B5.051 - Scholarly Projects

Question 1

In preparation for applying for a grant, researchers needed to determine if there was an association between spanking in childhood and adult physical health. Using publicly-available data from a national survey, researchers constructed a plan for a secondary data analysis. Survey participants had answered questions about their exposure to spanking before the age of 18 years and whether they had a variety of medical conditions in the past year. Participants had also answered questions about sociodemographic characteristics and family medical history. When researchers were describing the results of their analysis, they acknowledged several limitations including the lack of ability to make inferences about causality, no data on long-term health outcomes, and the use of self-reported data as opposed to medical record data (reporting bias, recall bias). When writing their grant, what study design could the authors propose to conduct that would address some of these limitations?

Answer 1

Cohort study.

A cohort study is an observational study, just like a cross-sectional study, but the study has a forward directionality. The exposure of interest is observed before the outcome of interest, so temporality can be definitely established.

Question 2

To assess the value of early cardiopulmonary resuscitation, researchers conducted a retrospective cohort study using data from the Swedish Cardiac Arrest Registry (January 1, 1990 through December 31, 2011). The researchers reported that the data showed a positive correlation (adjusted odds ratio, 2.15; 95% confidence interval, 1.88 to 2.45). What key piece of information is missing from this description?

Answer 2

A definition of the outcome being assessed.

The scenario described presents data that suggest that early cardiopulmonary resuscitation “works”, but not how that effectively is evaluated.

Question 3

A study is funded by the tobacco industry to examine the association between smoking and lung cancer. The researchers design a study with a prospective cohort of 1,000 smokers and 1,000 non-smokers between the ages of 20-30 years. The length of the study is ten years. After the study period ends, the researchers conclude that there is no relationship between smoking and lung cancer. Which of the following study features is the most likely reason for the failure of the study to note an association between tobacco use and cancer?

Answer 3

Latency period.

Lung cancer often takes several decades to develop, so a study that enrolls younger adults and only follows then for ten years may falsely conclude that there is no relationship between smoking and the development of lung cancer.

Question 4

what is a latency period

Answer 4

A latency period is seen when the negative effects of a disease take a long time to become clinically apparent. Importantly, this is not a type of bias. Rather, it is an intrinsic feature of a disease process.

Question 5

what type of study is useful to study diseases with a long latency period?

Answer 5

Case-control studies are useful to study diseases with a long latency period, such as cancer, as they do not require a long follow-up period at the time study is initiated.

Question 6

what is effect modification

Answer 6

Effect modification occurs when a third factor affects the magnitude of the relationship between the exposure and the disease (e.g. the increased risk of cancer in smokers is higher among those who also drink heavily).

Question 7

what is a pygmalion effect?

Answer 7

A Pygmalion effect occurs when an investigator inadvertently conveys his expectations to subjects, who then produce the expected result.

Question 8

what is a late look bias?

Answer 8

Late-look bias occurs when patients with severe disease are less likely to be studied because they are more likely to succumb to their disease before being studied.

Question 9

what is confounding bias

Answer 9

Confounding occurs when a third factor is either positively or negatively associated with both the exposure and outcome.

Question 10

Study X examined the relationship between coffee consumption and lung cancer. The authors of Study X retrospectively reviewed patients’ reported coffee consumption and found that drinking greater than 6 cups of coffee per day was associated with an increased risk of developing lung cancer. However, Study X was criticized by the authors of Study Y. Study Y showed that increased coffee consumption was associated with smoking. What type of bias likely affected Study X?

Answer 10

Confounding.

The study was subject to confounding, because the increase in coffee consumption was associated with increased cancer, however there was an additional variable that more likely caused the increase in lung cancer. This other variable was smoking; smoking was confounding the relationship between coffee drinking and lung cancer, throwing off the interpretation of the results in this study.

Question 11

when does confounding occur

Answer 11

Confounding occurs when a third factor is either positively or negatively associated with both the exposure and outcome of interest. Confounding can distort the true association between exposure and outcome and is a type of bias that must be addressed in observational studies.

Question 12

what is observer bias

Answer 12

Observer bias occurs when the observer is aware of what exposure group (intervention or control) the study subject has been assigned to, leading to possible bias when the observer interprets results in light of this awareness.

Question 13

what helps reduce observer bias

Answer 13

Blind ascertainment of outcomes helps reduce observer bias.

Question 14

when does lead time bias occur

Answer 14

Lead-time bias occurs with earlier detection of a disorder which makes the prognosis seem better when in reality the disease was merely detected earlier.

Question 15

when does selection bias occur

Answer 15

Selection bias occurs when the selected subjects are not representative of the population to be studied.

Question 16

Study X examined the relationship between coffee consumption and lung cancer. The authors of Study X retrospectively reviewed patients’ reported coffee consumption and found that drinking greater than 6 cups of coffee per day was associated with an increased risk of developing lung cancer. However, Study X was criticized by the authors of Study Y. Study Y showed that increased coffee consumption was associated with smoking. What study design technique could hypothetically be used to address the likely bias that was present in Study X?

Answer 16

Randomization.

The results of Study X were likely confounded by smoking. Smoking is associated with coffee drinking and with the development of lung cancer. Thus, a third factor distorted the relationship between the exposure of interest (coffee drinking) and the outcome of interest (lung cancer). The best way to control confounding in a non-observational study is to randomize participants into exposure groups; this approach balances the exposure groups on all factors (all possible confounders) except the exposure of interest.

Question 17

A new antihypertensive medication is studied in 3,000 white men with coronary heart disease who are over the age of 65 years. Results show that the new drug leads to improved morbidity and mortality as well as a decreased rate of acute coronary events with minimal side effects. After hearing about this new medication and supporting study at a recent continuing education course, a family physician elects to prescribe this medication to a 39-year-old Hispanic woman who presents with primary hypertension. After a one month trial and appropriate adjustments in the dosing, the patient’s blood pressure is not well controlled by this medication. What statistical concept could explain this patient’s poor response to the medication?

Answer 17

Generalizability.

Generalizability is the ability to use results from a study to draw conclusions about populations different than that used in the study (external study validity). Problems with generalizability can occur when a study is conducted in a very specific population and an attempt is made to apply the findings beyond the represented study population. Sometimes, the generalizability of study findings to a larger population is appropriate; sometimes is not.

Question 18

what is one of the key factors to consider in generalizability?

Answer 18

biologic plausibility
is there a biologic reason that the results can’t be applied to a broader population? Decisions about generalizability are subjective. However, in general, the stricter the inclusion and/or exclusion criteria for a study, the less generalizable the findings will be to the population at large.

Question 19

A team of researchers was interested in investigating the safety and efficacy of a standardized group clinic approach to management of heart failure. They designed a randomized controlled trial to test whether a group clinic intervention resulted in superior outcomes for heart failure patients. Patients were randomized to receive either standard of care or the experimental intervention. The primary outcome was time (in days) to the first rehospitalization or death due to heart failure within 12 months of randomization. What action can result in increased risk of type 1 error?

Answer 19

Upon seeing the data, the investigators decide to change the primary outcome from time to first rehospitalization or death due to any cause because a lack of heart failure deaths results in a negative (i.e., no statistically significant difference) study.

This reflects outcome switching. Any ad-hoc change to a study’s design can jeopardize the validity of its results. However, basing the decision to change a study on whether the change is likely to increase the chances of a significant finding can drastically increase the risk of a type I error.

Question 20

If investigators collect additional outcome measures, including outcomes that are not known or predicted to be associated with a patient’s response to treatment, what can this cause?

Answer 20

increased risk of a type I error (i.e., a false positive finding)
Only outcomes that are known or predicted to be associated with patient’s response to treatment should be collected in order to minimize the chances of outcome switching and p-hacking.