American Journal of Epidemiology Advance Access published June 24, 2014
American Journal of Epidemiology
© The Author 2014. Published by Oxford University Press on behalf of the Johns Hopkins Bloomberg School of
Public Health. All rights reserved. For permissions, please e-mail: [email protected].
DOI: 10.1093/aje/kwu134
Epidemiology in History
The Thompson-McFadden Commission and Joseph Goldberger: Contrasting
2 Historical Investigations of Pellagra in Cotton Mill Villages in South Carolina
Stephen J. Mooney*, Justin Knox, and Alfredo Morabia
Initially submitted November 22, 2013; accepted for publication April 29, 2014.
As pellagra reached epidemic proportions in the United States in the early 20th century, 2 teams of investigators
assessed its incidence in cotton mill villages in South Carolina. The first, the Thompson-McFadden Commission,
concluded that pellagra was likely infectious. The second, a Public Health Service investigation led by Joseph Goldberger, concluded that pellagra was caused by a dietary deficiency. In this paper, we recount the history of the 2
investigations and consider how the differences between the 2 studies’ designs, measurements, analyses, and interpretations led to different conclusions. Because the novel dietary assessment strategy was a key feature of the
Public Health Service’s study design, we incorporated simulated measurement error in a reanalysis of the Public
Health Service’s data to assess whether this specific difference affected the divergent conclusions.
epidemiology in history; measurement; multilevel epidemiology; nutrition
Pellagra
to relapse after active treatment ended (9)), it was reasonable
to expect to find another infectious cause of pellagra. Meanwhile, Christiaan Eijkman’s finding that thiamine deficiency
was the sole cause of beriberi disease provided a model for
disease caused exclusively by nutrient deficiency, even in a
diet with adequate caloric content (10).
With etiology uncertain, early investigations in the United
States described the pellagra epidemic patterns. Throughout
Appalachia, pellagra was prevalent in villages where residents were mainly employed by a cotton mill (11), and it was
associated with seasons, poverty, poor sanitation, and, potentially, diet (12). The growing epidemic led leaders in the affected states to pressure for more intensive investigations.
Pellagra, meaning “sour skin” in Italian, manifests through
the “4 Ds”: dermatitis, diarrhea, dementia, and death. It was
first formally described among Spanish agricultural workers
in 1735 but remained rare in America until the early 20th
century (1). Then, in the period from 1906 to 1940, it reached
epidemic proportions, resulting in more than 3 million cases
and 100,000 deaths (2). Cases were most predominant among
the socially disadvantaged: almost always among the very
poor and disproportionately among blacks and women (3, 4).
As of the early 20th century, pellagra’s etiology was unknown, allowing some to believe it had an infectious origin
(5). Others believed pellagra was caused by poor diet, either
following Casimir Funk’s emerging theory of vitamin deficiency (6) or the longstanding belief that spoiled corn contained a toxic agent (7, 8). Both the infectious and dietary
deficiency hypotheses were supported by recent discoveries.
In the period following Robert Koch’s identification of Bacillus anthracis in 1875, many diseases of unproven etiology,
including tuberculosis, cholera, pneumonia, scarlet fever,
and diphtheria, had been linked to bacteria. Given the resemblance pellagra bore to tuberculosis (e.g., endemic in environments marked by poverty and poor sanitation and
amenable to treatment with environmental change, yet prone
The Thompson-McFadden Commission
Through the influence of George Miller, president of the
New York Post-Graduate Medical School, mining magnate
Robert M. Thompson and cotton merchant Henry McFadden
together donated funds for a commission bearing their names
to investigate pellagra. The Thompson-McFadden Commission, led by Joseph F. Siler and Philip E. Garrison, 2 officers
in the Armed Services Medical Corps, and Ward J. MacNeal,
a doctor affiliated with New York Post-Graduate Medical
School, investigated the epidemic on a large scale.
1
Downloaded from http://aje.oxfordjournals.org/ at Pennsylvania State University on March 5, 2016
* Correspondence to Stephen J. Mooney, Department of Epidemiology, Mailman School of Public Health, 722 West 168th Street, Room 729,
New York, NY 10032 (e-mail: [email protected]).
2 Mooney et al.
Food Consumption Category
Period of
Consumption
7 Times/
week
2–6 Times/
week
Fewer Than 2
Times/week
0 Times/
week
Seasonal
Part-time
daily
Part-time
habitually
Part-time
rarely
Never
Year-round
Daily
Habitually
Rarely
The investigation, which lasted from 1912 to 1914, focused on the Spartanburg, South Carolina, area because of
local cooperation, personal interest of South Carolina Senator
Ben Tillman, and pellagra’s prevalence in the area (4, 13). An
initial survey of individuals with pellagra described the local
epidemic conditions (14, 15). Afterward, the Commission
chose 6 cotton mill villages (Inman Mills, Whitney, Pacolet
Mills, Saxon Mills, Arkwright, and Spartan Mills) to explore
both dietary and infectious hypotheses (16). Every household
in these villages was canvassed, with an interviewer asking 1
household resident (“usually the housewife” (16, p. 294)) for
the individual age, sex, occupation, residential history, and
pellagra status of each resident, as well as questions about
the consumption of corn meal, grits, wheat flour, fresh
meat, cured meat, lard, canned foods, milk, eggs, and butter.
A resident was considered to be a case if a skin lesion was
present at the time of canvass or if both the resident and his
or her treating physician confirmed the prior presence of a lesion (15). Food consumption was categorized as year-round
or seasonal, and as daily, habitual (less than daily but at least
twice a week, on average), rare (twice a week or less), or
never, for a total of 7 categories (Table 1).
The Commission analyzed individual- and householdlevel diets as causes of pellagra, dismissing a dietary cause
Goldberger and the Public Health Service investigation
As the Thompson-McFadden Commission’s work concluded, the pellagra epidemic continued, and the U.S. Congress appropriated funding for a major investigation. In
1914, Joseph Goldberger, who had distinguished himself in
the Public Health Service by combating yellow fever, dengue
fever, typhoid fever, measles, and typhus, was appointed to
lead the investigation. Goldberger reviewed the literature
and concluded that pellagra had a dietary origin because
1) outbreaks occurring in institutional settings never affected
the staff of the institutions (18); 2) among poor Southern
whites, outbreaks and diet followed a similar seasonal pattern
Table 2. Selected Results From the Thompson-McFadden Commission’s Investigation of Pellagra Incidence in
Cotton Mill Village Households in Relation to Self-Reported Consumption of Certain Foods, South Carolina,
1912–1913
Shipped Corn Meal
Timing and
Frequency of
Consumption
Total No. of
Households
Fresh Meat
Households
With
Incident
Case
No.
%
Total No. of
Households
Fresh Milk
Households
With
Incident
Case
No.
Total No. of
Households
%
Households
With Incident
Case
No.
%
6.8
Year-round
Daily
414
38
9.1
13
2
15.3
527
36
Habitually
161
17
10.6
151
21
13.3
112
9
8
85
9
10.6
337
33
9.7
121
9
7.4
Daily
69
6
8.7
8
1
12.5
0
0
Habitually
25
2
8
261
23
8.7
0
0
Rarely
11
4
36.4
45
4
8.9
1
1
100
9
9
46
1
2.1
117
24
Rarely
Seasonal
Never
100
20.5
Downloaded from http://aje.oxfordjournals.org/ at Pennsylvania State University on March 5, 2016
(Table 2). Most foods, including meat and corn, were
found to have no association with pellagra. An observed negative association between milk consumption and pellagra was
noted but dismissed because milk’s “use did not fully insure
against the development of the disease” (16, p. 373). Two
findings appeared to contradict a dietary hypothesis. First,
in contrast to beriberi, pellagra was almost never found
among nursing infants (17). Second, within the surveyed villages, pellagra was more common among whites than blacks,
in spite of blacks’ perceived worse diet (17).
The infectious hypothesis was explored using an ad hoc
classification of every house in the villages as being located,
as of 1912, in zone 1 (containing a pellagra case), zone 2 (adjacent to a house with a pellagra case), or zone 3 (not adjacent
to a house with a pellagra case). Individual zone assignment
of residents who had lived in the zone for at least 2 weeks provided zone-specific incidence rates among all villages, yielding a striking inverse gradient with distance from pellagra
cases consistent across villages (Figure 1). The lower incidence of pellagra in villages with better sanitation systems
was also consistent with the infectious hypothesis (17), and
the Thompson-McFadden Commission ultimately concluded
that an infectious etiology was more likely than a dietary one.
Table 1. Food Consumption Categories in the ThompsonMcFadden Commission’s Household Canvass, South Carolina,
1912–1913
Contrasting Historical Pellagra Investigations
(19); 3) although pellagra was associated with poverty, it
rarely affected poor farmers with cattle (1, 19); and 4) pellagra’s association with diet even among individuals with a
calorically adequate but monotonous diet bore similarities
to beriberi and scurvy (20). Goldberger’s prior failure to
transmit pellagra in monkeys may have given him a perspective the Thompson-McFadden Commission lacked (21).
Working from the vitamin deficiency hypothesis, Goldberger used broad dietary changes to prevent pellagra in an
orphanage and to cause and subsequently cure pellagra in a
prison (22, 23). He next worked to rule out the infectious hypothesis by experimentally using nasal secretions, scabs,
blood, urine, and feces from pellagrins (i.e., individuals
with pellagra) to attempt to transmit pellagra to willing subjects (including himself, his wife, and his close associates)
using techniques (e.g., ingestion, intramuscular injection,
subcutaneous injection) that propagated other diseases;
none of these attempts was successful (24). Yet despite this
evidence in favor of a solely dietary hypothesis, Goldberger’s
views were criticized, including by members of the ThompsonMcFadden Commission, who argued that the selected populations were different than those most at risk (25), and the diet
that Goldberger believed had induced pellagra in prisoners
had in fact merely weakened the body’s resistance to an as
yet unidentified infectious agent (1).
Therefore, Goldberger, with the support of the Public
Health Service, undertook to show that pellagra in the general
population could also be explained by diet. He followed up
on the Thompson-McFadden Commission’s study of pellagra in cotton mill villages by undertaking another study in
an overlapping set of cotton mill villages. The Public Health
Service gathered data at the individual, household, and village levels using a household-by-household canvass and assessed both the dietary and the sanitation-mediated infectious
hypotheses in a study closely resembling the ThompsonMcFadden Commission’s, yet with 2 notable differences.
The first difference between the 2 studies was the extraordinary attention given by the Public Health Service to diet
and pellagra assessment, including an integration of administrative and interview data that, to our knowledge, was unique
for its time. Cohort studies preceding these pellagra investigations were based either on vital statistics alone (e.g., the
work by Weinberg (26)) or on medical record abstraction
(e.g., the work by Lane-Claypon (27)). The case definition
required clearly defined, bilaterally symmetrical dermatitis
as assessed by Goldberger or Goldberger’s assistant, G.A.
Wheeler, who was also a physician. (Figure 2) (28). A
15-day household diet just prior to the peak pellagra season
(between April 16 and June 15) was derived from food supply assessed through company store records and supplemented by housewife surveys of the quantity and value of
food items obtained from other sources (e.g., produced at
home, given to the household, or purchased from neighbors,
farmers, or hucksters). Individual food supply was then
weighted using the Atwater scale (for more detail, see Web
Appendix 1 available at http://aje.oxfordjournals.org/).
The second major difference was the Public Health Service’s selection of a slightly different set of cotton mill villages. Spartan Mills and Pacolet Mills, the larger villages in
the Thompson-McFadden study, were replaced with Newry,
Seneca, and Republic. The change in villages was likely related to a desire to compare sanitation levels in relation to pellagra incidence between villages. The Thompson-McFadden
Commission had previously invoked poor sanitation as supporting an infectious hypothesis by contrasting Seneca’s
Downloaded from http://aje.oxfordjournals.org/ at Pennsylvania State University on March 5, 2016
Figure 1. The Thompson-McFadden Commission’s chart of pellagra
incidence rate by village and by zone, an ad hoc clustering analysis, as
published in the report by Siler et al. (16). Subjects in zone 1 resided in
the same household as a pellagrin (i.e., individual with pellagra), those in
zone 2 resided adjacent to a pellagrin, and those in zone 3 lived farther
away from any pellagrins. The Commission argued that the much higher
incidence rates in zone 1 supported an infectious hypothesis. A, Arkwright, South Carolina; Av, average; I, Inman Mills, South Carolina; P,
Pacolet Mills, South Carolina; SA, Saxon Mills, South Carolina; SP,
Spartan Mills, South Carolina; W, Whitney, South Carolina. Reprinted
from Arch Intern Med. XIV(3):293–373. Copyright ©1914 American
Medical Association. All rights reserved.
3
4 Mooney et al.
pellagra incidence with Newry’s in an analysis focused on sanitation alone (29). Republic, like Newry, was unusual among
cotton mill villages in having an improved sewage system.
The Public Health Service study confirmed several previous results. The peak season for pellagra was late spring (12).
Pellagra was positively associated with young age (between 2
and 10 years) regardless of sex and with female sex in adulthood (13, 30). Pellagra was much more common among poor
households (12) and was associated with diet (Table 3). At
the village level, pellagra was highly variable (Figure 3), but
that variability was not associated with sanitation (which
varied between villages but did not covary with pellagra incidence) (31, 32) or with poverty (which was roughly comparable in all villages) (33).
These findings did not conclusively implicate diet as the
cause of pellagra. No single food or food category was present in all nonpellagrous households and missing in all pellagrous households. Furthermore, because poverty was a cause
both of poor diet and of squalid conditions permitting infectious agents to proliferate, a dietary association with pellagra
did not fully distinguish between dietary and infectious
etiological hypotheses (Web Figure 1). To disentangle the effects of poverty and poor diet, Goldberger supplemented his
between-household comparison with a between-village comparison in an early form of multilevel epidemiologic analysis.
Having previously shown that sanitary conditions at the
village level were unrelated to pellagra incidence, and that
income distribution, food prices, and individuals of susceptible ages were comparable between villages, Goldberger compared a village with high incidence of pellagra, Inman Mills,
to one with a single case, Newry. What differed between the
2, he argued, was food supply. Inman Mills had a company
store with little fresh food and almost no nearby farms,
whereas Newry had a well-stocked store and a wealth of
nearby farmers selling food in town. Household-level purchase records indicated far more fresh meat and milk in
Newry diets, even among the very poor. Thus, community
conditions of food availability provided the best explanation
for the difference in pellagra incidence between Inman Mills
and Newry (33). By quantifying how the availability of food
within the village shaped the specific foods consumed by individuals within households, Goldberger identified dietary
Downloaded from http://aje.oxfordjournals.org/ at Pennsylvania State University on March 5, 2016
Figure 2. Reproduction of a drawing from a 1913 paper on pellagra in Africa with shading to show where lesions characteristic of pellagra were
found on the body (44). Ward J. MacNeal of the Thompson-McFadden Commission included this drawing in a letter to the editor of the Journal of the
American Medical Association arguing that the pellagra that Joseph Goldberger and G.A. Wheeler had induced in prison inmates did not follow the
characteristic lesion pattern (45). This criticism may have led to the case definition in Goldberger and Wheeler’s subsequent investigation of pellagra
in cotton mill villages in South Carolina. Reprinted from Trans Royal Soc Trop Med Hyg. 7(1):32–56. Copyright ©1913 Oxford University Press. All
rights reserved.
Contrasting Historical Pellagra Investigations
Table 3. Selected Results From the Public Health Service
Commission’s Investigation of Pellagra Incidence in Cotton Mill Village
Households in Relation to Supply of Certain Foods per Adult Male
Equivalent as Weighted by the Atwater Scale, South Carolina,
April–June 1916
Food Item and
Adult Male Unita
Supply
Total No. of
Households
Households With
Incident Case
No.
%
6.6
b
Corn meal, lbs
304
20
4.0–7.9
260
24
9.2
8.0–11.9
117
13
11.1
61
4
6.6
<1.0
495
54
10.9
1.0–1.9
131
4
3.1
2.0–2.9
61
2
3.3
3.0–3.9
36
1
2.8
≥4.0
18
0
0
<1.0
154
28
18
1.0–6.9
262
16
6.2
7.0–12.9
163
8
4.9
13.0–18.9
90
4
4.4
≥19.0
58
0
0
≥12.0
Fresh meat, lbsb
Fresh milk, qtsc
a
The Atwater scale accounts for the effect of household size on
resources available to each individual within the household. Any
male over the age of 16 years is 1 adult male unit, and women and
children are calculated as proportions of an adult male unit on the
basis of their expected consumption.
b
1 lb = 0.454 kg.
c
1 qt = 0.946 L.
deficiency patterns, the mechanisms of which that were
not fully explained until biochemical analysis techniques
developed.
From these findings and analogy with beriberi, Goldberger
wove a narrative highlighting the role of diet in explaining
pellagra variation through multiple levels of analysis: age
and sex were related to status at the dinner table and ability
to supplement diet outside the home; income determined
both the ability to purchase high-quality food and portion
sizes of those foods; and village determined access to food.
Furthermore, the recent emergence of pellagra could also be
explained by diet; changes in the Southern economy (which
included a transition from primarily growing food, which was
then available in local markets, to producing cotton) had led
to large-scale dietary changes (34). Overall, this narrative,
coupled with prior results of dietary experiments, made a
strong case for a dietary cause of pellagra.
Why the investigations reached different conclusions
Goldberger’s hypothesis was ultimately found to be correct: individuals with diets deficient in niacin (vitamin B3)
and tryptophan (which the body converts to niacin) develop
pellagra (35). Yet the Thompson-McFadden Commission, a
well-funded, well-intentioned, and professionally run study
reached the wrong conclusions. Why did the Commission
fail? We hypothesize that the failure was due to inaccurate dietary assessment, lack of between-village variability, and an
interpretation of data shaped by an a priori assumption of infectious etiology.
First, the Thompson-McFadden Commission’s dietary
canvassing strategy was not as accurate as the Public Health
Service’s use of company store records. By limiting dietary
assessment to a single per-household interview and not considering household composition, the Commission could not
account for seasonal variation in diet or for variations in portion size among individuals and households. Furthermore, in
a context of severe poverty, some respondents may have reported on idealized diets rather than their actual diets. It is
possible that, given the stigma attached to pellagra (2),
those with disease may have overreported eating meat for
social desirability reasons, although we have no data from
which to assess this hypothesis today. Additionally, the Commission’s acceptance of patient and physician reports of pellagra may have resulted in the inclusion of cases of other
diseases causing dermatitis. The most likely such disease is
ariboflavinosis, a disease caused by riboflavin deficiency,
which also results in skin lesions. Ariboflavinosis lesions
are characteristically located in the mouth and genitals rather
than on the extremities, as in pellagra (36); the Public Health
Service’s case definition likely would have excluded ariboflavinosis (19).
Second, the Public Health Service’s selection of villages to
canvass was also superior in terms of exposure and disease
variability. By including both Newry, with its broader food
supply and improved sanitation system but no cases of pellagra, and Republic, with an improved sanitation system but
prevalent pellagra, the Public Health Service was able both
to examine a greater range of dietary conditions than the
Thompson-McFadden Commission had been and to distinguish between the effects of improved diet and improved
sanitation.
Finally, the Thompson-McFadden Commission ignored
the evidence for the dietary hypothesis, downplaying an observed protective association with milk because some cases
of pellagra reported consuming milk. There is evidence, however, that the Commission could have interpreted its own data
differently. In 1914, Edward Vedder, a nutritionist, wrote a
report enumerating the resemblances between beriberi and
pellagra (37). Invited by the Commission to investigate
their data, Vedder noted that the age and sex patterning, the
seasonality of the disease, and the lack of correlation between
population density and pellagra incidence supported the
dietary hypothesis. Vedder also observed that residents of
Newry, whose dietary patterns the Thompson-McFadden
Commission had not assessed but who were known to have
a superior food environment, had little to no pellagra. He further pointed out that the Commission’s dietary survey did not
capture portion size, and that diet should be considered as a
whole rather than by individual foods. Finally, Vedder noted
2 flaws with arguments for the infectious hypothesis. First,
contact tracing was nearly meaningless in a context in
Downloaded from http://aje.oxfordjournals.org/ at Pennsylvania State University on March 5, 2016
<4.0
5
6 Mooney et al.
70
Public Health Service estimate
Thompson-McFadden Commission estimate
50
40
30
20
10
0
Inman Mills
Republic
Saxon Mills
Arkwright
Whitney
Village
Seneca
Newry
Spartan Mills Pacolet Mills
Figure 3. Pellagra incidence rates per 1,000 individuals in cotton mill villages in South Carolina in 2 surveys. The Thompson-McFadden Commission’s case ascertainment (in 1912–1913) may have included some ariboflavinosis along with pellagra, whereas the Public Health Service’s case
ascertainment (in 1916) was more specific. The Public Health Service’s inclusion of villages with a broader range of pellagra rates, most notably
Newry, also allowed for greater comparison of the role of context. Thompson-McFadden Commission data were extracted from Tables 53, 54, 57,
and 58 in the report by Siler et al. (16); the Public Health Service data were extracted from Table I of the report by Goldberger et al. (28).
which all residents had some contact with pellagra. Second,
given that hospital staff never contracted pellagra from patients after close contact, observed clustering within households was more consistent with causation due to shared diet
than to shared exposure to an infectious agent (38). Unfortunately, the Thompson-McFadden Commission dismissed
Vedder’s ideas, arguing that the dietary hypothesis would
predict simultaneous development of pellagra within a
household, whereas they had observed lags of months or
years between incident cases (25). However, the Public
Health Service was aware of Vedder’s work (19, 33); it
may have been that Vedder’s critique of the Commission’s
survey motivated the Public Health Service to use a dietary
assessment strategy that accounted for portion size.
Reanalysis
The Public Health Service’s attention to assessing household diet and validating the reported cases of pellagra indicates grave concern that measurement error might blur the
nature of the association between diet and pellagra (19). To
understand the specific importance of measurement differences between the 2 studies, we simulated analyses the Public
Health Service might have performed using methods only
as accurate as those used by the Thompson-McFadden
Commission. Our hypothesis was that inaccurate case definition and dietary assessment were 2 of the main factors
preventing the Thompson-McFadden Commission from
identifying the true cause of pellagra.
METHODS
Data collection
We were unable to retrieve original individual-level data
from the Public Health Service study. Our search targeted
the Goldberger archive at the University of North Carolina
(C. Gray, University of North Carolina at Chapel Hill, personal communication, 2013), the Goldberger archive at Vanderbilt University (C. Ryland, Vanderbilt University Medical
Center, personal communication, 2013), and the Public
Health Service archive. We also made personal inquiries to
the Pearl S. Buck family archive (the Public Health Service’s
chief statistician, Edgar Sydenstricker, was the brother of
Nobel laureate Pearl S. Buck), the descendants of G.A.
Wheeler, and the Milbank Memorial Fund, where junior statistician Dorothy Wiehl, who likely compiled results into the
final published tables, later worked.
We therefore restricted our reanalysis to summary data extracted from papers authored by the Thompson-McFadden
Downloaded from http://aje.oxfordjournals.org/ at Pennsylvania State University on March 5, 2016
Pellagra Incidence per 1,000 Residents
60
Contrasting Historical Pellagra Investigations
A)
B)
0.15
0.10
0.05
0.10
0.05
0.00
0.00
<1.0
1.0–1.9
2.0–2.9
3.0–3.9
≥4.0
Household Meat Supply per Adult Male Unit, lbs
<1.0
1.0–1.9
2.0–2.9
3.0–3.9
≥4.0
Household Meat Supply per Adult Male Unit, lbs
Figure 4. The association between household-level pellagra incidence and A) estimate of meat supply (slope = −0.022) in the Public Health
Service study, South Carolina, April 1916–June 1916 and B) in 25 data sets simulated from the Public Health Service data with pellagra measured
at 98.6% specificity and with 25% of household meat supply misclassified (median slope = −0.010). Each thin line represents the slope estimate
from 1 simulation; the thick line represents the median estimate. 1 lb = 0.454 kg. Public Health Service data were extracted from Table IX of the report
by Goldberger et al. (19).
Commission and the Public Health Service investigators. We
simulated measurement error in meat assessment because
milk and meat were the only 2 niacin-containing foods
considered by both the Public Health Service and the
Thompson-McFadden Commission. Data on the relationship
of fresh meat supply to pellagra were extracted from Table IX
of the Public Health Service’s diet paper (19) and from Tables 49 and 50 of the Thompson-McFadden Commission’s
diet paper (16). Data on village-specific pellagra incidence
were extracted from Tables I and III of the Public Health Service’s sanitation paper (31) and from Tables 53, 54, 57, and
58 of Thompson-McFadden Commission’s diet paper (16).
Statistical analysis
Our analysis treated the Public Health Service data on meat
supply and pellagra as the truth and simulated misclassification of both pellagra diagnosis and meat supply. Because the
Commission’s canvassing strategy was as inclusive as that
used by the Public Health Service (14), we assumed a sensitivity of 100% for pellagra diagnosis. However, because the
Commission’s case definition allowed for patient or physician reports of skin lesions and made no mention of requiring
bilateral symmetry (14), the Commission likely included
some cases of ariboflavinosis in their pellagra group. Because
pellagra’s yearly mortality rates among whites were relatively
constant in the mid-1910s (39), we assumed that the total
prevalence of pellagra in the 4 villages assessed by both investigations was equal in 1912 and 1916 and estimated specificity for the Commission’s diagnosis by assuming that the
excess pellagra incidence observed by the Commission was
due to ariboflavinosis misdiagnosed as pellagra. Further, because the Commission’s assessment of meat consumption
was based on self-reports and was not strictly seasonal
(16), we simulated misclassification of meat consumption
ranging from 0% to 30% of individuals. We then compared
the relation of partially misclassified meat supply to partially
misclassified pellagra to assess how much weaker the trend of
Downloaded from http://aje.oxfordjournals.org/ at Pennsylvania State University on March 5, 2016
% of Households With Pellagra
0.15
% of Households With Pellagra
7
8 Mooney et al.
Proportion of Simulations Resulting
in a Suggestive Correlation
1.0
0.8
0.6
0
10
15
20
25
5
% of Households for Which Meat Supply Was Misclassified
30
Figure 5. Relation of proportion of meat supply misclassified to the proportion of simulations with a slope of 9.1 cases of pellagra per 1,000 subjects for each increase in meat category, an association the Public Health Service described as suggestive of an inverse correlation. All simulations
assume 100% sensitivity and 98.6% specificity of pellagra case ascertainment, estimated by comparing the Thompson-McFadden Commission’s
observed pellagra cases to those observed by the Public Health Service, which used a more rigorous case definition. Based on data collected in
South Carolina in 1916.
increasing pellagra with decreasing meat supply became. We
ran 1,000 simulations at each level of meat supply misclassification. We also plotted 25 of these slopes (at 25% misclassification) to illustrate the variation in slopes in these
simulations.
Both studies predate the use of modern statistics; investigators compared incidence rates visually by category to determine whether a significant relationship existed. To simulate
this visual trend test in 1,000 simulations, we considered a regression line with a slope indicating an average change of 9.1
cases/1,000 subjects between categories to indicate a noteworthy relationship. We picked 9.1/1,000 as the change the
Public Health Service considered “a suggestion of an inverse
correlation” (19, p. 691). All analyses were performed using
R for Windows, version 2.15.3 (R Foundation for Statistical
Computing, Vienna, Austria). The code for simulations can
be found in Web Appendix 2.
RESULTS
Application of the pellagra incidence rate observed by
the Public Health Service to the Thompson-McFadden Commission data in Arkwright, Newry, Saxon Mills, and Whitney
resulted in an estimated 30 excess cases of pellagra in each
village. Assuming those cases were misclassified resulted
in an estimated specificity of 98.6%.
Figure 4 displays the pellagra incidence rates by misclassified meat supply in the Public Health Service results and in
25 simulations, with the median slope highlighted. Simulations incorporating measurement error resulted in notably
weaker associations between meat supply and pellagra. Figure 5 displays the proportion of simulations with a slope of
9.1 cases of pellagra per 1,000 subjects for each increase in
meat category. At 20% of meat supply misclassified, half of
the slopes were below the level the Public Health Service
considered a “suggestive correlation.”
DISCUSSION
The Thompson-McFadden Commission and the Public
Health Service both investigated pellagra incidence in cotton
mill villages in South Carolina in the 1910s using similar
survey methods, yet they came to starkly different conclusions. We believe that 3 characteristics allowed the Public
Health Service to correctly determine the relation of pellagra
to diet.
First, a carefully planned approach to data collection and
measurement avoided misdiagnosis of ariboflavinosis as pellagra, prevented inaccuracy in dietary measurement, and
accounted for variation in household composition. Our simulations showed that small amounts of exposure and disease
misclassification sufficed to nullify the meat consumption–
pellagra association. Although we have no evidence as to
the rate of misclassification in the Thompson-McFadden
Commission’s data, this may explain why the Commission
found no association between pellagra and the consumption
of foods we now know to contain niacin. Furthermore, this
may explain why Edward Vedder, making roughly similar
claims to those Goldberger made later, was not widely
heeded. Second, the Public Health Service’s selection of
Downloaded from http://aje.oxfordjournals.org/ at Pennsylvania State University on March 5, 2016
0.4
Contrasting Historical Pellagra Investigations
CONCLUSIONS
The Public Health Service’s dedication to collecting accurate data and its insight into the contextual shaping of risk enabled the second cotton mill village investigation to succeed
where the Thompson-McFadden Commission’s did not. The
Public Health Service’s investigation is an excellent example
of 2 key features of effective epidemiology: careful data collection and formulation of analyses that differentiate between
competing hypotheses. We believe that Goldberger deserved
the Nobel Prize he might have won had he lived longer (43);
his approach to studying pellagra represents a model we
would do well to follow today.
ACKNOWLEDGMENTS
Author affiliations: Department of Epidemiology, Mailman School of Public Health, Columbia University, New
York, New York (Stephen J. Mooney, Justin Knox, Alfredo
Morabia); and Barry Commoner Center for the Biology
of Natural Systems, Queens College, City University of
New York, New York, New York (Alfredo Morabia).
This work was supported by the National Cancer Institute
(grant T32 CA09529 to S.J.M.), the National Library of
Medicine (grant 1G13LM010884-01A1 to A.M.), and the
Columbia University EPIC Fund.
We thank Chris Gray for checking the Goldberger archive
at the University of North Carolina for data and Dr. Daniel
Westreich for putting us in touch with Chris Gray. We thank
Dr. Andrew Rundle, Dr. Sandro Galea, Dr. Sharon Schwartz,
Dr. Jeffrey Haydu, and M. Katherine Mooney for their insightful comments on an earlier version of this work.
Conflict of interest: none declared.
REFERENCES
1. Kraut AM. Goldberger’s War: the Life and Work of a Public
Health Crusader. New York, NY: Farrar, Straus and Giroux; 2003.
2. Bollet AJ. Politics and pellagra: the epidemic of pellagra in the
U.S. in the early twentieth century. Yale J Biol Med. 1992;
65(3):211–221.
3. Rajakumar K. Pellagra in the United States: a historical
perspective. South Med J. 2000;93(3):272–277.
4. Marks HM. Epidemiologists explain pellagra: gender, race, and
political economy in the work of Edgar Sydenstricker. J Hist
Med Allied Sci. 2003;58(1):34–55.
5. Sambon LW. Remarks on the geographical distribution and
etiology of pellagra. Br Med J. 1905;2(2341):1272–1275.
6. Funk C. The etiology of the deficiency diseases. J State Med.
1912;20:341–368.
7. Silver DR. Corn and pellagra: a contribution to our knowledge
of their relation as probable cause and effect. JAMA. 1910;
LIV(6):452–453.
8. Randolph JH. Notes on pellagra and pellagrins. With report of
cases. Arch Intern Med. 1909;II(6):553–568.
9. Siler JF, Garrison PE, MacNeal WJ. Further studies of the
Thompson-McFadden Pellagra Commission—a summary of
the second progress report. J Am Med Assoc. 1914;LXIII(13):
1090–1093.
10. Lavinder CH. The salient epidemiological features of pellagra.
Public Health Rep. 1911;26(39):1459–1468.
11. Grimm RM. Pellagra: a report on an epidemiologic study.
Public Health Rep. 1912;27(8):255–264.
12. Grimm RM. Pellagra: A Report on its Epidemiology.
Washington, DC: Government Printing Office; 1913.
13. Siler JF, Garrison PE, MacNeal WJ. Pellagra: a summary of the
first progress report of the Thompson-McFadden Pellagra
Commission. JAMA. 1914;LXII(1):8–12.
14. Siler JF, Garrison PE. An intensive study of the epidemiology of
pellagra. Report of progress. Am J Med Sci. 1913;146(2):238–277.
15. Siler JF, Garrison PE. An intensive study of the epidemiology of
pellagra. Report of progress. Am J Med Sci. 1913;146(1):42–66.
16. Siler JF, Garrison PE, MacNeal WJ. A statistical study of the
relation of pellagra to use of certain foods and to location of
domicile in six selected industrial communities. Arch Intern
Med (Chic). 1914;XIV(3):293–373.
Downloaded from http://aje.oxfordjournals.org/ at Pennsylvania State University on March 5, 2016
villages with a wider range of diet and pellagra incidence
augmented the advantage of superior accuracy in data by presenting stronger between-village contrasts. Third, flexibility
of analytical thought to understand that multiple levels of influence could be examined allowed the Public Health Service
to identify the proper contrasts within the data.
A contemporary perspective also allows us to consider the
Public Health Service investigation as an early example of
multilevel epidemiologic analysis (40). Diet at the individual
level is notoriously difficult to assess; instead, the Public
Health Service leveraged contrasts in village-level food availability and in household composition to reconcile known
patterns in disease incidence with a dietary hypothesis. By
comparison, the Thompson-McFadden Commission collected data at multiple levels but did not leverage the multilevel
potential of the data, instead using lack of between-village
variation in geographical clustering of cases as an argument
for an infectious hypothesis (16).
In hindsight, and using modern terminology, we can say
that, for the first time that we are aware of, the Public Health
Service team led by Goldberger combined 2 main features:
1) containment of misclassification using carefully collected
data on exposure and outcome and 2) a multilevel analysis
including individual level and several group (households
and villages) levels, with the understanding that the village
was what we would now term an effect modifier of the relation of poverty and pellagra (mediated by diet). Had modern
statistical tools been available, the Public Health Service
might have considered both the “neighborhood” (i.e., village)
association with the individual-level outcome and individuallevel risk factors within the same analysis. If the original
Public Health Service data are ever recovered, it would be
of interest to know whether modern multilevel analytical
tools would provide any insight beyond the published
analyses.
Unfortunately, although the cotton mill village study
fully convinced Goldberger of pellagra’s dietary etiology,
his critics remained skeptical (1). Deciding that further
epidemiologic studies would not convince them, Goldberger instead focused on finding the agent that prevented
pellagra (2). Nearly 20 years after the cotton mill village
studies, Elvehjem and Koehn (41) identified nicotinic acid
(later renamed niacin) as the vitamin whose deficiency
caused pellagra. Pellagra was fully eradicated when the niacin supplementation of flour became widespread in the
early 1940s (42).
9
10 Mooney et al.
31.
32.
33.
34.
35.
36.
37.
38.
39.
40.
41.
42.
43.
44.
45.
of the disease and its relation to race, age, sex and occupation.
Arch Intern Med. 1915;XV(1):98–120.
Goldberger J, Wheeler GA, Sydenstricker E, et al. A study of
the relation of factors of a sanitary character to pellagra
incidence in seven cotton-mill villages of South Carolina in
1916. Public Health Rep. 1920;35(29):1701–1714.
Jobling JW, Petersen W. The epidemiology of pellagra in
Nashville, Tennessee. J Infect Dis. 1916;18(5):501–567.
Goldberger J, Wheeler GA, Sydenstricker E. A study of the
relation of family income and other economic factors to
pellagra incidence in seven cotton-mill villages of South
Carolina in 1916. Public Health Rep. 1920;35(46):2673–2714.
Sydenstricker E. The Prevalence of Pellagra: Its Possible
Relation to the Rise in the Cost of Food. Washington, DC:
Government Printing Office; 1916.
Sydenstricker VP. The history of pellagra, its recognition as a
disorder of nutrition and its conquest. Am J Clin Nutr. 1958;
6(4):409–414.
Carpenter KJ, Lewin WJ. A reexamination of the composition
of diets associated with pellagra. J Nutr. 1985;115(5):
543–552.
Vedder EB. Some further remarks on beri-beri. Am J Trop Dis
Prev Med. 1914;1:826–847.
Vedder EB. Dietary deficiency as the etiological factor in
pellagra. Arch Intern Med (Chic). 1916;XVIII(2):137–172.
Marks HM. Epidemiologists explain pellagra: gender, race, and
political economy in the work of Edgar Sydenstricker. J Hist
Med Allied Sci. 2003;58(1):34–55.
Diez-Roux AV. Bringing context back into epidemiology:
variables and fallacies in multilevel analysis. Am J Public
Health. 1998;88(2):216–222.
Elvehjem CA, Madden RJ, Strong FM, et al. Relation of
nicotinic acid and nicotinic acid amide to canine black tongue.
J Am Chem Soc. 1937;59(9):1767–1768.
Park YK, Sempos CT, Barton CN, et al. Effectiveness of food
fortification in the United States: the case of pellagra. Am J
Public Health. 2000;90(5):727–738.
Carpenter KJ. The Nobel Prize and the discovery of vitamins.
http://www.nobelprize.org/nobel_prizes/themes/medicine/
carpenter/index.html. Published June 22, 2004. Accessed
October 22, 2013.
Stannus HS. Pellagra in Nyasaland. Trans R Soc Trop Med Hyg.
1913;7(1):32–56.
MacNeal WJ. The alleged production of pellagra by an
unbalanced diet. JAMA. 1916;LXVI(13):975–977.
Downloaded from http://aje.oxfordjournals.org/ at Pennsylvania State University on March 5, 2016
17. Siler JF, Garrison PE, MacNeal WJ. The relation of recurrent
attacks of pellagra to race, sex and age of the patient and to
treatment of the disease. Arch Intern Med (Chic). 1916;
XVIII(5):652–691.
18. Goldberger J. Public Health Reports, June 26, 1914. The
etiology of pellagra. The significance of certain
epidemiological observations with respect thereto. Public
Health Rep. 1914;29(26):1683–1686.
19. Goldberger J, Wheeler GA, Sydenstricker E. A study of the
relation of diet to pellagra incidence in seven textile-mill
communities of South Carolina in 1916. Public Health Rep.
1920;35(12):648–713.
20. Goldberger J, Waring CH, Willets DG, et al. The Treatment and
Prevention of Pellagra. Washington, DC: Government Printing
Office; 1914.
21. Anderson JF. An attempt to infect the rhesus monkey with
blood and spinal fluid from pellagrins. Public Health Rep.
1911;26(26):1003–1004.
22. Goldberger J, Waring CH, Willets DG. The prevention of
pellagra: a test of diet among institutional inmates. Public
Health Rep. 1915;30(43):3117–3131.
23. Goldberger J, Wheeler GA. Experimental pellagra in the human
subject brought about by a restricted diet. Public Health Rep.
1915;30(46):3336–3339.
24. Goldberger J. The transmissibility of pellagra: experimental
attempts at transmission to the human subject. Public Health
Rep. 1916;31(46):3159–3173.
25. Siler JF, Garrison PE, MacNeal WJ. Relation of pellagra to
location of domicile in Spartan Mills, S. C., and the adjacent
district. Arch Intern Med (Chic). 1917;XX(2):198–315.
26. Weinberg W, Von Gruber M. Die Kinder der Tuberkulösen.
Leipzig, Germany: S. Hirzel; 1913.
27. Lane-Claypon JE. Report to the Local Government Board Upon
the Available Data in Regard to the Value of Boiled Milk as A
Food for Infants and Young Animals. London, United
Kingdom: Her Majesty’s Stationary Office; 1912.
28. Goldberger J, Wheeler GA, Sydenstricker E. Pellagra incidence
in relation to sex, age, season, occupation, and “disabling
sickness” in seven cotton-mill villages of South Carolina during
1916. Public Health Rep. 1920;35(28):1650–1664.
29. Siler JF, Garrison PE, MacNeal WJ. The relation of methods of
disposal of sewage to the spread of pellagra. Arch Intern Med
(Chic). 1914;XIV(4):453–474.
30. Siler JF, Garrison PE, MacNeal WJ. Statistics of pellagra in
Spartanburg Country, S. C., including geographical distribution