Eliciting individual preferences for health care: a case study of perinatal care

Abstract

Objective  To demonstrate how a discrete choice experiment (DCE) can be used to elicit individuals’ preferences for health care and how these preferences can be incorporated into a cost–benefit analysis.

Methods  A DCE which elicited preferences for three perinatal services: specialist nurse appointments; home visits from a trained lay visitor; and home‐help. Cost was included to obtain a monetary measure of the value that individuals place on the services. In total, 292 women who had previously participated in a randomized trial of alternative forms of pre‐natal care were interviewed.

Results  The most preferred service configuration consisted of three nurse appointments and two home visits before birth and 4 h of home‐help per week for the first 4 weeks after birth. On average, women are willing to pay $371 for this package. A package that excluded home‐help was valued at $122 whilst provision of three nurse appointments only was valued at $97. The predicted uptake of the services ranged from 37% to 93% depending on the woman’s experience with the service, whether or not it was her first child and her level of education.

Conclusion  The willingness to pay values were much higher than the costs for nurse appointments, suggesting this service produces a net social benefit. The willingness to pay for the package including both the nurse appointments and home visits only just exceeded the costs of the package, suggesting there is a relatively high chance that this package produces a net social loss.

Introduction

It is increasingly recognized that people’s views and preferences should be taken into account when deciding what health care to provide. ¹ However, obtaining information on preferences that can be readily applied to the relevant decision‐making context can be challenging. Information is required on what attributes of a health‐care service are important to individuals and, more importantly, the strength of their preference for these attributes. This is crucial as trade‐offs often have to be made. For example, generally, there is a trade‐off between providing local services with longer waiting times or providing central services with shorter waiting times. Therefore, information is required on the relative value of each of the attributes of a service (that is their strength of preference) to understand how people trade one off against the others.

An increasingly popular tool for measuring strength of preference for health care is the discrete choice experiment (DCE). ² DCE present individuals with a series of hypothetical choices between different configurations of the services in question (e.g. waiting time and travel distance). Participants indicate which of each pair of service configurations they prefer. The relative importance of the different attributes is then estimated using regression analysis. The DCE results can also be used to determine the optimal configuration of the service(s), to predict the demand for the service given the different configurations and, by including cost as an attribute, one can derive a monetary value that individuals place on the services. The advantage of having a monetary valuation (willingness‐to‐pay measure) is that the costs and benefits are then expressed in the same currency and can therefore be compared directly within a cost–benefit analysis. One strength of the DCE approach is that preferences for non‐health outcomes, such as waiting time, reassurance, etc. as well as for health outcomes can be included in the evaluation. Many (economic) evaluations consider health outcomes only, which to the extent that non‐health outcomes are important can lead to sub‐optimal policy recommendations.

The aim of this study is to demonstrate how a DCE can be used to elicit individuals’ preferences for health care and how DCE can be used to incorporate individuals’ preferences into a cost–benefit analysis. Whilst other studies have used DCE to estimate preferences, very few have incorporated the values in a cost–benefit analysis. ³ We also aimed to demonstrate how a DCE can be used to determine optimal service configuration and to predict uptake of the service to predict demand. This information is crucial for service planning.

The context of the study is the community prenatal care (CPC) trial in Calgary, Canada. The CPC program consists of two components in addition to standard prenatal care provided to all low‐risk pregnant women: a specialist nurse intervention and a paraprofessional home visitor. ⁴ The nurse intervention was based on a comprehensive and universal approach that acknowledged the physical, emotional and spiritual aspects of pregnancy. Home visitors provide non‐medical, practical support to women in their own homes. Whilst these services appeared to have limited impact in terms of ‘hard’ outcome measures such as health outcomes of the baby or the mother, women who had participated in the trial used more pregnancy‐related information and used more resources than women in standard care. ⁴ The challenge is how to capture these more ‘intangible’ benefits in such a way that can inform the evaluation of the effectiveness and efficiency of the services. This is carried out through the use of a DCE to elicit women’s preferences for pre‐natal care. The DCE approach has been used previously to elicit the preferences of women of childbearing age for perinatal care, ⁵ , ⁶ , ⁷ fertility treatment ⁸ and miscarriage management. ⁹

Study design

The DCE was conducted alongside a community‐based‐randomized trial that evaluated the effectiveness of the CPC program. Preferences were elicited from a sub‐set of the women who had previously taken part in the CPC trial.

Community Pre‐Natal Care trial

The trial had three arms: (1) standard care; (2) standard care plus appointments with a Community Nurse Specialist; (3) standard care plus appointments with a Community Nurse Specialist augmented with home visits by trained lay‐advisors. The design of the two intervention arms was informed by several previous studies on home visitation, including the research of David Olds et al. ¹⁰ , ¹¹ , ¹² in the USA, where the provision of home visitation to ‘high risk’ women was found to improve some maternal health outcomes, reduce rates of domestic violence and reduce emergency department visits. A characteristic of this approach is the integration of additional pre‐natal care provided by professionals, such as specialist nurses, psychologists or social workers, and specially trained lay visitors, into obstetric and family physician care. The nurse‐intervention was based on a comprehensive and universal approach that acknowledged the physical, emotional and spiritual aspects of pregnancy. All nurses were experienced public health nurses with training in pre‐ and post‐natal care. Additional training in competency‐based approaches to pregnancy, ¹³ solution‐focused counselling, ¹⁴ the humanistic perspective to learning ¹⁵ and the ‘community as partner’ model ¹⁶ was provided prior to the study. Nurse consultations were conducted primarily in an office within the maternity clinic, and the frequency, length and content of the consultations were determined jointly by participating women and their nurse. Home visitors provided non‐medical, practical support to the women in their own homes. The home visitors focused on social support, practical help and providing a bridge between the family and resources available in the community. Home visitors received training in culturally appropriate care, and the role, philosophy and skills required for home visitation. The intervention was based on three models: ‘Invest in Kids’, ¹⁷ Wisconsin ¹⁸ , ¹⁹ and ‘Within our Reach’. ²⁰

The study population consisted of pregnant women who had booked to attend one of three clinics in Calgary for their first pre‐natal appointment. Women under 18 years of age were excluded from the study because of the ethical issues associated with confidentiality and informed consent, and because a pre‐natal program was already in place to serve specialized pre‐natal needs in this population. Women who were unable to speak English, French, Cantonese, Mandarin, Punjabi, Urdu or an Arabic dialect were also excluded.

The discrete choice experiment

The services to be evaluated in the DCE were those offered in the CPC trial (that is specialist nurse appointments and home visits) as well as one that was not part of the original trial, namely home‐help provided after birth. The latter was included to investigate the relative importance of practical help vs. information and social support. The home‐help was described as providing help with a range of household duties including home cleaning, meal preparation, shopping and laundry. A thorough description of each of the services was presented to the participants (questionnaire is available from authors).

The discrete choice experiment questions

The attributes included in the DCE were the number of appointments with the nurse specialist, the number of home visits, hours of home‐help provided after birth and cost. Appointments with the nurse specialist and home visits were included as these were the services provided in the CPC trial. Home‐help was included as an attribute to investigate the relative importance of practical help vs. information and social support. Cost was included as an attribute so that a monetary valuation of the services could be obtained. The levels assigned to the different attributes were: 0, 1, 3, or 5 nurse appointments; 0, 1, 2, or 4 home visits; 0, 2, 3, or 5 h/week of home‐help (for 4 weeks after birth); and CAD $0, $75, $140 or $250 for costs. The levels chosen for nurse appointments and home visits reflect the average and range in the CPC trial. The levels chosen for home‐help and costs were more arbitrary. The levels of the cost attribute were initially based around the cost of the services and were subsequently tested and revised in a pilot study which sent postal questionnaires to a sample of 100 women.

Given the number of attributes and their levels, the total number of possible combinations is equal to 256. This was reduced to 48 discrete choices using the methods outlined in Burgess and Street. ²¹ Deborah Street (UTS Sydney) produced the design. The design allows for a 2‐way interaction between nurse appointments and home visits, which allows us to investigate whether the optimal number of appointments with the nurse specialist is a function of whether and how many home visits women receive and vice versa. A block design was used that randomly allocated the choices across three versions of the questionnaire, bringing the number of choices for each participant down to a more manageable level of sixteen choices.

Figure 1 provides an example of one of the choices presented to each woman. Participants were asked to choose between package A or B or they could choose neither. It was stressed that if they chose neither then they would not receive any of the services. Data were collected through face‐to‐face interviews carried out in the woman’s own home. The choices were explained to the participants and visual aids were used to simplify the exercise. The latter were descriptions of the services and these were placed in front of participants throughout the DCE exercise. They were then given a paper copy of the sixteen choices. The interviewer only provided assistance if the participant had difficulties answering any of the questions.

Figure 1.

 Example of a choice.

Data analysis

Conditional logit regression in Limdep was used to model individuals’ preferences as a function of the attributes included in the DCE, that is, nurse appointments, home visits, home‐help and cost. A nested logit specification was also estimated but the IV parameter was equal to one indicating that a nested structure was not appropriate. The standard errors of the coefficients were adjusted to allow for correlation of observations within individuals. To investigate the possibility of nonlinearity, utility functions were modelled using second‐degree polynomials, that is, squared terms of the attributes were included. This investigates whether the value of additional services decreases as a function of the number of services individuals receives. For example, the fifth home visit may provide less additional value than the fourth home visit. The diminishing increase in the value or utility that an individual receives from each successive unit is called diminishing marginal utility. As described above, the design allows the investigation of an interaction between nurse appointments and home visits. Because of the squared terms, the total number of interaction terms to be included in the model is four (nurse appointments × home visits; nurse appointments ²  × home visits; nurse appointments × home visits ² ; and nurse appointments ²  × home visits ² ). An alternative‐specific constant for the neither option was also included to investigate whether anybody had a tendency to opt out. We hypothesized that first time mothers were less likely to opt out: the argument being that new mothers with less experience of raising a child would place a higher value on the services. Evidence suggests that experience with the service increases the value that people place on the service. ²² Therefore, it can be hypothesized that experience of receiving home visits and/or nurse appointments during the trial is likely to influence whether they opt out or not. We also hypothesized that women with higher levels of education and those with better connected social networks were more likely to opt out and select the ‘neither’ option in each of the choices: the argument being that women with these characteristics wee more likely to have access to alternative sources of information and support, and would therefore place less value on the services being offered. We also hypothesized that the women’s social networks may influence the value they place on the services. The CPC services provide two types of support to new mothers: emotional support; and information about the services that were available to mothers. These types of support can also be provided by women’s social networks. Therefore, it is hypothesized that women whose social network already provides these types of support place a lower value on the services. Characteristics of the women’s social networks were measured using three variables: network size, mean closeness and density. Mean closeness measures how tightly connected an individual is to the people in her network, whilst density measures how connected the people in her network are to each other. ²³ Looser, more extensive networks, with less redundancy (more variability) among network members, are better for providing access to information. Dense networks, which are better for coordination, are more suited to the provision of social support. The hypotheses were investigated through interaction terms with the alternative‐specific constant for the neither option.

The coefficients generated by the regression analysis were then used to estimate the optimal number of nurse appointments, home visits, hours of home‐help and the associated average willingness to pay for these services, as well as to predict the probability of uptake of the service. Confidence intervals were estimated using the nonparametric method of bootstrapping. Bootstrapping is extensively used in economics to estimate confidence intervals for ratios. ²⁴

The estimated willingness to pay values were also used within a cost–benefit framework and compared with the costs of providing the services. The costs of providing the nurse appointments and home visits were taken from Au et al. ²⁵

If the willingness to pay values are greater than the costs, then it can be concluded that there is a positive net benefit. Whether the services should be provided then depends on the financing system. ²⁶ If the service is financed privately and the values are elicited from service users, then the service will be provided if the net benefit is positive. If the service is financed through a public system, the service will compete with others services because of a fixed budget. Whether the service will be provided will then depend on the size of the net benefit.

Data

The economic evaluation was not part of the original trial protocol. Thus one further round of recruitment took place among women who completed the first outcomes evaluation at 6–8 weeks post‐partum, and whose baby was at least 1‐year old (31 May 2004). We excluded women from the long‐term follow‐up if they had given birth to a very low birth weight baby (gestational age <32 weeks) primarily because of ethical concerns related to responder burden and also because the specialist resources provided to these women and their infants are well‐established. Our own resource constraints also prevented us from following up women who needed the use of a translator in the previous interview (n = 98). In total, 292 women out of the 450 women who completed the CPC trial and were eligible for this study were recruited to the study. Full details of the CPC trial are available elsewhere. ⁴ Women were recruited by telephone around 1 year after birth. Table 1 shows the characteristics of the sample.

Table 1.

 Descriptive statistics of sample (n = 292)

	Mean (SD)	Range
Age of mother (years)	31.4 (4.92)	20–48
Age of infant (years)	1.3 (0.08)	1.1–1.7
Social network
Network size	9.88 (2.65)	2–16
Mean closeness	0.69 (0.13)	0.3–1.0
Density	0.46 (0.21)	0.0–1.0
	n	%
First child	150	51.4
Education
Other	167	57.2
High (university degree)	125	42.8
Arm of the trial
1: Standard care	104	35.6
2: Nurse appointments	89	30.5
3: Nurse appointments and home visits	99	33.9

Results

All 292 women completed sixteen discrete choices and therefore no missing values were generated. Table 2 shows the DCE regression results for the full model and a reduced model, estimated using backward stepwise elimination with a 10% significance level. The interaction terms between nurse appointments and home visits were not statistically significant at a 5% level (indicating that women perceive the two forms of service as supplementary rather than substitutes) and were therefore not included. The coefficients on nurse appointments, home visits and home‐help are positive and significant indicating that individuals place a positive value on these services. The squared terms are significant and negative reflecting diminishing marginal utility. This means that the value of each successive service decreases the more services the women receive. Cost is negative indicating that the higher the cost of the package the lower the utility associated with it. All other things equal, women who have no previous children, and have experience with (part of) the services are less likely to opt out. Women with high levels of education are more likely to opt out. It should be noted that some of these interaction terms are statistically significant at a 10% level rather than 5% level. None of the social network variables were associated with the decision to opt out of service provision.

Table 2.

 Regression results

	Full model		Reduced model
	Coefficient	P‐value	Coefficient	P‐value
Nurse	0.601	0.001	0.435	0.001
Nurse2	−0.093	0.001	−0.069	0.001
Home visitor	0.457	0.001	0.218	0.001
Home visitor2	−0.111	0.001	−0.065	0.001
Nurse × home visitor	−0.255	0.056
Nurse2 × home visitor	−0.008	0.173
Nurse × home visitor2	0.051	0.117
Nurse2 × home visitor2	0.039	0.098
Home‐help	0.796	0.001	0.811	0.001
Home‐help2	−0.092	0.001	−0.093	0.001
Cost	−0.007	0.001	−0.007	0.001
Neither	0.608	0.361	0.725	0.001
Neither × first child	−0.294	0.087	−0.296	0.086
Neither × experience	−0.475	0.008	−0.440	0.012
Neither × high education	0.553	0.001	0.493	0.004
Neither × network size	−0.033	0.371
Neither × mean closeness	0.862	0.215
Neither × density	−0.102	0.793
n	14061		14061
Pseudo R 2	0.179		0.177
Count R 2	0.585		0.580

The most preferred package consists of three nurse appointments, two home visits and 4 h of home‐help (Table 3). On average, women were willing to pay $371 for this package. A package which excluded the home‐help was valued at $122 whilst provision of three nurse appointments only was valued at $97.

Table 3.

 Optimal package

	Optimal package	WTP, mean (95% CI)
Full package	Three nurse appointments Two home visits Four hours of home‐help	371 (314−428)
Excluding home‐help	Three nurse appointments Two home visits	122 (87–161)
Nurse appointments only	Three nurse appointments	97 (68–127)

Mean [95% confidence interval (CI)] values are expressed as dollar ($). WTP, willingness to pay.

By comparing the willingness to pay values with the costs of the CPC services (the part of the service that the health region has direct responsibility for providing), we can assess whether the service provides a positive net benefit. The average cost of providing three nurse appointments is $67. ²⁵ This is well below the average willingness to pay of $97 suggesting that the nurse appointments are worthwhile. Providing the two home visits in addition to the three nurse appointments increases the cost by $52. ²⁵ The total cost of providing both services is $119 which is just below the average willingness to pay suggesting that both services are worthwhile. However, it is important to choose the uncertainty surrounding the net social benefit into account to explore how much confidence can be placed on these point estimates. The results of the sensitivity analysis suggest that there is only a 1% chance that the true ‘willingness to pay’ value for the three nurse appointments would be lower than the cost (a net loss to society). For the package that includes both nurse appointments and home visits, the chances that costs exceed the value of the benefits is much higher namely 44%.

As receipt of the services (should they be introduced) is not compulsory, it is important to be able to predict uptake. Table 4 shows the probability of taking up the different packages according to experience of the service and family circumstances, using the regression results in Table 2. The probability of uptake ranges from 37% for the three nurse appointments only [among women with high levels of education, more than one child, and who had not experienced (part of) the services] to 93% for the full package [among women who were less well‐educated, for whom this was the first child and who had experienced (part of) the service].

Table 4.

 Probability of uptake

	Low/medium level of education (%)		High level of education (%)
	First child	Not first child	First child	Not first child
Three nurse appointments
No previous experience with services	56	49	44	37
Experience with (part of) services	67	60	55	48
Three nurse appointments and two home visits
No previous experience with services	61	53	48	41
Experience with (part of) services	71	64	59	52
Three nurse appointments, two home visits and 4 h of home‐help
No previous experience with services	90	87	85	80
Experience with (part of) services	93	91	89	86

Discussion

This paper demonstrated how a DCE can be used to elicit individuals’ preference for health care. In particular, we showed how DCE can be used to incorporate individuals’ preferences into a cost–benefit analysis and how the results can be used to determine optimal service configuration and to predict uptake of the service. Women’s preferences for perinatal care, including the services offered in the Community Pre‐natal Care program, were elicited. The results showed that the optimal service configuration consisted of four nurse appointments, two home visits and 4 h of home‐help. Cost was included as an attribute allowing the estimation of willingness to pay values. These values were used in a cost–benefit analysis. In case of nurse appointments, the willingness to pay values were much higher than the costs suggesting that the service produces a net benefit. The willingness to pay for the package including both the nurse appointments and the home visits only just exceeded the costs of the package ($122 vs. $119) and with a wide margin of error around the willingness to pay estimate, there is a relatively high chance (44%) that the value women place on the service is less than its costs. The predicted uptake if the services were to be introduced ranged from 37% to 93% depending on the woman’s experience with the service, whether or not it was her first child and her level of education.

The technique of DCE is of particular value if there are important non‐health outcomes associated with the health‐care service. The CPC program was designed to better understand the opportunities for supporting pregnant women at low medical risk who attended a primary care clinic, and was not hypothesized to influence birth weight or gestational length. An evaluation based on health outcomes only would have concluded that the CPC program should not be introduced. However, this ignores important non‐health outcomes associated with the CPC program. The DCE showed that women positively valued the services. This clearly demonstrates the value of DCE for eliciting individuals’ preferences for health care. Moreover, the DCE produced willingness to pay values which could be directly incorporated into the cost–benefit analysis. Whilst DCE are being increasingly used, the results have not been applied extensively within a cost–benefit framework. ³ This should receive more attention in future applications. Other techniques are available to elicit willingness to pay values, such as contingent valuation, but they do not provide information on the relative importance of different attributes and as a result no information on optimal service configuration or predicted uptake.

We should note several limitations of the study. First, our measures of the characteristics of the women’s social networks were limited to two, each reflecting density or closeness of network relationships. Dense networks are more suited to the provision of social support. Looser, more extensive networks, with less redundancy (more variability) among network members, are better for providing access to information. The CPC intervention aimed to provide both sorts of support to new mothers – emotional support as well as information about the services that were available to mothers. Our network measures reflect potential for social support, though this appears to have had no effect on preferences or values.

Secondly, there are limitations inherent in the stated preference approach. Individuals are asked what they would do in hypothetical circumstances. The question arises whether they will behave in real settings as they state they would. There is only little evidence available on this with regards to DCE. ² Thirdly, it could be argued that preferences should be obtained from women who have experienced the services. These are the most appropriate values to use when determining the optimal service configuration from a users perspective. Not all women in our study experienced the services. Those who did were expressing their preferences for a service that they received more than 1 year previously. The time lag was the consequence of carrying out the DCE as part of a more extensive economic evaluation of the CPC service, where we also wanted to capture mid‐term outcomes on the mother and baby. It is possible that the time delay has watered down the women’s experience of the benefits of the service and/or distorted their perceptions of the ‘good’ and the ‘not so good’ parts of the service. That said, part of the substantive value of pre‐natal services lies in the effect that they have on the ongoing experience of child‐raising. This is not something that becomes immediately apparent after the service has been provided, and so a period of reflection is essential. The optimal time at which one should elicit the value that someone might assign to services such as those provided by the CPC is not known. Neither do we know how stable are those values over time. Both questions warrant further analysis.

It could be argued that within the cost–benefit framework the views of the tax paying, general population may be more appropriate, at least, in publicly funded systems such as those of Canada, Europe and Australasia. The validity of tax‐payer values depends in part however on the extent to which preferences are influenced by experience with the service. In line with what other studies have shown, ²² , ²⁷ our results show that individuals who experienced (part of) the service were more likely to assign a positive value to the service (i.e. they were less likely to opt out). We now need to examine why the differences in value arises: do they reflect strategic bias on the part of potential beneficiaries of new services or differences in information between tax‐payers and service users?