Dementia as a Predictor of Mortality in Dialysis Patients

Materials and Methods

A national registry (the US Renal Data System [USRDS]) was analyzed in a retrospective cohort study of the association between a predialysis diagnosis of dementia and subsequent survival after initiation of dialysis. The variables included in the USRDS standard analysis files (SAF), as well as data collection methods and validation studies, are listed at the USRDS web site (http://www.usrds.org) under “Researcher’s Guide to the USRDS Database,” Section E, “Contents of all of the SAF’s” (Standard Analysis Files), and published in the USRDS. The demographics of the dialysis population have been described previously (2002 USRDS report). The study cohort was all dialysis patients who initiated dialysis from April 1, 1995, to December 31, 1999, and had evidence of Medicare or Medicaid as primary payer at or before the date of initiation of dialysis, which included dates for initiation of dialysis as well as subsequent death as reported to the USRDS through multiple mechanisms as specified in the USRDS researcher’s guide. These files were merged with the study cohort using unique patient identifiers.

Inpatient and outpatient Medicare claims records were searched by International Classification of Diseases, Ninth Revision code for the following diagnoses, with dates of coding before the initiation of dialysis: Alzheimer’s disease (331.0x), vascular dementia (290.40x through 290.43x), dementia not otherwise specified (290.0x through 290.43x), and presenile dementia (290.x). These codes are grouped together in the term “dementia” for the remainder of the article. For the purposes of this study, a diagnosis of dementia was considered established by the finding of one claim for inpatient and two codes for outpatient claims for the particular diagnosis. All analyses were performed using SPSS 13.0 (SPSS, Inc., Chicago, IL). Files were merged and converted to SPSS files using DBMS/Copy (Conceptual Software, Houston, TX). Statistical significance was defined as P < 0.05. Univariate analysis was performed with χ² testing for categorical variables (Fisher exact test for violations of Cochran’s assumptions) and t test (Mann-Whitney for non-normal distributions) for continuous variables. Independent associations with dementia were assessed using logistic regression. Variables with P < 0.10 in univariate analysis for a relationship with dementia, as well as those that were thought to have a defined biologically plausible relationship with dementia, were entered into logistic regression analysis as covariates. Patterns of missing values of continuous variables were explored and missing values were interpolated using SPSS Missing Values 7.0 (both regression and expectation-maximization method). Missing values of categorical values were coded as an additional category in multivariate analysis but for parsimony in presentation are not presented in unadjusted analysis. A sensitivity analysis was performed using patients with nonmissing values for all variables.

Patient survival curves were calculated using the life table analysis method. Time to death was defined as the time from the date of dialysis initiation until the date of death, censored at the date of renal transplantation, loss to follow-up, or the end of the study period, December 31, 2001; all patients had a potential of 2 yr minimum follow-up. Cox regression models were performed testing associations with dementia with model building on the basis of both univariate testing for association with survival and forced entry of variables that were shown in previous analyses of the USRDS to be associated with patient survival. Covariates included were age by quartiles, body mass index (BMI) by quartiles, serum albumin by quartiles, diabetes, gender, ethnicity, heart failure, chronic obstructive pulmonary disease, stroke, peripheral vascular disease, serum creatinine, lupus, smoking status, cancer, alcohol or drug use, Medicaid as primary payer, inability to ambulate, inability to transfer, and year of dialysis initiation. For validation purposes, an additional model was run using renal transplantation as a time-dependent variable. The association of dementia with survival was calculated as a time-dependent variable in Cox regression, with all patients with dementia having values of 1 and all other patients having values of 0. For purposes of comparison with other diagnoses for which patients undergo dialysis, such as multiple myeloma and light-chain disease, this cause of death was entered into the model as a predictor variable for death as well.

Discontinuation of dialysis also was assessed as an outcome variable, because patients with dementia may develop difficulties in cooperating with the dialysis process. This variable was missing for 65% of patients in the database. However, this is a special variable that usually is not tracked in the treatment history files, and patients who did not discontinue dialysis would be far less likely to have a value of “no” entered in the database as long as they are routinely continuing dialysis. Therefore, for purposes of analysis, we considered all missing values as 0. In a sensitivity analysis, we also performed analysis limited to patients with either yes or no (collapsed from individual categories of why dialysis was discontinued, as indicated in the variable RXSTOP).

Results

From April 1, 1995, through December 31, 1999, 366,738 patients started long-term dialysis. Of these, 272,024 were documented as having Medicare or Medicaid as primary payer at the time of the first dialysis session. This cohort therefore composed the study population.

A total of 1725 patients were identified as having dementia before the initiation of dialysis. Patient characteristics are demonstrated in Table 1 for patients who had a diagnosis of dementia versus the rest of the cohort. The patients with dementia were older, more likely to be female, and more likely to be black. In addition, patients with dementia had lower BMI, albumin, and creatinine and had a higher hematocrit. Patients with dementia also were more likely to have coronary artery disease and congestive heart failure as additional comorbidities. They were less likely to have HIV or to abuse tobacco.

Patients with dementia (12.4%) were significantly more likely to discontinue dialysis before death than those without dementia (7.5%; P < 0.001 by χ²). The same was not true for patients with mental retardation (9.8 versus 7.5%; P = 0.18).

Logistic regression of factors that were associated with dementia yielded significant findings. As shown in Table 2, advancing age was independently associated with dementia. The odds ratio (OR) for the oldest quartile was 19.65 (95% confidence interval [CI] 14.48 to 26.67; P < 0.05), 10.22 for the second oldest (95% CI 7.50 to 13.92; P < 0.05), and 2.93 for the third oldest (95% CI 2.09 to 4.12; P < 0.05). The inability to transfer (OR 1.65; 95% CI 1.30 to 2.10; P < 0.05) and the inability to ambulate independently (OR 2.04; 95% CI 1.71 to 2.44; P < 0.05) both were positively correlated with dementia. Heart failure (OR 1.14; 95% CI 1.03 to 1.26; P < 0.05), diabetes (OR 1.17; 95% CI 1.06 to 1.30; P < 0.05), and cancer (OR 1.28; 95% CI 1.05 to 1.57; P < 0.05) were the final parameters with a positive association with dementia. Higher serum albumin seemed to be negatively associated with dementia, although these data were missing for approximately 30% of the cohort. The OR for the highest quartile for albumin was 0.44 (95% CI 0.38 to 0.52; P < 0.05), 0.62 for the next highest quartile (95% CI 0.54 to 0.71; P < 0.05), and 0.77 for the third highest quartile (95% CI 0.68 to 0.86; P < 0.05). Higher quartiles of BMI likewise were negatively associated with an OR of 0.52 for the highest quartile (95% CI 0.45 to 0.60; P < 0.05), 0.62 for the next highest (95% CI 0.55 to 0.71; P < 0.05), and 0.74 for the third highest (95% CI 0.66 to 0.84; P < 0.05). Female gender (OR 0.87; 95% CI 0.79 to 0.96; P < 0.05) and white race (0.76; 95% CI 0.69 to 0.84; P < 0.05) also were negatively associated. Transplantation also was less likely in patients who had dementia (adjusted OR 0.29; 95% CI 0.15 to 0.55). Discontinuation of dialysis was significantly more common among those with dementia (adjusted OR 1.37; 95% CI 1.19 to 1.58). A total of 73.1% of patients had complete data for the variables analyzed via logistic regression.

The average time to death for patients with mental retardation was 1.1 versus 2.7 yr for the rest of the cohort (P < 0.001 via log rank test). Univariate Cox regression yielded an adjusted hazard ratio of 3.2 for dementia (P < 0.001). Multivariate Cox regression for dementia as a time-dependent variable shown in Table 3 produced an adjusted hazard ratio for death of 1.91 (95% CI 1.77 to 1.98; P < 0.001). Results that incorporated renal transplantation into the model were not substantially different. The 2-yr survival rate was calculated because this was the shortest follow-up period that theoretically was possible. The 2-yr survival rate for patients with dementia was 24 versus 66% for those without dementia (P < 0.001).

Discussion

Many authors have discussed the importance of assessing patients who are dialysis candidates for cognitive impairment. Cognitive impairment that is severe enough to preclude reliable compliance is an accepted contraindication to kidney transplantation (11). Among dialysis patients, however, as long as the patient is cooperative enough to allow the procedure, dialysis is not withheld. This study was unable to determine whether the inability of patients with dementia to cooperate directly with the dialysis process may have been a reason for the decreased survival of these patients after initiation of dialysis. The Renal Physicians Association/American Society of Nephrology Clinical Practice Guidelines on Shared Decision Making in the Appropriate Initiation of and Withdrawal from Dialysis (12) offer several useful recommendations that are appropriate to patients with cognitive dysfunction. In particular, recommendations 7 (special patient groups) and 8 (time-limited trials) may apply to those with cognitive dysfunction as identified in this study. There is no way to know from our study design how many patients with dementia were not started on dialysis. However, dementia certainly could be considered “terminal illness from a nonrenal cause or… a medical condition [that] precludes the technical process of dialysis,” for which it may be reasonable to consider not initiating or withdrawing from dialysis. Recommendation 8 opens the possibility of a time-limited trial for patients with an uncertain prognosis or those for whom a consensus cannot be reached. It also is unclear whether such time-limited trials were performed or were more common for patients with dementia, compared with patients without this condition. However, as a time-dependent variable, the adjusted hazard ratio for death associated with a diagnosis of dementia actually was lower than that for other diagnoses that are considered acceptable indications for initiation of dialysis (13), such as multiple myeloma/light-chain disease in Table 3. Discontinuation of dialysis before death was significantly more common for patients with dementia.

A recent review estimated that as many as 60% of dialysis patients may meet diagnostic criteria for cognitive dysfunction (14). The much lower prevalence of dementia in our study suggests underdiagnosis by coding. In Sweden, coding has been shown to be insensitive (43%) compared with chart review (15). However, in the same study, the specificity of coding for dementia was 98%. A similar finding of low sensitivity but high specificity has been reported for the Medical Evidence Form (16). It therefore is less likely that patients who were identified as having dementia in our study do not have dementia but likely that patients who do have dementia would not be identified as such. Such misclassification generally would bias findings toward the null, that is, lack of statistical significance (17). The persistently poor prognosis that is associated with cognitive dysfunction in this study, even in adjusted analysis, also may be related to the practice of coding only of patients with the most severe manifestations of cognitive dysfunction. We cannot exclude the possibility that coding for dementia represents a “middle ground” of severity, and patients with the most severe dementia were not initiated on dialysis. This would have less predictable effects on study findings. This also would make comparison with studies of dementia in the general population difficult. However, the diagnosis of dementia carries important prognostic information in other populations as well. One large, prospective study described a mean survival time of 7.5 yr for patients who were younger than 75 yr and had received a diagnosis of Alzheimer’s disease (18). Population-based reports have shown a survival time of 3.3 yr for prevalent cases of dementia with a mean survival of 7.6 yr after the initial diagnosis was made (19). Mean survival cannot be compared between this study and other studies because of differences in time of follow-up. Nevertheless, it seems that survival for patients with dementia after dialysis is lower than for such patients who are not on dialysis, just as life expectancy for patients who are on dialysis generally is lower than for comparable cohorts of the general population who do not undergo dialysis.

Our data give a new importance to assessing dialysis candidates for mental impairment as a result of the extremely short life expectancy of this subpopulation. The reasons for this undoubtedly are multifactorial. On average, patients with dementia in our cohort were hypoalbuminemic and had a lower BMI than the overall population; both of these have already been associated with mortality (20). The subgroup with dementia was also older than the cohort in general. These patients also were more likely to have coronary artery disease and congestive heart failure. Without reliable data on cause of death for our study population, it is unclear whether this is of any significance to overall mortality. However, as presented earlier, patients with dementia in the general population still have a longer life expectancy than was found in our data. A thorough assessment of a patient’s cognitive function may be one method of predicting which patients will do poorly on dialysis and will affect how patients are counseled about renal replacement therapy. As stated previously, put in the context of other serious illnesses for which dialysis is offered, the adjusted mortality of dementia actually was less than that found in this study and previously reported for multiple myeloma (13).

Our study has the benefit of using a large cohort of dialysis patients. Theoretically, this cohort is not a sample that estimates a population but actually includes the entire dialysis population in the United States. The USRDS collects data in a prospective manner throughout the United States.

There clearly are many limitations to our study. These data were limited to patients with Medicare or Medicaid insurance because of the methods of ascertaining Medicare claims. They may not be representative of patients with other types of insurance, particularly for these diagnoses. Another potential limitation to our study is limitation of data reporting. Cause of death analysis and a subgroup analysis by type of dementia was attempted but was hampered by missing entries. Although it is not possible to prove, we believe that dementia most likely was underreported and underdiagnosed in this cohort. To preserve temporal relationships between diagnosis and subsequent survival, we did not assess dementia that occurred after the initiation of dialysis; in any case, new-onset (incident) cognitive dysfunction likely is less common than prevalent cognitive dysfunction. As previously cited, Periera et al. (21) believe that cognitive dysfunction may exist in up to 60% of dialysis patients. Although this may be an aggressive estimate, our patients who had a diagnosis of dementia accounted for <1% of the overall dialysis population. This might be plausible if these diagnoses represent only the extremes of cognitive dysfunction.

It is likely that social support systems, which are not assessed well by the USRDS, well may be related to the subsequent survival for such patients. Because of the rarity of these diagnoses in our study, analysis using the USRDS special studies (which have some metrics for social support) was not feasible.

Despite the limitations as presented here, our data are the first assessment of longevity for patients who have cognitive dysfunction and are started on dialysis. The increased risk for mortality in this group of patients needs to be discussed openly to provide truly informed consent before initiation of renal replacement therapy. Such patients should be considered for the options presented in clinical practice guidelines on shared decision-making, particularly careful consideration of whether initiation of palliative care truly is the best decision for the patient, and consideration of time-limited trials. Further exploration as to the causes of decreased survival among these vulnerable patients is warranted.