Parfait Munezero, PhD student, Department of Statistics.
Parfait Munezero, PhD student, Department of Statistics.

The goal in the statistics area of survival analysis, is to analyze the time of observing a certain particular event. In medicine, the event could be death from a certain disease. Here, you want to analyze the time from when you get infected until you die.

With survival methods a doctor can, for instance, evaluate if a certain kind of treatment would increase the likelihood for a patient to survive.

"Will know what treatment to give"

- It is important for doctors to make decisions. If the doctor can predict the life expectancy of a patient, then the doctor can know what treatment to give, says Parfait Munezero. 

He is trying to analyze the effect of some external characteristic on the time, here called survival time, until an event that you are interested in happens. However, since the effects can change over time, Parfait Munezero is working with methodology that takes into consideration effects that change over time. Popular models used in this area usually don’t do that - and that is the starting point of his research.

Not true that effects remain the same

One illustrating example would be when you want to analyze the effect of gender on the time a person gets an academic degree, and you have a very long study period. Here, Parfait Munezero points out the fact that the gender effect of the 1950s is not the same as the gender effect of today.

- To assume that the effect of some characteristic is constant over time is not realistic. In some sense it is not true, he says.

In order to take that into consideration (the change in effect over time) Parfait Munezero researches into what is so called dynamic hazard models. The new thing is to apply an algorithm called Sequential Monte Carlo (SMC) on these models.

In what area is your research most applicable?

- Now it is most applicable in clinical researches such as cancer- and HIV-studies; most of these incurable diseases. But whenever time is involved it can be applied.

The models can be used for both estimation and forecasting. Parfait Munezero explains one possible kind of prediction you can make, when you know that the effect of a characteristic change in a certain way:

- You try to predict, if an individual has characteristic A, B, or C - is it most likely that he will survive one year, two years or three years?

Easy to apply in many areas

The methodology, based on Bayesian inference, can be applied in many areas where you deal with time as a dependent variable, not only medicine. You can use it to analyze the effect of some parameter on a bankruptcy. It can be applied in insurance companies, software engineering and telecommunications dealing with smartphones, Parfait Munezero explains. His goal is to see that this methodology is used outside of the academia.

- What is good is that this methodology is very easy to apply. You might not understand the theory behind, but if someone gives you an algorithm to follow, it is very easy for practitioners, says Parfait Munezero.

Parfait Munezero’s supervisor is Gebrenegus Ghilagaber, and Mattias Villani, Linköping University, is his second supervisor.