A treatment trade‐off based decision aid for patients with locally advanced non‐small cell lung cancer

Abstract

Purpose To describe the structure and use of a decision aid for patients with locally advanced non‐small cell lung cancer (LA‐NSCLC) who are eligible for combined‐modality treatment (CMT) or for radiotherapy alone (RT).

Methods The aid included a structured description of the treatment options and trade‐off exercises designed to help clarify the patient’s values for the relevant outcomes by determining the patient’s survival advantage threshold (the increase in survival conferred by CMT over RT that the patient deemed necessary for choosing CMT). Additional outcome measures included each patient’s strength of treatment preference, decisional conflict, objective understanding of survival information, decisional role preference, and evaluation of the aid itself.

Results Twenty‐five patients met the eligibility criteria for study. Of these, seven declined the decision aid because they had a clear treatment preference (four chose CMT and three chose RT). The remaining 18 participants completed the decision aid; 16 chose CMT and two chose RT. All 18 patients wished to participate in the decision to some extent. All patients reported that using the decision support was useful to them and recommended its use for others. No patient or physician reported that the aid interfered with the physician‐patient relationship. Patients’ 3‐year survival advantage thresholds, and their median survival advantage thresholds, were each strongly correlated with their strengths of treatment preference (ρ=0.80, P < 0.001 and ρ=0.77, P < 0.001, respectively). For all but one patient, either their 3‐year or median survival threshold was consistent with their final treatment choice. Eight patients reported a stronger treatment preference after using the decision aid.

Conclusions We conclude that a treatment trade‐off based decision aid for patients with locally advanced non‐small cell lung cancer is feasible, that it demonstrates internal consistency and convergent validity, and that it is favourably evaluated by patients and their physicians. The aid seems to help patients understand the benefits and risks of treatment and to choose the treatment that is most consistent with their values.

Introduction

Patients with locally advanced non‐small cell lung cancer (LA‐NSCLC) are often eligible for treatment either with radiation alone (RT), or a combination treatment using chemotherapy and radiation (CMT). The clinical setting offers an opportunity for structured decision support because CMT, when used to treat patients with a high performance status, is associated with slightly higher patient survival rates but RT has fewer treatment related side‐effects. Thus, the ‘preferred’ treatment for each patient is sensitive to how the individual values the relevant treatment outcomes. ¹ , 2, ^–3 The purpose of this paper is to describe a decision aid for patients with LA‐NSCLC facing a choice between the two treatments. The decision aid is based on the treatment trade‐off approach, and is designed to help patients understand information relevant to the decision and to help clarify their values pertaining to pertinent outcomes.

We decided to explore a treatment trade‐off method ^{4 , 5} as a means of appraising patients’ values concerning quantity and quality of life, because the method has been studied in both oncologic ^{4 , 6} , 7, ^–8 and non‐oncologic ⁵ settings, and because it provides a method for the structured presentation of treatment‐related information, allowing the patient to integrate the information in their own way and to make a specific holistic judgement. ⁵ Although the name of the method closely resembles the ‘time trade‐off’ utility assessment method developed by Torrance, ⁹ the treatment trade‐off method does not generate utilities, but instead relies on holistic comparisons between treatment options.

Locally advanced (or surgically unresectable clinical stage III) NSCLC accounts for approximately one‐third of incident NSCLC cases. ¹⁰ A meta‐analysis of clinical trials comparing Cisplatin‐based CMT to RT, estimates that patients receiving CMT have higher survival rates after 3 years (about 9%, compared to about 6% with RT alone). ¹¹ Some individual trials have demonstrated absolute increases in 3‐year survival as high as 11 per cent, ¹² whereas others have shown a much smaller improvement or even inferior patient survival with combined‐modality therapy. ¹¹

While the differences in patient survival between treatments are statistically significant, their clinical significance is less certain. American clinical guidelines have supported the routine use of CMT, ¹³ while guidelines produced by UK ¹⁴ and Canadian ¹ groups vary in their recommendations. Additional lack of consensus among physicians about the optimal management strategy has been illustrated by variation in physicians’ treatment preferences, ¹⁵ , 16, ^–17 by diversity in both management approaches ¹⁸ and in clinical trial design, ¹⁹ and by variation in patterns of practice. ²⁰ In addition, we previously demonstrated that cancer patients, acting as surrogate decision‐makers, vary considerably in their willingness to choose CMT over RT. ²

In light of the variation in surrogate patients’ preferences revealed by the trade‐off method, we explored the method as part of a decision aid in actual LA‐NSCLC patients facing the treatment choice. The decision aid is intended to improve the decision‐making process by informing patients using a structured process, and by helping them to clarify their values pertaining to the decision. ²¹ The intended improvements are in keeping with the ethical principles underlying informed consent: autonomy, beneficence, and distributive justice. ²² Existing evidence suggests that the current practice of informed consent falls short of its goals ²³ ; thus, improvement in the decision process is necessary. For example, previous work has shown that lung cancer patients often cannot report information central to their condition, ^{24 , 25} suggesting that some patients may not have acted autonomously in giving consent. Our findings with surrogate decision‐makers ² suggesting exercises designed to help patients become clear about their attitudes may result in improving the decision‐making process. In addition, the demonstrated variation in physicians’ patterns of practice emphasizes the need for a systematic method of empowering the patient to ensure that his or her values are considered appropriately.

The purpose of this report is to describe a decision aid designed for patients in this setting, and to report its implementation in a regional cancer centre. The specific primary objectives of the study were: (a) to determine if using the decision aid in the setting of a regional cancer clinic was feasible. That is, to determine if patients wished to participate in their treatment decisions, and if the aid was acceptable to patients and to physicians, and (b) to determine the internal consistency of different components of the decision aid. The secondary objective was to report preliminary data regarding: the consistency of patients’ treatment preferences with their values; the degree to which patients acted with understanding and whether the decision aid improved their understanding of information; and whether patients’ treatment preferences or level of decisional conflict were influenced by using the decision aid.

Methods

Design and utilization of the decision aid

A research associate interviewed each participant individually using a structured format. ²⁶ The interview began by collecting baseline data (as stated below). The two treatment options were then described: RT (5 fractions to the chest) and CMT (induction Cisplatin and vinorelbine plus 30 fractions to the chest). The description of each treatment was divided into seven components: details of the actual treatment regimen; early side‐effects (and their frequencies); late side‐effects (and their frequencies); possible effects of the treatment on personal functioning, emotional state, and social interactions; and symptoms caused by the cancer. The treatment description components were those used in a previous study of surrogate stage III NSCLC patients, ²⁶ and were developed using a process involving interviews with patients and health care providers, blended with available information in the existing quality of life and oncologic literature, as described previously. ²⁶ We did not want to present information that was discordant with that provided by the consultant oncologists. Hence, all material was reviewed and approved by the Centre Tumour Site Group. In particular, clinicians arrived at a consensus view on the survival data that would be presented to patients, based on the meta‐analysis of clinical trials comparing Cisplatin‐based chemotherapy to radiotherapy alone ¹¹ and the results of influential individual trials. ^{12 , 27} Each component was printed on an individual card, which was read and then placed on a display board in front of the patient. For each component, the two treatment descriptions were placed side‐by‐side to facilitate direct treatment comparisons ( Fig. 1, Panel A). Examples of the description components for the early and late side‐effects of each treatment are shown in Table 1. Participants were given the opportunity to ask for clarification at any point during the presentation of information.

Figure 1 A schematic representation of the presentation of the decision aid. Figure 1a shows the set‐up and Fig. 1b shows the detail of the survival information displays used in the two trade‐off exercises. The grey bars in Fig. 1b were displayed in yellow (representing the number of people out of 1.

00 alive at a point in time) and the black bars were displayed in blue (representing the number of people dead at the same point in time).

Table 1.

 Two examples of treatment descriptor cards

Two treatment trade‐off exercises were designed to clarify participants’ values pertaining to survival outcomes in light of the treatment descriptions. One exercise was designed to determine the absolute increase in 3‐year survival probability that the participant would want before choosing CMT; the other was designed to determine the increase in median survival time that the participant would want before choosing CMT. In the exercises, survival information was expressed as a frequency and was explained to participants as the number of people out of 100 who would be alive at a given point in time. ²⁸ The display of the time‐line is shown schematically in Fig. 1a, and details of the trade‐off displays are shown in Fig. 1b.

The trade‐off involving the probability of surviving three years was performed first. At the 3‐year point on the time line, the survival frequency for RT was shown to be 5/100 and that for CMT was shown to be 10/100. This initial difference in the display of three‐year survival frequencies was designed to avoid an anchoring bias that might arise from starting the trade‐off exercise with no difference in survival outcomes. Each participant was asked which treatment they would prefer, given the treatment descriptions and the displayed 3‐year survival probabilities. To determine their trade‐off threshold, the frequency of surviving 3 years with CMT was then systematically increased (if the initial preference was for RT) or decreased (if the initial preference was for CMT) until the participant indicated a change in treatment preference. The survival percentage with CMT at the participant’s point of treatment preference switch is the threshold that represents the minimum survival advantage required by the participant in order to choose the more toxic CMT treatment ²⁶ ( Fig. 1b).

The trade‐off involving the median survival time was then performed. The median survival was illustrated to the participant as the time when 50 of the initial 100 patients had died and 50 remained alive. We explained that this time represented the ‘average’ length of life after treatment, emphasizing that some people live longer than the average and some people live not as long. We further explained that differences in median survival time between treatments were representative of the ‘average’ increase in length of life afforded by the more toxic treatment. The median survival times were initially displayed as 12 months with RT and as 14.5 months with CMT. The participant was asked which treatment they would prefer, given the treatment descriptions and the displayed median survival probabilities. The median survival time with CMT was then adjusted (as described for the 3‐year survival trade‐off) until the participant reached a threshold increase in median survival ( Fig. 1b).

The participant was then told that the initial displays of 3‐year survivals (5% vs. 10%) and of median survival differences (12 months vs. 14.5 months) represented the actual estimated survival benefits of CMT, and these differences were again displayed ( Fig. 1b). The participant was asked if they wished to revise either of the declared preference switch points, given the new knowledge of the actual benefit conferred by CMT. Finally, the participant was asked which of the two survival end points (3‐year survival, or median survival time) was more important to them when making a treatment choice.

For each trade‐off, each participant’s survival advantage threshold (SAT) was calculated as illustrated in Fig. 1b. For the 3‐year trade‐off, the SAT is the percentage survival conferred by CMT minus the 5% survival conferred by RT. For example, if a participant preferred CMT when the survival (with that treatment) was 7%, the SAT for that participant was (7% − 5%)= 2%. The SAT was less than 0% if the participant preferred the more toxic treatment with a survival percentage lower than that offered by RT. Likewise, for the median survival time trade‐off, the SAT is the median survival time conferred by CMT minus that conferred by RT.

Evaluation of the decision aid

Study participants

Between January 1997 and December 1998 consecutive eligible patients at a regional cancer centre were invited to participate. Eligible patients were those with locally advanced, unresectable non‐small cell lung cancer who were candidates for both RT and CMT. Patients with any of the following characteristics, in the opinion of the attending physician, were deemed ineligible: insufficient fluency in English; diagnosis of major affective disorder, at high risk for severe emotional distress and any other condition that placed major limitations on the patient’s ability to understand the content of the interview. Patients were evaluated for study eligibility by the attending oncologist after the initial medical consultation. Eligible patients then met individually with a research associate and were invited to participate. All participants provided written consent prior to commencing the interview.

Integration of the interview into usual clinical practice

Patients meeting the eligibility criteria were invited to participate in the study. It was stressed to the patients that their actual treatment decision would be made with their oncologist. The study interview was conducted on a day after the initial consultation but before a second consultation when the treatment decision was made. The participant was alone with the interviewer in order to limit biases that could occur due to the presence of a family member or friend; participants’ family were invited to attend a debriefing following the interview. A take‐home package containing the treatment description components and a graphical representation of the survival probabilities was provided to all participants. A summary of the interview results was provided to the attending oncologist for those participants who consented to share their information. A follow‐up interview was conducted with the participant following the decision with the oncologist.

Additional study endpoints

The baseline assessment included an evaluation of the participant’s decisional‐role preference, strength of treatment preference, decisional conflict and understanding of the information. The participant chose the decision‐role statement (as listed in Table 2) that best described their preferred role in decision‐making, based on Degner and Sloan’s card sort method. ²⁹ The participant indicated their strength of treatment preference on a 7‐point Likert scale. O’Connor’s Decisional Conflict Scale ³⁰ was used to evaluate participants’ degree of decisional conflict, with particular attention to the decisional‐uncertainty subscale (1=low uncertainty, 5=high uncertainty). We evaluated each participant’s objective understanding of information using four questions of the following format: ‘Out of 100 people treated with radiotherapy alone, how many would be alive after three years?’. Analogous questions targeted the same survival endpoint for CMT and the median survival endpoint for each treatment.

Table 2.

 Participant characteristics and decision role preferences

Physician and patient acceptance of the decision aid was assessed by two questionnaires analogous in design to the decisional conflict scale, employing a balance of positive and negative frames and requesting responses on a 5‐point Likert scale. We asked participants to respond to three questions (listed in Table 3) and asked each patient’s primary attending physician to respond to three corresponding questions (listed in Table 3).

Table 3.

 Participant and physician evaluation of the decision aid

Our experience implementing the study caused us to make two changes to our method. Initially, we intended to test post‐decision aid measures at a second study interview conducted after the decision was made but before treatment started; this strategy proved difficult to integrate into a busy oncology clinic, thus, some post‐interview measures were not obtained on six participants enrolled early in the study period. In addition, events occurring between our initial and follow‐up interviews caused us to be concerned about confounding effects on patient’s understanding of the information presented. Thus, we moved the evaluation of understanding to the end of the main study interview after the sixth study participant.

Evaluation of the study objectives

We evaluated the feasibility of the decision aid based on: the proportion of eligible patients agreeing to use the aid; the proportion of participants who completed all aspects of the interview; the proportion who preferred to participate in decision making; and the proportion of physicians and participants who responded favourably on their respective evaluations of the aid. Ninety‐five per cent confidence limits (95% CI) on these proportions were calculated using the binomial distribution. Our sample size was based on obtaining sufficient experience with the aid to make any required changes in design or evaluation methods, and on obtaining the responses from a sufficient number of participants to demonstrate the feasibility of the aid to other cancer centres interested in studying the intervention.

We evaluated the internal consistency of the aid’s components by comparing each participant’s SAT for 3‐year and median survival to his/her strength of treatment preference. We anticipated an association between participants 3‐year SATs and median survival SATs and calculated a Spearman correlation coefficient (ρ) accordingly. We also anticipated associations between participants’ strengths of treatment preference and their SATs for each of the two time points, and tested each by calculating Spearman correlation coefficients.

We determined if each participant’s expressed treatment preference was consistent with their values by comparing it to their SATs. For example, consistent outcomes would occur when a participant with a 3‐year SAT lower than the actual survival benefit estimated for CMT had a preference for CMT, or, when a participant with a SAT higher than the estimated benefit of CMT had a preference for RT.

We examined if participants acted with understanding, and if the decision aid improved their understanding of information, by comparing participants’ answers to questions about expected survival before and after using the decision aid. We refer to these responses as measures of participants’‘objective’ understanding. We compared these responses to participants’ sense of being informed as determined by item #5 of the Decisional Conflict Scale: ‘I feel I know the benefits of each treatment for my lung cancer’. We refer to these responses as a measure of participants’‘subjective’ understanding.

Finally, we examined the influence of the decision aid on participants’ degree of decisional conflict, and on their treatment preferences, by determining decisional conflict scores and strength of treatment preferences scores, respectively, before and after using the decision aid.

Results

Feasibility

Twenty‐five patients met the eligibility criteria for the study and all 25 were invited to participate. Seven (28%) patients declined to participate in the study because they had already developed a clear preference; four chose CMT and three chose RT. The remaining 18 (72%) patients agreed to participate; 16 chose CMT and two chose RT. Characteristics of participants are listed in Table 2. All participants completed the interview, which typically required 1 hour. Participants completed all tasks with the exception of one patient who could not declare a threshold for median survival time. All participants wanted a role in making the treatment decision, but they held a variety of decisional‐role preferences, as shown in Table 2.

All participants had favourable or neutral responses to the three questions addressing their evaluation of the decision aid, as shown in Table 3. For all but one participant, physicians had favourable or neutral responses to the three questions addressing their evaluation of the decision aid (shown in Table 3). For one patient with a strong preference for CMT, the physician ‘disagreed’ with the statement indicating the aid helped the participant understand the risks and benefits of each treatment.

Patients’ survival advantage thresholds and internal consistency

Participants’ initial 3‐year SATs ranged from choosing the toxic treatment (CMT) if it offered a survival frequency lower than that of RT (SAT=−1) to choosing CMT only if it offered a 55% survival advantage over the less toxic treatment (RT). Fourteen participants (78%, 95% CI=[52%–94%]) would choose CMT for an absolute 3‐year survival advantage of 5% (the actual estimated benefit). After disclosure of the actual estimated benefit, one participant revised his 3‐year SAT (from 15% to 40%) and no other participant revised his/her threshold values. Participants’ initial median SATs were also distributed over a range of values ranging from zero to 24 months. Twelve participants (67%, 95% CI=[41%–87%]) would choose CMT for a median survival advantage of 10 weeks (the actual estimated benefit). No participant revised their median SAT after disclosure of the actual estimated benefit. Ten participants felt that the 3‐year survival end point was more important than the median survival end point, eight participants felt the end points were of equal importance, and no participant felt the median survival end point was more important.

Seventeen patients were able to state thresholds for both time points and indicate a treatment preference. As anticipated, the median and 3‐year SATs were associated (ρ(15)=0.73, P=0.001). The relationship between each participant’s final 3‐year SAT and strength of treatment preference is shown in Fig. 2a. Across participants, the two variables were associated (ρ(15)=0.80, P < 0.001). Figure 2b, shows the relationship between each participant’s final median SAT and strength of treatment preference. Again, across participants, the two variables were associated (ρ(15)=0.77, P < 0.001).

Figure 2 3‐year ( Fig. 2a) and median ( Fig. 2.

b) survival advantage thresholds as a function of patients’ treatment preferences. The shaded quadrants represent survival threshold values that are discordant with the patient’s stated treatment preference.

Patients’ survival advantage thresholds and their treatment preferences

As Fig. 2 illustrates, most participants had SATs that were consistent with their treatment preferences. With regard to 3‐year SATs, 16 participants had SATs that were consistent with their preference. Of the two participants who indicated discordant 3‐year SATs, one had a median SAT that was consistent with his preference. He indicated that both end points were important to his decision. For the other participant, both his SATs were discordant with his preference for CMT. With regard to the median SATs, 10 participants indicated median SATs that were consistent with their treatment preference. Of the four participants with discordant median SATs, three indicated 3‐year SATs that were consistent with their preference and further indicated that this was the more important end point to their decision. The fourth participant was the patient with both SATs discordant with his treatment preference.

Patients’ understanding of survival information

Thirteen participants completed the assessments of ‘objective’ and ‘subjective’ understanding, before and after using the decision aid. Table 4 summarizes participants’ objective understanding of the two survival outcomes. Prior to using the aid, only two participants (15%) knew the frequency of survival with RT alone (within 5%), and only one of these knew the median survival with RT alone. No participant knew the magnitude of benefit conferred by CMT, but five knew that it offered improved 3‐year survival. As the table shows, after using the aid, most participants knew the actual estimated 3‐year and median survival with RT and the magnitude of benefit conferred by CMT. After using the aid, most patients reported a low score (reflective of good ‘subjective’ understanding) on the Decisional Conflict subscale (mean=2.2, range 1–5).

Table 4.

 Patients’ knowledge of treatment outcomes before and after the decision aid interview

Influence of the decision aid on treatment preferences and decisional conflict

The influence of the aid on participants’ treatment preferences is illustrated in Fig. 3. The concentric circles in the figure illustrate the 10 participants who had no change in their preference after using the aid. For three participants with no initial preference, all developed a preference for CMT. For all participants expressing a change of treatment preference, the change was toward a stronger preference for one of the treatments. For the 13 participants on whom post‐aid decisional uncertainty scores were measured, one (8%) reported an increase in score and the remaining 11 (92%) reported lower uncertainty after using the aid.

Figure 3.

 Change in patients’ strength of treatment preference from before to after the decision aid. The larger grey circle is the patient’s pre‐decision aid strength of preference and the smaller black circle is the post‐decision aid preference. All changes were to a stronger preference.

Discussion

The treatment trade‐off approach to decision support offers a method for the structured presentation of the treatment options, and their respective risks and benefits. It provides patients with the opportunity to integrate this information and to clarify their personal values regarding the risks of CMT in light of its benefits. In this report, we have shown that the aid can be successfully implemented in a multidisciplinary clinic, that many patients desire a role in making the treatment decision, that patients can complete the required tasks, that patients seem to benefit from using the aid, and that both patients and their physicians find the aid to be acceptable. While the small study sample size limits the conclusions that can be drawn regarding the overall effectiveness of the aid, the preliminary findings reported here suggest that further evaluation of the aid in a larger confirmatory study would be appropriate.

The variation among patients’ survival advantage thresholds observed in this pilot study highlights the importance of patients participating in decisions that are sensitive to their values. The variation in preferences that we observed is consistent with that illustrated by similar studies in other oncological contexts. ^{2 , 31} , 32, ^–33 The variation suggests that guidelines addressing the use of CMT in this setting ought to acknowledge the value‐sensitive nature of the treatment choice, ^{1 , 34} and thus, that methods to facilitate decision support may be of value.

The treatment decision in locally advanced NSCLC represents a situation in which patients must judge the acceptability of CMT under conditions of irreducible uncertainty. ³⁵ The trade‐off method may be considered a method on a continuum between the prescriptive approach of decision analysis at one end, and intuitive decision‐making at the other. In this sense, the trade‐off approach is analogous to Hammond’s adaptation of the ‘lens model’ ³⁶ in that it offers a structured process encouraging the patient to integrate the indicators available to them. The structured presentation allows for rational integration, but the judgement also allows for intuition. Admittedly, the judgements may be subject to heuristics and bias ³⁷ and are compromised by the same missing information about health states that limits the assessment of utilities. The strength of the method, however, is that it provides a structured process for presenting the best available evidence to the patient, and uses a similarly structured exercise to encourage the patient to carefully weigh his or her values pertaining to that evidence.

We do not know, on the basis of the present findings, whether the aid improves decision‐making over less intensive approaches; clearly studies comparing two or more approaches will be required to address this question. The three patients with no initial treatment preference may have gained the most benefit from using the aid, as they all developed a preference after the interview. An overview of decision aid research suggests that the patients who benefit most from decision support are those patients with decisional uncertainty prior to using the support. ³⁸ Other patients in our study may have benefited by consolidating their initial treatment preference or reducing their initial decisional conflict, as our results suggest. We have shown that patients’ expressed thresholds for accepting CMT were consistent with their choices, and that most patients could recall the direction and magnitude of survival differences between treatments; it would therefore appear that some elements of autonomous decision‐making were met in patients using the aid. Patients also reported high levels of subjective understanding, reporting that they felt they understood the risks and benefits of treatment. These findings, taken together, imply an improvement in decision‐making that requires confirmation in a larger study.

Explicit documentation of patients’ objective understanding revealed that some patients could not recall survival information after completing the aid. It is possible that some patients could not grasp the information, or that they forgot the exact details after the interview was completed. It is also possible that some patients could not recall the survival information because their treatment preferences were based on considerations other than the potential survival benefit.

Our findings that participants’ survival advantage thresholds were highly correlated with their stated treatment preferences provides evidence of convergent validity of the study measures. We note that both correlations remained significant (at P= 0.01) when the one patient with a strong preference for radiotherapy was excluded, indicating that the statistical significance of the results were not dependent on the extreme response. Moreover, discordance between participants’ survival thresholds and their stated treatment preferences were explained, in most cases, by that survival end point being deemed by the participant to be less important to their decision.

In this study we did not intend to systematically ascertain which elements of the treatment descriptor components were most salient to patients making this decision. We recognize that many different attributes of the treatment choices may influence patients choices. In a decision aid using a trade‐off approach, many different attributes with probabilistic outcomes, such as the risk of severe toxicity, or the risk of hair loss and so on, might be used in the trade‐off exercises. We chose to focus on survival outcomes because: few randomized clinical trials have reported symptom response rates ³⁹ or quality of life evaluations; ⁴⁰ meta‐analyses of clinical trials have focused exclusively on survival outcomes; and existing treatment guidelines tend to focus on survival benefits of combined modality therapy. ^{1 , 13} In contrast, Silvestri reported that over two‐thirds of surrogate lung cancer patients would prefer chemotherapy if it substantially reduced symptoms without prolonging life, ³³ and Davidson ⁴¹ demonstrated patients’ need for information on quality of life and symptom response in the setting of locally advanced NSCLC. A larger study is required to research the influence of these clinical end points on actual patients’ decisions. The internal consistency of the SAT scores, and the correlation of these scores with treatment preferences, provide indirect evidence that survival outcomes are relevant to many patients.

Some limitations in the interpretation of our findings are noteworthy. First, some early study patients did not have the opportunity to have post‐decision measures evaluated, hence the before‐ and after‐interview outcomes are based on a subset of the enrolled patients. In addition, moving the post‐decision aid measures closer to the time of interview may result in patients reporting short‐term recall rather than durable understanding. Nevertheless, our goal was to assess patients’ ability to understand the information provided in the interview, even if limited to the short time around decision‐making. Second, we used survival data based on a meta‐analysis of Cisplatin‐based polychemotherapy, which some clinicians consider outdated. New combined‐modality regimens promise higher response rates and lower toxicity, ³⁹ but there is no reason to believe that all patients would willingly accept such treatment. Third, our use of median survival as a representation of ‘average’ benefit may be challenged. We note that while life‐expectancy, in contrast to median survival, provides an arithmetic average of survival time after treatment, we chose not to display life‐expectancy information because the clinical trials typically report median survival differences between treatment regimens, and because physicians are more familiar with the median survival end point. Fourth, we did not examine the effect of altering the frame used in describing the treatment outcomes (for example, the probability of death, or the probability of not developing side‐effects) because, while we assume that a framing effect could be demonstrated, ^{42 , 43} the frames used in our treatment descriptions represent the conventional way clinicians would present the information to patients. Fifth, in gathering information on the treatments for the descriptions, it became evident that there is a paucity of information on certain aspects of the quality of life of NSCLC patients undergoing these therapies. Therefore, not all information used in the treatment descriptions could not be based on methodologically rigorous clinical studies, and some was based instead on clinician interviews and survey responses. Sixth, we did not examine the possibility that an explicit display of the uncertainty in treatment descriptions (for example, the variance of survival outcomes) would influence patients’ willingness to accept chemotherapy differently than would a display of the point estimates of survival, as was used in this study. Finally, we do not know to what extent our pilot findings might be generalized to all patients eligible for combined‐modality therapy, since some patients were excluded on the basis of the clinicians’ judgements, or on their declining of the interview. A larger study in other cancer centres is now underway in an effort to address these issues.

We conclude that a decision‐support strategy for patients with locally advanced NSCLC is feasible, that it demonstrates convergent validity, and that it is favourably evaluated by patients and their physicians. The aid seems to help patients understand the benefits and risks of treatment and to choose the treatment that is most consistent with their values. In doing so, it appears to improve the process of informed consent. In addition, it may reduce decisional conflict and increase patients’ satisfaction with their decision. Further evaluation of the aid is warranted.

Acknowledgements

The authors wish to acknowledge the patients who participated in the study, Ms Roxanne Cosby and Ms Judith Davidson for their contributions in the early stages of this study, and the anonymous reviewers for their helpful comments on an earlier draft of this manuscript.

This study was supported in part by a grant from the National Cancer Institute of Canada and the Canadian Cancer Society.