FormalPara Key Points for Decision Makers

Family preferences and barriers may impact broader health service design of neurodevelopmental follow-up care for children; however, there is a gap in eliciting families’ preference, particularly regarding health service design and delivery perspective.

A latent class analysis revealed two classes, with a higher proportion of families with full-time employees, a higher income, and metropolitan residents belonging to class 1 compared with class 2. Most of the decision making in class 1 was driven by the location of follow-up care services, while that in class 2 was influenced by cost of services.

A scenario analysis noted about 24% increase in the uptake of follow-up care when families’ ‘ideal’ preferences were considered.

1 Introduction

Neurodevelopmental disorders are a leading cause of morbidity impacting children and their families as well as healthcare and education systems [1]. These consist of a group of disorders, including diagnoses of intellectual disorders (ID), communication disorders, autism spectrum disorder (ASD), attention-deficit/hyperactivity disorder (ADHD), learning disorders (LD), motor disorders, and tic disorders [2]. In existing literature, few studies have measured the prevalence of neurodevelopmental disorders. While there are estimates for some individual diagnoses, collective prevalence estimates are lacking. A national survey in the United States estimated the weighted prevalence of ADHD, ASD, ID, and LD to be 8.5%, 2.9%, 1.4%, and 6.4%, respectively, in 2019–2020 [3]. These estimates are comparable to reports from other high-income countries; for example, the Australian Institute of Health and Welfare reported the prevalence of ADHD in Australian children to be 8.2% [4].

In this manuscript, ‘neurodevelopmental follow-up care’ refers to the care provided to children at risk of or diagnosed with neurodevelopmental disorder or delay and their families, to monitor the developmental outcomes and provide suitable support and services [5, 6]. Effective follow-up care programmes are likely to require sustained engagement with healthcare services. To promote sustained engagement with services, follow-up care programmes ought to be designed considering the preferences and needs of families.

Despite the importance of including family preferences in the design and delivery of neurodevelopmental care programmes, this topic has received little attention in research to date. Previous neurodevelopment care-related studies have primarily focussed on diagnosis-specific issues without consideration of how family preferences and challenges may impact broader health service design [7,8,9,10]. For example, waiting time has frequently been reported as a major barrier to accessing developmental services for children [11,12,13]. Another important healthcare-service-related factor for families is the cost associated with neurodevelopmental assessments, therapies, medications, as well as travelling or accommodation for accessing services [11, 14]. A lack of educational resources [11, 15] and mental health support for parents [16] have also been noted. These known factors are likely to be priority considerations for families of children with neurodevelopmental needs, with implication for the design and improvement of health services.

Preferences can be identified using several methods. Previous studies in this clinical area have elicited preferences using interviews, focus groups, and surveys [17,18,19,20]; however, quantifying weighted preference of attributes (features) cannot be captured using these methods. Additionally, the design and analysis associated with these methods may not be robust enough to disentangle preferences for the purpose of health service design. Discrete choice experiments (DCE) are an increasingly used preference method that can quantify people’s preferences where individuals are presented with several hypothetical health scenarios (choice tasks), each containing several alternatives (options) with different attributes for the individuals to choose from [21]. A recent systematic review by our research team indicated that DCE is a suitable method for studies related to neurodevelopmental follow-up care [22]. However, limitations were noted in the methodology of the studies included in the review, such as not providing reasons for forced choices, not conducting pilot tests, and using numerous attributes or choice sets, which made it cognitively burdensome for respondents.

Given the methodological limitations and the lack of eliciting preferences in this field, particularly from the perspective of health service design and delivery, this study aimed to identify families’ preferences for neurodevelopmental follow-up care for children in Australia using a DCE.

2 Methods

This study was approved by the Human Research Ethics Committee, Queensland University of Technology (approval number: 6468). Written consent was obtained from all the participants. The DCE design and analysis were reported based on the checklist and reports of the International Society for Pharmacoeconomics and Outcomes Research (ISPOR) Conjoint Analysis Task Forces (Table 1 of Online Resource 1, see electronic supplementary material [ESM]) [23,24,25]. Recent guidance published by the International Academy of Health Preference Research (IAHPR) was also used to ensure that the methodological approaches aligned with current best practices in health preference research [26, 27].

Table 1 Attributes and levels
Full size table

A DCE has several stages that makes it a robust technique for eliciting preferences. These stages broadly include attribute identification and selection, experimental design strategy, final survey administration, and econometric model and data analysis [28,29,30,31]. Several components within these stages were specifically adopted for our DCE, and this process is outlined in the flowchart in Fig. 1.

Fig. 1
figure 1

Flowchart of discrete choice experiment study process. MNL multinomial logit, MMNL mixed multinomial logit, LCM latent class model, AIC/N Akaike information criterion/number of observations, LL Log likelihood function

Full size image

2.1 Stage 1: Attribute Identification and Selection

It is typically recommended to conduct qualitative groundwork to support the identification and selection of attributes [23, 32, 33]. We conducted a systematic review and in-depth interviews to explore a broad range of attributes, followed by expert panel discussions to narrow down the attribute list for inclusion in the DCE. These are explained in detail below.

2.1.1 Systematic Review

We undertook a systematic review to identify a list of attributes. We systematically searched PubMed, Embase, CINAHL, Scopus, and Google Advance databases. After the title and abstract (n = 6618 papers) and full-text (n = 16 papers) screening, we obtained a comprehensive list of attributes (n = 102 attributes, Table 2 of Online Resource 1, see ESM) from international literature. This review is published elsewhere [22].

2.1.2 Interviews

We interviewed families of children with neurodevelopmental delays to obtain their first-hand experiences regarding follow-up care. Due to the sensitive nature of the condition, recruiting a large number of participants was challenging [34, 35]. Moreover, we faced challenges due to attempted imposter participation after recruiting via open postings on social media platforms associated with a neurodevelopment organisation [36]. This initially caused disruption and delays in the study, and we chose not to use any data collected from the attempted imposter participants. We promptly implemented strategies to prevent such occurrences in subsequent recruitment with a stricter screening process and recruitment via a closed social media page affiliated with the same organisation. In doing so, we were able to recruit two participants for in-depth interviews that provided valuable insights into the topic. Given time and resource limitations we were not able to recruit additional participants. Content analysis of the interview transcripts by PS (and confirmed by BA) identified 14 attributes/themes which aligned with the attributes identified in our systematic review (Table 3, Online Resource 1, see ESM).

2.1.3 Expert Panel

Limitations, including the cognitive burden on respondents and sample size constraints, prevented the inclusion of all identified possible attributes via the preliminary qualitative work [37]. Typically, five to seven attributes are ideal for a DCE [38, 39]. Consequently, before the expert panel discussion, a team of health economists and a qualitative research expert on neurodevelopment-related care (SK, BA and PS) reviewed the attributes identified from the review and interviews to prioritise their inclusion based on their relevance to the research question and contextual importance in Australia. Irrelevant attributes were excluded, and similar attributes were combined, reducing the list to 11 attributes. (Table 4, Online Resource 1, ESM).

This was followed by expert panel discussions to further narrow the attribute list. The expert panel (n = 5) included health economists and qualitative research experts with extensive experience in conducting interviews with a similar cohort, that is, families of children needing neurodevelopment-related care. This exercise was conducted individually with each expert. The 11 attributes were presented to the experts, who discussed and ranked the attributes in the order of their importance, relevance to neurodevelopmental follow-up care for children, and suitability to our research question. Then, all experts gathered to finalise the attributes to be included in the DCE, aiming to reach a consensus via unanimity, ensuring no objections persisted. At this point, seven attributes were finalised. This exercise also included defining the levels of attributes. The final list of attributes and levels and their descriptions are provided in Table 1.

2.2 Stage 2: Experimental Design Strategy

This stage includes the construction of choice tasks using a suitable experimental design. A choice task includes hypothetical scenarios, each containing a certain number of alternatives [21]. Those alternatives will comprise the finalised attributes and levels from Stage 1, allowing individuals to choose from them. This is followed by pre-testing and pilot testing of the design to check the functionality of the survey and to obtain suitable priors for the final DCE, respectively.

2.2.1 Constructing Choice Tasks

We included two unlabelled alternatives (Follow-up care A and Follow-up care B) and a ‘neither’ alternative. In this case, neither indicates that respondents do not prefer any of the provided options (A and B), but it does not necessarily mean that they do not prefer follow-up care at all. We employed a dual response approach where respondents were first given three options, ‘A’, ‘B’ and ‘neither’ (unconditional response). If they selected ‘neither,’ they were presented with a forced choice between ‘A’ and ‘B’ (conditional response). The inclusion of a ‘neither’ option was to make the choice tasks more realistic, as in the real world, there may be many scenarios where people do not prefer any available choices. To prevent bias due to the placement order of the alternatives, the position of the ‘neither’ alternative in the choice task was varied randomly.

The below formulae were used to estimate the total number of profiles and pairwise choice tasks, respectively:

$$P = {L}_{1}^{M1}+ {L}_{2}^{M2}+ {L}_{3}^{M3}$$
$$S = P\left(P-1\right)/2$$

Where P is the possible number of profiles and M1, M2, M3 are the number of attributes corresponding to the number of levels (L1, L2, L3) for different attributes [40]. Given three attributes with three levels (L1 = 3, M1 = 3), two attributes with two levels (L2 = 2, M2 = 2), and two attributes with five levels (L3 = 5, M3 = 2), this would result in 2700 (33 × 22 × 52) possible profiles and 3,643,650 combinations of pairwise choice tasks [2700 (2700 – 1)/2]. A full factorial design (a design encompassing all possible combinations of attribute levels) was not possible, so we used a fractional factorial design (a subset of the full design). A d-efficient design was generated using NGENE software [41].

To estimate the minimum number of choice tasks required for our design, we used the formula:

$${S}_{\text{min}} = K/(A-1)$$

Where Smin is the minimum number of choice tasks, K is the total number of parameters to estimate, and A is the number of alternatives [40]. According to this, our minimum requirement was eight choice tasks. Usually, this estimate is multiplied by two or three to ensure an adequate sample size for the study. We tested different numbers of choice tasks including 20, 30, 40, 50, and 60 and compared the normalised d-error of each design (Fig. 1 of Online Resource 2, see ESM). Even though the normalised d-error reduced with the increase in number of choice tasks, the design with 30 choice tasks was selected considering it had near-best balance. A design achieves balance when each level of an attribute occurs an equal number of times [23]. While there is limited existing literature on determining the number of choice tasks per person, eight to 16 tasks are usually considered suitable [23]. For large designs, it is recommended to use ‘blocks’, which divides the main design into groups to ease the cognitive burden on participants [42]. Therefore, the 30 choice tasks were divided into three blocks of ten. In the absence of prior information on the coefficients of different attributes, small positive or negative priors (non-informative priors) were used. The design codes are provided in Fig. 2 of Online Resource 2 (see ESM). An example choice task is given in Fig. 2.

Fig. 2
figure 2

Example of choice set

Full size image

2.2.2 Pre-Testing

We conducted two think-aloud exercises with ten randomly selected choice tasks from the design [43]. Participants of the two think-aloud exercises were parents from the general population (n = 10 and n = 8, respectively). The participants were presented with one choice task at a time and asked to verbalise their thought process in selecting an option from ‘A’, ‘B’, or ‘neither’. They were also requested to indicate if they had problems understanding specific attributes or levels. Their feedback was used to refine the comprehension of choice tasks, including restructuring wording for specific attributes, clarifying the meaning of some levels, providing examples alongside levels for better understanding, and improving the hypothetical scenario provided. Participants also noted that the choice tasks were easy to understand, with seven attributes not overwhelming and ten choice tasks not overly burdensome. After addressing these adjustments, we proceeded with the first pilot test.

2.2.3 Pilot-Testing

The survey was administered to families of children aged below five years, not specifically targeting those with neurodevelopmental disorders. This broader approach was considered to engage a larger audience and obtain a comprehensive perspective. We used PureProfile, an online survey panel, to administer our survey. The survey had four parts: eligibility questions, ten DCE choice tasks, demographic details, and questions related to the comprehensibility of the survey. In addition to these ten choice tasks per participant, a repeated choice task was also included to assess the internal reliability and consistency of responses, creating 11 DCE choice tasks presented to each participant. We received 53 responses. After analysing the data using multinomial logit (MNL) model, using NLOGIT software [44] (analysis codes are provided in Fig. 3 of Online Resource 2, see ESM), we found that the cost attribute was dominant as its parameter estimate was higher than the rest of the attributes by a large margin. However, about 26% of the respondents, identified through demographic questions, were families or caregivers of children with neurodevelopmental disorders. In this subgroup, cost was not a dominant attribute. Because of this divergent finding, we decided to focus on this population and conducted another pilot study.

For the second pilot, we used the same experimental design as the first pilot and recruited families of children aged under five years with neurodevelopmental delays via PureProfile. The survey was administered, and we received 48 responses. After analysing the data using the MNL model, we found no issues regarding the dominance of any attributes. We used the parameter estimates obtained from this pilot as priors in the main DCE.

2.3 Stage 3: Final DCE Survey

For the final survey, choice sets were generated again in NGENE using priors obtained from the second pilot. We used a Bayesian d-efficient design optimised using a swapped algorithm with two Gauss draws. We compared this design to the second pilot survey design using the ‘eval’ function in NGENE (second pilot d-error = 0.43), and the efficiency of the Bayesian design was improved by 9% (Bayesian d-error = 0.39). The codes are provided in Fig. 4 of Online Resource 2 (see ESM). Bayesian efficient designs are known to enhance the reliability of parameter estimates and reduce sample size requirements [45]. After a design was produced, the survey was administered. The eligible survey participants were families or caregivers of children < 5 years of age with neurodevelopmental needs. They were recruited via PureProfile. Full DCE survey, participant information sheet, and more information about PureProfile is provided in Online Resource 3 in the ESM.

We used S-estimates to determine our sample size. The S-estimate is the minimum sample size required for a parameter, with Sb mean estimates incorporating the full range of Bayesian priors for a comprehensive interpretation [46]. In our study, Sb mean estimates indicated a minimum sample size of 11,772, which was impractical. Therefore, we opted for the next largest Sb mean estimate of 105. With a three-block design, our target sample size was approximately 315 participants. We ultimately collected 301 responses from the final DCE survey. More details and explanations on the design, priors, and sample size can be found in Online Resource 4 (see ESM).

2.4 Stage 4: Econometric Model and Data Analysis

A panel mixed multinomial logit (MMNL) model was used to identify the overall preferences at a population level; however, this output was presented in the supplementary information and not in the main analysis (Table 5, Online Resource 1, ESM). A panel latent class model (LCM) was used and presented as the main analysis due to its ability to account for preference heterogeneity and identify preferences within different groups of the sample [47]. In this study, we aimed to determine if decisions can be linked to certain demographic groups. By identifying 'latent' or ‘hidden’ groups with distinct preferences and needs, neurodevelopmental care services can be tailored to meet the specific requirements of each group, which may ensure that access to care is provided to maximum families and their children. Moreover, we considered the theoretical interpretability of different models and decided to use panel LCM due to its clearer understanding and potential implications.

Since LCM is probabilistic, respondents belong to a group up to a certain probability and can be allocated the group they are most likely to be a member of [40]. Posterior probabilities were estimated to describe the differences in sociodemographic characteristics across groups. For each respondent, the posterior probability that they belong to each latent class was calculated and they were assigned to the class with the highest probability. Class membership was determined solely from the estimated preferences from the LCM [48]. The panel LCM codes are provided in Fig. 5 of Online Resource 2 and information on model fits are provided in Table 6 of Online Resource 1 (see ESM).

The below utility was estimated:

$$V={\beta }_{0}+ {\beta }_{1}{\text{LOCATION}}_{\text{Public}}+{\beta }_{2}{\text{LOCATION}}_{\text{CityBased}}+{\beta }_{3}{\rm MODE}_{\rm Face-to-face} +{\beta }_{4}{\rm MODE}_{\rm Online} +{\beta }_{5}{\rm COST}_{ 0} +{\beta }_{6}{\rm COST}_{ 100} +{\beta }_{7}{\rm COST}_{ 500} +{\beta }_{8}{\rm COST}_{ 1000} +{\beta }_{9}{\rm MH}_{\rm Yes} +{\beta }_{10}{\rm EDUC}_{\rm Pamphlets} +{\beta }_{11}{\rm EDUC}_{\rm Face-to-face} +{\beta }_{12}{\rm EDUC}_{\rm Online} +{\beta }_{13}{\rm EDUC}_{ \rm Group \,\, Activities} +{\beta }_{14}{\rm APPOINT}_{ \rm Health \,\, Provider} +{\beta }_{15}{\rm WAIT}_{ 3 {\rm months}} +{\beta }_{16}{\rm WAIT}_{ 6 {\rm months}}$$

Where V is the observable relative utility and is a function of constant β0 and 16 parameters measuring the effect of neurodevelopmental follow-up care attributes described in Table 1.

Descriptive statistics were used to describe demographic characteristics of the overall sample and each latent class. The DCE data were categorical, therefore they were dummy coded. We conducted a scenario analysis (NLOGIT codes provided in Fig. 6 of Online Resource 2, ESM). By controlling attribute levels within hypothetical scenarios, researchers may assess respondents' sensitivity to variations, providing insights into potential policy outcomes, and informing decision-making processes [49, 50]. Moreover, to evaluate the extent to which each attribute contributes to the overall utility of follow-up care, we calculated the relative importance of attributes within each class [51]. To estimate the relative importance of each attribute, the difference between the highest and lowest coefficient of each attribute was divided by the sum of these differences across all attributes [48].

We performed Monte Carlo simulation to address the uncertainty associated with the parameters used in estimating the relative importance of attributes. For this purpose, a probability distribution was assigned to each parameter, reflecting the uncertainty surrounding its value [52]. This study used normal distributions (mean and standard error) to represent parameter estimates for each attribute level. Our Monte Carlo simulations consisted of 1000 iterations, whereby values were randomly sampled from the parameter distributions to generate simulated observations. These simulated observations were then used to estimate the mean and 95% confidence intervals of the relative importance of attributes in each class.

All analyses were done using NLOGIT Version 6 software [44] and R Version 1.3.1093 [53]. The output from the unconditional dataset was reported as the main result. Due to the small and incomparable sample size, the output from the conditional dataset is presented in Table 7 of Online Resource 1 in the ESM.

3 Results

3.1 Respondents’ Characteristics

Demographic details of the respondents are presented in Table 2. Overall, there were more female respondents than male (56.4% and 42.5%, respectively). Most respondents had completed a higher education qualification, including both undergraduate and postgraduate degrees. Over half of the respondents had full-time jobs and earned 60,000–120,000 Australian dollars (AUD). Approximately 64% of the respondents lived in metropolitan cities, and some lived in regional (27%) and rural (7%) areas. The median age of family members was 35 years. Most respondents’ children were diagnosed with ADHD (34%) or ASD (19%).

Table 2 Participant characteristics of overall sample and each class, number and percentage
Full size table

3.2 DCE Results

3.2.1 Latent Class Analysis

The results of the panel LCM are provided in Table 3. We tested the model with two and three classes, and ultimately selected the model with two classes because of its enhanced comprehensibility and clearer interpretability. The probability of individuals being in class 1 is more (63%) than them being in class 2 (37%).

Table 3 Latent class model with two classes (n = 301)
Full size table

Class 1 had more families with postgraduate degrees, full-time employees, and more individuals with higher incomes living in metropolitan areas when compared with class 2. Conversely, class 2 had more families with undergraduate degrees, part-time or casual employees, slightly lower income, and higher representation of individuals from regional areas when compared with class 1. Both groups showed similar age distributions and types of neurodevelopmental disorders, with most respondents' children diagnosed with ADHD and ASD. However, class 2 had a higher proportion of ADHD and ASD diagnoses, while class 1 had a higher representation of learning disabilities, intellectual disability, and conduct disorder.

The alternative specific constant (ASC) reveals that preferences for follow-up care options A and B were positive in class 1, indicating a tendency for individuals in this class to choose these options over the reference option (neither). On the contrary, preferences were negative in class 2, suggesting a preference against options A and B compared with the reference option. There was a significant heterogeneity regarding these preferences among respondents.

In class 1, local public health community clinics were preferred (0.97, 95% CI 0.45–1.50) when compared with private services. A similar trend was observed in class 2; however, it was not statistically significant. Moreover, city-based hospitals were disliked by families in class 2 (− 0.21, 95% CI − 0.44 to 0.02). Regarding modality of follow-up care, families from class 1 preferred face-to-face follow-up sessions (0.23, 95% CI 0.08–0.38); however, no significant difference was observed in preferences regarding this attribute in class 2. Regarding out-of-pocket costs, even though both classes 1 and 2 preferred to pay at least AUD100, the cost was more crucial in the decision-making process for families in class 2 (2.72, 95% CI 2.27–3.16) compared with those in class 1 (0.50, 95% CI 0.13–0.85). Both class 1 and class 2 preferred having mental health support provided to them (class 1: 0.13, 95% CI 0.01–0.25 and class 2: 0.43, 95% CI 0.23–0.64). While class 1 preferred group activities (0.27, 95% CI 0.06–0.48) and online learning (0.22, 95% CI 0.005–0.43) as ways of receiving educational information, there was no significant difference in the preferences regarding this attribute in class 2. Class 1 preferred health service providers initiating reminders and scheduling appointments for them (0.33, 95% CI 0.21–0.46), whereas class 2 was found to be indifferent to this aspect. Lastly, both classes preferred waiting <3 months for follow-up appointments compared with a waiting time of around 12 months.

3.2.2 Relative Importance

The relative importance of all attributes was estimated for each class and is reported in Table 4. In class 1, the most important attribute was location, driving 28% of the decision making, followed by cost (26%) and waiting time (16%). In class 2, cost accounted for nearly half of the importance when selecting an alternative (49%).

Table 4 Relative importance of attributes
Full size table

3.2.3 Scenario Analysis

The scenarios were discussed and finalised by the research team considering their practicality in the real world and alignment with the research question. The scenarios and analysis results are presented in Table 5. The probabilities of uptake for three hypothetical scenarios were compared with the probability of uptake for the base scenario (with reference levels of all attributes). For the overall sample, the base scenario showed uptake probability of 88.7%. For scenario 1, the probability of uptake increased to 98.6%, making it the most preferred or an ‘ideal’ follow-up care scenario. The probabilities of uptake for scenario 2 and scenario 3 are 88.9% and 96.8%, respectively. While there was not a considerable difference in preferences among all scenarios in class 1, class 2 showed a significantly higher preference for scenario 1 (99%) and a lower preference for scenario 2 (66%).

Table 5 Scenario analysis
Full size table

4 Discussion

The primary aim of this study was to identify families’ preferences for neurodevelopmental follow-up care for children. Overall, the findings suggested that both class 1 and class 2 families preferred accessing local public health community clinics and disliked city-based hospitals. Both classes favoured receiving mental health support and waiting < 3 months. Even though both classes preferred paying AUD100, the next preferred choice for class 1 was AUD500, whereas class 2 preferred no cost. While class 1 liked face-to-face follow-up, group activities for receiving educational information, and health service providers initiating reminders and scheduling appointments for them, no significant differences were observed in preferences regarding these attributes in class 2. The heterogeneity of follow-up care preferences may be of value to policy makers and health services officials that may facilitate care management service improvements for children and families.

Families in both classes preferred local public health community clinics, but this preference was not statistically significant in class 2. There is limited published literature on public care services, particularly for neurodevelopmental follow-up care for children. Some available services, such as Child Development Services in the state of Queensland, Australia, offer formal evaluation and diagnosis of neurodevelopmental delays, which are crucial for obtaining referrals for early interventions [54]. In a general healthcare context, these facilities may be located closer to home, which may save travel time and improve access, especially in rural or regional areas, and sometimes offer more services compared with private care [55, 56]. The cost of private services is relatively high; however, waiting time may be less and some families may be able to pay for private services to bypass public waiting lists. City-based hospitals may not be preferred due to their centralised nature, leading to extra travel time for patients, particularly if the families are from rural or regional areas [57]. Since there is a higher representation of families from regional areas in class 2, this may explain why they strongly dislike city-based hospitals. Additionally, a higher proportion of lower-income families in class 2 may contribute to their significant dislike, as increased travel time and costs pose a greater burden.

Families in both classes preferred face-to-face follow-up, although this was not statistically significant in class 2. This was somewhat contradictory, as recently, an increased trend of virtual care has been noted due to enhanced convenience and reduced waiting times [58,59,60]. Additionally, recent reports from the United States recommend integrating a telehealth model in neurodevelopmental assessments [61, 62]. There may be several potential barriers however, including technical issues, difficulty in ensuring confidentiality, limited control over distractions at home, difficulties in administering standardised testing virtually, and access to technology and electronic devices. However, virtual care may be further explored for those services where specialists do not need to physically examine the child. Even though it may improve access and efficiency in follow-up care, facilitating quicker connections for patients needing support and services, further investigation is therefore still required about its implementation to meet both family and heath service needs.

It is likely that the presence of higher-income individuals and more full-time employees in class 1 meant that cost was not the driving factor of decision making in this group. They preferred to pay at least AUD100 or AUD500 for their child’s treatment. Interestingly, the preference for AUD1500 over no cost in class 1 may be perceived as receiving more service benefits and better quality. However, the coefficient for AUD0 was statistically insignificant, so a definitive conclusion should not be made. However, for class 2, which was associated with comparatively lower-income individuals with more part-time or casual jobs, cost was the most crucial deciding factor (nearly 50% relative importance for this attribute). They preferred to pay AUD100 followed by paying no cost. Other studies have shown cost as a crucial factor in decision making [63,64,65,66], especially for a lower-income group as they often encounter significant financial burden [67, 68]. In Australia, the National Disability Insurance Scheme also financially supports those who qualify, but the qualification process may be time consuming and difficult to access for some individuals [69].

Families of both classes preferred mental health support. Mental health for families may be a lower priority when they are concerned about their children. Previous studies have indicated that parents of children with neurodevelopmental needs often experience poor wellbeing, anxiety, and depression [70,71,72,73]. Despite this information, very few neurodevelopmental follow-up services have included mental health support for families as a crucial consideration. For example, a cardiac neurodevelopmental programme in Canada counselled 100% of the families in that programme on neurodevelopmental outcomes [74]. Moreover, studies have reported that financial stress may also negatively influence mental health and wellbeing [75, 76], which may be the reason that mental health support was relatively more important for class 2 (more low-income families), but the least important attribute for class 1 (more high-income families). This may inform decision makers to prioritise families’ mental health to reduce burden, improving their resilience related to the neurodevelopmental outcomes of their children.

Providing appropriate educational resources to families regarding neurodevelopmental delays and problems specific to their children is important. Even though class 2 families did not think this attribute was important, class 1 families considered group activities such as playgroups or peer support classes and online learning material helpful. Prior evidence shows that playgroups or peer support activities have been beneficial for families as they offer opportunities to connect with others with similar experiences, feel accepted, enhance their knowledge and skills, as well as provide an environment for child development, socialising, and enjoyment [77, 78]. Online learning material or training for families may also be valuable as it may make them self-reliant and skilled at addressing parenting stress or issues with their children, especially if they do not have the time to attend traditional classes [79, 80]. Preliminary evidence from a study in Canada indicated high parent satisfaction and improved parenting confidence, parenting skills, and child behaviour with the development of an online training programme that included educational modules and videoconference sessions with a therapist [81]. On the contrary, educational information provided face-to-face via a professional was not as effective, possibly due to its higher cost or the time commitment required, with sessions typically lasting 45–90 minutes [79, 82]. Moreover, studies have shown that using a phone application can be a useful tool for educational information purposes and accessing support related to neurodevelopmental follow-up care [83, 84]. It may be interpreted for class 2, which had more families with a lower income, that these educational information methods were not deemed necessary as they may have an additional cost.

In terms of managing appointments, it had no significant effect on class 2 families, but class 1 families preferred health service providers initiating reminders and scheduling appointments for them. A recent study reported that when health service providers scheduled appointments for neurodevelopmental evaluation, adherence was higher compared with when families scheduled appointment themselves [85]. This may lead to more effective and quick formal evaluations for children, identifying delays earlier and providing appropriate treatments, thus facilitating better patient-centred care. Our interviews confirm that implementing such systems could serve as a safety net. However, flexibility should also be provided based on the availability of family members, hence forming a hybrid care management system.

Waiting time was crucial for both classes. In healthcare, this issue has persisted where people often wait excessively to access services. This may be detrimental in the case of neurodevelopmental issues, as by the time families seek assistance for their child, the issue may have already progressed, with considerable literature highlighting this concern [86,87,88,89,90,91]. The latest findings from the Sydney Child Neurodevelopment Research Registry suggest that the average age that caregivers identified developmental concerns was 3 years of age, but the average age for receiving a developmental assessment was about 6 years [92]. These waiting delays may challenge implementing early intervention and support policies that depend on timely developmental assessments. This is further confirmed by one of our interview participants, who reported waiting for 3 years for their child's follow-up visit. During this waiting period, they experienced anxiety due to uncertainty about their child's condition and the lack of access to suitable treatment or support.

The scenario analysis suggests that there may be a likely impact of different scenarios on programme uptake. It shows the importance of considering various system-level aspects to optimise the effectiveness of neurodevelopmental follow-up care for children and families. Even though the different scenarios did not make notable differences in preference for class 1, the significant uptake increase observed in the first scenario—nearly 10% for the overall sample and nearly 24% for class 2—indicates that future programmes may be designed to account for the preferences identified in that scenario. This may include policies designed to reduce out-of-pocket costs, improve access to local public health community clinics, minimise waiting times, provide families online learning material, promote mental health counselling, and enhance coordination between healthcare providers and families. Other studies have conducted scenario analysis including DCEs regarding maternal health, doctor attendance, vaccination preferences, hypertension management, and environmental health fields [49, 50, 93,94,95,96], consequently informing policy simulations.

4.1 Further Steps

This study has a range of implications. The findings may be useful for Child Health Services and Child Development Services to enhance their neurodevelopmental follow-up care services. Moreover, health insurance policymakers may use these findings to step up to cover the costs of more services, including providing education and mental health support to families. Methodology-wise, DCEs may be conducted considering different perspectives (e.g., taxpayers’ perspective). Since taxpayers fund these services as a society, eliciting their preferences regarding the structure and delivery of care may also be an essential voice. The findings may also be integrated with economic evaluations to predict service uptake or cost effectiveness of neurodevelopmental care services. Furthermore, virtual care solutions may be explored to enhance service accessibility and reduce waiting times. Implementing such strategies may ensure practical use of healthcare resources and improved overall patient experiences. While including children's preferences was outside the scope of this study, it is a crucial area for future research, as also suggested by another study [97].

4.2 Strengths and Limitations

While this may not be considered as a strength or a limitation, it is worth noting that most of the sample in this study were better educated than the general population, which might imply that they may be more informed and proactive in seeking neurodevelopmental care for their children; this may have influenced preferences and decision-making processes.

It can be challenging to conduct DCEs in neurodevelopment-related contexts. Generating an apt design that focuses on the specific population and is methodologically correct remains imperative. Previous DCEs conducted in this context have evident limitations in their designs. For instance, absence of justification for forced choices (i.e., respondents not having the freedom to opt out if they were not interested in an option [98, 99]); using many attributes (e.g., n = 20) and choice tasks (e.g., n = 30) [100,101,102]; and not conducting pilot tests [10, 103]. These shortcomings were addressed in our study. The inclusion of a ‘neither’ option made the choice task realistic without forcing a participant to choose from the given choices. Our pre-test determined cognitively appropriate numbers of attributes, levels, and choice tasks to ensure that they do not cause a cognitive burden on respondents. The pilot studies informed us about the availability of the target population, due to which the priors we obtained may be comparatively efficient. Several trials in designing the DCE were useful for various reasons, including gaining an understanding of the sample size, obtaining priors, finalising the structure or function of the choice tasks and the overall DCE survey.

Despite several strengths of this study, there were some limitations. Firstly, the qualitative attribute development phase; even though our literature review and expert advice were appropriate for the qualitative groundwork for the DCE, we experienced challenges with recruitment as we encountered a case of participant fraud in online focus groups and interviews. Although it interrupted the study, we identified it early and implemented strategies to prevent this in further recruitment. Moreover, we identified ‘red flags’ for identifying imposter participants in the future [36]. Secondly, we reported the unconditional dataset as the main output. Even though combining the unconditional and conditional datasets increases the dataset's quality, the combined behavioural model would imply something completely different compared with the unconditional one, which is the true choice of the sample. Thirdly, another limitation of our study is the complexity of the ‘location’ attribute, which may encompass more than just the geographical aspect of healthcare services (e.g., the type of provider). In another instance, the implication of travel time may vary depending on the respondent's location as ‘local’ clinics are often presumed to be nearby for all respondents, while ‘city-based’ hospitals may be more accessible for urban respondents than for those in rural areas. This variability could influence preferences based on individual travel constraints and convenience. Moreover, different levels of the location attribute might imply differences in service quality, and some respondents may have ideological preferences for public versus private providers. Future studies could benefit from disentangling these factors to provide clearer insights into preferences related to healthcare service locations. Fourthly, in terms of analysis, we considered conducting a sub-analysis for various neurodevelopmental disorders, but the sample size (overall and of the two classes) for each diagnostic group were insufficient. Additionally, we did not use a statistical significance test to assess differences between classes as the primary goal of our study was to provide a descriptive overview of the characteristics of the identified latent classes. Given the exploratory nature of our analysis, we aimed to offer an understanding of the profiles within each class rather than to test for statistical differences. Furthermore, the use of statistical significance tests may not be appropriate in this context due to the assumptions required by these tests, which may not be met by the latent class analysis-derived classes. Therefore, we chose to present the descriptive characteristics to highlight the practical relevance and implications of our findings. Lastly, this study represents the combined preferences for all neurodevelopmental disorders or delays. The scope of this study was limited to providing an overview of preferences for informing a healthcare service design perspective rather than individual diagnosis. Given the diversity within neurodevelopmental disorders and individual circumstances, preferences may vary widely among consumers within the same diagnostic category and across different diagnostic groups. Moreover, health system designs internationally have not typically established different health systems and services for individual diagnostic groups, but rather aim to provide comprehensive services that cater to a spectrum of at-risk children regardless of whether they have specific diagnoses upon entering services. Our analysis also supports this information as the sociodemographic characteristic 'types of neurodevelopmental disorders' did not show any evidence suggesting that families of children with different conditions preferred different service models. We have tried to be inclusive of a range of diagnoses and disorders in this study. Therefore, the findings arising from this study should not be interpreted as representing the preferences of all families who may access neurodevelopmental follow-up care.

5 Conclusion

The design and delivery of neurodevelopmental care services for children have room for improvement. We conducted a DCE that identified families’ preferences for neurodevelopmental follow-up care for their children. The findings suggest that health services may benefit from prioritising aspects including making local public health community clinics more accessible, reducing waiting times and costs, providing mental health and educational resources to family members, and organising better care coordination for follow-up visits. Implementing these evidence-based findings may support resource allocation and contribute to improved outcomes for children with neurodevelopmental needs and their families.