On the possible effects of free GP care for under 6’s.

(This is a guest post by Dr Kevin Denny of the School of Economics & Geary Institute, UCD).

The extension of free GP care to under 6’s has raised the issue of what the effect will be on GPs’ practices. Understandably they are concerned about the increased demand on a sector that appears to be under strain. I am not aware of any specific research for this age group (mea culpa if I have missed it).

There are several studies for the general populations (various convex combinations of Anne Nolan, Brian Nolan & David Madden). I heard a very interesting recent interview on TodayFM with a GP Ciara Kelly. In the course of this, she said that children with medical cards visit a doctor 6 times a year while those without visited twice. From this she inferred that the new scheme would triple the demand of those currently without free care.A GP that I was once on a radio panel with said something very similar. I don’t know the origins of these numbers. However the average difference between the two groups is not relevant here: you need to know the marginal effect. Children currently without medical cards are different from those with: on average they are, inter alia, healthier and wealthier.

Estimates from other countries are not informative. I searched in vain for a government document that discussed this. We now have very good data in the form of the Growing Up in Ireland study. Here I report some estimates of what the effect of the reform might be. I use the second wave of the infant cohort – the three year olds who should be reasonably representative of the 0-5 age range. As dependent variable I use the question on “Number of times seen or talked with a general practitioner in the last 12 months”. This is top-coded at 20 but there are very few in that category. The variable of interest is whether they are covered by a medical card. I combine GP-only and the full medical card for simplicity.

I have a long list of controls which covers the usual suspects. They include health of the child and mother, income, education and other demographics. I also have a variable that indicated whether their GP visits are covered by private health insurance. Changes to the controls do not make much difference.

There are numerous ways of estimating such models and I used three. For the cognoscenti these are poisson & neg-bin2 regressions and a finite mixture of poissons. The marginal effects are very similar as is often the case. Before we consider those, what is the average difference in the data? In the GUI the population weighted mean of doctor visits for children covered and not-covered by medical cards is: 3.13 and 2.18 respectively. This is very different from the 6 and 2 mentioned above, the source of which I don’t know.

The marginal effects for the different models vary between 0.632 and 0.713, less than the average difference (as you would expect) and a lot less than the difference of 4 mentioned above. For simplicity I will take 0.68 as a ballpark value. So giving free GP care for under 6’s should increase the number of GP visits per child by less than one per annum. We are assuming homogenous effects: you could generalize this to allow the effects differ in various ways. The marginal effect of having private insurance is about 0.34. Since these are probably the better-off of non-medical card holders, this suggests that 0.68 is on the high side i.e. the effect on the kids of the rich of free GP care is probably lower, if anything.

I also estimated the model using the child cohort (the 9 year olds) for whom the marginal effect is about 0.33 incidentally. Estimates for adults tend to be in the 1-2 visits per annum range. So what might this mean in practice? The maximum number of children who could be covered by the present reform is about 270,000. Multiply by 0.68 & this suggests an extra ~183,600 GP visits a year. There are around 2,500 GPs in Ireland so this is about 73.5 visits a year each. If they work on average 47 weeks a year this would mean about 1.56 extra visits a week from the under-6’s for a GP. It would be interesting to know how much of a GP’s time this is likely to require.

The mean is not the only parameter in town. For doctors whose patients are already covered there will be little or no difference. Doctors in more affluent areas will likely bear the brunt. Doubtless there are additional complications.

For example, not all GP’s will sign up. I am ignoring general equilibrium effects, such as any ensuing change in the number of GPs. Perhaps the main known unknown is the labour supply responses of GPs to a switch from a per-visit fee to a capitation grant which encourages them to take on patients but spend as little time as possible with them. Extrapolation is difficult, especially about the unknown.

I don’t mean to suggest that my estimates are best, I can’t explore every possibility in a blog post. I think they are credible though. Readers may have better knowledge of some of these parameters.

A file with the full results is available at http://tinyurl.com/p9cu8ax

Comments

comments

Author: Stephen Kinsella

Senior Lecturer in Economics at the University of Limerick.

10 thoughts on “On the possible effects of free GP care for under 6’s.”

  1. Looks like a sound analysis. Given that we will find out the true answer very soon, you should consider writing a paper about the extent to which your modelling correctly predicted the true impact of this policy change (and if you get it wrong, where the errors arose, and if the model was correctable with better assumptions). That could be turned into a broader analysis of the value of using datasets like GUI and TILDA to model policy changes.

  2. I am not sure that we will find out the answer soon. Is there data regularly collected on GP visits by age and payment type? Anecdotal accounts are not to be trusted.
    GUI, TILDA and other data sources provide real insight here – that is why they are collected.
    I have no idea what the government’s expectations are with respect to the effect of the policy. Some GPs I have come across seem to be locked into the notion that there will be a huge effect despite there being no evidence of this and plenty of evidence to the contrary. Would they prescribe a drug on this basis?

  3. Really good analysis.
    Think you presume all doctors will take up scheme though. Probably safer to assume 70% of doctors will have 2.23 extra child visits. Also, are there not 2 set visits for check ups being set up for kids at 2 and 5. This will increase the number of visits not based on any perceived illness but just based on a free health check

  4. Why did the HSE not either do some research along these lines before proceeding with the proposal? Or if they were not capable of doing it why didn’t they commission it? It could have made for easier implementation of the under 6s scheme.

    Of course nothing should surprise one about the incompetence of the HSE. I recall that when the original over-70s medical card scheme was introduced the HSE underestimated the cost by some huge amount.

  5. My GP schedules an appointment every 15 minutes, so if that is representative an extra 1.56 visits a week would add an average of 23.5 minutes work per week. It could be, what, 3 times that for a GP in a well-off area who accepts the contract? An extra hour or so of work a week would not be insignificant.

    On the general equilibrium question, I presume the supply response would be significantly lagged, unlike the shock to demand.

  6. Am assuming 100% for simplicity yes. It gets complicated when you have less. Will people switch – people can be quite attached to their GPs? For some people switching may be difficult.
    Incidentally I just learned that Alex White TD said in the Dáil last year:
    “Regarding visitation rates, research carried out for the Department in 2013 indicates that fee-paying children under 6 years of age have an annual GP visit rate of 2.7, whereas, medical card/GP visit card holding children in the same age cohort have an annual visitation rate of 3.1”.
    That’s a smaller average difference that I found but similar magnitude. Its pretty likely that the marginal effect will be smaller than the average difference.
    So this lends some support to my estimates – if anything I am probably overestimating the effect.

  7. @Kevin Denny

    PCRS has good data for patients with medical cards. I’m not sure if the new scheme for under-6s will come under PCRS though.

    Here is a good paper on projections for future GMS costs.

    http://www.ncbi.nlm.nih.gov/pmc/articles/PMC4283081/

    Re Government expectations, one of the things they hope will happen is parents with febrile children will be more likely to visit their GP in the future (as opposed to going straight to the ED).

  8. Using the data from the ‘Growing up in Ireland’ study immediately nullifies this otherwise excellent study. Why Dr Denny failed to examine and use the plentiful data available in the Irish Medical Journal (IMJ) and the British Medical Journal is a not satisfactorily explained.

    Both the methodology and the conclusions of Dr Denny’s work have been robustly invalidated by Dr William Behan. This is unfortunate as it would have been helpful to have further evidence to show that this government’s health policies continue to degrade the quality and availability of health-care delivery to patients.

    In this case, a fixed number of providers (GPs) with a fixed number of hours of availability (contracted) will have to provide consultations to satisfy the extra demands of the under 6 cohort. These demands can only be met at the expense of other patients’ needs. To meet this increased requirement, consultation times will either shorten, there will be reduced availability of appointments for other patients or, most probably, both.

    Current national health policies, often based on predictive models based on flawed statistical evidence, spiced with executive mismanagement and driven by short-term political gain, have resulted in 415,000 patients languishing on hospital waiting lists and anywhere from 400 to 600 people lying forlornly on trolleys in accident and emergency departments on any given day.

    These figures are indisputable and culpability for this inexcusable and irresponsible, gross mismanagement of our health services must lie somewhere.

    Ask any GP what the net effect of providing free care to any patient of any age group has on visiting rates and the reply will indicate that Dr Behan’s study published last year in the IMJ reflects what GPs actually witness first hand. The ESRI data is certainly not reflective of the real world and to use or rely on this data cannot but lead to interpretations that will result in predictable and negative outcomes for Irish patients.

  9. Simple reply to Behan
    They say brevity is the soul of wit. Likewise, there is a lot to be said for keeping blog comments concise and to the point. Recall, the issue at stake here is: what is the effect of a change in price on the demand for a service, a common question in economics. There are numerous econometric analyses of this in general and a significant body that looks at this particular issue: the demand for GP services. Based on this, one can produce a guestimate of the likely impact on the number of GP visits for children under 6 from the new policy. Before replying to specific points, I should state a qualifier: I have no vested interest in this issue. In addressing the points raised I face the significant constraint that Dr Behan appears to have a very limited knowledge of statistics and no economics whatsoever.
    In brief, Dr Behan thinks the numbers from his co-authored paper (Irish Medical Journal 2014) are the basis for extrapolating the effect and that mine are not.
    1) He argues that the GUI data is subject to recall bias. Unfortunately for him – and this is a crucial point- that does not imply that my results are wrong. If we were doing a simple multivariate regression (OLS) then classical measurement error in the dependent variable, of course, does not bias the coefficients. For the count data models that I used, adding a constant (say 3) has no effect on the estimated marginal effect i.e. if everyone understated their number of visits by 3. That’s a rather extreme example. Instead, I generated a random variable taking integers values distributed uniformly on the [0,5] range as a pseudo-measurement error in the dependent variable. This causes the marginal effect to FALL in any count data model- albeit very slightly. Using a [0,10] range doesn’t make much difference. So none of this supports Dr Behan’s arguments. Of course one trivially devise some form of non-classical measurement error which would lead to attenuation bias of any particular coefficient but I think the onus is on him to show that that is (or is likely to be) the case & that it is sufficient to render my results unreliable.
    2) The work that I reported uses standard techniques that have been used in the extensive literature on this subject and on modelling counts generally. That Dr Behan regards this as “very sophisticated mathematical modelling” is not a good omen. The dataset I used is one that has been collected at huge effort and expense by trained survey statisticians and other researchers (including medics) to be reliable and population representative. An international panel of 45 experts provide advice. There is no question that it is more representative of the population than data from a handful of GP’s practices about which we know nothing. Critically, GUI contains essentially all or most of the other variables we would need to estimate the relevant model.
    3) Behan et al’s data is administrative records from 6 GP practices. Although this implied a study size of 1931 these observations are not independent: they are clustered in the 6 GP practices. This is somewhat academic since, remarkably, the paper presents no inferential statistics and no descriptive statistics. We simply have no way of knowing how representative this data apart from the authors’ assertions to that effect nor can we draw any statistical inferences. Since the data is from patients who first attended practice more than a year previously and one in their lifetime, the authors are essentially sampling on the outcome so sample selection bias may be a feature: very healthy children are less likely to be included. How important this is empirically is hard to tell but it biases upwards the numbers. This is why you need to sample individuals. Comparing the mean number of visits between the two groups (with/without medical cards) might be useful as a descriptive exercise. However it tells you nothing about the effect of extending the medical card scheme to the general population for reasons that will certainly be obvious to anyone with a basic knowledge of economics or multivariate data analysis. Note that there is a lot to be said for using administrative data, as is common in the Nordic countries, but it needs to be sufficiently rich.
    4) In short
    a. If you want to know how many GP visits the relevant patients of those six practices had then their study is probably perfect.
    b. If you want to draw an inference about the average number of visits of such children (by medical card status) in the population it may or not be reliable, we have no way of knowing.
    c. Even if the estimate of the mean difference was population representative, if you want to estimate the effect of having a medical card on the number of visits then this study is useless.
    d. The presence of under-reporting does not invalidate the estimates reported by me (a paper is on the way).
    5) The GUI data is available from the Irish Social Science Data Archive. My Stata code is available on request.
    6) At this point, my marginal product is much higher doing other things so I won’t be reading or writing on this thread any further. Thank y’all.

  10. Directly state sponsored GUI/TILDA/CSO 2007/2010 and LIIS 2001 data are all very similar to each other (1 year degradation of recollection).

    Mixture of origins CSO 2001/NUI Galway 2006 and 2010/IMJ 2013/2014, UK QRESEARCH 2008/09 are equally similar because they are audits/based on 2 week recollection.

    There is a 30-40% difference in visiting rates between the two groups of work above that appears to be solely as result of the two different methods of data collection, with GUI being worse because of misattributing general practice nurse activity to public health nurse visiting figures.

    Is Alex Whites figure of 3.1 attendance rate in the under 6 GMS population when the UK figure is 6.8 for an overall less deprived population really credible?

    I wouldn’t have spent so much time trying to figure out why the official figures were totally unreliable if I didn’t have an index of suspicion to start with. What I would like to know is how come nobody else suspected that there is a problem?

Comments are closed.