Writing and Reviewing Abstracts

1
Writing and Reviewing Abstracts

• Society for Epidemiologic Research Student
  Caucus

2
What is an abstract?

• Concise summary of a full scientific research
  project describing the study and its results
• Abstracts generally include the most important
  aspects of the studys purpose, design, findings,
  and implications

3
What do I include in an abstract?

• Background and significance
• Why do we care about the problem and results?
• Include the importance of your work, challenges
  in this research area, and the potential impact
  of your work
• Hypothesis
• What question are you trying to answer?
• What is the scope of your work?

4
What do I include in an abstract?

• Methods
• How did you address your hypothesis?
• How did you collect your data?
• What analytic methods did you use?
• Results
• What are some characteristics of your data?
• Sample size
• Age range
• Racial breakdown
• What did you find?
• Report noteworthy results and findings

5
What do I include in an abstract?

• Conclusions
• What are the implications of your study?
• Are the results generalizable?
• What is the take home message for the audience?

6
Abstract considerations

• Stay within the word limit
• Most journals and conference submissions will not
  accept abstracts that are over the word limit
• Use active voice and simple sentences
• Spell out any abbreviations
• Think of key words for searches and be sure to
  use them in your abstract
• Have someone else review your abstract before
  submission

7
Example Abstracts

• Disclaimer Abstract reviews are the opinions of
  the SER-SC Executive Board

8
Sample 1 Home Use of Paints and Petroleum
Products and Risk of Childhood Leukemia

• Associations between childhood leukemia and
  parental occupational exposures to paints,
  solvents, and other petroleum products have been
  reported, however, little is known about the
  effect of these products when used at home. The
  role of home exposures to paints and petroleum
  products on the risk of childhood leukemia was
  evaluated in the Northern California Childhood
  Leukemia Study (NCCLS), an ongoing
  population-based case-control study. The current
  analyses include 382 incident leukemia cases
  (aged 0-14) enrolled from 1995-2002, and matched
  to 482 controls on age, sex, Hispanic status, and
  maternal race. A detailed history on home use of
  paints and petroleum products, time of exposure,
  and user was collected during an in home
  interview. Exposures were censored one year
  before diagnosis. Conditional logistic regression
  was performed adjusting for annual household
  income. An increased risk of childhood leukemia
  was observed with parental use of paints anytime
  pre- or postnatally (Odds Ratio (OR)1.35, 95
  Confidence Interval (CI)0.99, 2.02), with a
  higher and significant risk when paints were used
  during both time periods (OR1.72, 95 CI1.07,
  2.74). Maternal use of paints anytime pre- or
  postnatally conferred a significantly increased
  leukemia risk (OR1.44, 95 CI1.03, 2.02), while
  non-significant increased risks were detected for
  paternal use of paint anytime before or after
  birth. Exposures to paints by both parents after
  birth resulted in a two-fold significant
  increased risk (OR1.97, 95CI1.18, 3.30).
  Subjects with acute lymphoblastic leukemia had
  greater risks compared to subjects with acute
  myeloblastic leukemia. No significant
  associations were seen with exposure to petroleum
  products across all time windows for either
  parent.

9
Why is sample 1 a good abstract?

• Used active voice
• Indicated the importance of their study
• Described the larger study, years of use, and
  sample size
• Included information on study design and matching
  variables
• Provided statistical methods and used appropriate
  methods to analyze matched data
• Described results and appropriate conclusions
• What could have improved this abstract?
• Including summary conclusions

10
Sample 2 Association between body composition
and blood pressure in a contemporary cohort of
9-year old children

• Background Elevated blood pressure in children
  is an early risk factor for cardiovascular
  disease and is associated with body mass index
  (BMI). However as BMI does not distinguish
  between fat and lean, little is known about the
  relationship of blood pressure in children to
  different elements of body composition.
  Objective This study aimed to investigate the
  association of blood pressure with total body
  fat, lean mass and trunk fat in a cohort of
  9-year-old children. Design Blood pressure, BMI
  and body composition were measured in 6, 863
  children enrolled in the Avon Longitudinal Study
  of Parents and Children (ALSPAC). Fat mass, lean
  mass and trunk fat were assessed using
  dual-energy X-ray absorptiometry (DXA). Results
  Total body fat and BMI were strongly associated
  with systolic blood pressure (SBP) ß3.50, 95CI
  3.27 to 3.74 mmHg/standard deviation (SD) and
  ß3.96, 95CI 3.76 to 4.16 mmHg/SD, respectively
  and weakly associated with diastolic blood
  pressure (DBP) (ß1.39, 95 CI 1.22 to 1.57
  mmHg/SD and ß1.37, 95 CI 1.22 to 1.52 mmHg/SD,
  respectively). SBP was also positively associated
  with lean mass (ß3.60, 95 CI 3.22 to 3.97
  mmHg/SD) and trunk fat (ß2.14, 95 CI 0.82 to
  3.46 mmHg/SD, independent of total fat mass).
  Conclusion Blood pressure in 9-year-old children
  is independently associated with fat mass and
  lean mass and, to a lesser extent, trunk fat. In
  this analysis, because both fat and lean mass are
  associated with blood pressure, BMI predicts
  blood pressure at least as well as these
  components of body composition.

11
Why is sample 2 a good abstract?

• Authors
• Used active voice and divided the abstract by
  section
• Indicated the importance of their study
• Included their study objective
• Described the study sample and measurements
• Described results
• Provided appropriate conclusions
• What could have improved this abstract?
• Description of the analytical methods
• Inclusion of methods for measure and control of
  potential confounding factors

12
Sample 3 A Major Determinant of Recent
Increases in HIV Incidence Among Men who Have Sex
with Men (MSM) in British Columbia (BC) Deferred
Initiation of Antiretroviral Therapy

• Background Recent clinical guidelines endorse
  deferral of highly active antiretroviral therapy
  (HAART) until later in the course of HIV
  infection. The effect of this change on
  population levels of high viremia, infectivity
  and, in turn, sexual transmission of HIV is
  unknown. Objectives Describe populational
  trends in HIV viremia among MSM in BC in relation
  to a 1999 change that deferred HAART from CD4
  lt500 to CD4 lt200 cells/mL. Estimate relative
  increases in annual HIV diagnoses among MSM
  attributable to this change. Methods Linked
  provincial registries provided records of all
  HAART, CD4 counts, and plasma HIV RNA viral load
  (pVL) determinations from 1997-2003. We assessed
  trends in MSM with high viremia (pVL gt10 000
  copies/mL) across 6-month intervals. Mathematical
  models predicted relative changes in infections
  among MSM following introduction and deferral of
  HAART. Results The number and proportion of
  highly viremic MSM declined steadily from 443
  (27.6) for two years following introduction of
  HAART. This trend has reverted, increasing from
  358 viremic MSM (17.8) in the second half of
  1999 to 713 (26.7) at the end of 2003 the
  increase occurred among men with CD4 gt200. Models
  predicted a 50 reduction in HIV diagnoses among
  MSM from 1997-1999 but rapid 83 increase
  following deferral of HAART. Conclusion
  Deferral of therapy appears to be a major
  determinant of the 75 increase in annual HIV
  diagnoses among MSM in BC from 1999-2004.
  Substantial increases in HIV prevention among MSM
  seem warranted.

13
Why is sample 3 a good abstract?

• Authors
• Used active voice and defined acronyms
• Divided the abstract by section
• Indicated the importance of their study
• Described their objective
• Included information on their sample and study
  period
• Described results and methods
• What could have improved this abstract?
• Sample size information
• More detailed data collection methods

14
Overall Summary of Sample Abstracts

• All abstracts were under the 250 word limit
• All authors used active voice
• Defined any acronyms used
• All abstracts included background and importance
  information
• Described their study population, methods, and
  results
• Provided appropriate conclusions for their results

15
Reviewing an Abstract
16
How to Review an Abstract?

• Questions to consider when reviewing an abstract
• Is the work appropriate for the journal or
  conference?
• At conferences, there is usually a theme or
  multiple areas of science that are sought
• Journals have specific aims and goals
• Is the work original?
• Is the work complete?
• Is the work ready to publish or present at a
  conference? Are the results complete?

17
How to Review an Abstract?

• Did the authors describe the important features
  of an abstract?
• Background and significance
• Hypothesis
• Methods
• Data collection methods
• Sample size and population information
• Analytic techniques
• Results
• Demographic characteristics of sample
• Significant results
• Conclusions

18
Deciding Between a Poster or Oral Presentation

• If you are asked to determine what type of
  acceptance an abstract should receive
• Ask for guidelines from the conference organizers
• Other considerations
• What is the conference theme?
• If so, does the abstract address the theme?
• Are there special topic areas for the conference?
• If so, does the abstract address any of these
• Is the quality of the abstract above others
  reviewed?

19
Practice Abstract Review
20
Abstract Evaluation Guidelines

• Assumptions
• Reviewing for Society for Epidemiologic Research
  (SER) Conference
• Areas of interest epidemiology, biostatistics,
  and public health
• Did the author include the important components
  of an abstract
• Follow the review outline provided in slides 19
  and 20

21
Review Sample 1 Bayesian propensity score
analysis for observational data

• Propensity scores analysis (PSA) involves
  regression adjustment for the estimated
  propensity scores, and the method can be used for
  estimating causal effects from observational
  data. However, confidence intervals may be
  falsely precise because PSA ignores uncertainty
  in the estimated propensity scores. We propose
  Bayesian propensity score analysis (BPSA) for
  observational studies which involve a binary
  exposure, binary outcome and measured
  confounders. The method uses hierarchical
  modelling with the propensity score as a latent
  variable. The first level models the
  relationship between the outcome, exposure and
  propensity score, while the second level models
  the relationship between the propensity score and
  measured confounders. Markov chain Monte Carlo
  is used to study the posterior distribution of
  the exposure effect. Our objective is to
  implement BPSA using computer programs and
  investigate the performance of BPSA compared to
  PSA using Monte Carlo simulations. Synthetic
  datasets, of sample size n250, 1000, 4000, were
  simulated by computer for various realistic
  parameter values. The datasets were analyzed
  using BPSA and PSA, and we estimated the coverage
  probability of 80 credible intervals. The
  estimated coverage probabilities ranged from 78
  to 84 for BPSA, and from 42 to 82 for PSA,
  with simulation standard errors less than 2. The
  simulation results indicate that BPSA provides
  improved inferences for exposure effects compared
  to PSA, in the sense that interval estimators
  have the correct frequentist coverage levels
  under repeated sampling of the data. We
  demonstrate BPSA in an observational study of the
  effect of statin therapy on all-cause mortality
  in patients discharged from hospital following
  acute myocardial infarction.

22
Abstract Evaluation Review Sample 1

• Is the work appropriate for the SER conference?
• Yes this describes analytic methods that can be
  used for epidemiology studies
• Is the work original?
• Yes - the authors developed a more appropriate
  method to calculate confidence intervals
• Is the work complete?
• Yes the authors simulated data to test their
  hypothesis

23
Abstract Evaluation Review Sample 1

• Are the major components included?
• Background and significance?
• Yes the authors described the need for their
  work and why it is important
• Hypothesis?
• Yes they proposed the use of Bayesian
  propensity score analysis (BPSA) for
  observational studies which involve a binary
  exposure, binary outcome and measured confounders
• Methods?
• Yes they described how they were going to
  address the hypothesis above

24
Abstract Evaluation Review Sample 1

• Results?
• Yes the authors described their results and the
  accuracy of their measures
• Conclusions?
• Slightly unclear We demonstrate BPSA in an
  observational study of the effect of statin
  therapy on all-cause mortality in patients
  discharged from hospital following acute
  myocardial infarction.
• No where else in the abstract do they mention the
  use of these methods for the effect of statin
  therapy
• Thus, the conclusions may not be appropriate
• Overall Good abstract and would accept it and
  point out concern about conclusions to authors

25
Review Sample 2 Exposure to H. pylori-positive
siblings and persistence of H. pylori infection
in early childhood

• H.pylori infection is a common chronic infection,
  yet transmission pathways are unclear evidence
  suggests that siblings play a role in
  transmission. Transient H.pylori infection is
  observed in children, but determinants of
  persistence are unknown. We examine the effect of
  exposure to H.pylori-positive siblings on the
  establishment of persistent H.pylori infection
  using data from the Pasitos Cohort Study, which
  recruited pregnant women from maternal-child
  clinics in El Paso, Texas, and Juarez, Mexico
  during 1998-2000 and followed 472 children after
  birth to identify predictors of H.pylori
  infection. Infection was detected at target
  intervals of 6 months in index children and
  younger siblings born during follow-up, using the
  13C-urea breath test (UBT) corrected for
  variation in CO2 production. We used proportional
  hazards regression to estimate hazard ratios for
  the effect of having H.pylori-positive younger
  siblings on the rate of developing a persistent
  H.pylori infection in index children with 1
  younger siblings. Persistent infection was
  defined as 3 consecutive positive UBT results. We
  modeled two exposure definitions infected
  younger sibling (1 younger siblings with 1
  positive UBT results) persistently infected
  younger sibling (1 younger siblings with a
  persistent infection). Adjusting for mom's
  education (the strongest H.pylori risk factor in
  this cohort), the hazard ratio was 4.0 (95
  CI1.8,8.2) for infected younger sibling and 9.3
  (95 CI3.2,26) for persistently infected younger
  sibling. Having H.pylori-positive younger
  siblings, particularly with persistent infection,
  was strongly associated with developing
  persistent H.pylori infection in this cohort.

26
Abstract Evaluation Review Sample 1

• Is the work appropriate for the SER conference?
• Yes study describes an important infectious
  disease
• Is the work original?
• Yes the authors are looking at an area of this
  infection that is not well understood
• Is the work complete?
• Yes the authors analyzed existing data to
  address their question

27
Abstract Evaluation Review Sample 1

• Are the major components included?
• Background and significance?
• Yes the authors described the need for their
  work and why it is important
• Hypothesis?
• Yes the authors examined the effect of exposure
  to H.pylori-positive siblings on the
  establishment of persistent H.pylori infection
• Methods?
• Yes authors described their study population,
  defined their study variables, and included the
  analytical methods used to address the hypothesis

28
Abstract Evaluation Review Sample 1

• Results?
• Yes the authors described some of their results
• However, the authors did not provide the number
  of events that were included in the Cox Models
• Conclusions?
• Yes authors described appropriate conclusions
  for the results presented
• Overall
• Good abstract
• Should be accepted for presentation at the
  conference

29
Resources

• Pierson DJ. How to Write an Abstract That Will Be
  Accepted for Presentation at a National Meeting,
  Respiratory Care, October 2004, 49(10)1206-1212
• Kiipman P. How to Write an Abstract.
  http//www.ece.cmu.edu/koopman/essays/abstract.ht
  ml
• Sample Abstracts Obtained From
• American Journal of Epidemiology, June 1, 2006
  163 (Supplement 11).

30
AppendixAdditional Abstracts
31
Sample A Lifetime Recreational Physical Activity
and Breast Cancer Risk

• While a number of studies have observed a reduced
  risk of breast cancer with increased levels of
  physical activity, these findings have been
  inconsistent regarding the optimal frequency,
  intensity, and timing of exercise. We performed a
  population-based case-control study in
  Massachusetts, New Hampshire, and Wisconsin. From
  the states' cancer registries, we enrolled 6421
  women (80 of eligible) aged 20-69 diagnosed with
  invasive breast cancer between 1995 and 2000.
  Controls (7683 76 of eligible) were selected at
  random from lists of licensed drivers and
  Medicare beneficiaries. In structured telephone
  interviews, women reported lifetime history (age
  14 to present) of recreational physical activity.
  After adjustment for potential confounding risk
  factors using logistic regression, women
  averaging greater than six hours of activity per
  week over their lifetime had a reduced odds ratio
  (OR) of breast cancer when compared to women
  reporting no recreational activity (OR 0.81, 95
  CI 0.68-0.97 P trend0.20). Analyses considering
  metabolic equivalents of energy expenditure (MET)
  produced similar results. While an inverse
  relation between physical activity averaged over
  the lifetime and breast cancer risk was observed,
  the relation was null when limited to activity in
  a single decade. The inverse association with
  lifetime activity was consistent regardless of
  menopausal status and body mass however, it was
  limited to women with no family history of breast
  cancer (OR 0.75, 95 CI 0.61-0.91 for 6 vs. 0
  hrs/week, P-interaction 0.02). The results from
  this large study provide further evidence that
  long-term recreational physical activity may
  reduce breast cancer risk.

32
Why is sample A a good abstract?

• Authors
• Used active voice
• Indicated why their study is important and
  interesting
• Gave a clear statement of purpose
• Included information about the selection of cases
  and controls
• Discussed data collection measures
• Described methods and results
• Provided appropriate conclusions

33
Sample B Mapping and measuring social
disparities in premature mortality the impact
of census tract poverty within and across Boston
Neighborhoods, 1999-2001

• The documentation of health disparities is an
  important function of public health surveillance.
  These disparities, typically falling along lines
  defined by gender, race/ethnicity, and social
  class, are often visible in urban settings as
  geographic disparities in health between
  neighborhoods. Recognizing that premature
  mortality is a powerful indicator of social
  disparities in health, we undertook an analysis
  of spatial variation in premature mortality in
  Boston (1999-2001) across neighborhoods and
  sub-neighborhoods in relation to census tract
  (CT) poverty. Using a multilevel model based
  framework, we estimated that the incidence of
  premature mortality was 1.39 times higher (95
  confidence interval 1.09-1.78) among persons
  living in the most economically deprived CTs
  (gt20 below poverty) compared to those in the
  least impoverished tracts (lt5 below poverty). We
  present maps of model-based standardized
  mortality ratios that show substantial
  within-neighborhood variation in premature
  mortality and a sizeable decrease in spatial
  variation after adjustment for CT poverty. To aid
  interpretation, we also map smoothed
  age-standardized rates that can be compared to
  externally published rates. Additionally, we
  present maps of the model-based population
  attributable fraction that show that in some of
  Boston's poorest neighborhoods, the proportion of
  excess deaths associated with CT poverty reaches
  25-30. We recommend that these methods be
  incorporated into routine analyses of public
  health surveillance data to highlight continuing
  social disparities in premature mortality.

34
Why is sample B a good abstract?

• Authors
• Used active voice and defined acronyms
• Indicated the importance of their study
• Described their objective and their analytic
  strategy (spatial mapping)
• Included the geographic area
• Described the maps generated
• What could have improved this abstract?
• Explanation of their results and how to interpret
  the maps