Spatial and temporal modeling for healthcare applications


Student thesis: Doctoral Thesis

View graph of relations


  • Yizhen HAI


Awarding Institution
  • Kwok Leung TSUI (Supervisor)
  • Wai Tat Peter TSE (Co-supervisor)
  • Michael Gerard PECHT (Co-supervisor)
Award date2 Oct 2015


Compared with the situation only a few years ago, the face of data analysis has been changed utterly. All kinds of intellectual devices keep records of many activities, but they are always biased, incomplete and unbalanced. Making good use of this type of data is the key of today’s “big data analysis”. We attempt and propose a preliminary method to deal with the observational data, which is the propensity score matching method. Propensity score was introduced as the conditional probability of assignment to a particular treatment verse control, given a vector of observed covariates. This procedure thus compensates the missing of randomization and helps to draw conclusions based on the available data. Meanwhile, as more wealth are accumulated and the global living standards are improving steadily, more and more attention has been attracted to public health and health care surveillance. The outbreaks of several infectious disease such as SARS, H1N1, H7N9, bird flu and Ebola Virus Disease have seized everyone’s nerves. Fast detection of increases in the incident rate of an adverse event is more-thanever desirable. Motivated by the successful applications in industry quality control, the technique of control chart has been introduced to the domain of health care surveillance from traditional statistical process control (SPC). Compared with SPC, healthcare surveillance is more sophisticated and involves geography, human behavior and other factors. Therefore, the outbreaks may be in varied forms, and happen simultaneously. Those traditional methods somehow simplify the problem and overlook the multiple-cluster scenarios. Our proposed adaptive order statistic plan and FDR-based method tried to fill this blank in two ways. The former tries to sum up all “significant” ones after standardizing the observed counts and comparing them with the quantiles of the Normal distribution. The latter applies the multiple tests method and makes “discoveries” while controlling the false discovery rate. They can detect the outbreak quickly and actually, as well as identify the exact outbreak area.

    Research areas

  • Public health surveillance, Statistical methods