To understand the very diverse experiences of rare disease patients, of whom some have had very long journeys before any suspicion of disease, it is useful to stratify patients into cohorts with similar experiences. We find that there are often more than a few such cohorts of undiagnosed patients.
We then identify these cohorts using the locus of “clinical focus”. This stratification is accomplished with a sophisticated computational analysis, consisting of:
- leveraging important patient attributes/features were identified through interpretable modelling
- computational analysis of EHRs to discover the main trajectories patients follow through the healthcare system and to discover what the primary experiences of patients are in the clinical setting
- refinement of this patient stratification based upon consideration of the differences between the target disease patients and a random sample of ‘typical’ patient EHRs
These groups of patient records are analysed manually to understand what other clinical encounters were being experienced and these encounters are then evaluated for their frequency of occurrence. The clinical notes are also assessed through semantic analysis for cognitive biomarkers and the understanding of the disease at hand by the clinicians. Lastly, the length of the patient experience is measured.
While a patient experience may centre around a certain clinic, this is not necessarily the first clinical touchpoint for a given patient, as we are basing the evaluation on the actual flows experienced by that patient. This means that our methodology focuses the intervention opportunities in a meaningful way, based on which clinical practices see the patients more.
We see that even if patients seem to have experience centred around a single clinical focus, often there are common referral patterns that can picked up that are unique and specific to the healthcare system we are working with. Insights are often derived suggesting that interventions could be most effective if focused on education in these clinical areas, for example, and by creating tooling to help clinicians triage at-risk patients more quickly.