ISSN: 2332-0761
+44 1300 500008
Research Article - (2014) Volume 2, Issue 1
This paper is a contribution to current controversies on major statistical methodological issues of stated preference studies. Professor Hausman and Professor Carson, both engaged in the litigation of the Exxon Valdez case, recently discussed again three major issues: embedding/scope issue, the existence of hypothetical biases since data are generated in experiments and willingness to pay versus willing to accept. However, their papers mainly targeted contingent valuation approach in willingness to pay studies and no other types of stated preference studies such as conjoint studies. I use a unique application of another type of stated preference study, a reversed conjoint analysis on physicians’ cost sensitivity to patient economics. It aims to show the value of such stated preference study especially with this topic of patient economics, not limited to out of pocket cost or copayment as one of the attributes in a set of products’ characteristics. Investigating the research question of physicians’ preferences and their value judgment also leads to discuss more the statistical contributions of clinical judgment analysis and the mathematical formulae used to assess overall judgment achievements. Bringing new methodological contributions (e.g. using S and D efficiency statistical tests on experimental designs in combination with econometric tests on effective data) can enhance stated preference studies which are increasingly useful in health care research especially with genomic medicine. Such stated preference studies can provide good predictors of revealed preference analysis, since statistical estimates of biases already calculated with collection of micro conjoint data over minimum of 5 to 10 years showed to be reliable.
Keywords: Contingent valuation; Conjoint analysis (reversed conjoint); State preference studies; Revealed preference studies; Predictive models; Clinical judgment; Economics; Sensitivity analysis
Market survey methods are considered reliable and extensively used in marketing sciences and behavioral economics. They provide data on stated preference, Contingent Valuations (CV), and some utility measures. In the field of health sciences, these methods are widely used to help health care decision makers adjust budgets or control of medical expenditures, in the light of different valuations of subpopulations ’preferences. Recently, there has also been a renewed interest in the use of heuristics and judgment analysis to complement rational decision tools in medical decision making (e.g. SMDM)/SJDM joint event (Los Angeles, 2009), in order to be more explicit on what kind of economic information is processed by providers of care and patients, due to different financial pressures or crisis.
Health service research communities, especially in Europe and Japan, have used these procedures since they both face similar demographic trends and critical needs to revise funding mechanisms with the growth of the elderly population. More recently, US scientists have also used more heuristics and stated preference studies to analyze the role of cost on choices in health care. This work has mainly been driven by outcome research societies, in order to remove financial barriers to the adoption of treatments and to optimize clinical strategies for uptake of new medical technologies [1,2].
(1) “The term contingent valuation is used because studies based on these methods represent values that are contingent on a hypothetical market” extract from Chap 19 of economics of regulation and antitrust on valuation of life [3]. This paper addresses two important criticisms of contingent valuation, presented by Professor J Hausman in the Journal of Economic Perspectives and partially addressed by the second paper in the same journal [4]. The criticisms concern the scope and embedding issue and hypothetical and upward biases in answers of surveyed individuals and are mainly targeted at using surveys to measure willingness to pay, WTP. During a meeting with Professor Hausman at MIT in May 2013, these two issues raised on contingent valuation were discussed for conjoint studies instead of WTP studies. This paper continues this discussion which uses research findings from a unique application of “reversed conjoint design” for series of physicians’ cost sensitivity studies in various national health systems and series of prototypical and statistical developments in collaboration with Prof McFadden in the US in order to investigate how and what economics influence physicians ‘choices.
A third critique raised for WTP studies in the Hausman’s paper concerns differences between Willingness to Pay and Willingness to Accept. This is outside the scope of this paper. Since my experience is in conjoint analysis in health care research (as scientific coordinator and Principal Investigator (PI) in international health) in different national health systems, I will use these studies to illustrate the discussion.
The original limitations of contingent valuation methods were discussed by econometricians, developing statistical methods and survey designs for valuation of goods and service in the early 1980’s. Such studies were widely used for evaluation of damage in several sectors of the economy: environment, forestry, transportation, agriculture. As contingent valuation has been often used in litigation, its role in evaluation of damages has been extremely important. Therefore, it is not surprising that the methodologies are under scrutiny, since they provide estimates of losses that may lead to major fines.
Valuation methods and health care
The debate between Hausman and Carson around the Exxon case where both were engaged in the litigation may benefit from an understanding of the debate on the methodological limitations of stated preference studies in Health Care applications. In health care, several options also exist for valuation of services and pricing methodologies. Usually, revealed preference studies are preferred, whenever it is possible. For instance, most American econometricians specialized in pricing of pharmaceuticals and medical services, use hedonic pricing methods [5]. The use of psychology in pricing methodologies has also existed for a long time under the name of psychological pricing in revealed preference studies. However, it has mainly been limited to changes like modifications of numbers that affect differently consumers’ purchasing behaviors (e.g. using 99 instead of 100).
For pricing, conjoint studies are classified in the UN pricing methodologies, under stated preference methodologies like direct contingent valuation methods versus revealed preference studies (including direct and indirect market prices, Huttin [6]. Zweifel, Breyer and Kifmann classified them as a “variant of Discrete Choice Experiments”, dating the origin of conjoint analysis in psychology to the late 60’s [7,8].
Stated preference methods, especially conjoint valuation have been widely used by commercial companies since the early 70’s in marketing, for explicit trade off between products’ attributes, especially for pricing and product marketing policies [8,9]. Like in other industries, they have been very popular in pharmaceutical marketing [2]. Market survey data are used in different market simulators and validated usually with relative market shares since data are not shared between competitors [10].
Other uses of conjoint valuation, with patient or consumer attributes are more recent in Health economics (see for instance, systematic reviews by Ispor task forces, such as Ryan [11]. Economists may use it for elicitation of Willingness to Pay (WTP) [12]. In such a study however, patient economics is limited to “out of pocket cost” or copayment associated to the product. It is one attribute in the tradeoff between product attributes.
In this paper, the value of stated preference studies is shown, especially for this topic of patient economics, not limited to an “out of pocket” cost per medical product. Patients ‘economics is broader. My contribution is also different. The main research question is how physicians’ preferences and their value judgment may be changed because of patient economics and not only product attributes (the approach is called “reversed conjoint”). The application helps to capture additional economic concepts that represent either more complex grouping of billing or financial information related to products and services, or additional economic information processed by physicians during an encounter about financial restraints to comply with a treatment or clinical strategy. Such economic information is not only measured with numbers (like for out of pocket costs), but described in the verbal language of a physician. The conjoint survey is used to obtain explicit measures of the “implicit restraints” that may interfere with physicians’ clinical strategies and clinical guidelines; these measures are called cognitive cost cues. Therefore, the conjoint measurement helps with the elicitation of physicians’ preferences measures with patients’ economic characteristics and not only their clinical profiles, if they are cost sensitive at the point of care.
More complex conjoint designs should include both patient characteristics and product attributes. So far, study designs testing both consumer and product’s characteristic are limited [13]. A continuation of the “reversed conjoint” application on physicians, presented in Table 1, lead for instance to an integration of newness of products (new versus old medicines) and mode of IT equipment for billing/modes of payment (ebilling). It is an application of an adaptation of this kind of design for an health care application [14].
Earlier adopters of IT processes (Ebilling processes and EMR) | Later Adopter of IT processes | |
---|---|---|
Sensitivity to newness (1) New versus old medicines | High (Clinical strategies from insulin only towards oral/insulin combinations and TZDs) | Low |
Sensitivity to interpersonal communication | Low | High |
Table 1: Integration of Physicians’ sensitivity to Newness of medicines and types of Communication Technologies (Examples on diabetes type II drug therapies).
(1) derived for instance from qualitative research such as focus groups or Nominal groups can be expressed in the economic wording of physicians. Companies such as Nvivo, specialized in lexical analysis can also integrate other communication forms.
However, despite this statistical development described in conjoint analysis, the predominant approach of stated preference studies by health economists for patients ‘economics, remain for the moment the Willingness to Pay (WTP) approach, in population studies [15]. The methodologies also usually come from marketing sciences, their use is more recent and less sophisticated than the one used by life sciences companies.
In Health care research, stated preference methods are used for evaluation research. However, health care markets are much regulated; they are surveyed for different types of stakeholders, often driven for major political health care agendas. They use them with very different perspectives and methodological constraints. I was invited, for instance, to review the Ispor Task force report on conjoint studies in Washington at Ispor [16]. This report sets best practices and standards for the use of conjoint analysis for their members. My perspective however with patients’ economics in the study design was considered more as a study from the perspective of a payer (public, private third parties or the patient himself). Different perspectives also mean that the stated preference studies are used for different purposes. For life science companies, the objective is to help the sales of a new technology in an existing or new market (originally, a new brand in the “evoked set of brands” in the physicians’ memory). For governmental agencies and payers, the main objectives are very different. Departments of Justice may use them in anti-discrimination practices of access to health care for minorities and ethnic groups [2,17]. Public and private payers use them for risk adjustment methods in insurance schemes (actuarial methods), since stated preference studies help to measure more variability inside different groups of populations or providers of care (e.g. medical professions). Stated preferences are used to calculate Qualys and for different valuations of medical products and services [18] for WTP based value of Qualys). They complement top down approaches where conventional technology assessment reviews are used (such as cost benefit or cost effectiveness studies) for competing technologies. Advances in genomic medicine and biomedical research in early prevention (e.g. biomarkers) also lead to an increase of such studies, especially if they are used in adaptive technology platform for integration of evidence in development [19] for a systematic review on tests on oncology).
Finally, decisions makers involved in macroeconomic policies and budget forecasts integrate results of calibration methods in revealed preference models. They help to provide interpretations on trends of medical expenditures when traditional econometric forecasting methods (e.g. with effective panel data) cannot explain them. For example, the study on various drug classes in the US for 2 types of employers’ benefit designs showed that trends of expenditures with different tier copayment mechanisms were limited to some drug classes. On other drug classes, such trends cannot be interpreted with pre and post evaluation disease econometrics and could benefit from behavioral modeling and stated revealed modeling. Moreover, if intention data (micro conjoint data, CVM surveys or DCE experiments) are generated over 5 to 10 years at least, they provide reliable estimates of potential systematic biases and are very useful as calibration methods to complement forecasting methods for trends of health care expenditures [20-22].
Before discussing some of the statistical issues related to sources or forms of biases associated to stated preference studies with hypothetical data on markets, I will address first the following question: whether and when judgment analysis is preferable to rational decision analysis? The use of judgment analysis in medical sciences can be justified because of incoherencies either in clinical studies or in markets, such as lack of evidence or poor quality of evidence, inconsistencies in study results for different patient subgroups, discrepancies of efficacy or outcome measures according to various degrees of risks.
This question also reopens a larger controversy between coherence and correspondence theories in judgment analysis, and not only in the medical field. The debate has been reopened in particular, since the September 11, 2001 event in the US in order to explain incoherencies in markets due to broken intelligence services, especially in the Brunswik society, where developments of the mathematical formulation of the Lens ‘model from Brunswik, and the application for medical decision processes [23] have been regularly discussed (www.brunswik.org). Independently of September 11, market incoherencies have Related philosophical discussion on search for truth For interesting readers, an update on the debate between coherence and correspondence theory is provided in the special issue of Judgment and Decision Making [24].
Anyway existed in highly regulated markets (e.g. in European pharmaceutical markets), mainly because of the existence of complex bureaucratic systems and various layers of local, regional, national and European level of decision-making (e.g. higher prices for generics than branded products for some medicines, higher out of pocket prices paid by patients than producer prices).
In medicine, correspondence based approach has been largely predominant in “evidence based medicine” since the early 1990s [25]. However, controversy still exists in medical sciences, since traditional medical research tends to be coherence-oriented and driven by theory [24]. Judgment analysis is used with a coherence approach in clinical research. Judgment studies in medical sciences can provide very good results and can be highly rated with a performance measure such as the Judgment Achievement Rate (Ra).
The Judgment achievement is measured with the Tucker formula:
Where G=Knowledge, Re=environment predictability,
Rs=Consistency, C=Knowledge [26].
For instance, the recent systematic review on judgment studies in four fields of science by Kaufmann [27,28], including 6 clinical studies with an idiographic approach, showed that performance measure (Ra) for some clinical
• See Series of research projects from the think tank: Evaluation Network of European Drug Policies on the website www.endeplux.org.
• For more information on the mathematical formula, see Tucker [26] and Hammond et al. [29] or updated discussions from Brunswik society meeting, Toronto 2010 and the website of the society (www.brunswik.org) tasks could reach 0.7 to 0.9; even if the average performance measure was an overall moderate achievement of 0.40 in comparison with the studies selected from other fields: business, education and psychological sciences (the highest average value was in the sampled business studies with an average rate of 0.50).
For more information on the mathematical formula, see Tucker and Hammond [26], Hurch and Todd [29] or updated discussions from Brunswik society meeting, Toronto, 2010 and the website of the society (www.brunswik.org) The systematic review is based on single judgment tasks, the overall sample includes 1055 persons judging 49 tasks across 31 studies. The sample for the medical field does not include cognitive feedback type of judgment studies. Additional information on the meta-analysis and inclusion of additional studies since 2010 are available on request to E Kaufman, PhD.
In addition, this meta-analysis also included nomothetic studies (aggregated judgment achievement). For instance, for the medical science, it was based on 221 clinical oriented educations (e.g. clinical psychologists) and 258 analyzed judgment achievements for 10 medical tasks. A few examples can illustrate what is called a coherence approach versus the correspondence approach in medicine. In the history of medicine, there are still important disease areas which lead to both approaches. The most cited is for the rhythm management of atrial fibrillation [30]; other examples also exist now in early prevention of genetic testing, such as for genetic tests of breast cancer [31].
The example on Atrial Fibrillation (AF) is the most cited especially in Judgment and decision societies. The Lens model was used in this case because of a controversy on rhythm versus rate control. The first clinical trial using this approach for the trial design was the AFFIRM trial (2002). The randomized trial of 4060 patients with AF patients with high risk of stroke or death was designed to compare rhythm control versus heart rate control and the use of different drug therapies on both arms. It showed that “no presumed benefits of rhythm control were confirmed, such as lower mortality, less hospitalization, less adverse effects”. Following trials also showed “detrimental effects of rhythm control”. For example, results on the use of Digoxin are recently discussed by Murphy for the TIMI study group [32].
Other fields of applications of judgment analysis are also very relevant for health policy, especially judgment analysis in political sciences and law (e.g. Study of the Brazilian congress, Brunswik society, 2010; www.sas.upenn.edu/tetlock)
The second example in the field of early genetic tests for breast cancer has a lot of evidence in development, with late adoption in some foreign countries. It was identified that, in the case of the HER2/neu test, where only 30% of patients overexpressed HER2/neu and were expected to respond to Herceptin, only 40% of those selected patients responded [31]. Judgment research and stated preference studies were integrated to adjust clinical decision point indices with cost modules. This gave an adaptive technology platform (AKP platform) for helping diagnostic decisions whether or not to use such tests. This enhanced evidence based evaluation and identified/detected relevant cost or economic cues that could interfere with the decisions (under NDA between Liebman m Strategic Medicine and Huttin C Endepusresearch in 2009).
“Translational medicine aims to integrate on adaptive platforms, existing evidence based knowledge from clinical trials and ongoing generation or collection and integration of external data” into the knowledge used by physicians in clinical practice for prognostic, diagnostic and treatment decisions [31].
If judgment research is selected to explore critical decision points in clinical practice and interpret unexplained trends in health care, then it becomes relevant to discuss the statistical methodological issues raised by survey designs and types of intentional data for hypothetical markets. As explained in the first section of this paper, I will mainly address arguments from the current controversy on embeddedness/ scope and hypothetical biases between Hausman and Carson, based on the type of “reversed conjoint” design used on the economics and clinical judgment. It originated from the series of pilot studies on physicians’ cost sensitivity in Europe [1] and has been sustained with different research statistical steps under ENDEPUSresearch in the US. The research lines aim to identify what and how economic information is processed by physicians in clinical practice under different types of financial and business constraints coming from patient economics, product economics and the organizations or health systems. In particular, the use of such type of stated preference study can be extremely useful to measure physicians’ preferences and decision shifts to treatment options for patients covered with different cost sharing mechanisms and to use these estimates in revealed preference models of care.
The introduction of judgment research and design of conceptual frameworks that can better enhance the contextualization of the experiment, instead of debating methodologies linked to statistical problems, may contribute to the embedding/scope issue.
In the reversed conjoint design, for example, when patients need to provide an evaluation of what they are willing to pay for new medical services, they often cannot integrate “what it meant” to be in an alternative payment mechanism (e.g. Under a deductible scheme versus a graduated system). For physicians, it was even less realistic, since they did not usually face directly economic choices. Surveyed physicians in various reimbursement and financing systems, could not think in terms of a system of reimbursement they never experienced. The concept of WTP did not apply. However, other economic concepts were identified both for cases of acute and chronic conditions: patient affordability, patient demand for cheaper treatment. In addition, physicians tried to help patients in order to choose a clinical strategy that could work with the patient financial constraint, especially in cases of shared decision making (e.g. Hormone Replacement Therapy (HRT)).
So, the type of valuation method for investigating such economic topics in Medical Decision Making confirms professor Hausman’s argument against WTP but leads to the use of judgment research and more complex designs (representative designs) to explore the complexity of physicians ‘decision making processes when facing patients covered with different cost sharing mechanisms. The conceptual framework of the lens model and its revisited forms, recent developments of clinical judgment approaches and critiques of prospect theory assumptions (e.g. on loss aversion), were used.
The argument on embedding provided by professor Hausman mainly refers to the scope issue and his recommendation to use a scope test to make sure stated preference studies such as WTP are consistent according to different magnitudes of the experiments. (e.g. WTP measures for willingness to pay to clean one lake up to 5 lakes). He developed with Professor Diamond, the Diamond-Hausman test as a consistency test to make the study useful for evaluators [33]. However, Desvouges et al. [34] and Train also reviewed 109 CVM studies on the period 1994-2012 and found many of these studies did not collect the necessary data for such scope test.
Using the example of a “reversed” conjoint study instead of a conjoint study in this paper, it is suggested to also rediscuss more the representative design of the study in addition to statistical tests such as the scope tests.
In judgment research, in order to have a representative study design, two dimensions are usually considered: task familiarity and task congruence (Table 2). They describe what is called the overall judgment context. Task congruence describes whether the task is concrete or abstract. The use of task congruence terminology comes from psychology [35] and Hammond was the first to apply it to clinical judgment tasks [23]. Task familiarity represents a familiar/unfamiliar task.
Task Congruence | |||
---|---|---|---|
Concrete | Abstract | ||
Familiar | A physician has judged a patient without additional insurance in previous cases “ no mutual” : example of physician wording A physician has judged patients with hay fever and asthma complications | B A physician has judged a patient with low income “ the term income is seldom used during a patient encounter” A physician has judged patients with hay fever and asthma complications | |
Task Familiarity | Unfamiliar | C | D |
Table 2: Task Congruence.
For a judgment study in the medical field, the task familiarity can be described as a familiar clinical case or clinical vignette, faced by a physician in his daily practice (e.g. patients with hay fever and asthma complications). Physicians are used to clinical vignettes since their medical education, and the experimenter can do a selection of existing cases. Then task familiarity is not such an issue.
On the contrary, task congruence is critical to develop an information system design that integrates economics into the system in the field of clinical decision making. It is the most relevant dimension to assess the performance of the description of the context for the Other scope tests are proposed in the literature. Interested readers can refer for instance to scope tests in a meta-analysis of contingent valuation studies (EAERE 2008 conference).
Adapted from Cooksey categorization of judgment analysis, for a cost sensitivity analysis of physicians 1996. The closer the wording and the more familiar the task described in a clinical vignette or case, the better the congruence of the frame. The accuracy level in this study measures “physicians–judges” responses to patients’ economics.
In such a judgment study, instead of using a consistency test such as the Diamond-Hausman test, contextual differences of experimental designs, where judges face concrete tasks versus abstract tasks are rated with a consistency index called Rs [26,29]. For instance, in the meta analysis of judgment studies by Karelaiai, the “Rs of studies with abstract tasks had a lower level of judgment consistency than studies with concrete tasks (Rs of 0.76 vs 0.82)”.
In the example of the “reversed conjoint study” on physicians’ cost sensitivity, the clinical vignette including patients ‘economics with various combinations of cost cognitive cues is described in the physician’s own economic wording (e.g. “no mutual” in Table 2). This economic information, identified during the research process helped to explicit the cost cues. They were directly extracted from the transcripts of physicians’ focus groups in each system for clinical cases in acute and chronic conditions. They represented verbal descriptions of the underlying economic concepts of patient affordability and patient demand for cheaper treatment. The table provides an example of an economic narrative in one system, for one of the cues associated with the patient affordability concept: a patient with “no mutual”.
In addition, in this special application of the reversed conjoint for physicians and economics, main differences between study designs also existed between countries (e.g. national characteristics of health care financing systems). So it brought an additional issue, not discussed so far in the controversy, on the translation of common economic concepts into narrative languages close to the practice of physicians embedded in different cultural medical practices and facing patients embedded in different consumption/behavioral habits. For clinical narratives usually, the medical language is well recognized by system developers internationally. The difficulty is in this case, for experimenters to find economic narratives that match a similar meaning in the local wording of physicians in each country. The level of abstraction that allows the same representation of an economic concept, such as patient affordability, via a set of cost cues, in different national systems may impede some local physicians to recognize a familiar clinical case. For instance, even if there is no available statistics on the comparison between the French and the British cost sensitivity analysis of physicians under the endep/biomed project, several differences in the congruence of the frame could explain the differences in the identification of groups of cost sensitive versus non cost sensitive physicians. These include the additional financial constraint in the UK design to differentiate physicians’ practices, such as fund holders versus non fund holders for prescribing budgets or systematic exemptions with age after 65.
Contrary to meteorological cues in environmental research, cues developed in the context of a physician office are easier to describe in the daily practice of a physician and in familiar tasks such as treatment choices [36]. The study results demonstrate that it can lead to good accuracy. Results of the analysis of variance on sampled physicians were statistically significant for cost cues associated to patient demand for cheaper treatment and patient copayment for comedications [2].
Cue is the usual terminology in judgment research instead of attributes or characteristic; it is a word from psychological sciences (cognitive psychology) and aims to identify when there a decision shift in a judge’s intention. In this case, when there is a treatment decision shift because of cost cues. For the interesting reader, this is the definition provided in Cooksey ‘s clinical judgment analysis book: “ any numerical, verbal, graphical, pictorial or other sensory information which is available to a judge for potential use in forming a judgment for a specific case and/or which is available in the ecology for making predictions about the value of a distal criterion”.
Therefore, a reliable metric can be measured based on the verbal description of cues and can potentially complement pricing and billing/financial information on patients ‘economics. The results showed that such type of economic information captures physicians’ sensitivity to patients’ economics that will restraint his compliance to treatment. Additional econometric studies on physicians’ CDC survey also confirms physicians’ cost awareness of patient economics at the point of care [2].
The use of judgment research also help with other series of tests on the efficiency of the study design: the statistical efficiency of experimental conjoint designs is tested with a test called D efficiency test (The test statistics can be found in Lusk and Norwood [37]. This test should also be used and discussed to assess the reliability of state preference studies and should be added in the controversy on statistical limitations of stated preference studies.
For example for the reversed conjoint discussed in this paper, complex reimbursement systems lead to generations of combinations of cost cues. As cost sharing mechanisms are complex, too many possible combinations of cost cognitive cues exist and require the use of quasi experimental and factorial designs to administer the series of clinical cases to physicians. The D efficiency scores on the reversed conjoint studies for physicians cost sensitivity analysis were the following Table 3.
Reversed Conjoint Designs | D efficiency value |
---|---|
Design with no prohibition on 100 physicians | 96.858374 |
Design with prohibition on 101 physicians | 90.976426 |
Results are taken from Iosbook-Part III, physician study.
Table 3: D efficiency scores for “reversed conjoint in endep/biomed project.
This test is also affected by the use of restrictions to exclude some impossible combinations between cost cues, in addition to the problem of too many combinations. Social security codes provide most of the restrictions on national reimbursement systems. Their inclusion in the study leads to introduce statistical prohibitions in the conjoint experimental design. For instance, the combination “high income and supplemental insurance” was excluded; it did not make sense to explore treatment decision shifts for the selection of clinical cases of the experiment. Exemption categories for disease severity also lead to exclude some combinations where no level of cost cue can be associated. If such prohibitions are not well addressed, it is a main reason for hypothetical biases since physicians cannot make choices for impossible combinations of cues in clinical vignettes. The calculation of the D efficiency test was provided with Sawtooth software that includes the procedure to run the test. Several runs of sensitivity of the D efficiency tests were performed with or without the restrictions and with variations in number of cases per physician. In addition to the D efficiency test, the recent review of best practices of the Ispor conjoint task force recommended, in order to increase the overall precision, to also calculate a Response Efficiency test, called S test, to estimate the measurement error of respondents’ inattention to the choice questions or other contextual influences.
If this is beyond the scope of this paper, it however reflects a critical trade off choice on the experimental design between the need to have the maximization of statistical efficiency which requires more answers to trade off questions, versus S efficiency which aims to reduce the number of cases and questions to avoid inattention due to respondents’ fatigue for instance. Physicians are especially well known for being busy and lacking time. A reversed conjoint survey dealing with economic topics is even more difficult to administer and to retain physicians ‘attention. Such a reversed conjoint survey cannot be administered to physicians more than 20 mn per questionnaire (according to best practices of market survey firms).
Several types of biases exist because of the use of micro conjoint data instead of effective data. However, such biases however can be quantified with reliable estimations over time. The experience of Japanese researchers especially before SARS pandemics in Asia helped to anticipate the number of flu shots and adjustments of supply and demand.
This may lead to a discussion with the indices of the Tucker formulae in the future in time. For instance, reliable estimators of biases for demand for drugs in cold care were in the range of 12% and helped to adjust for elasticity measurements [38]. However, such Japanese conjoint surveys were consumer surveys and not physicians’ surveys.
The reversed conjoint study on physicians discussed in this paper also raises a second type of issues associated to biases on hypothetical cases for this type of preference study, not addressed by the use of longitudinal series for reliable estimators for biases. In a comprehensive “reversed conjoint” study, interactive designs cannot be limited to a competition analysis between companies (represented by different sets of product attributes of life science/medical devices companies in a conjoint study); reversed conjoint also means that the perspective of payers (public, private payers and patients) is integrated in addition to competition between products, may be necessary. Biases may be non-systematic or systematic in studies according to the perspectives of economic actors: especially whether it is the perspective of a public health actor or the perspective of industry stakeholders. For instance, patients’ surveys run for the AARP company in the US , with choice experiments for insurance in long term care on various modes of administration, showed that systematic biases may exist between internet and other modes of administration [39]. On the contrary, market surveys for industry, using conjoint designs, usually never face such an issue (one exception was cited in Europe for systematic bias with internet survey from a company, in the Balkans region for women diagnosed with cancer).
AARP is an abbreviation for an American Retail Pharmacy organization
So, types and magnitude of systematic or non-systematic biases in conjoint surveys may vary according to the perspective of an economic stakeholder such as a company for the launch of a new technology, versus a perspective of a sick fund (e.g. conjoint design that address a national representation of a population (e.g. CDC)). Therefore , the issue of systematic or non-systematic biases in some countries, some populations and some diseases may also need to be addressed for physicians’ reversed conjoint studies in different health systems. Biases such as variations in computer literacy of physicians groups or variations between practices in IT security (e.g. disruption due to hacking) may have to be quantified.
This paper aims to add some statistical methodological developments to the current discussion on limitation of stated preference studies. The Diamond Hausman test was one major test to contribute to scaling problems, especially for classification of risks, or external validation of experimental results in laboratory or with limited sets of scenarios.
The discussion over the reversed conjoint study on economics and medical decision making in this paper shows that special applications of state preference studies and conjoint especially, can help with reliable statistical estimators, based on additional tests such as the D efficiency and S scores. When an analysis of variance provides significant statistical results, they should be discussed with the contextual limitations, with overall judgment achievements indices and efficiency indices on task congruence.
The limits are more and more quantified with meta-analysis contributions such as Kaufmann, Karelia or series of systematic reviews from specialized task forces such as the Ispor conjoint taskforce.
A better understanding of contexts is needed especially in genomic medicine for diagnostic and treatment choices where human judgment still prevails in areas with poor evidence or evidence in development. Global trial research, especially in the agenda for Comparative Effectiveness Research (CER), has interesting statistical developments to provide reliable study results, in difficult trials. For instance, trial designers use analysis by blocks, in order to control for modifications of patients ‘health status with several medical events during the trial. Similarly, several modules could be used with sets of cost cognitive cues, in order to analyze how some physicians ‘groups may be influenced in their clinical strategies by patients ‘economics or other financial constraints.
The current evolution in biological sciences and economics of IT imposes a paradigm change in research, especially to integrate more and more process research and evidence in development to enhance decisions. The aim of this paper, taking the opportunity of the controversy between Hausman and Carson [40] around the Exxon case, is to demonstrate the value of stated preference studies with more comprehensive designs and advanced statistical tests or meta-analysis to quantify biases. The reversed conjoint study design, with generation of cost cues on patients’ economics, products ‘economics and elements of health care financing systems that interfere with prognostic, diagnostic and treatment decisions in clinical practice, can help to adjust the social contracts and welfare packages and avoid exclusions of many subpopulations from access to care.
Other examples can be found in Phase IV clinical trial studies where additional data are collected for additional population groups in evidence in development, individual level data at the clinical practice level is critical. Judgment at clinical practice level can complement global or societal levels and top down decisions in insurance listing and clinical guidelines from technology assessment agencies and national agencies. The topic of economics in medical decision making is a sensitive topic. It can lead to serious court decisions against the medical profession (e.g. in Canada, the case at the supreme court of Alberta, about the economic information during a patient encounter on new technologies excluded from the provincial formulary).
Prof C Huttin is grateful to Prof S Felder who invited her to the faculty of business and economics, University of Basel and provided useful comments on different versions of the paper and to Esther Kaufman, PhD from University of Zurich for discussion on the met review of judgment studies in the medical field.
(1) ENDEPUSresearch, Inc, Founder and CEO, www.endepusresearchinc.com
(2) Professor Dr, University Aix Marseille (P3), Doctoral school social sciences and Institute of business and applied economics; also teaching economics of advance genomics at University of Luxembourg, Doctoral School in Systems and Molecular Biomedicine, life science research unit (2012-2013).