Natural language word embeddings as a glimpse into healthcare language and associated mortality surrounding end of life

Abstract

ObjectivesTo clarify real-world linguistic nuances around dying in hospital as well as inaccuracy in individual-level prognostication to support advance care planning and personalised discussions on limitation of life sustaining treatment (LST).DesignRetrospective cross-sectional study of real-world clinical data.SettingSecondary care, urban and suburban teaching hospitals.ParticipantsAll inpatients in 12-month period from 1 October 2018 to 30 September 2019.MethodsUsing unsupervised natural language processing, word embedding in latent space was used to generate phrase clusters with most similar semantic embeddings to ‘Ceiling of Treatment’ and their prognostication value.ResultsWord embeddings with most similarity to ‘Ceiling of Treatment’ clustered around phrases describing end-of-life care, ceiling of care and LST discussions. The phrases have differing prognostic profile with the highest 7-day mortality in the phrases most explicitly referring to end of life—‘Withdrawal of care’ (56.7%), ‘terminal care/end of life care’ (57.5%) and ‘un-survivable’ (57.6%).ConclusionVocabulary used at end-of-life discussions are diverse and has a range of associations to 7-day mortality. This highlights the importance of correct application of terminology during LST and end-of-life discussions.

The use of natural language processing in palliative care research: A scoping review

“…Clinical notes and electronic medical records were the most common primary data sources, used in 57 studies (69.5%). 21,23–27,29,30,33–36,40,42–46,48–52,54,55,57–64,66–68,70–73,75,79–93,95 Other primary sources included audio recordings ( n = 6, 7.3%), 6,28,32,38,39,65 administrative data ( n = 5, 6.1%), 37,47,53,77,...…”

“…Brizzi et al, 25 Kern et al, 49 Lau et al, 53 Lee et al, 54,55 Lindvall et al, 58 Poort et al, 71 Udelsman et al 84 Conversation analysis/conversational dynamics during serious illness conversations 7 (8.5) van den Broek-Altenburg et al, 6 Clarfeld et al, 28 Durieux et al, 32 Gramling et al, 38,39 Manukyan et al, 65 Ross et al 72 Symptom identification 6 (7.3) DiMartino et al, 30 Forsyth et al, 35 Heintzelman et al, 43 López-Torrecilla et al, 60 Taggart et al, 83 Yang et al 91 Decision-making 4 (4.8) Barrett et al, 24 Ouchi et al, 70 Saeed et al, 73 Almasalha et al 85 Identification of patients likely to benefit from palliative care or palliative needs or palliative status 4 (4.8) Murphree et al, 68 Sandham et al, 75 Song et al, 79 Zhang et al 97 Clinical phenotypes identification 3 (3.6) Ernecoff et al, 34 Jay et al, 45 natural language processing software to complete data analysis. This suggests that each software has its own strengths, and that weaknesses may be overcome by the combined use of more than one software solution to achieve the desired goal of the analyses.…”