Evaluation of Learning-Based Screening Policies for Early Detection of Hepatocellular Carcinoma (HCC)
This project is a replication of the following paper: Lee, Elliot, et al. "Applying reinforcement learning techniques to detect hepatocellular carcinoma under limited screening capacity." Health care management science 18.3 (2015): 363-375. I replicate the simulation flow presented in the paper using a toy dataset that I created. All code is written in R, and a presentation report is available under the presentation folder.
Cancer, when detected early, increase the chance of survivals significantly. 50% of patients diagnosed with early-stage cancer survive beyond 5 years, as opposed to 10% of patients diagnosed with late-stage cancer. Hepatocellular Carcinoma (HCC) is commonly known as liver cancer, its risk has been increasing over the past decades. In Canada, HCC frequency tripled in men and doubles in women since 1970.
The standard screening process consists of: (1) Ultrasound image of patient liver; (2) Blood-test measure for alphafeto-protein level (AFP). This standard (type of) screening does not consider the changes in blood tests done at each screening visit. However, it has been shown that AFP past patterns significantly improve estimation of patient's risk of developing HCC.
Screening is costly, and the number of patients at risk is often greater than the number of screening resources available.
To overcome the above challenges, Lee et al. [1] proposes a framework to test different strategies for selecting individuals to be screened for HCC in an optimal manner, by assigning a probability of cancer risk based on each individual's demographic information and blood-tests, while taking into account the AFP blood-test fluctuations.
- n: List of n patients
- k: number of k patients to subset (i.e. from those n patients, how many individuals do we want to select for screening?)
- T: iteration time
- E: number of early stage cancers detected
- L: number of late stage cancer detected
- X: number of screenings spent on patients who would eventually develop HCC
- C: set of patients who will develop cancer
- N: set of patients who will not develop cancer
- D: Departing Patients (Patients are eliminated and replaced)
- E/(E + L): Proportion of cancers detected in early stage
- X/(k * T): Proportion of resources spent on patients who will eventually develop cancer
[1] Lee, Elliot, et al. "Applying reinforcement learning techniques to detect hepatocellular carcinoma under limited screening capacity." Health care management science 18.3 (2015): 363-375.
[2] Lee, Elliot, et al. "Improving screening for hepatocellular carcinoma by incorporating data on levels of α-fetoprotein, over time." Clinical Gastroenterology and Hepatology 11.4 (2013): 437-440.
[3] O'Connor, S., et al. "Hepatocellular carcinoma-United States, 2001- 2006." Morbidity and Mortality Weekly Report 59.17 (2010): 517-520.
[4] Okada, S., et al. "Follow-up examination schedule of postoperative HCC patients based on tumor volume doubling time." Hepato-gastroenterology 40.4 (1993): 311- 315.
It appears that in Ontario, cervical cancer is predominant in women who experience imprisonement, and a lack of cancer screening is contributing to this phenomena as shown in the following paper:
Kouyoumdjian, Fiona G., et al. "Cervical Cancer Screening Access for Women Who Experience Imprisonment in Ontario, Canada." JAMA network open 1.8 (2018): e185637-e185637.
Implementing a similar type of strategy to assess risk of cancer within imprisoned women in Ontario, and choosing the most at risk subset of women for screening could be a potential solution to the scarce screening resources available.