I just published three documents supporting my big Mental Health Shallow Review, published at the effective altruism forums.  Check it out here.

Thanks to Peter Hurford for funding this research. If you would like to hire me for a research project, please reach out at elizabeth – at – this – domain .

Cost Effectiveness of Mindfulness Based Stress Reduction

The Problem

The WHO estimates that depression and anxiety together account for 75,000,000 DALYs annually, making up ~5% of total DALYs. In “Measuring the Impact of Mental Illness on Quality of Life”, I argue that there is good reason to think that the system used to generate these estimates severely underestimates the impact of mental illness, and thus the true damage may be much higher. To try to get an estimate on the harms of mental health and the benefits of alieviating mental health problems, I did a preliminary cost-effectiveness analysis of Mindfulness Based Stress Reduction (MBSR).

The Intervention

MBSR is an eight week class that uses a combination of mindfulness, body awareness, and yoga to improve quality of life and perhaps physical health for a variety of conditions.

MBSR was created by Jon Kabat-Zinn at the University of Massachusetts in the 1970s, but has spread widely since then. The exact extent of this spread is hard to measure because no official registration is required to teach mindfulness and many classes and books claim to be mindfulness inspired. For the purpose of this evaluation I looked only at things that were officially MBSR or adhered very closely to the description.

Cost of MBSR

Herman, et al. (2017) estimated the marginal cost of an MBSR class participant at $150. The first three hits on google (run in an incognito browser but suspiciously near the location from which I ran the search) for MBSR listed a cost of $395-$595, $275-$425, and $350. The difference between the top of the range and the marginal cost indicates that the high end of that range probably covers all of the costs involved with MBSR (space rental and instructor time for eight weeks of classes plus one eight hour retreat) and then some, so I will use $600 as the ceiling on costs and $150 as the floor.

MBSR has an unusually high time ongoing cost (one hour per day). To model this, I included a range of DALYs as a cost, ranging from 0 (indicating no cost) to 1/24 (as if the participant were dead for that hour). It is unclear how the one hour duration was chosen and I could not find any studies on the comparative impact of different lengths of meditation; it’s quite plausible one could get the same results in less time. For the purpose of this document I used the official program, because it was the most consistently studied.

Cost Effectiveness Analysis of MBSR

Both depression and anxiety are measured with a variety of clinical surveys. To estimate impact, I assumed that the top score on each survey caused a DALY loss equal to severe depression/anxiety, as estimated by the World Health Organization, and that a drop of N percentage points led to a drop of disability weight * N. For example, a drop of 8 points on an 80 point scale of anxiety (disability weight of severe anxiety: 0.523) causes a gain of .0523 DALYs.

For a survey of papers showing potential impact, see this spreadsheet. The estimates range from 2% to 11%, clustered around 7%.

I have created a Guesstimate model to estimate the impact of MBSR. Results were quite promising. On a randomly selected guesstimate run, the average cost was $290/DALY, with a range from $43/DALY to $930/DALY. This is close to but better than Strong Mind’s $650/DALY and overlaps with estimates of antimalarial treatment ($8.15-$150/DALY). Note that the MBSR estimate may understate the impact due to systemic biases in how DALYs are calculated. However it may also overstate the impact, as medical studies tend to overstate intervention impacts for a variety of reasons.

The model makes no attempt to account for co-morbid disorders. Individuals with depression and anxiety would likely see higher benefits, since the same hour of meditation would impact both.

This model makes the rather optimistic assumption the benefits persist for life. This assumes that the participant would have been counterfactually depressed forever without treatment.  In reality the average depressive episode lasts six months, and of people who have suffered at least one episode, the average lifetime number of episodes is four. If we assume the participant gets two years of benefit out of treatment the cost becomes $1200 to $14,000/DALY, with an average of $5200/DALY.


All of the effectiveness studies cited were done on developed world citizens with only mild to moderate mental illnesses. Most were middle aged, and access to MBSR implies a minimum SES bar. It is possible that more severe depression is not amenable to MBSR, or that it is amenable and shows a larger absolute change because there is farther to improve.

I could find no studies on MBSR in the developing world, although since mindfulness meditation was originally created before there was such a thing as the developed world, there is a higher than typical chance that its usefulness will survive cultural translation.

All of the studies referenced had small sample sizes. They all show a consistent effect, but it’s possible publication bias is keeping negative studies out of view.

Official MBSR has an unusually high time cost compared to medication and therapy. The costs are high both upfront (eight weeks of classes and an all day retreat) and ongoing (one hour of meditation/day). Some patients may be able to get the benefits of MBSR with less time; others may not be able to practice at all due to the time demands.


For more on this see my shallow review of mental health .

Measuring the Impact of Mental Illness on Quality of Life


I am currently evaluating multiple interventions aimed at mental illness. In order to compare these to each other and interventions in other areas, it is important to make an estimate of severity of the problem and of the impact of interventions. Several standard systems for evaluating health interventions exist, each of which has strengths and weaknesses. How accurate/useful are these systems for mental illness?

Death Rate

Mental illness has a death toll (primarily from suicide and overdoses) that can be compared to deaths from physical ailments. Death has the advantage of being a binary state subject to very little measurement error or differing definitions across culture. However it is an imperfect proxy for suffering inflicted by mental illness. Depending on culture one country may have a higher depression rate but lower suicide rate. A country with better medical services may have a worse drug problem but fewer deaths from overdoses. Cause of death is subject to manipulation. Mortality is also a very poor measure of anxiety, since anxiety is almost never the immediate cause of death (although it may shorten lifespan).

Disability Adjusted Life Years

Disability adjusted life years (DALYs) are an attempt to use a single number to express the health of a population. The calculation method can vary from study to study; for purposes of this post I will be referring only to the methods used in the Global Burden of Disease 2010 (hereafter GBD 2010) study.

Aggregated DALYs for a population are calculated by multiplying the [disability prevalence] x [disability weight] x [years until remission or death]. Some surveys (but not all) include further discounts for age, assuming that a year lived as a 70 year old is less valuable than a year lived as a 25 year old. This is known as age-weighting. Disability weight is calculated by asking individuals to compare two scenarios and rate which person seems “healthier.” GBD 2010 surveyed approximately 14,000 individuals from five countries (Bangladesh, Indonesia, Peru, the United Republic of Tanzania and the United States of America) and offered a web based survey as well, which was eventually taken by approximately 16,000 people. Previous versions of the GBD exclusively used the evaluations of health care practitioners.

Because they are only are a measure of health, DALYs are not a good measure of suffering. For example, a loved one dying is an obvious cause of suffering via grief, but has no impact on the DALY metric of the survivors. DALYs also deliberately exclude the availability of mitigations: vision impairment has the same DALY cost regardless of the availability of corrective lenses (Voight & King, 2010). These choices make DALYs highly legible and comparable, at the cost of excluding many things one might care about. Additionally, “Healthy” is a highly ambiguous term, which many cultures consider to refer only to physical health. This suggests that if one cares about suffering, or includes mental health in their definition of health, DALYs are likely to severely underrate the impact of mental illness.
Quality Adjusted Life Years

QALYs are explicitly designed to evaluate quality of life, not just health. Instead of choosing which of two individuals is healthier, survey participants may choose which situation they would rather live in (e.g., five years of blindness or four years of deafness), what risk of death they would accept in order to cure an ailment (e.g. 10% risk of death for surgery to restore function to your leg), or “how bad does this sound to you on a scale of 1-100?”

QALYs are noticeably better than DALYs for measuring the impact of mental illness, in that everyone agrees mental illnesses lower quality of life. However there is still concern that they underestimate the impact because people are bad at imagining themselves in different situations, and bad at imagining mental illness in particular. Dolan (2008) argues that any rating based on trade-offs is inherently weak, because humans are so bad at remembering the past and anticipating the future. He favors using ratings of subjective well being from people currently suffering from a condition. Brazier, et al. (2008) cites data that the general public rates mental health issues as less important than physical health, less so than those who suffer from mental illness (Brazier (2008), which if true would lead to an underestimate of the cost of mental illness. Meanwhile De Wit, Busschbach, and De Charro (2000) argue that people underestimate their ability to adapt to situations, and thus all QALY cost estimates are overestimates. Michael Plant argues that this applies only to physical ailments, and that this leads people to underestimate the severity of mental illness relative to physical illness.

Issues Comparing DALYs/QALYs for Mental Illness with Other Illnesses

The cost-effectiveness estimates for malaria nets are based solely on the averted physical suffering. In order to truly compare malaria QALYs with depression QALYs, we must take into consideration the mental health toll of malaria. This turns out to be a very complicated question that can’t be answered without getting into moral ontology, which is beyond the scope of this document.

For a very, very crude idea of the effect on bednets on suffering, see this guesstimate model, which lets you estimate the mental illness cost of malaria from mourning and mental-health related side effects. Ultimately the DALY/$ (guesstimated in the range of 10^-3 and 10 ^-5) are insignificant next to the DALY/$ gain from deaths averted (in the range of 10^-1).

Financial Cost

Illness (mental or physical) can exact an enormous physical toll on sufferers, in both cost of treatment and lost productivity. Productivity loss is more difficult to measure than death and thus not as precise a metric, but it is significantly more objective and comparable across ailments than DALYs or QALYs. For more information on the productivity costs of mental illness, see this post.

A second issue is that using productivity loss as a metric will bias interventions towards people with higher potential incomes, which is the opposite of most people’s instincts.


None of these measurements met my goals of being easy to measure and capturing the entire impact of mental illness. This is not surprising, since even the impacts of physical ailments are hard to measure. The only clear conclusion is that QALYs are better than DALYs for any purpose I can think of. Of the options available, death and financial cost are the most objective, easiest to measure, and easiest to compare to other ailments, but lose a lot of data around suffering. QALYs capture that data, but are still of questionable suitability for comparing to other ailments.

Impact of Depression and its Treatment on Productivity


One argument for prioritizing treatment of mental illness is that the secondary effects (such as higher productivity and improved health-related behavior) may be especially impactful. Illnesses like depression and addiction are incredible drains on productivity, which can be reversed with treatment. In this essay I investigate the productivity cost of untreated (or unsuccessfully treated) mental illness and the impact of treatment on productivity.

How Bad is it?

World Health Organization Data

Alonso, et al. (2011) surveyed workers to determine how many days they missed work due to a variety of chronic illnesses, including depression and anxiety. Their sample included 63,000 people spread across 24 countries, with a range of cultures and income levels. Across all countries, the following disorders caused the average person with that disorder to lose the following days of work. Note that comorbidity is common and days-missed are additive- e.g. a person with depression and generalized anxiety in a lower income country would miss 26.6 days of work.
Days of Productivity Lost to Illness

Lower income countries Medium income countries Higher income countries All countries
Additional days Additional days Additional days Additional days
Mean s.e Mean s.e. Mean s.e. Mean s.e.
Depression 13.1 5 14.7 4.1 4.1 3.2 9 2.5
Bipolar disorder 36.5 15 23.2 9.6 9.6 5.8 17.3 4.9
Panic disorder 24.3 12.9 17.7 5.5 11.7 4.1 14.3 3.5
Specific phobia −6.6 5.2 4.2 4.7 6.7 3.3 3.9 2.5
Social phobia 5.7 10 9 8.4 7.5 2.9 7.3 2.8
GAD 13.5 9.1 24.6 8.4 7.6 4.9 7.7 3.6
Alcohol abuse −2.8 7.2 8.2 5 −0.3 4.5 1.9 3.2
Drug abuse 14.7 13.9 3.9 12.2 1.2 5.5 2.5 4
PTSD 15.3 11.3 −1.1 9.5 16.2 4 15.2 3.5
Insomnia 5.7 5.3 4.6 5.4 9.4 3.2 7.9 2.7
Headache or migraine 10 3.6 6.5 3.3 4.5 2.1 7.1 1.5
Arthritis 6.1 4.4 0.8 5 1.8 2.4 2.7 1.8
Pain 0.9 3.1 11 2.4 19.6 2.1 14.3 1.5
Cardiovascular 2.7 6.7 1 3.6 7.2 2.7 5.7 2.1
Respiratory 10.7 3 −1.1 2.6 0.9 1.4 2.6 1.3
Diabetes 4 6.4 0.5 5.6 9.6 3.8 8.6 2.8
Digestive −4.3 4.8 −0.4 4 16.6 4.8 7.6 3
Neurological 33.7 23 18.6 7 15.3 7.4 17.4 5.8
Cancer 19.4 17.9 −4.2 12.9 6.9 3.6 5.5 3.5


[Note that negative numbers mean the condition is associated with an increase in number of days worked.]

Alonso, et al (2011) did not attempt to measure workers who attended work but were less productive due to illness (presenteeism), or control for average number of days of work for a given country.

Chrisholm, et al. (2016) attempted to estimate the economic impact of depression and anxiety, including the cost of lost productivity, using primarily the data above. They estimate that treatment for depression leads to a 5% increase in attendance (in any country) and 5% increase in productivity while present. This implies a normal worker has 180 working days in high income countries and 260 in low income countries, which is low (see OECD data), meaning the 5% estimate for absenteeism is too high. However I believe their estimate for presenteeism is much too low. Just the diagnostic criteria of depression suggests more than a 5% drop in productivity.


Comparison to Sleep Deprivation

The effects of depression can be similar to sleep deprivation, in part because depression can cause either insomnia or a need for excess sleep, and in part because both produce a “brain fog” (weirdly, sleep deprivation may also treat depression). Given the paucity of information on the relationship between depression and productivity and the abundance of information on the relationship between sleep and productivity, I turned to sleep deprivation as a model for the effects of depression on productivity, contingent on a given a worker making it to their job. The following are mostly small studies but unsurprisingly all show sleep deprivation having a large negative impact on productivity.


Kessler, et al. (2011) estimate that insomnia causes presenteeism equivalent to 7.8 days of missed work per year, an estimated financial loss of $2,280 per person. This used the WHO Health and Work Performance Questionnaire, which relies entirely on workers self-reports of their productivity relative to co-workers. It is also designed only to measure whether someone is more or less productive than average, not the magnitude of the difference.


Gibson & Shrader (2014) estimated that a one hour increase in average nightly sleep led to a 16% increase in wages (on average, $6,000). I will use that as my lower bound for the benefits of treating depression. I assume the actual increase productivity is larger than the increase in wages, because some of the benefit is captured by the employer. If we assume the employer and employee capture equal value, this implies an actual productivity increase of 32%. And if we assume depression is equivalent to 2-3x the cost of missing one hour of sleep, that is almost a halving of productivity (note that for actual sleep, the costs of missed sleep probably increase exponentially). This study is especially promising because it is rather large and used a natural experiment (distance from timezone line) to establish study conditions.


What Does Treatment Accomplish?

Strong Minds

[When not otherwise stated, data comes from Strong Mind’s 2015 report.]

Strong Minds is an NGO in Africa that runs 12 week group therapy classes in Uganda. Their three month month program produces a noticeable drop in depression.

Strong Minds monitors its effect on depression using a modified version of the PHQ-9 (Patient Health Questionnaire- 9). The scale of this test is unknown, making it hard to evaluate the absolute improvement, but lower scores are relatively better (less depression) than higher scores. This questionnaire is an accepted tool for monitoring severity of depression.

Of women participating in Strong Mind’s 12 week pilot program, 92% had reduced scores on the PHQ-9; 11% of the control group had reduced scores. Most of the other effects reported in Strong Mind’s report are given in absolute terms, with no reference to the control group. Based on the reduction in PHQ-9 scores, I will assume 88% of any result is due to participation in the program. Key results:

  • 15 percentage point increase in participation in primary occupation (79% -> 94%).
  • 40 percentage point reduction in families going 24 hours without a meal (53% -> 13%).
  • 17 percentage point reduction in medical care visits (58% -> 41%). This is likely to understate the improvement in health, as some participants probably had physical problems they had previously been too depressed to treat.
  • 18 percentage point increase in families sleeping in protected shelters (65% -> 83%).
  • 10 percentage point increase in school attendance (33% -> 43%).

Income is not reported in this study. The authors do not say this explicitly, but it is common in developing world studies to examine consumption, because income is so variable.

Qualms about data: the study recorded 46 variables, of which less than 10 were reported in their report (not all of which made it into this report). The report included different metrics from phase one studies (eating 3 meals/day, ability to save any amount of income).  Given that it appears this data was still collected in phase two, the absence of results in the report raises concerns about cherry picking. I included this study despite my qualms because so little data was available about the effect of treatment of depression in developing countries.

Cost: $240/12 women in the program = $20/person. This is almost certainly an underestimate of even the marginal cost of the program.

Schoenbaum, et al.

In The Effects of Primary Care Depression Treatment on Patients’ Clinical Status and Employment, researchers reported that six months after their intervention (treatment for depression by a primary care physician, in the USA), 24% (vs 70% in control group) were depressed, and 72% (vs 54%) were employed.


Translating these productivity impacts into dollars is difficult because we can’t assume they hit all incomes equally, however the WHO estimates that in aggregate depression and anxiety together cost one trillion dollars US/year in lost productivity worldwide, slightly more than 1% of total GDP. On an individual level, there is no satisfying answer here. Depression has a very broad definition: the worst cases can destroy all productivity. The typical case destroys somewhere between 5% and 50% of productivity. Treatment of depression can restore that lost productivity in some but not 100% of participants.  


Areas for Further Investigation

I used sleep deprivation to generate heuristics for how damaging depression might be, with the answer being “quite bad”. Those numbers are even more accurate for estimating the effect of sleep deprivation. Because the scope of this paper was limited to economic effects stemming from workplace productivity, I have left out many other costs of sleep deprivation, including health costs and developmental damage to children. Given the costs and prevalence of sleep deprivation, sleep-promoting interventions, especially in children and adolescents, may be a promising area for intervention.