/Spatiotemporal-epidemiology-and-analysis-of-mental-health-conditions-in-Thailand

Analyzed mental health service utilization in Thailand using spatial pattern analysis and cluster analysis.

Primary LanguageJupyter Notebook

Spatiotemporal epidemiology and analysis of mental health conditions in Thailand project ❤📊

This study aims to:

  • Advocate government policies that allocate resources to regions where people seek mental health services.
  • Help Thai people realize the importance of mental disorders.
  • Provide an overview of mental health in Thailand.

This study is part of the course CPE 301: Work Integrated Learning, in the Health Data Science program at the Princess Srisavangavadhana College of Medicine, Chulabhorn Royal Academy and the Department of Computer Engineering, King Mongkut's University of Technology Thonburi, during the third semester of the academic year 2021. It involves practical training with professionals in a company, hospital, or other health-related organization for a minimum of six weeks during the summer.

🙋‍♀️Group (เกือบจะอ๋อ)

  1. Papin Thanutchapat
  2. Chiraphat Phoncharoenwirote
  3. Ornrakorn Mekchaiporn

📚Advisor

  1. Asst. Prof. Dr. Chawarat Rotejanaprasert; Department of Tropical Hygiene, Faculty of Tropical Medicine, Mahidol University
  2. Asst. Prof. Dr. Peerut Chienwichai; Princess Srisavangavadhana College of Medicine, Chulabhorn Royal Academy
  3. Asst. Prof. Dr. Warut Aunjitsakul; Division of Psychiatry, Faculty of Medicine, Prince of Songkla University

🏢Internship placement

Mahidol-Oxford Tropical Medicine Research Unit (MORU) Faculty of Tropical Medicine Mahidol University in Bangkok, Thailand

Data source 📁

  1. Reported mental health service cases were collected from the health data center (HDC), Department of Mental Health, Ministry of Public Health (MOPH). All the cases were aggregated yearly mental health service cases notified during the years 2015 to 2021 at the provincial level, classified into eighteen categories in the health data center, and only eleven categories were used in the study (Dementia, Alcoholism, Drug addiction (excluding Amphetamine addiction), Schizophrenia, Depression, Anxiety disorder, Intellectual disabilities, Learning disabilities, Autistic disorder, Self-harm, and Epilepsy). ICD-10 mental disorders were assigned the codes F00.X - F99.X. Case report data includes the year and province of diagnosis.

  2. Thailand’s mid-year population statistics were collected from official statistics registration systems. This study was conducted in Thailand, which collected data on a population of approximately 70 million people across 77 provinces with different geographies.

  3. The geographic coordinates and provincial boundary data (shapefile) are obtained from the GEO package file in the Global Administrative Region Database (GADM), a high-resolution territorial database containing provincial to subdistrict data for all countries across the world

📍note that: The details of the Data Preparation can be found in the Code .

Methods ⚙

1. Descriptive Analytics

2. Cluster analysis

  1. Correlation Matrix

    • The Spearman's Correlation Coefficient is a nonparametric measure of a non-linear and monotonic relationship between two ranked variables

  2. Unsupervised Machine Learning

    • K-Prototype Clustering is a cluster analysis method derived from the k-means and k-modes algorithms. Therefore, the k-prototypes algorithm can cluster with mixed numerical and categorical values. K-means handle the numerical values, and k-modes handle the categorical values. These clustering algorithms are unsupervised learning, which is used with unlabeled data to find groups in the dataset. In this study, the algorithm is used to find groups among the disorders.

    • (Agglomerative) Hierarchical Clustering is a method of cluster analysis. There are two methods of hierarchical clustering, Agglomerative and divisive. In this study, we used the agglomerative hierarchical clustering method. In agglomerative hierarchical clustering, clusters are built from the bottom up.

3. Spatial pattern analysis

  1. Global Spatial Detection (Global Moran’s I) is used to determine whether the spatial correlation is positive or negative. It is shown as a clustered, fragmented, or random pattern

  2. Local Spatial Detection (Local Moran’s I) is used to evaluate the autocorrelation within local neighborhoods at the local level.

Results 📊

1. Descriptive Analytics

linechart
Figure 1 The patients accessing mental health services rate in Thailand since 2015-2021 (per 100k population)

There are 9,496,342 cases in the patient data for 11 mental health services collected from the Department of Mental Health's data system between 2015 and 2021. As shown in Figure 1, the total number of cases is calculated per 100,000 population and displayed in line chart format. The graph's data can be displayed in two periods: The number of mental health services provided prior to the COVID-19 outbreak until COVID-19 was detected (2015-2019). The number of patients accessingmental health services for Depression, Autism, Alcoholism, and Drug addiction (excluding Amphetamine addiction) increased steadily until 2018 and then decreased. The COVID-19 period's alcohol-control measures are expected to reduce the number of alcoholism cases. The second section displays the number of patients accessing mental health services following the detection of COVID-19 (2020–2021), found that the number of patients Depression, Alcoholism, and Drug addiction are expected to rise slightly, in 2019

barchart
Figure 2 Total of the number of patients accessing services in Thailand since 2015-2021 (per 100k population).

Figure 2 depicts the percentage of the number of patients accessing mental health services in Thailand calculated per 100,000 population from 2015 to 2021. The total number of patients is 9,496,342 and is shown in bar chart format. Anxiety disorders, Schizophrenia, and Depression were the three disorders with the most mental health patients, with 2,259,434, 2,016,200, and 1,773,618 patients, respectively. Intellectual disabilities, Learning disabilities, and Self-harm were the three disorders with the fewest mental health patients, with 110,469, 118,088, and 147,807 patients, respectively.

  • Mapping

map
Figure 3 The map Shows the location of mental health services rate at region, namely the provinces of Thailand, which has a total of 77 provinces.

Figure 3 depicts the study region's geographical location, namely the provinces of Thailand, which has a total of 77 provinces. The number of patients accessing mental health services for all 11 mental health services from 2015 to 2021 was: Alcoholism, Drug addiction (excluding Amphetamine addiction), Schizophrenia, Depression, Anxiety disorder, Autistic disorder, Intellectual disabilities, Learning disabilities, Dementia, Self-harm, and Epilepsy. A geographical heat map was used to present the spatial distribution of the number of patients over a seven-year period, allowing the mapping to highlight the geographic distribution of disease-identifying mental disorders using different colors. The darker the color displayed in a province, the greater the amount of traffic in that province. The GEO package file in the Global Administrative Region Database (GADM) Version 3.6 was used to create all the maps.

2. Cluster analysis

  • Correlation matrix

correlation
Figure 4 Total correlation of patients accessing mental health services in Thailand since 2015-2021 (per 100k population).

The correlation matrix shows 3 groups of disorders that have a positive linear relationship (correlate) with each other. The first group includes Alcoholism and Drug addiction (excluding Amphetamine addiction). The second group includes Schizophrenia, Depression and Anxiety disorder. The last group includes Intellectual Disabilities, Learning Disabilities, Autistic disorder and Self-harm. Dementia and Epilepsy are not highly correlated with the rest of the disorders.

  • Unsupervised Machine Learning
  1. K-Prototype

Below are the results of finding the optimal number of clusters using the Elbow method (Figure 5A) and the Silhouette coefficient (Figure 5B). In the elbow method, we can see a bend where the graph looks like an elbow at K = 2 and K = 3 in the left graph. In the right graph, it shows that the group of disorder is well-separated from other groups at K = 2 and K = 3 due to the higher silhouette coefficient than the other number of clusters (K). It indicates that 2 and 3 are the optimal number of clusters.

K-Prototype
Figure 5 The results of (A) finding the optimal number of clusters for K-prototype clustering using Elbow method and (B) Silhouette coefficient.

We applied K-Prototype to cluster the mental disorder. The algorithm clusters the disorders into groups based on the similarity of the number of patients accessing mental health services (per 100,000 population in every province 2015-2021) and the dissimilarity of the disorders’ names. When using K = 2, it divides 11 disorders into 2 groups. The first group includes Schizophrenia, Depression and Anxiety disorder, and another group includes the rest. The result of the K-Prototype using K = 3 as represented in Figure 6 is quite similar to the correlation matrix, which also clusters disorders into 3 groups. In the correlation matrix, Dementia and Epilepsy are not highly correlated with the rest of the disorders. For the K-prototype clustering result, Dementia was assigned to the same group of Intellectual disabilities, Learning disabilities, Autistic disorder and Self-harm. Epilepsy was assigned to the same group of Alcoholism and Drug addiction (excluding Amphetamine addiction). As we can see, the first group using K = 2 and K = 3 has the exact same disorders. That means Schizophrenia, Depression and Anxiety disorder are well-separated from other disorders and have a comparable number of patients accessing mental health services to each other

clusterreult
Figure 6 The result of K-Prototype.

  1. Agglomerative hierarchical clustering

We applied agglomerative hierarchical clustering and Ward's method to cluster the mental disorders. As a result, the dendrogram in Figure 7 shows two major branches, A and B, by using a line cut-point to obtain clusters. We found three clusters, represented in Figure 8. The first group includes Schizophrenia, Depression and Anxiety disorders. Subsequently, the second group includes Dementia, Intellectual disabilities, Learning disabilities, Autistic disorder and Self-harm. The third group includes Alcoholism, Drug addiction and Epilepsy.

hierarchical
Figure 7 The Dendrogram of mental disorder shows two major branches: A and B by using line cut-point to obtain three clusters.

clusterreult
Figure 8 The result of Agglomerative hierarchical clustering.

3. Spatial pattern analysis

  • Global Spatial Detection

tableglobal
Table 1 The value and p-value of Moran’s I hypothesis test of patients accessing mental health services in Thailand during 2015-2021 (per 100k population).

According to the description, Global Moran's I is a statistic used to measure variable spatial autocorrelation. The table below shows the value and p-value of Moran's I hypothesis test of mental health services accession in Thailand over a 7-year period from 2015 to 2021, calculated per 100,000 population. The GEO package file in the Global Administrative Region Database (GADM) Version 3.6 was used to create this analysis. According to Table 1's interpretation, there are 9 disorders with a p-value of 0.05 (significant), indicating that we can find the pattern of cluster or location in neighboring positions that tend to cluster but were unable to clearly identify the area.

However, there is a slight difference when looking at the yearly calculated p-value data. That is, the yearly p-value has an interesting cluster pattern and exceeds several years in only 8 disorders: Alcoholism, Drug addiction (excluding Amphetamine addiction), Schizophrenia, Depression, Anxiety disorder, Autistic disorder, Intellectual disabilities, and Learning disabilities. It is divided into three subgroups: addiction, mood disorder, and brain disorder. This is why the researchers decided to investigate the relationship by calculating the correlation coefficient between seven disorder pairs, as shown in Figure 10.

  • Local Spatial Detection

This map was created using the LISA cluster technique and analysis. It was used to assess the autocorrelation within local neighborhoods at the local level. Local measures can be used to identify spatial clusters and spatial outliers of each disorder. The interpretation of the defined color results is described in the method's section 2.2.3 Local Spatial Detection (Local Moran's I). The GEO package file in the Global Administrative Region Database (GADM) Version 3.6 was used to create all the maps.

Figure 9 shows a cluster pattern of detectable mental health services for each disorder in Thailand. For spatial interpretation, it was found that most clusters in the northern region had access to all 7 disorders, namely Depression, Anxiety disorder, Intellectual disabilities, Learning disabilities, Autistic disorder, Self-harm and Epilepsy. In the northeast region, there were clusters of the number of patients using mental health services for all 4 disorders, namely Alcoholism, Drug addiction, Depression and Schizophrenia. In the central region, there was a cluster of patients accessing mental health services for all 2 disorders, namely Dementia and Self harm. For the cluster, there were 2 clusters of patients accessing mental health services in the southern region: Anxiety disorder and Learning disabilities. In the eastern region, there is also a cluster of patients accessing mental health services for 2 disorders: Autistic disorder and Epilepsy. The last area, the western region, had a cluster of patients accessing mental health services only for learning disabilities. As shown in Figure 9,

mapLocal
Figure 9 Maps showing 11 clusters of disorders of the number of patients accessing mental health services in Thailand at the provincial level using Local Moran’s I statistic (LISA) used for a total of 7 years from 2015 to 2021.
Note: For interpreting clusters, High-High were labeled by red, High-Low were labeled by yellow, Low-High were labeled by sky blue, Low-Low were labeled by blue and Not significant were labeled by gray.

  • Disorder pairs

The correlation coefficient was calculated at a national level. The results are shown in Figure 4. In addition to arranging the disorders with similar correlation coefficients into the same cluster together. If determined using correlation data and pattern finding data calculated from global Moran's I analysis together, it is interesting to carry out a further study to find the correlation between pairs of disorders by calculating correlation detailed at the provincial level and using 7-year cumulative patients accessing mental health services data from 2015 to 2021. For interpretation, if a dark area is found in any province, it means these disorder pairs have a similar number of patients accessing mental health services that were continued for many years in such provinces. In other words, a relationship was found between these disorder pairs. Which found a total of 7 relationships with interest from 8 disorders as shown in Figure 10: Alcoholism and Drug addiction (excluding Amphetamine addiction) with a national correlation coefficient of 0.69. The second pair is Schizophrenia and Depression, with a correlation coefficient of 0.19. The third pair is Schizophrenia and Anxiety disorder, a correlation coefficient of 0.62. The fourth pair is Depression and Anxiety disorder with correlation coefficient of 0.32. The fifth pair is Autistic disorder and Intellectual disabilities with correlation coefficient of 0.67. The sixth pair is Autistic disorder and Learning disabilities with correlation coefficient of 0.37, and last disorder pairs is Intellectual disabilities and Learning disabilities, with a correlation coefficient of 0.39.

In terms of interpreting spatial outcomes for interrelated disorder pairs when calculating correlation at province level. It was found that most provinces in the northern region were related to Alcoholism and Drug addiction, Schizophrenia and Anxiety disorder. In addition, Autistic disorder, Intellectual disabilities and Learning disabilities that are the part of brain-related disorders were also found. In most provinces in the Northeast, there was a relationship between Alcoholism and Drug addiction, Schizophrenia and Anxiety disorder. The central and southern regions found the same pairs of disorders, namely, brain-related disorders (Autistic disorder, Intellectual disabilities and Learning disabilities). For the western ans eastern region, it was found a relationship of disorder pairs; Schizophrenia and Depression and the second pair is Depression and Anxiety disorder are shown in Figure 10.

mapdisorderpairs
Figure 10 Map of Spearman’s correlation of disorder pairs of the number of patients accessing mental health services in Thailand at the provincial level for a total of 7 years (2015-2021)

Discussion 🔍

In the discussion of spatial outcomes, it was found that the clusters of spatial autocorrelation patterns occurred in various regions. For example, the major clusters of spatial autocorrelation depression were discovered in several northern provinces where the Mind organization provided information on Seasonal Affective Disorder (SAD). SAD is a type of depression that occurs during specific seasons or times of year. Depression is characterized by a persistently low mood that interferes with daily activities. If you have SAD, you will experience depression during certain seasons or due to certain types of weather or temperature. It is available in both winter and summer (50). That is why the North has a large number of patient depressive services compared to other regions. The second disorder is drug addiction. According to an analysis of the current drug situation in 2017, most of the northern regions, especially the fringe areas, are conducive to the smuggling of drugs from neighboring countries that produce drugs, namely countries. Laos, Myanmar, and Cambodia (51). As a result, it may be the cause of the drug epidemic and a large number of patients seeking drug addiction services in the north. Then there was the cluster of spatial autocorrelation anxiety disorders, which were mostly found in the Northeast. It's a very hot climate. A description of cortisol: it's a hormone secreted by stress or anxiety. Cortisol levels are higher in the summer than in the winter (52). The northeastern region of Thailand has the highest average temperature. A rise in temperature can also cause agitation, palpitations, nausea, and fatigue, which is a common anxiety symptom. There are other disorders that were not discussed because there is limited information.

From all the methods we used (correlation matrix, K-prototype clustering, and agglomerative hierarchical clustering), there are three groups of disorders that cluster based on characteristics of the number of patients accessing mental health services in every province from 2015-2021. The first group includes Alcoholism, Drug addiction (excluding Amphetamine addiction), and Epilepsy. These disorders are highly correlated due to the fact that abusing alcohol has a potential risk of other substance use, such as marijuana, cocaine, and heroin (at least one) (53). For the association between Alcoholic and Epilepsy, the available evidence suggests that the prevalence of epilepsy among alcoholics is at least triple that in the general population, and that alcoholism may be more prevalent among epileptic patients than in the general population (54).

The second group includes Schizophrenia, Depression and Anxiety disorder. From the result of the correlation, it was found that Schizophrenia and Anxiety disorder are more highly correlated than Schizophrenia and Depression. In this research, the prevalence of anxiety disorders was studied. The incidence of anxiety disorders was 45.16% in schizophrenia compared to controls. The research findings show that anxiety disorders are higher in schizophrenia than in the general population. It is also said that symptoms of anxiety tend to precede symptoms of schizophrenia (55). Likewise, the Conrad and Chapman study found that anxiety disorders are one of the early causes of schizophrenia (56). On the other hand, anxiety disorders may occur after treatment for schizophrenia due to concerns in daily life. The stress of being judged because of admission for schizophrenia is not widely accepted by society. This is reason enough to explain the relationship between schizophrenia and anxiety disorders. There was also a relationship between schizophrenia and depression. which had lower correlation values compared to schizophrenia and anxiety disorders. Therefore, when diagnosed with this type of disorder, it is less likely to be identified as schizophrenia (57). But when diagnosed with schizophrenia, depression may also be present. Therefore, the correlation calculation showed that the correlation was low.

The last group includes Intellectual disabilities, Learning disabilities, Autistic disorder, Self-harm and Dementia. The correlation coefficient between Autistic disorder and Intellectual disabilities is higher than the correlation coefficient between Autistic disorder and Learning disabilities because Intellectual disabilities are the most common co-occurring disorder with Autistic disorder, and the greater severity of one of these two disorders appears to have effects on the other disorder (58). Dementia correlates with Autistic disorder and also correlates with Intellectual Disabilities. Vivanti et al. used US medical insurance records for 1.2 million individuals aged 30–64 years. The prevalence was highest among people with intellectual disability alone (7.10%), but was also considerably higher among people with ASD (4.04%) and people with ASD and intellectual disability (5.22%) than among the healthy population (0.97%) (59). For the association with Self-harm, the study of Blanchard et al. shows that Autistic disorder was associated with a substantially increased risk of self-injurious behaviors and suicidality, and people with Autistic disorder had 2.26-times higher odds of self-harm than those without ASD (60).

There are several limitations to this study. First, the data source was a mental health accessed services report and stigmatization of mental disorders may prevent patients from seeking treatment. Therefore, the number of patients accessing mental health services may not be the same as the actual number of patients that occur. Furthermore, the number and location of mental health services in each province may affect the difficulty of accessing the service and may explain why the number of patients accessing mental health services is low. Another limitation is that Bangkok Metropolis's mental health accessed services data was missing in 2017. Missing data can reduce the efficiency of statistical analysis (61). Moreover, many people may have more than one mental health diagnosis (62). All of the above limitations can cause bias in the data. As for the future work, we desired to reduce the limitations to the least.

Conclusions 📄

The results of this study provided information about the spatiotemporal characteristics of mental disorders in Thailand and also revealed the limitations of the data, which have an impact on the overall analysis and further applications. For more accurate and interpretable results, such as geographic heat maps and associations between two disorders, the data used for analysis should represent the real characteristics of each disorder, which can lead to change, such as mental health policy by the government, attitudes and public interaction, and access to prevention and treatments.

References 📖

  1. What is seasonal affective disorder (SAD)? [Internet]. [cited 2022 Jul 24]. Available from: https://www.mind.org.uk/information-support/types-of-mental-health-problems/seasonal-affective-disorder-sad/about-sad/

  2. Jandeang B. Analysis of Current Drug Situation Problem [Internet]. Journal of Research and Development Institute,Rajabhat Maha Sarakham University. 2017;4(2):37–56. Available from: https://so03.tci-thaijo.org/index.php/rdirmu/article/view/209737

  3. Geall L, Crockett M. You’re not imagining it: hot weather really can make anxiety worse – here’s how to deal with it 27 [Internet]. Stylist. The Stylist Group; 2022 [cited 2022 Jul 24]. Available from: https://www.stylist.co.uk/life/summer-anxiety-makes-worse-heat-wave-social-causes-symptoms-advice/219735

  4. Galbicsek C. Drinking And Drugs [Internet]. Alcohol Rehab Guide. [cited 2022 Jul 19]. Available from: https://www.alcoholrehabguide.org/alcohol/drinking-drugs/

  5. Chan AWK. Alcoholism and Epilepsy [Internet]. Vol. 26, Epilepsia. 1985. p. 323–33. Available from: http://dx.doi.org/10.1111/j.1528-1157.1985.tb05658.x

  6. Chaudhury S, Kiran C. Prevalence of comorbid anxiety disorders in schizophrenia [Internet]. Vol. 25, Industrial Psychiatry Journal. 2016. p. 35. Available from: http://dx.doi.org/10.4103/0972-6748.196045

  7. Grillo L. A Possible Link between Anxiety and Schizophrenia and a Possible Role of Anhedonia [Internet]. Vol. 2018, Schizophrenia Research and Treatment. 2018. p. 1–8. Available from: http://dx.doi.org/10.1155/2018/5917475

  8. Tandon R, Gaebel W, Barch DM, Bustillo J, Gur RE, Heckers S, et al. Definition and description of schizophrenia in the DSM-5. Schizophr Res [Internet]. 2013 Oct;150(1):3–10. Available from: http://dx.doi.org/10.1016/j.schres.2013.05.028

  9. Matson JL, Shoemaker M. Intellectual disability and its relationship to autism spectrum disorders [Internet]. Vol. 30, Research in Developmental Disabilities. 2009. p. 1107–14. Available from: http://dx.doi.org/10.1016/j.ridd.2009.06.003

  10. Vivanti G, Tao S, Lyall K, Robins DL, Shea LL. The prevalence and incidence of early-onset dementia among adults with autism spectrum disorder. Autism Res [Internet]. 2021 Oct;14(10):2189–99. Available from: http://dx.doi.org/10.1002/aur.2590

  11. Blanchard A, Chihuri S, DiGuiseppi CG, Li G. Risk of Self-harm in Children and Adults With Autism Spectrum Disorder: A Systematic Review and Meta-analysis. JAMA Netw Open [Internet]. 2021 Oct 1;4(10):e2130272. Available from: http://dx.doi.org/10.1001/jamanetworkopen.2021.30272

  12. Kang H. The prevention and handling of the missing data [Internet]. Vol. 64, Korean Journal of Anesthesiology. 2013. p. 402. Available from: http://dx.doi.org/10.4097/kjae.2013.64.5.402

  13. Ellis ME. The importance of a complete diagnosis: Managing multiple mental illnesses [Internet]. Constellation Behavioral Health. [cited 2022 Jul 18]. Available from: https://www.constellationbehavioralhealth.com/blog/the-importance-of-a-complete-diagnosis-managing-multiple-mental-illnesses/