|Year : 2021 | Volume
| Issue : 2 | Page : 120-126
Investigation of the complex structure between the severity of alzheimer's disease and influencing factors using latent class cluster analysis
Didem Derici Yildirim1, Mumine Bozdag Kiraz2, Bahar Tasdelen1, Aynur Ozge3
1 Department of Biostatistics and Medical Informatics, Mersin University, Mersin, Turkey
2 Department of Neurology, Tarsus State Hospital, Mersin, Turkey
3 Department of Neurology, Faculty of Medicine, Mersin University, Mersin, Turkey
|Date of Submission||22-Jun-2020|
|Date of Decision||18-Nov-2020|
|Date of Acceptance||26-Dec-2020|
|Date of Web Publication||15-Jun-2021|
Didem Derici Yildirim
Department of Biostatistics and Medical Informatics, Faculty of Medicine, Mersin University, Ciftlikkoy, Campus, 33343, Mersin
Source of Support: None, Conflict of Interest: None
Objective: The cognition of Alzheimer's disease (AD) has a heterogeneous pattern. It is useful to obtain more information about specific subgroups of patients to prevent disease progression. For better identification of the population, we aimed to detect latent groups based on cognitive test scores using latent class (LC) cluster analysis and influencing factors of latent severity groups to assist practitioners in outpatient departments who have restricted time and instrumentation. Materials and Methods: Data for 630 patients with AD in the Mersin University Dementia Outpatient Unit were collected, and cognitive test scores, demographic variables, and other factors such as comorbidities and family history of dementia were obtained. Initially, LC cluster analysis was performed to distinguish subgroups considering clinical dementia scores, age, and sex as covariates. Second, univariate analysis was used to detect the relationship between latent subgroups and influencing factors. Finally, multinomial logistic regression was performed to identify the magnitude of risk for significant factors. Results: Four severity groups were defined as mild, moderate, severe, and very severe cases of AD, and severity was significantly related to educational level, hyperlipidemia, diabetes mellitus, and sarcopenia (P < 0.001, P = 0.001, P = 0.043, and P < 0.001, respectively). Family history also influenced severity (P = 0.024). Disease severity increased with decreased education levels. Family history predicted a 1.555-fold increase in the risk of being in the moderate group versus the mild group. Moreover, diabetes mellitus predicted a 3.690-fold increase of being in the very severe group versus the mild group. Conclusion: LC cluster analysis is effective for determining severity groups for AD, and study results will help prepare a guide for an optimum evaluation tool for the disease.
Keywords: Alzheimer's disease, cluster analysis, cognitive subgroups, latent class model
|How to cite this article:|
Yildirim DD, Kiraz MB, Tasdelen B, Ozge A. Investigation of the complex structure between the severity of alzheimer's disease and influencing factors using latent class cluster analysis. Neurol Sci Neurophysiol 2021;38:120-6
|How to cite this URL:|
Yildirim DD, Kiraz MB, Tasdelen B, Ozge A. Investigation of the complex structure between the severity of alzheimer's disease and influencing factors using latent class cluster analysis. Neurol Sci Neurophysiol [serial online] 2021 [cited 2021 Sep 18];38:120-6. Available from: http://www.nsnjournal.org/text.asp?2021/38/2/120/318502
| Introduction|| |
Dementia is a disease that affects the daily life functions and independence of patients and causes serious learning and memory disorders., The burden of diseases such as dementia, diabetes, osteoarthritis, hypertension, heart failure, and cancer rises, especially for people over aged over 60 years. According to the World Health Organization, dementia is one of the diseases that causes the most loss of function and financial burden, especially in high-income countries. Unfortunately, the number of people with dementia throughout the world has been increasing along with the growing elderly population.,
The most common cause of dementia is Alzheimer's disease (AD), a heterogeneously structured disease that occurs as a result of genetic, environmental, and demographic factors. Therefore, AD should be treated while considering all factors affecting the disease burden. These factors can be summarized as comorbidities, the family history of the patient, and demographic variables, according to previous studies. Even though AD has no curative treatment, its severity can be reduced by considering influencing factors. In addition, AD is a heterogeneous disorder because of the disease's progression, characteristics, and problems. Hence, latent subgroups should be identified to examine the complex relationship between the severity of AD and influencing factors. This is essential to provide a natural history of each case and to organize supportive mechanisms including care. In our study, latent class cluster analysis (LC cluster) was used to detect homogeneous subgroups of patients based on cognitive test scores, using Clinical Dementia Rating (CDR) scores, age, and sex as covariates. Then, differences between subgroups were evaluated in terms of demographic variables, comorbidities, and the family history of the patients.
| Materials and Methods|| |
In total, 630 Turkish patients with AD dementia aged 50 years and older were included in the study. The patients visited the neurology clinic of Mersin University Hospital between January 2002 and February 2018. Baseline neuropsychological assessments, medical history, and demographic information were recorded for all patients. Dementia diagnosis has been made using a comprehensive history of the patients provided by the case and other associated persons, complete neurological examinations, and required laboratory.,
All patients presented to the physician with various levels of forgetfulness affecting their daily lives. Patients with depression and other types of dementia (e.g., frontotemporal and vascular) were excluded. The differential diagnosis of depression and dementia (also comorbidity) was made according to the clinical consultation in the psychiatry department.
The flow diagram of the study is shown in [Figure 1]. The local research ethics committee approved the study on June 6, 2018 (Decision number: 2018/257). Informed consent was not required because the dataset consisted of de-identified secondary data released for research purposes.
Cognitive functioning tests were given to all participants to evaluate the severity of AD in all patients at once.
Mini-Mental State Examination (range: 0–30)
Participants were screened using the Turkish version of the Mini-Mental State Examination (MMSE; score 0–30) to evaluate global cognition. This test consists of five subparts: consistency (10 points), record memory (3 points), attention and calculation (5 points), recall (3 points), and language (9 points).
Clock Drawing Test (range: 0–10)
The Clock Drawing Test (CDT) was applied by giving patients a white piece of paper and asking them to draw a circle with the numbers 1–12 in the correct positions for a clock; the patients were then asked to draw the time, ten past eleven (11:10). This test was important for assessing participants' skills in planning, memory, attention, abstraction ability, and executive functions. The test results were scored by the neurologist according to the 10-point clock test developed by Sunderland and taking into consideration the studies for use of the Turkish version of the test.,
Word Memory Test (range: 0–10)
The Word Memory Test (WMT) consists of learning essays 1, 2, and 3. The list of 10 words used in the learning trials (oil, building, arm, coast, letter, cat, stick, ticket, grass, and motor) was described verbally by a practitioner in three sets sorted in different ways. After each reading, they were asked to repeat the words they could remember. The number of words remembered was recorded as the score.
Boston Naming Test (range: 0–15)
The Boston Naming Test (BNT) was used to evaluate the visual nomenclature performances and lexicon-semantic skills of the patients. It consisted of 15 simple drawings that used a black color on a white background. The patients were asked to say the names of these drawings. The number of pictures they identified correctly was recorded as their score. The total score of the test was 15. For all cognitive tests used in the study, higher scores indicated stronger cognitive levels.
Latent class cluster analysis
Some classifying methods, such as clustering, discriminant, and logistic regression analysis, have been used by researchers to classify individuals. Although these methods are popular, there has recently been considerable interest in using latent class analysis (LCA) for determining latent subgroups using manifest variables. LCA explores homogeneous subgroups from heterogeneous data by maximizing the similarity within the groups and minimizing it between groups, like all cluster analysis methods. There are important differences between traditional methods and LCA. Traditional cluster analysis algorithms are based on the nearest distance, but LCA is based on the probabilities of group allocation. Latent clustering is more flexible than other methods. There are no restrictions on the scale of observed variables, normal distribution, or unknown variances for LCA. The group number and membership must be known in discriminant or logistic regression analysis, but unknown groups may be detected with LCA. The key assumption of LCA is local independence. This means that the indicator variables are not correlated and are only related to each other through the latent variable.
LC cluster models are used for ordinal indicators and are equivalent to LC models for Poisson counts. An important extension of our study is the inclusion of covariates to estimate class membership. It is an iterative method. The marginal probabilities of the combinations of responses given to different variables are calculated as follows. The main LC cluster model is given in Equation 1. In the model, yi is the score of observed variables set, the latent number of clusters is K, πk is the probability of being cluster k, zi is the covariate value of i object, and J is the total number of indicators.
Next, posterior probabilities are calculated using marginal probabilities with the formula given in Equation 2. Individuals are classified as LCs according to these posterior probabilities. The posterior membership probability can be obtained using Bayes' theorem.
In our study, the statistical analysis was conducted in two steps. First, the LC cluster method was used to identify the latent severity subgroups in the population according to the cognitive test scores, as explained above. The decision of the group number was the most important step. It was then necessary to compare the models by changing the number of groups (from one to five) to determine which was most appropriate for the data. Model fit statistics, such as Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC), were used to determine the best model; those with the lowest information criteria were considered the best. Entropy R2 was calculated to evaluate the success of latent classification; higher entropy indicated a better classification performance. Bivariate residuals were used to detect the local dependence assumption. LCA was performed using the Latent Gold 5.1 software. Second, the Chi-square test was used to detect the influence of all types of factors. The factors identified by univariate analysis (P < 0.10) were further entered into the multinomial logistic regression to calculate the odds ratios for significant factors.
We performed these analyses using the STATISTICA version 22.214.171.124 software (TIBCO Software Inc. CA, USA). Bonferroni adjustment (P < 0.013) was used because of the pairwise comparisons. Two-tailed P < 0.05 was considered statistically significant.
| Results|| |
In our patient population (n = 630), the mean age was 71.44 ± 8.52 years. The majority of the patients were female (58.1%) and had no education (38.3%). The baseline characteristics of the study sample are given in [Table 1].
|Table 1: Baseline characteristics of patients with Alzheimer's disease (n=630)|
Click here to view
Results of latent class cluster analysis
According to the information criteria of AIC Log-likelihood (LL), BIC (LL), and LL, there were four LCs of dementia severity defined as mild, moderate, severe, and very severe. These classes were named according to the summary statistics of the cognitive test scores. If AIC and BIC disagreed, the group number was assigned based on the BIC criterion because it is less likely to be affected by sample size. Entropy R2 for the four-class model was 0.8156. While deciding the LC number, we considered the confounders of AD dementia (each patient's CDR, age, and sex). Summary statistics of cognitive test scores are shown in [Table 2].
|Table 2: Summary statistics of cognitive test scores based on latent severity subgroups|
Click here to view
In support of our findings, all cognitive test scores were successful enough to distinguish the severity subgroups (P < 0.001, calculated using one-way analysis of variance for each cognitive test). In particular, CDT, MMSE, and BNT distinguished groups more successfully than WMT because WMT scores were low for mild groups, as well as the others. The detection of the mild group was very important because early detection of dementia provides future care and treatment. The estimation graph of the four-class LC model based on cognitive tests is given in [Figure 2].
|Figure 2: Each group of red bars represents conditional probabilities of groups. The manifest variables are the cognitive tests. Mild and moderate groups have higher scores. Mini Mental State Examination has high values in all classes, so only Mini Mental State Examination is insufficient for evaluating the severity|
Figure Caption: Estimation of the four class latent class model based on cognitive tests
Click here to view
Results of univariate analysis
We conducted the univariate analysis to detect the statistically significant relationships and differences between subgroups and influencing factors of AD dementia. The relationship between the clinical characteristics of participants and the latent severity groups are summarized in [Table 3]. The results showed that no significant relationships were observed in trauma, thyroid diseases, depression, loss of consciousness, coronary artery disease, hypertension, and stroke between the latent severity subgroups (P > 0.05 for all). However, a significant relationship was found for hyperlipidemia, sarcopenia, and diabetes mellitus between both groups (P = 0.001, P < 0.001, and P = 0.043, respectively). The percentages of hyperlipidemia and diabetes mellitus increased significantly along with the increase of AD severity. Sarcopenia was observed to be significantly higher in the severe group compared with the mild group (P < 0.001).
|Table 3: Univariate analysis of latent severity subgroups via influencing factors|
Click here to view
Among patients with higher levels of education, the severity of dementia decreased significantly (P < 0.001). In particular, the percentage of patients with no education was significantly higher than those with other education levels. A significant relationship between family history and severity of groups was also determined (P = 0.024). The percentage of family history was significantly higher in the mild group than the moderate group (P = 0.006). A family's recognition of dementia allows them to detect symptoms early and to present for treatment sooner. Accordingly, patients with no family history of dementia were diagnosed as having late-onset AD.
| Discussion|| |
In this study, we identified four latent cognitive AD classes: mild, moderate, severe, and very severe. The patients in the mild group remembered less than half of the words included in the WMT list, although the MMSE, BNT, and CDT results were good. In intensive outpatient clinics, these patients presenting with symptoms of forgetfulness are usually evaluated only with the MMSE because of its usefulness. However, this study shows that patients who were accepted as the mild group had near-total scores in the BNT, CDT, and MMSE, but low scores in the WMT, which evaluates short-term memory; this indicates that these patients should be examined more thoroughly. In many studies, neurocognitive screening tests compare patients with AD with the normal population and do not include patients with mild cognitive impairment. These tests allow physicians to recognize patients in the middle stages of AD. For example, in a report of the Quality Standards Subcommittee of the American Academy of Neurology, in 1367 patients with a clinical dementia rate (CDR) of 0.5, the MMSE was found to have a sensitivity of 49% with a specificity of 92%. Further, in a similar study that included 150 patients, MMSE with a sensitivity of 63% was found to have a cutoff value of 24. In our subgroup analysis, patients in the mild and moderate groups remained above the cutoff value in terms of the MMSE. When all the tests were evaluated together, they were observed in four groups, but if these patients had been evaluated only with the MMSE, the mild and moderate groups would have all been placed in the same nondementia group. Evaluation of patients using the MMSE alone seems to be risky in terms of delaying diagnosis and treatment.
Patients with AD have some comorbid diseases. Comorbidities have an accelerating effect on mental and physical disorders before the diagnosis and treatment of AD. At the same time, these comorbidities increase health expenditures, hospitalization, and mortality rates, together with care organization and cost. Therefore, the effect of comorbidities must be identified., In our study, we used a database that included all comorbidities and demographic factors. Our older population had two or more chronic diseases, such as hypertension, diabetes, depression, coronary artery disease, and others. We investigated the relationship between severity groups and comorbidities. According to the results, hyperlipidemia, diabetes mellitus, and sarcopenia had the worst effects on dementia. In support of the pathologic findings, high cholesterol levels in midlife may increase the risk of dementia. Furthermore, physicians overlook clinically significant dementia symptoms due to their workload. They recognize the influencing factors of AD and should be more careful after our study. The odds ratios of the influencing factors were calculated by multinomial logistic regression with subgroups as the dependent variable. The results are provided in [Table 4].
|Table 4: Multinomial logistic regression analysis with subgroups as a dependent variable|
Click here to view
Education and sarcopenia had the highest odds of having a relationship with AD severity. In particular, the confidence intervals for education were wide, and there was no calculation for the very severe group because of the insufficient sample size. The severity of disease among patients with no education predicted a 31.94-fold increase in comparison with graduates from university. The severity of the disease increased with a decrease in education level. Family history predicted a 1.56-fold increase in the risk of being in the moderate group versus the mild group. Moreover, diabetes mellitus predicted a 3.69-fold increase of being in the very severe group versus the mild group.
The strengths of our study are given respectively. First, direct and indirect effects of influencing factors on AD dementia have been more thoroughly investigated recently. However, previous studies regarding dementia prevalence or risk factors were not able to reveal this complex structure. The main reason is that many AD dementia studies conducted in the medical literature were planned prospectively. However, the number of individuals with dementia at the end of the follow-up period is quite low. The low prevalence of dementia also reduces the validity and generalizability of the statistical methods., In our study, all individuals were diagnosed as having AD dementia at the beginning of the study. Second, the absence of dementia severity has been stated as a limitation in some studies. In our study, the severity subgroups were determined using LC cluster analysis. Finally, the strongest aspect of our study is the fact that it is the largest computer-based empirical study conducted in Turkey that evaluated the demographic and comorbidity effects on latent severity subgroups. The first limitation of our study is the fact that longitudinal data of the patients with AD were absent. We were not able to follow the progression of the disease. Second, our data do not include socioeconomic and disease burden measurements of the patients.
| Conclusion|| |
The prevalence of dementia increases rapidly and dangerously. There is no single gold standard for the diagnosis of dementia, so it is very important to determine the main indicators that guide the dementia severity and identify the cofactors affecting the severity of disease. The LC cluster method is an effective calculation method for determining the severity groups of patients with AD together with their prognosis for obtaining a useful tool for optimum diagnosis for practitioners who have restricted time and instrumentation. We are planning ongoing studies in this regard.
Financial support and sponsorship
Conflicts of interest
There are no conflicts of interest.
| References|| |
Poblador-Plou B, Calderón-Larrañaga A, Marta-Moreno J, Hancco-Saavedra J, Sicras-Mainar A, Soljak M, et al
. Comorbidity of dementia: A cross-sectional study of primary care older patients. BMC Psychiatry 2014;14:1-8.
Schubert CC, Boustani M, Callahan CM, Perkins AJ, Carney CP, Fox C, et al
. Comorbidity profile of dementia patients in primary care: Are they sicker? JAGS 2006;54:104-9.
World Health Organization, World Report on Ageing and Health. Geneva, Switzerland: WHO Press; 2015.
Keskinoglu P, Giray H, Picakciefe M, Bilgic N, Ucku R. The prevalence and risk factors of dementia in the elderly population in a low socio-economic region of Izmir, Turkey. Arch Gerontol Geriatr 2006;43:93-100.
Wilkosz PA, Seltman HJ, Devlin B, Weamer EA, Lopez OL, DeKosky ST, et al
. Trajectories of cognitive decline in Alzheimer's disease. Int Psychogeriatr 2010;22:281-90.
Li CI, Li TC, Liu CS, Liao LN, Lin WY, Lin CH, et al
. Risk score prediction model for dementia in patients with type 2 diabetes. Eur J Neurol 2018;25:976-83.
McKhann GM, Knopman DS, Chertkow H, Hyman BT, Jack CR Jr., Kawas CH, et al
. The diagnosis of dementia due to Alzheimer's disease: Recommendations from the National Institute on Aging-Alzheimer's Association workgroups on diagnostic guidelines for Alzheimer's disease. Alzheimers Dement 2011;7:263-9.
Güngen C, Ertan T, Eker E, Yaşar R, Engin F. Reliability and validity of the standardized Mini Mental State Examination in the diagnosis of mild dementia in Turkish population. Turk Psikiyatri Derg 2002;13:273-81.
Can SS, Kizil TE, Varli M, Turan E, Atli T. Psychometric properties of the Turkish versions of three different clock drawing tests in patients with dementia. Noro Psikiyatr Ars 2010;47:91-5.
Emek Savaş DD, Yerlikaya D, Yener GG. Validity, reliability and Turkish norm values of the clock drawing test for two different scoring systems. Turk J Neurol 2018;24:143-52.
Kalem SA, Oktem O, Hanagasi HA, Emre M. Determination of the descriptive statistics for Alzheimer's disease Assessment Scale-Cognitive Subscore (ADAS-cog) in a normal adult Turkish sample. Noro Psikiyatr Ars 2003;40:13-25.
Kurt M, Can H, Karakas S. Research and development study for Boston naming test Turkish form. NYS 2016;54:6-14.
Sasidharan L, Wu KF, Menendez M. Exploring the application of latent class cluster analysis for investigating pedestrian crash injury severities in Switzerland. Accid Anal Prev 2015;85:219-28.
Magidson J, Vermunt JK. Latent class models. In: Kaplan D, editor. The Sage Handbook of Quantitative Methodology for the Social Sciences. 1st
ed.. Thousands Oakes: Sage; 2004. p. 175-98.
Vermunt JK, Magidson J. Latent class cluster analysis. In: Hagenaars J, McCutcheon A, editors. Applied Latent Class Analysis. Cambridge: Cambridge University Press; 2002. p. 89-106.
Goodman LA. Exploratory latent structure analysis using both identifiable and unidentifiable models. Biometrika 1974;61:215-31.
Vermunt JK, Magidson J. Technical Guide for Latent GOLD 5.1: Basic, Advanced, and Syntax. Belmont, MA: Statistical Innovations Inc.; 2016.
Lin SW, Tai WC. Latent class analysis of students' mathematics learning strategies and the relationship between learning strategy and mathematical literacy. UJER 2015;3:390-5.
Ganguli M, Belle S, Ratcliff G, Seaberg E, Huff FJ, von der Porten K, et al
. Sensitivity and specificity for dementia of population-based criteria for cognitive impairment: The MoVIES project. J Gerontol 1993;48:M152-61.
Kukull WA, Larson EB, Teri L, Bowen J, McCormick W, Pfanschmidt ML. The Mini-Mental State Examination score and the clinical diagnosis of dementia. J Clin Epidemiol 1994;47:1061-7.
Bähler C, Huber CA, Brüngger B, Reich O. Multimorbidity, health care utilization and costs in an elderly community-dwelling population: A claims data based observational study. BMC Health Serv Res 2015;15:1-12.
Bunn F, Burn AM, Goodman C, Rait G, Norton S, Robinson L, et al
. Comorbidity and dementia: A scoping review of the literature. BMC Med 2014;12:1-15.
Mielke MM, Zandi PP, Shao H, Waern M, Östling S, Guo X, et al
. The 32-year relationship between cholesterol and dementia from midlife to late life. Neurology 2010;75:1888-95.
Barnes DE, Covinsky KE, Whitmer RA, Kuller LH, Lopez OL, Yaffe K. Predicting risk of dementia in older adults: The late-life dementia risk index. Neurology 2009;73:173-9.
Sindi S, Calov E, Fokkens J, Ngandu T, Soininen H, Tuomilehto J, et al
. The CAIDE Dementia Risk Score App: The development of an evidence-based mobile application to predict the risk of dementia. Alzheimers Dement (Amst) 2015;1:328-33.
[Figure 1], [Figure 2]
[Table 1], [Table 2], [Table 3], [Table 4]