Early-life infectious disease exposure, the “hygiene hypothesis,” and lifespan: Evidence from hookworm disease

Significance

This study uses a quasi-experimental approach to provide causal evidence regarding the long-term impacts of deworming in childhood on lifespan and morbidity. Deworming before age five is associated with increases in lifespan, and examining morbidity finds improvements in health related to hookworm, but no changes in unrelated domains. Results are consistent with influential but hard to rigorously test theories linking early-life infectious disease exposure to lifespan. They are inconsistent with theorized long-term benefits of early life parasitic infections, especially the potential role of reduced helminths in the rise of allergies over the 20th century. Taken together, deworming in childhood leads to improved health and lifespan, many years later.

Abstract

Exposure to infectious disease in early life may have long-term ramifications for health and lifespan. However, reducing pathogen exposure may not be uniformly beneficial. The rise of modern sanitation and reduction of infectious diseases has been implicated in increasing levels of allergy and immune dysregulation: termed, the “hygiene hypothesis.” This study leverages quasi-experimental variation from combining precampaign hookworm exposure with the Rockefeller Sanitary Commission’s deworming campaign in the early 20th century to rigorously examine the impacts of childhood hookworm exposure on adult lifespan and morbidity. Findings show deworming before age five leads to 2.5 additional months of life in a large sample of adult death records. Further, decreasing hookworm exposure is related to improvements in biomarkers for inflammation and skin-tested allergies, in contrast to predictions of the “hygiene hypothesis.” Placebo tests using health outcomes that should not be affected by deworming do not show similar patterns. Overall, childhood deworming leads to improvements in morbidity and lifespan decades later.

Over 400 million people globally live with hookworm (1). Seminal work finds human capital and short-term health benefits from mass deworming (2, 3). However, the value of large-scale deworming continues to be debated, and long-term evidence on the consequences of deworming is thin (4, 5). This study contributes rigorous evidence on the adult lifespan and morbidity consequences of childhood deworming, using the Rockefeller Sanitary Commission’s (RSC) deworming campaign in the US South as a natural experiment.^*

I also use deworming to test the pathways linking early life infectious disease exposure and adult lifespan. While in-utero conditions have well-known consequences, less is known about long-term health impacts of childhood exposures (6). Reductions in early-life infectious disease has been implicated in 20th-century improvements in older-age mortality, and prior work links cohort infant mortality and respiratory disease prevalence to adult outcomes (7–15). Nonetheless, causal evidence and mechanistic understanding of the link between early life health and later life outcomes remain sparse. One challenge is that separating in-utero maternal exposure from childhood exposure empirically is difficult. Additionally, infectious disease exposure may be endogenous to other socioeconomic factors (15, 16).^† This study’s approach addresses these concerns, leveraging rapid unanticipated reductions in hookworm prevalence that primarily affected children.

While prior work implicates lifelong improved inflammation and nutrition as mechanisms linking early life health and adult lifespan, mechanisms are rarely tested directly (7, 8, 18, 19). Further, it is not theoretically clear that reduced infectious disease burdens are uniformly positive. The “hygiene hypothesis” suggests pathogen, particularly helminth, exposure may play a distinctive beneficial role in immune regulation (20, 21). Helminth reductions have been linked to the rise of allergies and autoimmune conditions, but murine and human evidence is discrepant, theoretical mechanisms are contested, human evidence is mixed, and models and methods to critically evaluate the hygiene hypothesis have been limited (21–23).^‡ This paper finds long-term physiologic benefits of deworming, with implications both for the mechanistic link between early life infectious disease, and for the role of hookworm in the hygiene hypothesis.

I study the RSC mass deworming campaign, using an approach similar to Bleakley, 2007’s study of deworming on schooling (3). Hookworm, primarily Necator Americanus, in the early 20th century United States was principally a disease among children that caused very little acute mortality, but potentially serious chronic morbidity and inflammation (22, 30). Hookworm larvae infect the host through the skin of the feet, maturing eventually in the wall of the small intestine where they can survive between one and four years, excreting eggs via feces (31). Eggs hatch and larvae develop in the soil, preferably in wet and sandy conditions, creating variation in hookworm suitability. This life-cycle underpins the variation in hookworm prevalence, which is primarily related to soil type and number of frost-free days,^§ and also lays the groundwork for the work of the RSC: deworming individuals and preventing ongoing transmission.

In 1909, the RSC was established to eradicate hookworm in the American Southeast, with deworming at scale by 1913 (32). Deworming was pharmaceutical using thymol, but the campaign was broader. Given that hookworm’s recognition as a disease was relatively recent, a significant amount of education was necessary to encourage adoption (32). One common approach was a “dispensary” setup where larvae could be shown microscopically in children’s stool and administered treatment onsite. At the same time, the RSC partnered with states to develop systems and infrastructure (such as privy and latrine building) to prevent transmission and enable continued treatment and eradication hookworm after the campaign (32).

Precampaign hookworm prevalence from the Rockefeller Archive is plotted in Fig. 1 (29). Nearly 40% of children had hookworm, with substantial geographic variation (33). Rapid, unanticipated reductions in hookworm due to the RSC campaign provides a natural experiment to explore the long-term health effects of hookworm.

Fig. 1.

This figure shows estimates of childhood hookworm prevalence from systematic county surveys of children preintervention conducted by the Rockefeller Sanitary Commission. County data are aggregated and displayed at the State Economic Area (SEA). Prevalence data were digitized by Roodman, 2018 from the Rockefeller Foundation Archive (29).

My empirical approach compares individuals born in areas with varying baseline hookworm exposure, before and after the RSC campaign. I examine the impacts of deworming before age 5 on lifespan in a sample of nearly 4 million older adult deaths drawn from Social Security Administration records. Importantly, I use variation across counties, but within states, and account for time-specific changes in each state. In parallel analyses, I draw biomarker outcomes from nationally representative samples, including immunologic processes such as allergy skin tests and erythrocyte sedimentation rate (an inflammation biomarker), other hookworm-related outcomes including body-mass index (BMI) and hemoglobin, and a set of placebo health outcomes that are unrelated to hookworm exposure. This study provides rigorous evidence on the long-term impacts of hookworm on lifespan, as well as mechanistic insight into the ways hookworm may shape long-term health.

Results

Study Populations.

This study draws data from two populations. The primary lifespan analyses draw nearly 4 million death records from the Berkeley Unified Numident Mortality Database (BUNMD), derived from a 2013 release of Social Security Administration Numident mortality records (34, 35). In primary analyses, I draw data from deaths between 1988–2005, the BUNMD “high coverage” years where over 95% of older adult deaths in the United States are included. Thus, this paper estimates conditional lifespan: age at death conditional on death between 1988–2005. I relax this restriction and evaluate earlier and later death years with lower mortality coverage as a robustness check. I also use census data to verify that selection into the sample is unrelated to the RSC campaign (SI Appendix, Table S2). The sample is weighted to reflect the human mortality database. Critically, for each decedent these data include county of birth, as well as age and date of death. For empirical models, I aggregate birth counties into State Economic Areas (SEAs) with stable borders over time. I restrict to decedents born between 1900–1935 in areas where the RSC was active and where I have precampaign SEA-level hookworm prevalence. This yields an analytic sample of 3,980,291 individuals, 47% of whom are male, and 27% are Black (Table 1).

Table 1.

Summary statistics by sample

	Unified numident	NHANES
	Decedents	Survey participants
Lifespan (years)	77.6	–
	(7.2)	–
Age (years)	–	57.18
	-	(10.69)
Proportion male	0.47	0.46
	(0.50)	(0.50)
Proportion black	0.27	0.15
	(0.45)	(0.35)
Effective years	2.92	2.34
	(1.34)	(1.85)
% Born in RSC state	1	0.34
	–	(0.47)
% Major urban areas	–	0.46
	–	(0.50)
% Rural areas	–	0.39
	–	(0.49)
% HS grad+	–	0.47
	–	(0.50)
% College grad+	–	0.08
	–	(0.28)
% Low income	–	0.24
	–	(0.43)
N	3,980,291	18,377

This table presents means and (SDs) for individuals in both of the analytic samples used for these analyses. Mortality data are drawn from the Berkeley Unified Numident Mortality Dataset. Morbidity data are drawn from the NHANES I and NHANES II. Individuals represented were born between 1900 and 1935.

The distribution of birth years smoothly covers my birth cohorts of interest, the distribution of death years increases slightly over the years covered (SI Appendix, Figs. S1 and S3) (34).^¶ The average lifespan in this sample is 77.6, somewhat older than the expected lifespan for these birth cohorts, since my primary analysis of the BUNMD conditions on survival until the latter 20th century.^# It is important to consider that estimates are in a sample of older-age deaths. Selection into the sample on the basis of older ages of death may bias estimates toward 0 if additional individuals who died earlier than the period of observation due to hookworm were less healthy than those who survived into the observation window.^‖ I conduct additional analyses with census data and expanded samples to validate results.

My second sample draws from two waves of the National Health and Nutrition Examination Survey (NHANES), which were fielded in 1971–1974, and 1976–1980. Both surveys are nationally representative probability samples of the United States, and include state of birth ($N=18,377$ born between 1900 and 1935). Unlike the lifespan specification, I use state of birth to assign hookworm prevalence, and thus make comparisons across states. I restrict to individuals born in the contiguous United States, rather than just RSC-active areas. Morbidity outcomes are assessed at an average age of 57.2. 46% are male, 15% are Black, 39% live in rural areas, and 47% finished high school. Results should be interpreted as morbidity in mid-to-late life. I examine erythrocyte sedimentation rate (ESR), a biomarker for background inflammation; number of skin-tested allergies out of 8 skin pricks; body-mass-index, hemoglobin, and a series of placebo outcomes. Some outcomes were only measured in random subsamples of the population.

Impacts of Deworming on Lifespan.

I evaluate the impacts of deworming before age 5 on lifespan in the Unified Numident dataset using a primary specification defined in Eq. 1 that regresses lifespan on the interaction between baseline SEA-level hookworm prevalence and effective years of exposure, with fixed effects for birth SEA by sex, and state of birth-by-year of birth. Effective years of exposure, described in detail in the Materials and Methods, are constructed for each birth cohort to index years of exposure to the campaign before age 5. SEA fixed effects absorb time-invariant SEA characteristics (including baseline hookworm prevalence and time-invariant differences such as economic characteristics and other disease environment), and state-by-birth year fixed effects capture potential state-specific trends. Thus, variation in precampaign exposure to hookworm across SEAs, within states, is used to estimate long-term impacts on lifespan.

I find one year of deworming is linked to 0.042 additional life years (0.504 mo) (Table 2). This effect is qualitatively large–5 y of exposure adds 0.21 additional years of life at older ages, similar to the life expectancy gained by eliminating Alzheimer’s disease or half of cerebrovascular disease in older age, or removing lead in-utero (36, 37).^**

Table 2.

Change in lifespan per year of exposure to deworming campaign

	Lifespan (years)
	All	Males	Females
A. Primary specification
SEA prevalence	0.042***	0.034**	0.046***
$\times$ Effective exposure years	[0.012]	[0.015]	[0.016]
Constant	77.765***	76.100***	79.191***
	[0.012]	[0.016]	[0.015]
N	3,980,291	1,851,939	2,128,352
Implied effect of full exposure (years)	0.211	0.171	0.228
B. Allow for SEA-specific mean reversion
SEA prevalence	0.042***	0.035**	0.045***
$\times$ Effective exposure years	[0.012]	[0.015]	[0.017]

This table presents estimates of the effect of exposure to the deworming campaign on the age of death, first pooled then stratified by sex. Lifespan data are drawn from decedents in the Unified Numident file (BUNMD), and include deaths in high-coverage years from 1988 to 2005. The sample includes decedents born between 1900 and 1935 in the 11 states the RSC was active, in the SEAs where preintervention surveys were conducted. Hookworm prevalence is assigned at the SEA of birth. Estimates are conducted using fixed effects by sex that reflect birth SEA, and birth state-by-birth year. SEs presented in brackets, clustered at the SEA level. The coefficients presented are the interaction between baseline SEA hookworm prevalence and effective years of exposure and can be interpreted as the implied effect of one full year of exposure to deworming. Each individual received up to five years of exposure. Panel (B) shows the same model estimated allowing for SEA-specific mean reversion based off of SEA lifespan in the year prior to the campaign’s initiation.

The point estimates for females are larger than for males, though the difference is not statistically significant. This is potentially consistent with the fact that women have conventionally benefited more than men from improvements in the infectious disease environment (38). Ability to examine other dimensions of heterogeneity is limited in the mortality data, though exploratory analyses suggest that impacts may have been larger for males in lower-income states in 1900 (SI Appendix, Table S3) (39).

Two event-study style analyses support the argument that changes in lifespan are driven by deworming due to the RSC-campaign. In the Top panel of Fig. 2, an event-study with a full set of fixed-effects from the primary specification examines the change in the estimated relationship between baseline hookworm prevalence and lifespan within each birth cohort, relative to the last birth years before the intervention began. In the 12 preceding years, there is no evidence that the outcomes for areas with more hookworm prevalence were evolving differently from those with less, supporting the validity of the parallel-trends assumption. After the RSC campaign, the coefficient changes positively. On the Bottom panel, a similar analysis is conducted, but the SEA fixed effects are excluded, so the baseline hookworm prevalence–lifespan relationship can be estimated directly for each birth cohort. There are no systematic trends related to hookworm prevalence in the preperiod. In the post period, the coefficient goes from negative to essentially 0, suggesting that the relationship between baseline hookworm prevalence and lifespan did not just become more positive, but that large-scale treatment of hookworm rendered baseline hookworm prevalence and lifespan uncorrelated after it occurred.

Fig. 2.

These figures show estimates of the relationship between baseline state economic area (SEA) hookworm prevalence and lifespan for people born in a given SEA, estimated for different birth cohorts. The Top panel plots event study estimates, including fixed effects by sex that reflect birth SEA, and birth state-by-birth year. Plotted coefficients are the interaction term between birth year and hookworm prevalence, reflecting the hookworm prevalence coefficient in each birth cohort, relative to the years immediately precampaign. The Bottom panel relaxes this model, dropping the birth SEA fixed effects. Thus, the relationship between baseline hookworm prevalence and lifespan can be directly estimated and plotted for each birth cohort. Blue markers plot coefficient estimates and 95% CIs within each cohort over time, visualizing the trajectory of the changing relationship between baseline hookworm prevalence and lifespan around the time of RSC campaign initiation. Red squares show the normalized average effective years of exposure in each set of year bins, to map plotted results onto the functional form applied in the primary estimate in Table 2. In order to visualize the functional form of effective years of exposure assigned alongside the point estimates, I regress the effective years on the point estimate, and plot the normalized values. Detailed effective years for each birth cohort are plotted in SI Appendix, Fig. S5.

Results are robust to accounting for potential mean reversion.^†† Results are also robust to a variety of alternative specifications and functional forms including an extended event study, linear years of exposure before age 5, a binary pre/post specification, and a semiparametric specification using tertiles of baseline hookworm prevalence, as well as including controls for birth year interacted with rurality, climate, and soil suitability for cultivation from the FAO-GAEZ (SI Appendix, Tables S4 and S6 and Fig. S6).^‡‡ They are also robust to alternative approaches to calculate SEs that account for geographic correlation beyond the SEA-level, and a randomized inference procedure (SI Appendix, Table S5).

A potential concern is that these data do not record lifespans for deaths that occur earlier in the 20th century, or deaths after 2005. Censoring of deaths before 1988 is the primary concern as a much larger share of deaths for the relevant cohorts occurred before 1988 than after 2005. If hookworm led additional individuals to die earlier than the period of observation, and those individuals were less healthy than those who would have survived anyways, the effect I find is an underestimate of what one would find if all deaths were observed. Nonetheless, I conduct parallel analyses of survival until 1960 using state data from the census, focusing on the same RSC-campaign states, and find deworming significantly increases the probability of survival (SI Appendix, Table S7). I also use extended samples of the Unified Numident data, adding the “low coverage” years, to evaluate the potential impacts of missing deaths before 1988 and after 2005. The extended sample covers a smaller share of deaths in each given death year, but raises the sample size by over 30%, including years before 1988 and extending coverage to 2007. In the extended sample, findings are similar, but point estimates are larger, suggesting that the primary estimates from this paper may be underestimating total potential impacts (SI Appendix, Table S8).^§§ I conduct two further robustness checks to examine the extent to which right-censoring specifically may be a concern. Panel (C) only uses the extended data through 2006–2007, increasing the sample size by 21% but only adding right-censored years. Panel (D) truncates the last two years of the high-coverage sample, which alleviates a concern that changes in the distribution of death around the right-censoring threshold drive results. Results are similar in both alternative specifications.

Impact of Childhood Deworming on Morbidity.

Having established the impacts of deworming on lifespan, I examine morbidity to evaluate the pathways through which childhood hookworm exposure shapes long-term outcomes. Hookworm’s primary impacts are related to the immune reaction as the body responds, or due to the anemia and calorie loss caused directly by infection.

The direction of potential immunologic impacts is ambiguous and may be nuanced. In brief, prior evidence suggests exposure to chronic inflammatory processes in early life may resonate years later, and that there may be complementarities between nutritional status and inflammatory environment (8, 19). Other theories, popularized as the “hygiene hypothesis,” suggest hookworm in particular may have beneficial long-term immunologic consequences, though evidence is mixed (20, 22, 23).

Hookworm’s direct impacts on anemia, body mass, and nutritional outcomes may also be persistent. BMI and nutritional status in utero and in childhood may have persistent long-term impacts (6, 40).

I test these theories using data from the NHANES I and NHANES II, estimating a parallel model to the specification for lifespan, using birth-state hookworm prevalence precampaign in the contiguous United States. For immunologic outcomes I evaluate ESR, a biomarker for general inflammation, as well as number of skin-tested allergies, which may reflect background inflammation but also provide a direct test of the allergy predictions in the “hygiene hypothesis.” I also evaluate outcomes related to long-term nutritional status and anemia: BMI, venous measures of hemoglobin, as well as categorical outcomes of anemia (very low hemoglobin).

I find evidence for long-term immunologic improvements related to deworming (Table 3). Estimating the effect of 1 y of deworming on the number of skin-tested allergies finds reductions of 0.09. The implied reduction in the number of allergies is substantial: Five years of childhood deworming exposure is estimated to reduce the average number of allergies by nearly 0.45, approximately 1/3 of a SD. I also find each year of deworming reduces ESR by 0.94 mm/h later in life, which is significant at a 10% size of test. This implies a reduction by 0.4 SDs with full campaign exposure. This is approximately the gap between people born preintervention in the South versus the rest of the country.

Table 3.

Estimated change in adult morbidity per year exposure to deworming (before age 5)

	ESR			# of allergies
A. immunologic changes	All	M	F	All	M	F
Prevalence $\times$	-0.943*	-1.454*	-0.652	-0.090**	-0.056	-0.116*	-	-	-
Effective exposure years	[0.547]	[0.746]	[0.639]	[0.043]	[0.067]	[0.063]
N	6,218	2,660	3,558	8,639	4,219	4,420
Mean	16.95	13.26	19.73	0.55	0.59	0.52
(SD)	(11.67)	(10.6)	(11.67)	(1.31)	(1.33)	(1.30)
Implied effect of full exposure (SD)	-0.404	-0.623	-0.280	-0.345	-0.212	-0.441
	BMI			Hemoglobin			Anemia
B. BMI & hemoglobin	All	M	F	All	M	F	All	M	F
Prevalence $\times$	0.582***	0.663***	0.423**	0.090**	0.095	0.080***	-0.008	-0.012	-0.005
Effective exposure years	[0.170]	[0.181]	[0.205]	[0.037]	[0.062]	[0.029]	[0.007]	[0.011]	[0.008]
N	17,450	8,162	9,288	17,068	7,856	9,212	17,068	7,856	9,212
Mean	25.01	24.81	25.18	14.22	14.93	13.61	0.08	0.08	0.07
(SD)	(4.87)	(4.49)	(5.18)	(1.44)	(1.35)	(1.22)	(0.27)	(0.28)	(0.26)
Implied effect of full exposure (SD)	0.598	0.681	0.435	0.311	0.330	0.277	-0.143	-0.225	-0.090

This table presents estimates of the effect of exposure to the deworming campaign, first pooled then stratified by sex. Preintervention hookworm prevalence is assigned by state of birth. Data are drawn from individuals surveyed in the NHANES I and NHANES II. Models include fixed effects for birth year-by-sex and birth state, controls for age, rural/urban, status, Black race, educational attainment, and low income level. The coefficients presented are the interaction between baseline state hookworm prevalence and effective years of exposure and can be interpreted as the implied effects of one full year of exposure to deworming, and each individual received up to five years of exposure. State hookworm prevalence is rescaled to match county-level prevalence using the relationship in SI Appendix, Fig. S4, to facilitate direct comparison of coefficients with mortality coefficients. In panel (A): Erythrocyte sedimentation rate (ESR) is a biomarker for inflammation, while the number of allergies reflects the number of positive skin-tested allergies out of 8 potential allergens. In panel (B): Body mass index (BMI) is calculated as weight (kg) divided by height(m) squared. Hemoglobin is in grams per 100mL, and anemia is defined as hemoglobin $<$ 13 for men, and hemoglobin $<$ 12 for women. SEs presented in brackets, clustered at the birth state level

Beyond its immunologic impacts, hookworm also may cause persistent decreases in BMI and hemoglobin. I find that exposure to deworming is linked to increased BMI and hemoglobin in adulthood, with full exposure associated with nearly a 0.6 SD increase in BMI, and a nearly 0.3 SD increase in hemoglobin. This appears to be true for both males and females. The increases in BMI are apparent across the BMI distribution, but deworming appears to especially reduce incidence of low BMI and increase subobese overweight (SI Appendix, Table S9). The benefits for hemoglobin appear to be subclinical improvements in hemoglobin status. While the point estimates for anemia are negative, they are small and not statistically significant.

In the NHANES data, unlike the lifespan data, I can control for individual-specific socioeconomic and geographic characteristics at the time of measurement.^¶¶ I include these in the estimates shown, but estimates are minimally affected when a simpler specification with minimal controls is used (SI Appendix, Table S10).

Placebo Health Tests.

I examine health outcomes that are unlikely to be significantly affected by hookworm, in order to test whether other health changes occurring at the same time as the RSC campaign and spuriously correlated with deworming—but not caused by it—might compromise the identification of deworming’s potential long-term impacts (Table 4). I focus on markers of health status that may be sensitive to economic factors or behaviors.

Table 4.

Placebo tests

Variables	Height (m)	Total chol.	HDL	FEV1/FVC
Prevalence $\times$	0.001	-0.851	-1.149	-0.006
Eff. exposure years	[0.004]	[0.882]	[0.811]	[0.004]
N	17,714	15,526	5,695	3,338
Mean	1.64	229.23	50.11	0.76
(SD)	(0.14)	(48.40)	(14.95)	(0.08)

This table presents estimates of the effect of exposure to the deworming campaign on health outcomes that should not be directly affected by hookworm. Hookworm prevalence is assigned by state of birth. Data are drawn from individuals surveyed in the NHANES I and NHANES II. Total cholesterol and HDL (high-density lipoprotein) are measured in mg/dL. FEV1/FVC is a ratio that reflects the volume of air expelled in one second of spirometry over the total amount expelled. Models include fixed effects for birth year-by-sex and birth state, controls for age, rural/urban, status, Black race, educational attainment, and low income level. The coefficients presented are the interaction between baseline state hookworm prevalence and effective years of exposure, and can be interpreted as the implied effects of one full year of exposure to deworming. SEs presented in brackets, clustered at the birth state level

I first examine height in the NHANES. Height is not a perfect placebo outcome, as extreme calorie deprivation can reduce heights, and meta-analyses of deworming randomized trials at older ages find small but statistically significant impacts on height (point estimate approximately 0.3 cm) (5). However, economic shocks such as cash transfers, or parental socioeconomic factors have been shown to have larger impacts (15, 41–44). For example, the PROGRESA conditional cash transfer had impacts approximately three times that of deworming estimates (45, 46). Thus, while height is imperfect and potentially contaminated by hookworm exposure, ruling out large changes in height serves as a check on unobserved health and economic improvements. In this study, while point estimates are small and positive, I find no significant effect of exposure to the deworming campaign and height. Nonetheless, albeit imprecise, the coefficient’s implied deworming impact of 0.5cm is within the CI of meta-analyses of deworming randomized trials (5). These estimates are reassuring insofar as it indicates little in the way of large unobserved in-utero or childhood health changes that may have been occurring at the same time.

I also examine cholesterol measures, which reflect not only health endowments but also health-related behaviors such as diet, exercise, and health care access. I find no impacts of deworming on cholesterol measures.

Finally, I find no effects of deworming on FEV1/FVC, a common pulmonary function test which could be related to smoking, exercise, environment, or obstructive lung conditions (47, 48).^##

Discussion

I examine the impact of deworming before the age of 5 on adult lifespan and morbidity later in life. In doing so, I provide evidence on the long-term impacts of deworming, as well as tests of 1) the linkage between early-life inflammatory exposure with later life morbidity and mortality, and 2) the role of hookworm in the “hygiene hypothesis” in the 20th century. I find substantial gains in lifespan at older age, as well as long-term reductions in a biomarker of inflammation and the number of skin-tested allergies, as well as increases in BMI and hemoglobin. Similar changes are not observed in health outcomes that should be unaffected by deworming.

The estimated impact of deworming on lifespan at older age is large, corresponding roughly to the older-age lifespan lost due to Alzheimer’s Disease. These findings are consistent with, and provide rigorous evidence for, influential theories that infectious disease burdens early in life shape mortality across the life course (7, 18). Men and women benefit similarly from deworming, though point estimates for mortality and some morbidity outcomes are larger for women.

The morbidity findings provide mechanistic support for many of the theorized links between early life infectious disease and adult lifespan. Changes in ESR are consistent with long-term changes in chronic, low-grade, inflammation that may raise cardiovascular disease and cancer risk (50, 51). Further, changes in early-life health such as nutritional status and hemoglobin may be persistent, leading to long-term benefit (8, 52). In this sample of older adults with relatively low BMI in the 1970s, reductions in BMI $\le$ 20 are likely protective against frailty. Changes at higher BMI may have mixed impacts—while BMI over 25 is considered overweight and may have adverse consequences, especially with long-term exposure, BMI of approximately 25 has been considered historically mortality-optimal, and in recent cohorts “optimal” BMI has been between 25 and 30 as medical technology has changed (53–55).

While the improvements in mortality can be attributed to meaningful observed improvements in underlying health changes from early childhood, it is possible that deworming may lead to economic benefits that may also play a role (2, 56). In a study of the economic ramifications of under-18 exposure to deworming, Bleakley finds no long-term impact on educational attainment, but small increases in income. In my analyses of morbidity, I directly control for a variety of potentially relevant socioeconomic pathways including education, poverty status, and rurality. These controls minimally impact observed results. Additionally, the lack of impact on placebo health outcomes suggests that potential economic improvements were not driving generalized improvements in health status.

The joint reductions in background inflammation as well as allergies further an argument that helminths may in fact raise allergies through chronic inflammatory processes (23). Notably, these findings contrast sharply with a potential role of hookworm in the “hygiene hypothesis” (20–23, 25, 57). This study provides rigorous evidence of long-term impacts between hookworm and allergies, in a space where mouse models and short-term human trials are conflicting. However, the interpretation of these findings is limited by the context of the RSC deworming campaign in the early 20th century. The role of hookworm in autoimmune processes may vary in contexts with differing levels of background pathogens (58).

This study also has substantial implications for mass deworming programs around the globe. The efficacy and cost-effectiveness of mass deworming has been debated in recent years, with mixed results on the impacts of hookworm on short and medium-term outcomes, and theoretical arguments made that the potential role of hookworm in autoimmune conditions may undermine the value of deworming (4, 21, 59). Prior work on the RSC campaign has found long-term improvements in income and contemporaneous improvements in school attendance (3), and contemporaneous work on male deaths finds health impacts of deworming and education may be complementary for adult lifespan (60). (This study complements independent, simultaneous work by Noghanibehambari and Fletcher (60). Mortality findings are qualitatively similar, though several important differences are notable. I use a sample four times as large, and include both men and women, rather than just men. This is particularly important, as some effects are larger in women. Additionally, this study uniquely explores the long-term health impacts, and potential physiologic mechanisms by which deworming may affect long-term mortality.) This study provides evidence for life-long benefits to deworming, with persistent impacts on health and age at death. Impacts in lower-income states appear to be similar to if not larger than in wealthier states, furthering the case for deworming in low-income settings. These benefits make a strong case for expanded deworming programs in endemic areas around the globe.

This study has several limitations. One limitation is that the primary source of mortality data primarily draws from the later 20th century, potentially missing earlier-life impacts of hookworm, or to a lesser extent impacts after 2005. Hookworm should minimally impact childhood mortality, but may have impacted mid-life mortality, which would not be measured in this paper’s primary analysis. Analyses at the state level using census data on survival rates until 1960, which should reflect mid-life mortality, and morbidity analyses that reflect effects in midlife, both suggest impacts of hookworm occurring before the left-censoring occurred. Extensions of the Unified Numident sample into earlier and later years of death, find larger impacts of hookworm exposure relative to the primary sample. Primary estimates are thus possibly underestimates of true lifespan effects.

Another limitation is potentially unobserved changes in the health status across cohorts before and after deworming begins, in a way that is related to hookworm prevalence. I address this in several ways. 1) Using a model that includes state-by-birth year fixed effects in the primary mortality specification, I am able to account for potential state-specific trends and changes over time, and localize the variation used to across-SEAs within states. I also extend the model to include controls for time trends that may be related to SEA characteristics such as rurality, climate, and soil suitability for agriculture. 2) Event-study analyses suggest that in the years before the campaign, and the years after full effects are realized, mortality dynamics are not evolving in a way related to baseline hookworm prevalence. 3) I provide evidence from placebo health outcomes that there were not major changes in health status unrelated to hookworm-specific morbidity.

I conclude that deworming in early childhood has large life-long benefits for morbidity and mortality. Mechanistic evidence finds support for long-term immunologic improvements due to deworming, and other health benefits. In this context, hookworm reduces allergies in a way that is not consistent with a role in the “hygiene hypothesis.” Potential long-term benefits should be considered when evaluating mass deworming programs.

Materials and Methods

The intuition for these analyses is analogous to a difference-in-differences approach, similar to Bleakley, 2007. While conventional difference-in-differences uses a treatment variable that takes the value of 1 for groups eventually treated, and 0 untreated, I use continuous hookworm prevalence from prior to the intervention that takes values between 0 and 1. Comparing birth cohorts before and after the deworming campaign from areas of varying hookworm prevalence enables a causally interpretable estimate of the long-term consequences of childhood deworming.

Data Sources for Hookworm Prevalence.

Prior to the initiation of the deworming campaign, the RSC sampled at least 200, but often many more, children in each county. They examined stool microscopically for hookworm ova. Using data digitized by Roodman from the Rockefeller foundation archives, I aggregate counties into SEAs with stable borders over time (29, 30). As shown in Fig. 1, prevalence varied within and across states before the campaign. Hookworm suitability was primarily a function of land characteristics and temperature: Sandy soils were preferable for hookworm reproduction. Note that many of the largest urban centers were excluded from the preintervention prevalence data and activity. People born in these areas are also excluded from analysis. Prior analyses of preintervention hookworm prevalence found more rural and less Black counties had higher prevalence, but the primary variation was predicted by soil type and number of frost-free days (32). I use this preintervention hookworm prevalence data in my primary mortality specification.

In other analyses, I take a complementary approach comparing hookworm prevalence across states. To expand my sample to the contiguous United States, I reference complementary hookworm data from army recruits (61). For consistency, I only use the army data as a state-level measure in these analyses in RSC and non-RSC states.

Analyses of the 11 overlapping states suggests that SI Appendix dataset well-approximates systematic county-level data collection from the RSC (SI Appendix, Fig. S4). The primary difference is that army prevalence is lower, on average, than the county data for a given prevalence level. This is likely due to differences in the age of the samples. (Kofoid notes that the burdens identified were the remaining light-intensity residual hookworm burden as hookworms in young soldiers died and they were not reinfected.) In order to make the coefficient estimates from the SEA-level mortality analyses and state-level morbidity analyses comparable, I rescale the army prevalence estimates to match the RSC distribution using the regression presented in SI Appendix, Fig. S4. Note this adjustment only impacts the magnitudes of the coefficients, not statistical inference.

Defining Exposure to the RSC Campaign.

In order to account for 1) partial treatment of children born right before the initiation of the RSC campaign, and 2) that it took several years to achieve high penetrance of deworming, I construct a continuous measure of “effective years” of exposure to the campaign before age 5. To model the increasing efficacy of the campaign over time, I use aggregated data from a subset of counties where repeated follow-ups occurred. In these samples, in 1910, 59.7% of children had hookworm, falling to 39.7 in 1915, to 21.7 by 1920. This suggests yearly reduction in hookworm prevalence by a factor of 0.9. The functional form is as follows: $\sum _{y=Y1}^{Y5}1[y\ge 1912](1-0.{90}^{(y-1912)})$, where $Y1,Y5$ are the years of birth through age 5 respectively. Thus, each individual gets the sum of “effective years” of exposure to the campaign before they reach the age of 5. The functional form is plotted in SI Appendix, Fig. S5. Analogous to more simpler approaches that use “years of exposure,” in the preperiod individuals have 0 effective years, and by 1920 the functional form is essentially flat. Results are robust to using simpler functional forms (SI Appendix, Table S4).

Primary Analyses of Mortality.

The empirical strategy rests on two core ideas. First, it leverages the varying hookworm prevalence across different areas at baseline. Higher prevalence areas stood much more to benefit from the RSC treatment than lower-prevalence. Second, the study design treats the initiation of the RSC campaign as an exogenous natural shock—given poor baseline understanding of hookworm and the amount of capital investment needed to intervene, the timing and initiation of the RSC campaign is considered exogenous to later health outcomes.

The core specification for the lifespan outcome is as follows:

$$\begin{array}{c}\hfill Y_{\mathit{igcst}}=\beta (W_c\ast EffectiveYrsExpose{d}_t)+{\theta }_{\mathit{st}}+{\lambda }_{\mathit{gt}}+{\gamma }_{\mathit{cg}}+{ϵ}_{\mathit{igcst}},\end{array}$$ [1]

Where $Y_{\mathit{igcst}}$ is lifespan for an individual $i$ of sex $g$, born in state economic area $c$ and state $s$, in year $t$. $W_c$ refers to the hookworm prevalence at baseline in SEA indexed by $c$, and $\mathit{EffectiveYearsExposed}$ is based on birth year and is the effective number of years exposed to the RSC campaign. I allow the fixed effects for year and geography to vary by sex $g$ as potentially important sex-specific trends may have been different for males and females during this time (62, 63). In regressions separated by sex, these effects collapse to just one term for birth SEA, state, or year. Thus, ${\lambda }_{\mathit{gt}}$ reflects sex-by-birth year fixed effects and ${\theta }_{\mathit{st}}$ represents the state of birth-by-birth year fixed effects, which capture state-specific shocks and time-trends. ${\gamma }_{\mathit{cg}}$ are gender-specific fixed effects for SEA $c$.

Note, ${\gamma }_{\mathit{cg}}$ absorbs the baseline worm prevalence, along with observed and unobserved SEA-level characteristics that may be relevant for lifespan and are not changing over time. While there are some concerns for this study about limited coverage of deaths at younger ages and selection into the sample, the birth state by birth year fixed-effect capture cohort-level endogeneity in the probability a given individual is reflected in the data, to the extent it is fixed across a given cohort. ${\theta }_{\mathit{st}},{\lambda }_{\mathit{gt}}$ also accounts for observed and unobserved state and cohort effects in lifespan. The analytical sample is restricted to RSC-surveyed SEAs.

Taken together, the coefficient of interest $\beta$ can be interpreted as the effect of an additional year of exposure to deworming on age at death, estimated by comparing across SEAs, but within states, that have differing baseline prevalence of hookworm. SEs are clustered at the SEA level.

State-Level Survival Estimates.

In order to address concerns that the primary mortality data are at older ages and may miss earlier-life mortality, I conduct a second series of analyses using survival rates I construct from census data. In brief, I construct survival rates until 1960 from the census for birth cohorts from 1900 to 1925, where the numerator is population in 1960 and the denominator the population alive at ages 5 to 14. I use population observations after age 5 for the denominator to avoid infant mortality. Using these state level survival rates, I estimate a similar approach with state-level hookworm prevalence, and a state fixed effect instead of SEA.

Morbidity Analyses.

I use a parallel strategy to the primary analyses of lifespan, in order to evaluate the impacts of deworming on morbidity. I use a series of biomarkers from the NHANES. Since the NHANES does not have county or SEA of birth, I construct treatments at the state of birth, using state level hookworm prevalence from Kofoid and Tucker, 1921. I use the state-level measures for all states in these analyses.

The core specification for each morbidity outcome is as follows:

$$\begin{array}{c}\hfill Y_{\mathit{igst}}=\beta (W_s\ast EffectiveYrsExpose{d}_t)+{\lambda }_{\mathit{gt}}+{\gamma }_{\mathit{gs}}+\varphi X_i+{ϵ}_{\mathit{ist}},\end{array}$$ [2]

Where $Y_{\mathit{igst}}$ is the outcome of interest for an individual $i$ born in state $s$ in year $t$, of sex $g$. $W_s$ refers to the hookworm prevalence at baseline in states indexed by $s$, drawn from army recruit data, rescaled as described in the Data Sources for Hookworm Prevalence to be interpretable similar to the RSC county-level estimates (61). $\mathit{EffectiveYearsExposed}$ is based on birth year and is the effective number of years exposed to the RSC campaign. ${\lambda }_{\mathit{tg}},{\gamma }_{\mathit{gs}}$ are birth state and birth-year fixed effects that vary by sex, which capture time-invariant state characteristics, and cohort-specific effects. One strength of the NHANES is richer socioeconomic data. I include $X_i$ as a vector of covariates of interest, including NHANES wave, age, rural/urban status, Black race, educational attainment, and low income level. Note that ${\gamma }_{\mathit{gs}}$ will absorb baseline state hookworm prevalence, along with observed and unobserved state-level characteristics. ${\lambda }_{\mathit{gt}}$ will absorb birth year-specific effects. $\beta$ retains a similar interpretation to the primary specification. SEs are clustered at the state level.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

Previously published data were used for this work (https://censoc.berkeley.edu) (35).