Article Text

PDF

Decision analysis in NSCLC
  1. F Macbeth1
  1. 1Velindre Hospital, Cardiff CF14 2TL, UK; fergus.macbeth{at}velindre-tr.wales.nhs.uk
  1. J Dowie2,
  2. M Wildman2
  1. 2Public Health and Policy Department, London School of Hygiene and Tropical Medicine, London WC1E 7HT, UK; jack.dowie{at}lshtm.ac.uk

Statistics from Altmetric.com

The paper by Dowie and Wildman1 exploring the decision analysis for surgically high risk patients with stage Ia non-small cell lung cancer (NSCLC) was interesting. However, I have reservations about the figures used to inform the decision tree.

There has never been a randomised trial comparing surgery with radical radiotherapy for these patients and the authors have taken 5 year survival figures for quite different groups of patients. The surgical survival figures come from a very large series of patients of all ages and levels of fitness,2 while the radical radiotherapy figures come from one relatively small retrospective series of patients who were considered unfit for (or refused) surgery.3 These patients have, for reasons of age and co-morbidity, a lower background probability of 5 year survival. The Cochrane Review on this topic4 identified 26 studies, including over 2000 patients. For patients with T1 tumours from six studies, 5 year survival ranged from 29% to 37%, but there was a high rate of intercurrent deaths (median 25%). So, in the model, either the 5 year survival probability for surgery needs to be adjusted down or that for radical radiotherapy adjusted up to allow for the different case mix.

Also, we have no evidence on the natural history of untreated stage Ia disease. The probability of 5 year survival without treatment of 1% that is used in the model may be unduly low because it is the figure generally quoted for patients with advanced disease at presentation treated palliatively.

Why are the utilities for survival 10 points lower after radical radiotherapy than after surgery? A patient who is alive at 5 years with a given level of disability should express the same utility whatever the original treatment. The utility after surgery could, in fact, be lower because some patients are troubled by post-thoracotomy chest wall pain.

Like Treasure,5 I have the feeling that this model was derived from a theoretical study of the literature without adequate discussion with clinicians. It is, however, a start. I would be happy to cooperate on a more clinically based further exploration of the topic.

References

Authors’ reply

We are grateful for the specialist input of Dr Macbeth who uses his expertise to contribute to the refinement of the probabilities and utilities in the decision tree. It was clearly not our intention to provide a definitive tree, simply an illustrative one which would demonstrate the potential of decision trees in integrating the best evidence with individual patient preference in a coherent and transparent way. But we were able to use this tree to confirm that the surgical mortality threshold for an individual patient should be determined by their expected mortality (and health related quality of life) under the alternative options, rather than by considering the surgical consequences alone.

The study by Gauden et al1 was selected to provide a point estimate of the 5 year survival possible following radiotherapy (range of 5 year survival in the Cochrane review2 29–37%, Gauden study 5 year survival 32%). The Cochrane review highlights the fact that survival following radiotherapy is influenced by patients’ performance status, presence or absence of weight loss, and tumour size. The tree provides a structure to allow this complexity to be explicitly taken into account. We see decision trees allowing experts in the various disciplines involved in cancer care to operationalise textual guidelines and evidence for multidisciplinary decision making in a transparent and analytically coherent way. Consensus techniques would be required to identify the important patient archetypes, the best evidence applying to these archetypes, and the implications of the archetypes for each of the candidate treatment modalities. Macbeth illustrates the necessity of this process by suggesting the need to alter the 5 year survival probability in the light of case mix. A similar consensus process (taking patient characteristics into account) could be used to refine the probability of 5 year survival without treatment.

As far as utilities are concerned, these are best elicited directly from patients using techniques that also take account of attitudes to risk. In the tree in the paper the utility of 5 year surgical survivors is set at 100 in patients with only mild shortness of breath, but at 70 in 5 year surgical survivors with severe shortness of breath. In order to keep the illustrative tree relatively simple, 5 year survivors of radiotherapy include all levels of dyspnoea and the utility chosen lies between the two groups of surgical survivors to take account of this.

Both Macbeth and Treasure3 are absolutely correct to observe that the paper represents a largely theoretical demonstration of the potential of decision analysis, though that does not mean it has not already yielded some important insights. We agree that a substantive decision analysis requires extensive discussion and collaboration with both clinicians and patients, and we would welcome an opportunity to take this work forward in such a cooperative environment. In our experience clinicians—and funding bodies—vary enormously in their assessment of the value of analysis based decision technologies, but it is reassuring to see that there is increasing recognition of their future potential.

References

View Abstract

Request permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.