analysis, mortality was associated with age, gender, MMSE score, Blessed DRS score, rate of symptom progression, wandering or agitation, heart disease or diabetes, and sensory impairment affecting reading or hearing. In a multivariate model adjusted for age and gender, the Blessed DRS score (indicating cognitive and physical impairment due to dementia) had the strongest association with shortened survival. After inclusion of the Blessed DRS score in the model, only sensory impairment affecting reading contributed significantly to the survival model. Patients with heart disease or diabetes had a modest increase in mortality, which was not statistically significant at the 0.05 level.

Several measures of debility were associated with shortened survival in our screening analysis. Although often used as a measure of overall cognitive decline, the Blessed DRS also measures debility affecting activities of daily living. The baseline Blessed DRS had the strongest association, with subsequent mortality, of any variable. Measures of debility that have been associated with increased mortality in other studies include the Blessed DRS, ^[31] the clinical dementia rating scale, ^[21] an activities of daily living scale, ^[15] decreased grip strength, ^[20] and severe physical impairment, unstable gait, and falls. ^[19]

Sensory impairment not only impacts a patient's interest and pleasure in life but may also have a particularly detrimental effect in demented patients who are already suffering from a decreased ability to process environmental stimuli. Sensory impairment affecting reading or hearing was associated with decreased survival on univariate analysis. Other investigators have found a similar association between survival and debility. Walsh et al. ^[16] reported that hearing but not visual loss was associated with increased mortality, but the association was not retained in a multivariate model including age. Van Dijk et al. ^[8] reported both hearing and visual impairment to be associated with mortality. These variables likely serve as markers for general debility rather than affecting mortality directly.

Wandering or agitation was associated with mortality on univariate analysis. Behavioral changes have been associated with mortality in a number of studies, ^[9,22] including wandering or falling, ^[16] irritability or nocturnal disruptive behaviors, ^[15] and dependency, physical disability, and inactivity. ^[8]

Increased dementia severity has been associated with increased mortality. ^{[4,7,9,10,12,15-17]} MMSE score was associated with mortality in this study, but the association was weaker than that seen with the Blessed DRS, possibly because the Blessed DRS includes measures of functional ability as well as cognition. The estimated rate of MMSE decline was associated with increased mortality on univariate analysis but not in a multivariate model. The difficulty in estimating the age of dementia onset may have attenuated the effect on survival of rate of MMSE decline. Direct measures of cognitive decline over time would be preferable. Specific areas of cognitive dysfunction, such as language dysfunction, have been associated with mortality in some studies, ^[10,18,31] but not in our cohort of patients with early dementia.

Other comorbid conditions have frequently been associated with mortality in dementia. We found a vascular disease variable (presence of myocardial infarction, angina, diabetes, or hypertension) to be associated with increased mortality only after adjusting for age, gender, and Blessed DRS score. This variable was not associated with mortality on univariate analysis. Although related to mortality in the present study, comorbid vascular conditions were overwhelmed by the effects of age, gender, and debility. Other workers have similarly found an association between mortality and vascular factors, such as hypertension, ^[20] cardiovascular disease, ^[8] pulse pressure, and ischemia score, ^[18] but these findings are not universal. ^[8,9,20] Nonvascular comorbid conditions were associated with mortality in some studies ^[8,19] but not in others. ^[16] Cachexia has been associated with poor survival in some studies, ^[4,18] but we did not find weight loss to be significant as defined in this study, probably because our subjects showed early signs of dementia. Cachexia may only become a significant problem in relatively long-surviving cases. The weak association with these comorbid conditions in our study might be strengthened by larger numbers of subjects or longer follow-up.

The mortality we observed among patients with AD was about twice that of the age- and gender-adjusted death rate of the surrounding county. Others have also noted similar increases in mortality with dementia. ^[5,9,10,19] Schoenberg et al. ^[3] found an increase of 23% after 5 years and 57% after 10 years in a study of population-based incident cases. In incident subjects <60 years of age, Treves et al. ^[2] found a 60% increase in mortality with dementia. In a prevalence study of noninstitutionalized subjects, Evans et al. ^[4] found a 53% mortality rate in those with probable AD compared with 24% in control subjects.

The cohort selected for this study overcomes limitations found in many previous studies. Our use of sequential newly recognized cases from a health maintenance organization minimizes the selection bias noted in studies of dementia clinic, psychiatric hospital, and nursing home populations. ^[1] The use of incident cases allows survival to be evaluated in earlier stages of the disease, thus avoiding to some extent, the obvious increased mortality of those with advanced dementia. ^[1] Rapidly progressing cases are not excluded from the present study, although they may be under-represented in prevalence-based cross-sectional studies. ^[1] Our population should be more representative of patients seen in the offices of primary care providers after their first diagnosis of AD, compared with the populations used in many previous studies.

This study has some limitations. Our survey methods are limited in that only those patients with suspected dementia were screened; thus, some demented patients in the population may not have been identified as cases. Because these cases did not come to medical attention, their exclusion should not affect the applicability of these results in most clinical settings. Our results apply only to short-term mortality in those newly diagnosed with AD and should only be extended to patients with advanced disease with caution. Although our cohort is large, we may have had insufficient power to detect a true difference with some variables that were uncommon in this population (e.g., individual vascular variables). False associations between survival and baseline variables could occur by chance owing to the analysis of multiple variables. We minimized this risk by including only a reduced subset of variables selected a priori from a larger list of variables and by selecting variables to address specific hypotheses. Many of these variables were suggested in previously published studies, such as duration and severity of dementia, ^{[4,7,9,10,12,15-17]} cachexia, ^[4,18] language skills, ^[10,18] hearing deficit, ^[16] and wandering or falling. ^[16] For other characteristics, we based selection on a likelihood that they might affect survival, including a history of heart disease, hypertension, and diabetes.

This study should provide useful information about survival in patients newly diagnosed with AD. Although there are many risk factors for mortality in this elderly population, the factor with the greatest potential for modification is general debility. For care givers, our results suggest that efforts should be made to minimize debility. Providing hearing and visual aids, improving physical conditioning, optimizing function in activities of daily living, and addressing behavioral changes of dementia may be considered. Further study will be needed to determine whether minimizing general levels of debility in those with early AD affects longevity or improves functional well-being.