Q J Med 2003; 96: 583-591
© 2003 Association of Physicians
Variation in intubation decisions for patients with chronic obstructive pulmonary disease in one critical care network
From the 1Health Services Research Unit, London School of Hygiene and Tropical Medicine, London, 2Department of Primary Care. University of Birmingham, 3Department of Critical Care, University Hospital Birmingham, and 4Department of Critical Care, City Hospital, Birmingham, UK
Received 9 April 2003 and in revised form 21 May 2003
 |
Summary |
|---|
 Top Summary IntroductionMethodsResultsDiscussionConclusionReferences |
|---|
Background: Anecdotal evidence suggests variation in intubation decisions for chronic obstructive pulmonary disease (COPD) patients with respiratory failure, but little is known about the extent of or reasons for this variability.
Aim: To describe clinician decision-making for patients with exacerbations of COPD considered for intubation.
Design: Telephone simulation study.
Methods: Consultants responsible for COPD admissions in the Heart of England Critical Care network were asked to decide whether or not to admit three patients with COPD to ICU on the basis of information conveyed over the telephone. Consultants were also asked to predict patients survival in ICU hospital and at 180 days on the assumption that the patient did receive ICU care.
Results: Of the 120 consultants, 98 (82%) took part; 89% would admit patient 1, 64% patient 2, and 40% patient 3. The prediction of survival if ICU admission had occurred differed significantly between admitters and non-admitters. Mean predicted post-ICU hospital survival for patient 1 was 46% (95%CI 43–49) for admitters, and 13% (95%CI 6–19) for non-admitters (p < 0.001). The respective figures for patient 2 were 38% (95%CI 34–42) vs. 12% (95%CI 8–15) (p < 0.001), and for patient 3, 28% (95%CI 24–33) vs. 13% (95%CI 10–16) (p < 0.001). For a housebound COPD patient in their mid 70s, the mean (SD) threshold of predicted hospital survival below which consultants would recommend not admitting to ICU was 22% (13.2%).
Conclusions: Consultants differed markedly in their admitting decisions about identical patients. Objective outcome prediction models might improve equity in ICU bed use for patients with COPD.
|
Introduction |
|---|
|
TopSummaryIntroductionMethodsResultsDiscussionConclusionReferences |
|---|
Intensive care unit (ICU) gate-keeping has life-or-death significance for individuals, yet the decision-making involved might well be subject to the widespread variability documented to permeate clinical practice.1 Patients admitted to ICU should be those expected to receive sustained benefit in terms of quality and length of life,2 and critical care networks have been called upon to develop consistent admission thresholds.3 For many patients, consistency will already exist. All units will admit patients with clearly potentially reversible illness, and not those with obvious terminal disease. Difficulties arise for patients with progressive diseases such as chronic obstructive pulmonary disease (COPD) where the natural history spans a continuum from early mild disease to late terminal disease. COPD patients requiring ICU admission for invasive ventilatory support will usually die if they do not receive this treatment. Gate-keeping decisions influenced by patients, potential to benefit will involve the complex task of prognostication. Such decisions involve uncertainty. Though the normative approach to decisions under uncertainty is Bayesian, clinicians tend to use shortcuts (heuristics) and these heuristics are prone to bias.4,5 American studies using simulated paper cases have suggested that clinicians may vary in their predictions of outcome for identical COPD patients.6 However, paper-based cases may not catch real-life decision-making, which typically involves responding to an oral presentation of patient characteristics. There are no studies investigating decision-making for patients with COPD across UK critical care networks. We present the results of a study to investigate intubation decisions for COPD patients in one critical care network. The study exploits the fact that on occasion consultants are required to make intubation decisions on the basis of patient details conveyed by telephone. Simulation of these phone calls allowed all consultants within a critical care network to be presented with identical patients, to formulate prognostic estimates and to make intubation decisions using information presented in a way that mimicked real life.
|
Methods |
|---|
|
TopSummaryIntroductionMethodsResultsDiscussionConclusionReferences |
|---|
The study received multi-centre research ethics committee approval. Consultants taking part in the emergency care of patients with COPD within the hospitals of the Heart of England critical care network were identified and a letter outlining the purpose of the study was sent. An appointment was made to contact consultants willing to take part in the study by telephone. The investigator contacting participants took on the role of a registrar phoning them at 2 am in the morning requesting advice about whether or not to intubate a COPD patient and admit them to ICU. A single interviewer with clinical experience of intensive care carried out all interviews. Consultants considered the cases in the same order, with patient 1 first and patient 3 last. It was made clear that each patient should be considered as though they were the only COPD patient requiring intubation and ICU admission that night. Consultants were informed that the circumstances following hospital admission and prior to the registrars phone call were identical for each of the three patients. Each patient had been admitted to hospital 18 h previously at 0800 h, and had deteriorated despite appropriate therapy that had included a failed attempt at non-invasive ventilation. Consultants were informed that the patient had been fully assessed, and any information required in deciding the patients’ further management could be obtained from the registrar. It was also explained that though the patient was too ill to converse, the patients spouse was available to answer questions. All three patients had single-organ respiratory failure but had important differences in key characteristics identified as having independent prognostic significance in the SUPPORT study.7 Brief summaries of the three cases are outlined in Table 1. Following the completion of data collection, the SUPPORT COPD model7 was used to estimate the probabilities of survival of the three patients.
|
Consultants were asked to request the information that they needed to make the intubation decision. The cues requested were noted. It was explained that their predictions of ICU, hospital and 180-day survival would also be requested once they had decided whether or not to intubate. If consultants chose not to intubate a patient, they were asked to predict the patient’s outcome following ICU admission as if their registrar had intubated the patient and admitted to ICU without discussion.
When all three patients had been considered consultants were asked about the predicted hospital survival threshold at which they would tend to move from intubating a patient, to recommending palliative care on the wards. Consultants were asked to formulate this threshold considering patients with COPD who were in their mid-70s and housebound, but who had expressed the wish to go ahead with intubation if they could get back to how they were prior to the exacerbation leading to hospitalization.
Finally, consultants were asked how well they felt the exercise reflected real-life experience.
Outcome predictions and the number of end-of-life decisions made in the past year were compared for each patient scenario in turn. Outcome predictions were compared with the two sample T-test and number of end-of-life decisions with the Mann-Whitney test.
|
Results |
|---|
|
TopSummaryIntroductionMethodsResultsDiscussionConclusionReferences |
|---|
Response rates
We identified 125 consultants with responsibility for the emergency care of COPD patients admitted to the eight hospitals of the Heart of England Critical Care network. Five retired before the study started, leaving 120 eligible, of whom 98 (82%) took part. Responders had been qualified for a mean (SD) of 21.1(7.4) years and non-responders for 24.4 (7.2) years (p = 0.06). Of the 98 responders, 78 (80%) were male, as were 20 (90%) of the 22 non-responders.
Responders had been consultants for a mean (SD) of 9.6 (8.1) years and reported making intensive care gate-keeping decisions on a median (IQR) of 10 (6.0–20.0) patients in the past 12 months. By speciality, 34/37 (92%) intensivists responded, 17/21 (81%) respiratory physicians responded and 47/62 (76%) other physicians responded.
Admitting decisions and survival predictions
Consultants choosing to intubate a patient were significantly more optimistic about the patients probability of survival at ICU discharge, hospital discharge or 180 days than consultants predicting those outcomes who would choose not to intubate (Table 2). It is important to emphasize that non-intubaters were providing predictions that estimated the outcome if they had chosen to intubate the patient. For all three patients there was no significant difference between intubaters and non-intubaters in terms of the number of COPD admission decisions made in the past year, number of years since qualification and years since attaining consultant status. The study was not adequately powered to examine difference in intubating decision between physicians grouped by speciality.
|
Patient 1
Eight-seven (89%) of the consultants would intubate patient 1. Intubating consultants predicted the patient to have a higher probability of survival than non-intubaters at ICU discharge (56% vs. 23%, p < 0.0001), hospital discharge (46% vs. 13%, p < 0.0001) and at 180 days (39% vs. 10%, p < 0.0001). Both intubaters and non-intubaters were more pessimistic than the survival probabilities calculated by the SUPPORT model, which gave a 30-day survival of 90%, an 85% probability of 60-day survival and a 74% probability of 180-day survival. Figure 1 displays the decision making for patient 1.
|
Patient 2
Sixty-three (64%) of the consultants would intubate patient 2. Once again, intubating consultants predicted the patient to have a higher probability of survival than non-intubaters at ICU discharge (47% vs. 19%, p < 0.0001), hospital discharge (38% vs. 12%, p < 0.0001) and at 180-days (32% vs. 8%, p < 0.0001). Once again both intubaters and non-intubaters were more pessimistic than the survival probabilities calculated by the SUPPORT model, which gave a 30-day survival of 94%, a 91% probability of 60-day survival and a 84% probability of 180-day survival. Figure 2 displays the decision making for patient 2.
|
Patient 3
Thirty-nine (40%) of the consultants would intubate patient 3. Again, intubating consultants predicted the patient to have a higher probability of survival than non-intubaters at ICU discharge (37.% vs. 20%, p < 0.0001), hospital discharge (28% vs. 13%, p < 0.0001) and at 180-days (21% vs. 8%, p < 0.0001). Once again both intubaters and non-intubaters were more pessimistic than the survival probabilities calculated by the SUPPORT model, which gave a 30-day survival of 68%, a 57% probability of 60-day survival and a 34% probability of 180-day survival. Figure 3 displays the decision making for patient 3.
|
Cue use
Overall, consultants requested a median of 10 (IQR 8–12) cues. When analysed by patient, there was no significant difference in the number of cues requested by intubaters and non-intubaters. Although > 99% of consultants requested information about the patient's exercise tolerance, < 10% requested information about the patient's albumin and < 7% requested information about the patient's body mass index. Only one consultant requested all the information required for the SUPPORT model.
Intubation threshold
Five out of 98 consultants felt unable to provide a predicted hospital survival below which they would not intubate a COPD patient in their mid 70s with an expressed wish to receive intubation if they could get back to their prior level of function. For consultants willing to provide such an estimate, the mean (SD) predicted hospital survival threshold below which consultants would move from intubating to not intubating a patient to critical care was 22% (13.2%). Thirty-two (32.7%) consultants would move from intubating to not intubating when they predicted the patients probability of survival following critical care admission was between 1% and 10%, 23 (23.5%) at a probability between 11% and 20%, 22 (22.4%) at a probability between 21% and 30%, seven (7.1%) at a probability between 31% and 40%, eight (8.2%) at a probability between 41% and 50%, and 1(1%) at a probability between 51% and 55%. Comparison of intubation threshold between intubaters and non-intubaters for each patient did not reveal a statistically significant difference (Figure 1).
Consultants’ view of simulation
Sixty (70%) of the consultants considered the phone call caught their decision-making very well, 24 (24%) quite well, three (3%) not very well and one (1%) poorly with one consultant not giving a view. Consultants’ main criticism was that the registrar providing the description would usually have an opinion, and discussion with other consultants may well occur.
|
Discussion |
|---|
|
TopSummaryIntroductionMethodsResultsDiscussionConclusionReferences |
|---|
Our main finding was that consultants who regularly have responsibility for making intubation decisions for COPD patients showed important differences in their intubation practices, despite considering identical patients with identical preferences in the face of uniform resource availability. Intubaters and non-intubated did not differ in the number of cues requested, but intubaters were significantly more optimistic about patients’ survival after ICU admission than non-intubaters. In comparison to estimates provided by the US SUPPORT study model, consultants were pessimistic about COPD patients’ chances of surviving to leave hospital.
The response rate in this study was 82%, which compares favourably to the average response rate of 61% by physicians to mailed questionnaires in a random selection of all published studies between 1985 and 1995.8 However, non-responders typically differ systematically from responders and in this study they tended to have been qualified longer and be male. Table 2 shows that non-admitters tended to be males who had been qualified longer than admitters, and hence our study may have underestimated the extent of non-admission.
A criticism of simulation studies is that they do not reflect real life and that clinicians may not act as they say they will act. It is reassuring that 70% of clinicians considered that the phone calls elicited their decision-making very well. Simulation studies may be useful when they provide insight into an area that is otherwise impossible to study. Gaining insight into clinicians judgements about identical patients intubated in routine practice is impossible. A meaningful sample of clinicians will never make decisions about an identical patient. Risk-adjustment models available in intensive care are currently inadequate to define identical patients across units. Models such as APACHE II are generic, and do not discriminate well between COPD patients with differing functional reserve, since the vast majority will be in the worst category of the respiratory impairment measure. The SUPPORT model is not yet available in the UK. In addition, it would be impossible to standardize unmeasured confounders such as bed availability and the views of relatives and patients.
Order effects will tend to influence decisions about patients, so that clinicians may well respond differently to a moderately ill patient considered after a very sick patient, than if the moderately sick patient were considered after a mild patient. However the study was not seeking to determine the ‘true' admission threshold that attended a patient with a given set of characteristics. Instead, the aim of the study was to investigate variations in clinician decision-making under identical conditions. It was for this reason that all consultants were asked to consider the patients in the same order (1–3). Order effects may therefore influence the absolute probabilities the clinicians gave to the different patient profiles, but since all clinicians considered the patients in the same order, comparisons between admitters and non-admitters will not have been distorted by order effects. A study using randomly ordered cases and thus free-form order effects would have required a sample size beyond the scope of the current study.
Intuitively, one might expect that hospitals may develop a tacitly agreed approach to difficult management decisions through discussions over time. It would have been interesting to look at comparisons between the eight hospitals, but the study was underpowered to consider this area.
A common reason for distorted responses in simulation studies is social desirability bias that results in respondents tending to reply in such a way as to avoid the possibility of criticism.9 Although this might influence clinicians stated intubation policy, it seems unlikely to the authors that the clinicians who decided to intubate patients would perceive pessimism about outcome to be a socially desirable response. In any case, the potential significance of the results is not only the absolute survival probabilities, but also the differences between clinicians predictions in the face of identical information. The decision to intubate may be a gestalt reaction to the patient as a whole, and the survival probabilities provided may be constructed post hoc to support this decision. Nevertheless, when it comes to conversations with patients or relatives it is those ‘supporting probabilities' that will be translated into predictions of a good or poor outcome.
The other potentially interesting finding of the study is the pessimism of the clinicians’ survival predictions in comparison to the predicted outcome of the patients using the COPD SUPPORT model. Caution must be taken in extrapolating risk adjustment models between health-care systems.10 The SUPPORT study recruited 1016 consecutive COPD admissions with type II respiratory failure with the outcome model developed on the first 600 admissions and tested on the next 416. The receiver-operating characteristic (ROC) for the SUPPORT COPD model predicting 6-month survival was 0.75.11 In the SUPPORT study, 342 COPD patients were ventilated, with an overall hospital survival of 75% and 180-day survival of 57%, broadly similar to the 66% hospital survival and 64% 180-day survival for 242 COPD patients admitted to UK ICUs.12 SUPPORT made it clear that outcome prediction in COPD patients is difficult; however, the SUPPORT model differed from physician estimates of outcome in that the physicians tended to make more extreme predictions (very high or very low likelihood of survival) than the patients experienced.13 Wennberg has recently highlighted the need to understand the marked variations in supply-sensitive services such as ICU admissions for patients with chronic diseases.1 In the UK health-care system, which has a fraction of the ICU beds available to US clinicians, a pessimistic view of COPD patients’ prognosis might ease the cognitive dissonance experienced by clinicians forced to manage in such an environment.
|
Conclusion |
|---|
|
TopSummaryIntroductionMethodsResultsDiscussionConclusionReferences |
|---|
We have shown that in simulated phone scenarios consultants within one critical care network show important variations in intubation decisions and survival predictions for identical COPD patients. If the variation observed in the simulation occurs in clinical practice it is likely to undermine equity in provision of intubation for these patients. The study raises the possibility that the development and assessment of models capable of objective prognostic estimation might have the potential to improve the equity of ICU bed usage for patients with COPD.
|
Acknowledgments |
|---|
Martin Wildman is funded by a Medical Research Council Training fellowship in Health Services Research.
|
Footnotes |
|---|
Address correspondence to Dr M.J. Wildman, Health Services Research Unit, Department of Public Health and Policy, London School of Hygiene and Tropical Medicine, Keppel Street, London WC1E 7HT. e-mail: Martin.Wildman{at}lshtm.ac.uk
|
References |
|---|
|
TopSummaryIntroductionMethodsResultsDiscussionConclusionReferences |
|---|
1. Wennberg JE. Unwarranted variations in healthcare delivery: implications for academic medical centres. Br Med J 2002; 325:961–4.
2. Department of Health. Guidelines on the Admission and Discharge from Intensive Care and High Dependency. London, Department Of Health, 1996.
3. Department of Health. Comprehensive Critical Care. A review of adult critical care services. London, Department Of Health, 2000.
4. Elstein AS, Schwartz A. Clinical problem solving and diagnostic decision making: selective review of the cognitive literature. Br Med J 2002; 324:729–32.
5. Dawson NV, Arkes H. Systematic errors in medical decision making: Judgement limitations. J Gen Intern Med 1987; 2:183–7.[Web of Science][Medline]
6. Pearlman RA. Variability in physician estimates of survival for acute respiratory failure in chronic obstructive pulmonary disease. Chest 1987; 91:515–21.
7. Connors AF, Jr., Dawson NV, Thomas C, Harrell FE, Jr., Desbiens N, Fulkerson WJ, et al. Outcomes following acute exacerbation of severe chronic obstructive lung disease. The SUPPORT investigators (Study to Understand Prognoses and Preferences for Outcomes and Risks of Treatments). Am J Respir Crit Care Med 1996; 154:959–67.[Abstract]
8. Cummings SM, Savitz LA, Konrad TR. Reported response rates to mailed physician questionnaires. Health Serv Res 2001; 35:1347–55.[Web of Science][Medline]
9. Marlowe D, Crowne DP. Social desirability and responses to perceived situational demands. J Consult Clin Psychol 1961; 25:109–15.
10. Justice AC, Covinsky KE, Berlin JA. Assessing the generalizability of prognostic information. Ann Intern Med 1999; 130:515–24.
11. Lynn J, Ely W, Zhong Z, Landrum K, Dawson NV, Connors AF, et al. Living and dying with Chronic Obstructive pulmonary disease. J Am Geriat Soc 2000; 48:S91–100.[Web of Science][Medline]
12. Wildman MJ, Goldfrad C, Rowan K. Functional health following intensive care unit admission for patients with acute respiratory failure due to chronic obstructive pulmonary disease in the UK. Am J Resp Crit Care Med 1998; 157:A18.
13. Freeborne N, Lynn J, Desbiens NA. Insights about dying from the SUPPORT project. J Am Geriat Soc 2000; 48:S199–205.[Web of Science][Medline]
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  B Chakrabarti, R M Angus, S Agarwal, S Lane, and P M A Calverley Hyperglycaemia as a predictor of outcome during non-invasive ventilation in decompensated COPD Thorax, October 1, 2009; 64(10): 857 - 862. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M.J. Wildman, C. Sanderson, J. Groves, B.C. Reeves, J. Ayres, D. Harrison, D. Young, and K. Rowan Predicting mortality for patients with exacerbations of COPD and Asthma in the COPD and Asthma Outcome Study (CAOS) QJM, June 1, 2009; 102(6): 389 - 399. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 B. Creagh-Brown and C. Shee Noninvasive ventilation as ceiling of therapy in end-stage chronic obstructive pulmonary disease Chronic Respiratory Disease, August 1, 2008; 5(3): 143 - 148. [Abstract] [PDF]
|
|
|
|
 |
|
 B Chakrabarti and P M A Calverley Management of acute ventilatory failure. Postgrad. Med. J., July 1, 2006; 82(969): 438 - 445. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 Minerva BMJ, August 30, 2003; 327(7413): 510 - 510. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||