Improving access to healthcare among low-income populations: Operational research modelling approaches to support the outpatient service delivery process

Abstract

This thesis studies no-show behaviour for medical appointments. It comprises four research papers, each of which addresses a different aspect of the problem. The case study is an outreach program designed to overcome access barriers affecting low-income patients in Bogotá, Colombia. The research uses a range of approaches, both qualitative and quantitative, and represents a scientific contribution in terms of the novel methodology developed to tackle some of these problems. However its key feature is its relevance to real world decision making through a longstanding collaboration with the Secretaria Distrital de Salud in Bogotá, who have supported the research throughout.

First, in Chapter 2, we assess the effectiveness of three machine learning models to predict individual attendance probabilities using routinely collected administrative data. Although all three models allow us to identify those patients at higher risk of no-show, due to the limitations of the data it is not possible to understand the reasons behind patients’ health-seeking behaviour. Therefore, in Chapter 3 we show the benefits of combining these machine learning models with an in-depth qualitative methodology. Particularly, we aim at understanding patients’ experience with the cervical cancer screening program in Bogotá. This paper uses a mixed methods approach, in which qualitative data are used to explain quantitative results. Sixty semi-structured interviews were conducted, and the Health Belief Model (HBM) used as a conceptual framework to build
second order categories. The Framework method was used to analyse the qualitative data. Then, in Chapter 4, we validate the use of the HBM to explain and predict no-show behaviour for cervical cancer screening appointments among low-income hard-to-reach women in Bogotá. A randomly selected sample of 1699 women was surveyed using a 37-item instrument. We quantify the relationship between each construct of the HBM and the attendance probabilities for cervical cancer screening. Additionally, we propose a sequential approach to improve the accuracy of the no-show prediction, using the survey results. Finally, in Chapter 5 we develop a model to select which patients will receive a given behavioural intervention to increase attendance, in situations where funding is limited. Our aim is to classify patients into three groups, based on their attendance probabilities: one group at high risk of no-show who will receive a more costly personalized intervention; a medium-risk group who will receive a cheaper mass intervention; and a low-risk group who will not receive any intervention at all. To do this in a fair way, i.e. one that does not disadvantage specific subgroups, we develop a novel optimization-based post processing approach aimed at addressing machine learning bias in the algorithmic classification problem.

More information

Identifiers

ORCID for Honora Smith: orcid.org/0000-0002-4974-3011

ORCID for Sally Brailsford: orcid.org/0000-0002-6665-8230

ORCID for Steffen Bayer: orcid.org/0000-0002-7872-467X

Catalogue record

Export record