IVF and the evaluation of policies that don’t affect particular persons

Over at the CLAHRC West Midlands blog, Richard Lilford (my boss, I should hasten to add!) writes about the difficulties with the economic evaluation of IVF. The post notes that there are a number of issues that “are not generally considered in the standard canon for health economic assessment” including the problems with measuring benefits, choosing an appropriate discount rate, indirect beneficiaries, and valuing the life of the as yet unborn child. Au contraire! These issues are the very bread and butter of health economics and economic evaluation research. But I would concede that their impact on estimates of cost-effectiveness are not nearly well enough integrated into standard assessments.

We’ve covered the issue of choosing a social discount rate on this blog before with regards to treatments with inter-generational effects. I want instead to consider the last point about how we should, in the most normative of senses, consider the life of the child born as a result of IVF.

It puts me in mind of the work of the late, great Derek Parfit. He could be said to have single-handedly developed the field of ethics about future people. He identified a number of ethical problems that still often don’t have satisfactory answers. Decisions like funding IVF have an impact on the very existence of persons. But these decisions do not affect the well-being or rights of any particular persons, rather, as Parfit terms them, general persons. Few would deny that we have moral obligations not to cause material harm to future generations. Most would reject the narrow view that the only relevant outcomes are those that affect actual, particular persons, the narrow person-centred view. For example, in considering the problem of global warming, we do not reject its consequences on future generations as being irrelevant. But there remains the question about how we morally treat these general, future persons. Parfit calls this the non-identity problem and it applies neatly to the issue of IVF.

To illustrate the problem of IVF consider the choice:

If we choose A Adam and Barbara will not have children Charles will not exist
If we choose B Adam and Barbara will have a child Charles will live to 70

If we ignore evidence that suggests quality of life actually declines after one has children, we will assume that Adam and Barbara having children will in fact raise their quality of life since they are fulfilling their preferences. It would then seem to be clear that the fact of Charles existing and living a healthy life would be better than him not existing at all and the net benefit of Choice B is greater. But then consider the next choice:

If we choose A Adam and Barbara will not have children Charles will not exist Dianne will not exist
If we choose B Adam and Barbara will have a child Charles will live to 70 Dianne will not exist
If we choose C Adam and Barbara will have children Charles will live to 40 Dianne will live to 40

Now, Choice C would still seem to be preferable to Choice B if all life years have the same quality of life. But we could continue adding children with shorter and shorter life expectancies until we have a large population that lives a very short life, which is certainly not a morally superior position. This is a version of Parfit’s repugnant conclusion, in which general utilitarian principles leads us to prefer a situation with a very large, very low quality of life population to a smaller, better off one. No satisfying solution has yet been proposed. For IVF this might imply increasing the probability of multiple births!

We can also consider the “opposite” of IVF, contraception. In providing contraception we are superficially choosing Choice A above, which by the same utilitarian reasoning would be a worse situation than one in which those children are born. However, contraception is often used to be able to delay fertility decisions, so the choice actually becomes between a child being born earlier and living a worse life than a child being born later in better circumstances. So for a couple, things would go worse for the general person who is their first child, if things are worse for the particular person who is actually their first child. So it clearly matters how we frame the question as well.

We have a choice about how to weigh up the different situations if we reject the ‘narrow person-centred view’. On a no difference view, the effects on general and particular persons are weighted the same. On a two-tier view, the effects on general persons only matter a fraction of those on particular persons. For IVF this relates to how we weight Charles’s (and Diane’s) life in an evaluation. But current practice is ambiguous about how we weigh up these lives, and if we have a ‘two-tier view’, how we weight the lives of general persons.

From an economic perspective, we often consider that the values we place on benefits resulting from decisions as being determined by societal preferences. Generally, we ignore the fact that for many treatments the actual beneficiaries do not yet exist, which would suggest a ‘no difference view’. For example, when assessing the benefits of providing a treatment for childhood leukaemia, we don’t value the benefits to those particular children who have the disease differently to those general persons who may have the disease in the future. Perhaps we do not consider this since the provision of the treatment does not cause a difference in who will exist in the future. But equally when assessing the effects of interventions that may cause, in a counterfactual sense, changes in fertility decisions and the existence of persons, like social welfare payments or a lifesaving treatment for a woman of childbearing age, we do not think about the effects on the general persons that may be a child of that person or household. This would then suggest a ‘narrow person-centred view’.

There is clearly some inconsistency in how we treat general persons. For IVF evaluations, in particular, many avoid this question altogether and just estimate the cost per successful pregnancy, leaving the weighing up of benefits to later decision makers. While the arguments clearly don’t point to a particular conclusion, my tentative conclusion would be a ‘no difference view’. At any rate, it is an open question. In my rare lectures, I often remark that we spend a lot more time on empirical questions than questions of normative economics. This example shows how this can result in inconsistencies in how we choose to analyse and report our findings.



Chris Sampson’s journal round-up for 28th March 2016

Every Monday our authors provide a round-up of some of the most recently published peer reviewed articles from the field. We don’t cover everything, or even what’s most important – just a few papers that have interested the author. Visit our Resources page for links to more journals or follow the HealthEconBot. If you’d like to write one of our weekly journal round-ups, get in touch.

How did medicaid expansions affect labor supply and welfare enrollment? Evidence from the early 2000s. Health Economics Review [PubMed] Published 22nd March 2016

In the early 2000s, a number of states in the USA expanded Medicaid while others did not. These expansions covered similar populations to those that are likely to benefit from the Affordable Care Act. This study combines county-level statistics on unemployment, wages and enrollment in welfare programmes and uses a difference-in-differences regression to look at the effects of the expansion. The study is based on the assumption that some counties would be more affected than others, and that these groups can be identified based on the level of poverty before the expansion. Counties above the 75th percentile of poverty rates are considered the ‘treatment group’. The poorer counties saw a decrease in labour force participation rate, a decrease in working hours, an increase in wages and an increase in the use of food stamps. The main identification strategy isn’t very convincing, but the author argues that the findings are robust to alternatives. If we believe it, then the Affordable Care Act might have some modest negative consequences for employment and welfare enrollment.

Extrapolating survival from randomized trials using external data: a review of methods. Medical Decision Making [PubMed] Published 22nd March 2016

Modelling is built on assumptions. Many of these will be about how we extrapolate outcomes beyond what is observed in a trial. This study reviews the ways in which cost-effectiveness studies have extrapolated survival data using external information to inform the extrapolation. Such external information might come from national statistics such as life-tables, or from large cohort studies. The authors started with NIHR HTA reports and adopted a ‘pearl growing’ approach to identifying relevant studies. Based on their findings, the authors present a framework for the assumption process. This can be followed to determine which kind of approach to extrapolation should be adopted; for example, depending on whether the control group is assumed to have the same mortality as the external population or whether this differs in the short term. The authors describe the approach that should be taken in each circumstance. It’s also important to think about uncertainty. In particular, there is likely to be uncertainty regarding the effectiveness of the treatment. No studies were identified that formally used external data to quantify future changes in treatment effects. The authors discuss the potential for the use of expert elicitation to inform survival extrapolation using Bayesian inference. If you’re building a model that requires survival data from a trial to be extrapolated, you’ll find this review to be very helpful.

Harsh parenting, physical health, and the protective role of positive parent-adolescent relationships. Social Science & Medicine Published 21st March 2016

There’s plenty of evidence showing that being loved by one’s parents is crucial to development. A new study of 451 adolescents followed into adulthood supports this. 12 year olds and their families were recruited for a study in Iowa, and data were collected at multiple time points up to age 20. Harsh parenting was identified by coders who watched videotapes of parent behaviour. Harsh parenting behaviours were hostility, angry coercion, physical attacks and antisocial behaviour. Adolescents’ self-assessed health and BMI were recorded throughout the study, as was their own judgment of parental warmth for each parent. The authors use a latent change score model to investigate associations between these variables. Harsh parenting was associated with a negative impact on future self-reported health and BMI. In terms of self-reported health, a positive relationship with the father helped mitigate the health impact of having a harsh mother. But then the effect on BMI seemed to increase with warmth from the other parent. The evidence suggests that as well as preventing harsh parenting, it may be worthwhile focussing on a child’s relationship with the less harsh parent as a means of buffering against the negative effects.

Analyzing health-related quality of life in the EVOLVE trial: the joint impact of treatment and clinical events. Medical Decision Making [PubMed] Published 17th March 2016

This study reports on the EQ-5D data from a clinical trial of cinacalcet for secondary hyperparathyroidism. Using the normal approach of estimating QALYs based on the area under the curve, no difference was identified between the two arms of the trial. But then trials like this aren’t designed to identify differences in health-related quality of life. This study explores an alternative approach. EQ-5D-3L was collected at baseline and 6 follow-up points from 3547 subjects. It was additionally collected after particular clinical events. A regression model using a generalised estimating equation (GEE) approach was fitted with EQ-5D index scores as the dependent variable and clinical events as explanatory variables along with baseline utility and trial allocation. The analysis looked at acute effects (on utility within 13 weeks) and chronic effects (on utility in all subsequent months). A regression analysis with just trial allocation as an explanatory variable only found a non-significant treatment effect. However, the GEE regression that controlled for the acute and chronic effects of clinical events was able to identify a small but significant beneficial effect of the treatment. The effect could be observed independent of the effect of clinical events. Whether such results will be as convincing as traditional trial comparisons will remain to be seen, but adopting an approach of this sort could be far more informative when determining parameters for a decision model.