Personalising high-risk perioperative care with dynamic models of clinical trajectory

At the beginning of my fellowship...

My intention was to characterize the determinants and trajectory of post-operative complications after major surgery and develop a machine learning algorithm to predict them and their impact on length of stay.

During my fellowship...

A successful computational phenotype was developed and validated so that post-operative complications can be identified at scale without human intervention as basis for the Topol fellowship work (Development and Validation of an Electronic Postoperative Morbidity Score. Stubbs DJ et al (Ercole A). Anesth Analg. 2019;129(4):935-942.)

This was applied to characterise the surgical pathway of chronic subdural patients (Identification of factors associated with morbidity and postoperative length of stay in surgically managed chronic subdural haematoma using electronic health records: a retrospective cohort study. Stubbs DJ, (Ercole A). BMJ Open. 2020;10(6):e037385). This digital project has led to a nationwide project to intervene and improve this pathway.

Methodological work in developing a suitable timeseries machine learning methodology suitable for incorporating unselected variables has been published (Dynamic survival prediction in intensive care units from heterogeneous time series without the need for variable selection or curation. Jacob Deasy, (Ercole A) Sci Rep. 2020;10(1):22129).

Although not part of my fellowship, I have completed (as PI) a multicentre phase II study of a digital intervention for individualized treatment of severe traumatic brain injury (COGITATE study). This will report shortly.

As an intensive care consultant, work has been severely disrupted by COVID-19. I led a group predicting ICU surge capacity. We were able to create a model and make it available within a week using agile techniques ( and this was used to the inform NHS response locally in March 2020.

I have published a number of other peer-reviewed publications in machine learning methodology.

Lessons learned:

  1. Methodological challenges are visible and obvious targets but non-technical skills are crucial in implementing transformation.
  2. COVID-19 has been a disruptive influence. However, it has brought the benefits of data-driven decision-making to the forefront and removed some barriers to change.
  3. Agile data-driven techniques are not embedded in the NHS. However, they can be implemented and useful in defining national policy.