Curves were fit using second-order polynomial regression. Data are from
all Ohio deaths 1989-2000 according to death certificate–specified cause
Young DC, Hade EM. Holidays, Birthdays, and Postponement of Cancer Death. JAMA. 2004;292(24):3012-3016. doi:10.1001/jama.292.24.3012
Author Affiliations: Comprehensive Cancer Center
and Center for Biostatistics, The Ohio State University, Columbus.
Context Articles in the medical literature and lay press have supported a belief
that individuals, including those dying of cancer, can temporarily postpone
their death to survive a major holiday or other significant event, but results
and effects have been variable.
Objective To determine whether, for the patient dying of cancer, a “death
takes a holiday” effect showing a reduction in deaths in the week before
a significant event was associated with Christmas, the US holiday of Thanksgiving,
or the date of the individual’s birthday.
Design, Setting, and Subjects Analysis of death certificate data for all 1 269 474 persons
dying in Ohio from 1989-2000, including 309 221 persons dying with cancer
noted as the leading cause of death.
Main Outcome Measure We measured the total number of cancer deaths in the 2 weeks centered
on the event of interest and the proportion of these deaths that occurred
in the week before the event to determine whether this proportion was significantly
different from 0.5 by using an exact binomial test.
Results The proportion of persons dying of cancer in the week before Christmas,
Thanksgiving, and the individual’s birthday was not significantly different
from the proportion dying in the week after the event (P = .52, .26, and .06, respectively). However, among black
individuals there was an increase in cancer deaths in the week before Thanksgiving
(P = .01), whereas women showed an increase
in cancer deaths in the week before their birthday (P = .05).
There was no statistically significant excess of deaths in the postevent week
in any subgroup.
Conclusion We found no evidence, in contrast to previous studies, that cancer patients
are able to postpone their deaths to survive significant religious, social,
or personal events.
Health care workers and others involved with patients dying of cancer
commonly recall those who apparently held on to life and defied the odds by
surviving a major holiday or significant event, only to die immediately thereafter.1,2 An apparent dip or peak mortality pattern
associated with significant religious and social events has been reported
in a series of recently reviewed studies.1 One
study examined mortality related to the Jewish holiday of Passover in Californians
with Jewish surnames and found that overall mortality was 8% higher in the
week after Passover compared with the week before, an effect that was observed
separately for death caused by cerebrovascular, cardiac, and neoplastic diseases;
the effect size was greatest in white men but nonexistent in women.3 Similarly, a study in New Haven, Conn, found that
elderly Jewish men were able to postpone death in the 30 days preceding Passover
but not Rosh Hashanah or Yom Kippur and that for all elderly Christians there
was a 24% total dip-peak difference around Christmas but not Easter.4 Another study reported a similar dip on the Saturday
Sabbath compared with individual weekdays in all Jewish men and women dying
in Israel but no dip or peak effect before and after 9 Jewish holy days, including
All-cause mortality in 1288 Chinese American women was found to dip
35% in the week before the Harvest Moon Festival, with a corresponding peak
in the following week, with no effect in a non-Chinese control group.6 However, a reanalysis of these data, which included
an additional 2437 Asian American deaths, found no evidence that elderly Asian
women were able to prolong their lives until after the festival.7
An individual’s birthday has also been associated with patterns
of mortality in a study of all-cause deaths in California.8 Birthdays
apparently served as a “lifeline” for women, who were more likely
to survive their birthday and die in the following week; in contrast, men’s
birthdays functioned as a “deadline,” showing a mortality peak
before the event. A 19% dip in the month before the birthday for famous Americans
and Englishmen has also been reported.9
We used a large database to examine whether cancer deaths would demonstrate
a dip or peak phenomenon around 3 events with potential religious, secular,
and personal importance to the individual: Christmas, the US holiday of Thanksgiving,
and the person’s birthday. We analyzed cancer deaths because the concept
of intentional death postponement appears to be most tenable with a chronic
disease. Previous studies have reported a significant effect in this subgroup,
and there is no superimposed seasonal pattern.
We examined data from death certificates representing all deaths in
Ohio from 1989-2000. These data were obtained from the Ohio Department of
Health as the Ohio Mortality Public Use Statistical File, as submitted to
the National Center for Health Statistics (NCHS). The leading cause of death
was encoded by using the NCHS 113 coding system. The file includes individual
data for 61 variables, including date of birth, date of death, sex, race,
ethnicity, and leading cause of death. The race and ethnicity information
on all cases was that recorded by the physician on the individual’s
death certificate and subsequently entered into state and national databases.
We included such information because previous studies reported a death postponement
effect in racial subgroups.
Because no individuals were living, under 45 CFR 46 §46.102(f)
institutional review board approval was not required.
Our study included all persons with NCHS 113 codes of 18-40 (corresponding
to International Classification of Diseases, Ninth Revision codes of 140-208, malignant neoplasm) as the leading cause of death.
For comparison, noncancer mortality includes the remaining NCHS 113 codes
1 to 17 and 41 to 113 for all other causes of death. These causes include
infectious diseases (codes 1-17), cardiovascular diseases (codes 49-60), cerebrovascular
diseases (code 61), respiratory infection and diseases (codes 67-76), and
accidents and unintentional injuries (codes 96-104), among others. Individuals
born or dying on February 29 were excluded from analysis.
We chose 3 events as potentially meaningful or symbolic: Christmas,
the US holiday of Thanksgiving, and the individual’s birthday. Christmas,
celebrated on December 25, and Thanksgiving, observed on the fourth Thursday
of November, are major religious and secular holidays in the United States,
respectively. To determine whether there was a decrease (“dip”)
in mortality before a significant event, followed by a subsequent increase
(“peak”), we tabulated the number of deaths per week for the 2
weeks centered on the specific event. The week before the event was defined
as the 7 days leading up to and including the event, consistent with previous
studies. The magnitude of the dip-peak effect is defined as the sum of the
percentage of reduction in the number of deaths in the week before the event
plus the percentage of increase in the number of deaths in the subsequent
week, both compared with the mean number of deaths per week for the 2-week
We hypothesized that the significant event has no effect on mortality
and that the proportion of individuals dying of cancer in the week before
a significant event will be equal to half of the total number of deaths in
the 2-week interval centered on the event. We compared the proportion of people
dying in the week before each event to our expected proportion of 0.5 by using
an exact binomial test with statistical significance defined as P<.05. With approximately 12 000 deaths in the 2-week period
centered on each event, our sample had 90% power to detect a 1.5% dip in the
observed proportion of individuals dying in the week before the event, or
a combined 3.0% dip-peak effect, using a.05 2-sided level of significance.
The American Religious Identification Survey10 shows
that approximately 80% of Americans identified themselves as Christian during
the period 1990-2001. Excluding individuals for whom Christmas likely held
little or no significance, our remaining sample could detect a dip-peak effect
of at least 3.4% for this holiday. As a secondary post hoc analysis, the number
of deaths on the day of the event was compared with the mean number of deaths
per day per year for the 2-week period centered on the event by using analysis
of variance. STATA version 8 (StataCorp LP, College Station, Tex) was used
for all statistical analyses.
In Ohio from 1989-2000, 1 269 474 persons died, including
309 221 persons dying with cancer as the leading cause of death (Table 1). The distribution of the mean number
of deaths caused by cancer and noncancer causes by week of the year (Figure) shows the usual winter increase in noncancer
mortality, with a peak in the last week of December and first 2 weeks of January.11- 13 Cancer mortality
shows little seasonal variation.
For Christmas, Thanksgiving, or the individual’s birthday, during
the 12-year period there was no significant difference in the proportion of
patients dying in the week after the event compared with the proportion dying
in the week before the event (Table 2).
No significant effects were observed during the Christmas period according
to sex, race, or age (<70 years vs ≥70 years). However, black Ohioans
were more likely to die of cancer during the week before Thanksgiving than
during the week after the holiday, unlike white persons, who showed similar
proportions dying each of the 2 weeks. Although overall birthday data showed
no effect, women dying of cancer were more likely to die during the week before
their birthday compared with the following week. Men showed no significant
differences. In no subgroup was a statistically significant decrease of deaths
observed in the week before the event.
In the secondary analyses, there were no differences in the numbers
of deaths on either Christmas (P = .83)
or Thanksgiving (P = .08) compared with
the number of deaths on each day in the 2-week period centered on the holiday.
There was an increase in the total number of deaths on a person’s birthday
(P = .008), consonant with the
P value of .06 for the dip-peak effect for the period.
For cancer deaths, our study failed to substantiate previous reports
that dying persons can intentionally postpone death to survive personally
significant events. The size of our sample makes it unlikely that we failed
to detect an important dip or peak effect because previous studies have demonstrated
large dip-peak differences, up to 70%.4,6 Although
we cannot eliminate the possibility that a small number of dying cancer patients
have the ability to control the timing of their death, the proportion would
have to be much smaller than that previously reported. In contrast to studies
that used select, small samples of individuals representing specific racial
or religious groups, our study of a large racially and ethnically mixed sample
offers generalizability to the even larger population of more than 500 000
people dying each year of cancer in the United States.
We focused on death caused by cancer rather than all-cause mortality
or mortality due to unintentional natural causes for a number of reasons.
A concept or potential mechanism of intentional death postponement is most
tenable in persons dying of a chronic disease during an extended time. Previous
studies have included analyses indicating an ability to delay death in subgroups
of persons noted as dying of cancer.3,6 Although
the cause-of-death coding system we used to select our sample easily identifies
persons dying of cancer, differentiating between acute and chronic forms of
noncancer causes of death is more difficult. Because relatively few patients
dying of cancer are maintained on life support, by studying this group we
have minimized the impact of family decisions to maintain or discontinue life
support on the timing of death. Restricting our study to cancer deaths facilitates
analysis because of the lack of confounding seasonal variation in mortality.
Finally, without evidence of geographic differences in seasonal patterns of
cancer mortality, our study of cancer deaths in Ohio should be generalizable
to the larger population of persons dying of cancer.
The tendency for blacks to die more frequently in the week before Thanksgiving
is surprising. Similarly, the increase in cancer deaths for women the week
before their birthday appears to contradict previous studies showing that
women survived their birthday more frequently than men.8 We
believe that the best explanation for these observations is that they represent
artifacts of multiple significance testing. The increase in the number of
deaths on the actual birthday and not on the other holidays is from a post
hoc analysis, raising the likelihood that this represents an artifact. Without
an analysis of the actual time of death and its relationship to any celebration,
it is difficult to attach particular significance to this observation.
There are a number of factors that could have contributed to the differences
from previous studies. First, the positive results in many of the articles
have represented multiple comparisons in small sample sizes, which could introduce
statistical error. One reported positive effect represents a deviation of
71 deaths from the expected total of 621 deaths,3 another
represents a deviation of 36 deaths from an expected total of 103,6 and a third study performed 109 tests of significance
in samples of 354 and 58 subjects.4 Two of
the studies related to Jewish holidays3,9 have
been criticized in that decedents were not necessarily Jewish and that the
significant events, Yom Kippur and Passover, were not both analyzed in each
The unspecified psychological mechanisms proposed in many articles associating
death with symbolic occasions are not supported by direct evidence that such
processes exist.1 Similarly, there has not
been convincing evidence of coping mechanisms or optimistic attitude affecting
survival in cancer.15,16 Plausible
nonpsychosomatic mechanisms exist for the ability of a patient to temporarily
delay or hasten his or her impending death. Patients may choose to either
forgo or accept good supportive care, including nutrition, hydration, and
the use of antibiotics and palliative medications. Nevertheless, regardless
of these choices, the time scale on which these interventions may exert their
effect is unlikely to permit a patient to select a specific date before or
on which to avoid death. Common perceptions of this effect (among physicians,
as well as the public) may represent an example of the availability heuristic,17 a cognitive bias in which we recall more easily deaths
that occurred immediately after important events because they were so striking,
compared with the greater number that occurred at random times, and thus mentally
assign them an exaggerated prevalence.
Several potential limitations of our study need to be addressed. Although
we have chosen 3 specific and easily denoted events of general impact, other
personally salient events such as weddings, anniversaries, and graduations
could present greater emotional impact for the individual with cancer and
thus have a greater effect on the timing of death. Unfortunately, no practical
means exist to ascertain the impact of these events in large population-based
registries for the hundreds of thousands of persons dying of cancer. We are
additionally unable to determine from our database the proportion of individuals
who were comatose before death and therefore were incapable of knowing the
approach of a holiday. Although children may have no association with a landmark
date, only 881 cases (0.29%) of our sample were aged 12 years or younger at
their death from cancer, so their impact on these results would be minimal.
The potential influence of the location of an individual’s death
on the timing of death is difficult to evaluate. Although our database includes
information on whether the individual died in a hospital, nursing home, or
at home, we have no information on the length of time a person was in that
setting before death. Although observing a holiday or birthday at home and
surrounded by family and friends might have a more pronounced effect on death
postponement until after the occasion, home care with increased stress on
family caregivers could have the opposite effect.
The proximity of the Christmas and New Year’s holidays could confound
results in that a post-Christmas peak in deaths may be obscured by a pre–New
Year’s dip in deaths. Were this to have occurred, the “double
dip” would have decreased the total number of deaths throughout the
2-week Christmas period. However, the Christmas period had the largest number
of deaths of the 3 events (Table 2).
Additionally, we saw no dip or peak effect with either Thanksgiving or birthdays.
Our data are subject to error from the misclassification of cause of
death on death certificates. An 18% underestimation of cancer deaths because
of other causes of death being noted on the death certificates has been reported.18 The impact on the power of this study would have
been at most a 0.3% reduction. There is no reason to believe that an increased
dip or peak effect would be observed in potential subjects lost because of
misclassification. On the other hand, deaths due to noncancer causes may be
erroneously classified as caused by cancer. Given that deaths due to noncancer
causes showed a dramatic seasonal pattern, if a large number of noncancer
deaths were included in our sample, we would have expected a seasonal mortality
pattern to emerge.
In conclusion, analysis of thousands of cancer deaths shows no pattern
to support the concept that “death takes a holiday.” We find no
evidence that cancer patients are able to postpone their death to survive
Christmas, Thanksgiving, or their own birthdays.
Corresponding Author: Donn C. Young, PhD,
The Ohio State University Comprehensive Cancer Center, M410 Starling Loving
Hall, 320 W Tenth Ave, Columbus, OH 43210 (firstname.lastname@example.org).
Author Contributions: Dr Young had full access
to all of the data in the study and takes responsibility for the integrity
of the data and the accuracy of the data analysis.
Study concept and design: Young.
Acquisition of data: Young.
Analysis and interpretation of data: Young,
Drafting of the manuscript: Young.
Critical revision of the manuscript for important
intellectual content: Young, Hade.
Statistical analysis: Young, Hade.
Obtained funding: Young.
Administrative, technical, or material support:
Study supervision: Young.
Funding/Support: Dr Young and Ms Hade were
supported by a cancer center support grant (P30CA16058, Dr Michael Caligiuri,
principal investigator) from the National Institutes of Health, National Cancer
Institute, to The Ohio State University Comprehensive Cancer Center.
Role of the Sponsor: The funding organization
did not participate in the design and conduct of the study; the collection,
analysis, and interpretation of the data; or the preparation, review, or approval
of the manuscript.