Pathology testing underpins the diagnostic and treatment pathways in primary care. NHS England highlighted nearly 10 years ago that around 95% of clinical pathways rely on patients having access to efficient, timely and cost-effective pathology services. Increasingly, it has become evident that inequalities in healthcare linked to pathology test findings exist at many different levels, from the provision of tests through to interpretation of findings. AI and clinical decision support systems (CDSS) have been cited as promising tools to improve pathology test ordering behavior. At a population level, the ability for machine learning algorithms to analyse large data sets allows for evidence-based (historical data, published guidelines, post analytic peer review) evaluation of test order patterns and outcomes linked to demographics and referral indication(s). This proposal seeks to evaluate pathology test ordering patterns and outcomes with unbiased AI algorithm predictions against three main themes – demand optimisation, identifying health inequalities, and identifying unwarranted variations in care.
Looking at and applying ML algorithms to pathology test results and patient data together creates a powerful tool for developing prediction models, identifying and addressing health inequalities, quantifying the impact on patient care, e.g., is there clinical value? And if so, does this help direct guidelines and compliance to guidelines. Considerable work has been done looking at demand optimisation in pathology as well as applying machine learning to clinical outcomes, but no one appears to be looking at how, from a big data perspective, test outcomes and population level patient data can be utilised together.
The thesis aims are to :
• Identify and select a cohort of tests (B12, Liver panel for example) where it is suspected that health inequalities are prevalent (inadvertent or otherwise) in Lambeth borough.
• Propose hypotheses as to why these inequalities are thought to exist and use the model to see if the hypotheses are correct.
• Formulate a plan to address above and ideally work with GP’s to support the model’s predictions whilst addressing inequalities.
• Quantify the impact of the above on metrics such as clinical value to patients.
• Assess findings against guidelines and compliance to guidelines asking if compliance is poor can one identify why? (e.g. cost, waste, education, lack of engagement).