A larger coefficient indicates a greater impact on preferences, a positive coefficient indicates more preferred/acceptable attributes, and a negative coefficient indicates less preferred/acceptable attributes.
aModel fit: log likelihood, −16 422.35; χ2 = 44 172.28 (P < .001, 65 df); pseudo r2 = 0.573; Akaike information criteria, 1.534.
eTable. Attributes and Levels
eAppendix 1. Example of scenario shown to participants in the cancer frame
eAppendix 2. Technical appendix – Analysis
eAppendix 3. Willingness to pay
Customize your JAMA Network experience by selecting one or more topics from the list below.
Nickel B, Howard K, Brito JP, Barratt A, Moynihan R, McCaffery K. Association of Preferences for Papillary Thyroid Cancer Treatment With Disease Terminology: A Discrete Choice Experiment. JAMA Otolaryngol Head Neck Surg. 2018;144(10):887–896. doi:10.1001/jamaoto.2018.1694
Which factors, including cancer or noncancer terminology, influence preferences and trade-offs for treatments for papillary thyroid cancer, including surgery and active surveillance?
Individuals were willing to tolerate more severe adverse effects if the term “cancer” was used to describe the condition, and low health literacy was found to be associated with a lower preference for less invasive treatments.
Health literacy plays an important role in treatment preferences, which may contribute to health inequalities; changing the terminology of papillary thyroid cancer may be 1 strategy to help reduce overall preferences for more invasive treatment options and patient willingness to endure more serious unnecessary adverse effects from treatments.
Given recent evidence of overdiagnosis and overtreatment of small papillary thyroid cancers (PTCs) and other low-risk cancers, strategies are needed to help patients consider less invasive treatment options.
To determine which factors influence treatment preferences for patients with PTC, and the trade-offs in treatment factors people are willing to accept, and to understand how terminology influences preferences and benefit-harm trade-offs.
Design, Setting, and Participants
Preferences in PTC treatment were evaluated using a discrete choice experiment (DCE) conducted as a web-based survey using an existing public online research panel. Participants were randomized to receive 1 of 2 frames of information based on the terminology used to describe the condition: “cancer” or “lesion.” Participants chose between 3 treatment options for PTC (thyroidectomy, hemithyroidectomy, and active surveillance). Analyses were conducted using a mixed logit model.
Main Outcomes and Measures
The main outcome variable was treatment preference; attributes of treatment options and sociodemographic characteristics were explanatory variables.
The DCE was completed by 2054 participants (993 [48.3%] men and 1061 [51.7%] women; mean [SD] age, 46.0 [16.5] years) with no history of thyroid cancer. Participants preferred options with less frequent follow-up, lower out-of-pocket costs, lower chances of having voice and calcium level problems, and a lower risk of developing invasive thyroid cancer and of dying of thyroid cancer. When trading benefits against harms, participants were willing to accept a higher number of extra patients experiencing adverse effects to avoid a thyroid cancer death when the condition was described as a cancer compared with a lesion. Specifically, participants for whom the condition was described as a cancer were willing to accept more patients requiring lifelong medication (mean, 273; 95% CI, 207-339 vs mean, 98; 95% CI, 66-131), experiencing calcium problems (mean, 110; 95% CI, 77-144 vs mean, 56; 95% CI, 55-58), and fatigue (mean, 958; 95% CI, 691-1224 vs mean, 469; 95% CI, 375-564). For both the cancer and lesion terminology, health literacy consistently was associated with preferences for treatment options. Those with lower health literacy had a significantly lower preference for less invasive treatment options.
Conclusions and Relevance
This study makes an important contribution to understanding how attributes of treatment options, terminology, and patient characteristics, in particular health literacy, influence treatment decision making for PTC. As a result of increasing evidence of the indolent nature of PTC and other low-risk cancers, strategies to deal with potential overtreatment are critically needed.
Australian New Zealand Clinical Trials Registry: ACTRN12617000066381
Thyroid cancer is the most common endocrine malignant abnormality. In recent years the incidence of thyroid cancer has substantially increased worldwide,1 which has largely been driven by the increase in the detection of papillary thyroid cancers (PTCs).2 Screening and autopsy studies indicate that asymptomatic PTCs are present in more than 10% of the adult population,3 and with the surge in the use of diagnostic ultrasound, other imaging modalities, and fine-needle aspiration, these cancers are now more frequently identified.4 As a result of this, overdiagnosis and subsequent unnecessary treatment has now been recognized to occur in patients with PTC.5
Surgery, including total or partial removal of the thyroid, is the mainstream treatment for all biopsy-proven PTCs, although Japanese studies have demonstrated that active surveillance can be an effective treatment with comparable rates of growth, metastases, and progression to invasive cancer.6,7 Active surveillance can reduce the possibility of overtreatment of PTC, which is particularly important because thyroid surgery is invasive and has the potential to result in complications or adverse effects.8,9 Furthermore, total removal of the thyroid involves the reliance on lifelong thyroid replacement medication, with those who have had their thyroid partially removed having more than a 30% chance of also requiring medication.10,11 Although active surveillance studies are beginning to be acknowledged in clinical practice guidelines, there are still strong recommendations and clinician preferences for surgical intervention.12,13 At this time, patient treatment preferences for PTC are largely unknown and hard to test because most patients are still only being recommended surgical interventions.13
Preferences about treatment are individual and can be driven by personal beliefs and experiences as well as systemic influences. In PTC, there may be a number of different factors associated with treatment options that are important to patients and could affect their decision making (eg, need for lifelong thyroid replacement medication, adverse effects or complications of surgery, risk of mortality, etc). Another factor that has been shown to influence treatment preference is the terminology used to describe the condition.14 In 2 experimental studies that tested how terminology impacted treatment preferences for ductal carcinoma in situ (DCIS), it was shown that removing the term “cancer” in descriptions of the condition lead to greater preferences for nonsurgical treatments.15,16 Ductal carcinoma in situ is a condition similar to small PTCs in that there is only a small possibility that it will cause harm if left undetected and untreated. In both conditions, there have been calls to change the terminology by removing the “cancer” term because this may more appropriately convey their indolence and, in turn, make it easier for patients to consider and choose less aggressive interventions for these tumors.4,17
There is currently little evidence on the impact of changing the terminology for PTC or on what trade-offs between benefits and downsides patients are willing to accept with their treatment choice. The proposed experimental study aims to: (1) determine which factors influence treatment decisions for PTC, (2) determine the trade-offs in treatment attributes (ie, characteristics of the treatment options) people are willing to accept, and (3) understand how terminology (cancer terminology vs noncancer terminology) influences preferences and trade-offs.
Providing a better understanding of community preferences for PTC treatment and the impact of terminology can help improve shared decision making between clinicians and patients and can provide evidence on how terminology may impact treatment decisions. This information may help inform future strategies to reduce the potential for overtreatment for this low-risk condition and ensure that treatments offered are aligned with community and patient preferences.
Papillary thyroid cancer treatment preferences were evaluated using a discrete choice experiment (DCE),18,19 a quantitative technique that is underpinned by strong theory and based on the premise that a health care good/service/intervention can be described by its characteristics or attributes. In a DCE, levels of each attribute (chances of potential adverse effects, for example) presented to participants are varied in a series of questions and respondents choose the option that they prefer for each question. Discrete choice experiments determine which attributes are driving preferences and the trade-offs between attributes that people are willing to accept.
The DCE presented participants with a series of questions, asking them to choose between 3 treatment options for PTC: full surgery (thyroidectomy), partial surgery (hemithyroidectomy), and monitoring (active surveillance). The primary outcome measure was treatment preference.
Participants were randomized to receive 1 of 2 frames of information based on the terminology used to describe the condition: “papillary thyroid cancer” (frame 1) or “papillary thyroid lesion” (frame 2). The 2 frames were identical, other than where the term for the condition is used.
The study was registered with the Australian New Zealand Clinical Trials Registry (ACTRN12617000066381), and ethical approval was obtained from the University of Sydney Human Ethics Research Committee.
The DCE was conducted as a web-based survey using an existing general public online research panel that was administered by an independent external organization (Survey Sampling International). This organization maintains an extensive database of over 100 000 Australians who have indicated their willingness to be involved in online survey research in return for points that go toward a small reward (eg, coupons or discounts). From their panels, Survey Sampling International alerted men and women aged 18 years and older that there was a survey available to them. Interested participants who had been previously diagnosed or treated for thyroid cancer were excluded from participating. Survey Sampling International used quota sampling to ensure representativeness in response by sex, age-group, and Australian state in which the participants lived.
Attributes were based on published literature, and qualitative interviews conducted by study authors with 25 patients recently diagnosed with small papillary thyroid cancers.20 The 8 attributes used in this study to describe PTC treatment options included: chance of requiring lifelong thyroid replacement medication,10,11 follow-up required,12 out-of-pocket costs,21 chance of having problems with your voice,8,9 chance of having problems with your calcium levels,8,9 chance of suffering from tiredness,22 chance of invasive cancer, and chance of dying in the next 20 years6,7 (eTable in the Supplement).
All of the attributes were presented as simple percentages, with a natural frequency also presented for “chance of dying in the next 20 years” because the denominator was 1000.23,24 Out-of-pocket costs were presented as the amount not covered by Medicare that participants would personally need to pay over the next 10 years for any clinician visits, tests, and/or treatments, and follow-up was presented in number of months or years.
An initial DCE design was tested in a pilot study to assess understanding and comprehension of attributes, using a convenience sample. Ten participants were asked to complete the experiment to provide lay feedback in terms of understanding, formatting, or text changes. Following any language modifications, a second online pilot was conducted with 203 respondents recruited through Survey Sampling International to collect response data to calculate prior parameter estimates with which to inform the design of the main study and slight modifications to attributes levels were made. Based on the model results from the pilot study, a Bayesian d efficient DCE design was created using NGENE software (d error, 0.0013; s estimate, 157). The full design contained 60 questions, each comparing 3 alternative treatment strategies (full surgery, partial surgery, and monitoring); each participant saw 1 block of 12 questions. Participants were randomly allocated to see questions framed as either cancer or lesion.
Before participants began the DCE they were given lay definitions of each attribute and specifically told that although each question may look similar, each is slightly different and to try and think about what treatment factors are most important to them. Participants were then presented with an example question (eAppendix 1 in the Supplement). In designing the experiment, all attributes and their definitions were discussed with the entire study team which included a thyroid cancer expert (J.P.B.).
Respondents were then presented with 12 DCE questions, showing 3 treatment choices: full surgery, partial surgery, and monitoring. Following the DCE questions, participants were asked additional sociodemographic questions including education, employment, income, relationship status, health insurance, language, personal history of being diagnosed with a thyroid nodule, personal history of cancer (not including thyroid cancer), and family history of cancer. Additional characteristics including participants’ self-rated general health, cancer worry,25 anxiety,26 health literacy27 and numeracy28 were also measured.
Analyses were conducted using a mixed logit model using 2000 Halton draws with a panel specification to account for correlated choices within an individual.18,29,30 All attributes were specified as random and coded as linear variables with normal distributions. Respondent demographic characteristics were specified as alternative-specific, nonrandom categorical variables and were effects coded. We examined interactions between the information frame presented and the attributes, between attributes themselves, and between attributes and population characteristics (eg, age, sex, education) before deciding on a final model specification. Results are reported as β coefficients and 95% CIs. A positive β coefficient indicates an alternative is more preferred as the attribute level increases; a negative coefficient indicates an alternative is less preferred as the attribute level increases. The base alternative was full surgery. Goodness of fit was examined using pseudo r2 and AIC (Akaike information criteria) with the log likelihood = −16 422; pseudo r2 = 0.5735; AIC, 1.534; 2054 respondents (24 648 observations). All analyses used NLOGIT statistical software (version 5.0, Econometric Software, http://www.limdep.com/products/nlogit/). Benefit-harm trade-offs between attributes were calculated from individual-specific parameters as willingness to pay and willingness to accept additional harms to avoid 1 invasive thyroid cancer, or 1 thyroid cancer death. Technical details of the analyses are presented in the supplementary materials (eAppendix 2 in the Supplement).
The DCE was completed by 2054 participants (993 [48.3%] men and 1061 [51.7%] women aged 18 years and older with no history of thyroid cancer. Participants were randomized to the cancer terminology frame (n = 1029) or lesion terminology frame (n = 1025). Participant’s characteristics are presented in Table 1. Across the sample, participants were broadly representative of the Australian public in terms of age, sex, and state of residence. Overall, participants had slightly higher levels of educational attainments than the general Australian public; however, they had comparable household incomes and relationships status.36 There was no difference in participant characteristics by terminology frame.
All attributes significantly influenced preferences for treatment options (Table 2 and Figure 1), except the chance of experiencing fatigue. Attributes also behaved in the a priori expected directions, for example as the chance of needing lifelong medication increased, we expected that there would be a lower preference for that treatment option.
Overall, a treatment option was less preferred when the chance of needing lifelong medication increased; when the chance of having voice or calcium problems increased; and when patient out of pocket costs were higher. A treatment option was also preferred when there was a longer duration between follow-up visits. Unsurprisingly, as both the risk of developing invasive thyroid cancer, and the risk of dying from papillary thyroid cancer increased, a treatment option became less preferred. The chance of experiencing fatigue did not influence respondent preferences for treatment of PTC (Table 2 and Figure 1).
Regardless of whether the term “cancer” or “lesion” was used, all attributes influenced preferences in the expected directions (Table 2). For the most part, the terminology frame did not significantly impact the relative influence of attributes on treatment preferences. The exceptions were the chance of needing lifelong thyroid medication, out-of-pocket costs, and risk of developing invasive thyroid cancer. For these attributes, the use of the word “lesion” resulted in significantly lower β coefficients, meaning these harms had a significantly greater negative influence on treatment preferences compared with when the cancer terminology was used (Table 2 and Figure 1).
The influence of sociodemographic characteristics on treatment preferences is also presented in Table 2. For use of both the “lesion” and “cancer” terms, the influence of health literacy consistently impacted preferences for treatment options. Those with lower health literacy were significantly less likely to prefer less invasive treatment options (hemithyroidectomy and active surveillance compared with total thyroidecomy). Other sociodemographic characteristics (English language, employment, income, and health insurance) had variable impact on treatment preferences (Table 2 and Figure 2).
Benefit harm trade-offs were calculated as the willingness to accept extra people experiencing harms to avoid (1) an additional person with invasive thyroid cancer and (2) an additional person dying from thyroid cancer (Table 3). Although an increased chance of harms were associated with lower preferences for a treatment option, respondents were willing to accept more people experiencing harm to avoid thyroid cancer diagnoses and deaths.
There was a difference across terminology frames in trade-offs participants would be willing to accept. Overall, respondents were willing to accept more harms when the term “cancer” was used compared with the term “lesion” (Table 3), and this was significant for trade-offs to avoid thyroid cancer deaths. Participants were willing to accept a higher number of additional people experiencing fatigue to avoid 1 person with an invasive thyroid cancer diagnosis and 1 person dying from thyroid cancer than any of the other harms (lifelong mediation, voice problems, calcium problems) across both the cancer and lesion terminology frames (Table 3).
Willingness to pay results also varied with terminology frame and are presented in eAppendix 3 in the Supplement.
When comparing terminology frames, people were more willing to accept potential treatment harms when the condition was described using the term “cancer.” Those participants for whom the condition was described as a lesion judged the need for lifelong thyroid replacement medication, higher costs, and a higher risk of developing invasive thyroid cancer more negatively compared with those who received the cancer frame. This indicates that when the condition was described as a lesion, participants’ acceptance of these specific treatment attributes was lower compared with when the condition was described as a cancer. When the condition was described as a cancer, there was a lower preference for active surveillance and when the condition was described as a lesion, there was a higher preference for hemithyroidectomy. These findings are congruent with previous literature that demonstrates that the use of the cancer term may elicit higher patient preferences for more invasive treatment options.15,16
Importantly, participants’ level of health literacy also impacted preferences for treatment options. This is in line with a growing body of literature demonstrating that health literacy influences patients’ attitudes toward shared decision making and their underlying preferences and understanding during health-related decisions.37 However, for the first time we found an association of health literacy with treatment preferences. Across both terminology frames participants with lower health literacy (compared with higher health literacy) had significantly lower preference for less invasive treatment options (hemithyroidectomy and active surveillance). This novel finding indicates that regardless of the terminology used to describe a condition, health literacy may play an important role in patient’s ability to individually weigh potential benefits and harms of treatment options. This emphasizes the importance of the need for clear communication about treatment options and their associated attributes to patients, so that they can make a decision that fits their preferences.
When trading benefits against harms, participants were willing to experience higher levels of fatigue compared with other harms. One possible explanation for this may be that troublesome fatigue may be difficult for individuals to comprehend and quantify if they have never experienced it, making them less worried about this particular harm, although it has been found to be a common burdensome complaint of patients with thyroid cancer.22,38 Notably, terminology (cancer vs lesion) also made a difference to participants’ willingness to accept harms to avoid 1 person with an invasive cancer or dying from thyroid cancer. Across most attributes, the mean number of people participants were willing to accept experiencing harms was higher in the cancer frame compared with the lesion frame. This demonstrates that overall when the condition is described as cancer participants are more willing to accept more harms from treatment than when it is described as lesion. The terminology did not seem to impact the willingness to pay to the same degree, with estimates varying across both terminology frames.
Taken together, findings suggest that the terminology (cancer vs lesion) used to describe PTC makes a difference to treatment preferences. Although results varied in the degree by which the terminology consistently impacted treatment decision making, we found a difference in participant’s overall treatment preferences, a difference between the value attached to some of the treatment attributes and a difference between the benefit-to-harm trade-offs. These findings provide important evidence to inform the conversation about changing the terminology of low-risk conditions currently labeled as cancer.17 Importantly, the health literacy finding highlights the lack of understanding the consequences that use of the cancer terminology may have on individuals and the need for further education about the real and diverse biology of cancer.
Our study has important limitations. By virtue of this being a community sample and not a sample of patients diagnosed with PTC, actual treatment preferences may differ for patients. However, our sample did allow us to assess preferences for attributes and treatment, and the effects of terminology on individuals who were unbiased by previous knowledge or experience of PTC, as would be the case in a newly diagnosed patients sample. With the DCE design it is not feasible to include every attribute that may be important to every patient. Instead, we ensured that we included the attributes that were most prevalent in the literature and were discussed in qualitative patient interviews.20 With these attributes we used standardised levels of risk for each of the different treatment choices presented, when in reality these levels may vary based on the age of the patient, the patient’s comorbidities, and surgeon experience. Our assessment of health literacy relied on a widely used single-item measure27 that asked respondents to self-report their ease completing medical forms. This measure is effective in identifying adults with low levels of health literacy but has poor discrimination for adults with marginal levels of health literacy. Further research could use more extensive measure of health literacy to examine its influence on preferences of treatment. Last, although Survey Sampling International’s online panel is one of the largest in Australia and we set quotas for age, sex, and state of residence to be broadly representative of the Australian adult population, it is possible that respondents may not be fully representative of patients most at risk of developing PTC.
Overall, this study makes an important contribution to our understanding of how attributes of treatment, terminology, and patient characteristics, in particular health literacy, influence treatment decision making. It is important that patients make decisions that align with their preferences, so clinician understanding of what drives these patient treatment preferences and decision making is fundamental here. As guidelines supporting more conservative treatments for PTC and other cancers continue to emerge12 as a result of continuing evidence of the indolent nature of PTC39 and other cancers,40-42 dealing with potential overtreatment is critically needed. Our finding that patients with lower health literacy consistently had lower preferences for less invasive treatments is significant and worthy of more investigation to ensure inequalities are not inadvertently created. However, changing the terminology of PTC may be 1 strategy to help reduce overall preferences for more invasive treatment options and patient willingness to endure more serious unnecessary adverse effects from treatments.
Corresponding Author: Kirsten McCaffery, PhD, University of Sydney, Edward Ford Bldg (A27), Rm 128B, University of Sydney, NSW 2006, Australia (email@example.com).
Accepted for Publication: July 2, 2018.
Published Online: August 22, 2018. doi:10.1001/jamaoto.2018.1694
Author Contributions: Ms Nickel and Professor Howard had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Nickel, Howard, Brito, Barratt, McCaffery.
Acquisition, analysis, or interpretation of data: Nickel, Howard, Barratt, Moynihan, McCaffery.
Drafting of the manuscript: Nickel, Howard.
Critical revision of the manuscript for important intellectual content: All authors.
Statistical analysis: Nickel, Howard.
Obtained funding: Barratt, McCaffery.
Study supervision: Howard, McCaffery.
Conflict of Interest Disclosures: All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Professor Barratt reports she is chief investigator on a Australian National Health and Medical Research Council (NHMRC) Centre of Research Excellence Grant (1104136). Dr Moynihan and Professor McCaffery report that they are coorganizers of the Preventing Overdiagnosis conferences, and the 2017 Australian National Summer on overdiagnosis. B. Nickel is supported by Sydney Catalyst Research Scholar Award. J. P. Brito is supported by the Karl-Erivan Haub Family Career Development Award in Cancer Research at Mayo Clinic in Rochester, honoring Richard F. Emslander, M.D. R. Moynihan is supported by an NHMRC Early Career Fellowship Grant (1124207). K. McCaffery was supported by an NHMRC fellowship (1029241). No other disclosures are reported.
Funding/Support: The project was supported by an NHMRC Centre of Research Excellence Grant (1104136).
Role of the Funder/Sponsor: NHMRC Centre of Research Excellencehad no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
Meeting Presentation: This paper was presented at the Preventing Overdiagnosis 2018 Meeting; August 22, 2018; Copenhagen, Denmark.
Create a personal account or sign in to: