Societal preferences for fertility treatment in Australia: a stated preference discrete choice experiment

Abstract

Objective: To investigate preferences for fertility treatment from the Australian general population with the aims of calculating the willingness to pay in tax contribution for attributes (characteristics) that make up treatment and for an “ideal” fertility treatment program. We also assessed whether willingness-to-pay varies by the relationship status or sexual orientation of the patient.

Methods: A stated preference discrete choice experiment was administered to a panel of 801 individuals representative of the Australian general population. Seven attributes of fertility treatment under three broad categories were included: outcome, process, and cost. Attributes were identified through published literature, focus group discussions, expert knowledge, and a pilot study. A Bayesian fractional experimental design was used, and data analysis was performed using a generalized multinomial logit model. Further analyses included interaction terms and latent class modeling.

Results: Six of the seven attributes influenced the choice of a treatment program. Under process attributes, individuals preferred: continuity of care of clinic staff, where patients are seen by the same doctor but different nurses at each visit; “alternative” treatments being offered to all patients; and onsite clinic counseling and peer-support groups. Personalization and tailoring of the treatment journey were not important. Among outcome attributes, the improved success rate of having a baby per cycle and significant side-effects were considered important. Cost of treatment also influenced the choice of treatment program. Individual preferences for fertility treatment were not associated with patients’ relationship status or sexual orientation. Latent class modeling revealed sub-groups with distinct fertility treatment preferences.

Conclusion: This study provides important insights into the attributes that influence the preferences of fertility treatment in Australia. It also estimates socially-inclusive willingness-to-pay values in tax contributions for an “ideal” package of treatment. The results can inform economic evaluations of fertility treatment programs.

Keywords:

• Stated preference discrete choice experiment
• socially-inclusive willingness-to-pay
• generalized multinomial logit model
• Bayesian fractional experimental design

JEL classification codes:

• J13
• C35

Background

Infertility affects ∼15% of couples at any one time, causing significant personal suffering to an estimated 50 million couples worldwide¹. The treatment for infertility has been revolutionized over the last generation with the introduction of assisted reproductive technologies (ART), including in vitro fertilization (IVF), whereby oocytes (eggs) are fertilized outside of the body using partner or donor sperm and transferred into a woman’s uterus with the aim of achieving a pregnancy. At least 2.4 million ART cycles are performed each year globally². Other fertility treatments include artificial insemination and ovulation induction. Artificial insemination involves inserting male sperm (partner or donor) into the woman’s uterus close to the time of ovulation. Ovulation induction involves taking hormonal tablets or sometimes injections to induce ovulation by encouraging eggs to develop in the ovaries and be released. Often ovulation induction and stimulation are also used during artificial insemination to increase the chance of pregnancy.

Public and third-party funding for fertility treatment varies extensively internationally. Australia has a relatively supportive funding environment for fertility treatment, with almost all ART and other forms of fertility treatments attracting a rebate through the universal health insurance scheme, Medicare. There are no restrictions to access or reimbursement based on female age or number of treatment cycles, which are criteria used to ration funding in most other countries³. For example, most European Union countries provide government support to fund a limited number of fertility treatment cycles with restrictions on eligibility based on number of cycles, age, sexual orientation, and relationship status⁴. In contrast, Japan provides income means-tested funding for a limited number of cycles, while no public funding for fertility treatment exists in the US, South America, and most developing countries. However, in recent years an increasing number of states in the US have introduced insurance mandates to cover the provision of fertility treatments^5–7.

The cost of a fresh embryo ART cycle in Australia ranges between $9,000 and $11,000 AUD with ∼ $6,000–$7,000 reimbursed to the patient through Medicare. Similarly the cost of an artificial insemination cycle in Australia is between $2,000 and $3,000 with ∼ $1,000–$2,000 reimbursed through Medicare⁶. The total cost to taxpayers of funding fertility treatment in Australia is estimated to be $400 million AUD per year, or ∼ $30 per average Australian taxpayer. Whether this represents good value for money from a general population perspective is largely unknown, as is the societal preferences for funding fertility treatment.

A stated preference discrete choice experiment (SPDCE) approach is a survey-based technique which examines expected behavior and is a particularly well-suited methodology for exploring individuals’ preferences for fertility treatment, which has both substantial health and non-health outcomes^8–10. In a SPDCE, individuals are asked to consider a number of attributes (characteristics) of a healthcare intervention or treatment program and make trade-offs when making a choice between alternatives based on different combinations of attributes and levels. “Levels” in the context of a SPDCE are the different values that can be assigned to an attribute. The choices made reflect the opportunity cost which is fundamental in determining the value of an intervention¹¹. The SPDCE approach is based on random utility theory (RUT), which assumes that individuals make choices which give them the highest level of satisfaction¹², and the theory of economic value, which posits that individuals value any non-market priced commodity based on a bundle of its attributes as opposed to its whole consumption per se¹³. When a cost or a proxy attribute for price is included as part of the model, it is possible to indirectly estimate willingness-to-pay (WTP) values for positively or negatively preferred attributes or their combination, through the marginal rate of substitution (MRS)^14–16. The MRS is calculated as the ratio of the statistically significant coefficients of the attributes of interest to the negative of the coefficient of the cost attribute¹⁷.

A small number of studies in ART have used SPDCE to elicit preferences for various aspects of fertility treatment^18,19, but rarely has this approach been extended to estimate WTP for fertility treatment programs in its entirety. The aim of this study was to elicit the Australian general population’s preferences and WTP for fertility treatment, specifically to estimate WTP values for attributes that make up fertility treatments; to determine WTP for an “ideal” package of treatment to be offered to Australians based on the current level of public insurance expenditure on fertility treatment; and to determine if WTP varies by the relationship status (single or couple) or sexual orientation of the patient receiving treatment.

Methods

Attributes and levels

Seven fertility treatment attributes, each with three possible levels, were identified through qualitative work^20–23. The methods for the qualitative work have been described previously²⁴. In summary, it included reviewing relevant published literature, two focus group discussions (FGDs) with members of the general population (n = 17). A third FGD was conducted with fertility patients attending a large fertility clinic in Sydney, Australia (n = 4).

Published literature revealed three broad categories of attributes for fertility treatment: outcome attributes related to the quality of treatment as determined by its effectiveness and safety^25–32; process attributes concerned with the delivery of treatment which include, among other things, patient-centeredness, provision of information on treatment, continuity and transition of care, patient’s involvement in decision-making about treatment^{25,26,28,32–37}; and cost attribute related to how fertility treatment is funded^38–41. These findings were supported by information provided from the FGDs and expert consultations.

All FGD participants were recruited through poster advertisement on notice boards in public places (for the general public) and on the fertility clinic’s notice board (for patients). Participants in the FGDs were required to be aged 18 years or older, Australian citizens or residents, and able to speak English. For the IVF clinic FGD, participants had to also be considering, receiving or had previously received fertility treatment. Interested individuals responded to the FGD advertisements by contacting the researchers through a dedicated email address and were screened for eligibility. FGDs used the nominal group technique⁴² to identify and rank attributes of fertility treatment. FGDs lasted ∼1 h, with discussions audio recorded for later transcription without identifying information. The transcribed data were analyzed thematically into the three broad categories of fertility treatment attributes using NVivo software version 11⁴³. Subsequently, the results from the FGDs were consolidated and supplemented by expert opinion through interviews with two fertility doctors, one fertility nurse and one women’s health academic on what they perceived as key characteristics of fertility treatment.

The appropriateness of the identified list of attributes and levels was further validated through an online pilot study with members of the general population (n = 52) recruited by a commercial survey company, Survey Sampling International, SSI (https://www.surveysampling.com/). SSI used members of its general population panels of 100,000 members following the same inclusion criteria as with the FGDs of the general population. The commercial survey panels provided a convenient and engaged sample that mirrored the demographics of the Australian general population in terms of age and sex, Australian Bureau of Statistics (ABS) 2016 Census demographic distribution, as shown in Table 1, for the final survey.

Table 1. Socio-demographic characteristics of respondents participating in the SPDCE.

Category	n (%)	Australia (%)^a
Gender
Male	389 (49%)	(49.3%)
Female	411 (51%)	(50.7%)
Other	1 (< 1%)
Age range (years)
18–24	84 (11%)	15–24^b (12.8%)
25–34	162 (20%)	(14.4%)
35–40	112 (14%)	35–39^b (13.5%)
41–54	233 (29%)	40–54^b (20.1%)
55–64	131 (16%)	(11.8%)
65+	79 (10%)	(15.8%)
Do you have children?
Yes	445 (56%)
No	356 (44%)
Have you ever tried unsuccessfully to have a baby?
Yes	123 (15%)
No	678 (85%)
Would you like to have children in the future?
Yes	274 (34%)
No	527 (66%)
Marital status
Never married	251 (31%)
Married	367 (46%)	(47.7%)
De facto	69 (9%)
Separated/Divorced	82 (10%)	(11.7%)
Widowed	24 (3%)	(5.2%)
Prefer not to say	8 (1%)
Do you have any private insurance?
Yes	450 (56%)
No	351 (44%)
Employment status
Work full time	314 (39%)	(57.7%)
Work part-time	169 (21%)	(30.4%)
Retired/unable to work	141 (18%)
Full time student	43 (5%)
Unemployed	68 (9%)	(6.9%)
Full time care giver/stay at home parent	66 (8%)
Level of education
Did not finish year 12 (High School)	95 (12%)
Graduated from year 12	153 (19%)	(15.7%)
Certificate/trade/TAFE	255 (32%)
University qualification	298 (37%)	(22–30.9%)^c
Family income before tax (Australian dollars)
< $25,000	85 (11%)
$25,000–$49,999	160 (20%)
$50,000–$99,999	234 (29%)
$100,000–$150,000	149 (18%)
> $150,000	80 (10%)
Prefer not to say	93 (12%)
Are you currently or have you previously received fertility treatment?
Yes	52 (7%)
No	749 (93%)
Do you think fertility treatment should be subsidized by Australian government through Medicare?
Yes	545 (68%)
No	256 (32%)
Do you think fertility treatment should be available to same-sex couples?
Yes	348 (44%)
No	259 (32%)
Undecided	194 (24%)
Do you think fertility treatment should be available to single people?
Yes	366 (46%)
No	242 (30%)
Undecided	193 (24%)
Do you have a religious, moral or philosophical objection to fertility treatment?
Yes	64 (8%)
No	666 (83%)
Undecided	71 (9%)

^aFrom 2016 Australian Census data: http://quickstats.censusdata.abs.gov.au/census_services/getproduct/census/2016/quickstat/036

^bSlightly different age ranges are used in the census and the closest range is provided for comparison.

^cA range is provided because the census data splits education into two categories – Bachelor’s degree level and above and Advanced Diploma and Diploma level, the latter can be provided in universities or TAFE/Colleges

The identified attributes from the different qualitative components were iteratively revised for clarity, until the researchers reached consensus on their framing and wording. The attributes with their associated levels are presented in Table 2. The attribute-levels were designated based on good practice principles of SPDCE design^44,45.

Table 2. Identified attributes and levels of a fertility treatment program.

Attributes	Description	Levels
1. Outcomes attributes
Side-effects	This is the extent of nausea, headaches, mood swings and pain associated with the fertility treatment	Negligible, causing no problems performing usual activities
		Moderate, causing some problems performing usual activities
		Significant, causing major problems performing usual activities
Success rate per treatment cycle	This is the likelihood that each treatment cycle will result in a baby	10% chance of having a baby
		20% chance of having a baby
		40% chance of having a baby
2. Process attributes
Counselling and peer-support	This refers to the level of professional counselling provided and whether the treatment program offers peer-support groups	No information on counselling or peer-support groups provided
		Brochures provided on external counselling and peer-support groups
		Onsite clinic counsellors and peer-support groups provided
Treatment journey	This is the extent that the treatment journey is personalised and tailored for each patient	Same treatment pathway for all patients
		Some level of personalised treatment based on upfront tests
		High level of tailored, personalised treatment based on detailed patient tests
Continuity of care of clinic staff	This relates to how often patients are seen by the same staff during each treatment cycle	Patients are seen by different doctors and nurses each visit
		Patients are seen by the same doctor but different nurses at each visit
		Patients are seen by the same team of fertility doctors and nurses each at visit
Availability of ‘alternative’experimental treatments	This refers to whether there is an option to have ‘alternative’ experimental treatment included as an add-on to standard treatment	No ‘alternative’, treatments available
		‘Alternative’, treatments may be considered if requested by patient
		‘Alternative’, treatments offered to all patients
3. Cost attribute
Cost of treatment	This is the additional amount in tax that a taxpayer would have to pay to fund the fertility treatment program for all patients in a year (Currently, each taxpayer pays about $30 per year as part of the Medicare levy to subsidise fertility treatment in Australia)	No additional annual tax
		Additional $5 in annual tax
		Additional $12 in annual tax

The derivation of WTP values used the cost attribute which was framed as the additional amount of tax that a taxpayer would have to pay to fund fertility treatment for all patients in a year, reflecting the way the general population funds fertility treatment in Australia. Currently, the total Medicare Benefit Scheme (MBS) and Pharmaceutical Benefits Scheme (PBS) expenditure for ART is ∼ $250,000,000 and $150,000,000 AUD per year, respectively. With an estimated 14,000,000 taxpayers, a taxpayer contributes on average ∼ $30 AUD per year as part of the Medicare levy towards fertility treatment.

Experimental design

The seven attributes, each with three levels, resulted in 2,187 choice scenarios (= 3⁷) in the full factorial experimental design. This number of choice scenarios would be cumbersome and not feasible to be presented to respondents in a survey^17,46,47. Hence, we followed the recommended practice of using a statistically reduced experimental design that is still capable of estimating the main effect in the SPDCE model^48–50. A Bayesian fractional experimental design in Ngene software⁵¹ was used, which generated 12 choice scenarios based on the priors from a pilot study which used a near-orthogonal fractional design⁵².

Questionnaire construction and data collection

The output of the experimental design was used to construct the SPDCE questionnaire. Given our sample population, we framed the preference elicitation questions following the conceptual framework of Dolan et al.⁵³, which resulted in obtaining socially-inclusive preferences for fertility treatment from an ex-ante context. Figure 1 is an example of the choice set and its preference elicitation question.

Figure 1. An example of a choice set for fertility treatment programs.

As with the pilot study, an online SPDCE was administered to a sample of the Australian population (n = 801) by a commercial survey company, Survey Sampling International (SSI). Respondents were English speaking Australian citizens or residents. The sample was drawn from SSI’s general population panel members and was selected to reflect at least the gender, age, and marital status distribution of the Australian general population aged 18 years and older, as presented in Table 1.

Sample size calculations for a SPDCE is a developing field of research in healthcare. There is no consensus or guidance on the appropriate sample size for a SPDCE. Several factors, such as the format of the questionnaire, the complexity of the choice tasks, the desired precision of results, and the need for sub-group analysis determine the minimum sample size to be used⁵⁴. Recently, de Bekker-Grob et al.⁵⁵ proposed an approach to calculating the sample size for a SPDCE but it requires prior information or knowledge about some parameters, such as expectations of the relative sizes of the preference weights to be estimated, for which we had no information. Previous studies have shown that sample sizes of 40–100 respondents may be enough for reliable statistical analysis⁵⁶. Orme⁵⁷ proposes a “rule of thumb” of using 300 respondents or more for robust results, and a minimum of 200 per group for sub-group analysis^54,58. Researchers commonly use Orme’s⁵⁷ “rule of thumb”, such that we considered a sample size of above 800 to yield enough observations for robust results for our study on the basis that it is considerably larger than the recommended guidance.

Respondents were provided with background information about fertility treatments and how they are funded in Australia, and then asked the following:

Thinking about the characteristics of fertility treatment programs described, which of the following programs—Treatment A or Treatment B—would you prefer others, or yourself, had access to should it be needed to help have a baby. You may also select “Neither” of the options if you would not choose either.

Each participant completed 12 hypothetical choice scenarios constructed from the experimental design where a fertility treatment program varied on attribute-levels. A series of socio-demographic questions was also included, together with a practice question to familiarize the participants with the survey format^9,16,59,60. With the cost of fertility treatment attribute included, attribute trade-offs reveal how much individuals are willing to pay for marginal changes in the attributes^{14–16,61,62}. The experimental design allowed some attribute-levels to be the same in the scenarios being compared (attribute-level overlap) which reduced respondents’ task complexity and facilitated trading-off of attributes^63,64. This study is part of a wider project that has been previously described²⁴ for which ethical approval was obtained from the University of New South Wales, Sydney (HC17255).

Econometric model

Within the random utility theory (RUT) framework, the utility that an individual i derives from choosing alternative j (Treatment A, B, or Neither) in choice set t is given by: $$U_{ijt}=X_{ijt}\beta +{\epsilon }_{ijt};i=1,\mathrm{ }2,\dots ,n^i;t=1,\mathrm{ }2,\dots n^t$$ where$\mathrm{ }{U}_{ijt}$ is the observed utility$,\mathrm{ }{\mathrm{ }X}_{ijt}$ is a vector of variables representing attributes of alternative j, and $\beta$ is a vector of coefficients. $X_{ijt}\beta$ is the deterministic component, assumed to be linear and additive, ${\epsilon }_{ijt}$ is the random component^17,65, $n^i$ is the number of respondents, and $n^t$ is the number of choice scenarios. In this study, the deterministic component is represented as: $$X_{ijt}\beta =\mathit{\text{ASCa}}+\mathit{\text{ASCn}}+{\beta }_{1(\mathit{\text{side}}\_\mathit{\text{effects}})}+{\beta }_{2(\mathit{\text{couns}}\_\mathit{\text{peer}}\_\mathit{\text{support}})}+{\beta }_{3(\mathit{\text{success}}\_\mathit{\text{rate}})}+{\beta }_{4(\mathit{\text{treat}}\_j)}+{\beta }_{5(\mathit{\text{cont}}\_\mathit{\text{care}})}+{\beta }_{6(alt\_\mathit{\text{treat}})}+{\beta }_{7(\hspace{0.17em}\text{cos}\hspace{0.17em}t\_\mathit{\text{treat}})}$$ where ${\beta }_{1(\mathit{\text{side}}\_\mathit{\text{effects}})}$ to ${\beta }_{7(\hspace{0.17em}\text{cos}\hspace{0.17em}t\_\mathit{\text{treat}})}$ are preference parameters or marginal utilities of individuals for given levels of attributes with reference to the base level. A choice outcome variable was created as a dichotomous dependent variable to signify the choice decision made for each fertility treatment program scenario, with the independent variables being the combinations of each attribute-level. Effects coding was used for all categorical attributes to allow effect of each attribute to be uncorrelated with the constant, while continuous attributes assumed their values⁶⁶. Coding of “Neither” option was the same throughout choice sets.

We progressed from Model 1, the classical conditional logit (clogit) model, also known as multinomial logit (MNL) (see Supplementary Appendix), which estimated fixed effects with the assumption of homogeneous preferences across respondents^67,68 to Model 2a, a generalized multinomial logit (GMNL) model^69–71. The GMNL model is an extension of the commonly used mixed logit (MXL) model which accounts for both taste and scale heterogeneity. Taste heterogeneity implies that individuals’ preferences for observed attributes are different, while scale heterogeneity is the non-consistency of individuals’ choices between attributes. In the GMNL model, all parameters were specified as random and normally distributed except cost parameter, which was assumed to be fixed. This specification avoids undefined moments and heavily skewed distributions when calculating WTP values through marginal rate of substitution (MRS)^72–74. An additional model (Model 2b) was estimated where Model 2a was extended to include an interaction which estimated fixed effects of responses to the question on whether fertility treatment should be available to same-sex or single individuals and all the attributes. Latent class (LC) analysis (Model 3) with discrete distribution of parameters across individuals was also undertaken to identify classes with similar preferences, but different across classes^17,65. The optimal number of model classes is theoretically unknown and was determined through the estimation of a series of models with an increasing number of classes from 2 to 10⁷². An iterative procedure was used to retain four classes based on comparisons of the Log-likelihood (LL), Akaike, Consistent Akaike, and Bayesian Information Criteria (AIC,CAIC and BIC) after which an additional class did not significantly improve the model fit⁷⁵. Moving from the base MNL model to more advanced GMNL and LC models resulted in an improved model fit, as shown by the LL, AIC, and BIC scores in the lower section of Table 3 ⁷⁶.

Table 3. GMNL and LC model results.

Coefficient (Standard error)	GMNL (Model 2a)		GMNL interaction-availability of treatment to same-sex couples (Model 2b)	GMNL interaction-availability of treatment to single people (Model 2b)	LC (Model 3) Class membership
					1 (18%)	2 (18%)	3 (48%)	4 (16%)
ASCa	0.0200 (0.0451)		0.0205 (0.0462)	0.0205 (0.0462)	0.0055 (0.1068)	0.0497 (0.0951)	–0.0093 (0.0385)	–2.0134 (1.2659)
ASCn	–0.2629 (0.2382)		–0.1715 (0.3053)	–0.1715 (0.3053)	3.4691*** (0.3671)	–0.8511*** (0.2573)	–2.5618*** (0.2895)	4.2875 (3.1261)
Side-effects (base level = Negligible) Moderate	–1.4714*** (0.2171)		–0.8341 (0.5090)	–0.8341 (0.5090)	–0.5463 (0.2652)	–1.6411*** (0.1972)	–0.7865 (0.3271)	–3.7349 (1.4320)
Significant	–3.0340*** (0.2488)		–2.7836*** (0.5064)	–2.7836*** (0.5064)	–2.2764*** (0.2266)	–2.3373*** (0.1865)	–0.6771*** (0.1034)	–2.0548 (1.7761)
Counseling and peer-support groups (base level = No information)
Brochures	–0.5356*** (0.1407)		0.2127 (0.3591)	0.2127 (0.3591)	–0.0055 (0.3178)	–0.3537 (0.1818)	–0.3368 (0.1539)	6.2952 (3.9422)
Onsite	0.2423*** (0.0643)		0.5008*** (0.1678)	0.5008*** (0.1678)	0.2466 (0.1216)	0.3395 (0.1487)	–0.0833 (0.0636)	2.9504 (1.6305)
Success rate per treatment cycle	0.1221*** (0.0099)		0.1041*** (0.0178)	0.1041*** (0.0178)	0.1277*** (0.0088)	0.0201*** (0.0050)	0.0525*** (0.0045)	0.1689 (0.1171)
Treatment journey (base level = Same treatment pathway)
Some level of personalized treatment	–0.1515 (0.1192)		0.3041(0.3017)	0.3041 (0.3017)	0.1272 (0.2966)	0.3354 (0.1918)	–0.6600 (0.1909)	–2.4771 (1.6019)
High level of tailored, personalized treatment	0.0288 (0.1160)		0.4819 (0.3040)	0.4819 (0.3040)	0.2469 (0.2296)	–0.0957 (0.1822)	–0.0657 (0.1470)	–0.6362 (2.4711)
Continuity of care (base level = Patients are seen by different doctors and nurses)
Patients are seen by the same doctor, but different nurses	0.5842*** (0.1111)		0.2663 (0.2892)	0.2663 (0.2892)	0.6121*** (0.2028)	0.4325*** (0.1499)	–0.1518 (0.1252)	–6.5584 (7.1429)
Patients are seen by the same team of fertility doctors and nurses	0.3442*** (0.1161)		0.1274 (0.2643)	0.1274 (0.2643)	0.5416*** (0.1518)	0.3371 (0.1467)	0.4581*** (0.1527)	–1.4560 (2.4032)
Availability of “alternative” experimental treatments (base level = None)
“Alternative” treatments may be considered if requested	0.5447*** (0.1464)		0.4198 (0.3653)	0.4198 (0.3653)	0.7143*** (0.2159)	0.0302 (0.2085)	0.3071 (0.1389)	–5.8660 (4.0773)
“Alternative” treatments offered	0.5545*** (0.1229)		0.5687 (0.3350)	0.5687 (0.3350)	0.8893*** (0.2134)	0.2700 (0.1804)	0.2932 (0.1617)	–0.5167 (1.0205)
Cost of treatment	–0.0548*** (0.0083)		–0.0176 (0.0207)	–0.0176 (0.0207)	–0.0345 (0.0082)	–0.0946*** (0.0133)	0.0049 (0.0065)	–0.1271 (0.1270)
GMNL with interactions
Side-effects (base level = Negligible) Moderate			0.2230 (0.2546)	–0.5898 (0.2828)
Significant			0.4358 (0.6463)	–0.6063 (0.5303)
Counseling and peer-support groups (base level = No information)
Brochures			–0.2909 (0.2354)	–0.1091 (0.2215)
Onsite			–0.0054 (0.1262)	–0.1350 (0.1184)
Success rate per treatment cycle			–0.0001 (0.0189)	0.0090 (0.0169)
Treatment journey (base level = Same treatment pathway)
Some level of personalized treatment			–0.0252 (0.2980)	–0.2250 (0.2921)
High level of tailored, personalized treatment			–0.1481 (0.1526)	–0.1141 (0.1591)
Continuity of care (base level = Patients are seen by different doctors and nurses)
Patients are seen by the same doctor, but different nurses			0.2581 (0.2576)	–0.0965 (0.2588)
Patients are seen by the same team of fertility doctors and nurses			0.1185 (0.1621)	–0.0109 (0.1396)
Availability of “alternative” experimental treatments (base level = None)
“Alternative” treatments may be considered if requested			0.0454 (0.2541)	0.0062 (0.2284)
“Alternative” treatments offered			0.1194 (0.2438)	–0.1318 (0.2347)
Cost of treatment			–0.0188 (0.0198)	–0.0016 (0.0180)
Standard deviation of random parameters
ASCa	0.0336		0.0187	0.0187
ASCn	4.7809***		4.6998***	4.6998***
Side-effects (base level = Negligible) Moderate	0.8856***		0.8351***	0.8351***
Significant	1.6088***		1.7663***	1.7663***
Counseling and peer-support groups (base level = No information)
Brochures	0.0094		0.0881	0.0881
Onsite	0.0279		0.0879	0.0879
Success rate per treatment cycle	0.1127***		0.1151***	0.1151***
Treatment journey (base level = Same treatment pathway)
Some level of personalized treatment	–0.0308		–0.0195	–0.0195
High level of tailored, personalized treatment	0.1684		0.1322	0.1322
Continuity of care (base level = Patients are seen by different doctors and nurses)
Patients are seen by the same doctor, but different nurses	0.0210		0.1264	0.1264
Patients are seen by the same team of fertility doctors and nurses	–0.0576		–0.1849	–0.1849
Availability of “alternative” experimental treatments (base level = None)
“Alternative” treatments may be considered if requested	–0.1073		–0.1124	–0.1124
“Alternative” treatments offered	–0.1708		–0.1469	–0.1469
Tau (scale parameter)	0.7079*** (0.0633)		0.6929*** (0.1060)	0.6929*** (0.1060)
Goodness-of-fit
LL	–6,534.51		–6,504.49	–6,504.49	–6,708.25
Degrees of freedom	28		52	52	59
AIC	13,125.01		13,112.98	13,112.98	13,534.49
BIC	13,356.55		13,542.99	13,542.99	14,022.39
No. of observations	28,836		28,836	28,836	28,836

*** Significant at ρ < 0.001.

This SPDCE was unlabeled, hence we included an alternative specific constant for choosing treatment program A as ASCa and the “Neither” option as ASCn. These parameters were specified as random with normal distribution and allowed us to capture the average effect on utility for a systematic tendency of choosing treatment program A and “Neither” option, respectively⁷³.

Data analysis

All analyses were performed in Stata 14⁷⁷. The estimated coefficients and their statistical significance indicate individuals’ relative strengths of preferences for the different fertility treatment attribute-levels, where more preferred attribute-levels have higher preference weights. Changes in the preference weights between the most and least preferred levels of each attribute can be interpreted as the importance of an attribute relative to that of all other attributes, conditional on the ranges for each attribute⁷⁸. A positive coefficient implies that the change in the attribute-level increases the utility derived and the converse for the negative coefficient.

Subsequently, WTP for attributes of fertility treatment were indirectly calculated using the MRS. The positive/negative WTP estimate indicates individual’s willingness to pay/to be compensated for a change in a given attribute level, all factors being held constant. The 95% confidence intervals (CIs) for WTP values were derived using the delta method⁷⁹, and the WTP values are reported in 2018 Australian dollars.

Results

Survey respondents

A total of 801 respondents completed the SPDCE survey in June 2018, which resulted in 28,836 observations. The average time spent to complete the questionnaire was 9.46 min. A total of 27% chose the “Neither” option, while 73% made at least a choice of a treatment program across all choice scenarios. Those who did not want to have children in future, were not on full-time work, and those without private insurance were more likely to choose none of the treatment programs presented to them in the choice scenarios. Most respondents (83%) indicated that the SPDCE questionnaire was very easy, easy, or OK to complete. There were slightly more females (51%) than males, and 45% were aged between 18 and 40 years. Most respondents had children (56%). Table 1 presents the socio-demographic characteristics of the respondents.

SPDCE results

Table 3 reports estimates for the GMNL model (Model 2a), an extension of the GMNL to include interaction effects (Model 2b) and the latent class model (Model 3). The results of the preliminary MNL model 1 are presented in the Supplementary Appendix. The estimated coefficients in the models are log-odds associated with each attribute and can be interpreted as relative individual preference weights. The size and sign of the coefficients indicate the extent of individual preference for the attributes.

In Model 1, the ASCa was statistically insignificant, suggesting that, holding all other factors constant, individual choices of a treatment program were not influenced by the systematic tendency to choose only the first alternative (Treatment Program A). The ASCn was positive and statistically significant, which implies that respondents derived utility in choosing the “Neither” option. At least one level of all seven attributes was statistically significant at ρ < 0.001 which implies that they were considered important when choosing a treatment program.

Of interest are the results in Models 2a, 2b, and 3. The ASCa coefficients were statistically insignificant for these models. The ASCn coefficients related to choosing “Neither” option were also insignificant, except for class 1, 2, and 3 of Model 3. The positive and statistically significant ASCn suggest utility derivation from the “Neither” option and the converse for the negative significant ASCn, all other factors being the same.

In Model 2a, at least one level of only six attributes influenced individuals’ choice of a fertility treatment program. Examining process attributes first, continuity of care of clinic staff where patients are “seen by the same doctor but different nurses” at each visit, relative to “seen by different doctors and nurses” each visit, was preferred the most. This was followed by a treatment program that offers “alternative” treatments to all patients compared to a treatment program that offered none, and onsite clinic counseling and peer-support groups compared to non-provision of information, in that order. The extent that the treatment journey is personalized and tailored for each patient did not influence the treatment program choice decision. Improved success rate (likelihood) of having a baby per cycle was considered most important. Both moderate and significant side-effects had a strong negative association with the decision to choose a treatment program.

Cost of treatment was also an important consideration. The scale parameters from Models 2a and 2b were statistically significant (Model 2a = 0.7079, ρ < 0.001 and Model 2b = 0.6929, ρ < 0.001), indicating the presence of scale heterogeneity. Holding all other attributes constant in Model 2a (GMNL model), the ratio of treatment preferences to cost of treatment (marginal rate of substitution) revealed that respondents were willing to pay an additional annual tax contribution over and above what they pay now of:

• $10.66 (95% CI = $6.60-$14.73) for continuity of care of clinic staff, where patients are seen by the same doctor but different nurses at each visit;

• $10.12 (95% CI = $5.51–$14.73) for a treatment program that offers “alternative” treatments to all patients;

• $4.42 (95% CI = $1.87–$6.98) for onsite clinic counseling and peer-support groups; and

• $2.23 (95% CI = $1.64–$2.82) for a 1% improvement in the success rate per cycle of having a baby.

Respondents had a considerable disutility for a treatment program with significant side-effects which cause major problems performing usual activities’ relative to negligible side-effects, such that they were willing to pay an additional annual tax of $55.38 (95% CI = $39.64–$71.13) to avoid significant side-effects.

There was little evidence to suggest that the general population’s preferences for fertility treatment differed by patients’ sexuality or relationship status.

Latent class modeling (Model 3) revealed similar preference class shares for fertility treatment that were different across classes of 18% (class 1), 18% (class 2), 48% (class 3), and 16% (class 4). There was no systematic tendency of choosing treatment program A in all the four classes. Individuals in class 1 derived utility in choosing “Neither” option compared to a treatment program. On the other hand, classes 2 and 3 preferred choosing at least a treatment program, while class 4 individuals appear not to have cared about the “Neither” option. Individuals in classes 1–3 were influenced by outcome attributes. They exhibited a dislike for a treatment program with significant side-effects relative to negligible side-effects, and considered improved success rate per cycle of having a baby as important. Individuals had varied preferences related to process attributes. Class 1 preferred a treatment program with continuity of care of clinic staff, where patients are seen by the same team of fertility doctors and nurses each visit compared to being seen by different doctors and nurses each visit, and a treatment program that offers “alternative” treatments to all patients compared to one that does not. Only continuity of care of clinic staff where patients are seen by the same doctor but different nurses at each visit relative to having different doctors and nurses each visit was considered important by class 2 individuals. Class 3 preferred continuity of care of clinic staff where patients are seen by the same team of fertility doctors and nurses each visit compared to different doctors and nurses each visit and a treatment journey with some level of personalized treatment based on upfront tests compared to having the same treatment pathway for all patients. Only class 2 individuals cared about cost of treatment. On the other hand, individuals in class 4 (16%) did not consider any of the attributes important, and perhaps did not care about fertility treatment in general. Table 4 presents overall preferences for a fertility treatment program from results of latent class analysis. The standard deviations relative to the coefficients of ASCn, moderate, and significant side-effects, and success rate per treatment cycle for Models 2a and 2b were statistically significant, implying significant variability in preferences for these attributes.

Table 4. Results of latent class analysis.

Parameter	Class 1 (18%)	Class 2 (18%)	Class 3 (48%)	Class 4 (16%)
ASCa	No systematic tendency of choosing Treatment A	No systematic tendency of choosing Treatment A	No systematic tendency of choosing Treatment A	No systematic tendency of choosing Treatment A
ASCn	Utility in choosing “Neither” option	No Utility in choosing “Neither” option rather chose Treatment A or B	No Utility in choosing “Neither” option rather chose Treatment A or B	Did not care
Outcome attributes	Disliked a treatment program with significant side-effects relative to negligible side-effects	Disliked a treatment program with significant side-effects relative to negligible side-effects	Disliked a treatment program with significant side-effects relative to negligible side-effects	Did not care
	Improved success rate per cycle of having a baby was important	Improved success rate per cycle of having a baby was important	Improved success rate per cycle of having a baby was important	Did not care
Process attributes	Preferred continuity of care of clinic staff where patients are seen by the same team of fertility doctors and nurses each visit compared to being seen by different doctors and nurses each visit	Preferred continuity of care of clinic staff where patients are seen by the same doctor but different nurses at each visit relative to having different doctors and nurses each visit	Preferred continuity of care of clinic staff where patients are seen by the same team of fertility doctors and nurses each visit compared to different doctors and nurses each visit	Did not care
	Preferred treatment program that offers “alternative” treatments to all patients compared to one that did not		Preferred treatment journey with some level of personalized treatment based on upfront tests compared to having the same treatment pathway for all patients
Cost attribute		Cared about the cost of treatment		Did not care about the cost of treatment

Discussion

This study found that societal preferences for attributes of a fertility treatment program and subsequently their WTP are influenced mostly by process and outcome attributes. For process attributes the level of continuity of care of clinic staff, whether “alternative”/experimental treatments are offered, and whether there is counseling and peer-support groups were the most important, in that order. For outcome attributes, improved success rate (likelihood) of having a baby per cycle and side-effects were considered important. An increase in the cost of treatment would result in disutility for a treatment program. This study has also found that those who did not want to have children in the future, were not in full-time work, and those without private insurance were more likely to choose none of the configuration of the treatment program scenarios presented to them.

This study has produced WTP estimates of the attributes of a fertility treatment program from a socially-inclusive Australian general population perspective and an ex-ante context. These WTP estimates can be used in an economic evaluation of fertility treatment to assess value for money for the public investment in fertility treatment. For example, a cost-benefit analysis framework where the marginal WTP estimates or the total of the marginal WTP estimates for the preferred attributes or most preferred configuration of fertility treatment program could be compared with the cost of delivering those attributes, if known^80–82. The total of the marginal WTP estimates can be derived using the compensating variation (CV) approach which accounts for the probability of making a choice, hence consistent with the RUT framework and has a solid basis in welfare theory^83,84.

The common assumption used in healthcare when applying the CV approach to an unlabeled SPDCE is that individuals have a preference for one treatment over another, which would be attributed solely to differences in one or more attributes of that treatment^68,85. Hence, there is certainty in choosing the treatment, with the probability of making a choice being one. In this case, the CV approach reduces to the sum of marginal WTP for each attribute⁸⁰.

On this basis, this study has found that a representative sample drawn from the Australian general population, as shown in Table 1 (at least by gender, age distribution, marital status), is willing to pay an additional $27.43 in annual tax contribution, to fund the preferred configuration of a fertility treatment program for everyone who might benefit from treatment in Australia. Currently, in Australia, on an average $30 per year per taxpayer from the Medicare levy subsidizes fertility treatment with undefined attributes for everyone who needs it. The total cost to Australian taxpayers is around $400 million. The societal WTP of $27.43 as an additional annual average tax contribution to fund the “ideal” fertility treatment program translates to about a 90% increase from the current level. Given that respondents were told that the average Australian taxpayer contributes about $30 per year to Medicare towards providing fertility treatment to all Australian citizens and residents, these WTP estimates indicate the current level of funding for fertility treatment is viewed favorably, and that society would be willing to pay more in tax for an “ideal” package of treatment to be offered to Australians.

Individuals preferred a fertility treatment program with the following process attributes: continuity of care of clinic staff where patients are seen by the same doctor but different nurses at each visit compared to having different doctors and nurses each visit, “alternative” treatments offered to all patients rather than none, onsite clinic counseling and peer-support groups compared to non-provision of information counseling and peer-support, and with the following outcomes attributes: improved success rate of having a baby per cycle, and without significant side-effects relative to negligible side-effects. Cost of treatment also influenced the choice of a treatment program such that the marginal utility of increases in cost was negative. There was little evidence to suggest that individual preferences for fertility treatment vary according to relationship status (single or couple) or sexual orientation of the patient being treated. The results of the latent class analysis revealed that the general population inevitably consists of sub-groups with different preferences, which are different across the groups.

This research makes a considerable contribution to informing policy and resource allocation decisions related to the provision of fertility treatment. However, as is the case with all SPDCEs, there are methodological limitations such that the results should be interpreted with caution. It is impractical and cognitively not plausible to include all relevant attributes and levels for fertility treatment in the SPDCE; however, the qualitative and pilot stages of the study were used to identify the most relevant attributes for inclusion. In addition, this study was undertaken in Australia, where funding for fertility treatment is subsidised by government through Medicare, hence the absolute results are not directly transferable to other settings. It is further acknowledged that the SPDCE approach has inherent hypothetical bias⁸⁶. In this regard, effort was made to mitigate hypothetical bias by reminding respondents to make choices between the hypothetical scenarios of fertility treatment presented to them as they would in real life. The use of a market research panel sample of a survey company may also be a drawback because they are rewarded for taking part. However, it is argued that professional respondents of a panel are more attentive and responsive, hence engage with the survey, than volunteers⁸⁷.

Conclusion

To our knowledge, this is the first study that has attempted to elicit preferences and subsequently calculated WTP values for attributes of a package of fertility treatment. This study provides important insights into the attributes that influence the choice of a fertility treatment program from the perspective of the Australian general population. It further produces socially-inclusive WTP estimates for preferred fertility treatment attributes and treatment programs which can be used in economic evaluations, such as cost-benefit analyses to assess the value for money of fertility treatment programs. The results will contribute to the development of an evidence-based policy framework for the provision of cost-effective and patient-centered fertility treatment in Australia.

Transparency

Declaration of funding

This paper is part of the study being funded by the Australian National Health and Medical Research Council (NHMRC) project grant AP1104543 and has been approved by the University of New South Wales Human Research Ethics Committee (HEC 17255).

Declaration of financial/other interests

WB, ND, MS, RN, and GMC declare that they have no conflicts of interest. JME peer reviewers on this manuscript have no relevant financial or other relationships to disclose.

Acknowledgments

We wish to thank members of the public and patients who agreed to participate in our focus group discussions which informed the design of this study. We also thank Professor Denzel Fiebig of the University of New South Wales, whose comments on the first draft assisted greatly in the revision of this paper.