No effect of thyroid hormones on 5‐year mortality in patients with subjective cognitive decline, mild cognitive disorder, and Alzheimer’s disease

Abstract The present study aimed to investigate differences in circulating thyroid hormone levels, gender, education, depressive symptoms, and cognitive performance among patients with cognitive impairment, and also to examine their associations, as well as that of cognitive decline, with 5‐year mortality. Between 1998 and 2017, a hospital‐based, single‐centre (Neurology Department of the General Hospital in Vienna, Austria), retrospective follow‐up study enrolled 2102 patients with mild to severe cognitive impairment (grouped into subjective cognitive decline, mild cognitive impairment, and Alzheimer's disease). Cox proportional hazards models were used to calculate hazard ratios (HRs), with 95% confidence intervals (CIs), as well as to calculate stepwise adjustments for demographic variables (age, gender, and education), depressive symptoms (Geriatric Depression Scale, GDS‐15), and neuropsychological abilities (four domains of a neuropsychological test battery of Vienna, NTVB). In univariate analyses, total triiodothyronine (TT3) levels differed significantly between Alzheimer's disease and mild cognitive impairment patients (p diff = .001). No other differences in cognitive impairment subgroups with any of the measured thyroid hormones were observed. Furthermore, in multivariate models, circulating TT3 was not associated with mortality (multivariable‐adjusted HR per unit increase in TT3 = 0.56; 95% CI = 0.29–1.07). In multivariate models, we observed significantly lower 5‐year mortality among women (HR = 0.56; 95% CI = 0.43–0.73) and those who scored higher on any of the four domains of the NBTV (e.g., attention and perceptual speed, HR = 0.63; 95% CI = 0.54–0.72); we also observed significantly higher 5‐year mortality among patients with depressive symptoms (HR per one point score increase in GDS‐15 = 1.06; 95% CI = 1.02–1.10), regardless of cognitive impairment subgroup. Among patients with various degrees of cognitive impairment, we found no associations of thyroid hormone levels with 5‐year mortality. Gender, neuropsychological abilities, and depressive symptoms were each significant predictors of 5‐year mortality. These results suggest that a neurocognitive test performance could serve as an important predictor of 5‐year mortality among patients with cognitive impairment, although further studies with a more complete adjustment for comorbidities are needed to confirm these findings.


| INTRODUC TI ON
Dementia is one of the most common causes of death in the senior population, 1-3 with Alzheimer's disease (AD) representing more than half of all diagnosed senile dementias. 4 However, cognitive decline does not always happen rapidly and there are transitional stages.
Subjective cognitive decline (SCD) as the earliest diagnostic stage of dementia on the trajectory of cognitive decline is identified as perceived cognitive decline, in which only the afflicted person notices differences of present cognitive abilities compared to the previous ones. 5 Patients with more serious mild cognitive impairment (MCI) 6 (i.e., the next stage on the dementia trajectory) exhibit a higher risk of developing Alzheimer's disease (i.e., the final stage on the trajectory to dementia), as well as other forms of cognitive impairment. 7 Previous studies have consistently shown higher mortality rates in senior populations with MCI or AD. 2 Early symptoms of AD are mostly pronounced as episodic memory problems, followed by difficulties in executing a complex task, followed by the inability to perform everyday activities. Behavioral changes set in, followed by disorientation, confusion, impaired mobility, and potential hallucinations. 8 Ultimately, motoric difficulties lead to swallowing and malnutrition, which leads to the acquisition of often fatal pneumonia among the already weak patients. 9 Additional factors such as age, gender, and comorbidities, including cardiovascular diseases and diabetes mellitus, influence the increased mortality of patients with dementia, regardless of the severity of their cognitive decline, 10 although cognitive decline in itself as measured by various cognitive test batteries has also been associated with an increased mortality in the geriatric population. 11 Furthermore, depression is also an important factor because Byers et al. 12 found that veterans with depression or dysthymia have a higher change of developing dementia and have a higher mortality rate than veterans in an euthymic state. Another factor of interest comprises the circulating thyroid hormones.
Thyroid dysfunctions are very common in the general population and are associated with an increased risk of cardiovascular diseases, diabetes, various cancers, lung diseases, and psychiatric diseases. 13 Women and the elderly are most at risk of having thyroid dysfunction. 14 Similar to most organs in our body, the thyroid gland is affected by the passing of time, which can be beneficial or disadvantageous for overall survival. Thyroid-stimulating hormone (TSH) levels generally rise in older adults, and a subclinical hypothyroid state may be beneficial to longevity as reported by Lang et al. 14 Gusseklo et al. 15 showed that unusually high levels of TSH and low levels of free thyroxin (fT4) might be associated with a low mortality rate among the senior population compared to older patients with low TSH levels and high fT4 levels, who had the highest mortality rate. Another study 16 found that older men between the ages of 73 and 94 years with lower free fT4 serum levels had a higher 4-year survival rate than older men with normal fT4 serum levels. By contrast, Laulund et al. 13 reported that thyroid dysfunctions, such as overt and subclinical hyperthyroidism, as well as overt hypothyroidism, have a significant association with an increased mortality rate.
In light of this mixed evidence for thyroid hormones, we hypothesized that there is a difference in circulating thyroid hormones between SCD, MCI, and AD patient groups, as well as for demographic factors (gender, education), depressive symptoms, and cognitive performance. We also hypothesized that the above mentioned factors and cognitive decline will have an effect on the 5-year mortality of patients affected with SCD, MCI, and AD.

| MATERIAL S AND ME THODS
The present study represents a retrospective, single-centre analysis, with all data having been collected over a period of almost 20 years (January 1998 to December 2017) from patients within the Neurology Department of the Medical University of Vienna. Of note, subjects with SCD were only documented starting in 1999.
The relevant data for the present study were extracted from the Research, Documentation, and Analysis (RDA) system of the Medical University of Vienna, or the Allgemeines Krankenhaus Information's Management system (AKIM) in a corresponding SPSS (IBM Corp.) data matrix, and subsequently analyzed. If no death certificate was confirmed until 31 December 2017 in either the RDA or AKIM, then the patient was assumed to be alive.

| Participants
Patients were diagnosed and referred by physicians from the Department of Neurology or they had a self-complaint about memory problems in recent times. Patients were diagnosed on the basis of their cognitive performance into three diagnostic groups: SCD, MCI, or AD. All participants had a neurological examination, using criteria from Jessen et al. 5 for the diagnosis of SCD, and Petersen's criteria from 2004 for the diagnosis of MCI. 17 NINCDS-ADRDA guidelines were taken into account for the diagnosis of AD. 18,19 Serum thyroid important predictor of 5-year mortality among patients with cognitive impairment, although further studies with a more complete adjustment for comorbidities are needed to confirm these findings.

K E Y W O R D S
5-year mortality, Alzheimer's disease, MCI, neuropsychological battery, SCD, thyroid hormones hormone levels were measured in blood samples collected from the patients at one of their visits to the Department of Neurology.
Study participants were included in the study if they were at least 50 years of age at the time of neuropsychological testing. People who had a stroke, a serious head injury or psychiatric problems causing pseudo-dementia were excluded from the study. The subjects were given the Mini-Mental State Examination (MMSE), 20 the Geriatric Depression Scale (GDS-15), 21

and the Neuropsychological
Test Battery Vienna (NTBV-short). 22 The study was approved by the ethics committee of the Medical University of Vienna and followed the tenants of the Declaration of Helsinki.
Information about circulating thyroid levels was taken from the AKIM-System. Only values that were acquired ±3 months within the neuropsychological setting were taken into account. Some patients had incomplete blood parameters, with one or more missing thyroid parameters. If patients did not have a TSH value, or the existing thyroid parameters did not match the neuropsychological test date within a ±3-month period, they were excluded from the analyses.
Of the 2102 individual patients whose data were retrieved at the beginning, 591 did not meet the required criteria of the study.
Specifically, 459 individuals lacked thyroid parameters (specifically TSH parameters) and 127 had invalid data or were too young to participate in the study. An additional five patients had to be removed as a result of double counting, leaving a total of 1511 patients as our base population. Figure 1 illustrates the exclusion protocol for patients who did not meet the required criteria of the present study.

| Neuropsychological instruments
Neuropsychological tests were used to determine and evaluate the cognitive performance among the subjects in the study. MMSE score was not included in descriptive statistics or further analyses because of its redundant role as a neuropsychological test in the place of the NTBV-15, a modified short version of NTBV that is also used to determine Parkinson's disease. 22 Furthermore, NTBV-short was included, which measures categories such as psychomotor speed, concentration or attention, language, memory, or executive functions. 22 Finally, the GDS-15 was included, which is a questionnaire with yes/no answers designed to measure geriatric depression. 21

| Statistical analysis
For descriptive purposes, we calculated the mean ± SD, and present the range of minimum (min) and maximum (max). If the distribution of data was skewed, we present medians and interquartile ranges (IQR) (25%-75%). To deduce whether the frequency of distribution was skewed, Kolmogorov-Smirnov and Shapiro-Wilk tests were used.
Additionally, analyses were performed utilizing the nonparametric Kruskal-Wall test and the Mann-Whitney U test to compare the diagnostic subgroups.
Cox proportional hazards models were used to examine associations of thyroid parameters and varying levels of cognitive impairment (SCD, MCI, and AD) with mortality. Because, in univariate analyses, the results of Kruskal-Wallis testing revealed significant differences by diagnostic subgroups only for TT3 F I G U R E 1 Flowchart, showing the exclusion of cases from the study and the remaining participants. AD, Alzheimer's disease; MCI, mild cognitive impairment; SCD, subjective cognitive decline; TSH, thyroid-stimulating hormone (χ 2 = 11.332, df = 2, p = .003), and all other thyroid parameters (TSH, fT3, fT4, and TT4) showed no significant difference between diagnostic subgroups (p ≥ .168), only TT3 was retained for multivariable testing.
Patients were followed until they were deceased or until the end of follow-up (31 December 2017), whichever came first.
Multivariable models adjusted for age, gender, and number of years in school (representative of educational status), NTVB-short (representing neurocognitive functioning), and GDS-15 score (representing depressive symptoms). The calculation was performed using five hierarchical blocks via the enter method, where, in each block, the predictors were inserted simultaneously. All statistical analyses were performed using SPSS, version 26 (IBM Corp.). p < .05 (twosided) was considered statistically significant. Bonferroni adjustment was performed to account for multiple testing.

| RE SULTS
Demographics, clinical and circulating thyroid parameters data are summarized in Table 1.

| Neuropsychological functions and depressive symptoms
NTBV-15 short subtest results were first analyzed utilizing a principal component analysis, using the varimax rotation method with Kaiser normalization to achieve a dimensional reduction. The rotation converged in nine iterations, and a four-dimensional solution can be suggested, which represents the information contained in an optimized way, as shown in Table 2 Subsequently, NTBV factor z-score performances were compared between diagnostic subgroups using Kruskal-Wallis testing.
The test value H of the Kruskal-Wallis test showed a significant difference in factor scores between the diagnostic subgroups in each case (p < .001). Likewise, with post-hoc pairwise comparisons using the U test procedure between diagnostic subgroups, the Mann-Whitney U test also showed a significant difference between ing GDS-15 did not show any further significant differences for the subgroup comparisons.

| Model for influencing factors on 5year mortality
To summarize all the previous results, we used a Cox proportional hazards model. In the model, 61.6% of the patients, in total 931 participants (n SCD = 106, n MCI = 612, n AD = 213), could be included. Of these, 245 (26.3%) were deceased, whereas 686 (73.7%) were cen- In the first block of the Cox regression, only the TT3 thyroid parameters (ng mL -1 ) were analyzed. The result was significant, showing that patients with lower circulating TT3 levels had a higher risk of dying sooner (p < .001).
In the second block of the model, we included cognitive diagnostic groups. The result was significant, showing that patients with a more severe diagnosis had a higher 5-year mortality risk (p ≤ .013).
TT3 levels remained significant (p = .005), meaning that lower TT3 levels are still a risk factor.
In the third block, we included the sociodemographic parameters of school years and gender. In the case of gender, men tended to die sooner than women (p < .001). In the case of school years, those with higher education lived longer, indicating lower educational sta-  5-year mortality of the patient collective. The prototypical TT3 difference between deceased and censored cases can be assumed as −0.58 ng mL -1 ; a lowered TT3 has a trend (reciprocal value of HR = 1.79) in predicting the 5-year mortality of patients with cognitive degeneration, as shown in Table 3.  40 also revealed that poor neuropsychological cognitive performance is correlated with a higher death toll, although it lost its significance after adjusting for comorbidities such as heart disease.

| DISCUSS ION
Although it did have a significant impact on 5-year mortality until the last block of the Cox regression model, cognitive diagnosis lost this after taking NTVB factor scores into account. This is also substantiated by the fact that the age of death in the cognitive diagnosis groups is relatively the same regardless of the severity of the condition. This contrasts with several studies suggesting that cognitive decline has an impact on mortality. 2,7,9,10,[41][42][43][44] This could be because diseases such as pneumonia, dehydration, and cachexia are more likely to occur in individuals dementia than in individuals from the healthy senior population. 9,10 One possible explanation is that social and psychological factors, rather than dementia itself, influence mortality. 10 Furthermore, Mooldijk et al. 45 showed that, in the general population, individuals with MCI die sooner than those without MCI, although the life expectancy is higher than in studies done in clinical settings. As mentioned above, the patient population with AD is significantly older than the rest of the study collective. Therefore, it is more deductive to attribute the mortality of the patients to age rather than cognitive impairment. Age and other variables, including gender, diabetes, cancer, and stroke, were considered as risk factors in mortality for patients with dementia in a longitudinal study in Taiwan. 34 It should be noted that the present study compares the mortalities in individuals of a population already diagnosed with cognitive decline amongst each other and not with a healthy population, similar to that conducted in most previous studies reporting a coincidence between mortality and dementia. 2,9 The present study has several limitations. Although the study had 1511 participants, a large portion of the patient collective had incomplete datasets. In the final multi-regression model, only 931 participants (61.2% of the whole patient collective) had all of the required data needed to be included in the regression model. One of the reasons why the thyroid data were incomplete was a result of the change in the laboratory results over the decades; TSH, TT3, and TT4 were regularly measured in the late 1990s up until the middle 2000s, whereas, after that point, thyroid examinations changed to TSH, fT3, and fT4. Second, the present study used multiple testing, which is susceptible to data-dredging and bias-finding of statistically significant results. 46 To counter this bias, we used the Bonferroni adjustment. Finally, the present study investigated thyroid hormone levels in the serum and not in the cerebrospinal fluid.
In conclusion, the present study shows that TT3 has no significant impact on the survival and 5-year mortality of patients afflicted with cognitive degeneration. To understand the relationship between thyroid parameters and dementia, further studies are needed.
The data showed that female patients with cognitive impairment tend to live longer than male patients with cognitive impairment.
The cognitive diagnosis did have an impact, until the last block of the Cox regression. Depressive symptoms appear to have had an impact on predicting the survival of patients, but the influence is rather small. Neurocognitive functioning is a valuable 5-year mortality predictor in patients with cognitive impairment, and further exploration of their predictive value is recommended.

ACK N OWLED G M ENTS
DB owes special thanks to the Registrar's Office of the Medical University of Vienna for financial support.

CO N FLI C T O F I NTE R E S T S
The authors declare that they have no conflicts of interest. A positive vote of the Medical University of Vienna from January 2020 with the EK-Nr. 2040/2018 is available.

DATA AVA I L A B I L I T Y
The data that support the findings of this study are available from the corresponding author upon reasonable request.