The GiveWell Blog

Disability-Adjusted Life Years II: Variations

(updated on: ) | by Holden

Previously, I outlined the basics of the Disability-Adjusted Life Year (DALY) metric. It takes the approach of converting all health burdens into equivalent “years of healthy life lost”: a year of blindness is counted as .6 lost years, a year of severe malnutrition is counted as .053 lost years, etc.

This post discusses two common “variations” on DALYs, meant to deal with relatively thorny disagreements about how different years of life should be valued. As before, page numbers refer to the Global Burden of Disease 2000 report.

Age-weighting

One variation has to do with the intuition some people have that a 20-year-old’s death is more tragic than an infant’s. (I expressed this intuition myself back in November, and I still hold this view.) In an attempt to square with this intuition (which is common and well-documented, as Pg 400 shows), the DALY metric includes an optional age weighting feature that lowers the value of a healthy year of life lived at very young and very old ages, relative to the value of a healthy year of life around age 20. DALYs can be computed with or without age-weighting (“without” just means that all years of healthy life are valued the same).

Discounting

The other variation has to do with valuing present vs. future benefits of aid. DALY calculations apply a discount rate to future benefits; for example, when using a discount rate of 3%, one would count a year of healthy life saved ten years from now as being worth only 74% as much as a year of healthy life saved this year (74% = 1/1.03^10).

I confess that I don’t fully follow the justification for discounting given in the Global Burden of Disease Report, which claims that “the strongest argument for discounting is … [that] not discounting future health would lead to the conclusion that all of society’s health resources should be invested in research programs or programs for disease eradication” (400), which apparently is considered obviously wrong by the authors. Personally, the most appealing argument I can think of for discounting is that helping a person can help them help others, so helping a person sooner is literally “worth more” than helping a person later.

Notation

DALYs(0,0) refers to DALYs calculated with a 0% discount rate and no age-weighting. DALYs(3,1) refers to DALYs calculated with a 3% discount rate and age-weighting. (The first number in parentheses is the discount rate; the second is a 1 if age-weighting is being used, and a 0 if not.) See Pg 401 for the specifics of how varying these numbers affects the valuation of different years.

In theory, you can calculate DALYs using whatever parameters best fit your own philosophical values. In practice, the reports we’ve seen using this metric (Global Burden of Disease Report, Copenhagen Consensus, Disease Control Priorities Project) will give you, at most, DALYs(0,0), DALYs(3,0) and DALYs(3,1), and will rarely give you the inputs into these numbers so you can calculate your own versions. That means that if you want to use a 6% discount rate, you’re completely out of luck; there’s no way to convert DALYs(3,0) to DALYs(6,0) without having more information. More importantly, it means that:

  • You can’t use your own version of age-weighting. Even the age-weighted version of DALYs still rates an infant death as about equally tragic to a 20-year-old death (it values a year more for a 20-year-old, but when you work it all out the value of a life comes out the same). There is evidence (see pg 401) that people find a 20-year-old’s death to be far worse; if you share that intuition, then DALYs as they are usually presented won’t reflect your values, and there will be no way to convert them into a unit that does.
  • You can’t use your own disability weights. Personally, this is the area I’d most like to see some variation in – the official disability weights disagree violently with my personal intuitions about, for example, how bad it is to be severely malnourished (current weights put it at only 5.3% as bad as a year of life lost – see Pg 121) or how bad it is to go through an abortion (it appears that this is counted as “no cost” by DALYs – see Pg 121 again).

The DALY metric does have some flexibility to accommodate different personal values, but in practice it ends up being pretty rigid. More on this in a future post.

Next in series:

Previously in series:

I like it. How do I fund it?

| by Holden

The Community-Led Total Sanitation program looks like a potentially good target of funding.

But I can’t find out how to fund it.

The program’s summary page links to three organizations. One appears devoted to research rather than replication. Another is a water and sanitation omnibus program whose activities include many of the activities I’m less confident in. CLTS is nowhere on its list of global initiatives. The third organization is CARE, a giant organization whose website barely mentions CLTS (and its only use of the program appears to be as an “entry point” to other programs).

We think programs that are proven, cost-effective and scalable should be popular. But in many cases, there isn’t even a mechanism for this to happen – some of the most promising interventions aren’t even on a donor’s menu.

Disability-Adjusted Life Years: Introduction

(updated on: ) | by Holden

We’ve had many discussions in the comments about the metric known as Disability-Adjusted Life-Years (DALYs). The DALY essentially converts the burdens imposed by all health issues – from premature death to blindness to injuries – into a single, consistent unit. It is the metric of choice for the Disease Control Priorities Project as well as a centerpiece of the Copenhagen Consensus analysis, and is used widely by the World Health Organization – yet it isn’t, and likely won’t be, the central metric in our analysis.

At this point I want to start a more thorough discussion of why this is. I’m going to start at the beginning, with a full description of what DALYs are (and the different ways of calculating them). Some readers will already be familiar with what’s below, but we want to make sure we clearly describe the metric and give examples of its implications before discussing its strengths and weaknesses.

The most complete account of DALYs I know of is in the Global Burden of Disease report. Page numbers below refer to this report.

The basics: burden of health problems in terms of years of life

A DALY is a measure of the “burden” of a health problem; two common uses of this measure are (a) ranking diseases and risk factors (from most to least burdensome), as the Global Burden of Disease report does, and (b) ranking different interventions (in terms of how much they can be expected to reduce burdens, “per dollar”), as projects including DCPP do. The basic DALY formula is on page 48:

DALY = YLL (Years of Life Lost) + YLD (Years of Life lost due to Disability)

YLL is the more straightforward component. Putting aside discounting/weighting issues (to be discussed later), the death of a male infant (life expectancy 80 years) would be counted as 80 years of life lost, while the death of a 45-year-old female (life expectancy 83.72 years) would be counted as 38.72 years of life lost (see page 402 for the life expectancy figures). Without further adjustments, this implies that the death of a single infant is considered about as bad in and of itself as the death of two adults.

Quantifying morbidity

YLD represents an attempt to convert years of life affected by a disability into the same terms as years of life lost due to premature death. For example:

  • A year spent with blindness (as opposed to a year spent with “normal health”) is counted as 60% as “bad” (i.e., as much burden) as a year of life lost due to premature death. So the metric would count a condition that permanently blinds five 30-year-olds as about equally “burdensome” to a condition that results in the death of three 30-year-olds.
  • A year spent with protein-energy malnutrition to the point of wasting (i.e., being severely underweight) is counted as 5.3% as “bad” as a year of life lost due to premature death. This implies that if a child is malnourished to the point of being severely underweight and having a lower life expectancy (say 30 years), the burden in DALYs is equal to about 51.59 (50 years of life lost due to early death; 30 years of malnutrition * 5.3% = 1.59 YLD), which is about 60% the burden of an infant death.

As for where these numbers come from (why is a year of blindness 60% as bad as a year lost, and a year of wasting 5.3% as bad?), they were obtained through a variety of methods usually involving surveying groups of people on their subjective attitudes (Pg 50 has more on this). The complete list of disability weights – giving a conversion factor for every kind of health condition analyzed by the GBD – is found on pages 119-125.

This basic framework – evaluating all health burdens in terms of “life-years,” with a year lost to death counted as a full year and a year otherwise afflicted counted according to the disability weights – is common to all DALY calculations. In the next post on this topic, I’ll discuss some of the variations between different versions of DALYs; some versions “discount” life-years that are early in a person’s life, late in a person’s life, or far in the future. After that, I will explain what we think the limitations of this metric are as it applies to our work.

Sources

  • Copenhagen Consensus Center. Copenhagen Consensus 2008. http://www.copenhagenconsensus.com/Home.aspx (accessed April 15, 2010). Archived by WebCite® at http://www.webcitation.org/5p0sJczhJ.
  • Jamison, Dean T. et al., eds. 2006. Disease control priorities in developing countries (2nd Edition) (PDF). New York: Oxford University Press.
  • Lopez, Alan D. et al., eds. 2006. Global burden of disease and risk factors (PDF). New York: Oxford University Press.
  • World Health Organization. Global burden of disease (GBD). http://www.who.int/healthinfo/global_burden_disease/en/index.html (accessed April 15, 2010). Archived by WebCite® at http://www.webcitation.org/5p118giwH.

Next in series:

Infant mortality and overpopulation

(updated on: ) | by Holden

When looking at programs that mostly target infant mortality, I’ve mostly thought of them as “population-increasing” programs. I’ve sympathized with donors who say that bigger populations might be the last thing poor villages need, and I’ve also assumed that “strict utilitarians” are likely to value such programs more than I do. It’s interesting to see the Report of the Commission on Macroeconomics and Health try to turn this issue on its head:

…high mortality rates of children tend to provoke high fertility rates among poor couples. In general, the high fertility more than compensates for the high mortality … In one numerical illustration, households whose children have a 75-percent survival rate choose to have six children, of whom an average of 4.5 survive. The households whose children have a 95-percent survival rate have two children, of whom an average of 1.9 survive … This pattern helps us to understand the surprising fact that countries with high infant mortality rates have the fastest growing populations in the world…

The report includes charts (see Pgs 35-38) showing a pretty strong association between low infant mortality and low population growth across the world’s nations.

This statistical association doesn’t necessarily mean the above logic is correct; it could be that something else, such as economic prosperity or education, is associated with both low infant mortality and low population growth, and that simply lowering infant mortality wouldn’t reduce fertility.

Not being aware of any studies on this specific relationship, I looked a little further using Gapminder. Rather than looking across the world’s countries (as the Report does), I looked specifically at countries within sub-Saharan Africa over time, which seems more relevant to the hypothesis that lowering infant mortality in high-mortality, high-fertility countries (such as those in sub-Saharan Africa) is associated with lowering fertility.

Since 1950, these countries have seen noticeable declines in both infant mortality and fertility (children per woman). However, the trends don’t sync up. In particular, since 1990, infant mortality has largely remained flat while fertility has continued to decline. (Note that literacy has improved over this time period as well.)

I’m hesitant to go as far as the report in calling infant mortality a primary, or the primary, driver of lower fertility. Still, it is clear at least that reducing infant mortality need not result in population growth. I don’t know of any more thorough studies on the link, and would be interested in any references.

A few thoughts on water

| by Holden

The cause of “water” is one of the more (initially) emotionally appealing, and probably marketable, causes in developing-world aid. Here are some thoughts on the cause, fresh off of reading the Copenhagen Consensus report on it:

From what I’ve seen – both in terms of water-related literature and in terms of general morbidity data – outright lack of water (i.e., dehydration in otherwise healthy people) is not a widespread problem. If you know of data showing otherwise, even for particular parts of the world, please share. However, I think most water and sanitation projects are instead concerned with:

1. Access to convenient water sources. Some people lose hours to maintaining their filters and/or boiling their water for cleanliness; people who live far from water sources can lose far more time (see Pg 11 of the Copenhagen Consensus report for a stark example). Improving water infrastructure may therefore free up time and make them economically better off. However, when this is the goal, it seems important to consider not only how much time potential beneficiaries would save, but how much this time is worth (i.e., what else they could do with it). Depending on market and/or weather conditions, extra time may not translate into extra money, or into much extra quality of life.

2. Access to clean water. Contaminated water can contribute to a variety of diseases that generally cause severe diarrhea (see Pg 34). However, it’s important to note that:

  • Water is not the only source of contamination, and clean water alone – when unaccompanied by other sanitation interventions – can only dent the burden of these diseases (again see Pg 34).
  • There are a variety of ways to purify water at the “point of use,” some of which – like boiling – are extremely simple and relatively inexpensive (see Pgs 90-91).
  • Communities that suffer from contaminated water may also suffer from a host of other health problems (such as malaria and malnutrition) that can be at least as damaging as waterborne infections, while having solutions (such as supplementation, bednets, etc.) than are far cheaper and simpler than the provision of clean water.

In our first year, we saw no cases of well-documented water-focused projects that address key questions such as whether water quality and use were verified, whether an effect on quality of life was documented, etc. Literature on past programs’ effects also seems relatively thin.

At this point, I think of water projects as being pretty far from the sort of “proven, effective, scalable” programs we are looking for. If I change my mind, it will likely be for a program in the first category – providing water to people who otherwise would be spending inordinate amounts of time retrieving it – rather than for a program focusing exclusively on clean water.

Significant life change

| by Holden

If you could accomplish any of the following for the same cost, which would you choose?

(1) Prevent 100 deaths-in-infancy, knowing that in all likelihood these 100 people will grow up to have consistently low income and poor health for their ~40-year-long lives.

(2) Provide consistent, full nutrition and health care to 100 people, such that instead of growing up malnourished (leading to lower height, lower weight, lower intelligence, and other symptoms) they spend their lives relatively healthy. (For simplicity, though not accuracy, assume this doesn’t affect their actual lifespan – they still live about 40 years.)

(3) Prevent one case of relatively mild non-fatal malaria (say, a fever that lasts a few days) for each of 10,000 people, without having a significant impact on the rest of their lives.

For me, the answer is definitely #2. I am very excited by the idea of changing someone’s life in a lasting and significant way (2); I’m much less excited by the idea of a temporary, less significant life change (3), and I don’t think that the quality of a life equals the sum of the quality of the days in it. (1) excites me the least – I just don’t put that much value in “potential lives” (I think the death of a 20-year-old is more tragic than the death of an infant), and I especially don’t put much value in saving “potential lives” riddled with health problems.

I’m not interested in having a long philosophical argument about the validity of my views. I believe that different donors likely have fundamentally different values that you can’t change by throwing any number of thought experiments or philosophical abstractions at them. Our research will aim to serve as many different sorts of donors as possible, rather than holding up one philosophical value set as the “rational” one. But I am interested in what others think, and whether my attitude is common or rare.

To give a quick sense of the practical relevance of this question: programs targeted directly at under-5 mortality (including some vaccination programs and some micronutrient programs) are much more likely to get you (1)-type results; programs that distribute bednets or other health materials en masse are more likely to get you (3)-type results; an economic empowerment program (particularly focused on improved farming techniques) may aspire to (2)-type results, but I believe that these types of results are the most difficult and expensive to bring about.