STA 105: Study Designs for Medical Research

1
STA 105 Study Designs for Medical Research

• Lecturer Dr. Daisy Dai
• Department of Medical Research

2
Design will impact the study results and
inferences
3
When it comes to a design

• Consult statisticians
• Conduct literature review
• Assess strength and limitation of a design
• Think outside of a box.

4
Observational Studies

• Prospective Cohort Studies
• Retrospective Cohort Studies
• Case-Control Studies
• Cross-Sectional Studies
• Pre-post Tests

5
Cohort Studies

• Cohort studies involve following groups of
  subjects over time.
• The goal of the descriptive cohort study is to
  describe the occurrence of certain of outcomes
  over time.
• The analytic cohort study is to analyze
  associations between predictors and outcomes.

6
Prospective cohort study

• The investigator defines the sample and measures
  predictor variables before undertaking a
  follow-up period to observe outcomes.

7
Nurses Health Study

1. Goal to examine incidence and risk factors for
   common diseases in women.
2. Assemble the Cohort In 1976, the investigators
   obtained lists of registered nurses aged 25 to 42
   in the 11 most populous states and mailed them an
   invitation to participate in the study those who
   agreed became the cohort.
3. Measure Predictor Variables and Potential
   Confounders. They mailed a questionnaire about
   weight, exercise and other potential risk factors
   and obtained completed questionnaires from
   121,700 nurses. They send questionnaires
   periodically to ask about additional risk factors
   and update the status of some risk factors that
   had been measured previously.
4. Follow-up the Cohort and Measure Outcomes. The
   periodic questionnaires also included questions
   about the occurrence of a variety of disease
   outcomes.
5. Findings Heavier women has a higher risk of
   breast cancer after menopause, and those who
   gained more than 20kg since age 18 had a twofold
   increased risk of developing breast cancer.

8
Strengths and Weaknesses

• Time sequence of risk factors
• Information on baseline
• Complete measure of risk factors and outcome
  variables
• Recall error is minimized

• Hard to make causal inference
• Expensive
• Low incidence rate and long waiting period for an
  event
• Sensitive to attrition

9
Retrospective cohort study

• The follow-up period and outcomes have already
  occurred when the study takes place.

10
Case Study aortic aneurysm

• Identify a Suitable Cohort. The investigators
  used the residents of Olmsted County, Minnesota.
  They searched a database of diagnoses made
  between 1980 and 1995 and found 133 residents who
  had a diagnosis of aortic aneurysm.
• Collect Data about Predictor Variables. They
  reviewed patients records to collect gender,
  age, size of aneurysm, and risk factors for
  cardiovascular disease at the time of diagnosis.
• Collect data about Subsequent Outcomes. They
  collected data from the medical records of the
  133 patients to determine whether the aneurysm
  ruptured or was surgically repaired.
• Findings The 5-year risk of rupture was 20 and
  that women were 6.8 times more likely to suffer a
  rupture than men. 31 of aneurysms with diameters
  of more than 6cm ruptured, compared with none
  with diameters of less than 4cm.

11
Strengths and Weaknesses

• Less costly
• Time saving

• Existing data may be incomplete and inaccurate
• Confounding factors
• Some information might be outdated

12
Strategies for Minimizing Losses during Follow-Up

• During enrollment
• Exclude those likely to be lost
• Obtain information to allow future tracking
• During follow-up
• Periodic contact with subjects to collect
  information, provide results, express care, and
  so on.
• For those who are not reached by phone or mail,
  consider home visit.
• At all times
• Treat study subjects with appreciation, kindness
  and respect, helping them to understand the
  research question so they will want to join as
  partners in making the study successful.

13
Cross-sectional Study
14
Cross-sectional study

• The structure of a cross-sectional study is
  similar to that of a cohort study except that all
  the measurements are made at about the same time,
  with no follow-up period.

15
Case Study Smoking in Movies

• Goal Investigate the association between actor
  smoking and teenage audience smoking
• Selected the Sample a random-digit-dial survey
  of 6,522 children aged 10 to 14 years.
• Measured the Variables They quantified smoking
  in 532 popular movies and for each subject asked
  which of a randomly selected subset of 50 movies
  they had seen. Subjects were also asked about a
  variety of covariates such as age, rage, gender,
  parental smoking and education, sensation-seeking
  (e.g., I like to do dangerous things) and
  self-esteem (e.g., I wish I were someone else.)
  The outcome variable was whether the child had
  ever tried smoking a cigarette.
• Findings Adjusted for age and other confounders,
  38 of smoking initiation was attributable to
  exposure to movies in which the actors smoke.

16
Strengths and Weaknesses

• One time measure.
• Fast.
• Less costly.

• Hard to make causal conclusions.
• No time trend

17
Case-Control Study
18
Case-control Study

• In a case-control study, the investigator (a)
  selects a sample from a population with the
  disease (cases), (b) selects a sample from a
  population at risk that is free of the disease
  (controls), and (c) measures predictor variables.

19
Case Study

• Objectives Because intramuscular (IM) vitamin K
  is given routinely to newborns in the United
  States, a pair of studies reporting a doubling in
  the risk of childhood cancer among those who had
  received IM vitamin K caused quite a stir. To
  investigate this association further, German
  investigators
• Selected the Samples 107 children with leukemia
  from the German Childhood Cancer Registry. 107
  children matched by sex and date of birth and
  randomly selected from children living in the
  same town as the case at the time of diagnosis.
• Measure the Predictor Variable. Reviews medical
  records to determine which cases and controls had
  received IM vitamin K in the newborn period.
• Findings 69 of 107 cases (64) and 63 of 107
  controls (59) had been exposed to IM vitamin K
  with an odds ratio 1.2 and 95 CI 0.7 2.3.
  Therefore, the study did not confirm the
  existence of an association between the receipt
  of IM vitamin K as a newborn and subsequent
  childhood leukemia.

20
Strengths and Weaknesses

• Reduce sample size.
• Reduce the variation within the pair

• Only study one outcome.
• Susceptible to bias.

21
Nested Case-Control and Case-Cohort Studies

• In a case-control study, the investigators (a)
  identifies a cohort with banked specimens, images
  or information (b) identifies those participants
  who developed the outcome during follow-up (the
  cases) (c) selects the rest of the cohort (the
  controls) and (d) measures predictor variables
  in cases and controls.
• The nested case-cohort approach is the same
  design except that the controls are a random
  sample of all the members of the cohort
  regardless of outcomes.

22
Sex hormones in patients with breast cancer

• Identify a Cohort with Banked Samples. The
  investigators used serum and data from the Study
  of Osteoporotic Fractures, a prospective cohort
  of 9,704 women age 65 and older.
• Identify Cases at the End of Follow-up. Based on
  responses to follow-up questionnaires and review
  of death certificates, the investigators
  identified 97 subjects with a first occurrence of
  breast cancer during 3.2 years of follow-up.
• Select Controls. The investigators selected a
  random sample of 244 women in the cohort who did
  not develop breast cancer during that follow-up
  period.
• Measure Predictors on Baseline Samples from Cases
  and Controls. Levels of estradiol and
  testosterone were measure in serum specimens from
  the baseline examination that had been stores at
  190C by laboratory staff who were blinded to
  case-control status.

23
Case Study

• To determine whether significant neonatal
  jaundice or dehydration has any significant
  adverse effects on neurodevelopment,
  investigators from UCSF and the Northern
  California Kaiser Permanente Medical Care Program
  (5) undertook a triple-cohort study. The basic
  steps in performing the study were to
• Identify Cohorts with Different Exposures. The
  investigators used electronic databases to
  identify term and near-term newborns who (1) had
  a total serum bilirubin level of 25 mg/dL, or
  (2) were readmitted for dehydration with a serum
  sodium of 150 mEq/L or weight loss of 12
  from birth, or (3) were randomly selected from
  the birth cohort.
• Determine Outcomes The investigators used
  electronic databases to search for diagnoses of
  neurological disorders and did full
  neurodevelopmental examinations at the age of 5
  for consenting participants.

24
Advantages and Disadvantages of the Major
Observational Designs
25
Randomized Controlled Trials
26
Randomized controlled trial (RCT)

• A study in which people are allocated at random
  (by chance alone) to receive one of several
  clinical interventions. One of these
  interventions is the standard of comparison or
  control. The control may be a standard practice,
  a placebo ("sugar pill"), or no intervention at
  all. RCTs seek to measure and compare the
  outcomes after the participants receive the
  interventions.

27
Case Study FEV1 Changes

• A new compound, ABC-123, is being developed for
  long-term treatment of patients with chronic
  asthma. Asthma patients were enrolled in a
  double-blind study and randomized to receive
  daily oral or a placebo for 6 weeks.

asthmatic patients
Placebo
Test
FEV1 after 6-week treatment
28
FEV1 Data
Test Group Test Group Test Group
Patient ID Baseline week 6
101 1.35 n/a
103 3.22 3.55
106 2.78 3.15
108 2.45 2.3
109 1.84 2.37
110 2.81 3.2
113 1.9 2.65
116 3 3.96
118 2.25 2.97
120 2.86 2.28
121 1.56 2.67
124 2.66 3.76
Placebo Group Placebo Group Placebo Group
Patient ID Baseline week 6
102 3.01 3.9
104 2.24 3.01
105 2.25 2.47
107 1.65 1.99
11 1.95 n/a
112 3.05 3.26
114 2.5 2.55
115 1.6 2.2
117 .77 2.56
119 2.06 2.9
122 1.71 n/a
123 3.54 2.92
29
What is the difference between std and std error?
P-value
P-value
30
Mean and Error Bar

• Conclusion
• As compared to placebo, the new drug did not
  show any effect on FEV1.

31
Paired t-test

• Compare the means of a normally distributed
  interval dependent variable for two related
  groups.

Test Group Test Group Test Group
Patient ID Baseline week 6
101 1.35 n/a
103 3.22 3.55
106 2.78 3.15
108 2.45 2.3
109 1.84 2.37
110 2.81 3.2
113 1.9 2.65
116 3 3.96
118 2.25 2.97
120 2.86 2.28
121 1.56 2.67
124 2.66 3.76
32
Conclusion For subjects on the new drug, FEV1
at week 6 is significantly higher than baseline.
P-value
33
Strengths and Weaknesses

• RCTs are quantitative, comparative, controlled
  experiments in which investigators study two or
  more interventions in a series of individuals who
  receive them in random order. The RCT is one of
  the simplest and most powerful tools in clinical
  research.

• Ethics
• Recruit of participants
• Cost

34
Case Studies and Discussions
35
Probiotics

• A recent study identified 63 medical trials that
  compared probiotics to placebo or to no
  probiotics for treating short-term diarrhea in
  8,014 people, mostly children. Probiotics,
  commonly found as a supplement in many yogurts
  and other foods, are thought to fight diarrhea in
  a number of waysperhaps by competing with the
  pathogens for nutrients or by spurring a stronger
  immune response. Whatever the reasons, probiotic
  treatments shortened the course of diarrhea in
  nearly all the studies, and by an average of 25
  hours. (The average total duration of diarrhea
  in the studies ranged from 39 to 161 hours.)
  Patients taking probiotics were also 59 less
  likely to experience diarrhea lasting longer than
  three days.
• Caveat The trials tested a range of probiotic
  strains and doses the meta-analysis could not
  determine whether certain strains were more
  effective than others.

36
Cataracts

• Neither vitamin E nor vitamin C protests men
  against cataracts, according to a study in the
  Archives of Ophthalmology. The antioxidant
  vitamins had once been candidates to reduce the
  risk of developing cataracts (cloudy areas in the
  eye that can impair vision), since they already
  play a protective role within eye membranes.
  Though several recent randomized trials have
  found little evidence in favor of the treatment,
  none have followed their participants for more
  than 6½ years. In this study, researchers
  randomly assigned 11,545 male doctors to take a
  daily dose of vitamin C or placebo daily. The
  same group was also randomly assigned to receive
  either vitamin E or placebo every two days. After
  an average of eight years of follow-up, 1174 of
  the participants developed cataracts, evenly
  spread between the vitamin and placebo groups.
  The vitamins also had no effect on any subtype of
  cataract.
• Caveat Doctors might have healthier diets than
  the general population, and thus might derive
  less benefit from vitamin supplementation. Its
  also possible that the protective effects of
  vitamins E and C only emerge after a period even
  longer than eight years.

37
Fish Oil

• Omega-3 fatty acids, such as those found in fish
  oil, dont prevent atrial fibrillation from
  recurring, according to a study in the Journal of
  American Medical Association. Researchers
  recruited 542 patients who had been diagnosed
  with a short-term form of atrial fibrillation.
  Each patient was randomly assigned to take daily
  capsules of either omega-3 fatty acids or placebo
  for six months. Though omega-3s have been
  associated in observational studies with improved
  heart health, they provided little benefit in
  this study By the end of the six months, atrial
  fibrillation had returned in 52 of the patients
  taking omega-3s and 48 of those taking placebo,
  a statistically insignificant difference.
  Omega-3s also did not prevent atrial fibrillation
  recurrences among an additional 121 patients who
  had a more persistent form of the condition.
• Caveat Its possible that omega-3s protect
  against atrial fibrillation only in the
  longer-term. Because the study recruited only
  otherwise healthy patients, the results might not
  apply to the elderly, patients with other forms
  of heart disease, or those who have recently had
  heart surgery.

38
Satiety Study

• Satiety is subjective, variable and difficult to
  measure. Two experiments were conducted to assess
  the impact of Crossover Design and Parallel
  Design to the same stimuli and speculate the
  sensitivity and reliability of continuous and
  dichotomous measures in satiety studies.

39
Problem Statements

• Visual analogue scale has been widely used to
  measure satiety. However, the outcome variables
  were often noisy and sensitive to design
  manipulation.
• Many psychometric rating systems have been
  developed to measure satiety and sensation. But
  there are few experiments conducted to compare
  different measures.
• In this work, we conduct two experiments to
  answer following questions
• Is continuous measure superior to dichotomous
  measure?
• What factors do we consider in selection between
  cross-over design and parallel design?
• What are the halo effects in cross-over design?

40
Methods

• Experiment 1
• Crossover design (n23)
• Frozen meals with different sensation agents,
  Control and Test, were randomly assigned to
  subjects.

• Experiment 2
• Stratified Parallel Design.
• Subjects were stratified by baseline hunger and
  randomized to Control (n27) and Test (n29).
• We provided a reduced calorie diet. By comparing
  the results with and without this experiential
  manipulation, we can assess the degree of
  influence from this confounding factor and also
  evaluate the satiety measures sensitivity and
  robustness.
• Other conditions remained same as Experiment1.

41
In both experiments, satiety was measured by

• Continuous Hunger Likert Score Hunger,
  fullness, thirst, desire for a snack, thirst and
  general mood were evaluated at baseline and every
  30 minutes for up to 5 hours post meal using
  11-point Likert scale.
• Dichotomous Measure Subjects were asked whether
  they would decline their favorite food or dessert
  at baseline and every 30 minutes for up to 5
  hours post meal.

42
(No Transcript)
43
(No Transcript)
44
(No Transcript)
45
(No Transcript)
46
Results

• Significant treatment and time effects were found
  using Dichotomous Measure in both Crossover
  Design (Figure 1) and Stratified Parallel Design
  (Figure 2). Only the time effect but no treatment
  effect was detected in Hunger and Fullness Likert
  Scales measured every half hour (Figures 3 4).
• Both Dichotomous Measure and continuous Likert
  Measures encountered the adaptation effect in
  Crossover Design (Experiment 1) the satiety
  perception was significantly different between
  two visits after adjusting for the treatment
  effect.
• The Hunger Likert Scores had an undesirable
  treatment by day of visit interaction at
  baseline, which indicates Continuous Measures
  could be less robust or more variable as compared
  to Dichotomous Measure (Figure 3).
• In Experiment 2, the stratification technique
  objectively calibrated the psychometric ratings
  among subjects at baseline. Also thanks to the
  removal of the adaptation effect, Hunger Liker
  Scores under Stratified Parallel Design were
  cleaner (Figure 4).

47
Conclusions

• Our study suggests that discrete measures, with
  less psychological/rating variability, could be
  more powerful and robust than continuous measures
  in assessment of small to moderate satiety
  effects. A Stratified Parallel Design might help
  reduce the experimental halo for studies with
  confounding adaptation effects.

48
Summary

• Observational Studies vs. Randomized Controlled
  Trials (RCTs)
• Cohort Studies vs. Cross-Sectional Studies
• Prospective Studies vs. Retrospective Studies
• Case-control Studies and Pre-Post Tests

49
References

• Designing Clinical Research 3rd edition by Hulley
  et al.