J Am Acad Audiol 26:229-246 (2015)

Case Factors Affecting Hearing Aid Recommendations by Hearing Care Professionals DOI: 10.3766/jaaa.26.3.4 Carmine Gioia*t+ Moshe Ben-Akivaf Matilde Kirkegaard*§ Ole J0rgensen** Kasper Jensen* Don Schumtt

Abstract Background: Professional recommendations to patients concerning hearing instrument (HI) technology levels are not currently evidence-based. Pre-fitting parameters have not been proven to be the primary indicators for optimal patient outcome with different HI technology levels. This results in subjective decision­ making as regards the technology level recommendation made by professionals. Purpose: The objective of this study is to gain insight into the decision-making criteria utilized by pro­ fessionals when recommending HI technology levels to hearing-impaired patients. Research Design: A set of patient variables (and their respective levels) was identified by professionals as determinant for their recommendation of His. An experimental design was developed and 21 representative patient cases were generated. The design was based on a contrastive vignette technique according to which different types of vignettes (patient cases) were randomly presented to respondents in an online survey. Based on these patient cases, professionals were asked in the survey to make a treatment recommendation. Study Sample: The online survey was sent to approximately 3,500 professionals from the US, Germany, France, and Italy. The professionals were randomly selected from the databases of Oticon sales companies. The manufacturer sponsoring the survey remained anonymous and was only revealed after completing the survey, if requested by the respondent. The response rate was 20.5%.

Data Collection and Analysis: Data comprised of respondent descriptions and patient case recommen­ dations that were collected from the online survey. A binary logit modeling approach was used to identify the variables that discriminate between the respondents’ recommendations of HI technology levels. Results: The results show that HI technology levels are recommended by professionals based on their per­ ception of the patient’s activity level in life, the level of HI usage for experienced users, their age, and their speech discrimination score. Surprisingly, the patient’s lifestyle as perceived by the hearing care professional, followed by speech discrimination, were the strongest factors in explaining treatment recommendation. An active patient with poor speech discrimination had a 17% chance of being recommended the highest tech­ nology level HI. For a very active patient with good speech discrimination, the probability increases to 68%. Conclusions: The discrepancies in Hi technology level recommendations are not justified by academic research or evidence of optimal patient outcome with a different HI technology level. The paradigm of lifestyle as the significant variable identified in this study is apparently deeply anchored in the mindset of the professional despite the lack of supporting evidence. These results call for a shift in the professional’s technology level recommendation practice, from nonevidence-based to a proven practice that can maxi­ mize patient outcome.

Key Words: Hearing aid, hearing instrument, professionals, technology level, vignette Abbreviations: HI = hearing instrument; SRT = speech reception threshold

'Center for Decision Sciences, Oticon A/S; Smorum, Denmark; -fCivil and Environmental Engineering, Massachusetts Institute of Technology Cambridge, MA; ^Department of Statistics, Copenhagen Business School, Copenhagen, Denmark; §Strategic Management and Globalization Copen­ hagen Business School, Copenhagen, Denmark; "S a le s and Marketing, Oticon A/S, Smorum, Denmark; f+Audiology, Oticon Inc, l\IJ, USA Carmine Gioia, PhD, Center for Decision Sciences, Oticon A/S, Kongebakken 9, 2765 Smorum, Denmark; E-mail: [email protected], [email protected]

229

Jo u rn a l of th e A m erican Academy of Audiology/Volume 26, Number 3, 2015

INTRODUCTION rofessionals who recommend and fit hearingimpaired patients with hearing instrum ents (His) have long known that, to obtain a success­ ful selection of His for their patients, it is necessary to take into account both the audiometric and nonaudiomet­ ric parameters of the individual patient (Pichora-Fuller and Singh, 2006). When a decision is made to recommend His to a patient, the professional must make a decision about which type of HI to recommend. HI manufacturers offer a broad range of products in different styles and price ranges with different levels of technology. New His are often introduced, resulting in the professional having to consider a substantial amount of information and numbers of products before making a decision about the correct rec­ ommendation for the patient in question. This can be a complex choice with many initial factors to consider. Both audiometric and nonaudiometric parameters—such as vision, manual dexterity (Erber, 2003), and vanity—affect both the need for amplification and considerations for the specific HI style. The parameters defining the correct HI technology level selection are less clear. No known studies define any audiometric or nonaudiometric parameters as being significant as prefitting indicators determining patient outcome or satisfaction as a function of HI technology levels. From studies in health care, we know that the treat­ ments offered to patients are often affected by the psycho­ logical heuristics of professionals. Examples of these heuristics include ignoring statistical base rates, the ten­ dency to anchor decisions on the first piece of information offered (Kahneman, 2011), the tendency to stick to the sta­ tus quo (Samuelson and Zeckhauser, 1988), eliminating by aspect when basing a decision on a minimum criterion of weighted attributes and immediately eliminating the options that do not meet the criteria (Tversky, 1972), inde­ cision because of an abundance of information and large choice sets (Schwartz, 2004), stereotype activation (Wheeler and Petty, 2001), and schemas of thoughts and behaviors forming categories on which to base deci­ sions (Chi et al, 1982). This results in biased judgment in decision making, even when no purely medical contra­ indications are presented (Ryynanen et al, 1996; Charny et al, 1989; Arber et al, 2006; Taylor, 2006; Kahneman, 2011). Research shows that the three dominant stereo­ types influencing our immediate impressions are race, gender, and age (Kite et al, 1991; Bargh et al, 1996; Ory et al, 2003; Steffensmeier et al, 1998) and that the age stereotype has become one of the most socially con­ doned and institutionalized forms of prejudice (Nelson, 2005). Schemas are useful in decision making because they allow people to make sense of experiences, form expectations, help make predictions (Beck, 2010), and enhance processing. However, the schemas and automa­ tion working in subjective decision making can also inter­ fere with comprehension and recall (Bartlett, 1932).

P

830

Given th at the prefitting param eters for recommend­ ing different HI technology levels have not been described in any known literature, this study investigated which audiometric and nonaudiometric param eters act as the primary influencers when professionals are deciding which HI technology level to recommend to a patient. We further wished to examine how the psychological effects of subjective decision m aking is affecting how professionals recommend HI technology levels for hearing-im paired patients.

MATERIALS AND METHODS Creating Patient Cases To understand which factors play a role in influenc­ ing how professionals make decisions when recommend­ ing HI feature levels, we designed an online experiment aimed at approximating the professional’s real-life de­ cision m aking when recommending H is to patients in the clinic (B arnett et al, 1994). We constructed a vignette experiment, a survey design th at experimen­ tally controls a set of factors selected by the researcher and put together in a format aimed at mimicking an aspect of reality (Wason et al, 2002; Caro et al, 2010), to present hypothetical patients to professionals that could provide insights about their behavior in realchoice situations (McFadden, 1974). “People respond to seemingly neutral stimuli in ways th at reflect biases in how they process information and make choices based on th is inform ation, and given how power­ ful these processing biases are we should expect th at they emerge when people encounter vignettes that makes vignettes a valuable research tool” (Beck, 2010). We used the type of vignette experiment called contras­ tive vignette techniques (Cavanaugh and Fritzsche, 1985), where the vignette structure is systematically varied based on a specified experimental design. Two profes­ sionals, one with 6 yr and the other with 5 yr work expe­ rience in audiological clinics/retail stores, were asked to list the patient variables (audiometric and nonaudio­ metric) th at they would typically consider when recom­ mending His. To avoid bias, the professionals were not told about the scope of the study. After the list of vari­ ables was created, 25 professionals (with work experience between 2-25 yr in audiological clinics and retail stores) from the United States, Germany, France, and Italy were interviewed by an experimental design expert (a senior analyst at the Center for Decision Sciences, Oticon A/S) to review the list of variables. After two iterations, the list was finalized based on these inputs. Table 1 lists the approved variables with the respective levels. It was a deliberate choice not to include the patient’s financial sit­ uation among the variables. The financial situation of the patient will normally not be directly available to the pro­ fessional when he or she makes a recommendation in the

C ase F a cto rs A ffectin g H ea rin g A id R ecom m endations/G ioia et al

Table 1. List of Variables and Levels Used for Creating Vignette Patient Cases Patient Case Variables Degree of Hearing Loss Binaural/Monaural Symmetry Type of Hearing Loss SRT (in dB HL) Speech Discrimination (Score in %) Middle Ear Pressure Age Gender Employment Status Marital Status Young (Grand) Children (age 0-10 yr) Sporting Activities Activity Level Problem Context Use of HI Discreteness Concerns Living in Nursing Home Experience with HI (yr) Renewal Extra Features/Devices Tinnitus (Continuous/lntermittent) Vision Impaired Dexterity Problems

'

Levels

# Levels

Mild; Mild to moderate; Moderate; Moderate to severe; Severe; Severe to profound; Profound Binaural; Monaural Symmetrical; Asymmetrical

7 2 2

Conductive; Sensorineural; Mixed Range from 0-100 dB HL; NA Range from 0-100%; NA Range from -100 to + 50 daPa; NA

3 6 6 4

Intervals from 0-100 yr old Male; Female

5 2 5 2 2 2 3

Student; Full-time; Part-time; Retired; Unemployed Single; Married/(in a relationship) Yes; No Yes; No Very active; Active; Not active* One-on-one conversations; Meetings; Party effect (large groups); TV; Phone; NA Every day; Certain situations; Rarely; NA Yes; No Yes; No

6 4

Range from no experience to 30 yr Yes; No

2 2 6 2

FM; Telecoil; Wireless accessories Yes; No Yes; No Yes; No

3 2 2 2

'The variable activity level was constructed by a team of expert audiologists based on the patient case information. It reflects the mental algorithm of perception of activity level as influenced by the patient case material. This well reflects the reality in the professional mindset when encountering a patient. NA = not applicable.

clinic, and the focus of this test is not the patient’s actual purchase, but the professional’s recommendation to the patient. The variables were then mapped to construct the vignette patient cases. When dealing with such complex choices and an immense variety of possible patient cases that could be generated mathematically (i.e., full facto­ rial), the challenge became to generate an experimental design that would reproduce relevant patient cases in order to mimic the professional recommendation choice. One approach was to generate a random fraction from a full-factorial design (Louviere et al, 2000). Although this allows a subset of all possible combinations, respondents should only consider reasonable ones (Wason et al, 2002). Therefore, a team of two professionals (work experience in audiological clinics/retail stores: 6 and 5 yr) and two experimental design experts (one a senior analyst and the other a senior director, Center for Decision Sciences, Oticon A/S) worked on generating a sample of patient cases based on the variables and levels in the table cover­ ing a broad spectrum of actual cases in clinics by mapping the variables. This was done to ensure the inclusion of all key patient case combinations. The patient cases were then pretested with the 25 professionals in the selected countries to assess whether the professionals responding

to our online test would believe the patient cases to be realistic and consistent (Finch, 1987). Patient cases found nonrealistic were changed until approved or removed. In all, 21 patient cases were chosen for this test (see Appendix 1, which contains all 21 patient cases). The survey was administered online. After a pilot launch of the survey with a small sample of respondents (n = 15), followed by interviews with the respondents, we determined that the choice task was feasible in length and the questions understood such that a full-scale survey could be launched with a total sample of professionals. The survey was sent to approximately 3,500 professionals from the United States, Germany, France, and Italy ran­ domly selected. The company sending the survey was kept anonymous and revealed only after completing the survey if requested by the respondent. The respondents were kept anonymous. The response rate was 20.5%. The interviewer was a senior researcher at the Center for Decision Sciences, Oticon A/S. C h oice Task

The steps used in the experiment were the following: A patient case was randomly assigned to a respondent. After reading the case, the respondent would choose

231

Journal o f the American Academy of Audiology/Volume 26, Number 3, 2015

Choice Model

100%! 90%80%70%60%50%40%30%' 20%10% -

o%- 14 6

1 3 7 18 21513105 17198 16129 1122 4 20 Patient case number

■ Technology level 1 ■ Technology level 2

■ Technology level 3 » Technology level 4

Figure 1. HI recommendations for each patient case split in per­ centage between HI technology levels (Levels 1, 2, 3, and 4), ordered in descending order by the amount of Level 1 technology HI recommendations.

We used the Binary Logit modeling approach to iden­ tify the variables th at discriminate between choosing whether or not to recommend Level 1 technology. Con­ sider the Uin as the utility of the respondent n of recom­ mending a solution i (Label i equals 1 for Level 1 technology His and 2 for lower technology His). Each utility can be decomposed into a systematic or deter­ ministic (V) and random (e) part: Um = Vln + The V includes a vector of independent variables xin (variables from patient case characteristics and charac­ teristics of professionals). Following Ben-Akiva and Lerman (1985), the probability th at individual n recom­ mends a Level 1 technology HI (1) is: P„(l) = P(Uln > U2n) = P(Uln - U2n) & 0

from among a list of 14 manufacturers. The respondent could then choose an HI which represented the family, style, and technology level relev an t for the chosen m anufacturer. In all, 217 products were included in th e survey. Respondents were asked to make firstand second-choice recommendations. A set of questions was then asked. This process was repeated for each respondent. After the recommendation experiment, the respondents were asked how often they would see a similar patient case in their clinics. An estimated 51% of the respondents saw the presented patient case very often, 33% saw the patient cases from time to time, and 14% saw the patient cases rarely, whereas only 2% indicated never seeing this type of patient. This validated the usage of the selected patient cases for the analysis. Even if this procedure was already performed qualitatively with a set of professionals from the different countries (n = 25), this step was important to fur­ ther validate the relevance of the patient case presented in the experiment. Recom m endation o f HI Technology Levels We defined four categories of HI feature levels based on the HI m anufacturers’ respective definitions of performance classes; Technology Level 1 being the premium/high-end solution and Technology Level 4 the basic/entry-level solution. These four categories were then used to split the results for technology level recommendations for the 21 patient cases. Figure 1 illustrates the type of recommendation pro­ posed as first choice by th e professionals for each p a tie n t case. The vertical axis illu stra te s th e split between technology level recommendations, and the horizontal axis shows the specific patient case ordered by the number of Technology Level 1 recommendations. There was general agreement on which technology level to recommend to the patient in few of the patient cases. The 21 patient cases are presented in the Appendix.

232

(1)

Assuming th a t the variations of utility differences are independently logistically distributed, the probability that Respondent n chose to recommend the premium alter­ native (1) is given by (Ben-Akiva and Lerman, 1985): prf'x

Pn( 1) =

i» - t - e nP'x2

,1 + e - r f ' ( X l n ~ X2n)

(2)

for linear in param eter utilities, as the param eter p, cannot be distinguished from the overall scale of p’s, we therefore normalize the scale param eter p = 1. The model above is estimated via maximum likelihood (see McFadden,1974 and McFadden, 2002 for a review). RESULTS Survey R espondents The sample included a total of 733 professionals from the United States, Germany, France, and Italy. The analysis is performed on the overall sample for which data were available in the experiment. The United States 100%

90% 80% 70% 60% 50% 40% 30% 20% 10% 0%

2*70%69%69%

[65%&4's, 51%50*.

» s 21% 8%

Patient case number Figure 2. Percentage of Level 1 technology HI recommendations for each patient case, going from the patient case most likely to be recommended a Level 1 technology HI to the patient case least likely to be recommended a Level 1 technology HI.

Case Factors Affecting H earing Aid Recommendations/Gioia et al

Table 2. Binary Logit Model Overview of Estimates, Standard Errors, Test of Significance, and Confidence Interval for the Estimated Parameters Binary Logit Model: list of Independent Variables Age Age2 Lifestyle: Very Active Lifestyle: Active United States Everyday Usage Speech Discrim.: R Speech Discrim.: L Constant

Coef.

SE

Z

P>lz|

0.099 -0.080 1.651 0.319 0.501 1.081 0.014 0.011 -6.664

0.031 0.037 0.483 0.421 0.234 0.358 0.016 0.007 1.556

3.150 -2.180 3.420 0.760 2.150 3.020 0.870 1.600 -4.280

0.002 0.029 0.001 0.449 0.032 0.003 0.383 0.110

95% Cl

0.037 to 0.161 -0.153 to -0.008 0.704 to 2.598 -0.506 to 1.143 0.044 to 0.959 0.379 to 1.782 -0.017 to 0.044 -0.003 to 0.026 0.000 -9.715 to -3.613 Notes: Maximum likelihood estimation results: Log likelihood: -229; LR X2 (8): 94.13; Pseudo R*: 0.17; Prob'>X2: 0.000; n = 411

had the highest representation with 420 respondents, followed by France (148 respondents), Germany (139), and Italy (37). The mean age of the respondents was 44.5 yr (SD = 11.8 yr). Hearing care clinics are often small entities, and the professionals employed in them often have overlapping functions. The respondents in our study had a very rep­ resentative split between profiles (respondents could select more than one profile if relevant): 50% audiologists, 34% fitters, 25% owners, 12% general managers, 12% hearing aid dispenser, 5% other. Most of the respondents’ workplaces were characterized as “independent retail” or “independent clinic” (63%). There were 20% that were characterized as “retail/clinic part of a chain” and 18% as “other.” The mean num ber of hearing aids sold per year a t the respondent’s clinics/retail stores was 248.5 units (the smallest clinics sell

Case factors affecting hearing aid recommendations by hearing care professionals.

Professional recommendations to patients concerning hearing instrument (HI) technology levels are not currently evidence-based. Pre-fitting parameters...
9MB Sizes 0 Downloads 9 Views