Part 1: Empirical analysis

Methods

Participants and procedure

American participants (n=242) were recruited online through Amazon Mechanical Turk (MTurk) and paid $1 for their time. MTurk is a website that facilitates payment for completing Human Intelligence Tasks (HITs) posted by researchers, and is widely used in social science research because it provides participants who vary in age, socioeconomic status and ethnic background10 and are broadly representative of the US population. A power analysis revealed that a minimum of 128 participants was required to detect a moderate effect size (d=0.5) at the 5% level with 80% chance. Participants were excluded from data analysis if they failed to complete the survey (n=33) or failed a simple attention check11 (n=31). Thus, our final sample used in analysis consisted of 178 participants.

Participants completed the study through the online survey tool Qualtrics. The survey was piloted for face validity with a cohort of students and colleagues, and the order of the questions within all sections was randomised to reduce order effect.12 Ethics approval for this study was provided by the Social Sciences and Humanities Inter-Divisional Research Ethics Committee of the University of Oxford.

Design

We hypothesised that when faced with a request for a substitute medical treatment, members of the public would be increasingly inclined to refuse parents’ requests as the comparative efficacy decreases, and as the expense increases. We predicted that survey respondents would be more likely to allow requests based on religious rather than non-religious reasons.

Sections 1 and 2

The first two sections of the study had a mixed-subjects design: a between-subjects factor of whether the reasons for requesting the substitute treatment were ‘religious’ or ‘ethical’; and a within-subjects factor of whether this treatment was less efficacious or more expensive (figure 1).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Flow chart of survey structure (see full survey; Appendix A). SRT, surfactant replacement therapy.

Supplementary file

[SP1.pdf]

Respondents were asked to imagine themselves as health professionals facing a request for substitute treatment (using the example of surfactant replacement). Participants were informed that without treatment the infant was stated to have a 30% absolute risk of mortality, and with the standard form of treatment this risk was reduced to 15% (these figures were based on data13 on the effectiveness of surfactant replacement therapy, but slightly exaggerated and simplified for the sake of clarity). Respondents were randomised to receive one of two versions of these sections of the survey; 50% were told the parents’ reason for objecting was ‘religious’ and the other 50% were told the parents’ reason for objecting was ‘ethical.’

To look at the influence of efficacy and expense on support for substitute treatments, we asked participants to respond to a series of variations of the case example. First, participants were told that the hypothetical parents had either refused treatment or had requested treatments with varying degrees of reduced efficacy. The substitute treatments had a higher risk of pneumothorax needing drainage (higher risk of a painful procedure), intraventricular haemorrhage (higher risk of disability) or of death (4%, 9% or 14% absolute increase in risk: see figure 2A). Next, participants were told that the substitute treatments requested were equally effective, but associated with varying degrees of additional cost (see figure 2B). To provide context for the cost of treatment, indicative examples were provided of other medical treatments with similar expense.14 Respondents were asked to indicate their level of agreement with allowing these parental requests on a 7-point Likert scale (1=strongly disagree, 7=strongly agree).

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

(A) Sample question from the survey, investigating willingness to provide less effective treatment. (B) Sample question from the survey, investigating willingness to provide more expensive treatment.

Section 3

This part of the study assessed all participants’ agreement to different reasons behind parental requests. There were five versions in a within-subjects design presented to all participants: three religious reasons for objection (Muslim, Christian and Jewish parents) and two non-religious reasons (vegan parents and parents with an irrational fear their child will develop an allergy to the medication). For each, respondents were asked whether they would agree to provide either a less effective substitute (with a higher, but unspecified chance of death) or a more expensive substitute (double the cost of the standard treatment).

Section 4

In order to allow comparison between requests for suboptimal treatment and other examples of parental discretion around treatment, the fourth part of the survey assessed respondents’ willingness to accede to two other widely discussed controversial parental requests. They were asked if they would agree to prolong treatment deemed futile by medical professionals, or allow parents to decline a blood transfusion in a severely anaemic child. For each case, the respondents were provided with two versions, one where the parents’ request was based on a religious reason and the other on a non-religious reason.

Section 5

Finally, to look at how beliefs about the acceptability of suboptimal treatments might differ across different groups, we collected basic demographic information. Given that some of the reasons the hypothetical parents gave for refusing treatment were religious ones, we measured how religious participants considered themselves using the Central Religiosity Scale (CRS).15 The CRS (α=0.93) consists of five items tapping different dimensions of religious belief (eg, private practice, public practice and religious experience). Similarly, because some of the reasons the hypothetical parents gave were based on ethical views relating to the use of animals, we measured participants’ concern with animal welfare using an 8-itemed Speciesism Scaleⁱ (α=0.87).

Statistical analysis was conducted using SPSS Statistics V.22 for Windows (IBM). Paired samples t-tests were used to compare the participants’ responses between questions. Independent samples t-tests were used to investigate the differences between the participants given religious-based requests and those given non-religious versions. A p value of <0.05 was considered significant.

Part 2: Ethical analysis

Our data show how some members of the general public respond to hypothetical cases of parental requests for substitute medical treatment, but how should we respond? A detailed theoretical analysis is beyond the scope of the paper, however, as a starting point we draw on a recently published systematic review.9 We summarise several existing ethical frameworks that have been used to decide about parents’ medical decision making for their child and apply them to the clinical scenario summarised at the start of this paper (box 1). We compare empirical findings to theory, with the goal of achieving coherence between ethical principles and data gathered of people’s beliefs and intuitions, working towards wide reflective equilibrium.28 We will propose a preliminary framework building on this analysis.

Several normative models have previously been proposed to assess whether or not it is justified for a physician to over-ride parents’ medical decision making for their child. McDougall and Notini, in a recent paper, identified nine different frameworks.9 Table 1 summarises a subset of the most relevant of these frameworks and includes another, more recent, one which deals explicitly with parental requests as well as refusals.29

Applying these frameworks to the case example yields varying outcomes (table 2). A strict interpretation of the Best Interests Standard would mean denying any parental request for a less effective treatment.30 It is not clear whether this standard would permit more expensive (but equally effective) substitutes. However, other frameworks did not yield a clear answer to whether parents should be permitted to choose suboptimal substitute treatment. Although Rhodes and Holzman’s ‘Not Unreasonable Standard’31 classifies beliefs held by only subgroups of the population (such as religion) as third domain judgements which can be reasonably rejected (table 1), the authors argue that parents’ decisions should only be over-ruled if the prognosis is poor; and it is unclear how poor that prognosis must be. DeMarco et al’s ‘Balance of Costs and Benefits’ framework32 does not state how one ought to weigh the costs and benefits of treatment to come to a decision. Diekema’s Harm Principle and Gillam’s Zone of Parental Discretion appear to map most closely onto current clinical decision makingⁱⁱⁱ but do not clearly state what constitutes a ‘significant risk’ or a ‘serious harm’ and thus fail to yield a clear answer.

Where should we draw the line on parental requests for substitute medical treatment? There is no clear theoretical answer to this question. Although empirical surveys do not settle the question either, the intuitions of the general public provide a potential starting point for achieving coherence between intuitions and theory using the reflective equilibrium.28 Empirical evidence that is inconsistent with one’s position should encourage further analysis, and re-evaluation of previous reasoning. For example, if the general public believe that a 5% risk of death is acceptable, yet an ethicist may only accept 0.01%, this says that the general public value parental autonomy more than the ethicist. The ethicist ought to assess the value of parental autonomy and consider what arguments could justify raising the threshold above 0.01%, even though those arguments might not support going all the way to 5%. That is, the empirical evidence causes us to reconsider arguments and reasons, and perhaps choose another point within a justifiable options space.

There are some apparent inconsistencies between the ethical frameworks and the results of our empirical survey. First, a significant proportion of our cohort judged it acceptable for parents to make decisions that were associated with serious risks for the child. Even if the frameworks are unclear about requests for less effective treatments, it seems that a 15% absolute increase in the risk of death for a child (as was the case with an outright refusal of treatment) represents a significant risk of serious harm, and would not be supported by any of the models. Second, in our survey, there was a clear bias towards religious reasons and against non-religious ethical beliefs such as veganism. This appears to demonstrate discrimination against parents with sincere non-religious beliefs. However, ethicists have argued that patient or parent’s religion should not be considered a relevant difference when considering otherwise alike cases.29 33 Of the models described in table 1, only the ‘Not unreasonable framework’ makes a clear difference based on the reason behind parental requests. In that model, central core judgements provide an acceptable reason for parents to make requests for suboptimal treatment. In contrast to our survey findings, that framework suggests that a secular preference for non-animal-based treatment should be given more weight than a religious reason.

A majority of the general public surveyed were comfortable allowing parents to choose treatment that would result in up to a <5% increase in the absolute risk of death. If this were used as the threshold for over-ruling parental decisions, 20 parents (or fewer) would need to be over-ruled to prevent one death. Five per cent represents a statistical threshold often used in scientific studies for judging low probability of error, as well as the threshold that some have argued should be applied to judgements ‘beyond reasonable doubt’ in criminal prosecutions,34 however, some may regard 5% as still too high a level of risk to impose on a child for the sake of parental values. Other risks that children are subjected to by their parents, such as passive smoking or unsafe driving, would not increase risk of death by as much. The risk of death from driving in a year is approximately 1/10 000,35 and is universally regarded as an acceptable risk for parents to impose on their children. Perhaps the threshold should be set at 1%, 0.1% or 0.01%? We cannot definitively answer that question here.

In fact, decision theory would suggest that there are two variables in setting a threshold level of harm: the severity of harm and the probability of this harm occurring.36 It may be reasonable to allow a higher probability where the severity of the harm is less. If an expected harm approach (drawing on decision-theoretic consequentialism) is the correct approach to medical uncertainty, we could extrapolate from the empirical results to create a set of potential harm thresholds to decide when a risk should be tolerated (table 3).

What cost would be acceptable for parental choice? The results of this study demonstrate a willingness to provide treatment up to $500 more expensive at parental request. The plausibility of this threshold will depend on the availability of resources within a public health system. Very wealthy countries may be able to afford a higher amount, while countries that struggle to afford basic health services might decide to prohibit any parental choices that would be more costly. In practice, the $500 threshold would potentially rule out many substitute treatments, since small differences in the efficacy of treatment could lead to additional medical costs including additional bed days or need for future medical interventions.

Wherever we set the threshold for more expensive treatment, one option will be to permit parents to pay the excess cost of their preferred treatment. This would largely remove the ethical reason to decline parental request (assuming the treatment was not more harmful), and we would propose this should be allowed. However, this option would conflict with strict egalitarian policies like those adopted in the National Health Service.37

How should doctors take into account the reasons behind parental requests? One option would be to vary thresholds for cost or harm depending on the reason. On this basis it may be ethical to allow a more harmful or more costly choice where parents are motivated by a particularly weighty reason. However, such variable thresholds might well be anticipated to lead to significant inconsistency in practice, as well as bias against certain groups due to physicians’ variable interpretation of the weightiness of reasons. On the other hand, it appears unacceptable to allow parents to choose suboptimal treatment for any reason whatsoever: parents should not be able to choose more harmful or more costly treatments for trivial or irrational reasons. In such cases, the cost borne by the child will outweigh the benefit gained by the parents, if any. (The latter might yield genuine questions about the competence of parents to make decisions for their child.)

The Reasonable-Choice Threshold Model (figure 6) combines the above ethical conclusions into a model for arbitrating parental decisions around refusal of treatment or requests for suboptimal treatment.^iv

In this model, the doctor should first assess the competency of the decision maker, next consider the acceptability of their reason and finally, if the reason is acceptable, the request should be subjected to fixed cost and harm thresholds. Which reasons are acceptable? Intuitively, those that are clearly articulated, motivated by genuine concern for the child’s well-being, persistent and represent core values associated with long-standing tradition would be deemed acceptable by many, this may include religious values. However, secular reasons may also potentially fulfil these criteria, and the model should not discriminate between religious and non-religious adherents. A full articulation of the distinction between acceptable and unacceptable reasons requires further research and lies beyond the scope of this paper.

• Download figure
• Open in new tab
• Download powerpoint

Figure 6

Reasonable-Choice Threshold Model for parental choice around treatment.

Applying the Reasonable-Choice Threshold Model to our case from the beginning of this paper (box 1) yields the conclusion that the parents’ request should arguably be permitted. We did not provide information about parents’ competence to decide, but this may be assumed for the sake of argument. Is the reason acceptable? Islam is a well-established religion with many millions of followers practising the same customs for thousands of years. Respect for the large number of adherents and longevity of the tradition would arguably make such a reason acceptable. The request must then be subjected to the cost and harm thresholds. There are relatively little contemporary data on the relevant costs of surfactant therapy, however, one study comparing the costs of acquisition, administration and adverse events found beractant (bovine) to be $493 more expensive than poractant alfa (porcine)38; on the model above it sits just below the cost threshold. Is the level of harm acceptable? Focusing just on the most serious risk—that of death, the absolute risk reduction in mortality associated with poractant alfa is 6.4% and beractant is 10.9%,39 which is about a 4.5% absolute increase and sits below the harm threshold as set out in table 3.