Covid-19 survival analysis - Hong Kong case study

Predicting whether a patient will recover from Covid-19 using data of individual patients.

Comment

Inspiration

We are often provided with high-level, aggregated statistics about Covid-19 cases, e.g., patients by gender, age group, city of residence. However, I believe there is so much more we can uncover by analyzing data on an individual-patient level. And this is what this project about.

What it does

The goal of this project is to analyze patient at an individual level and see how these factors interact and affect the outcome (i.e., deceased or discharged from hospital).

How we built it

I looked into two main data sources: One is the daily update released by the Hospital Authority in Hong Kong, each of which contains a list of patients discharged, in critical/serious condition during the day. Another is the background of the patient such as age, gender, source of infection. The sources are compiled and I developed a random tree classifier to predict patient outcome.

Challenges we ran into

The daily updates have to be obtained from separate URLs and there is no directory that contains all the list of URLs, so it was challenging to locate all the necessary data. Also, the format of the updates changed over time, so turning them into usable information was challenging.

Accomplishments that we're proud of

I managed to tally the number of days each patient is in critical or serious condition (if they ever fall into these conditions), and I haven't seen such data available from other sources before.

What we learned

A lot, examples are processing data into usable formats and creating plots to present the data.

What's next for Covid-19 survival analysis - Hong Kong case study

I did not explore all of the variables available in my dataset in this project. For future work, I can look deeper into these variables, e.g., the district the patient lives in, the nature of the case (e.g., imported or local) and see how they affect outcomes. I can also collect new public information about the patient such as which cluster they belonged to.

Built With

Share this project:

Updates