Navigating Sex-Specific Disease Dynamics in Incident Dementia

Abstract

Dementia is among the leading causes of cognitive and functional loss and disability in older adults. Past studies suggested sex differences in health conditions and progression of cognitive decline. Existing studies on the temporal trajectory of health conditions for patient characterization after dementia diagnosis are scarce and ambiguous. Thus, there’s limited and unclear research on how health conditions change over time after a dementia diagnosis. To this end, we aim to analyze the shift in medical conditions and examine sex-specific changes in patterns of chronic health conditions after dementia diagnosis. We centered our analysis on a 15-year window around the point of dementia diagnosis, encompassing the 5 years leading up to the diagnosis and the 10 years following it. We introduce (i) MedMet, a network metric to quantify the contribution of each medical condition, and (ii) growth and decay function for temporal trajectory analysis of medical conditions. Our experiments demonstrate that certain health conditions are more prevalent among females than males. Thus, our findings underscore the pressing need to examine differences between men and women, which could be important for healthcare utilization after a dementia diagnosis.

I. Background

In 2023, the Alzheimer’s Association reports that roughly 6.7 million Americans above 65 have Alzheimer’s, with about two-thirds of them being women. Despite some treatments that can manage or slow dementia symptoms, there is no cure for the disease. Current research in healthcare informatics concentrates on early detection and diagnosis using learning algorithms, studying risk factors and patterns, remote monitoring of patients, and interactive tools for assessment and support [1]. Artificial intelligence (AI) and machine learning are enhancing early disease detection by analyzing vast health data. However, dementia patients also experience changes in many other health conditions after their diagnoses, and understanding these changes is important for caring for the large number of patients with dementia.

Clinical informaticists have used electronic health records (EHRs) to identify risk factors associated with post-dementia diagnosis, such as kidney functionality [2], vascular indicators [3], and psychosocial isolation scores [4] over 10-year follow-up after dementia diagnosis. However, these studies have focused on single conditions or organ-specific disease. Surprisingly, current literature lacks studies investigating the most common health condition patterns after dementia diagnosis. Addressing this gap, our investigation delves into patterns of both chronic and acute conditions for 10-years after the initial dementia diagnosis. In addition, we provide data on sex-specific post-dementia-diagnosis health trajectories.

Contributions:

Our study is focused on two important questions: (i) What are the pressing health issues in the decade after a dementia diagnosis? (ii) Do the health issues and patterns differ in women and men?

Objective:

Given that dementia is non-curable, our analysis with this follow-up study provides sex-specific early warning signs and information about pressing health issues to those recently diagnosed with dementia. Our study shall support dementia patients by describing health concerns that the patient may have to face in 10-years and allow the health care provider to plan and suggest precautionary measures for such conditions.

II. Design and Method

In this study, we analyze participants’ medical visits over a 15-year period, ranging from five years before a dementia diagnosis to ten years after.

A. Study Population

We utilized Mayo Clinic Study of Aging (MCSA) [5] and Rochester Epidemiology Project (REP) [6] data for this study. The MCSA is a large, ongoing population-based study that aims to understand mild cognitive impairment (MCI) and its progression to dementia and Alzheimer’s disease among aging individuals. Since 2004, participants undergo 15-months periodic examinations that include neurological evaluations, neuropsychological testing, and other diagnostic procedures. The REP is a unique medical records linkage system which captures virtually the entire population of Olmsted County of the population of Olmsted County, Minnesota [7]. This project allows researchers to study various diseases and conditions over long periods, and its database contains health care information for residents spanning several decades. From the MCSA dataset, we included participants diagnosed with dementia and their respective diagnosis dates. We excluded participants with a dementia diagnosis prior to their involvement in the MCSA. Finally, 729 participants with a known dementia diagnosis date were identified, comprising 356 females and 373 males. Using the REP data, we procured International Classification of Diseases (ICD) codes for every participant visit from 5 years prior to 10 years after the dementia diagnosis date. Among 729, records were available for 665 (91%) of participants in REP data for a given time frame. Each ICD code was categorized in one of 35 categories defined by the Agency for Healthcare Research and Quality clinical classification software (CCS)¹. Table 1 summarizes the participant’s the sex and age distribution of the study participants. We categorized participants into two age groups: those aged <=80 years and those >80 years at the time of dementia diagnosis.

Table 1:

Sex and Age based distribution of participants.

Sex	Age	N (%)	median (IQR)
M	<=80	147 (22.16)	74.86 [60.69, 79.92]
	>80	191 (28.72)	83.31 [80.04, 91.65]
F	<=80	124 (18.65)	75.76 [63.84, 79.95]
	>80	203 (30.53)	84.16 [80.08, 91.82]

B. Data Collection

To understand the trajectory of health concerns for each participant, it’s essential to have a unique identifier (ID) for every participant, allowing for a longitudinal analysis of their health. Each record includes information about the health concerns discussed during their visits, categorized into one of the 35 CCS groups. Consider a dataset $D$ consisting of participants’ medical visits over a 15-year period, denoted as the range [−5, 10]; 5 years prior to dementia date to 10 years after dementia date. $D$ is a collection of individual records $d_i$, where $i$ denotes the record index. Each record $d_i$ is defined as a tuple containing four attributes:

$$d_i=\left(I{D}_i,Y_i,A_i,S_i,C_i\right)$$ (1)

Where $I{D}_i$ is a unique ID of a participant, $Y_i$ is the year of visit and $\in [-5,10],A_i$ is the age of a participant at $Y=0$ (the year when dementia was diagnosed), $S_i$ is the sex of a participant, and $C_i$ is the medical code of the health concern during visit.

C. Temporal Analysis

Time-varying analysis considers patterns that evolve over time to identify how patterns change from one period to the next. The dataset is divided into consecutive time periods, denoted as $T_1,T_2,T_3,\dots ,T_n$ where $n$ represents the total number of time periods. Within each time-period $T_i$, Let $F_i$, represents the set of frequent patterns in $T_i$. We further examine (i) patterns in conditions that occurred in the 5 years prior to dementia diagnosis, and (ii) sex-specific health concerns emerging after a new (incident) dementia diagnosis.

Frequency distribution:

Among 35 medical conditions, we obtain top $n$ of the most frequently occurring health concerns among males and females, separately, as $f_m$ and $f_f$, respectively. Next, we obtain the union of both the groups as the most pressing health issues during participants’ visits. To achieve the approximation of the range of the percentage of participants undergoing the most pressing health concerns, we plot the boxplots for each category among males and females, separately. Furthermore, standard Frequent Pattern Mining (FPM) techniques (‘apriori’ algorithm) are applied to discover frequent patterns.

D. Frequent Pattern Mining

Frequent Pattern Mining (FPM) identifies common patterns or relationships in medical records. It is crucial for understanding the intricate and interrelated medical conditions in healthcare, helping professionals pinpoint prevalent health issues that often coexist. One of its notable strengths is the reduction of data noise by eliminating rare or random occurrences. By establishing a minimum frequency threshold, FPM ensures that only significant and recurrent patterns are highlighted, making the findings more trustworthy. We employ FPM to detect the most observed patterns among medical conditions during participants’ visits, annually. The ‘apriori’ algorithm employs support and confidence measures. Consider a set of medical codes $C=\left\{C_1,C_2,\dots ,C_{35}\right\}$ consisting of 35 CCS categories. Let there be a subset of $C$, defined as $X$. Support $\left(\mathit{\text{Supp}}\right(X\left)\right)$ measures how frequently a pattern $X$ occurs in the dataset among the total number of records. Confidence measures the conditional probability of pattern Y occurring given that pattern X has already occurred.

$$\mathit{\text{Conf}}(X\to Y)=\frac{\mathit{\text{Supp }}(X\cup Y)}{\mathit{\text{Supp }}\left(X\right)}$$ (2)

Association rules are employed to extract meaningful connections/ties between medical conditions in healthcare data, answering important questions such as, “If health concern A is present, how likely is it that health concern B will also be present?” Understanding such associations is vital for healthcare professionals and researchers as it can provide insights into potential comorbidities or related health issues. An association rule is typically represented as $X\to Y$, where X and Y are subsets of concerning medical conditions during visits. The minimum support threshold, denoted as $m$, is set to a predefined value of 20%, depicting the minimum frequency or occurrence required for a pattern to be considered significant.

E. Network Analysis

Network graphs are a visual representation of the association rules. They help to uncover complex relationships and visualize the strength of associations between health concerns. Network graphs make it easier to interpret the results. Consider a network graph that is represented as $G=(V,E)$, where: $V$ represents the set of nodes (CCS categories), and $E$ represents the set of edges (association rules). Edges in the graph represent associations among medical conditions, and their weights (based on lift) indicate the strength of these associations through ‘lift’ metric, which quantifies the strength of an association rule. Lift is a measure of how much more likely two health concerns are to occur together compared to their individual probabilities. By incorporating ‘lift’ values as edge weights, network graphs highlight the relative importance of associations. Thus, ‘Lift’ metric is calculated as follows:

$$W_{XY}=\frac{\mathit{\text{Conf}}(X\to Y)}{\mathit{\text{Supp}}\left(Y\right)}$$ (3)

Here, $\mathit{\text{Conf}}(X\to Y)$ represents the confidence of the association rule between $X$ and $Y$, and $\mathit{\text{Supp}}\left(Y\right)$ is the support of category $Y$. Researchers can utilize network graphs to identify clusters of related health concerns and assess their importance within different demographic groups and time periods, aiding in data-driven healthcare decision-making. To this end, we calculate useful network metrics to observe the significance of each node (medical condition) in a graph. We highlight (i) degree centrality, (ii) clustering coefficient, and (iii) betweenness centrality, as three important metrics suggesting the importance in terms of the association of a given medical condition with other medical conditions.

According to network analysis a general assumption is that nodes with higher degrees (more connections) are more central or influential within the network, perhaps indicating that it has shared underlying causes or risk factors with many other medical conditions. Thus, degree centrality $\left(Deg\right(v\left)\right)$ quantifies the importance or prominence of a node $\left(v\right)$ (medical condition) in each network $\left(G\right)$ based on its connections with other nodes. In an undirected graph $G$, $\left(Deg\right(v\left)\right)$ is defined as the sum of the edges connecting $v$ to other modes. The clustering coefficient of a node is measure of the degree to which nodes adjacent to a given node are interconnected with each other. It represents the extent to which the neighbors (adjacent nodes) of a given node link to each other, forming a cluster or tight-knit group. For a given node v with degree k (number of connections or edges), the clustering coefficient $\left(CC\right(v\left)\right)$ is defined as:

$$CC\left(v\right)=\frac{2E}{k(k-1)}$$ (4)

where $E$ is the number of edges between the neighbors of node $v,k$ is the degree of node v, indicating how many connections it has, $k(k-1)/2$ is the maximum possible number of connections between the neighbors of node $v$. A high clustering coefficient for a specific medical condition within the network of associated medical conditions could imply the condition that sits within a closely related group of other medical conditions, possibly suggesting shared risk factors, symptoms, or treatment protocols. These clusters could indicate areas of comorbidity or interconnected health challenges, making them particularly important for comprehensive healthcare management and research. the betweenness centrality of a medical condition (node) measures how often this condition plays a central or intermediary role in the associations of other conditions. A high betweenness centrality would suggest that the medical condition often lies on the shortest path of associations between various other conditions, indicating its pivotal role in the overall network. For a given node $v$, the betweenness centrality $BC\left(v\right)$ is defined as:

$$BC\left(v\right)={\sum }_{s\ne v\ne t}\frac{{\sigma }_{st}\left(v\right)}{{\sigma }_{st}}$$ (5)

where the total number of shortest paths from node $s$ to node $t$ is represented as ${\sigma }_{st}$, The number of shortest paths from node $s$ to node $t$ that pass-through node $v$ is represented as ${\sigma }_{st}\left(v\right)$. Addressing or understanding this condition might have cascading effects or implications for several other conditions, given its central role in the associations network.

MedMet:

We collect the advantage of each metric by defining MedMet, a single metric for studying contribution of health conditions towards network of associated medical conditions for medical grounds, and average all three metrics (degree centrality, clustering coefficient, and betweenness centrality) to deduce goodness of each of these metrics. First, we normalize the scores by adjusting each measure so that they all fall between 0 and 1. We achieve this by subtracting the smallest value of the measure from each value and then dividing by the range of that measure (the difference between its largest and smallest value). For a given metric M,

$$M_{\mathit{\text{Norm}}}\left(v\right)=\frac{M\left(v\right)-\text{min }\left(M_G\right)}{\text{max}\left(M_G\right)-\text{min }\left(M_G\right)}$$ (6)

Once each measure is adjusted, we find the average of all three metrics for a given health conditions. MedMet gives us a single score that represents the combined importance of a health condition as:

$$\mathit{\text{MedMet}}=\frac{\mathit{\text{De}}{g}_{\mathit{\text{Norm}}}+C{C}_{\mathit{\text{Norm}}}+B{C}_{\mathit{\text{Norm}}}}{3}$$ (7)

where ${Deg}_{\mathit{\text{Norm}}}$ is the set of normalized scores of degree centrality, $C{C}_{\mathit{\text{Norm}}}$ is the set of normalized scores of clustering coefficient, and $B{C}_{\mathit{\text{Norm}}}$ is the set of normalized scores of betweenness centrality.

F. Temporal Analysis

To determine the evolving frequent patterns and associations among CCS categories, we construct a network for each year and the strength of ties (edges) depict the strength of association among different medical conditions (nodes). We examine the evolution of such ties over the time through Dumbbell Plot, heatmap analysis, and network dynamics.

Dumbbell Plot

Dumbbell Plot, also known as a Dot-Dash Plot, is a data visualization tool used to compare two distinct values for a set of categories. It effectively highlights the difference or change between two data points (male and female), suggesting disparities among males and females 10-years post dementia diagnosis. We keep the threshold for FPM as 20 and for association mining as 0.25 after grid-search mechanism and experts’ observations. To construct this plot, we introduce two functions, $K_f\left(C\right)$ and $K_m\left(C\right)$, which give the year values for females and males, respectively for each category $c$ in $C$.

$$K_f\left(C\right)=G_f(c,i)$$ (8)

$$K_m\left(C\right)=G_m(c,i)$$ (9)

where $G_f(c,i)$ and $G_m(c,i)$ are the medical conditions present in the Graph $G_f$: Graph constructed from association mining for visits by females and $G_m$: Graph constructed from association mining for visits by males, for a given year $i$, respectively.

Heatmap Analysis:

A heatmap displays the magnitude of a phenomenon as color in two dimensions. The variation in color may correspond to the variation in the values of a network metric, MedMet as observed in Equation 7.

Network Dynamics:

Network visualization allows for the visualization of complex relationships in a system and is often used to intuitively understand and analyze the structure, behavior, and dynamics of networks. Consider a set of association rules $\left(A_t\left(G\right)\right)$ for Graph $G_t\left(V_t,E_t\right)$ at time $t$. Correspondingly, we obtain a set of association rules $\left(A_{t+1}\left(G\right)\right)$ for Graph $G_{t+1}\left(V_{t+1},E_{t+1}\right)$ at time $t+1$. Now, the evolving graph from time $t$ to $t+1$ is given as

$$\Delta G=G_{t+1}-G_t$$ (10)

As the graph evolves, we observe changes in the association rules $\Delta A$ where we introduce growth (added association rules) and decay (deleted association rules) in a new constructed graph $G_{t+1}$ [8]. The growth mechanism $\left(G_{g,t}\right)$ for a graph evolving $\left(G_{t+1}\right)$ from the graph $\left(G_t\right)$ at time $t$ contains the addition of edges $\left(E_{g,t}\right)$ and nodes $\left(V_{g,t}\right)$. The growth of $G_t\left(V_t,E_t\right)$ at time $t$ is given as:

$$G_{g,t}=\left\{\Delta V_{g,t},\Delta E_{g,t}\right\}$$ (11)

where

$$\Delta V_{g,t}=\left\{V\mid V\in V_{g,t}\right\}$$ (12)

$$\Delta E_{g,t}=\left\{E\mid E\in E_{g,t}\right\}$$ (13)

The decay mechanism $\left(G_{d,t}\right)$ for a graph evolving $\left(G_{t+1}\right)$ from the graph $\left(G_t\right)$ at time $t$ contains the deletion of edges $\left(E_{d,t}\right)$ and nodes $\left(V_{d,t}\right)$ from Graph $G_t$. The decay of $G_t\left(V_t,E_t\right)$ at time $t$ is given as:

$$G_{d,t}=\left\{\Delta V_{d,t},\Delta E_{d,t}\right\}$$ (14)

where

$$\begin{array}{l}\Delta V_{d,t}=\left\{V\mid V\in V_{d,t}\right\}\end{array}$$ (15)

$$\begin{array}{l}\Delta E_{d,t}=\left\{E\mid E\in E_{d,t}\right\}\end{array}$$ (16)

Thus, we define evolving graphs as the function of growth $\left(G_g\right)$ and decay $\left(G_d\right)$ mechanism in Graph visualization.

$$G_{t+1}-G_t=G_{g,t}+G_{d,t}$$ (17)

$$\Delta G=G_{g,t}+G_{d,t}$$ (18)

We study the dynamics of evolving association among medical conditions through growth and decay mechanism in evolving networks of medical conditions after different time intervals.

III. Experimental Results

A. Sex-specific variation among number of participants

The total number of participants who visited hospital per year before incident dementia diagnosis remains more than 300 for both males and females. The average number of participants who visited hospital (i) in 15 years of duration centered around dementia diagnosis are 23 males and 22 females, (ii) during 5 years before dementia diagnosis are 67 males and 65 females, (iii) in 10 years following dementia diagnosis are 34 males and 33 females. There is a discernible decline in the number of visits by participants in the years following a dementia diagnosis, maybe due to mobility issues and/ or death (see Fig. 1). Although male participants tend to visit more frequently than female participants before a dementia diagnosis, the frequency of visits by females shows a significant increase post-dementia diagnosis. The scope of our study covers the changing patterns in the medical conditions that are linked to the complaints or health issues raised by participants when they visit hospitals.

Fig. 1:

Changes in the total number of participants visited for a span of 5 years before and 10 years after dementia diagnosis (x=0).

B. Boxplot

Over a 15-year timeline, we analyzed the most common health concerns through boxplot (refer to Fig. 2). We observe ten most frequently occurring medical conditions among males and females around the timeline of dementia diagnosis in current settings. Interestingly, there are a few medical conditions which have no outliers, and approximate an accurate range of the percentage of participants such as chronic respiratory diseases, and arthritis, joint and back issues. We further observe higher variation in percentage of participants among certain medical conditions for males as compared to females, including hyperlipidemia, chronic respiratory diseases, and arthritis, joint or back issues indicating reduced certainty in diagnostic prevalence. For the most medical conditions, the percentage of participants increase after dementia diagnosis, except a few, which shows non-increasing trends such as hyperlipidemia, arthritis, joint or back issues, eye issues and skin issues, and some specifically for males (heart issues). We further conduct Dumbbell plot analysis to compare the timeline of frequent patterns among male and female, in detail.

Fig. 2:

Frequent health concerns during hospital visits for a span of 5 years before and 10 years after dementia diagnosis.

C. Dumbbell Plot

We further examine the progression of the most pressing health concerns during participants’ visit in Fig. 3. The x-axis spans from 5 years before to 10 years after the diagnosis. Infectious diseases, hyperlipidemia, and heart issues are some of the most frequently occurring conditions across the timeline for both sexes, while issues like thyroid/parathyroid/endocrine disorder and blood disorders appear less frequently, supporting the boxplot analysis. It offers insights into the trends and differences in medical conditions between males and females over time. Noticeably, certain healthcare conditions that are more common among females than in males such as mental health disorders, osteoporosis, and fractures, however, while certain health concerns are more likely to be seen among males such as cerebrovascular, cancer, neuro-muscular disorders, and circulatory diseases. Other than likelihood, we observe recent changes in the pattern, within the window of 5 years before dementia is diagnosed. Among females, the most concerned health conditions in the vicinity of dementia diagnosis are mental health disorders, injuries, pain and sleep disorders, cerebrovascular diseases, and circulatory issues.

Fig. 3:

Timeline for sex-specific changes in the health conditions during hospital visits for a span of 5 years before and 10 years after dementia diagnosis. Males and females are represented by solid blue and red dotted lines, respectively and color intensity indicates its frequency over the years. A vertical dotted green line at x=0 marks the dementia diagnosis year.

D. Heatmap Analysis

Fig. 4 shows two heatmaps representing the “Pressing Medical Issues over Time” for both males and females, suggesting the need of in-depth analysis of sex-specific studies for temporal trajectory analysis. Each heatmap visualizes the prevalence or frequency of certain medical conditions over different years. These heatmaps offer a comprehensive view of how different medical conditions manifest and evolve over time for both sexes. The distribution of conditions varies between males and females, suggesting that some medical conditions may be more prevalent in one sex compared to the other. Infectious diseases and chronic respiratory diseases, continue to show high frequency over the years after dementia diagnosis, consistent with the previous observations with boxplot and dumbbell plot. Some medical conditions contribute to association with other health concerns more among males such as Skin issues, circulatory issues, arthritis/ joint or back issues and kidney/bladder/urinary disorders, especially during later stages.

Fig. 4:

Heatmap Analysis for sex-specific changes in the health conditions during their hospital visits for a span of 5 years before and 10 years after dementia diagnosis. Darker shades indicate higher values, while lighter shades represent lower values for MedMet metric. The empty spaces and ‘0’ values show the absence of a given medical conditions and isolated nodes in the graph of corresponding year, respectively.

Contrastingly, Mental health disorders, hypertension and heart issues also display higher values, particularly in the mid to later years, signifying a notable concern among the female population. Both sexes demonstrate a noticeable contribution of Pain and sleep disorders, hyperlipidemia, chronic respiratory diseases, Infectious diseases, though the values might differ slightly.

Overall, Osteoporosis and fractures, mental health disorders and hypertension seem more pressing health concerns in females compared to males. As we observe individual contribution of each medical condition among other health concerns during 15 years of participants’ visit through Heatmap analysis, we further delve into dynamics of evolving networks. As the starting point of our study and for clear visualizations, we consider timeline for four segments: (i) five years before dementia diagnosis, (ii) the year of diagnosis, (iii) five years after diagnosis, (iv) ten years after diagnosis. We examine the network dynamics two-fold: (i) strength among medical conditions: strength of prevailing ties with blue color (light suggesting weak ties and dark color suggesting strong ties) five years before diagnosis, (ii) growth and decay mechanism: the growth mechanism (solid green) and decay mechanism (dotted red) for each subsequent graph as shown in Fig. 5.

Fig. 5:

Network dynamics for medical conditions among male (a-d) and females (e-h) population. (Green: newly added association, and red: the deleted association in new graph) 1: infectious diseases, 2: cancer, 3: thyroid/parathyroid/endocrine disorder, 4: diabetes, 5: hyperlipidemia, 6: gout, 7: blood disorders, 8: neuromuscular disorders, 9: headaches, 10: dizziness, 11: heart issues, 12: hypertension, 13: cerebrovascular disease, 14: chronic respiratory diseases, 15: gastrointestinal tract issues, 16: liver disease, 17: pancreatic disorders, 18: kidney/bladder/urinary disorders, 19: male genitourologic, 20: non malignant breast disease, 21: female genitourologic or reproductive issues, 22: arthritis, joint, or back issues, 23: osteoporosis and fractures, 24: lupus and connective tissue disease, 25: other injuries (sprains, burns, poison), 26: mental health disorders, 27: substance use disorders, 28: delirium and dementia, 29: immune disorders, 30: circulatory issues, 31: eye issues, 32: ear issues, 33: pain and sleep disorders, 34: skin issues, 35: mouth or teeth issues

E. Network Dynamics

Fig. 5 shows many changes among ties of medical conditions in evolving temporal trajectory for males (a-d) and females (e-h). The first phase of five years before diagnosis includes heart issues, hypertension, hyperlipidemia, eye issues, arthritis, joint or back issues, and skin issues. In addition to these, females have increased association with osteoporosis and fractures. During the year of dementia, the number of medical conditions for men and women increases from 10 to 20 and from 9 to 18, respectively. The darker green color indicates stronger connections among medical conditions for men that for women. Five years after diagnosis, we observe consistency/growth $n\left(G_g\right)$ or decay $n\left(G_d\right)$ of medical conditions among males $\left(n\left(G_g\right)=15,n\left(G_d\right)=5\right)$ and females $\left(n\left(G_g\right)=\left.12,n\left(G_d\right)=6\right)\right.$. At this stage, we found different health conditions which have strong association among other medical conditions in men (cerebrovascular diseases) and women (lupus and connective tissue disease, and skin issues). Finally, ten years after diagnosis, we found strong associations among 11 and 15 medical conditions among men and women, respectively, with denser connections among females, suggesting increased complexity.

IV. Discussion and Conclusion

Our study used the ICD codes that were observed during the hospital visits of participants for 5 years before and 10 years after dementia diagnosis. In general, our observations with participants’ visits suggests increase in the number of visits by females as compared to males, post dementia diagnosis. As we observe the percentage of participants having the most frequent medical conditions, the range is different for men and women. For the temporal trajectory analyses, Dumbbell plot and heatmap plot show notable prevalence and contribution of certain medical conditions during 15-year timespan centered around dementia diagnosis. Interestingly, these conditions are different for men and women. Our most notable findings with network dynamics demonstrate the increase in complexity among medical conditions of females in comparison of males. We show the evolving trends of pressing medical concerns as shown in Fig 5, highlighting the need of fine-grained sex-specific analysis for temporal trajectory of cognitive decline.

The most frequent health concerns in the decade after dementia diagnosis are infectious diseases, heart issues, hyperlipidemia, hypertension, chronic respiratory diseases, gastrointestinal diseases, kidney/bladder/urinary disorders, arthritis as well as joint and back issues, eye issues, and skin issues as shown in Fig. 2. However, their prevalence differs among men and women at different timespan after dementia diagnosis.

We propose that the existing observations with clustering medical conditions and its association with cognitive decline [9] requires enhanced visualization towards timeline analysis. The prevalence of certain medical conditions that evolve within five years before dementia diagnosis might be an indication of associated risk-factors with dementia. Although some medical conditions are common among men and women, there exist differences among evolving medical conditions in five years before dementia diagnosis among men (neuromuscular disorders, osteoporosis, and fractures, male Genito-urologic, cancer) and women (thyroid, parathyroid, endocrine disorder, cerebrovascular diseases, and circulatory diseases) as observed in Fig. 3. As we explore the associations among the most frequent patterns, the diagnostic prevalence of certain medical conditions is more among men during early phase of decade after dementia diagnosis as compared to women, such as, heart issues, circulatory issues, and cerebrovascular issues. Interestingly, women follow men at later stages as shown in Fig. 4. We further observe higher associations of medical conditions with hypertension, and osteoporosis and fractures among females as compared to men, supporting existing studies [3].

Unlike the existing latent class analysis [10], neural network [11] and hidden Markov models [12] for longitudinal studies in healthcare informatics, our network dynamics model visualizes the growth and decay of network among medical conditions in various time windows, demonstrating its utility to show the changing medical condition patterns. Among male population, the network graph appears moderately connected, suggesting that there are some prominent medical conditions closely associated with others, such as heart issues and hypertension, indicating they might be central or crucial in this medical conditions’ network five years before diagnosis. The network has adopted a denser structure, which can imply the emergence or strengthening of co-morbidities or that certain conditions might lead to or be symptomatic of others. However, there exists many other additions and deletions of associations among other medical conditions during evolution. Interestingly, some initial strong ties among health concerns fades away with time such as association between arthritis, joint, or back issues and lupus and connective tissue disease. In the future, we plan to enhance our experiments on different set of participants including progression from normal to MCI, MCI reversion (MCI to normal) and MCI to dementia. Furthermore, it would be interesting to observe age-stratified control groups.

Limitations:

Our observations are based on the participants’ health concerns/diagnoses during their visit, and we expect break between corresponding years, even for the chronic conditions such as diabetes. The percentage of population presented in our experimental studies are based on the total number of participants visited in corresponding year. As the number of participants’ visit decreases at later stages, there might be some potential limitation to reflect actual disease presentation. In addition of these, we do not have sufficient information about the date of death for all the participants.