Is mild lifelong acne really worse than stunted growth?

A preliminary estimate, and a request for better ones.

Summary

I believe the standard literature estimates for the number of DALYs attributable to a case of stunting are too low, largely because they don’t account for the long term effects. This means that childhood nutritional interventions that reduce the prevalence of stunting may be substantially more cost-effective than previously believed.

Epistemic status

Exploratory and back-of-envelope, ~100 hours invested in an area well outside my expertise. I have moderate-to-low confidence in any single point estimate, but high confidence that the standard literature value of 0.23 DALYs greatly underestimates the lifetime burden of stunting. I’m posting this to be red-teamed. If you have a better method, or know of data I’m missing, please let me know.

AI use

I’ve had many conversations with Claude about this topic, and used Claude to develop the ideas, find references, and help generate the accompanying model. I’ve also previously discussed the topic with Grok, Gemini, and ChatGPT, all in their respective research / long-answer modes. The text of this post is all mine.

Conflict of interest

This post is written in my professional capacity at Semilla Nueva, an organization that makes more nutritious corn. As such, I am biased towards thinking that malnutrition is a big deal. For what it’s worth though, I already thought malnutrition was a big deal - more specifically “one of the largest scale and most tractable causes of human suffering and lost potential for flourishing” - long before I started working at Semilla Nueva.

Why should we care?

Humans need a good diet to grow. Not just enough calories, but a long list of other macro and micro nutrients, including quality protein and minerals like iron and zinc. When those are missing, a child can fail to reach their genetic potential. That outcome is called stunting.^[1] Stunting has been common throughout human history. It became rare in rich countries only in the last couple of centuries, and these days affects about 22% of children under five years of age worldwide. Since stunting is permanent^[2], it affects a similar proportion of adults, actually somewhat higher since their critical early-growth period took place when stunting prevalence was higher.

Malnutrition serious enough to interfere with growth causes a lot of other problems along the way. It weakens immune function, raises the risk of chronic disease, and, at least on the maternal side, raises the risk that the next generation is stunted too. The same improvement in nutrition that made stunting rare in rich countries is probably part of what drove the Flynn effect^[3], because the poor nutrition behind stunting also impairs cognition and lowers wages. As countries continue to develop, the prevalence of stunting will likely continue to decline.

Improving children’s diets to reduce stunting looks straightforwardly good. Stunting is also an unusually convenient thing to measure compared to other effects of an improved childhood diet. Immune function is hard to measure at population scale, IQ is fraught to measure at population scale, but height is trivial to measure and is already being measured ~everywhere. So reduction in stunting, or acceleration of growth, has become a standard outcome for dietary interventions: give some number of children a nutritional supplement, and a fraction of them will shift from below the stunting threshold to above it. There are many studies on the topic. Quantifying total global philanthropic spending is difficult, but my best guess is that about 3-5% of global health funding goes into childhood nutritional interventions^[4].

To bring that outcome into the standard EA framework, we need one more number: how many disability adjusted life years (DALYs) should you attribute to one case of stunting? That, alongside org-specific numbers on costs and reach, makes it possible to calculate DALYs per dollar. There is a widely-used value, but it looks surprisingly low: with 3% discounting, 0.23 DALYs per case^[5], justified in the source with nothing more than “from the assumption.” I emailed the first author of that paper, and he replied that he no longer stands by that estimate, but declined to give me an updated one. For comparison, all using 3% discounting, clubfoot is generally estimated at 7.4 DALYs, lifelong moderate iron deficiency anemia^[6] or lifelong Kwashiorkor (moderate wasting with edema) is 1.5 DALYs^[7]. GBD stopped discounting future DALYs in 2010 and I generally think that’s the right accounting, but include both discounted and undiscounted figures here to enable comparison.

My own current best guess is that the average moderate or severe^[8] case of childhood stunting costs on the order of 2.3 DALYs^[9] (subjective 80% confidence interval 1.2 - 4.2)^[10]. This estimate comes from starting with the Years of Life Lost (YLLs) from stunting, then estimating the Years Lived with Disability (YLDs) by comparing stunting to similar conditions that already have disability weights (like mild acne or anemia) and multiplying that weight by how many years the person is expected to live afterward.

Background on DALYs and GBD (feel free to skip if those are familiar acronyms)
The Global Burden of Disease study is a pillar of the global health and development world; GBD 1990 was the first serious attempt to quantify global disease, and invented the concept of DALYs. The core idea can be summarized in one line: disability-adjusted life years equal years of life lost plus years lived with disability (DALYs = YLL + YLD). YLL is relatively simple, comparing the age someone actually dies with their life expectancy. YLD relies on a disability weight (DW) from 0 (perfect health) to 1 (death) multiplied by the duration of disability. Older GBD versions set these DWs by expert judgment; newer ones essentially poll people on which health state they’d prefer^[11] and convert the answers into a weight. Moderate acne, for instance, has a DW of 0.067, meaning that state is treated as about 93% as good as perfect health. A disability weight for stunting existed in GBD 1990^[12], but has been removed and has not been added back^[13].

YLL Component

There’s an excellent new GBD paper on total costs of child growth failure, which computed the total annual DALY burden of stunting on children under age 5 at 33 million per year, basically all as YLLs^[14] - stunted children are more likely to die young due to disease. It’s not that stunting causes increased disease death per se, just that the same nutritional deficiencies that lead to stunting also lead to reduced immune function. Specifically, about 23% come from increased risk of death from diarrheal disease^[15], 52% from increased risk of death from lower respiratory infections^[16], 13.5% from increased risk of death from malaria^[17], and 11% from increased risk of death from measles^[18]. Combining that with the fact that about 132.5 million children are born per year^[19] and that 22.3% of children are stunted, we can calculate stunted children, on average, lose 1.1 DALYs^[20] during their first 5 years of life^[21]. However, that GBD paper is making the unusual decision of including mild cases of stunting (height for age Z score or HAZ < -1), whereas most literature only includes moderate and severe stunting - so just naively dividing them uses too large a numerator. To address this, I model attributable burden across the HAZ distribution, and estimate that 59% of the burden comes from HAZ < -2 (Tabs E and F of the model), i.e. moderate and severe stunting, which brings YLLs to 0.66. In practice, this comes out to a bit under 1% of stunted children dying before age 5. The lack of accounting for any long-term effects like reduced cognitive function makes sense given that stunting doesn’t have any disability weight. 

YLD Component

If a stunted child survives, how might we quantify the disability state of being stunted? Without running the massive surveys GBD uses to establish the DWs, the best I can do is to consider various other health states, anchoring on some that seem worse and some that seem better, with the goal of triangulating in on a DW. This of course involves some judgement calls, which is why I wanted to do the thinking in public and invite other perspectives. It’s widely accepted that stunting (or more precisely the nutritional deficiencies that lead to it) has long-term effects on cognition, but the standard GBD framework doesn’t account for the long-term effects because it’s challenging to do well^[22].

To make that comparison, I excerpted 27 (of about 2100) rows of the health sequelae DW table from GBD 2019, choosing conditions either related to stunting and malnutrition or to things I feel able to imagine, like acne or hair loss. I’m including 8 here, with more in the “DW Comparison” tab of the model.

Intellectual disability

Stunted people have lower wages - perhaps somewhat due to the stunting itself via the same effects that cause a wage return to people in the middle of the height distribution in rich countries and due to reduced capacity for physical labor from the smaller body frame, but almost certainly primarily due to cognitive impairment not caused by stunting but caused by the same childhood malnutrition that caused the stunting. Many papers and press releases simply state that stunted people have lower wages by 10-20% compared to people of similar SES who are not stunted. Victora et al found that 1 Z-score of height is associated with an 8% increase in wage^[23], which I use in the model (Tab D). This means that an average case of moderate stunting implies a 12% reduction in wage, whereas the average severe case implies a 21% reduction^[24]. In light of this information, my best guess is that the average case of stunting is associated with intellectual losses roughly equivalent to the average borderline intellectual functioning^[25] - though of course the cognitive effects of malnutrition, like the height effects, vary significantly in severity. Following GBD convention, I’m not including lost wages as a health effect, so the direct DALY contribution from wage is 0, and is instead used to help inform my reasoning about how bad the cognitive effects are through other quality of life pathways.

Table 1: Some conditions one might compare stunting to to assess YLDs.

Upper and lower anchors

In estimation and forecasting, I often find it helpful to also establish bounds that I’m confident the true value must lie between - depending on how the remainder of the process goes, these can either comprise the final uncertainty interval, or can be a broader superset of the plausible range of values.

Lower anchor: Moderate and severe stunting cause intellectual disability that seems roughly equivalent to the GBD health state “borderline intellectual function”, in addition to all the other sequelae of stunting like shorter stature, reduced physical strength, and higher probability of having offspring below a healthy weight range. As such, I’m happy to take that DW of 0.011 as a lower anchor^[26]. Multiplied through low-SES in the developing world life expectancy of 67 years^[27], we get 0.74 YLDs as the floor. The same math applies for lifelong mild alopecia areata (spotty hair loss) or mild acne. To account for the possibility that this judgement is wrong and mild acne is in fact worse, I applied a further 25% haircut, expanding the bottom of the CI down to 0.55.

Upper anchor: Severe wasting is an acute state of near-starvation, and never sustained long - people either recover or die. It’s definitely a worse state than being stunted. Moderate anemia is also almost certainly worse, and comes in at 3.5 YLDs if sustained for life, which I think is a good upper bound for at least moderate stunting. To account for the possibility that moderate anemia is actually less bad than stunting, the upper edge of the CI is a bit higher, at 4. For more comparisons, see Tab B of the model.

Table 2: Headline estimates

Conclusion

Stunting is definitely bad, and hopefully something that will continue to become rarer during the coming decades. Stunting kills a bit under 1% of those it afflicts before the age of 5, which puts the average YLLs for a case of stunting at 0.66. Those who survive have some level of mild ill-health throughout the remainder of their lives, a combination of shorter stature, reduced strength, and a mild cognitive impairment. Weighing the YLDs of those sequelae is challenging and imprecise, but my best guess from modeling the major effects at various levels of stunting gives a population-weighted average for moderate and severe stunting of 1.6. Combining the YLLs and YLDs, this means the average case of stunting^[28] costs about 2.3 DALYs, well above the standard literature value and worse than a lifetime of mild acne, though still only about half the value for lifelong moderate acne. This means that early-life nutritional interventions that can reduce growth failure may be substantially more cost effective than previously believed, and it may therefore be justified to increase investment in such interventions.

Acknowledgements

Thanks to Kyle Scott, Sam Anschell, Katie Adams, Tony Senanayake, and others for reading and providing feedback on drafts of this post. All remaining errors are my own.

Model
Tab A: Central estimate

Tab B: DW comparison
Tab C: Stunting facts (built as a reference and scale-check)

Tab D: Cognitive comparison

Tab E: Continuous HAZ model

Tab F: Continuous HAZ result table

1. ^{^}

   Formally, height-for-age Z-score (HAZ) below −1 standard deviations relative to the reference population qualifies as mild stunting, below -2 is moderate, below -3 is severe. This essay, following a widespread convention, is primarily concerned with moderate and severe stunting.

2. ^{^}

    Catch up growth is possible at early ages, but someone stunted at age 5 is stunted for life.

3. ^{^}

   The Flynn effect is the long-run rise in measured IQ across the twentieth century. Better childhood nutrition is one of the leading candidate causes, and is almost certainly at least part of the story, in addition to other causes like greater familiarity with test-taking.

^{^}

 This comes from the Institute for Global Health Metrics and Evaluation (IHME), specifically their “Financing Global Health” data portal. If we look at Development assistance for health (DAH) flows in 2025, newborn and child health - nutrition is 2.19 billion of 49.7 billion, 4.4%. To check if 2025 was an outlier, I also checked 2024 (2.39/53.9 = 4.4%), 2020 (1.95/64.6 = 3.0 %, but of course an outlier year), 2018 (2.41/48.3 = 5.0%), 2015 (2.31/45.7 = 5.1%), 2010 (1.36/38.8 = 3.5%), and 2005 (0.707/23.3 = 3.0%). Overall, it looks like the range is roughly 3-5% over the last 2 decades.

^{^}

 In the original Bhutta et al. 2008 paper this value is specifically for children who survive to age 3 but are stunted, but in other papers it’s used for stunting without that carve-out for cases of stunting where the nutritional deficiency that leads to stunting also leads to early death (in most cases due to reduced immune function).

^{^}

 This assumes a 67 year life expectancy, following Trenouth et al methodology. Tab B of the model shows the math to get to this total, and here I’m applying the 3% discount to make a fair  comparison. Of course iron deficiency anemia is often not lifelong, whereas stunting generally is.

^{^}

 Same source and calculations.

^{^}

 Using a population weighted average. Like most stunting-focused literature, this analysis omits mild stunting.

^{^}

 Of course, many nutrition-related interventions will cause widespread shifts in average growth, which don’t perfectly map the question of “how many people will cross the threshold from stunted to not”. This estimate is a downward revision from the ~6 I estimated in an earlier more surface-level investigation, for two reasons. First, I now think the right cognitive anchor is borderline intellectual functioning (disability weight 0.011), not mild intellectual disability (0.043), which cuts the cognitive channel by about four times. Second, most stunted children are only moderately stunted, so a prevalence-weighted average case sits well below the severe-case figure. I tried modeling DALYs per case averted by untargeted interventions shifting the whole HAZ distribution (the case-averted multiplier on Tab A, following math from Tab E), but have low confidence in this modeling.

^{^}

 Modern GBD doesn’t discount health outcomes, but Bhutta et al 2008 did. To compare like with like, my YLD with 3% discounting comes to 0.69, compared to their 0.23, which is why my overall takeaway is “3x underestimate” rather than ”order of magnitude underestimate.”

^{^}

 Yes, one of the most important resources in the academic global health community boils down to a bunch of “would you rather” responses.

^{^}

 I haven’t actually been able to confirm this, many of the documents of GBD1990 are not digitally available. Secondary sources claim it was there with a DW of 0.002, and Table 1 from a summary document of GBD 1990 confirms that wasting was in “class 1” of health states, with DW 0 - 0.02. Given that my argument here doesn’t hinge on this 36 year old claim, I haven’t investigated further.

^{^}

 Stunting appeared in GBD 2013 as a risk factor, which is reasonable since it’s highly correlated with the bad things like reduced immune function. However, the DALYs calculated are only for children under 5, which explains why YLLs dominate and YLDs are nearly absent.

^{^}

 They break down the health burden of stunting into YLLs and YLDs in Tables S1 and S5. Table S1 tells us global all-cause stunting DALYs = 33 million, Table S5 tells us global all-cause stunting YLDs = 59 thousand. This means that only 0.2% of the DALYs are contributed by the vast majority of stunted children who don’t die at a young age - their YLD costs are rounded to 0, which is addressed in the next section.

^{^}

 84,800 deaths per year, Table S3

^{^}

 196,000 deaths per year, ibid

^{^}

 50,800 deaths per year, though note that here the CI crosses 0, so this effect is less certain. Ibid

^{^}

 42,000 deaths per year, ibid

^{^}

 The most recent stunting data is from 2022, so I’m also using that year for the total births. Annual births have varied by something like 20% over the last 3 decades, so there’s a bit of an error bar here, though it will be far from the largest contributor to our final uncertainty.

^{^}

 The math here is challenging with a stock-flow question (it’s easy to get 0.22 DALYs per child-year), I built a simulation to confirm it adds up to 1.1

^{^}

 There’s an accounting question of when to count those losses - if a 3 year old dies, when does the loss of their 21st birthday happen? From the individual perspective, arguably that loss happens in the future. However, the GBD standard is to count the loss in the year of their death, which has some useful accounting properties: for example, if there’s a big war or pandemic, annual DALY stats can show it as a spike rather than “smearing” the loss out over the next several decades.

^{^}

 A senior stunting researcher told me “I agree GBD (currently) misses the long-term cognitive impacts of stunting. That has been something we’ve wanted and tried to add for years, but it is genuinely hard to quantify”. I agree it’s hard to quantify (and my error bars are large), but I don’t think we should quantify it as 0, as GBD currently does.

^{^}

 For men. However, there's no reason to think that the cognitive effects of malnutrition differ substantially for men and women - the lack of significant wage results in women in most papers reflects women mostly doing unpaid forms of labor.

^{^}

 The wage differential is calculated assuming a local mean HAZ of -0.97 rather than from 0 (see Tab E of the model). Calculating from 0 would give substantially larger figures, but I think would overstate the effect in populations with widespread stunting.

^{^}

 The two GBD health state descriptions for comparison: borderline intellectual functioning (DW 0.011, UI 0.005–0.02): “is slow in learning at school. As an adult, the person has some difficulty doing complex or unfamiliar tasks but otherwise functions independently.” Mild intellectual disability (DW 0.043, UI 0.026–0.064): “has low intelligence and is slow in learning at school. As an adult, the person can live independently, but often needs help to raise children and can only work at simple supervised jobs.” Over the 67 years my model applies, the borderline weight would contribute about 0.7 DALYs and the mild weight about 2.9. Unfortunately, these GBD states are defined by those lay terms (that’s where the DWs come from), and are impossible to perfectly map to a given IQ distribution or wage band, so there is inherent uncertainty in picking which cognitive state to map to. The Cognitive comparison tab of my model goes into more detail on why I think borderline is a better fit than mild.

^{^}

 This doesn’t necessarily mean that CEAs based around biofortification should use this exact DW for everyone that crosses the line from moderate to mild stunting, ultimately this is all continuous and the HAZ cutoffs are artificial (if not entirely arbitrary since they’re integer Z scores). To think about this in more detail, I constructed continuous DW models by HAZ as tabs E and F of the model. The model CI goes a bit below this lower bound to account for potential bias in the papers I used to establish the cognitive effects of stunting.

^{^}

 Following Trenouth et al using numbers from Pakistan, averaging male and female life expectancies, 67 years. The model is set up to enable modeling stunting that occurs later in life (e.g. at age 2) and thereby shortening the length of time that the DW applies, but by default it’s set to apply from birth since that’s the most common scenario.

^{^}

 Of course severe cases are worse than moderate, this is a prevalence-weighted average

Show all footnotes