Jason Shafrin’s journal round-up for 3rd February 2020

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.

Capacity constraints and time allocation in public health clinics. Health Economics [PubMed] Published 14th January 2020

Capacity constraints are a key issue in many health care markets. Capacity constraints can be due to short-run fluctuations in available labor or long-term resource scarcity where available supply does not meet demand at prevailing prices. One key issue to understand is how providers respond when capacity constraints appear. There are a number of potential options for health care providers to address capacity constraints. The first would be to decrease the number of patients seen. A second option would be to decrease quality. A third option would be to access additional funding to increase capacity. A fourth option would be to demand workers to work longer hours at no additional pay, which would be an implicit hourly wage reduction (and may not be legal).

In an attempt to find out the answer to this question, Harris, Liu and McCarthy examined data from a clinic in Tennessee. One issue with capacity constraints is that workforce allocation may be endogenous. Harris and co-authors look at a case where two nurses were removed from the clinic on selected mornings to administer flu shots at local schools. They claim that the assignment of nurses to schools was done by the local health department and those decisions were plausibly exogenous to any clinic decisions on volume or quality of care. Further, these nurses were not replaced by staff from other clinics.

The authors use data covering 16 months (i.e. two flu seasons) of visits. They first conducted a nurse-level analysis to evaluate if nurses on flu days had lower productivity. ‘FluNurse’ and ‘FluDay’ indicator variables were used. The former was used to control for whether the nurses selected to administer the flu shots at the schools were systematically more or less productive than the other nurses; the latter was used to measure the impact of the actual day when nurses were assigned to the school. The authors also conduct a visit-level analysis to see how time spent in the clinic varies on the days when nurse capacity is reduced compared to when it is not.

The authors find that the extensive margin is most important. On days when nurses visited schools, capacity is reduced by about 17%. On these days, providers do see fewer patients overall and prioritize scheduled visits over walk-ins. On the intensive margin, providers do decrease time spent with patients (by about 7%), but this time reduction is largely achieved by reducing some of the administrative aspects of the visit (i.e. expedited check-out times). The authors conclude that “providers value spending sufficient time with patients over seeing as many patients as possible.”

It is unclear whether these results would translate to other settings/countries or cases where capacity constraints are more long-term. The study does not discuss in detail how the clinic is reimbursed. A fee-for-service reimbursement may incentivize prioritizing volume over quality/time spent whereas under capitated or salaried reimbursement they may prioritize quality. Since many of the services examined by the study were provided by nurses, and nurses are more likely to be paid a salary rather than compensated based on clinic volume, it is unclear whether these results would translate to capacity constraints involving physicians.

Outcome measures for oncology alternative payment models: practical considerations and recommendations. American Journal of Managed Care [PubMed] Published 11th December 2019

Value-based payment sounds great in theory. Step 1: measure health outcomes and cost. Step 2: risk adjust to control for variability in patient health status. Step 3: pay providers more who have better outcomes and lower cost; pay providers less who have worse outcomes and higher cost. Simple, right? The answer is ‘yes’ according to many payers. A number of alternative payment models are being adapted by payers in the U.S. In oncology, the Centers for Medicare and Medicaid Services (CMS) has implemented the Oncology Care Model (OCM) to reimburse providers based on quality and cost.

This approach, however, is only valid if payers and policymakers are able to adequately measure quality. How is quality currently captured for oncology patients? A paper by Hlávka, Lin, and Neumann provides an overview of existing quality measures. Specifically they review quality metrics from the following entities: (i) OCM, (ii) the Quality Oncology Practice Initiative by the American Society of Clinical Oncologists (ASCO), (iii) the Prospective Payment System–Exempt Cancer Hospital Quality Reporting Program by CMS, (iv) the Core Quality Measures Collaborative Core Sets by CMS and America’s Health Insurance Plans (AHIP), (v) the Oncology Medical Home program by the Community Oncology Alliance, (vi) the Osteoporosis Quality Improvement Registry by the National Osteoporosis Foundation and National Bone Health Alliance, and (vii) the Oncology Qualified Clinical Data Registry by the Oncology Nursing Society. So, how well do these entities measure health outcomes?

Well, in general, most are measuring quality of care processes rather than health outcomes. Of the 142 quality measures examined, only 28 (19.7%) were outcome measures; the rest were process measures. The outcome measures of interest fell into five categories: (1) hospital admissions or emergency department visits, (2) hospice care, (3) mortality, (4) patient reported outcomes (PROs), and (5) adverse events (AEs). The paper describes in more detail how/why these metrics are used (e.g. many hospital admissions are related to chemotherapy AEs, hospice care is underutilized).

While this paper is not a methodological advance, knowing what quality measures are available is extremely important. Further, the paper cites a number of limitations of these quality metrics. First, you need reliable data to measure these measures, even if patients move across health care systems. Additionally, administrative data (e.g. claims data) often appear with a lag. Second, risk adjustment is imperfect and applying a value-based payment may incentivize providers to select more low-risk patients and avoid the sickest cancer patients. Third, if we care about patients’ opinions—which we should!—PROs are important. Collecting PROs, however, is more expensive than using administrative data. Fourth, quality measurement in general takes time and effort. My own commentary in the Journal of Clinical Pathways argues that these limitations need to be taken seriously and these costs and accuracy issues may undermine the value of value-based payment in oncology if quality is measured poorly and is costly to collect.

Modeling the impact of patient treatment preference on health outcomes in relapsing-remitting multiple sclerosis. Journal of Medical Economics [PubMed] Published January 2020

We hear a lot about patient-centered care. Intuitively, it makes a lot of sense. Patients are the end users, the customers. So we should do our best to give them the treatments they want. At the same time, patients have imperfect information and rely on physicians to help guide their decisions. However, patients may value things that—as a society—we may not be willing to pay for. For instance, empirical research shows that patients place a high value on hospital amenities. Thus, one key question to answer is whether following patient preferences is likely to result in better health outcomes.

A paper by van Eijndhoven et al. (disclosure: I am an author on this paper) aims to answer this question for patients with relapsing-remitting multiple sclerosis. The first step of the paper is to measure patient preferences across disease modifying drugs (DMDs). This was parameterized based on a discrete choice experiment of patients with multiple sclerosis. Second, a Markov model was used to estimate the impact of each individual DMD on health outcomes (i.e., number of relapses, disability progression over time, QALYs). QALYs were estimated utilities by health states defined by the Expanded Disability Status Scale (EDSS), caregiver disutility in each state, disutility per relapse, and disutility due to adverse events from the literature. Third, the authors compared two states of the world: one based on patient-driven preferences (as estimated in Step 1) and another based on current prescribing practices in the United Kingdom.

Using this approach, patient-centered prescribing practices lead to 6.8% fewer relapses and a 4.6% increase in QALYs. Applied to the UK population level, this would result in almost 37,000 avoided relapses and over 44,000 discounted QALYs over a 50 year period. Additionally, disease progression was slowed. For the typical patients, EDSS was -0.16 less each year under the patient-centered prescribing compared to the current market shares.

There are a few reasons for this finding. First, under current treatment patterns, many patients are not treated. About 21% of patients with RRMS in the UK do not receive DMDs. Access to neurologist care is often difficult. Second, patients do have a strong preference for more effective, newer treatments. Previous research indicates that prescribers in England generally viewed NICE guidelines as mandatory criteria they were obligated to follow, whereas neurologists in Scotland and Wales were more varied.

This study did have a number of limitations. First, the study did not examine costs. Patients may prefer more effective treatments, but this may have cost implications for the UK health system. Second, the impact of treatments was based on clinical trial data. DMDs may be more or less effective in the real world, particularly if adherence is suboptimal. Third, once people reach high levels of disability (EDSS ≥7), the study assumed that they were treated with best supportive care. Thus, the estimates may be conservative if treatment is more aggressive. Fourth, the treatment options were based on currently approved DMDs. In the real-world, however, new treatments may become available and the actual health trajectories are likely to deviate from our model.

While this study does not answer what is the optimal treatment mix for a country, we do see evidence that patient-preferred treatments are highly related to the health benefits received. Thus—in the case of multiple sclerosis—physicians should not fear that shared decision-making with patients will result in worse outcomes. On the country, better health outcomes could be expected from shared decision-making.


Chris Sampson’s journal round-up for 19th August 2019

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.

Paying for kidneys? A randomized survey and choice experiment. American Economic Review [RePEc] Published August 2019

This paper starts with a quote from Alvin Roth about ‘repugnant transactions’, of which markets for organs provide a prime example. This idea of ‘repugnant transactions’ has been hijacked by some pop economists to represent the stupid opinions of non-economists. If you ask me, markets for organs aren’t repugnant, they just seem like a very bad idea in terms of both efficiency and equity. But it doesn’t matter what I think; it matters what the people of the United States think.

The authors of this study conducted an online survey with a representative sample of 2,666 Americans. Each respondent was randomised to evaluate one of eight systems compared with the current system. The eight systems differed with respect to i) cash or non-cash compensation of ii) different sizes ($30,000 or $100,000), iii) paid by either a public agency or the organ recipient. Participants made five binary choices that differed according to the gain – in transplants generated – associated with the new system. Half of the participants were also asked to express moral judgements.

Both the system features (e.g. who pays) and the outcomes of the new system influenced people’s choices. Broadly speaking, the results suggest that people aren’t opposed to donors being paid, but are opposed to patients paying. (Remember, we’re talking about the US here!). Around 21% of respondents opposed payment no matter what, 46% were in favour no matter what, and 18% were sensitive to the gain in the number of transplants. A 10% point increase in transplants resulted in a 2.6% point increase in support. Unsurprisingly, individuals’ moral judgements were predictive of the attitudes they expressed, particularly with respect to fairness. The authors describe their results as exhibiting ‘strong polarisation’, which is surely inevitable for questions that involve moral judgement.

Being in AER, this is a long meandering paper with extensive analyses and thoroughly reported results. There’s lots of information and findings that I can’t share here. It’s a valuable study with plenty of food for thought, but I can’t help but think that it is, methodologically, a bit weak. If we want to understand the different views in society, surely some Q methodology would be more useful than a basic online survey. And if we want to elicit stated preferences, surely a discrete choice experiment with a well-thought-out efficient design would give us more meaningful results.

Estimating local need for mental healthcare to inform fair resource allocation in the NHS in England: cross-sectional analysis of national administrative data linked at person level. The British Journal of Psychiatry [PubMed] Published 8th August 2019

The need to fairly (and efficiently) allocate NHS resources across the country played an important part in the birth of health economics in the UK, and resulted in resource allocation formulas. Since 1996 there has been a separate formula for mental health services, which is periodically updated. This study describes the work undertaken for the latest update.

The model is based on predicting service use and total mental health care costs observed in 2015 from predictors in the years 2013-2014, to inform allocations in 2019-2024. Various individual-level data sources available to the NHS were used for 43.7 million people registered with a GP practice and over the age of 20. The cost per patient who used mental health services ranged from £94 to over one million, averaging around £2,000. The predictor variables included individual indicators such as age, sex, ethnicity, physical diagnoses, and household type (e.g. number of adults and kids). The model also used variables observed at the local or GP practice level, such as the proportion of people receiving out-of-work benefits and the distance from the mental health trust. All of this got plugged into a good old OLS regression. From individual-level predictions, the researchers created aggregated indices of need for each clinical commission group (CCG).

A lot went into the model, which explained 99% of the variation in costs between CCGs. A key way in which this model differs from previous versions is that it relies on individual-level indicators rather than those observed at the level of GP practice or CCG. There was a lot of variation in the CCG need indices, ranging from 0.65 for Surrey Heath to 1.62 for Southwark, where 1.00 is the average. You’ll need to check the online appendices for your own CCG’s level of need (Lewisham: 1.52). As one might expect, the researchers observed a strong correlation between a CCG’s need index and the CCG’s area’s level of deprivation. Compared with previous models, this new model indicates a greater allocation of resources to more deprived and older populations.

Measuring, valuing and including forgone childhood education and leisure time costs in economic evaluation: methods, challenges and the way forward. Social Science & Medicine [PubMed] Published 7th August 2019

I’m a ‘societal perspective’ sceptic, not because I don’t care about non-health outcomes (though I do care less) but because I think it’s impossible to capture everything that is of value to society, and that capturing just a few things will introduce a lot of bias and noise. I would also deny that time has any intrinsic value. But I do think we need to do a better job of evaluating interventions for children. So I expected this paper to provide me with a good mix of satisfaction and exasperation.

Health care often involves a loss of leisure or work time, which can constitute an opportunity cost and is regularly included in economic evaluations – usually proxied by wages – for adults. The authors outline the rationale for considering ‘time-related’ opportunity costs in economic evaluations and describe the nature of lost time for children. For adults, the distinction is generally between paid or unpaid work and leisure time. Arguably, this distinction is not applicable to children. Two literature reviews are described. One looked at economic evaluations in the context of children’s health, to see how researchers have valued lost time. The other sought to identify ideas about the value of lost time for children from a broader literature.

The authors do a nice job of outlining how difficult it is to capture non-health-related costs and outcomes in the context of childhood. There is a handful of economic evaluations that have tried to measure and value children’s foregone time. The valuations generally focussed on the costs of childcare rather than the costs to the child, though one looked at the rate of return to education. There wasn’t a lot to go off in the non-health literature, which mostly relates to adults. From what there is, the recommendation is to capture absence from formal education and foregone leisure time. Of course, consideration needs to be given to the importance of lost time and thus the value of capturing it in research. We also need to think about the risk of double counting. When it comes to measurement, we can probably use similar methods as we would for adults, such as diaries. But we need very different approaches to valuation. On this, the authors found very little in the way of good examples to follow. More research needed.


Considering time perception

In economic evaluations in health, time matters. More time in a bad health state is necessarily worse (or no better) than a smaller amount of time in said health state. Likewise, the value of an intervention increases with the duration of benefit. The standard QALY framework takes this into account. Furthermore, time preferences matter. Economists deal with this in a present value framework; discounting future costs and benefits. But there is another aspect of time which is not taken into account; time perception. Issues surrounding our perceptions of time recently appeared in the usual pop science outlets following the release of Claudia Hammond’s new book ‘Time Warped‘. As medical, neurological and psychological understandings of time perception improve, is it time economists weighed in?

Economists have considered the effects upon time perception of things like ‘awe‘ and the cognitive resource demands of tasks, while others have investigated the interaction between time preferences and time perceptions. It seems none have investigated the implications for an area in which time perceptions might play their most important role; health.

Why might it matter?

Time is an abstract idea, but economists rarely treat it as such. This means that people’s perceptions of time are overlooked. This may be a reasonable approach if we are working at the mean and people’s perceptions of time are consistent across health states. But what if they aren’t? Consider a basic illustrative QALY example:

  • Scenario A: 2 weeks in 0.8 health, followed by 2 weeks in full health
  • Scenario B: 1 week in 0.6 health, followed by 3 weeks in full health

Clearly these scenarios give the same QALY result. If one perceives time to pass more slowly in a worse health state then we might be able to add to the scenarios that in ‘A’ the first 2 weeks “feels like 3 weeks”, while in ‘B’ the first week “feels like 2 weeks”. If we were to weight the QALY values in these periods according to perceived time we would have a preference for scenario A. With these re-weightings, 2 weeks of 0.8 in scenario A would become equivalent to 2 weeks of 0.53, while 1 week of 0.6 in scenario B would become equivalent to 1 week of 0.3; increasing the difference between the two health states.


We all experience fluctuations in our own time perception. Time flies when we’re having fun. It might fly when we’re healthy too. It may be the case that a general and testable model of time perception exists in health. The most likely relationship seems to be that being in poorer health would be associated with a perception of time moving more slowly. There may also be differences in time perception between different groups of people (men and women, children and the elderly) that we may or may not want to adjust for. Some interesting implications of time perceptions for the burden of waiting times have already been identified.

Particular health conditions can affect an individual’s perception of time, and this is where the consideration of time perceptions could be crucial. Children with ADHD, for example, have been shown to perceive time in very different ways to those without ADHD. Likewise, cancer patients with evidence of disease have been found to perceive time as progressing more slowly, compared with those without. If a condition causes individuals to perceive time to move more slowly in a consistent and measurable way, then specific rules could be established to assign greater weight to treatments for said condition.


The relationships above are all testable and quantifiable. Time perceptions can be easily tested by denying well and unwell patients access to a clock and calendar, and then asking them about the ways in which they perceived their time in these states. It is important to note that during these times our perception of time may be affected by other things; whether we are confined to a hospital bed, stuck in the office or asleep at home. All of these things could be controlled for in an experiment.

The real question is whether or not we (the public) want to take this into account or not. If the public’s stated preferences do not reflect the effects of time perception then should we artificially weight their preferences to do so? The issue seems analogous to that of adaptation (to which time perceptions would no doubt be subject!). Public valuations of the effects of time perceptions could be captured with the usual time trade-off, standard gamble or discrete choice experiment techniques. One difficulty would be in ensuring that there is no double-counting. It seems likely that this would not apply in the specific applications, where the general public’s valuations would not consider ADHD’s effects on time perception, for example. However, it seems possible (likely, even) that the general public would consider time perception effects associated with being generally unwell.

Perceptions of time have been discussed in economic and health economic literature in respect to experienced utility and biases of memory in retrospective valuations. There has been little contribution to theories about individuals’ present time perception. I believe it may be time for these ideas to be explored further, particularly in relation to specific health conditions, and applied to preferences and expected utility in health.

How do you see time perceptions influencing quality of life? To what extent do you think differences in time perception either could or should influence health care decision making? Please comment below.