Activities & Exercises
Activities-Research Question
Question 1 :
What is the relationship between depression and health?
First, write a research question in a single sentence that specifies a study design, predictor, outcome, and population. Then evaluate whether this research question and the design you have chosen meet the FINER (Feasible, Interesting, Novel, Ethical, Relevant) criteria. Rewrite the question and design until you have overcome any problems in meeting these criteria.
Question 2 :
Does eating red meat cause cancer?
First, write a research question in a single sentence that specifies a study design, predictor, outcome, and population. Then evaluate whether this research question and the design you have chosen meet the FINER (Feasible, Interesting, Novel, Ethical, Relevant) criteria. Rewrite the question and design until you have overcome any problems in meeting these criteria.
Question 3 :
Can relaxation exercises decrease anxiety during a mammogram?
First, write a research question in a single sentence that specifies a study design, predictor, outcome, and population. Then evaluate whether this research question and the design you have chosen meet the FINER (Feasible, Interesting, Novel, Ethical, Relevant) criteria. Rewrite the question and design until you have overcome any problems in meeting these criteria.
Writing activity - PICO / PECO
The acronyms PICO and PECO sum up key elements of clinical and epidemiological studies, and can help focus the research question:
- P – who were the participants or population? what problem was addressed?
- I or E – what was the intervention or exposure?
- C – what was the comparison group?
- O – what was the outcome or endpoint?
Can relaxation exercises decrease anxiety during a mammogram?
“What was the impact of Brazil’s Family Health Program on age standardised mortality rates from heart and cerebrovascular diseases for 20-74 year olds between 2000 and 2009?” (longitudinal study). Davide et al BMJ 2014;349:g4014
- P
- I or E
- C
- O
“What was the effectiveness of antenatal continuous glucose monitoring on maternal glycaemic control and pregnancy related morbidity—namely, birth weight and risk of macrosomia in the offspring of mothers with type 1 and type 2 diabetes?” (RCT). Murphy et al BMJ 2008; 337: a1680
- P
- I or E
- C
- O
Question 4 :
Designing cross-sectional and case-control studies
The research question is, “Does maternal height and weight predict infant birth weight?” During a 12-month period, an investigator assembles data on consecutive newborns in a large maternity hospital. The study is limited to term newborns as defined by delivery 38 to 42 weeks after the mother’s last menstrual period. In the maternity ward, the investigator measures each infant’s birth weight and the mother’s height and weight. Based on the data obtained, the investigator concludes that birth weight is strongly dependent on both maternal height and weight.
- a. What kind of study is this?
- b. Explain why you agree or disagree with the investigator’s conclusions.
Answer 4 :
- a. The study is cross-sectional because the potential predictors (maternal height and weight) are measured at essentially the same time as the outcome (infant birth weight).
- b. Causal inference in cross sectional observational studies depends on the time sequence of the predictor and outcome variables, and on the possible role of confounding variables. In this study, the mother’s weight was measured just after delivery and reflects a combination of the mother’s weight before she became pregnant and the amount of weight that she gained during pregnancy. Since the amount of weight gained during pregnancy may depend on the weight of the fetus, it is not clear which variable is the predictor. However, the mother’s height measured just after delivery probably does not differ from her height if it had been measured before conception. It is reasonable therefore to conclude that maternal height is a predictor of birth weight. However, the phrase “dependent on” conveys a hint of causal inference that goes beyond the findings, since confounders like nutritional state of the mother may be operating.
Question 5 :
Designing cross-sectional and case-control studies
The research question is, “How much does a family history of ovarian cancer increase the risk for ovarian cancer?” The investigator plans a case-control study to answer this question.
- A. How should she pick the cases?
- B. How should she pick the controls?
- C. Comment on potential sources of bias in the sampling of cases and controls.
- D. How would she measure “family history of ovarian cancer” as the predictor variable of interest? Comment on the sources of bias in this measurement.
- E. What measure of association would she use, and what test of statistical significance?
- F. Do you think the case-control method is an appropriate approach to this research question? Discuss the advantages and disadvantages of the case-control design relative to other possibilities for this research question.
Answer 5 :
- a. The cases might consist of all women between 30 and 75 years of age with ovarian cancer reported to the Northern California Cancer Center Tumor Registry. This tumor registry has been shown to include nearly 100% of incident ovarian cancer cases in five San Francisco Bay Area counties.
- b. The controls might be a random sample of all women between 30 and 75 years of age from the same five counties in the San Francisco Bay Area. The random sample might be obtained by using random-digit dialing.
- c. The methods outlined for choosing cases and controls are aimed at obtaining all cancer cases and a random sample of those at risk in the target population (women 30 to 75 years old in five Bay Area counties). However, it is possible, since ovarian cancer requires intensive therapy and is deadly, that some cases may be unwilling to enroll in the study or may die before they can be interviewed. If a family history of ovarian cancer is related to more aggressive forms of ovarian cancer, then the study might underestimate its relative risk, because those cases with a positive family history are less likely to survive long enough to be included in the sample of cases. If familial ovarian cancer is more benign than other ovarian cancers, the opposite could occur. Similarly, it is possible that healthy women who have a family member with ovarian cancer will be more interested in the study and more likely to enroll as a control that women who do not have a family member with ovarian cancer. In that case, the prevalence of family history of ovarian cancer in the control group will be artificially high, and the estimate of the risk for ovarian cancer due to family history will be falsely low. This problem might be minimized by not telling the potential control subjects exactly what the research question is or exactly which cancer is being, if this can be done in a way that is acceptable to the IRB.
- d. Family history of ovarian cancer is generally measured by asking subjects about how many female relatives they have, and how many of them have had ovarian cancer. Recall bias is a possible problem with this approach. Women with ovarian cancer, who may be concerned about the possibility of a genetic predisposition to their disease, may be more likely to remember or find out about relatives with ovarian cancer than healthy women who have not had reason to think about this possibility. In this case, the estimate of the association between family history and ovarian cancer may be falsely high. In addition, women may confuse the gynecological cancers (cervical, uterine, and ovarian) and confuse benign gynecological tumors that require surgery with malignant tumors. This may cause misclassification (some women without a family history of ovarian cancer will report having the risk factor and be misclassified). If misclassification occurs equally in the cases and controls, the estimate of the association between family history and ovarian cancer will be falsely low. If this type of misclassification is more common in cases (who may be more likely to misinterpret the type of cancer or the reason for surgery in relatives), then the estimate of the association between family history and ovarian cancer will be falsely high. Misclassification could be decreased by checking pathological records of family members who are reported to have ovarian cancer to verify the diagnosis.
- e. The simplest approach would be to dichotomize family history of ovarian cancer and use the odds ratio as the measure of association. The odds ratio approximates the relative risk because the outcome (ovarian cancer) is rare. A simple chi-square would then be the appropriate test of statistical significance. Alternatively, if family history were quantified (e.g., proportional of first- and second-degree female relatives affected), one could look for a dose-response, computing odds ratios at each level of exposure.
- f. The case-control design is a reasonable way to answer this research question despite the problems of sampling bias, recall bias, and misclassification that are noted above. A nested case-control study in generally preferable if possible. The chief alternative would be a large cohort study, but because ovarian cancer is so rare, a cohort design is probably not feasible.
Question 5 :
Designing cross-sectional and case-control studies
The investigator wants to investigate the relationship between playing video games involving car racing and the risk of being involved in a real car crash (as the driver). She has two research questions, one that addresses the long-term effects of habitual use of the games, and the other that tests whether use of such games in the hour immediately preceding driving increases short-term risk. What are some designs she might consider?
Question 6 :
Designing a cohort study
The research question is, “Does vitamin B12 deficiency cause hip fractures in the elderly?”
- a. Briefly outline a study plan to address this research question with a prospective cohort study.
- b. An alternative approach would be to compare evidence of vitamin B12 deficiency in women who have had previous hip fracture with that in women who have not. Compared with this “case-control” approach, list at least one advantage and one disadvantage of your prospective cohort study. (If you like, you can come back to this question after you have read Chapter 8).
- c. Could the cohort study be designed as a retrospective study, and how would this affect these advantages or disadvantages?
Answer 5 :
- a. Measure serum vitamin B12 levels in a cohort of white women more than 70 years of age and without a history of hip fractures and analyze the association with incident hip fractures observed over the next 5 years. (Choice of study subjects is based on the fact that hip fracture is most common in white women; the age cutoff is somewhat arbitrary, but based on the age at which hip fracture incidence rises rapidly and to substantial levels.) The use of reported dietary intake of foods rich in vitamin B12 instead of the serum level might be more expensive because dietary histories take a lot of time to collect and score and would certainly be a less precise and less accurate measurement.
- b. Advantages of the prospective cohort design for studying vitamin B12 sufficiency and hip fractures:
Temporal sequence (i.e., the hip fracture follows the vitamin B12 deficiency) helps establish a cause-effect relationship. Women who fracture their hips might become vitamin B12 deficient after the fracture because they have reduced food intake or received treatments to suppress acid (and vitamin B12) production.
The prospective design allows you to design good methods for measuring the predictor variable (i.e., serum vitamin B12).
The cohort design avoids the sampling bias that is always a possibility in case-controlled studies if the women with fractures come from a different accessible population than those without fractures. Disadvantages of the prospective cohort design: A prospective cohort study will require many subjects followed for multiple years. The study will therefore be very expensive. - c. A retrospective cohort study could be done if you could find a cohort with stored serum or records on dietary vitamin B12 intake and with reasonably complete follow-up to determine who developed hip fracture. The main advantage of this design is that it would be less time consuming and expensive. The major drawback is that measurements of vitamin B12 in the serum might be altered by the storage, and that measurements of potential confounders (such as age, race, physical activity, cigarette smoking, etc.) may not be available.