MECHANISMS OF CLINICIAN-ATTRIBUTABLE VARIATION IN HEALTH CARE: A PILOT STUDY

Purpose: Patterns of healthcare delivery are known to vary among and within centers and regions. Although among-center variation may be partially attributable to a center's "culture", the concomitant within-center differences suggest that individual clinicians' approaches to medical decision making may drive much of this variability. We aimed to assess the variability among clinicians in selected attributes hypothesized to be associated with healthcare variation. We also explored associations between these attributes and clinicians' accuracy in predicting outcomes for intensive care unit (ICU) patients.

Methods: We administered an online questionnaire to ICU clinicians who had provided prognostic estimates of outcomes for at least 5 patients they cared for in the ICU as part of a prospective cohort study. We measured clinicians' scores on the Collett-Lester Fear of Death (CLFOD) scale, Physicians' Reactions to Uncertainty (PRU) scale, Jefferson Scale of Empathy (JSE) scale, Life Orientation Test -- Revised (LOT-R), and a modified End-of-Life Preferences (EOLP) scale. We computed clinicians' accuracy in predicting in-hospital mortality and 4 potential outcomes 6 months after ICU discharge (e.g. mortality, return to original place of residence, quality of life, and ability to toilet independently). We conducted multivariable logistic regression to assess relationships between clinicians' attributes and prognostic accuracy.

Results: 27 of 35 eligible clinicians (77%) completed the questionnaire. The descriptive statistics (median, IQR) for clinicians' scores were: EOLP (45, 31-60), LOT-R (18, 17-23), JSE (78, 77-80), PRU (84, 72-94), CFLOD (90, 83-107). Responses for JSE and LOT-R were fairly consistent among clinicians, whereas clinicians' CLFOD, PRU, and modified EOLP scores demonstrated variability (Figure). ICU clinicians' JSE scores were lower than previously reported scores for primary care physicians. Regression models did not reveal consistent associations between clinicians' attributes and their prognostic accuracy scores, although this pilot study had limited power to detect such associations.

Conclusions: Our findings suggest that ICU clinicians vary in attributes that are measurable and may influence their approaches to medical decision making. Scales demonstrating such variability represent candidates for use in future research to identify clinician-attributable mechanisms explaining variation in healthcare delivery. Understanding the causes of clinician-attributable variation will enable development and testing of communication and decision-making interventions to improve the alignment of health care with patients' true goals and preferences.