In a previous news blog I explicated a theory for improved access to services in countries of moderate income. I argued that services would always be limited in very low-income countries. On the other hand, in high-income countries the tax base is sufficient to support comprehensive public services, including health services. However, I explained that in countries of intermediate wealth, the tax base was often immature and small, relative to the total economy. In such a scenario it is necessary to rely on community contributions in order to improve access to services. I refer to this as the IKEA model.
I found a very nice example of the IKEA model recently. This is a pre-payment service implemented by Safaricom and other companies in Kenya called M-TIBA. These organisations provide an electronic wallet that families can contribute to, which provides ready cash at the point when services are required. Such a service can ensure that money is available for out-of-pocket payments. Electronic pre-payment thus acts as a kind of community insurance or risk-sharing system.
I see great potential in these services, but feel that they should be evaluated. Like microfinance, people may make exaggerated claims of how beneficial they can be. I expect that these services will be most useful when they are all linked to other services. Thus, I imagine that a pre-payment system will be much more effective in securing transport to hospital when it has grown up in parallel with an inexpensive (perhaps partially subsidised) local ambulance service. As part of my work as lead for the Access work package in the NIHR Global Health Research Unit on Global Surgery, I shall be investigating this type of system in improving access to acute surgical and medical care.
The CLAHRC WM Director is something of a pro-market libertarian. From this ideological standpoint he is sympathetic towards the idea of payment by results to supplement for lack of market incentives in the supply of non-market services. Just one problem – they don’t work. Well, they often don’t work. A recent, elegant paper uses threshold analysis to examine the effect of the Medicaid / Medicare Services payment by results – their ‘Value Based Modifier’. This is the largest health service in the world – much bigger than the NHS – so there are plenty of data points. The threshold used in studies was the size of the practice, at which various rewards and penalties cut in. So, in outline, the graphs look like this:
If the incentive worked then we would expect to see this:
Instead we see this:
In other words, the incentives did not work. Assuming that the threshold does not coincide exactly with a counterfactual break, this disproves any benefit (or disbenefit) of performance-based payment in this or similar contexts. But CLAHRC WM did find that an incentive was effective in increasing uptake of home haemodialysis. So here is an hypothesis: incentive systems only work when there is a clear objective that can be achieved by specific management / clinical action. We have said this before – Lilford’s rule if you insist – “incentives do not work when the people at whom the incentive is targeted do not have a clear and correct idea of how the objective may be achieved.” I also think this is a lovely example of threshold analysis, a topic mentioned in a recent News Blog.
Reimbursement levels for medical care in large US hospitals are reduced by up to 2% if compliance with evidence-based clinical care standards falls below threshold levels. Does this result in improved care compared to control hospitals not exposed to the financial incentive? To find out, intervention hospitals were compared to control hospitals. The ‘value based purchasing’ schemes were not introduced in a prospective experiment, and the controls (small rural hospitals) are very different in nature to those larger hospitals to whom the incentive applies. To mitigate potential bias, difference-in-difference approaches were used; hospitals were matched for previous performance; and the usual statistical adjustments were made. Adherence to appropriate clinical processes was increasing among both control and intervention hospitals before the intervention was implemented. Rates of adherence did not differ between intervention and control hospitals post-intervention. The clinical indicators related to three tracer conditions frequently used in studies of adherence to clinical standards – pneumonia, heart attack or heart failure. Patient experience measures also did not differ over intervention and controls, and while mortality was improved for pneumonia, it did not do so for the other conditions. The effect on pneumonia deaths was regarded as a chance finding (alpha error), given the null result on mediating variables (i.e. clinical process variables). Arguably these results were null because the incentive was low (only 2% of total reimbursement) and distributed over a large number of outcomes. Alternatively, doctors are largely intrinsically motivated and do not need financial incentives to moderate their performance. We will pick up on this issue in our next News Blog.
An interesting paper from the Berlin University of Technology compares the quality enhancement systems across the above countries with respect to measuring, reporting and rewarding quality. This paper is an excellent resource for policy and health service researchers. The US has the most developed system of quality-related payments (P4P) of the five countries. England wisely uses only process measures to reward performance, while the US and Germany include patient outcomes. The latter are unfair because of signal to noise issues, and the risk-adjustment fallacy. Above all, remember Lilford’s axiom – never base rewards or sanctions on a measurement over which service providers do not feel they have control. It is true, as the paper argues, that rates of adherence to a single process seldom correlate with outcome. But this is a signal to noise problem. ‘Proving’ that processes are valid takes huge RCTs, even when the process is applied to 0% (control arm) vs. approaching 100% (intervention arm) of patients. So how could an improvement from say 40% to 60% in adherence to clinical process show up in routinely collected data? I have to keep on saying it – collect outcome data, but in rewarding or penalising institutions on the basis of comparative performance – process, process, process.
Across low- and middle-income countries (LMICs), over 50% of total health care spending is derived from out-of-pocket expenses. Some of these are formal recognised tariffs in public health systems. However, a proportion are irregular or informal payments (bribes/kick-backs). It is hypothesised that these informal payments are used to subsidise the poor at the expense of the rich after the fashion of Robin Hood in English folklore. Enter results from a series of publically available repeated surveys called ‘Afrobarometer‘. Here public attitudes and experiences relating to democracy and governance are surveyed in 18 African counties. Nationally representative samples of over 25,000 individuals are selected randomly across participating countries. Afrobarometer provides the data for an important study  of the extent to which informal payments were elicited across people of different income levels (according to the Lived Poverty Index). Far from confirming the Robin Hood hypothesis, the authors find a higher occurrence of bribe paying among the poorest people across the countries studied – elasticity is negative in that the richer the person, the lower the probability that they will have paid a bribe on attending a health care facility. These results are similar to those obtained in a previous study in Hungary. There is some evidence that the problem is worse in cities where service providers are less likely to have known or have community affiliations with patients. This finding reminds me of the Bible scripture – “For whosoever hath, to him shall be given… but whosoever hath not, from him shall be taken away...” (Matthew 13:12).
In a previous News Blog we reported results showing that maintenance of communal lavatories was worse among people who had had a subsidy for lavatory maintenance withdrawn than among those who had never had the subsidy in the first place. The ‘free good’ idea was at work here, whereby people can develop a sense of entitlement. A recent study involved providing free shoes for poor people. Those who received the free shoes were more likely to feel that other people should provide family needs in general than those who had not been given free shoes – classic aid dependency. Handing out free shoes did not increase overall ownership of shoes, foot health or self-esteem; presumably because of ‘fungability’ – people used their shoe money for other purposes. Yet not all ‘free goods’ are bad – a recent paper co-authored by CLAHRC WM researchers showed that even very small user fees reduce access to services in Malawi and this hits the most vulnerable – children – the hardest. Taken in the round this leads to the CLAHRC WM Director’s axiom – “ration health care, but not access to health care.”
Introducing fines for readmission rates crossing a certain threshold has been associated with reduced readmissions. Distilling a rather wordy commentary by Friebel and Steventon, there are problems with the policy since it might not lead to optimal care:
The link between quality of care and readmission is not good according to most studies, so that there is a risk that patients who need readmission will not get it.
In support of the above, less than a third of readmissions are for the condition that caused the previous admission (which is not to say that none are preventable, but it suggests that a high proportion might not be).
Risk-adjustment is at best imperfect.
And this probably explains why ‘safety net’ hospitals caring for the poorest clientele come off worst under the pay-for-performance system.
I refer it my iron law of incentives – ‘only use them when providers truly believe that the target of the incentive lies within their control.’
Evidence is accumulating that pay for performance does not perform. The problems are both endogenous (what happens to the particular performance targets at which the intervention is targetted) and exogenous (what happens to performance criteria beyond the designated targets).
First, the endogenous effects. Pay for performance does not seem to work (not even on its own terms) or its effects are short lived. Studies on both sides of the Atlantic have found that initially positive effects were short lived.
Second, a recent article highlights the negative exogenous effects of payment for performance. The interventions particularly penalise hospitals at the bottom of the financial difficulty league, who tend to serve disadvantaged populations. They are unfair, since patients at these hospitals have greater morbidity (with more opportunities for error). Their patients are not as mobile as better-off people, and so they cannot as easily ‘vote with their feet’. Worse, there is a substantial volume of psychological work showing that financial rewards/penalties are demotivating, especially to intrinsically motivated people. Specific targets sometimes work, but burdensome lists of tick-box targets are often not effective in the long term, and can have negative spill-over effects.
The NHS is constituted to provide care that is free at the point of use. However, even in the NHS, patients sometimes have to contribute (make co-payments) – for example, a prescription charge is levied on patients who do not qualify for exemption. What about the reverse – paying patients to adopt healthy behaviours, such as adhering to recommended treatment? Pregnant women in some parts of France have been incentivised to attend antenatal clinics, for example, while Theresa Marteau’s team has found that financial incentives were superior to other methods in increasing cigarette quit rates. There are many examples of incentive payments in terms of cash or an opportunity to participate in a lottery in low-income countries.
CLAHRC WM is very interested in the effect of individual incentives and co-payments on uptake of services. CLAHRC WM collaborator Ivo Vlaev is co-investigator for a trial on financial incentives for diabetic retinal screening with three arms – control; money payment; and participation in a lottery. We have identified two recent systematic reviews dealing with this topic – one on the effect of co-payments on utilisation of services in high-income countries; and the other on incentive payments in low-income countries. The former study finds that even small co-payments suppress demand. The latter study appears to be a mirror image with the reciprocal finding that small incentive payments stimulate uptake. More data are needed; so far our evidence base is financial flows from patients in high-income settings, and financial flows to patients in low-income countries. However, the evidence suggests that money flow in either direction is associated with high elasticity of demand. This concept of reciprocal responses as money is made available or withdrawn is represented in the figure. The origin is the point where the service is free at the point of use and there is no incentive payment. This origin is represented at around 50% uptake of service, but could lie anywhere between 0% and 100%, depending on the service concerned.
We would be pleased to hear from other scholars who wish to collaborate with us on populating the above graph. This would answer many questions, for example, is the graph symmetric, or is the graph steeper for incentive payments than for co-payments?
In a previous blog I promised to disclose the results of the Oregon experiment on the effects of providing health insurance to previously uninsured people. To recap, in the USA all young and middle-aged adults below the poverty line are provided with national insurance called Medicaid. The state of Oregon wished to extend such insurance to a category of slightly better-off people, yet did not have enough money to provide insurance for all in this category. Eligibility was therefore determined by lottery, on the grounds that this would be a fair way to distribute a scarce resource – a reverse of the draft if you like. Of course, such distributional exercises constitute an unintended RCT if one can find out who was randomised to which condition and then follow them up. Baicker et al. did just that. The results are politically sensitive (as you will see if you search the internet), but wisely the authors published the protocol before analysing the data.
The results show that the group offered insurance sought more services, engaged in more preventive activities, had lower expenses, and were more likely to avoid catastrophic payments than those not offered cover. There was no difference between groups in proportions with high blood pressure or elevated glycosylated haemoglobin and death rates did not differ (under 1% in both groups). Patients offered insurance were more likely than those not selected in the lottery to report improved health and the mental component of the quality of life score was improved.
Only two year follow-up data are available. A health economic analysis was not attempted. Despite the study size (over 12,000 people), the power to detect changes in rates of diseases, such as diabetes and hypertension, was low. What is not in doubt is that insurance relieves financial stress and the anxiety that goes with it. This study is relevant for two reasons. First, the results are of policy relevance world-wide. Second, it is a fine example of a high quality academic output from an entirely service-led intervention. Indeed, it conforms with the CLAHRC model, where the service dog wags the research tail.