International Choice Modelling Conference, International Choice Modelling Conference 2017

Font Size: 
Agency in health care: who preferences count in Intensive Care Unit referrals?
Nicolas Krucien

Last modified: 28 March 2017


AUTHORS ((*) corresponding author)

Chris BASSFORD; Nicolas KRUCIEN (*); Mandy RYAN; Anne-Marie SLOWTHER



CONTEXT: Admission to an intensive care unit (ICU) improves the chances of survival for critically ill patients. However the burdens of ICU therapy, and the limited prognosis for many critically ill patients, mean that admission to ICU is not appropriate for all patients. In the context of limited information about patients’ views and health status, clinicians must decide whether the burdens of ICU care outweigh any potential benefit and thus whether admission is in the best interests of the particular patient. In this context a clinician’s personal values and views on withholding or withdrawing life prolonging treatment might affect an individual ICU consultant’s decision.

OBJECTIVE: The objective of this study is to investigate the effects of clinicians’ personal preferences for their own care if they were critically ill, in different illness contexts such as terminal illness or dementia, on their decisions to admit patients into ICU.

METHODS: We ran a discrete choice experiment (DCE) to measure the effect of eight patient-related characteristics on clinicians’ admission decisions: patients’ age; type of main comorbidity; severity of main comorbidity; severity of acute condition (NEWS score); patients’ mobility; patients’ safety at the hospital; family views; and patient assessment by colleagues. Identification of these characteristics was informed by a systematic review of the literature on ICU admissions and a focussed ethnographic study of the decision making process. A two-steps D-Efficient designing procedure was used to obtain 24 pairwise choice tasks (i.e., patient A vs. patient B). Bayesian updating of the initial design, with informative priors obtained from the quantitative pilot study (N=35), informed the final design. These final 24 choice tasks were divided into two blocks of 12 tasks and the participants were randomly allocated across blocks. The questionnaire also included one warm-up task (Task #1) to familiarise the participants with the format of the choice questions, one stability test (Task #14 as repetition of task #1), and one dominance test (Task #15). The order of the choice tasks was randomised. In addition to the pairwise choice, participants were asked to state, for each patient, whether s/he should admitted at all. This information was used to derive a rank-ordering of the patients (e.g., Patient A > No admission > Patient B) which served as a basis for the modelling of clinicians’ preferences. We included six questions to capture clinicians’ attitudes towards critical care (i.e., invasive ventilation; cardio-pulmonary resuscitation; organ support) in difficult situations (i.e., terminal illness; dementia). The effect of such attitudes on choices was estimated using a Hybrid Choice Multinomial Logit model (HC-MNL).

For HC-MNL model #1:

U(ntj) = b(n).ADMIT(ntj) + SUM[b(k).X(ntjk)] + ERROR(ntj); ERROR(ntj) ~ iid EV1

b(n) = b + j(1).ILLNESS*(n) + j(2).DEMENTIA*(n)

For HC-MNL model #2:

U(ntj) = b.ADMIT(ntj) + SUM[b(nk).X(ntjk)] + ERROR(ntj); ERROR(ntj) ~ iid EV1

b(nk) = b(k) + j(1k).ILLNESS*(n) + j(2k).DEMENTIA*(n)

The sample of respondents consisted in. The choice questionnaire was administered online to 303 clinicians working in UK-based ICU. It is expected that clinicians with positive views towards critical care are more likely to admit patients (H1a: j1 > 0; H1b: j2 > 0). We also expect to find a significant effect of consultants’ attitudes on their preferences for patient-related characteristics (H2a: j1k 0; H2b: j2k 0).

RESULTS: Despite the large increase in the number of model parameters, the HC-MNL model still outperforms its standard MNL counterpart (Log-likelihood ratio test: Deviance = 101.8; DF = 34; P5% = < 0.001), indicating that consultants’ personal preferences for critical care have a significant influence on their decisions to admit patients into ICU. Regarding the model estimates, 11 interaction effects out of 34 were significant at 95% confidence level. Consultants with more positive views regarding critical care in case of terminal illness are more likely to admit 66 and 79 years old patients (j = +0.252; j = +0.236), and also those with a less severe acute condition (NEWS score = 8) (j = +0.242). At the opposite they are less likely to admit patients suffering from COPD (j = -0.329) or dementia (j = -0.261). The pattern of results for the “attitudes towards critical care in case of dementia” is different. Consultants with more positive views are more likely to admit patients with COPD (j = +0.388), dementia (j = +0.445), very severe acute condition (NEWS score = 11) (j = +0.209), not safe in their current ward (j = +0.178), with less than good assessment by colleagues (j = +0.225). Regarding the attitudes-related results, we observe systematic differences between terminal illness and dementia (All forms of critical care appear to be more acceptable in case of terminal illness), and between the types of critical care (Organ support appears more acceptable than cardio-pulmonary resuscitation).

CONCLUSION: Our results indicate that what consultants want for them in case of terminal illness or dementia has a significant effect on their decision to admit patients into ICU. This raises a potential issue of mismatching between consultants’ and patients’ attitudes towards critical care.

Conference registration is required in order to view papers.