The Feasibility of Automating Assessment of Concordance Between Advance Care Preferences and Care Received Near the End of Life

摘要

Background: End-of-life care is patient centered when it is concordant with patient preferences. Concordance has beenfrequently assessed by interview, chart review, or both. These time-consuming methods can constrain sample sizes, preclud-ing population-level quality assessment. Concordance between preferences and care as measured by automated methods isdescribed.Methods: Automated processes extracted and analyzed electronic health record (EHR) data to assess concordance between15 advance care planning preference domains and 232 related end-of-life care events for 388 patients aged 65 years or olderwith an inpatient encounter at Kaiser Permanente Southern California who died during or after the encounter. Patientpreferences were recorded in advance directives or physician orders or reflected in hospital code status. Concordance, assessedin relation to the most recent documents, orders, or code status, occurred when patients received care they preferred or didnot receive nonpreferred care. Discordance occurred when patients received care they did not prefer or did not receive carethey preferred.Results: Overall concordance for 12,592 observed end-of-life care events was 97.7%. A total of 55 of 4,154 (1.3%) receivedcare events were nonpreferred, according to patient preferences in the EHR. Automated methods could not distinguishbetween medically nonbeneficial treatments, those that were not medically indicated, and potential undertreatment.Conclusion: Automating assessment of concordance between care near the end of life and preferences is feasible but re-quires model refinement and discrete care preference data. Automated methods may be most valuable as a screening tool toidentify potential overtreatment and undertreatment, with chart review to verify discordance.