February 13, 2014DOI: 10.1056/NEJMoa1311485
Pregnancy significantly increases the
risk of thrombosis. This heightened thrombotic risk rises further during the
postpartum period, which is conventionally defined as the 6 weeks after
delivery.1As
compared with the nonpregnant state, the 6-week postpartum period is associated
with increases by a factor of 3 to 9 in the risk of stroke, by a factor of 3 to
6 in the risk of myocardial infarction, and by a factor of 9 to 22 in the risk
of venous thromboembolism.2-8 It is unknown whether these risks remain increased after the
conventionally defined 6-week postpartum period. Guidelines for the treatment
of thrombotic disorders during pregnancy advise the discontinuation of
prophylactic therapy at 6 weeks after delivery in women at high risk for venous
thromboembolism.1However,
previous studies and isolated case reports have suggested that an increased
thrombotic risk may persist beyond 6 weeks after delivery.5,8-10 Therefore, more data are needed to
rigorously assess the risk after the 6-week postpartum period. We designed this
study to assess the duration of an increased postpartum thrombotic risk in a
large population-based cohort of women.
METHODS
Study
Design
We performed a retrospective
crossover-cohort study (a study design in which each patient serves as his or
her own control), using administrative claims data on all discharges from
nonfederal emergency departments and acute care hospitals in California. We
compared each patient's likelihood of a first thrombotic event during
sequential 6-week periods after delivery with the likelihood of an event during
the corresponding 6-week period 1 year later. Since exposure to pregnancy
varies discretely over time, this design allowed each patient to serve as her
own control, thereby minimizing unmeasured confounding.11,12 California was chosen because it is a
large and demographically heterogeneous state13 with administrative data that allow
tracking of individual patients across visits over numerous years,14 thereby providing sufficient statistical power to detect
associations among conditions with low absolute event rates. Analysts at each
facility used detailed reporting and formatting specifications and automated
online-reporting software to provide uniform data on all discharges to the
California Office of Statewide Health Planning and Development.15 After a multistep quality-assurance
process to flag invalid or inconsistent entries, these data were provided in a
deidentified format to the Healthcare Cost and Utilization Project.14The
institutional review boards at Weill Cornell Medical College and Columbia
University Medical Center certified that this analysis of publicly available,
deidentified data was exempt from review and from the need for informed
consent. All authors take responsibility for the integrity of the data and
analyses.
Study
Patients
We identified all women who had been
hospitalized for labor and delivery, using standard codes from the International Classification
of Diseases, 9th Revision, Clinical Modification (ICD-9-CM) for vaginal delivery (72, 73, 75, V27, or
650–659) and cesarean delivery (74).16 To
maximize longitudinal follow-up, we excluded non-California residents. We
included patients 12 years of age or older, given the infrequency of births
among patients younger than 12 years of age (<0.1% of all births17).
Post hoc sensitivity analyses that included patients regardless of age or
included only patients 18 years of age or older did not substantially alter our
findings.
For women with multiple labor-related
hospitalizations during a single 40-week period, we excluded cases of false
labor by identifying delivery as the latest hospitalization during that time.
Since women who have had a thrombotic event may be less likely to subsequently
become pregnant, we included only the first pregnancy captured in our database
for each patient. To focus on incident outcomes, we excluded patients who had
had any thrombosis diagnoses before their first recorded delivery (see the
Methods section in the Supplementary Appendix,
available with the full text of this article at NEJM.org, for definitions).
To compare thrombotic risk during the
postpartum period with the risk during nonpregnant periods, we excluded
patients with a second delivery during the follow-up period. We included
hospitalizations for labor starting on January 1, 2005, when patient-specific
longitudinal tracking identifiers were introduced in these databases.14 Data were available through December 31, 2011,14 so to accommodate analyses of the 24 weeks after delivery
and the same 24-week period 1 year later, we included patients with a
hospitalization for a first labor through June 30, 2010.
Study
Outcomes and Measurements
The primary outcome was a composite of
ischemic stroke, acute myocardial infarction, or venous thromboembolism. We
identified these outcomes using validated diagnosis-code algorithms that were
previously shown to have a positive predictive value of 90% or more (see the Supplementary Appendix).18-20 To maximize accuracy, we limited our case ascertainment to
stroke and myocardial infarction resulting in hospitalization but included
discharges from the emergency department as well as hospitalizations for venous
thromboembolism, since this condition is now often managed in the outpatient setting.21 To focus on incident outcomes and avoid bias from the
effects of antithrombotic therapy prescribed after the initial thrombotic
event, we included a maximum of one thrombosis diagnosis for each patient;
however, in a sensitivity analysis, we also included thrombosis diagnoses
subsequent to the index event.
In addition to the primary composite
outcome, we separately assessed arterial events (stroke or myocardial
infarction) as compared with venous thromboembolism. The definition of venous
thromboembolism in our primary analysis did not include cerebral venous
thrombosis because that condition lacks rigorously validated ICD-9-CM diagnosis
codes, but it was included among secondary outcomes, which consisted of the
primary outcome plus a broader set of other thrombosis diagnoses (see the Supplementary Appendix for
definitions).
We performed subgroup analyses
stratified according to thrombotic risk, using ICD-9-CM codes to identify
consistently reported risk factors for postpartum thrombosis: maternal age of
more than 35 years, primary hypercoagulable state, eclampsia or preeclampsia,
smoking, and cesarean delivery (see the Supplementary Appendix for
definitions).1,3,22,23
Statistical
Analysis
For each patient, we compared the
likelihood of a first-ever recorded thrombosis during postpartum days 0 through
41 versus the same period exactly 1 year later. We repeated this
crossover-cohort analysis for postpartum days 42 through 83, 84 through 125,
and 126 through 167. We used conditional logistic regression to calculate odds
ratios for each interval because each patient was matched to her own crossover
period 1 year later.11 Our a priori hypothesis was that the risk would
progressively decrease across sequential 6-week periods but remain
significantly elevated at least through the period of 7 to 12 weeks after
delivery. To help ensure that visits that were related to venous
thromboembolism did not represent the sequelae of previous outpatient
diagnoses, we performed a sensitivity analysis that excluded diagnoses of
venous thromboembolism with a concomitant bleeding-related diagnosis,24 since the event may have represented a complication of
preexisting anticoagulant therapy.
To assess the sensitivity of our
results to our baseline model structure, we inverted the model and performed a
case-crossover analysis. We identified all women who were 12 years of age or
older in whom the primary outcome had been diagnosed from July 1, 2006, to
December 31, 2011. We compared the likelihood of a first recorded labor and
delivery during days 0 through 41 before the thrombotic event versus the same
6-week period exactly 1 year earlier. We repeated this case-crossover analysis
for postpartum days 42 to 83, 84 to 125, and 126 to 167 before the thrombotic
event. In a sensitivity analysis, we included only cases that occurred beyond 1
year 24 weeks after a first documented delivery. This nested design ensured
that all patients were alive and under observation throughout the entire study
period, while also allowing us to assess the effects of the inclusion of
pregnancies subsequent to the first.
We performed a separate post hoc
case–control analysis to confirm whether any heightened risk of postpartum
thrombosis was associated with labor and delivery specifically, rather than
with hospitalization in general. We defined cases and controls on the basis of
the presence or absence of the primary outcome. The exposure variable was a
preceding hospitalization for delivery versus for any other diagnosis. To
account for potential confounders, these analyses were adjusted for age, race,
insurance type, the presence or absence of a primary hypercoagulable state,
smoking, and the Elixhauser comorbidity index.25
RESULTS
Study
Population
We identified 1,687,930 California
residents with a first recorded hospitalization for labor and delivery between
January 1, 2005, and June 30, 2010. This number was within 6% of the expected
number on the basis of birth certificates issued during that time.17 In
the 1 year 24 weeks after delivery, 1015 women had a thrombotic event (248
cases of stroke, 47 cases of myocardial infarction, and 720 cases of venous
thromboembolism). As compared with patients without postpartum thrombosis,
those with postpartum thrombotic events were older, were more likely to be
white or black than Hispanic or Asian, were less often privately insured, and
were more likely to have risk factors for thrombosis (Table 1)
TABLE 1: Baseline
Characteristics of the Patients, According to the Presence or Absence of a
Postpartum Thrombotic Event.).
Risk
of Thrombotic Events
Significantly more thrombotic events
occurred within 6 weeks after delivery than during the same period 1 year later
(411 events, or 24.4 events per 100,000 deliveries, vs. 38 events, or 2.3
events per 100,000 deliveries), corresponding to an absolute risk difference of
22.1 (95% confidence interval [CI], 19.6 to 24.6) per 100,000 deliveries and an
odds ratio of 10.8 (95% CI, 7.8 to 15.1). In the period of 7 to 12 weeks after
delivery, there was a modest but still significant increase in the number of
thrombotic events, as compared with the same period 1 year later (95 events, or
5.6 events per 100,000 deliveries, vs. 44 events, or 2.6 events per 100,000
deliveries), corresponding to an absolute risk difference of 3.0 (95% CI, 1.6
to 4.5) per 100,000 deliveries and an odds ratio of 2.2 (95% CI, 1.5 to 3.1).
The risk was no longer significantly
elevated after 12 weeks, with an odds ratio of 1.4 (95% CI, 0.9 to 2.1) for the
period of 13 to 18 weeks after delivery and an odds ratio of 1.0 (95% CI, 0.7
to 1.4) for the period of 19 to 24 weeks after delivery (Table 2)
TABLE 2: Number
and Rate of Postpartum Thrombotic Events during Sequential 6-Week Intervals
after Labor and Delivery.). In post hoc exploratory analyses,
the thrombotic risk was increased during the period of 13 to 15 weeks after
delivery (odds ratio, 2.0; 95% CI, 1.1 to 3.6) but was no longer elevated in
the period of 16 to 18 weeks (odds ratio, 1.0; 95% CI, 0.6 to 1.8) (Figure 1)
FIGURE 1Risk
of a Thrombotic Event, According to the Interval after Delivery.,
and Table S1 in the Supplementary Appendix).
The risk of thrombosis during the
period of 7 to 12 weeks after delivery appeared to be similarly elevated for
arterial events (odds ratio, 2.1; 95% CI, 1.0 to 4.3) and venous events (odds
ratio, 2.2; 95% CI, 1.4 to 3.3), although the absolute risk difference was
especially low for arterial events. We found a similar temporal pattern of
thrombotic risk in the secondary analysis, which included a broader set of
thrombosis diagnoses, including cerebral venous thrombosis (Table 2).
The period during which thrombotic risk
was significantly increased was also materially unchanged in sensitivity
analyses that excluded diagnoses of venous thromboembolism with accompanying
bleeding codes or that included thrombosis diagnoses subsequent to the first
recorded event. Except for a significantly higher risk within 6 weeks after
delivery among women who had undergone cesarean section than among those who
had undergone vaginal delivery, we found no significant variation in thrombotic
risk over the different time periods across subgroups with or without thrombotic
risk factors (Tables S2 and S3 in the Supplementary Appendix).
Case-Crossover
and Case–Control Analyses
In a case-crossover analysis of the
likelihood of labor and delivery before a first thrombotic event versus the
same periods 1 year earlier, we found that the odds of a first delivery were
markedly elevated in the period of 0 to 6 weeks before a thrombotic event (odds
ratio, 9.8; 95% CI, 7.0 to 13.9), significantly elevated in the period of 7 to
12 weeks before a thrombotic event (odds ratio, 2.2; 95% CI, 1.5 to 3.2), and
not significantly different in the periods of 13 to 18 weeks or 19 to 24 weeks
before a thrombotic event (Table 3)
TABLE 3Number
and Rate of Deliveries during Sequential 6-Week Intervals Preceding a
Thrombotic Event (Case-Crossover Analysis).). This pattern
was essentially unchanged in a nested analysis that included only patients who
were known to be alive and under observation for the entire 1 year 24 weeks
before the thrombotic event. In a separate case–control analysis, women with a
thrombotic event were more likely to have been hospitalized for labor and
delivery within the previous 7 to 12 weeks than to have been hospitalized for
another diagnosis (odds ratio, 1.9; 95% CI, 1.4 to 2.5) (data not shown).
DISCUSSION
Using administrative claims data from a
large state population, we found that the risk of a thrombotic event remained
elevated beyond the 6-week postpartum period, as compared with a similar time
period 1 year later, although absolute risk increases were small after 6 weeks.
As compared with the absolute increase in risk during the period within 6 weeks
after delivery (22.1 cases per 100,000 deliveries), the absolute increase
during the postpartum period of 7 to 12 weeks was much smaller (3.0 cases per
100,000 deliveries). During the latter period, odds ratios for thrombosis were
similar for women with recognized risk factors for thrombosis and those without
those risk factors, so the increased relative risk would be expected to be
especially important among high-risk patients (e.g., those with an inherited
primary hypercoagulable state or previous thrombosis).
To our knowledge, previous studies have
not reliably determined the relative risk of thrombosis beyond 6 weeks after
delivery. A population-based analysis of pregnancy-related venous
thromboembolism over several decades included events up to 3 months after
delivery, but only two cases were captured beyond 6 weeks, and relative risks
for this period were not reported.5 In
a population-based study of venous thromboembolism after in vitro
fertilization, thrombosis rates between 7 weeks and 1 year after delivery were
reported, but the study lacked suitable nonpregnant control patients or
intervals.26 In another population-based study,
there was no significantly elevated risk of thrombosis between 7
weeks and 1 year after delivery, but investigators did not assess risks across
discrete intervals during that time.4 Two
other studies suggested a possibly heightened risk of venous thromboembolism
between 7 and 12 weeks after delivery but lacked sufficient statistical power8 or had imbalances between cases and
controls, which probably resulted in an overestimation of postpartum risks.9
Despite this limited prior evidence,
our finding that increased risk for thrombosis persists beyond 6 weeks after
delivery has face validity. The magnitude of increased risk is high throughout
the 6 weeks after delivery,2,5 and it is unlikely that this
prothrombotic state would resolve suddenly. Our findings are consistent with a
more biologically plausible tapering of risk through at least 12 weeks after
delivery (Figure 1). This
pattern is concordant with data on laboratory coagulation markers after
delivery; most of these markers normalize by 6 weeks after delivery, but some
remain abnormal through at least 8 to 12 weeks after delivery.27-29
The validity of our study is buttressed
by its crossover-cohort design, which allowed each patient to serve as her own
control and thus reduced the unmeasured confounding that can occur with
traditional case–control or cohort studies.12 The validity of our study is further
supported by the consistency of our findings in a confirmatory case-crossover
analysis. Our study fully meets the assumptions of these crossover designs, in
that we modeled a transient, discrete exposure with stable prevalence over time
and an outcome that was defined by an acute event.30
Limitations of our study require
consideration, however. First, in the absence of prospective case ascertainment
and detailed clinical information, some outcome events may have represented
delayed sequelae of previous thrombotic events. For example, an outpatient in
whom venous thromboembolism is diagnosed at 2 weeks after delivery who is then
hospitalized with symptoms of venous thromboembolism 8 weeks later would have
incorrectly appeared to have had a first thrombotic event at 10 weeks after
delivery. This scenario would have artificially increased the apparent length
of time between delivery and outcome, thereby upwardly biasing our estimates
for later postpartum periods. However, we think that this possibility is
unlikely to have substantially affected our results. Although we may not have
captured some cases of venous thromboembolism that were diagnosed entirely in
the outpatient setting, almost all diagnoses of ischemic stroke and acute
myocardial infarction are made in the emergency department and result in
hospitalization,31,32 and our analysis of these arterial events alone was
consistent with our overall analysis. Furthermore, our estimates of the
magnitude of thrombotic risk within 6 weeks after delivery closely overlap with
those of previous studies that incorporated detailed clinical information,2,5 suggesting that we did not often miss thrombotic events and
incorrectly ascribe them to later periods. Second, patients may have been
progressively lost to follow-up during the 1 year 24 weeks after delivery owing
to unrecorded out-of-hospital deaths or emigration from California, and this
would also have upwardly biased our estimates. However, we think that this is
unlikely because we found the same results in a nested case-crossover analysis
that was limited to patients who were known to be alive and under observation
throughout the entire study period. Third, the sensitivities of the diagnosis
codes that we used to determine risk factors for thrombosis have not been
validated, and therefore our subgroup analyses may not have detected true
interactions between specific risk factors — especially between the presence of
a primary hypercoagulable state and smoking — and the duration of thrombotic
risk after delivery. Fourth, we lacked data from federal health care
facilities, which comprise 3.1% of the facilities in California.33
Current guidelines advise that
high-risk patients receive prophylactic anticoagulant therapy until 6 weeks
after delivery, but these recommendations are based largely on expert opinion.1 The
duration of therapy that best balances the risk of thrombosis with the risk of bleeding34,35 remains uncertain.36 Our findings suggest that the risks and
benefits of continuing treatment for high-risk women beyond 6 weeks after
delivery should be investigated. In addition, clinicians who are evaluating
possible symptoms of thrombosis in postpartum women should recognize that risk
remains increased for at least 12 weeks after delivery, although the absolute
risk of thrombotic events beyond 6 weeks after delivery is low.
The views
expressed in this article are those of the authors and do not necessarily
represent the official views of the National Institutes of Health (NIH).
Supported by a
grant from the National Institute of Neurological Disorders and Stroke of the
NIH (K23NS082367, to Dr. Kamel).
Disclosure forms provided by the authors are available with the full text of
this article at NEJM.org.
This article was
published on February 13, 2014, at NEJM.org.
SOURCE INFORMATION
From the
Department of Neurology (H.K., B.B.N., N.S., D.A.H.), Feil Family Brain and
Mind Research Institute (H.K., B.B.N.), and Division of Cardiology (R.B.D.),
Weill Cornell Medical College, the Department of Neurology, Columbia College of
Physicians and Surgeons (M.S.V.E.), and the Department of Epidemiology, Mailman
School of Public Health, Columbia University (M.S.V.E.) — all in New York.
Address reprint
requests to Dr. Kamel at 407 E. 61st St., 5th Fl., New York, NY 10065, or athok9010@med.cornell.edu.
No comments:
Post a Comment