Personalised uncertainty estimates for intelligent blood test ordering
Inefficiencies in the delivery of care currently have an enormous cost to the NHS. Frequent blood testing is unpleasant for patients, and possibly harmful by contributing to anaemia (up to 40-70ml/day of blood loss in Intensive Care Units) and delirium as it disrupts sleep and rest.
Many blood tests are not clinically necessary and bulk orders are often generated by ritual or habit. This includes routine postoperative testing and daily rote orders in critical care. These problems are international and contribute to the Institute of Medicine’s estimate of more $200 billion worth of unnecessary tests and procedures in the US alone.
This research aims to enhance patient care by assisting clinicians in ordering timely and pertinent blood tests. By integrating into electronic health record systems, this study hopes to provide clinicians with precise, timely data, enabling them to make more informed diagnostic decisions.
A variety of statistical methods will be used to analyse historic blood test trends and predict future values. Uncertainties of these predictions will be presented to help clinicians decide whether a new blood test may be necessary. Other sources of information such as clinical notes, medications and vital signs will be incorporated to help refine the predictions. Further work will then be done to generalise the developed methods so they may apply to other means of obtaining information about a patient’s condition.