Ching To Chung

[Ching To Chung]

I am trying to do assignment 2 of constructing a decision tree. However, I don’t see any labels that I can predict on using the dataset. The original dataset contained a malignant/benign label. The dataset I download from the assignment page has no such label. A tree cannot be constructed without the prediction label since it is not an unsupervised learning algorithm.

[Ching To Chung]

Here are some common methods that are used for the prevention of HPV.

### 1. Vaccination

Vaccines like Gardasil and Cervarix protect against the most common high-risk HPV types such as HPV 16 and 18. In places like Hong Kong, there is an emerging HPV vaccination capaign to provide free doses to primary school children in their adolescence (around ages 11-12). In general, vaccines are considered very effective with a 92%+ protection rate against high risk types. Two doses have to be taken, and each dose is estimated to be around $8.5 USD (Hussain et al., 2020). Vaccination is already considered in the existing model as an intervention to reduce transmission, and a value of DALY averted per dollar could be calculated from the model.

### 2. Safe Sex

While condoms do not provide perfect protection, they are shown to reduce the risk of HPV transmission (Winer et al., 2006). Partners who frequently use condoms, compared to partners who nearly never use condoms, have a reduced hazard ratio of HPV transmission at 0.3. This can be simulated with a reduced beta value in the epidemiological model. Behavioural intervention to push the usage of condoms have been shown to demonstrate effectiveness empirically (Scott-Sheldon et al., 2011). However, a budget value of how much these kinds of campaigns cost could not be found.

### 3. Chastity or Abstinence

Chastity or abstinence is the simplest and most effective method of preventing STIs, and is a common method in conservative education. However, it is arguably difficult to change an individual’s behaviour to lower her amount of sex partners or to convince someone not to have sex. I am unable to find any papers that investigated whether chastity education are effectiveness. Moreover, even though limiting the amount of sexual partners could indeed reduce HPV transmission risk, the transmission risk is still relatively high even with only one sex partner at over 25% (Winer, 2008). This makes the effectiveness of chastity questionable. This can be simulated with reduced sexual partners in the epidemiological model.

Hussain, R., Bukhari, N. I., Ur Rehman, A., Hassali, M. A., & Babar, Z. U. D. (2020). Vaccine prices: a systematic review of literature. Vaccines, 8(4), 629.

Scott-Sheldon, L. A., Huedo-Medina, T. B., Warren, M. R., Johnson, B. T., & Carey, M. P. (2011). Efficacy of behavioral interventions to increase condom use and reduce sexually transmitted infections: a meta-analysis, 1991 to 2010. *JAIDS Journal of Acquired Immune Deficiency Syndromes*, *58*(5), 489-498.

Winer, R. L., Hughes, J. P., Feng, Q., O’Reilly, S., Kiviat, N. B., Holmes, K. K., & Koutsky, L. A. (2006). Condom use and the risk of genital human papillomavirus infection in young women. *New England Journal of Medicine*, *354*(25), 2645-2654.

Winer, R. L., Feng, Q., Hughes, J. P., O’Reilly, S., Kiviat, N. B., & Koutsky, L. A. (2008). Risk of female human papillomavirus acquisition associated with first male sex partner. The Journal of infectious diseases, 197(2), 279-282.

[Ching To Chung]

Modelling HPV

The primary mode of HPV transmission is through vaginal, anal, or oral sex with an infected person. Many individuals with HPV are asymptomatic and often unknowingly transmit the virus to others. There are many different types of HPVs, some doesn’t cause much symptoms, but there are high risk types which may lead to cervical cancer such as HPV 16. Infections usually stay for at most two years until it is cleared up by the body’s immunity. HPV Vaccines are available to protect against the most common high-risk types.

I found this paper which modelled HPV transmission in Finland. It is one of the few papers that I could understand because other papers add way too many parameters and processes trying to imitate the dynamic process of reality. They deviate too much from the scope of this course.

Barnabas, R. V., Laukkanen, P., Koskela, P., Kontula, O., Lehtinen, M., & Garnett, G. P. (2006). Epidemiology of HPV 16 and cervical cancer in Finland and the potential impact of vaccination: mathematical modelling analyses. *PLoS medicine*, *3*(5), e138.

This paper used models based on the original SIR concept. Its end goal is to evaluate the impact of vaccination. It used different models for different genders. For females, it used a more complex model that was built upon the SIR model, but adding many processes such as progression and regression, screening and treatment, and those who died from cervical cancer directly caused by HPV. For males, it used the simplest SIR model (in this case, “recovery” is the immunity from HPV). The model population was stratified according to age in 5 year cohorts and classes of sexual activity (frequent partner change or few). The variables used include the mean age of sexual debut, lifetime number of partners, vaccination effectiveness, etc. The paper varied vaccine coverage and vaccination age to show differences in outcomes. Its main findings are that HPV vaccination has the potential to significantly decrease HPV type-specific cervical cancer incidence. In addition, vaccinating women alone is sufficient for this significant reduction and vaccinating men also has little additional benefits.

Similarly, an SIR model could be built like the paper, which uses the SIR process for both males and females for simplicity.

The Finnish paper also stated some of the variable values that were used. For example, the mean number of sexual partners in a lifetime is 7.7 for women and mean age of sexual debut is 16.6. The vaccine efficacy is estimated to be 100%, although the 95% CI is also large and one study reported 92%. The study used values of real-life cervical cancer incidence, and varied the model by % explainable by HPV. The transmission probability (beta) uses a similar approach, where the value is varied until an outcome is derived to match real-life incidence, which was 0.6.

I was a bit lost on how to write the R code so I used some aid from generative AI. These are very rough and definitely needs further work.

Simple reiterative approach:

`jsx
# Define parameters
N <- 1000 # Total population
N_female <- N / 2 # Number of females
N_male <- N / 2 # Number of males

# Initial states
female_susceptible <- N_female
female_infected <- 0
female_immune <- 0
male_susceptible <- N_male
male_infected <- 0
male_immune <- 0

# Vaccination parameters
vaccination_rate <- 0.7 # Proportion vaccinated
vaccination_effectiveness <- 0.9 # Vaccine effectiveness

# Transmission parameters
transmission_rate <- 0.1 # Rate of transmission upon contact
morbidity_rate <- 0.2 # Rate of morbidity

# Time parameters
time_steps <- 100 # Number of time steps to simulate

# Initialize a data frame to store results
results <- data.frame(
time = integer(),
female_susceptible = integer(),
female_infected = integer(),
female_immune = integer(),
male_susceptible = integer(),
male_infected = integer(),
male_immune = integer()
)

# Simulation loop
for (t in 1:time_steps) {

# Calculate vaccinations
vaccinated_females <- female_susceptible * vaccination_rate * vaccination_effectiveness
vaccinated_males <- male_susceptible * vaccination_rate * vaccination_effectiveness

# Update susceptible counts
female_susceptible <- female_susceptible – vaccinated_females
male_susceptible <- male_susceptible – vaccinated_males

# Model transmission
new_infected_females <- (male_infected * transmission_rate * female_susceptible) / N
new_infected_males <- (female_infected * transmission_rate * male_susceptible) / N

# Update infected counts
female_infected <- female_infected + new_infected_females
male_infected <- male_infected + new_infected_males

# Update immune counts (based on morbidity)
new_immune_females <- female_infected * morbidity_rate
female_immune <- female_immune + new_immune_females
female_infected <- female_infected – new_immune_females

new_immune_males <- male_infected * morbidity_rate
male_immune <- male_immune + new_immune_males
male_infected <- male_infected – new_immune_males

# Store results
results <- rbind(results, data.frame(
time = t,
female_susceptible = female_susceptible,
female_infected = female_infected,
female_immune = female_immune,
male_susceptible = male_susceptible,
male_infected = male_infected,
male_immune = male_immune
))
}

# View results
print(results)
`

SIR differential equations approach:

`jsx
# Load necessary package
library(deSolve)

# Define parameters
N <- 1000 # Total population
N_female <- N / 2 # Number of females
N_male <- N / 2 # Number of males

# Initial states
initial_state <- c(S_f = N_female, I_f = 1, R_f = 0, # Female compartments
S_m = N_male, I_m = 1, R_m = 0) # Male compartments

# Vaccination and transmission parameters
vaccination_rate <- 0.7
vaccination_effectiveness <- 0.9
mean_partners_female <- 2
mean_partners_male <- 2
base_transmission_rate <- 0.1
gamma_f <- 0.1 # Recovery rate for females
gamma_m <- 0.1 # Recovery rate for males

# Effective transmission rates
beta_f <- base_transmission_rate * mean_partners_female
beta_m <- base_transmission_rate * mean_partners_male

# Define the model equations
hpv_model <- function(t, state, parameters) {
with(as.list(c(state, parameters)), {
# Rates of change
dS_f <- -beta_f * S_f * I_m / N
dI_f <- beta_f * S_f * I_m / N – gamma_f * I_f
dR_f <- gamma_f * I_f

dS_m <- -beta_m * S_m * I_f / N
dI_m <- beta_m * S_m * I_f / N – gamma_m * I_m
dR_m <- gamma_m * I_m

return(list(c(dS_f, dI_f, dR_f, dS_m, dI_m, dR_m)))
})
}

# Time parameters
time <- seq(0, 100, by = 1) # Time from 0 to 100

# Parameters list
parameters <- list(beta_f = beta_f, beta_m = beta_m,
gamma_f = gamma_f, gamma_m = gamma_m)

# Solve the differential equations
output <- ode(y = initial_state, times = time, func = hpv_model, parms = parameters)

# Convert output to data frame
output_df <- as.data.frame(output)

# Plot the results
library(ggplot2)
ggplot(output_df, aes(x = time)) +
geom_line(aes(y = S_f, color = “Susceptible Females”)) +
geom_line(aes(y = I_f, color = “Infected Females”)) +
geom_line(aes(y = R_f, color = “Immune Females”)) +
geom_line(aes(y = S_m, color = “Susceptible Males”)) +
geom_line(aes(y = I_m, color = “Infected Males”)) +
geom_line(aes(y = R_m, color = “Immune Males”)) +
labs(title = “HPV Transmission Model”, y = “Population”, color = “Compartment”) +
theme_minimal()
`

SIR differential equations approach with cancer compartment:

`jsx
library(deSolve)
library(ggplot2)

# Define parameters
N <- 1000 # Total population
N_female <- N / 2 # Number of females
N_male <- N / 2 # Number of males

# Initial states
initial_state <- c(S_f = N_female, I_f = 1, R_f = 0, C_f = 0, # Female compartments
S_m = N_male, I_m = 1, R_m = 0) # Male compartments

# Vaccination and transmission parameters
vaccination_rate <- 0.7
vaccination_effectiveness <- 0.9
mean_partners_female <- 2
mean_partners_male <- 2
base_transmission_rate <- 0.1
gamma_f <- 0.1 # Recovery rate for females
gamma_m <- 0.1 # Recovery rate for males
delta_f <- 0.05 # Rate of cervical cancer development for females

# Effective transmission rates
beta_f <- base_transmission_rate * mean_partners_female
beta_m <- base_transmission_rate * mean_partners_male

# Define the model equations
hpv_model <- function(t, state, parameters) {
with(as.list(c(state, parameters)), {
# Rates of change
dS_f <- -beta_f * S_f * I_m / N
dI_f <- beta_f * S_f * I_m / N – gamma_f * I_f – delta_f * I_f
dR_f <- gamma_f * I_f
dC_f <- delta_f * I_f

dS_m <- -beta_m * S_m * I_f / N
dI_m <- beta_m * S_m * I_f / N – gamma_m * I_m
dR_m <- gamma_m * I_m

return(list(c(dS_f, dI_f, dR_f, dC_f, dS_m, dI_m, dR_m)))
})
}

# Time parameters
time <- seq(0, 100, by = 1) # Time from 0 to 100

# Parameters list
parameters <- list(beta_f = beta_f, beta_m = beta_m,
gamma_f = gamma_f, gamma_m = gamma_m,
delta_f = delta_f)

# Solve the differential equations
output <- ode(y = initial_state, times = time, func = hpv_model, parms = parameters)

# Convert output to data frame
output_df <- as.data.frame(output)

# Plot the results
ggplot(output_df, aes(x = time)) +
geom_line(aes(y = S_f, color = “Susceptible Females”)) +
geom_line(aes(y = I_f, color = “Infected Females”)) +
geom_line(aes(y = R_f, color = “Immune Females”)) +
geom_line(aes(y = C_f, color = “Cervical Cancer Females”)) +
geom_line(aes(y = S_m, color = “Susceptible Males”)) +
geom_line(aes(y = I_m, color = “Infected Males”)) +
geom_line(aes(y = R_m, color = “Immune Males”)) +
labs(title = “HPV Transmission Model with Cervical Cancer”, y = “Population”, color = “Compartment”) +
theme_minimal()
`

[Ching To Chung]

I am interested in exploring the transmission of HPV (Human Papillomavirus) in Hong Kong. HPV is one of the most significant sexually transmitted infections and may act as an oncovirus potentially causing cancer. However, there has been little attention on the topic by the public health authorities in Hong Kong until recently. Optional free HPV vaccination program is recently implemented in Hong Kong primary schools. Some Hong Kong teenagers become aware of the risk of HPV, and pay for vaccination themselves since free HPV vaccination were not available back then. There is also increasing acceptance of male homosexuality, which may lead to more prevalent male-male HPV transmission. It would be interesting to explore how these factors affect the pattern of HPV transmission.

Here are some research questions:

What are the trends and dynamics of HPV transmission both currently and in the future? (May focus on certain strains of HPV)

How does the new optional HPV vaccination program in primary schools and the proportion of students who accept them affect the trend of HPV?

How does the out-of-pocket vaccination in Hong Kong teenagers affect the trend of HPV transmission? (This may be incorporated with economic modelling, more subsidies may act as an incentive increasing the demand for vaccination)

How does the prevalence of homosexual behaviour affect the transmission of HPV?

[Ching To Chung]

Question 1: How will you apply the results of this research article into different health care departments/settings or chronically ill patients, if this would be beneficial in your future practices e.g., Mental Health, Diabete, Oncology or others?

This paper shows the potential and importance of integrating ePROs in modern day healthcare. There is often little time when doctors could observe or talk with their patients, which makes capturing patient condition outside of the clinical setting very important. ePRO serves as a great tool for data collection and capturing symptoms when the patient is at home so that the doctor will have more data to improve on a better diagnosis and treatment. As a cancer patient myself, I am often bombarded by a lot of questions about how I have been feeling at home after I leave the hospital. I also experience a lot of different symptoms, psychological or physical, that I would like to record and report to my doctor. This data is valuable for both the patient as means of self management and for the doctor to make treatment decisions.

Question 2: What additional features or improvements would you suggest for ePRO systems to make them more effective and user-friendly for both patients and healthcare providers?

I think one important point is to make the app easy to use and navigate. Like writing a diary, users of ePRO may find it difficult to keep using ePRO systems since they may become unmotivated in a short amount of time. This is why it is important to make the UI and UX simple and attractive to use. This will not only act as incentive for users to keep using, but is important to cater to a large variety of patients, since they may be of different age and education level. An incentivising idea would be to use notifications to remind patients to use the system regularly, or reward them visually each time they use the system (for example, they may grow a tree in a virtual garden after recording their symptom).

[Ching To Chung]

1. How can integrating environmental data improve the accuracy and timeliness of malaria outbreak predictions compared to relying solely on epidemiological data?

From what I learnt in my undergraduate studies, I had a seminar where we discussed that certain infectious diseases have been exacerbated by influences of climate change under the principles of one health and planetary health. For example, the El Niño cycle is shown to be related to diseases transmitted by mosquitoes, including malaria. While epidemiological data could only show us the picture after the first patients have already contracted the disease, by aggregating environmental data and using spatial approaches, we could know which regions are at increased risks before an outbreak emerges. This allows us to better prevent outbreaks ahead of time.

2. What are the key benefits and challenges of involving public health stakeholders continuously throughout the development and implementation of a system like EPIDEMIA, and how can their feedback shape the system’s effectiveness?

The key benefits is that stakeholder involvement could help integrate and optimise the system. Public health stakeholders have an interest in all parts of the system, from input to output. We need stakeholders to help share their data, so that we have sufficient knowledge from the front line workers. They may interested in knowing how the system works and its accuracy to make the discretion of how they would utilise the system in their work. When learning of the output, stakeholders could also provide feedback to constantly improve the system. However, a main challenge is that most public health stakeholders are not tech savvy enough to know the system well. There may need to be increased efforts to simplify technical communication.

[Ching To Chung]

Question 1: In your experience, what are the biggest challenges to achieving sustainability in health information systems, and how can they be addressed?

I think the biggest problem is planning ahead, since government officials tend to be rather short-sighted and see functions of EHIS as individual one-off projects. This means that as opposed to spending more money to design a more comprehensive, scalable, flexible system, officials would rather prefer to minimise spending and only add functions when it is deemed absolutely necessary. This leaves the whole system very fragmented, as each function may be proprietary in its own. As a result, the Hong Kong public hospital app have something like 30 different icons for users to navigate (their attempt to “integrate” things into one system), each icon is a different system, and is very difficult for the average person to use. I think it is very difficult to solve this problem however, as this is not only a problem in the healthcare sector, but every sector! Officials want to spend the minimal amount of money and brag about cost savings, but the delivery is poor. When the problem underlies the culture of whole government, and you could not elect your own leaders and officials, there is very little that could be done.

Question 2: How have EHIS been designed to adapt to changing needs and technologies in your experience? If you haven’t encountered this, what features do you think are important for adaptability?

I think one aspect that is important but frequently neglected is to involve people in the EHIS planning process, since ultimately the users of EHIS are the people. In view of population aging, it is a nuisance if the elderly still rely on the old paper documents for their medical records, especially when they tend to frequent multiple facilities for their health conditions, there needs to be information exchange between facilities. However, to enable EMR, patient consent is needed and a form has to be signed. Many elderly has no idea what EMR is or how it could help them. I think a good effort that is being done in Hong Kong, is that EMR is being promoted to the elderly. They are educated on what the EMR is, they sign the form, and more people of age are joining the EHIS ecosystem. This effort helps ensure that the main users (elderly people who is becoming the majority of Hong Kong’s patients) are joining and using the system. A system is useless when you can’t convince its main users to use it!

[Ching To Chung]

Seminar 9
1. Please discuss how you think the perceived ease of use and usefulness may differ among the different demographics.
– Age
– Gender
– Education level

For age, I think the younger generation would find it easier to use new technology, while the older generation generally struggle with the skills of using smartphones and operating apps. The perceived usefulness may however be opposite. The younger generation may be accustomed to the various utility that technology brings and they may not have serious health conditions that require frequent e-health usage. Meanwhile, the elderly who have more chronic conditions and are more oblivious to the benefits of technology may find e-health systems to be more useful.

For education level, I think it is a similar trend with age. People with higher education level will definitely find the system easier to use since they are more knowledgeable about various vocabulary and the rationale of system design, while people less educated and hence have less exposure towards technology may find it more difficult to use. The perceived ease of use might be the opposite. Higher educated people may be more familiar and used to the utility of such information system, resulting in lower perceived usefulness, while people who are less educated and have no experience in this domain may perceive higher usefulness.

As for gender, I think it is more difficult to tell. In terms of psychology, it is known that males and females are similar in temperament. Males are more interested in things while females are more interested in people. Therefore, it depends on the design of the system. If the system is more complicated to use, involving more complex processes, this may interest males and be easier for them to use. If the system is centred around patient care and demonstrates more empathy, this may interest females and make them perceive the system to be more useful.

2. In your experience of using e-health applications or systems, what are some external factors or variables that should be considered to extend the proposed model for assessing the intention to use the system?

I think one thing that is important to consider is the trust towards the system depending on its origin or its developer. People are more likely to trust impartial, transparent, and authoritative figures, such as universities, as opposed to opaque organisations with their own self-interests, such as private companies. I am very reluctant to input my personal health data into a system developed by private companies because I don’t know how they will store them or whether they will secretly sell my data. An example in Hong Kong is that many people are generally reluctant to use e-health systems due to a fundamental distrust in the government. Nearly all e-health systems are government-initiated, and people are afraid that their data would be secretly monitored for political purposes besides medical usage. Therefore, I think it would be interesting to investigate the perceived trust of users by varying the source/developer of the system.

[Ching To Chung]

1. From the results, What would you recommend to Tak Hospital to improve the syphilis surveillance system?

A recurrent theme in the paper is the difficulty to operate the system when it comes to non-Thais. This is important as the majority of the Syphilis cases is attributable to non-Thais, and cases are likely to be underreported when it comes to migrants. As Thailand is becoming an increasingly attractive country to work and seek affordable medical treatment, this situation is only going to get worse. There needs to be more robust efforts in immigration control to ensure that all incoming migrants provide clean test results before being allowed to enter the country. Currently, the measures in place are not stringent enough, allowing many immigrants to enter by simply showing proof that they have performed an examination without requiring them to wait for the actual test results. This is alarming as they may be hidden Syphilis patients which could spread the disease in the community. There also needs to be better protocols in handling the medical affairs related to non-Thais. Currently, many workers don’t know how to handle non-Thais with the current system due to the lack of established protocols. Non-thais should be given a unique identification number, in the manner similar to Thai ID, when they enter the country or seek medical care for the first time. This would enable the migrants to have their own EMR in Thailand and allow hospitals and authorities across the country to track their records for different purposes.

2. Do you have experience with disease surveillance systems? What are the strengths and weaknesses of that system?

In Hong Kong, we only have disease surveillance systems for a selected list of communicable diseases. When doctors diagnose cases of one of these diseases, they will report to the centre of health protection which will keep monitoring the situation and keeping track of cases. This limited surveillance system for selected diseases need little resources to sustain, which makes it cost effective and protects most patients’ privacy since only patients with diseases that apply to the list need to have their data accessed. Even though it is inadequate, it has successfully limited and controlled outbreaks in the past, such as HFMD in schools. However, the lack of surveillance system for many other diseases have become a difficulty and nuisance in research. It is imaginable that if the government sets up a surveillance system compiling data from all hospitals covering more diseases including non-communicable ones such as cancer, it may reveal surprising insights and important health policy implications.

[Ching To Chung]

1. How can implementing artificial intelligence technologies in epidemic surveillance systems be enhanced to better detect and respond to disease outbreaks?

I think a great potential is the power of the LLM models nowadays that allow real-time monitoring of online unstructured textual data. Nowadays our epidemic preparedness is dependent on official reporting, but this may be a long time lag from the outbreak patient zero, or the authorities may fail to report the outbreak honestly, especially in authoritarian countries. The epidemic surveillance system provides a potential early warning, providing transparency for authorities to investigate.

2. What potential benefits do you see in utilizing AI for public preparedness, and what challenges might arise in implementing these technologies effectively?

The benefit is that making use of social media data is one of most responsive sources of information. As seen in many previous examples, early warnings of a potential outbreak is often reported by voices within a community. For example, the reports of COVID-19 were first reported by citizens in Chinese community and doctors of Wuhan hospital on Chinese social media. This makes surveillance systems potentially be able to catch epidemics very early. However, the challenge is of course that it is difficult to filter quality and reliable information from social media. People frequently posts rumours and outrageous, untrue things on their accounts. Social media are also filled with bots that artificially paste fake content. It may be difficult to distinguish the true from the fake, leading to a system with frequent false alarms.

[Ching To Chung]

1. What additional factors should be considered to identify barriers and unmet needs in health information seeking among youth for HIV/STI and RH than in the paper?

I think peer influences and the attitude of friends could be included in the paper. I saw some respondents saying that they took HIV test because their friends are also taking it. But the data only included whether they live with friends or family and whether their family accept them as sex workers. I think the influence of friends is more important and the study could look at the interactions between the participants and their friends.

2. Which types of vulnerable people in your community are missing or left behind in receiving necessary health information, and why? How can we best reach these individuals and measure the real impact of health information on their health-seeking behaviors to ensure its effectiveness?

I think in Hong Kong definitely teenagers with mental health problems have trouble getting the information and the help they need. In Hong Kong, we have had serious problems of student suicide due to academic pressure, but schools still only emphasise on grades and performance, not student well-being. It is difficult for teachers and social workers to identify which students are secretly have mental health problems because students are scared of reaching out and asking for help. Mental health in general also have bad stigma among the society and the older generation refuses to acknowledge the problem. If students tell their parents that they think they may have depression, they may be scolded by their parents as to “how can you have depression when I provide everything for you?” or “you’re so young how can you have depression?”. To reach out to these teenagers, I think a way is to utilise social media and provide services of anonymous counselling via text. Students may be reluctant to speak but nowadays people are comfortable texting. There could be qualitative evaluation and feedback on how the counselling have helped the teenagers.

[Ching To Chung]

1. How can the decision tree model be integrated into clinical practice to assist surgeons in preoperative planning and decision-making?

While the decision tree model may not be a big game-changer in clinical practice, I think it can offer some valuable cautions for surgeons because they may have underestimated certain risks happening in the surgery. Using myself as an example, I experienced anastomotic leak after my surgery because I have had extensive chemotherapy before which makes my healing very slow. This risk was underestimated because my previous treatment was not taken much into account. The decision tree model could incorporate individual patient differences from patient EMR data and conclude different types of operative risks. This allows surgeons to keep certain risks in mind when operating, potentially improving the surgical process.

2. What are the potential benefits and limitations of using this model in a real-world clinical setting?

I think one of the best benefits of decision tree is that it is explainable AI and is very intuitive and visible. The surgeon could see how the model arrived at its conclusion and the criteria for each decision point. This enables the surgeon to better make the decision on what and whether to trust, as opposed to deep learning models that are essentially “black box” and surgeons may find them less credible because they don’t know how the conclusions are arrived at. The limitations is that decision model may have a lower accuracy than more recent and powerful models. It also cannot process textual data such as charting of previous doctors, which limits the types of variables that it is able to take.

[Ching To Chung]

1. Why was the author interested in investigating the suicide problem in Thailand during the time?
The author was interested in the investigation because the number of suicides in Thailand has seen a general increase over the past decade, ranging from 3600 to 4000 cases every year. While in other developed countries, scholars have conducted a huge number of empirical studies to assess whether economic factors lead to higher suicide rates, the issue has been relatively under-studied in Thailand. Thailand also differs significantly from other countries in both economic and social contexts, and it has limited macro-level data available. To the author’s knowledge, only one previous study has used time-series exploration to explain the suicide rates in Thailand. The author aims to use cross-sectional analysis, which would be able to better explain the impact of cultural, social, and economic differences among regions.

2. Each of students picks one potential risk factor mentioned in the paper and explains how the variable can contribute to the suicide rate?
An interesting variable is the percentage of woman that are counted as the head of the family. The author explained why this variable was chosen in the beginning based on the observation that females are often considered inferior to their male counterparts in Asian societies. The example of Japan was given, where the hardship on families and suffering the abuse of their husbands may lead to depression of women and thus depression. The source of this variable’s data in Thailand was from National Statistics Bureau of Thailand. Most provinces had percentage of female as head of family at around 31-45%. Angthong ranked first and Samutsongkran and Singburi ranked Second and Third. After performing the regression, the variable was found to demonstrate statistical significance at the 0.01 level in both models. The coefficients were -0.396 and -0.401. This means there is a negative association. When women are listed as the head of household, they are less overwhelmed by males, and less suppressed emotionally and financially in the family. Therefore, in provinces with higher ratio of female as the head of household, the suicide rates were significantly lower.

3. How statistical modeling can contribute to investigate the epidemiology and spatial aspects of Thai suicide problem?
The author constructed two models using multiple linear regression analysis. Both model included a variety of social and economic factors, only differing in the age range of the alcohol consumption variable. For each variable, data at the provincial level of precision is used. All data is from 2011 which served the purpose of cross-sectional investigation. The benefits of this approach is that it is able to capture both the general trend and the spatial distribution of certain risk factors. For example, it was known that provinces in the northern region has the highest suicide rates compared to other regions, while Narathiwas has the lowest suicide rate. By using provincial data, we would not only be able to know which variables are significant at the national level, but also those that are especially serious at the provincial level. This would help capture a better picture at a higher precision, give insight to the geographic patterns of various risk factors, and provide implications on policy suggestions. For example, Phayao is not only a northern province with higher suicide rate, it is also a province with the highest adult alcohol consumption rate. Therefore, when work has to be done to lower alcohol consumption, the government may think to start there first, where both the suicide problem and the risk factor is serious.

[Ching To Chung]

What are possible reasons locations in epidemiological research have not been incorporated as much as other components in epidemiological research? How can spatial epidemiology be considered as an interdisciplinary science?

As mentioned in the paper, I think two of the major reasons is the availability of data and the appearance of easy-to-use GIS systems.

First, there is limited high-quality geographic data, which makes it difficult to conduct GIS analysis. I speak of this from personal experience since I have used GIS before, and when conducting analysis in rural areas or middle east countries, I often struggle to find shapefiles and a complete region code database. Many countries also lack data on disease and residence of individuals. For example, one of the papers mentioned that not all countries have a cancer registry, which becomes a barrier to conducting ecological studies. Moreover, we only started adopting electronic health records in the last few decades. Before that, most data is only available in paper form, meaning there is substantial difficulty in retrieving data and conducting statistical analysis using computers.

Second, GIS systems that are easy to use have only emerged relatively recently. I have tried using GIS and they are powerful tools for analysis. They have statistical packages integrated and could handle complex what-if questions with lots of conditionals. With the right data, I could get my answer in a few clicks. However, I imagine a few decades ago, GIS system may have been a lot more difficult to use. Statistical packages may not have been available, requiring researchers to write their own codes from scratch. The need for so much manual labour may have barred researchers from conducting spatial research, as compared to other quantitative and statistical analysis which could be done easily on spreadsheets like Excel.

Spatial Epidemiology could be considered as an integrated science since it incorporates the ideas and principles of lots of discipline. The fundamental usage of spatial epidemiology was rooted in public health when John Snow mapped cholera cases in London, finding links of disease with water. Spatial epidemiology uses statistical methods such as Bayesian analysis and Moran’s I for spatial autocorrelation. It uses the tools of GIS from geography, studying the relationship between people and the environment. It also uses the idea of environmental science and social science, since it considers pollution and socioeconomic status as risk factors and determinants of disease.

Explain why it is widely recognized that the place where an individual lives or works should be considered as a potential disease determinant and give some examples?

It is because environmental factors could be important risk factors to a disease. Example given in the papers were the high incidence of hepatoma in Asia found to be linked to hepatitis B, higher risks of cancer when working in the furniture industry, risk of cancers when living close to an overhead power line, et cetera. These risks are often unknown or undiscovered until the relevant investigation is done. These studies may not reveal a definite causal relation, but they answer important public health questions, such that the public will know whether they have increased risk for a certain disease when they are exposed to a specific thing. I could think of a few more examples from where I live. For example, miners and construction workers in China show a much higher prevalence of COPD and lung cancer. This is because they are constantly exposed to ash, dust, and asbestos in their work environment. In Hong Kong, a building had a serious outbreak of SARS in 2003. When investigating why the virus spread so quickly, researchers found that this is due to the building design, where virus particles could travel easily from floor to floor in the central air vent. These examples get to show that our working and living environment could drastically influence our health and disease.

[Ching To Chung]

1. I think for hospitals and medical facilities that have decided to incorporate medical AI systems into their diagnosis or treatment process, they need to set up a dedicated department to handle ethical issues and problems surrounding the use of AI. The department should be in charge of oversight of all AI uses, and regularly review its security and reliability. This would allow the prompt decision to limit or terminate AI use when things go wrong.

2. As a patient myself, I appreciate the use of AI for better diagnosis. However, I would really like more transparency. When you take a drug, you can look at the drug label to see who manufactured the drug, what are the side effects, etc. Similarly, if AI model is used in my diagnosis, I would want to know who made the model, what data it is trained on, what is the sample size, what algorithm is used, what is the accuracy metrics, etc. Proper communication and transparency are essential for patients to make an informed decision for themselves to accept the AI system or not. I would also like to be ensured that the AI only aids clinician in making decisions, that it is limited in its use and ultimately it is my clinicians that make the judgments for me. This would reduce the concern about its safety.

[Ching To Chung]

1. Hong Kong is a very urbanized city, so there is no decision support system that is specifically developed for rural areas. However, in China, this is a very common problem as there are millions of people living in rural areas, and there is great discrepancy in medical technology and personnel. In rural areas, they have clinicians known as “barefoot doctors”, who are not really trained doctor but health workers with clinical training with only the education level of high school graduates. To narrow the knowledge gap, electronic-based clinical decision support systems which are common in China’s hospitals have been introduced to these rural areas. The advantage is that this could be powerful technology to improve clinical outcomes. However, the undereducated barefoot doctors may not be well trained in using this new technology, or they may develop overdependence the system. See this article from Lancet.

2. Certainly. Nurses or community health workers, paired with NCD management tools, could provide service to provide coverage of medical services for large remote regions. Telemedicine, more stipends and budgeting from the government to improve primary care technology, allowance and better salary to attract clinicians to work in rural areas are all equally as important measures to improve rural medical standards.

[Ching To Chung]

Here’s my infographic for this week: https://snipboard.io/k1iqzR.jpg

[Ching To Chung]

My infographic this week:
https://snipboard.io/7ZVo6j.jpg

[Ching To Chung]

Here is my infographic!
https://snipboard.io/iIDZb6.jpg

[Ching To Chung]

Hello! This is my infographic. (I guess it’s more like a poster than an infographic, I am too used to making posters)

https://snipboard.io/ATwrBu.jpg

[Ching To Chung]

I would like to talk about point 6, which mentions that a P-value of >0.05 does not necessarily means that no effects are observed. I think this is a very common point or error that a lot of people profoundly believes. Many people treat the p-value as a magical threshold to the point that they think anything below it is a YES and anything above it must be a NO. It is important to note that there are many potential factors contributing to the p-value, so something above the threshold does not mean that the null hypothesis must be true. It just means that we cannot rule out the possibility of chance. There could still be an effect demonstrated in the data. It’s just that the probability that this may happen by chance is a bit too high for us to definitely conclude that an effect must have caused this.

Also, the p-value is set differently in different research objectives and different disciplines. For example, in astronomy, most research set the p-value threshold to 0.001 or 0.0001. In the astronomical scales of the universe, a lot of things could happen by chance. Therefore, the threshold for rejecting the null hypothesis is set very stringently to ensure that there is enough confidence to conclude an effect is more likely to cause the observation than it happening by chance.

[Ching To Chung]

1. I am Toby and I am not working now because of my sickness. However, I have done statistics work during my university years. I have mostly used excel, Python, and R.
2. Yes, I have learnt mostly data science, machine learing, and deep learning. However, I am a bit blurry on traditional statistics theory. I have applied statistics in a lot of of my work. I studied linguistics so I did quite a lot of statistical work on language data (word frequency, phonology), and I have also used statistics in my internship with United Nations to help compile economic data.

[Ching To Chung]

Sex: Male
Age: 21
Place of Residence: Hong Kong

If the above is in a study about DSRCT or Sarcoma, people will already know it’s me because it is so rare and I am the only one in Hong Kong that is 21 years old!

If it is a more general study, then education would also be needed to identify me.

Education:
Current Student of Mahidol Masters of Science in Biomedical and Health Informatics
Graduate of HKU Bachelors of Arts and Sciences in Global Health and Development

[Ching To Chung]

1. Focus Groups: I would use focus group discussions and gather all individuals who do not use bednets to collectively share their thoughts and experiences. Through group interactions, it may be able to discover the common reasons for not using bednets in the community, such as economic difficulties, or cultural norms.

2. Observation: By observing the behavior of the community and the actual way they live, it can provide valuable insights into the reasons for not using bednets. For example, some families may be already using other ways or homeopathic remedies to repel mosquitoes.

[Ching To Chung]

TAM would suggest that the new technology must be at least as useful as the old technology for a potential switch. The Technology Acceptance Model suggests that perceived usefulness is a key factor of individuals’ acceptance and adoption of a new technology. Individuals are more likely to accept and use a technology if they perceive it to be useful.

[Ching To Chung]

I think there are several external variables that might influence an individuals’ perceived ease of use or perceived usefulness of a new technology

Age and generational differences: People in different generations have experienced different paradigms of technology. This is an evident phenomenon nowadays. Younger people expects everything to be automated and they have a shorter attention span. This can be shown from the social networks that are popular amongst the youth, such as Tiktok and Snapchat. It shows that they don’t want the input output process to take long and they don’t like textual input. Meanwhile, the older generation still uses apps like Facebook and are more conventional. They have a longer attention span and don’t mind textual input.

Socioeconomic status: This is evident in the digital divide. Poorer families may not have had exposure to new technologies, and as a result have low technological literacy. Meanwhile, richer families have the resources to purchase new technologies, and have higher technological literacy. This may mean that richer families may find new technology easier to use. However, the converse may be true for perceived usefulness. Richer families have access to the latest technology, and therefore may be unimpressed when presented with innovation, while the same innovation may be a big improvement for the poor family with little exposure beforehand.

Compatibility and integration: It would also depend on how the technology is able to be integrated to the person’s life and other devices. For example, when I purchase a Google smartwatch, and I also own a Google phone, I find the whole system very convenient and easy to use as it is compatible. Everything syncs up beautifully and I have my health data on the Google ecosystem. However, if I purchase the smartwatch from another brand, my perceived ease of use and usefulness may drop. For example, if I purchase an Apple Watch, it may not sync up with my Google phone, and I would have to spend the effort to set things up. Therefore, compatibility and integration is an important factor as well.

[Ching To Chung]

Efficacy is the capacity for producing a desired result or effect in an ideal setting. Does the agent or intervention “work” under ideal, “laboratory” conditions? For example, a medicine could be ready to improve a patient’s symptoms in a perfect environment, where they’re closely monitored in order that they stick perfectly to the prescription. In this case, the drug has demonstrated efficacy.

Effectiveness is the extent to which an intervention produces the desired effect in real-world or everyday conditions. While an intervention may have high efficacy in a perfectly controlled setting in experiements, it may not work as well to the same extent within the real world to a broader population.

Efficiency is comparing a solution’s input to its output, taking into the account the economic aspect. For example, two drugs might be equally effective. However, if one of them is far more costly than the other one, that drug won’t be considered as efficient.

[Ching To Chung]

I think occupation is a confounder. Young people, whether they are students or workers, have to commute to school or their workplace every weekday, which guarantees a high frequency of contact. Meanwhile, for older people, many of them may have retired and do not have an occupation. This would mean they would tend to stay home on weekdays. This factor explains the difference in contact patterns outside of age.

[Ching To Chung]

Infant Mortality Rate is the proportion of infant deaths under one year old per 1000 live births.

It is calculated by (**the number of deaths in the first year of life)/(the number of live births)*1000**

It is useful for comparing the quality of health systems, especially among less developed countries. If there is high IFR, it may mean that the health system is too poor to prevent infant deaths.

[Ching To Chung]

Sorry for the late reply of this discussion. I had complications from my surgery and was incapacitated for a while.

I would like to speak from the experience of Hong Kong and how my university handled COVID.

Privacy and transparency: An app is enforced on all citizens to track their locations by adopting the policy of “Vaccine Passport”. The app stores the vaccine records of the user, and the user is expected to scan QR codes whenever they enter a public place, for example a restaurant or a library. If the user has already been vaccinated, it will beep blue and they will be allowed in the place. If not, it will beep red and staff will reject them from entering. The problem with such design is that the app is of extremely high risk, because it contains extremely sensitive data, including the name, the national ID, the health records, and the location of that person, which puts the app under a lot of opposition and outrage by the public. The government has not explained the encryption and data projection measures used to enforce this app, which has led to a lot of distrust. This is why it is important to reinforce data security whenever there are risks of extremely sensitive data. As a public app, it is also important to remain transparent about the measures being done to its users to ensure them that the usage is safe and important.

Compassion and consideration: The university has also adopted the policy of “Vaccine Passport”, all students must have received three shots of the COVID vaccine before they could enter the campus. To prove, students have to upload their vaccine records to the school website. This is a problem of privacy and freedom. Some students may have concerns regarding the second or third dose of the booster vaccine, but is forced by the school to take it. Some students are of foreign origin, and may find it difficult to get vaccines without Hong Kong ID. Students are also forced to reveal their vaccines records by the school, where it is not transparent or guaranteed about how such student’s data is handled. The school needs to consider the conditions and thoughts of all of its students, instead of implementing a one-for-all draconian arrangement.

[Ching To Chung]

Hong Kong as a developed city has been doing fairly well in terms of approaching Universal Health Coverage, but it certainly does not have a fully implemented UHC
system.
Hong Kong has a fairly mature healthcare system with public and hospitals covering the vicinity of pretty much every district, be it urban or rural. It follows a mixed healthcare model, with both public and private healthcare sectors. The public sector,
includes public hospitals and clinics, provides subsidized healthcare services to people with very affordable fees . For example, consultation fees of public clinics are as low as 8 USD. Life-saving surgeries that are deemed medically necessary are also subsidized and could be done at essentially zero cost to the patient and is paid
for by the government. This ensures that even those with lower incomes can access necessary medical care.
The strength of Hong Kong healthcare system is that such public healthcare is funded at no additional costs, unlike the system of many other countries that use some type of public insurance system which requires residents to pay a regular fee
to fund the healthcare system. As long as you are a Hong Kong resident, you can use its public healthcare system. Also, Hong Kong’s healthcare is very strong with two world-leading medical schools partnering with the public healthcare system. Therefore, certain public hospitals which also act as the teaching hospitals of these medical schools are known to provide very advanced and high quality healthcare.

However, there are also weaknesses in Hong Kong system which lead me to think
that it still has a lot of work on before fully achieving UHC. First, due to the dense and aging population, demand of healthcare services often exceeded the capacity. This has led to long wait times at hospital and clinics. Surgeries that are deemed not
immediately life threatening also have very long queue times. For example, you may need to wait for a few years to get your hip or bone surgery done at a public hospital. Therefore, some people have pointed out that “if you are about to die, Hong Kong has the best healthcare, but if you have a chronic disease and are dying slowly,
Hong Kong’s system may not help you much”. If the services exist but is not
accessible by people in time, then it is not really true UHC. Second, although on paper Hong Kong’s public hospitals cover every district, there are great differences in their service quality. The teaching hospitals in urban areas provide very good
healthcare, but in rural areas it is common for hospitals to offer poor services with inexperienced doctors. This leads to geographical inequities, and is not true UHC.

[Ching To Chung]

I would definitely be thinking of data sharing if I am in charge of the data, given that the participants know about the data sharing and agree, and that the desensitivization procedures are properly done. As mentioned in the lecture, data sharing contributes a lot to science in general and is usually beneficial to both parties.

For example, if I am doing research based on the dataset I collected, releasing the data may be beneficial to myself. This is because other researchers are able to audit and verify my findings from the dataset, which may improve the credibility of my research. If I make the data publicly available, it would also likely attract more attention to my research. Even if other researchers may not be interested in the main content of my research, they may be interested in using the dataset that I collected.

There is a saying that “data is the new oil”. Data is extremely important and valuable in this day of age, especially with the advent of AI and machine learning which requires lots of data for training. With these new technologies we may be able to infer new insights, and develop new models and uses that is beneficial to the society and scientific community. Though data is important, it is also costly to collect, which is why we should share the data we collected to maximize and pool our resources. By combining different datasets and sources, we could achieve synergy and enable more research while lowering barriers and costs. Therefore, data sharing is for the greater good and for the advancement of scientific findings and mankind.

[Ching To Chung]

In my opinion, the condition of the health informatics workforce in Hong Kong is terrible. Hong Kong is extremely slow in integrating IT into its health functions. The health informatics workforce is mainly concentrated in the Centre of Health Protection and the Department of Health, which are governmental bodies doing public health, epidemiology, and health economics work. There is a limited amount of data analysts, data scientices, and technicians working with health projects. There is also a small amount of IT personnel in public hospitals upholding their information systems. However, the healthcare landscape in Hong Kong is by no means adequate nor comprehensive.

A major challenge in developing a more comprehensive health informatics workforce is high turnover rate in the core health workforce in Hong Kong to begin with. Hong Kong has very bad doctor-to-patient and nurse-to-patient ratio, with extremely long waiting times in public hospitals due to a low amount of healthcare workers. Many doctors and nurses have either emigrated due to the worsening political landscape in Hong Kong, or have opted to work for private hospitals instead of public hospitals due to a much better salary and working environment. As a result, government funding has mostly been placed on hiring doctors and nurses back to the public hospital system to alleviate the crisis of supply-to-demand, instead of spending it on innovations, health informatics, and preventative work, which is deemed as less important. The excessive focus on clinical functions and service delivery has posed a challenge on resource allocation. Since the benefits of putting funding on health informatics may not be immediately obvious, and the lack of doctors and nurses in the core health workforce is also a critical urgency, it would be difficult to justify putting funding into health informatics instead of prioritizing a more urgent crisis at hand.

[Ching To Chung]

In Hong Kong, the wide-spread transition and usage of EMR only began in 2016 through a governmental intiative. The advantages of using EMR is immediately obvious after the city-wide adoption. When traditional paper records were used, it was extremely difficult to trace a patient’s medical history when the patient uses different medical facilities. For example, a patient may use different outpatient clinics, public hospitals, and private hospitals depending on his condition at the time. The communication and information transmission between hospitals are very difficult, leading to challenges in diagnosis and data collection from the patient. The patient is expected to bring his medical documents with him whenever he visits a new facility. After the city-wide implementation of a common EMR system, all medical facilities are able to see and edit a patient’s EMR through a shared network after the patient has provided consent. This offers a tremendous amount of convenience and benefits in diagnosis and treatment. Moreover, the usage of EMR increases standardization and may be able to reduce the amount of missing data and error. EMR has standard vocabulary and fields that the staff have to fill in, and could give alerts when there are numbers that are out of range and unfilled spaces.

The disadvantage of using EMR mostly comes from its obstacles. For example, there is subtantial cost of setting up the network which could be used by all medical facilities in Hong Kong, though cost is not a major concern with Hong Kong being a developed city. The challenge mainly comes with the dissemination of new technologies. Even though the system is available and provided by the government for free, many private providers and clinics are slow in adopting the system, as old doctors and nurses are used to the old ways of doing things and have been writing paper records for decades. Asking them to suddenly transition to use EMR system is very intimidating, and there were no training and guidance provided to educate staff on using the system. There are also patients who are reluctant to give consent to use the EMR system as they are concerned about the security and privacy of their records. The government would need to be more transparent about the security and privacy measures of the EMR system.

[Ching To Chung]

**Missing Data**

One of the ways that I think could solve missing data is to educate and train healthcare staff to chart and document things more completely. As mentioned in the paper, one of the reasons that there is missing data is that healthcare professionals don’t think documenting certain items in the EMR is useful. If we train staff to understand that their data input may also be used for research, they may be more incentivized to document things completely. Another key solution mentioned in the data is data imputation. I have used this method personally, and if the missing data is not too much, imputation could help solve the problem without changing the distribution too dramatically. However, there is of course limitations with this method as it has limited accuracy, and it does not work when there is too much missing data for the imputation to be based on.

**Selection Bias**

Sometimes because of limited budget and research limitations, certain data would have limited reach and demographic coverage. A similar problem is discussed in the United States. Some AI models, such as face recognition, are trained on large datasets that have bias. There may be a lot of data on the white and Asian community, but not enough data on the black community. As a result, the AI model is not representative enough and does not work well among black people. Similarly, in our studies and research, it is important to take account of the population with different types of characteristics to ensure the validity and representativeness of our conclusions.

**Data Analysis and Training**

There needs to be more training and education on biostatistics, especially in researchers.

The traditional way of teaching statistics is on math formula and statistics. This leads to too many researchers not knowing how to do things by application on the computer. Dissemination of knowledge should be focused on teaching the application of new and powerful statistical methods, not old statistical formulas and mathematical theories. Distance learning methods and MOOCs should be considered for flexible and convenient education.

**Interpretation and Translational Applicability of Results**

This is difficult to solve as it is deeply rooted in the academic culture. Researchers may be pressured into producing papers, even if they are self-explanatory without any translational uses, to maintain their research output. It is therefore important to have transparent reporting. Researchers and analysts should provide detailed descriptions of the analytical methods used, including any assumptions, manipulations, or transformations performed on the data.

**Privacy and Ethical Issue**

It is important to implement stringent data collection and management protocol. All means of accessing data should only be available to authorized personnel only. All methods of data access and uploading should be encrypted using industry standards such as AES-256. The data collection process should be under the consent of the patients affected.

[Ching To Chung]

I do agree with the four points proposed in the paper.

First is to convene key stakeholders in the health system, including policy makers, health professionals, and managers, to seek agreement on the scale and nature of corruption in each health system. While this may seem obvious, this is indeed very important. In the health system where everyone’s focus is service delivery and to provide clinical functions, it may not be easy to get everyone to notice and agree that there is an internal problem to the system. The first step to solving a problem is to recognize it, and therefore this step is a key first step.

Second, and once agreement has been reached on the problem, it will be necessary to prioritize action. It is important to put emphasis in implementing the solution to solve the most damaging problem, as the continued existence of the problem will hurt the integrity and the reputation of the health system. As the paper has mentioned, the solution needs to be devised to be one which is likely to succeed, and the solution needs to be pragmatic.

Third, as with many complex problems, it is essential to take a holistic view. As mentioned in the paper, corruption is a sensitive topic which may be met with avoidance. It is important to understand why corruption exists in the first place, and how it relates to the broader health system, especially taking into account the systemic and cultural context. It may be a lead to bigger and more profound problems.

Fourth, it is important that the research community sets out what it can offer. The paper mentions that knowledge could be drawn from different disciplines, such as criminology, psychology, etc., and at different scales: individual, organization, country. Research is important because it tells us how corruption emerges. After learning about the mechanism, effective interventions and policies could be devise to combat and prevent it.

An additional suggestion to combat corruption is to set up an individual organization to oversee corruption in the health system. In Hong Kong, we already have an organization that has been effective in combatting corruption known as the Independent Commission Against Corruption (ICAC). It is an independent organization that oversees all governmental departments, and punishes occurences such as the unfair appointment of government officials or irregularities in government spending and funding. However, it is simply assumed that health system is a specialized function, and the ICAC indeed imposes little to no action on the corruption of health system. As our health systems are increasingly intertwined with businesses and the private sector, it is important to make sure that there are no unfair practices that may hurt the interests of patients. It is therefore important to set up an organization against corruption to specifically handle corruption cases in the health system.

[Ching To Chung]

In the Hong Kong health system, a few selected public hospitals are integrated with university medical schools to offer a special expert clinic. Patients can pay an extra amount of money out-of-pocket to consult the professional teaching staff of Hong Kong’s top medical schools. This opens up a way for the teaching staff of the universities to do clinical work, offer their clinical expertise in the hospital, and earn extra income. It also helps patients to improve their quality of care, since for certain complex cases or rare diseases, regular doctors in the hospital without advanced training may not have the knowledge on treatment. As a result, this is an improvement in Hong Kong’s health system in terms of the building blocks of healthcare workfore and service delivery.

However, there are also barriers to its implementation. Since the usage of such services are costly, and many Hong Kong people do not have insurance and have to pay out-of-pocket, not many families are able to afford to use such services. This leads to the disputes of inequity, as only rich families are able to use this special clinic. Second, the provision of this special clinic also requires substantial funding from the government. This is because the public hospitals are essentially competing with resource-rich private hospitals to hire the university professors. The teaching staff are under no legal obligations to work only in public hospitals, and they may be attracted by the higher salary from private hospitals. To keep the teaching staff within the public hospital system and prevent high turnover rate, the government has to provide funding to offer competitive salaries and maintain this special clinic, which leads to challenges in budgeting.

[Ching To Chung]

In the graduating project of my undergraduate degree, I developed a deep learning NLP system to read company’s annual reports and predict their bankruptcy risks. It is a very big and complicated project and I had to do it on my own. It involves the 12 steps of

Step 1: Define the project clearly.

I defined the goals and objectives of my project to successfully build a classification system that is comparable to existing research. The type of data used for the project and the packages used for the model are also defined.

Step 2: List all tasks.

I broke down the massive project into smaller tasks that I need to do, such as data mining, writing the deep learning model, training, collection of results, analysis, writing up the paper.

Step 3: Get tasks into the right order.

Each steps are given a proper order. Some tasks could be overlapped to speed up the process (for example, analysis and the writing of the paper could be done simultaneously), while some are strictly linear where I could only complete one after another (for example, I could only train the model after I have completed data mining, or else there will be no data to train the model)

Step 4: Add a safety margin.

Some buffer time and leeway is added to the schedule because the project also coincided with my exam and assignment periods, so I may be too busy to work on the project from time to time.

Step 5: Crash the project plan (if necessary).

The original project ambitions are great, but after considerations with the workload at that time, I have decided to shrink the scope of the project to a smaller scale

Step 6: Create a Gantt chart.

A Gantt chart is created in excel for each of the tasks I have to do, so that I could follow my work progress and make sure everything is on schedule

Step 7: Look at resources.

I looked at the resources I have in terms of data source, my computational resource for training (graphics card capability), and time

Step 8: Think about what might go wrong.

I considered potential risks in my project like the failure of convergence in the training of my model, dirty data that requires more time for data cleaning than expected, etc.

Step 9: Monitor progress.

I mark down my model training outcomes and cross-check with my Gantt chart to see if I am on progress or if I am falling behind

Step 10: Monitor cost (time and resources).

Due to my graphics cost being too weak, I used cloud computing services from Google to help me train my model. Since they charge a monthly fee, I had to complete model training as soon as possible to minimize the budget used.

Step 11: Readjust plan.

I make sure everything is ready before training to problem arising afterwards and minimize training costs. For certain models that are too expensive to run or did not work as well as imagined, I omitted them from the project scope and used other models instead.

Step 12: Project review.
The work output is reviewed for its quality, and submitted to the professor after I am confident about the outcomes.

[Ching To Chung]

I think an important thing would be giving feelings of affirmation to potential leaders, because competent people often have doubts about themselves. When I was in my undergraduate studies and are doing projects with my classmates, I often see people in my team who are very knowledgeable about the topic and should be the one who lead the team. However, because they are too humble and don’t want to be in the spotlight, they often choose to stay off to the side and let someone else leads. Some very smart students I met have never taken the leader role in a project. It is therefore important to give encouraging remarks to them to let them know they have the ability, highlight their strength and explain to them how their competencies could be much more valuable in a leader role, and create opportunities to coerce them to play the leader role for once so they know they can do it!

[Ching To Chung]

I will talk about a disaster recovery plan for the information system of the University of Hong Kong, which includes research data, university documents, and members emails.

Risk Assessment:

A risk assessment is crucial to identify the potential threats and vulnerabilities to the information system. The assessment should evaluate both internal and external risks, such as natural disasters, cyber attacks, hardware failures, and human error. For the University, cyber attacks are the common occurence with the university system becoming a frequent target for DDoS and phishing in member’s emails. Weak points should be identified and addressed, as any vulnerability in the end point could mean catastrophic failure. There should also be RTOs and RPOs established as target metrics.

Data Replication:

Data replication involves creating and maintaining multiple copies of data in different locations or systems. The University is a buyer of Microsoft’s services, and imaginably uses Microsoft’s Azure Clouds for data replication. There are also servers on site, and duplicates of magnetic tapes off-site. By having copies of data in three locations, such geographically distributed data replication strategy can help ensure data availability and minimize downtime in the event of a disaster.

Data Recovery:

Data recovery focuses on restoring and recovering data in the event of a disaster. The university would need to establish regular back-ups or snapshots of the servers such that data could be recovered when things go wrong. The university has experienced server crashes and hardware failure in the past, so it is important to learn from such mistakes. Cloud-based solutions could help reduce the cost and simplify the process in making back-ups.

It is also important to involve relevant stakeholders, including IT personnel, staff and students, school management, in the development of the disaster recovery plan. Their knowledge and feedback will help ensure that the plan is suitable for HKU’s resources.

[Ching To Chung]

Hong Kong’s hospitals nowadays are extremely reliant on its information systems, especially for communication and data exchange between different healthcare facilities. Patients’ appointments, medication records and vital signs are all recorded and tracked on the hospital information system. Therefore, any unplanned downtime for the hospital information system could lead to disastrous consequences and potentially cripple the entire hospital services. Doctors and nurses would not be able to see a patient’s appointment, would not know what medication a patient has had, and would not be able to trace the patient’s vital signs the day before. Communication and data exchange with other facilities will have to revert back to phone lines and fax machines. Therefore, it is crucial to avoid unplanned downtime for the system and strive to maximize availability for the system. By using high availability technology, it could help ensure that hospitals can reduce and eliminate downtime when the servers fail or overload. This protects patients from being affected or harmed by system outage, and helps sustain the smooth operations of the hospital.

[Ching To Chung]

I am good at recalling details since I have quite good short term and long term memory. I don’t have to write things down and I could still remember the intricate details of what others have said, like date, time, and location. However, I would need to improve on my emapthizing skills. Sometimes when others talk to me they complain about something in their lives that seems rather minor and insignificant to me, and it is easy for me to get very impatient. It would almost appear to me like they are bragging about how good and smooth their lives have been going because they are bothered by such small things. Sometimes I will commit listening crimes by giving unsolicited advice, or telling them that many people have experienced rougher things than them and they should be grateful for what they already have. I would need to practice supressing my mental filter and try to listen to them more patiently.

[Ching To Chung]

I remember an incident that happened in my primary school/secondary school days that is a breach of confidentiality and caused a big mess

What happened?
It was the beginning of the school year and students are given their username and passwords to log into the school system for the first time. However, the username and passwords were short and shown in plain sight, so students could easily peek at each other’s passwords. What is worse is that while most websites give a randomly generated password for users to log in for the first time, some “IT genius” at our school decided that the password be something based off of the username. So for example, if your username is “1234”, your first-time login password would be two times that number then plus one, so 1234×2+1=2469. Because of this extremely insecure pattern, some students figured out the mechanism themselves and started brute-forcing into other students’ accounts.

How did it affect the system or users?
The school system account contains each student’s sensitive information such as phone number, home address, and student photo. Because it is very easy to enter someone else’s account, some students keep on accessing other students’ accounts from the same class. Luckily, because students only plan on playing tricks, laughing at other’s student photos and do not have bad intentions, the privacy leak does not have severe consequences. However, if someone with bad intentions accessed students account, it could have a severe consequences as the personal information of students could be leaked and sold. The school apologized the next day and quickly locked all student accounts before new passwords are assigned.

And how to prevent it?
First of all, obviously the first-time login password should be something complicated and randomly generated, such that no one can deduce other’s passwords. If the password is long enough and complicated enough, even if others walk by and saw someone else’s password slip, it would be very difficult for them to remember the password and try to maliciously login. It would be even better if the first-time login password is something that only the student would know (for example their national ID number) such that the password does not have to be printed on a slip. Also, students should be reminded to quickly change their password after logging in for the first time using the first-time login password.

[Ching To Chung]

This is a very interesting topic indeed! A very original and creative idea. I find diseases related to toxicology very intriguing. I live in Hong Kong where mushrooms are not a common sight, but occassionally I do hear news about hikers trying toxic mushrooms resulting severe consequences.
The system could potentially be very useful such that we know what are the common toxic mushrooms in certain areas, and the doctors would know what antidote to use. This is one of the cases where GIS would be extremely valuable, and the fruits of the surveillance system would also potentially benefit research in other fields such agriculture, ecology, biology, etc. Great presentation and very informative!

[Ching To Chung]

Strength: Self Management
I think I am one of the very few people who say I am good at this. I am very slow to anger and remain calm at unexpected situations since I am good at deriving solutions. For example, when I was travelling to Japan with my friends, we missed the last train. While my friends were panicking, I was able to keep calm and calculate another alternative train route where the train was still available. After I am diagnosed with cancer, my ability to keep calm is even stronger since I have been through a lot of crazy things. Sometimes I feel bored when listening to others talk about their lives’ problem because what makes them sad seem very insignificant to me.

Weakness: Relationship Management
One thing that I find difficult is managing my relationship with others. I am unfortunate enough to have met a number of very “interesting” people who have betrayed me and ruined a part of my life. While it is usually best to believe the best in people, it is very frequently not the case. Some people will exploit your kindness and use you. Some people are unreasonable and project their own negativity on you. It is very difficult to tell who is kind-hearted and who is not. To improve this, I think I should improve my skills in telling who has good intentions and who does not, and become less gullible to other’s people tricks and deception.

[Ching To Chung]

From the paper, I think integration with an electronic information management system would improve the two indicators of flexibiliy and representativeness.
Flexbility refers to the ability to change information needs or operating conditions with little additional resources. For the system, it is able to adapt and comply with WHO’s new case definition, and the surveillance system would also be able to allow investigation of other viral respiratory infections other than influenzas. However, the integration with an electronic information management system would likely make the system more flexible. The system is currently only sentinel and is only getting samples from a limited number of locations. By switcihing to the electronic system, the system would be much more scalable and the system could be expanded to a larger scale and include larger parts of the country. The electronic system would also be much more customizable, so the system could be changed to collect more relevant data, or help detect other diseases and pathogens. An electronic system would also be very convenient whenever there are changes to case definition. With statistical softwares to manipulate the data, it would be easy to filter out which patients match the new definition after inputting the new criteria.
Representativeness is about the coverage of the data that the system is able to capture, like the total number of patients, the greographical coverage. The system is only able to collect 14 sentinel sites in 8 regions, and it involves a lot of manual register at the sentinel locations which is slow and ineffective. By integrating with the electronic system, it would lower the resource requirements for more locations to join the surveillance effort, and more healthcare facilities would be able to join. This would likely increase the coverage of the data collected, and increase the amount of patients that are abled to be covered since the process is digitized and sped up, allowing us to serve and test more patients.

[Ching To Chung]

1. Verification & preparation:
Information technology could help the collection of data for outbreak investigation. By collecting data electronically, it could help identify outbreak and clusters more easily. We can more easily judge if there is an outbreak through electronical medical records (EMR). For example, we could plot all suspected cases spatially on the map to see if there could be suspected epidemiological linkage. If medical records are recorded on paper, it would take a lot of time to digitize the data through data entry for future analysis. IT also helps in the diagnosis verification. The Laboratory Information Management System helps track samples and record test results. If the expertise is limited, the lab technicians could use telepathology to the images to other experienced pathologists to establish a timely diagnosis. Information also help to prepare for a field trip through enhancing coordination and communication with different stakeholders.

2. Describe the outbreak:
After a case definition is established, IT could help to efficiently analyze and filter the existing data for the identification of cases. This could be done through advanced statistical software or data visualization tools. For example, we may construct a line listing of suspected cases. Statistical software would allow us to search, sort, and manipulate the data easily. We may have defined a time, place, and person to act as a clear case definition, which we could use as criteria to quickly filter all the cases using statistical software. Visualization tools could help us quickly plot an epidemic curve, so that we could understand the pattern and trend of the spread, and learn about the incubation period of the disease.

3. Hypothesis & testing:

Advances in bioinformatics and data science approaches applied in genetics and genomics may mean that we could analyze samples or identify risk factors more easily. The usage of statistics packages in IT also help us confirm or reject the null hypothesis, such as doing statistical tests, computing the p-value, correlation or risk ratio.

4. Response & action:
IT is essential for communication and the dissemination of information. The internet and social media could be used by the health agency to communicate the public health message, warning citizens of a new epidemic and what measures they could take for disease prevention. IT could also help continued surveillance. For example, under an epidemic, a dedicated system could be set up for nearby residents to report any new cases to disease control. This opens up more pathways for passive reporting. IT also helps real-time communication between health facilities in different parts of the country to share findings, ensure capacity for incoming patients, and coordinate medical supplies.

[Ching To Chung]

Among the technologies, I think I like Data-visualization tools for decision support the most. Data dashboard is becoming more and more widely used in many fields today, including public health surveillance. Real-time data is visualized to provide an intuitive and clear view of the current situation to keep the public informed. I remembered that John Hopkins University made a very popular dashboard to provide information on COVID. The dashboard collects data from various sources, such as national and local health agencies and other reliable sources to provide information on the amount of confirmed cases, deaths, and other relevant metrics. Once the data is collected, it is cleaned to remove duplicates and missing data, and data is merged from different sources into a single source. The processed data is then visualized on the dashboard using interactive maps, charts, and graphs. This allows the public to access a global view of the pandemic through using the dashboard. Users can also access historical data and track trends over time. I think it is important as use cases like this are incredibly beneficial as a dissemination tool for real-time disease surveillance. I have a bit of experience of making dashboards myself too, and it is very convenient and low-cost to make a dashboard nowadays with tools like powerBI and Tableau. Therefore, I think this tool has huge potential and a lot of use cases in the field of public health.

[Ching To Chung]

Should you give the data out?
I would only consider giving the data out after sufficient steps are done to ensure that none of my participant’s privacy would be violated and that all of my actions are in line with the local privacy laws. If this is not possible, I would refuse to give out the data.

How do you not violate any of the General Principles of Informatics Ethics?
Checking compliance: under what agreement did the participants provide us with their data? if they only agreed for their data to be used in our study, it would be both illegal and unethical to pass on their data to some other research team. if they did agree that their data could be used in any purposes beforehand, we still need to know what are the ethical codes of the foreign research team? are they credible and have stringent data protection principles? there are a lot of questions we need to ask to ensure that our actions are in line with the principles of ethics
Anonymizing the data: are all the identifiers really necessary to the team? why would they need every partipicant’s individual address? if this is not required, all personal identifiers of participants could be erased to ensure that there is no breach of privacy
Informed consent: make sure that all participants from the study know that their personal data would be relayed to a foreign research team, whose action are not under the control of our current team

If you want to provide the data to them, what and how will you do it?
First, I would ask about the advice from the project head, as I am only the one who manages the data but do not necessarily have authority to decide what to do with it on my own. I don’t own the data. When it comes to big decisions like this, making the decision alone instead of consulting with the local research team first is a big taboo. Then, if everyone in the team is okay with the decision, we would review the current data agreement to see whether the participants gave us consent beforehand for their data to be used elsewhere. If they did not, we would need to re-obtain their consent. After that, we would try our best to erase all the unnecessary identifiers to the participant’s data. Only after all of these processes, I would send the data to the foreign team. I would also need to maintain regular communication with them to ask them about the usage of this data and ensure that everything is in line with code.

[Ching To Chung]

Thank you so much! You’re a genius!

I have also just noticed that the code generates an incorrect map of the local Moran’s I. I am opening another thread for it. Do you have any ideas why?

[Author]

Posts