• Tags
ABC
Background
Basic
Benjamen Simon
Clinical Trial
Epidemics
Ethics
Experiment Design
Exploratory Data Analysis
GLM
Introduction
Logistic Regression
Machine Learning
Masters
PhD
Portfolio
Python
R
Regression
SAS
Series
Simulation
SIR
Spatial
Supervised
Synthetic Data
Talk

Optimal pricing strategy for a Ride Hailing Service

I was tasked with proposing the highest profit yielding pricing strategy for a new Ride Hailing Service based on data from similar services in the area. I proposed 3 distinct pricing strategies and used simulation methods to explore the optimal parameters for each strategy, and then comparing the yield of each, identified the optimal strategy. I used a logistic regression model for the data to estimate the probability that a given pay rate for a ride request would be accepted.

January 14, 2023 Read

Approximate Bayesian Computation for Epidemics

Approximate Bayesian Computation (ABC) is a likelihood-free (sometimes called simulation) method of inference. In this project I present a series of ABC methods which can be used to make approximate inference for epidemic models, and demonstrate their effectiveness on a series of simple case study outbreak datasets. Following this, I apply a selection of the methods to a complex spatio-temporal outbreak dataset for Sugarcane Yellow Leaf Virus. Finally, I discuss some questions that have arisen throughout the project on the use of ABC, and give my judgement on its position as an alternative to MCMC for making inference on epidemic models.

September 7, 2018 Read

A Spatial Analysis of Cases of Non-Specific Gastrointestinal Infections in Southampton

I wished to investigate the relationship between the levels of deprivation in lower super output areas (LSOAs) in Southampton and the number of cases of non-specific gastrointestinal infections. To do this I utilised spatial models to make inferences about risk factors and areas of increased risk. I begin with exploratory analysis, followed by K-function analysis, intensity estimation, and fitting Poisson generalised additive models (GAMs), and finally I fit both non-spatial and spatial extra Poisson variation models.

March 25, 2018 Read

Investigating the Effect of Body Mass Index on the Risk of Pre-Eclampsia

Given a data set from a case-control study of 750 mothers, I wished to investigate the relationship between Body Mass Index (BMI) and pre-eclampsia. To do this I took known information about the variables associated with BMI and/or pre-eclampsia, and created a directed a-cyclic graph (DAG) to aid in building a logistic regression model. I also employed simulations to help find the model with the least bias. My findings suggest that higher BMIs increase the risk of contracting preeclampsia, however, there is a great deal of uncertainty in my estimates of the odds-ratios.

February 12, 2018 Read

Predicting readmission for high-risk infants within 1 year of initial discharge A generalised linear model

I utilised data on 1488 high-risk infants collected for a multi-centre study to create a generalised linear model to predict the probability that a high-risk infant will be readmitted to the neonatal unit within 1 year after their initial discharge. My model was built using backwards elimination and multiple methods of interaction selection, with the Hosmer-Lemeshow test and cross validation being the main metrics by which I judged its quality. I suggest a model, that could be used with some alteration to the sensitivity, to fit the needs of the hospitals.

February 12, 2018 Read

A double-blind, randomised, placebo-controlled study of Vinpocetine for increasing IQ

In this project I propose a double-blind, randomised, placebo-controlled study of Vinpocetine for increasing IQ points. I will measure the change from baseline of IQ after 28 days of receiving 10mg of Vinpocetine daily. Block randomisation with multiple block sizes will be utilised. The clinically relevant difference between the placebo and test groups will be 10 IQ points, in a population of healthy individuals aged 18-64 years. The primary analysis will be ANCOVA, and I estimate a required total sample size of 88 individuals, and 65 days to complete the study.

January 29, 2018 Read

The Ethical Issues of Placebo Control Trials

In this project I was tasked with reviewing the WMA declaration of Helsinki: Ethical principles for medical research involving human subjects, the ICH E10: Choice of control group and related issues in clinical trials harmonised tripartite guideline, and other sources to identify the ethical issues in placebo control trials and the methodological issues in active control trials, and explain how each might be overcome by design modifications. Length: 3 pages Report: The Ethical Issues of Placebo Control Trials This project formed part of my Masters in Statistics.

January 29, 2018 Read

Bayesian Ridge Regression: An Overview and Comparison to Classical Regression

In this project I explain the theory behind Ridge regression from a Bayesian perspective and suggest why one might use Ridge regression over classical methods. Then, using a data set on Diabetes I constructed a series of classical and Ridge models and compared their effectiveness, including an extension to a selection of ‘hybrid’ models. I found that, for this data set, the classical subset models were better at the prediction of new data than the Ridge models, but suggest situations in which the Ridge models may be preferable.

January 15, 2018 Read

A preliminary investigation regarding violence on a Yorkshire Hospital Psychiatric ward

In this project I was tasked with performing an initial exploratory analysis on data arising from an observational study in a Yorkshire hospital psychiatric ward, and returning my findings on some questions of interest. I found strong evidence of a relationship between the number of Restraint and Control staff present on a ward, and the proportion of incidents that were ‘Near Misses’. I utilised time series plots, group bar charts, grouped histograms, and pie charts to visualise the data.

December 15, 2017 Read

Investigating the efficacy of Formoterol compared to Salbutamol using adjusted analysis

This project tasked me with investigating the efficacy of two treatments for Asthma, Formoterol and Salbutamol, and highlighting any statistically significant differences. After investigating a range of possible methods that could be used, Analysis of Covariance was chosen for its unbaised estimates in the scenario of baseline imbalance. The analysis showed a non-significant effect for treatment group (p-value = 0.25688) on the response variable, and provided no evidence for a difference in efficacy of the two treatments, but the clinical trial itself had many flaws outlined in the discussion.

November 6, 2017 Read